Split (1)

train · 36.7k rows

title stringlengths 1 827 ⌀	uuid stringlengths 36 36	pmc_id stringlengths 5 8	search_term stringclasses 44 values	text stringlengths 0 8.58M
Verbal and Social Autopsy of Adult Deaths and Adult Care-Seeking Pattern in Mozambique, 2019–2020	2c5af270-c78a-4718-9575-25dc3ef8d993	10160866	Forensic Medicine[mh]	There is a scarcity of data to ascertain the causes of adult deaths nationally and regionally in Mozambique. A post-census mortality survey 2007–2008 reported that HIV/AIDS, malaria, circulatory diseases, tuberculosis, injury, and diarrheal diseases are among major causes of adult deaths in Mozambique. The distribution of adult causes of death varies among age groups and by sex. A systematic statistical model–based analysis to ascertain causes of death in 195 countries during 1980–2017 documented a continuing disparity in death rates by sex across different age groups amidst variability in causes of death in different age groups. Age-specific mortality patterns have changed over the years, which is documented in a study conducted in central Mozambique showing a dramatic shift to the age group 15–49 years (49%) in the post-war decade (1993–2003) compared with mortality in under-five children, which was the main contributor (58%) to all deaths during war time. In Mozambique, like many other sub-Saharan African countries, many people are born and die before being formally registered, and most demographic data are obtained using census or sample survey, including indirect estimations. Similar to most low-income countries, Mozambique does not currently have a functioning civil registration and vital statistics system that is able to produce complete and high-quality mortality data for monitoring recent trends in mortality and causes of death. Therefore, the country currently relies on national household surveys such as the Demographic and Health Surveys and the Multiple Indicator Cluster Surveys to measure mortality, but these sources are not able to generate recent and timely mortality data. The lack of the capacity to detect causes of death due to the fragility of the health system and a lack of adequate data makes it challenging to monitor mortality by cause among adults in Mozambique. , When innovative approaches are urgently needed to support the country to effectively monitor levels and trends in mortality as well as causes of death, a cause of death may be ascertained using information obtained from bereaved relatives through verbal autopsy (VA). A cause of death may be assigned either by physician review of the VA data or following a set of predefined diagnostic criteria given in an algorithm. Also, to understand the social barriers and factors related to the deaths, a social autopsy (SA) can be implemented to identify bottlenecks in the family and community related to the lack of prompt formal care-seeking and to increase and facilitate response by the community. The integration of the VA and SA tools permits a simultaneous collection of these two types of information and the ability to correlate them for improved understanding of social, economic, and cultural factors related to specific causes of death. In this context, in January 2017, Mozambique launched the Countrywide Mortality Surveillance for Action (COMSA) to establish a national sampling registration system to monitor mortality and causes of death at the national and subnational level, including the use of VA to ascertain the causes of death. Knowledge of the causes of death helps governments and their partners to allocate resources and make decisions to identify disease prevention priorities. This article analyzes data collected through verbal and social autopsy of adult deaths under the COMSA project and provides countrywide (national and regional level) cause-specific information of adult deaths and adult care-seeking that is useful for formulating policy and developing programs to improve adult health in Mozambique. The objectives of this study are to 1) identify the biological causes of adult deaths in Mozambique, 2) identify the social causes of adult deaths, and 3) examine care-seeking during illnesses among deceased adults in Mozambique. Our verbal and social autopsy of adult deaths includes the analysis of data collected through a structured questionnaire from a nationally representative sample of 700 clusters randomly selected within each of Mozambique’s 11 provinces. In this paper, we analyzed deaths of persons aged 18 years (as considered adult age in Mozambique) and above, stratified into 18–49 years and 50+ years, for a total of 4,040 deaths reported during years 2019–2020. We used the InSilicoVA method from the openVA R package, an improved version of the InterVA method, to classify biological causes of adult deaths. In our analysis, we grouped individual causes into broader groups. These include cancer, cardiovascular diseases, HIV, injury, maternal causes, pneumonia, tuberculosis, other causes, and other infections. Cardiovascular diseases included ischemic heart disease and stroke. Other causes of death included other and unspecified noncommunicable diseases, acute abdomen disease, acute cardiac disease, asthma, chronic obstructive pulmonary disease, chronic respiratory illness, diabetes mellitus, epilepsy, liver cirrhosis, renal failure, sickle cell with crisis, other and unspecified cardiac disease, and severe malnutrition. Other infections included dengue fever, hemorrhagic fever (non-dengue), measles, meningitis and encephalitis, other and unspecified infections, pertussis, sepsis (non-obstetric), tetanus, malaria, and diarrhea. In this analysis, we have examined social autonomy and social capital as independent variables if they played a role in care-seeking of the deceased. Social autonomy has been defined as being an active participant of community groups including vocational training group; savings group or microcredit program; community cooperative, such as an agricultural cooperative; political group; religious group; sports club; youth/student club; women’s group; and other groups. Social capital is characterized as having people in the community working together on community issues (education/schools, health services/clinics, paid job opportunities, credit/finance, roads, public transportation, water distribution, sanitation services, agriculture, justice/conflict resolution, security/police services, mosque/church/temple, and other issues) that affect the entire or part of the community. Principal component analysis was conducted to create a composite measure of wealth index from 20 variables concerning household’s ownership of assets. It was then divided into tertiles to categorize the lowest, middle, and highest levels in terms of household wealth. We used the Pathway to Survival model to analyze the steps of care-seeking and possible breakdowns in continuity of care that may have contributed to causes of adult deaths. Responses related to care-seeking were categorized as no care-seeking, care sought at home, and care sought outside home. Seeking care outside home included care sought from a formal provider (doctor, nurse/midwife, and trained community health worker), care sought from an informal provider (traditional providers, family members, and pharmacists), and care sought from formal or informal providers. In addition to Pathway to Survival analysis, we have further explored socio-demographic factors that are associated with care-seeking from formal and informal providers. We have conducted multinomial regression analysis (informal, informal or formal, and formal being the outcome variables) and other socio-demographic characteristics of deceased as independent variables. A description of independent variables that were used to characterize the deceased is provided in . Overall, in this paper we report findings that are obtained from descriptive data analysis. To ascertain the relationship between two categorical variables, we conducted χ 2 tests. The P value for significant statistical association was determined at the 0.05, 0.01, and 0.001 levels. Socio-demographic characteristics of deceased adults. Of the 4,040 total adult deaths, about half (51%) were males. The proportion of male deaths was higher in the age group 18–49 years (52.66%). The older age group was equally distributed between males and females. Over 45% of deceased adults did not have any education, 44% had primary level education, and the proportion was significantly higher in the age group 18–49 years compared with the age group 50+ years (51.0% versus 39.11%). About 56% of deceased adults were married, significantly higher in age group 18–49 years (61.95%) compared with the 50+ years age group (51.97%). Nearly 28% of all adults were used, although the proportion was significantly lower in the 50+ years age group (25.23%) compared with their counterparts (30.99%). The household wealth of deceased adults was divided into tertiles. Nearly half of deceased adults were reported to have social autonomy, and about 78% possessed social capital. Overall, 73% lived in rural areas and 27% in urban areas. The proportion of deceased adults in the 50+ years age group was significantly higher in rural areas compared with that in the 18–49 years age group (73.48% versus 71.94%). The majority of adults came from the central region (43.60%), and the proportion from this region is significantly higher in the 18–49 years age group compared with the 50+ years age group (50.21% versus 38.84%) ( ). Causes of adult deaths by age. The major causes of adult deaths were HIV (17.09%), cancer (13.3%), injury (9.92%), cardiovascular diseases (9.45%), pneumonia (6.84%), tuberculosis (4.78%), maternal causes (2.92%), other infections (16.25%), and other causes (19.45%). HIV causes more deaths at age 18–49 (28.36%) than at age 50+ (8.96%). Likewise, injury was more common at age 18–49 (13.43%) than at age 50+ (7.39%) On the other hand, deaths due to cardiovascular diseases were more common at age 50+ (13.52%) than at age 18–49 (3.79%), as were deaths due to cancer (16.56% versus 8.78%), tuberculosis (5.37% versus 3.97%), other infections (17.1% versus 15.07%), and other causes (24.21% versus 12.84%). Among those 18–49 years old, about 6.98% of deaths were maternal deaths; this category does not apply at age 50+ years ( ). Causes of adult deaths by sex and place of residence. When stratified by sex, the proportions of deaths due to cancer were significantly higher among women as opposed to men (13.81% versus 12.82%, P < 0.001), with similar patterns for HIV (19.12% versus 15.15%, P < 0.001) and cardiovascular diseases (9.81% versus 9.11%, P < 0.001). On the other hand, more male deaths were attributed to injury (12.6% versus 7.13%), pneumonia (8.27% versus 5.36%), tuberculosis (5.74% versus 3.79%), and other causes (20.09% versus 18.74%). The proportion of maternal deaths among women was 5.97% ( ). Comparing adults of rural residence with those of urban residence, rural residents had higher proportions of deaths due to cancer (14.3% versus 10.62%), maternal causes (3.2% versus 2.17%), and other infections (17.85% versus 11.95%). Residents of urban areas had higher proportions of deaths due to injury (12.15% versus 9.09%), cardiovascular diseases (11.64% versus 8.63%), and other causes (20.62% versus 19.01%) ( ). Causes of adult deaths by region and province. In the northern region, the proportion of deaths due to cancer (15.91%), HIV (18.9%), maternal causes (4.36%), and tuberculosis (5.85%) were higher than in other regions. The southern region included the highest proportions of deaths due to cardiovascular diseases (12.52%), injury (12.24%), pneumonia (7.65%), and other causes (23.23%) compared with other regions ( ). shows the distribution of causes of all deaths by 11 provinces of Mozambique compared with the national average. Proportions of cancer deaths were higher in Cabo Delgado (15.8%), Inhambane (14.81%), Zambezia (14.25%), Niassa (16.77%), and Nampula (15.53%). The proportions of deaths due to HIV were higher in Zambezia (20.08%), Nampula (18.44%), Gaza (17.26%), and Cabo Delgado (21.17%). The proportions of injury deaths were higher in Gaza (13.93%), Manica (12.08%), Maputo city (18.57%), Maputo province (14.87%), and Sofala (14.43%). The proportions of maternal deaths were higher in Cabo Delgado (5.16%), Manica (4.0%), Nampula (4.4%), Niassa (3.29%), Sofala (4.42%), and Zambezia (3.07%). The proportions of deaths associated with cardiovascular diseases were higher in Gaza (13.25%), Inhambane (12.35%), Maputo city (11.52%), Maputo province (12.28%), and Sofala (10.5%). The proportions of tuberculosis deaths were higher in Gaza (5.44%), Manica (6.01%), Maputo province (4.88%), Nampula (5.23%), Niassa (8.97%), and Tete (6.71%). The proportions of pneumonia deaths were higher in Ihambane (7.45%), Manica (7.25%), Maputo city (8.15%), Maputo province (13.23%), Niassa (7.21%), Sofala (8.43%), and Tete (8.48%), compared with the national average ( ). Causes of adult deaths by place of deaths. Of all deaths, 79.14% took place in the community (at home, en route to hospital, or other places in the community); the rest (20.86%) occurred in a health facility. Data showed variability in place of deaths (facility versus Community) when analyzed by causes of death. Higher proportions of deaths due to HIV, pneumonia, and tuberculosis took place in the health facilities compared with the community (20.89% versus 16.08%, 9.36% versus 6.18%, and 5.11% versus 4.7%, respectively). On the other hand, higher proportions of deaths due to cancer, cardiovascular diseases, injury, other infections, and other causes occurred in the community compared with the facility (14.3% versus 9.51%, 10.09% versus 7.01%, 10.32% versus 8.39%, 16.85% versus 13.96%, and 19.59% versus 18.93%, respectively) ( ). Pathway to survival indicators/components. illustrates the steps and possible breakdowns in the Pathway to Survival that may have contributed to death. shows the distribution of these indicators by age group. When the deceased adults or their caregivers first noticed that they (deceased adults) were ill, healthcare was obtained or sought outside the home as a first action for most deaths (71.62%). The distribution is similar across age groups 18–49 years and 50+ years. For nearly one-fifth of adult deaths, the caregiver did not seek care after the illness was noticed (19.75%), and the proportion is higher for adults aged 18–49 years (22.04%) compared with adults aged 50+ years (18.10%). About 4% of deceased adults received only home care. Receiving home care was much higher among deceased adults aged 50+ years (5.30%) as opposed to adults aged 18–49 years (2.27%). For those who sought or tried to seek any outside care, the majority sought formal care only (66.90%); however, the proportion of formal care-seeking was higher in the younger adult group (18–49 years) (72.85%) compared with older adult group (62.66%). Contrastingly, informal care-seeking was higher among adults aged 50+ years (9.64%) compared with adults aged 18–49 years (3.93%). A combination of informal and formal care was sought more among deceased older adults (50+ years) compared with deceased younger adults (18–49 years) (27.29% versus 22.59%). Among adults who reached the first health provider, 12.7% died at the facility. The vast majority of deceased adults left the first health provider/facility alive (85.90%). Only 18.16% of adults who reached and left the first provider alive were referred to a second provider, and the rest either received home care or did not receive any home care recommendation. A majority of those who were referred complied and went to a second health provider (85.19%), and the proportion was higher among deceased adults aged 18–49 years (88.42%) compared with deceased adults aged 50+ years (82.38%) ( ). Care-seeking among deceased adults. shows the proportional distribution of formal care-seeking by characteristics of deceased adults. We have analyzed the data between age groups 18–49 years and 50+ years. The distribution of care-seeking from formal providers by sex was similar. However, a higher proportion of female adults (50.41%) in age group 18–49 years and a higher proportion of male adults (50.75%) in age group 50+ years sought care from formal providers. Of 2,056 deceased adults who sought care from formal providers, 40% had no formal education, 46.59% had primary level education, and 13.43% had secondary or higher education. The proportions of primary or secondary or higher were higher among the adults in age group 18–49 years compared with older adults (51.54% versus 42.50% and 22.99% versus 5.51%). Nearly 60% of the deceased adults who were married or had a life partner had sought care from a formal provider, and the proportion was higher among those aged 18–49 years (63.67%) and lower among those aged 50+ years (54.07%). Less than one-third of the deceased adults who were employed and nearly two-thirds of the deceased adults who resided in rural areas sought care from formal providers. Over half of the deceased adults having social autonomy and nearly three-fourths having social capital sought care from formal providers. The majority of the deceased adults that came from central region sought care from formal providers, and the proportion was higher in the 18–49 age group (48.27%) than in the 50+ age group (41.55%). Factors associated with care-seeking among deceased adults. We conducted multivariate analysis to ascertain factors associated with seeking care among adults before their death. We compared the factors associated with seeking care from informal providers or from informal and formal providers combined ( ). Also, we compared care sought from formal providers as opposed to informal providers in relation to these factors ( ). Adults in the 50+ years age group and adults who had primary level education were significantly more likely to have sought care from informal or formal providers than to have sought care from informal providers (relative risk ratio [RRR]: 1.27, P < 0.05; RRR: 1.43, P < 0.05). Compared with deceased female adults, male adults were significantly less likely to have sought care from informal or formal providers compared with informal providers alone (RRR: 0.69, P < 0.01). Adults from the highest tertile were more likely than those from the lowest tertile to have sought care from informal or formal providers compared with informal providers (RRR: 2.36, P < 0.001). Deceased adults from southern region were significantly less likely to have sought care from informal or formal providers compared with those from the northern region (RRR: 0.59, P < 0.05) ( ). Comparing care-seeking from informal with formal providers, adults who had had primary, secondary, or higher level of education were more likely to have sought care from formal providers than to have sought care from informal providers (RRR: 1.57, P < 0.001; RRR: 1.70, P < 0.01). Compared with deceased female adults, deceased male adults were less likely to have sought care from formal providers than to have sought care from informal providers (RRR: 0.68, P < 0.001). Adults who were married or had had a life partner as opposed to those who were single, divorced, separated, or widowed were more likely to have sought care from formal providers as opposed to informal providers (RRR: 1.30, P < 0.01). Adults who had the highest household wealth were 3.14 times more likely to have sought care from formal providers than those having the lowest household wealth (RRR: 3.14, P < 0.001). Social capital seemed to have a negative association with care-seeking from formal providers as opposed to informal providers (RRR: 0.73, P < 0.05). Compared with adults from the northern region, adults from the southern region were less likely to have sought care from formal providers than to have sought care from informal providers (RRR: 0.43, P < 0.001) ( ). Of the 4,040 total adult deaths, about half (51%) were males. The proportion of male deaths was higher in the age group 18–49 years (52.66%). The older age group was equally distributed between males and females. Over 45% of deceased adults did not have any education, 44% had primary level education, and the proportion was significantly higher in the age group 18–49 years compared with the age group 50+ years (51.0% versus 39.11%). About 56% of deceased adults were married, significantly higher in age group 18–49 years (61.95%) compared with the 50+ years age group (51.97%). Nearly 28% of all adults were used, although the proportion was significantly lower in the 50+ years age group (25.23%) compared with their counterparts (30.99%). The household wealth of deceased adults was divided into tertiles. Nearly half of deceased adults were reported to have social autonomy, and about 78% possessed social capital. Overall, 73% lived in rural areas and 27% in urban areas. The proportion of deceased adults in the 50+ years age group was significantly higher in rural areas compared with that in the 18–49 years age group (73.48% versus 71.94%). The majority of adults came from the central region (43.60%), and the proportion from this region is significantly higher in the 18–49 years age group compared with the 50+ years age group (50.21% versus 38.84%) ( ). The major causes of adult deaths were HIV (17.09%), cancer (13.3%), injury (9.92%), cardiovascular diseases (9.45%), pneumonia (6.84%), tuberculosis (4.78%), maternal causes (2.92%), other infections (16.25%), and other causes (19.45%). HIV causes more deaths at age 18–49 (28.36%) than at age 50+ (8.96%). Likewise, injury was more common at age 18–49 (13.43%) than at age 50+ (7.39%) On the other hand, deaths due to cardiovascular diseases were more common at age 50+ (13.52%) than at age 18–49 (3.79%), as were deaths due to cancer (16.56% versus 8.78%), tuberculosis (5.37% versus 3.97%), other infections (17.1% versus 15.07%), and other causes (24.21% versus 12.84%). Among those 18–49 years old, about 6.98% of deaths were maternal deaths; this category does not apply at age 50+ years ( ). When stratified by sex, the proportions of deaths due to cancer were significantly higher among women as opposed to men (13.81% versus 12.82%, P < 0.001), with similar patterns for HIV (19.12% versus 15.15%, P < 0.001) and cardiovascular diseases (9.81% versus 9.11%, P < 0.001). On the other hand, more male deaths were attributed to injury (12.6% versus 7.13%), pneumonia (8.27% versus 5.36%), tuberculosis (5.74% versus 3.79%), and other causes (20.09% versus 18.74%). The proportion of maternal deaths among women was 5.97% ( ). Comparing adults of rural residence with those of urban residence, rural residents had higher proportions of deaths due to cancer (14.3% versus 10.62%), maternal causes (3.2% versus 2.17%), and other infections (17.85% versus 11.95%). Residents of urban areas had higher proportions of deaths due to injury (12.15% versus 9.09%), cardiovascular diseases (11.64% versus 8.63%), and other causes (20.62% versus 19.01%) ( ). In the northern region, the proportion of deaths due to cancer (15.91%), HIV (18.9%), maternal causes (4.36%), and tuberculosis (5.85%) were higher than in other regions. The southern region included the highest proportions of deaths due to cardiovascular diseases (12.52%), injury (12.24%), pneumonia (7.65%), and other causes (23.23%) compared with other regions ( ). shows the distribution of causes of all deaths by 11 provinces of Mozambique compared with the national average. Proportions of cancer deaths were higher in Cabo Delgado (15.8%), Inhambane (14.81%), Zambezia (14.25%), Niassa (16.77%), and Nampula (15.53%). The proportions of deaths due to HIV were higher in Zambezia (20.08%), Nampula (18.44%), Gaza (17.26%), and Cabo Delgado (21.17%). The proportions of injury deaths were higher in Gaza (13.93%), Manica (12.08%), Maputo city (18.57%), Maputo province (14.87%), and Sofala (14.43%). The proportions of maternal deaths were higher in Cabo Delgado (5.16%), Manica (4.0%), Nampula (4.4%), Niassa (3.29%), Sofala (4.42%), and Zambezia (3.07%). The proportions of deaths associated with cardiovascular diseases were higher in Gaza (13.25%), Inhambane (12.35%), Maputo city (11.52%), Maputo province (12.28%), and Sofala (10.5%). The proportions of tuberculosis deaths were higher in Gaza (5.44%), Manica (6.01%), Maputo province (4.88%), Nampula (5.23%), Niassa (8.97%), and Tete (6.71%). The proportions of pneumonia deaths were higher in Ihambane (7.45%), Manica (7.25%), Maputo city (8.15%), Maputo province (13.23%), Niassa (7.21%), Sofala (8.43%), and Tete (8.48%), compared with the national average ( ). Of all deaths, 79.14% took place in the community (at home, en route to hospital, or other places in the community); the rest (20.86%) occurred in a health facility. Data showed variability in place of deaths (facility versus Community) when analyzed by causes of death. Higher proportions of deaths due to HIV, pneumonia, and tuberculosis took place in the health facilities compared with the community (20.89% versus 16.08%, 9.36% versus 6.18%, and 5.11% versus 4.7%, respectively). On the other hand, higher proportions of deaths due to cancer, cardiovascular diseases, injury, other infections, and other causes occurred in the community compared with the facility (14.3% versus 9.51%, 10.09% versus 7.01%, 10.32% versus 8.39%, 16.85% versus 13.96%, and 19.59% versus 18.93%, respectively) ( ). illustrates the steps and possible breakdowns in the Pathway to Survival that may have contributed to death. shows the distribution of these indicators by age group. When the deceased adults or their caregivers first noticed that they (deceased adults) were ill, healthcare was obtained or sought outside the home as a first action for most deaths (71.62%). The distribution is similar across age groups 18–49 years and 50+ years. For nearly one-fifth of adult deaths, the caregiver did not seek care after the illness was noticed (19.75%), and the proportion is higher for adults aged 18–49 years (22.04%) compared with adults aged 50+ years (18.10%). About 4% of deceased adults received only home care. Receiving home care was much higher among deceased adults aged 50+ years (5.30%) as opposed to adults aged 18–49 years (2.27%). For those who sought or tried to seek any outside care, the majority sought formal care only (66.90%); however, the proportion of formal care-seeking was higher in the younger adult group (18–49 years) (72.85%) compared with older adult group (62.66%). Contrastingly, informal care-seeking was higher among adults aged 50+ years (9.64%) compared with adults aged 18–49 years (3.93%). A combination of informal and formal care was sought more among deceased older adults (50+ years) compared with deceased younger adults (18–49 years) (27.29% versus 22.59%). Among adults who reached the first health provider, 12.7% died at the facility. The vast majority of deceased adults left the first health provider/facility alive (85.90%). Only 18.16% of adults who reached and left the first provider alive were referred to a second provider, and the rest either received home care or did not receive any home care recommendation. A majority of those who were referred complied and went to a second health provider (85.19%), and the proportion was higher among deceased adults aged 18–49 years (88.42%) compared with deceased adults aged 50+ years (82.38%) ( ). shows the proportional distribution of formal care-seeking by characteristics of deceased adults. We have analyzed the data between age groups 18–49 years and 50+ years. The distribution of care-seeking from formal providers by sex was similar. However, a higher proportion of female adults (50.41%) in age group 18–49 years and a higher proportion of male adults (50.75%) in age group 50+ years sought care from formal providers. Of 2,056 deceased adults who sought care from formal providers, 40% had no formal education, 46.59% had primary level education, and 13.43% had secondary or higher education. The proportions of primary or secondary or higher were higher among the adults in age group 18–49 years compared with older adults (51.54% versus 42.50% and 22.99% versus 5.51%). Nearly 60% of the deceased adults who were married or had a life partner had sought care from a formal provider, and the proportion was higher among those aged 18–49 years (63.67%) and lower among those aged 50+ years (54.07%). Less than one-third of the deceased adults who were employed and nearly two-thirds of the deceased adults who resided in rural areas sought care from formal providers. Over half of the deceased adults having social autonomy and nearly three-fourths having social capital sought care from formal providers. The majority of the deceased adults that came from central region sought care from formal providers, and the proportion was higher in the 18–49 age group (48.27%) than in the 50+ age group (41.55%). We conducted multivariate analysis to ascertain factors associated with seeking care among adults before their death. We compared the factors associated with seeking care from informal providers or from informal and formal providers combined ( ). Also, we compared care sought from formal providers as opposed to informal providers in relation to these factors ( ). Adults in the 50+ years age group and adults who had primary level education were significantly more likely to have sought care from informal or formal providers than to have sought care from informal providers (relative risk ratio [RRR]: 1.27, P < 0.05; RRR: 1.43, P < 0.05). Compared with deceased female adults, male adults were significantly less likely to have sought care from informal or formal providers compared with informal providers alone (RRR: 0.69, P < 0.01). Adults from the highest tertile were more likely than those from the lowest tertile to have sought care from informal or formal providers compared with informal providers (RRR: 2.36, P < 0.001). Deceased adults from southern region were significantly less likely to have sought care from informal or formal providers compared with those from the northern region (RRR: 0.59, P < 0.05) ( ). Comparing care-seeking from informal with formal providers, adults who had had primary, secondary, or higher level of education were more likely to have sought care from formal providers than to have sought care from informal providers (RRR: 1.57, P < 0.001; RRR: 1.70, P < 0.01). Compared with deceased female adults, deceased male adults were less likely to have sought care from formal providers than to have sought care from informal providers (RRR: 0.68, P < 0.001). Adults who were married or had had a life partner as opposed to those who were single, divorced, separated, or widowed were more likely to have sought care from formal providers as opposed to informal providers (RRR: 1.30, P < 0.01). Adults who had the highest household wealth were 3.14 times more likely to have sought care from formal providers than those having the lowest household wealth (RRR: 3.14, P < 0.001). Social capital seemed to have a negative association with care-seeking from formal providers as opposed to informal providers (RRR: 0.73, P < 0.05). Compared with adults from the northern region, adults from the southern region were less likely to have sought care from formal providers than to have sought care from informal providers (RRR: 0.43, P < 0.001) ( ). The major causes of adult deaths in Mozambique are HIV (17.09%), cancer (13.3%), injury (9.92%), cardiovascular diseases (9.45%), pneumonia (6.84%), tuberculosis (4.78%), and maternal causes (2.92%). HIV and injury cause more deaths in younger adults (18–49 years), whereas cancer, cardiovascular diseases, pneumonia, and tuberculosis cause more deaths among adults 50+ years of age. When stratified by sex, among all adult deaths, HIV, cancer, and cardiovascular diseases appear to cause more deaths in females compared with males, for whom injury, pneumonia, and tuberculosis were the main causes. The post-census mortality survey 2007–2008, Mozambique conducted about 12 years prior to our study reported HIV/AIDS (40%), malaria (14%), circulatory diseases (7%), tuberculosis (6%), injury (6%), and diarrheal diseases (3%) as the major causes of deaths among adults age 15 and older. The post-census mortality survey also used VA but assigned cause of death through independent review of each VA questionnaire by two trained physicians. In our study, cardiovascular diseases include ischemic heart disease and stroke, whereas the earlier study reported circulatory diseases. One study conducted in northern Ethiopia identified causes of death using 723 verbal autopsy interviews of death of adults aged 15+ years (2009–2013), and the major causes of deaths were tuberculosis (15.9%), cerebrovascular diseases (7.3%), and accidental falls (3.9%). Even though the Ethiopian study used a methodology similar to ours, the context is different, having a lower prevalence of HIV. In our study, HIV as a single cause appears to be the leading cause of death for both the residents in rural and urban areas because nearly one-fifth of the adult population die of HIV/AIDS. The similar findings appeared in the post-census mortality survey conducted about a decade ago, suggesting that HIV/AIDS has been the major killer among the adult population in the last decade regardless of place of residence. Whereas cancer (14.3%) and injury (9.1%) have been the second and third causes of adult deaths in rural areas, injury (12.2%) and cardiovascular diseases (11.6%) take those places in urban areas. The post-census mortality survey conducted a decade ago also demonstrated the circulatory diseases (9%) and accidents and external causes (7%) were the third and fourth leading causes of adult death in urban areas; however, the data suggest that the proportion of injury-related deaths doubled (7%–12.2%) in the last decade. Study findings suggested monitoring of injury-related deaths and developing and implementing injury prevention programs to avert injury-related deaths among those living in urban areas. In this study, HIV and cancer appear as the leading causes of deaths for all the provinces. While comparing with 2007–2008 post-census mortality survey, we observed no change in the trend during the last 12 years. However, our study demonstrates that the proportion of some causes of deaths are higher in some provinces compared with other provinces ( ). This warrants development and strengthening of province-specific disease prevention strategies and implementation of the strategies. We defined social capital considering responses about having people in the community working together on community issues that affect entire or part of community. Findings from adjusted analysis using multinomial regression models showed that social capital was positively associated with deceased adults’ care -eking from informal or formal providers compared with care-seeking from informal providers (RRR: 1.222), even though the association was not statistically significant. Interestingly, when we examined adult care-seeking from formal providers as opposed to informal providers, we found a negative association with social capital (RRR: 0.674). Adults’ social autonomy, as defined as active participation with community groups, appeared to have positively associated with care-seeking from formal providers as opposed to informal providers (RRR: 1.046) and care-seeking from informal or formal providers compared with informal providers (RRR: 1.142), although the associations were not statistically significant ( and ). Bakeera et al. explored the role of social capital in utilization of healthcare services by children in Uganda. In this study, social capital was measured by assessing providers’ responses to questions related to civic trust, social support, reciprocity, or willingness of community to help each other out. Study findings from adjusted analysis controlling for potential confounding factors, including socio-demographic and socio-economic factors of service recipients and service providers, demonstrated that high levels of trust and medium levels of informational support (odds ratio [OR]: 2.75, 95% CI: 1.50–5.02; and OR: 1.68, 95% CI: 1.12–2.50, respectively) were positively associated with the use of a public facility compared with other treatment options (community medicine distributor, neighbor, drug shops, and others). Our study findings suggest significant variability in care-seeking across regions. Compared with adults from the north region, adults from the southern region are less likely to have sought care from formal providers than to have sought care from informal providers (RRR: 0.428, P < 0.001) ( ). Similarly, compared with adults from the north region, adults from the southern region are less likely to have sought care from informal or formal providers than to have sought care from informal providers (RRR: 0.592, P < 0.01) ( ). Care-seeking data from our study suggest that nearly one-fifth of the deceased did not seek any care before death, and 4% received home care. Nearly three-fourths sought care or tried to seek care outside the home, and the majority sought formal care. Care-seeking from formal providers was much higher among younger adults (18–49 years) compared with older adults (50+ years). The proportions of receiving home care and seeking informal care were much higher among older adults (50+ years) compared with their younger counterparts. The main barriers for care-seeking included physical distance of the health facility, unavailability of transportation to travel to health facility, and cost related to healthcare and transportation (data not shown). Whereas physical distance of health center and transportation appear to be the main barriers for maternity care-seeking, , access and utilization of available services depend on information, cost, and quality of services. These factors also play a critical role in healthcare seeking and healthcare utilization by adults. Furthermore, deep-rooted social beliefs, stigma, and family practices influence care-seeking for health, although practicalities of service availability and cost often outweigh the deeply held beliefs and practices. In this study, household wealth, a proxy indicator of respondent’s socio-economic status, appeared to be strongly associated with care-seeking during illnesses. Findings from multinomial regression analysis suggest, compared with the lowest tertile, respondents belonging to the highest tertile were more likely to have sought care from formal health providers as opposed to informal health providers (RRR: 3.14, P < 0.001). Similarly, compared with the lowest tertile, the highest tertiles were more likely to have sought care from informal or formal providers as opposed to informal providers (RRR: 2.363, P < 0.001). The positive association between socio-economic status of an individual and healthcare seeking behavior is documented in other studies. A study conducted in one subdistrict in Bangladesh revealed that people living outside the embankment (considered to be poorer) were significantly less likely to seek care from medically trained provider compared with people living inside the embankment (OR: 0.56, 95% CI: 0.286–0.834). The positive association of women’s socio-economic status and care-seeking from trained providers during pregnancy and delivery has been documented in earlier studies. – In our study, among those who reached and left the first provider alive, less than 20% were referred to a second provider, and the rest either received home care or did not receive any home care recommendation. Over four-fifths (83.24%) of those who were referred complied and went to a second health provider, and the proportion was higher among deceased adults aged 18–49 years (88.42%) compared with deceased adults aged 50+ years (82.38%) ( ). It is intriguing to note that despite the higher compliance rate, only 18.16% were referred. The lack of good physical infrastructure (roads, transports) and communication services especially in the rural areas of most low- and middle-income countries (LMICs) are barriers to referring and transporting patients in emergencies to higher-level health facilities. Quality of service delivery of a referral facility is also critical for the service provider for making a decision before referring a patient. Moreover, many LMICs lack an organized and active referral system. The Ministry of Health can demonstrate a leadership role in developing a functional referral system engaging public, private, and NGO health providers and facilities in LMICs. Our care-seeking analysis used the Pathway to Survival model to report sequential steps, possible breakdowns and itinerary of care, and failures in the pathway to survival that may have contributed to causes of adult deaths. Also, we have explored and reported socio-demographic factors that are associated with care-seeking from different providers. However, one of the limitations of our study is that it included the care-seeking information of the deceased only. Another limitation of our study is possible recall bias in providing information by the respondents on death and care-seeking of the deceased prior to death. Findings of this study about adult causes of death are useful for program planning and development of priority programs within the limited resources available for provision of healthcare. The information is also important for monitoring and evaluation of ongoing disease prevention programs, including malaria, HIV, and tuberculosis. Regional- and provincial-level cause of death information is pivotal for local level program planning and implementation and for strengthening provincial- and district-level health systems, including disease prevention programs and community outreach services. Study findings related to adult care-seeking not only identify demand and supply-side barriers but also help Mozambique health systems develop and strengthen affordable outreach services. The information emphasizes the need for developing community-driven programs to generate local-level resources and facilitate transportation to overcome the barriers in seeking care during illnesses. Financial support: The COMSA project is implemented through the generous support of the Bill & Melinda Gates Foundation through the Johns Hopkins University.
The Development and Evaluation of a Combined Infection–Rheumatology Assessment Service in Response to the Chikungunya Fever Epidemic	81fdc7e4-da77-439c-adc2-480f4fdcdadd	10160878	Internal Medicine[mh]
Chemical symphony of coumarins and phenazines in rhizosphere iron solubilization	122b7c63-2fbc-411b-b91b-4e28e98bf66d	10160995	Microbiology[mh]
Experiences with telemedicine among family medicine residents at king saud university medical city during the COVID-19 pandemic: a cross-sectional study	0da75ef8-8114-4503-b84b-84038017f77c	10161176	Family Medicine[mh]	Telemedicine has existed since the early 1960s. Over the years, it has developed and spread considerably, becoming a widely used tool for patient care. The World Health Organization defined telemedicine as “the delivery of health care services, where distance is a critical factor, by all health care professionals using information and communication technologies” [ p. 9]. Consulting through telephone consultation provides a promising alternative to in-person visits for general practice care . Moreover, telemedicine reduces inefficiencies in the delivery of healthcare, such as reducing patient travel and waiting time . In addition, telemedicine was found to be effective in specialty consultations, primary care assessments, preoperative assessments, and postoperative follow-ups . The World Health Organization declared a global pandemic after cases of coronavirus disease 2019 (COVID-19) were confirmed worldwide . During the COVID-19 pandemic, many hospitals in Saudi Arabia reduced the number of patients allowed to visit primary care clinics by more than 75% of their maximum capacity. This reduction led to the implementation of telemedicine clinics as a part of patient care. Telemedicine is an attractive solution to minimize the risk of virus transmission . The use of telemedicine is growing in clinical practice, and medical residents of the current generation grew up using technology as a major part of their daily lives. Medical residents believe that interactions with telemedicine during their training serve as an important educational tool that supports their understanding of core competencies in practice-based learning, medical knowledge, and patient care . In a 2015 survey of 207 family medicine residencies nationwide, the majority of program directors reported that their facilities had some form of telehealth services; however, actual use was limited and infrequent . The COVID-19 pandemic led the Ministry of Health of Saudi Arabia to accelerate the growth of digital health by creating and developing mobile health applications and adopting telemedicine in primary care clinics of many tertiary hospitals to improve patient care and minimize the risk of infection. Since many of the tertiary hospitals are teaching hospitals, the adoption of telemedicine introduced a new set of challenges to the family medicine residency program that family medicine residents are involved in as a part of their training. To the best of our knowledge, only one study, conducted in the USA, assessed the perceptions of family medicine residents regarding the use of telemedicine in their training . However, the sample of that study was small. Therefore, the present study aimed to assess experiences with telemedicine among family medicine residents during the COVID-19 pandemic. Study group A cross-sectional study was conducted with family medicine residents at King Saud University Medical City, Riyadh, Saudi Arabia. The family medicine residency program had 60 residents; all of them were included in the study, with a 100% response rate. Telemedicine clinic visits were introduced to the residents in October 2020. Prior to the clinic appointment, family medicine consultants sorted the booked cases based on the patient’s reason for booking and used their judgment to decide whether the patient required an in-person or telemedicine consultation. The cases included new and follow-up visits. Telemedicine clinics were conducted primarily through telephone rather than video calls. To assess the residents’ experience, a 20-item questionnaire survey was administered between March and April 2022. The questionnaire was anonymous to ensure confidentiality. Questionnaire design A validated electronic questionnaire from a similar study conducted in the USA was adapted after obtaining the author’s permission . A few adjustments were made to the questionnaire to match the family medicine program. The adjustments were reviewed and approved by three research experts of family medicine consultants. Questions 1 and 2 related to demographics, including gender and post-graduation year level (residency level). Questions 5, 6, 7, 16, and 17 focused on providing patient-centered care. Question 8 and its subsections assessed residents’ confidence. Common diseases managed in family clinics, including diabetes, dyslipidemia, hypertension, hypothyroidism, osteoarthritis, depression and anxiety, chronic headaches, urinary tract infections, back pain, and congestive heart failure, were chosen. Questions 9–11 evaluated system-based practice and work in interdisciplinary teams. Questions 12–15 focused on practice-based learning and improvement and asked family medicine residents about their clinical experience and level of supervision through telemedicine clinics. Questions 18–20 assessed the influence of experiences with telemedicine clinics on family medicine residents’ career plans and inquired their opinions on the acceptable ratio of telemedicine in residency training. Statistical analysis The data were analyzed using SPSS v. 23.0. Chi-square and Fisher’s exact tests were used to analyze the difference (1) between post-graduation levels (juniors and seniors) and practice-based learning parameters (amount of attending supervision, clinical experience gain, and acceptable percentage of telemedicine practice) and (2) between post-graduation levels and other variables including communication skill gain, preference of telemedicine practice, and effect of telemedicine on future career decision. P < 0.05 was considered indicative of a statistically significant difference. In addition, Spearman’s correlation was used to test the association between post-graduation level and residents’ confidence. A cross-sectional study was conducted with family medicine residents at King Saud University Medical City, Riyadh, Saudi Arabia. The family medicine residency program had 60 residents; all of them were included in the study, with a 100% response rate. Telemedicine clinic visits were introduced to the residents in October 2020. Prior to the clinic appointment, family medicine consultants sorted the booked cases based on the patient’s reason for booking and used their judgment to decide whether the patient required an in-person or telemedicine consultation. The cases included new and follow-up visits. Telemedicine clinics were conducted primarily through telephone rather than video calls. To assess the residents’ experience, a 20-item questionnaire survey was administered between March and April 2022. The questionnaire was anonymous to ensure confidentiality. A validated electronic questionnaire from a similar study conducted in the USA was adapted after obtaining the author’s permission . A few adjustments were made to the questionnaire to match the family medicine program. The adjustments were reviewed and approved by three research experts of family medicine consultants. Questions 1 and 2 related to demographics, including gender and post-graduation year level (residency level). Questions 5, 6, 7, 16, and 17 focused on providing patient-centered care. Question 8 and its subsections assessed residents’ confidence. Common diseases managed in family clinics, including diabetes, dyslipidemia, hypertension, hypothyroidism, osteoarthritis, depression and anxiety, chronic headaches, urinary tract infections, back pain, and congestive heart failure, were chosen. Questions 9–11 evaluated system-based practice and work in interdisciplinary teams. Questions 12–15 focused on practice-based learning and improvement and asked family medicine residents about their clinical experience and level of supervision through telemedicine clinics. Questions 18–20 assessed the influence of experiences with telemedicine clinics on family medicine residents’ career plans and inquired their opinions on the acceptable ratio of telemedicine in residency training. The data were analyzed using SPSS v. 23.0. Chi-square and Fisher’s exact tests were used to analyze the difference (1) between post-graduation levels (juniors and seniors) and practice-based learning parameters (amount of attending supervision, clinical experience gain, and acceptable percentage of telemedicine practice) and (2) between post-graduation levels and other variables including communication skill gain, preference of telemedicine practice, and effect of telemedicine on future career decision. P < 0.05 was considered indicative of a statistically significant difference. In addition, Spearman’s correlation was used to test the association between post-graduation level and residents’ confidence. The questionnaire was distributed to 60 residents (30 juniors and 30 seniors), with a 100% response rate. Of the participants, 48.3% (29) were men and 51.7% (31) were women (Table ). Prior to the COVID-19 pandemic, none of the residents had telemedicine experience. The results indicated that if a patient did not answer the call, 81.7% (49) of the participants attempted to call 2–3 times, while only 18.3% (11) attempted to call 4 times or more. Most (47; 78.3%) of the participants could handle 6–8 telemedicine visits per clinic (Table ). An overall 53.3% (32) of the participants thought that telemedicine may affect their future career decision. However, in-person visit practice was the preferred mode in residency training among most (43; 71.7%) of the participants, and only 10% (6) preferred telemedicine. In addition, 76.7% (46) of the participants accepted the implementation of telemedicine clinics as a part of their training as long as they constituted not more than 25% of the training program (Table )(Fig. ). Compared with in-person visits, most participants reported receiving less clinical experience, less supervision, and less discussion time with the attending supervisor when training in telemedicine clinics. However, most (41; 68.3%) participants gained communication skills through telemedicine (Table ). The participants were mostly confident in managing chronic conditions, such as hypertension, dyslipidemia, diabetes mellitus, and hypothyroidism, through telemedicine. In contrast, most participants were not confident in dealing with back pain through telemedicine. The participants stated that telemedicine does not provide the same quality of care as in-person visits because of an increased number of patients lost to follow-up, significant language barriers, and patients feeling uncomfortable with discussing their complaints through telemedicine (Table )(Figs. and ). Fisher’s exact test and chi-square test were conducted to investigate the association between post-graduation levels (juniors and seniors) and practice-based learning parameters (amount of attending supervision, clinical experience gain, and acceptable percentage of telemedicine practice). There were no statistically significant associations (p = 0.37, p = 0.42, and p = 0.47, respectively; Table ). In addition, the same tests were used to evaluate the relationship between post-graduation levels and other variables, including communication skill gain, preference of telemedicine practice, and effect of telemedicine on their future career decision. Similarly, there were no significant statistical associations (p = 0.78, p = 1.00, and p = 0.61, respectively; Table ). Spearman’s correlation was used to test the association between post-graduation level and resident confidence variables. The results revealed a statistically significant correlation between post-graduation level and treating hypothyroidism and urinary tract infections (p = 0.01, and p = 0.01, respectively; Tables and ). Residency training for all specialties was affected during the COVID-19 pandemic, leading to the use of technology as a part of medical education and training. During the COVID-19 pandemic, various fields of the healthcare system, such as telemedicine, medical education, and patient care, were affected . Our results demonstrated that most of the family medicine residents surveyed in this study preferred in-person visits rather than implementing telemedicine in their training program. In addition, our findings indicated that most participants felt confident in making decisions for managing chronic conditions. Apart from this is because acute conditions such as new onset proctological complaints require in-person visits for further physical examination . Therefore, through our observation, in terms of decision making, we believe that telemedicine should be implemented in residency training regardless of seniority level. Moreover, most residents stated that telemedicine increased the number of patients who did not complete the required labs and imaging studies or were lost to follow-up. In addition, many participants reported experiencing a language barrier caused by telemedicine. These problems can affect the continuity of care and patient trust in their physicians, which are core principles of family medicine practice. Nevertheless, most participants reported gaining communication skills while using telemedicine, which is another core principle of family medicine. Most of our findings were consistent with that of the Lincoln Medical Center study . Most participants reported gaining less clinical experience through telemedicine clinics and preferred in-person visits, and the majority of the participants reported that the inclusion of telemedicine might affect their future career decision. The participants stated that they prefer only 25% or less of their training to include telemedicine. This could be due to the lower amount of supervision and discussion time with the attending supervisor during the clinic. The reduced supervision and discussion time, in turn, could be due to the COVID-19 pandemic and the high number of attempted calls from the residents to the patients, leading to fewer patients answering and decreased time for discussion with the supervisor. The high number of telemedicine appointments per clinic could be another possible explanation. Previous research conducted in other developed countries identified the absence of a policy framework for telemedicine as a key factor that influenced the implementation and sustainability of telemedicine . One study showed that most residents expressed an overall concern regarding their overall preparedness to conduct telemedicine visits and their ability to provide high-quality care to their patients . The increased use of technology in medical education and training necessitates an organized policy and regulations for implementing telemedicine clinics, including requiring the attendant physician to physically supervise trainees and discuss cases with them. This could improve training for use of telemedicine. A study conducted in France reported that most residents acknowledged the growth and expansion of telemedicine, reported not being well trained to use telemedicine, and were aware that training is mandatory to provide high-quality care . Training in using telemedicine could increase the confidence level of residents during clinical practice . The acceptance of telemedicine practice in Saudi Arabia could result in its significant growth . Appropriate training in telemedicine could alleviate disparities in care access in rural and urban areas . Telemedicine plays an important role in digital health growth. Residents are a part of healthcare providers, and proper training during residency programs can help develop telemedicine. Residents need to advance their clinical skills and judgment throughout their training, as this will make them better physicians and allow them to approach and effectively manage patients remotely and provide high-quality care. Implementing telemedicine in residency training and medical education requires clear protocols and paradigms designed and tested by experts that help achieve optimal training and enhance patient care. Our study’s strength lies in being the first study in the Middle East with the largest sample size. However, this study had several limitations. First, it was a single-center study with a small sample involving only the department of family medicine; therefore, it may not represent the residents in other specialties or other hospitals. Second, the study regarded only phone consultations; thus, other modalities of telemedicine should be considered in the future. Third, this study was conducted during the COVID-19 pandemic, and future normalized circumstances should be considered. Finally, the future involvement of telemedicine in residency training requires follow-up studies to assess its effect on such training.
Protocol of an implementation study of a clinician intervention to reduce fear of recurrence in cancer survivors (CIFeR_2 implementation study)	234f7563-1da0-4811-aa3d-72ba3210ebf5	10161179	Internal Medicine[mh]	Improvements in cancer screening, diagnosis and management has resulted in substantial increased survival rates. Long-term survival is common after treatment for early breast cancer, with 5-year survival rates reaching 89% . One of the most prevalent unmet needs of breast cancer survivors is fear of cancer recurrence (FCR) . FCR is defined as ‘fear associated with the possibility the cancer will return or progress in the same or different part of the body’ . FCR is a significant problem affecting 50–70% of cancer survivors across all cancer subtypes which persists over time . A systematic review of FCR in adult cancer survivors found inconsistent relationships between cancer stage or objective markers of recurrence risk, and patients’ perceived level of FCR. However, there is moderate evidence that cancer survivors who reported higher FCR expressed lower healthcare satisfaction . High levels of FCR also affects patients’ quality of life and productivity and increases resource expenditure and health system utilization . Importantly, 30% of patients surveyed report significant unmet need for help with managing FCR . A number of studies evaluating evidence-based psychologist-delivered interventions, including a randomized trial conducted by this research team of ConquerFear, have demonstrated sustained efficacy in reducing FCR in patients with high baseline fear levels. However, these programs are resource intensive and time-consuming, not acceptable to all patients, and not tailored for the vast majority of patients with mild-moderate FCR, whose fears may be more appropriately managed by clinicians (such as oncologists) within the context of routine consultations. Furthermore, brief, targeted oncologist-delivered FCR interventions have the potential to improve patient-clinician communication and rapport and prevent the development of severe FCR. Cross-sectional surveys of cancer survivors indicate that many patients report a desire but reluctance to raise FCR with their doctors for fear of appearing ‘ungrateful’ or damaging the patient-physician relationship by suggesting their treatment may not have been successful . Moreover, > 70% of surveyed doctors indicated discomfort with managing FCR with the majority indicating interest in specific methods and education and training on how to better discuss and manage FCR as part of their routine clinical consultation . Current clinical practice guidelines on the identification and management of FCR by Cancer Australia recommends psychological interventions and involvement of family/carers to help address FCR, but provides no recommendations or guidance on how doctors can address FCR with the patient . A systematic review of doctor and nurse-led interventions for managing FCR revealed that no intervention trials currently exist to address FCR in the context of routine consultations . To address these gaps in evidence-based interventions addressing FCR, the investigators developed the 5-component CIFeR intervention based on current knowledge of existing interventions, results of cross-sectional surveys on FCR, FCR theoretical models and expert input (including psycho-oncologists, clinicians, and consumers). The 5-component CIFeR intervention entails: (1) FCR normalisation and reassurance delivered by the clinician during the consultation (2) Provision of concrete prognostic information (if desired by the patient) (3) A take-home education sheet on red-flag recurrence symptoms (4) Brief advice on strategies to manage worry (5) Referral to psychologist if FCR is severe or the patient requests additional help. CIFeR is delivered at any appropriate follow-up appointment to breast cancer survivors who are 6 months to 5 years after completion of treatment (with exception of hormone therapy) for early-stage breast cancer. CIFeR may be delivered either face-to-face, or via TeleHealth. To determine the usefulness, feasibility and efficacy of CIFeR, we conducted a multicentre, single-arm Phase I/II study involving five oncologists and 61 women with early-stage breast cancer . Patients were surveyed before (T0), one week (T1) and three months (T2) after the intervention on FCR, need for help with FCR and depression/anxiety, and at one week on satisfaction. Oncologists underwent one-hour face-to-face training on the steps and delivery of CIFeR. Overall, 58 women (95%) found CIFeR to be helpful and 59 (98%) would recommend it to others. Women noted that they very much appreciated FCR being addressed by their oncologist and found all components of CIFeR beneficial with 56/58 women (97%) reporting the intervention to be useful and 57/58 (98%) reporting that they would recommend it to other patients. FCR severity, and proportion of women with clinically significant FCR decreased significantly over time. Mean intervention length was 9 min (3–20 min). Average intervention fidelity by the oncologist was 82% (range 67–89%) using audio-recordings of consultations. The intervention was perceived as useful and feasible by oncologists, all of whom have used components of the intervention to help manage FCR in other breast cancer patients. Thus, it was concluded that CIFeR was feasible, acceptable and potentially efficacious. This brief and low-cost intervention may be effective in preventing FCR, as well as reducing its severity and duration in patients who develop FCR. However, it remains to be demonstrated that clinicians more widely will take up CIFeR in routine clinical practice. To guide implementation efforts, we need to understand the barriers and facilitators of implementing CIFeR in routine care. Thus, the C linician I ntervention to Reduce Fear of Recurrence (CIFeR_2) study aims to determine the uptake, adoption and sustainability, and perceived acceptability, feasibility, costs, barriers and facilitators to implementation of CIFeR with early-stage breast cancer patients who are 6-months to 5 years after completion of surgery/chemotherapy/radiotherapy. The resulting data will guide further intervention development, and future large-scale efficacy studies. This sequence of research is recommended by Proctor et al. who position implementation outcomes as preceding both service outcomes and client outcomes, with the latter outcomes being impacted by implementation outcomes. The PARiHS framework (Promoting Action on Research Implementation in Health Services) incorporating strong scientific evidence with a supportive context and implementation facilitation will be used to guide the study. The primary hypothesis of CIFeR_2 is that: > 50% of participating Medical and Radiation oncologists or surgeons will offer CIFeR to at least 1 early stage/curable breast cancer patient by 3 months after receiving training. Secondary hypotheses are that: 2. > 50% of participating Medical and Radiation oncologists or surgeons will offer CIFeR to at least 1 early stage/curable breast cancer patient in the last 3 months when surveyed 6-months after their training on the CIFeR intervention. 3. More than 20% of Medical and Radiation oncologists or surgeons invited to join the CIFeR implementation study will agree to participate. 4. Participating Medical and Radiation oncologists or surgeons will deliver 4/5 of the components of CIFeR (80% fidelity) to at least two of the first three patients to whom they deliver CIFeR. 5. Participating Medical and Radiation oncologists or surgeons will find CIFeR acceptable, appropriate and feasible in routine practice. 6. CIFeR implementation will be low in costs across consultation time, resources, and psychologist referrals. 7. Participating Medical and Radiation oncologists or surgeons will report a range of barriers and facilitators to CIFeR implementation in qualitative interviews at 6 months follow-up. 8. There will be no differences between Medical and Radiation oncologists and surgeons with respect to CIFeR implementation outcomes. 9. Oncologists’ or surgeons’ scores on a scale assessing self-efficacy to manage FCR in patients will increase from baseline to post-training, and to 3 and 6-months post-training follow-up. This multi-site implementation study is being led by the Chris O’Brien Lifehouse, Sydney Australia in collaboration with the Psycho-Oncology Co-operative Research Group based at the University of Sydney, Australia. This project was prospectively registered with the Australian New Zealand Clinical Trials Registry (ACTRN12621001697875). Ethics approval has been obtained from the St Vincents Hospital Research Ethics Committee (2021/ETH10908). Participants Participants will be Medical and Radiation oncologists and Surgeons, including oncology and surgical senior trainees who treat women with early-stage breast cancer. Oncologists or surgeons will be eligible if they are: Currently practising medical and radiation oncologists and breast surgeons or senior trainees and breast surgical trainees (with > 6 months training in clinical oncology at the time of recruitment) who treat women with early-stage breast cancer. Ability to commit to the study requirements and undertake online CIFeR training modules. Participants will be recruited through advertisements posted by breast cancer organisations (e.g. the Medical Oncology Group of Australia and Breast Cancer Trials Group) via email and through newsletters as per those organisations’ procedures; by email from the researchers; and by snow-balling recruitment techniques (participants informing colleagues about the study) and social media (e.g. professional Twitter accounts). Where possible, we will obtain estimates of the number of oncologists approached. Interested oncologists will be provided with the research assistant’s email if they would like to speak to a study staff member to obtain further information, and a link to the online Qualtrics portal where they will be provided with an information sheet and provide written online consent. Procedure Participants will be prompted to complete the online (Qualtrics) baseline questionnaire eliciting demographic and practice details, estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR over the past 3 months, and self-efficacy to manage FCR in patients (Supplementary File 1). Oncologists or surgeons who have not responded to invitations to participate or do not complete the baseline assessment will receive up to two emails and two phone calls from the research assistant to prompt completion spaced out over two weeks. After completing the baseline measure, participants will then be emailed a link to the online CIFeR training, which will be indefinitely available to oncologists or surgeons, allowing them to refresh their familiarity with CiFeR content at their convenience. Upon training completion, participants will click on a link redirecting them to the Qualtrics portal, where they will be asked to complete the post-training self-efficacy measure for managing FCR (Supplementary File 2). Participants will then be asked to use CIFeR with suitable patients for the next six months. An example script for participants to use in their consultations with patients when introducing CIFeR will be provided to participants. They will also be provided with patient hand-outs, online links to CIFeR resources and a 5-point checklist (paper or online version) (Supplementary File 3) which they will be asked to complete after delivering CIFeR to three patients, to assess intervention fidelity. Given CIFeR will be offered to patients as part of routine care (with no patient-reported outcomes), patients will not need to be consented to the study. Oncologists or surgeons will identify patients suitable for the CIFeR intervention by asking each patient as they attend for follow-up “Do you ever worry that your cancer may come back?” All patients who indicate any worry can be offered CIFeR. If the clinician determines during the consultation that the patient is significantly distressed by FCR, they will refer the patient to a psychosocial health professional or other intervention as per usual practice. A research assistant will contact participants 1 month after training completion by phone to prompt them to utilise CIFeR with their patients in follow-up, and to complete the checklist after 3 patients have received CIFeR. At 3 and 6-months post training participants will be emailed a link to an online questionnaire assessing Proctor outcomes, number of patients with whom CIFeR has been used and their self-efficacy in FCR management (Supplementary File 4). Participants who have not completed follow-up measures within 10 days will receive two emails and two phone calls to prompt completion spaced out over two weeks. See Fig. for study schema. Training The CIFeR training has been developed by an expert panel of FCR experts, online education experts, oncologists and consumers, with the aid of a videographer with expertise in creating brief online clinical education courses. The training features didactic material on the prevalence, severity, clinical features and outcomes of FCR, description and modelling of the CIFeR intervention, and evidence supporting its efficacy, captured in short videos of the study team, and videos of clinicians and patients modelling intervention delivery. Training is hosted on the Thinkific platform and will take approximately 15 min to complete. The training is being piloted using think-aloud techniques with 5–6 oncologists not participating in the main study and will be further refined in line with feedback if necessary. As an alternative to the online course, if participants request, the course will also be run as a facilitated one on one educational session with video-conferencing, hosted by a member of the research team. Intervention The CIFeR intervention components are described in Table . Further information is provided in the CIFeR phaseI/II pilot paper. As noted above, CIFeR is delivered by an oncologist or surgeon, during a face-to-face consultation or via TeleHealth if preferred during the COVID-19 pandemic or in rural contexts. Quantitative data Collection Using Proctor’s implementation outcomes and Shepherd’s article which described a conceptual approach to defining and operationalising implementation outcomes, we defined measures of success for the CIFeR_2 study (see Table ). Primary outcome The primary outcome of the CIFeR_2 study is adoption (percentage of participating oncologists or surgeons who report offering the CIFeR intervention to at least one early stage/curable breast cancer patient attending a follow-up appointment 3 months after receiving the CIFeR training). The CIFeR intervention will be defined as adopted if ≥ 50% of oncologists or surgeons deliver the CIFeR intervention to at least one early stage/curable breast cancer patient in that timeframe, whereas the intervention will be deemed not adopted if < 50% of oncologists deliver CIFeR in that timeframe. Secondary outcomes Secondary outcome measures are. Acceptability (percentage of participating oncologists or surgeons who report CIFeR to be acceptable or very acceptable) 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not acceptable, 2 = moderately acceptable, 3 = acceptable and 4 = very acceptable. Appropriateness (percentage of participating oncologists or surgeons who report CIFeR to be appropriate or very appropriate to their patients) 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not appropriate, 2 = slightly appropriate, 3 = appropriate and 4 = very appropriate. Feasibility (proportion of participating oncologists who report CIFeR to be feasible or very feasible in their practice 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not feasible, 2 = slightly feasible, 3 = feasible and 4 = very feasible. Fidelity (proportion of the first three patients receiving CIFeR to whom all 5 components of CIFeR are delivered, as assessed by oncologist-completed checklist self-reported 3 months after the CIFeR training). The checklist comprises 5 items assessing fidelity to each of 5 CIFeR intervention components, with yes/no response options and open-ended questions eliciting reasons for not delivering components if that occurs. Penetration: (percentage of oncologists or surgeons informed about CIFeR who express interest in using CIFeR in their clinical practice and agree to participate in the implementation study). This will be recorded as the difference between the total number of oncologists or surgeons informed about the CIFeR implementation study (recorded by the study research assistant) and how many agree to participate. Sustainability: (proportion of participating oncologists or surgeons who report having used CIFeR with at least one patient within the last 3 months, 6-months after the CIFeR training. Costs (Oncologist or surgeon-estimated mean time taken to deliver CIFeR in minutes; costs of printing CIFeR leaflet (recorded by study staff); Oncologist or surgeon-estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR; CIFeR will be determined to be low cost if the time taken to deliver the intervention is < 10 min, printing costs are < $1 per leaflet, and proportion of patient referrals to psychosocial health professionals does not increase. Barriers and facilitators to implementation generated from semi-structured qualitative interviews with oncologists or surgeons 6 months after the CIFeR training. Self-efficacy: (change in oncologists’ or surgeons’ self-efficacy to manage FCR scores from baseline to post-training, 3 and 6 months after the CIFeR training). This will be measured by the 12-item Self-efficacy Questionnaire (SE-12) , adapted to target self-efficacy in managing FCR in patients, with 4-point response scales. Data on demographics and professional characteristics will be collected at baseline. Participants will also be asked to report estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR, and the number of patients to whom they have delivered CIFeR at baseline and 3 and 6-month assessment points. Semi-structured interview At 6 months post CIFeR training, participants will be contacted to arrange a semi-structured telephone interview at a time convenient to them, to elicit their feedback about barriers and facilitators to using CIFeR in routine clinical practice, conducted by a trained qualitative researcher. Open questions will elicit discussion about the CIFeR training, the CIFeR intervention as a whole, specific components of the intervention, barriers and facilitators to implementing CIFeR, and ideas to improve translation of CIFeR into routine practice (Supplementary File 5). Probing questions will be used to deepen and extend responses. The recorded interviews will be transcribed verbatim. Sample size The sample size was determined using the power-based approach for the primary endpoint, assuming that the intervention may be adopted if H1: p > 50% (greater than 50% of consenting clinicians offer CIFeR to at least 1 patient at 3-months follow-up), and that H0: p < 30%, a level below which the intervention will not be regarded as adopted. If 30% is assumed for the participation rate under the null and 50% under the alternative, then based on a one-sided alpha = 5% and a power of 80%, the estimated sample size is n = 39. The intervention would be regarded as adopted, if at least 17 out of 39 clinicians offer CIFeR to at least 1 early stage/curable breast cancer patient by 3 months. Assuming a drop-out rate of 25% then 50 clinicians will need to be recruited to meet the primary endpoint of the study. Data handling Study data will be recorded on the password protected Qualtrics server. All required data entry fields will be completed. All completed questionnaires, audio-recordings and transcripts will be managed centrally at the University of Sydney. Electronic data will be collected securely by the Qualtrics database, and only the chief investigator or site principal investigator will have access to the study data. All information will be stored securely for seven years as per NHMRC and will only be available to staff directly involved with the study. Non-identified data will be analysed by the core research team (chief investigator, principle investigator and sub-investigators), with the possible assistance of additional research staff or research students. A Clinical Study Report will be issued which may form the basis of a manuscript intended for publication. All data collected for, used in, or generated by this project will be disposed by secure methods after seven years from the completion of the study. Paper files will be shredded and computer files will be deleted. Any major changes to the protocol will be updated to participants and relevant ethics committee. Statistical analyses Measures of Proctor outcomes (e.g. intervention adoption, acceptability) will be reported using descriptive statistics including proportions, means and standard deviations. Baseline demographics will be summarized in table format. Repeated measure t-tests will be used to examine changes in oncologist self-efficacy scores pre- and post-intervention. Proctor outcomes will be summarised using descriptive statistics. Exploratory predictors of higher adoption rates will be examined using linear models. Kolmogorov-Smirnov and Shapiro-Wilk tests of normality (tests statistic, degrees of freedom, p-value) will be performed. As there are no pre-specified instructions available for handling missing data for the SE-12, the averages for the remaining items for the scale in question will be calculated with the adjusted denominator. Missing data from the questionnaires will be descriptively reported and all available data will be included for analysis. Qualitative data will be analysed using Framework analysis . Line-by-line coding will be conducted on three transcripts by the research team to develop the preliminary coding framework, which will be iteratively refined following review of subsequent transcripts. Over-arching themes and sub-themes will be developed to summarize the data. Differences in researcher interpretation of the data will be resolved through discussion. Themes arising from medical and radiation oncologists and surgeons will be compared. We will use the consolidated criteria for reporting qualitative research (COREQ) to guide reporting . Recruitment timeline Recruitment has started as of March 2022 and recruitment is in process of finishing with final follow-up to be sent in May 2023. Participants will be Medical and Radiation oncologists and Surgeons, including oncology and surgical senior trainees who treat women with early-stage breast cancer. Oncologists or surgeons will be eligible if they are: Currently practising medical and radiation oncologists and breast surgeons or senior trainees and breast surgical trainees (with > 6 months training in clinical oncology at the time of recruitment) who treat women with early-stage breast cancer. Ability to commit to the study requirements and undertake online CIFeR training modules. Participants will be recruited through advertisements posted by breast cancer organisations (e.g. the Medical Oncology Group of Australia and Breast Cancer Trials Group) via email and through newsletters as per those organisations’ procedures; by email from the researchers; and by snow-balling recruitment techniques (participants informing colleagues about the study) and social media (e.g. professional Twitter accounts). Where possible, we will obtain estimates of the number of oncologists approached. Interested oncologists will be provided with the research assistant’s email if they would like to speak to a study staff member to obtain further information, and a link to the online Qualtrics portal where they will be provided with an information sheet and provide written online consent. Participants will be prompted to complete the online (Qualtrics) baseline questionnaire eliciting demographic and practice details, estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR over the past 3 months, and self-efficacy to manage FCR in patients (Supplementary File 1). Oncologists or surgeons who have not responded to invitations to participate or do not complete the baseline assessment will receive up to two emails and two phone calls from the research assistant to prompt completion spaced out over two weeks. After completing the baseline measure, participants will then be emailed a link to the online CIFeR training, which will be indefinitely available to oncologists or surgeons, allowing them to refresh their familiarity with CiFeR content at their convenience. Upon training completion, participants will click on a link redirecting them to the Qualtrics portal, where they will be asked to complete the post-training self-efficacy measure for managing FCR (Supplementary File 2). Participants will then be asked to use CIFeR with suitable patients for the next six months. An example script for participants to use in their consultations with patients when introducing CIFeR will be provided to participants. They will also be provided with patient hand-outs, online links to CIFeR resources and a 5-point checklist (paper or online version) (Supplementary File 3) which they will be asked to complete after delivering CIFeR to three patients, to assess intervention fidelity. Given CIFeR will be offered to patients as part of routine care (with no patient-reported outcomes), patients will not need to be consented to the study. Oncologists or surgeons will identify patients suitable for the CIFeR intervention by asking each patient as they attend for follow-up “Do you ever worry that your cancer may come back?” All patients who indicate any worry can be offered CIFeR. If the clinician determines during the consultation that the patient is significantly distressed by FCR, they will refer the patient to a psychosocial health professional or other intervention as per usual practice. A research assistant will contact participants 1 month after training completion by phone to prompt them to utilise CIFeR with their patients in follow-up, and to complete the checklist after 3 patients have received CIFeR. At 3 and 6-months post training participants will be emailed a link to an online questionnaire assessing Proctor outcomes, number of patients with whom CIFeR has been used and their self-efficacy in FCR management (Supplementary File 4). Participants who have not completed follow-up measures within 10 days will receive two emails and two phone calls to prompt completion spaced out over two weeks. See Fig. for study schema. The CIFeR training has been developed by an expert panel of FCR experts, online education experts, oncologists and consumers, with the aid of a videographer with expertise in creating brief online clinical education courses. The training features didactic material on the prevalence, severity, clinical features and outcomes of FCR, description and modelling of the CIFeR intervention, and evidence supporting its efficacy, captured in short videos of the study team, and videos of clinicians and patients modelling intervention delivery. Training is hosted on the Thinkific platform and will take approximately 15 min to complete. The training is being piloted using think-aloud techniques with 5–6 oncologists not participating in the main study and will be further refined in line with feedback if necessary. As an alternative to the online course, if participants request, the course will also be run as a facilitated one on one educational session with video-conferencing, hosted by a member of the research team. The CIFeR intervention components are described in Table . Further information is provided in the CIFeR phaseI/II pilot paper. As noted above, CIFeR is delivered by an oncologist or surgeon, during a face-to-face consultation or via TeleHealth if preferred during the COVID-19 pandemic or in rural contexts. Using Proctor’s implementation outcomes and Shepherd’s article which described a conceptual approach to defining and operationalising implementation outcomes, we defined measures of success for the CIFeR_2 study (see Table ). Primary outcome The primary outcome of the CIFeR_2 study is adoption (percentage of participating oncologists or surgeons who report offering the CIFeR intervention to at least one early stage/curable breast cancer patient attending a follow-up appointment 3 months after receiving the CIFeR training). The CIFeR intervention will be defined as adopted if ≥ 50% of oncologists or surgeons deliver the CIFeR intervention to at least one early stage/curable breast cancer patient in that timeframe, whereas the intervention will be deemed not adopted if < 50% of oncologists deliver CIFeR in that timeframe. Secondary outcomes Secondary outcome measures are. Acceptability (percentage of participating oncologists or surgeons who report CIFeR to be acceptable or very acceptable) 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not acceptable, 2 = moderately acceptable, 3 = acceptable and 4 = very acceptable. Appropriateness (percentage of participating oncologists or surgeons who report CIFeR to be appropriate or very appropriate to their patients) 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not appropriate, 2 = slightly appropriate, 3 = appropriate and 4 = very appropriate. Feasibility (proportion of participating oncologists who report CIFeR to be feasible or very feasible in their practice 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not feasible, 2 = slightly feasible, 3 = feasible and 4 = very feasible. Fidelity (proportion of the first three patients receiving CIFeR to whom all 5 components of CIFeR are delivered, as assessed by oncologist-completed checklist self-reported 3 months after the CIFeR training). The checklist comprises 5 items assessing fidelity to each of 5 CIFeR intervention components, with yes/no response options and open-ended questions eliciting reasons for not delivering components if that occurs. Penetration: (percentage of oncologists or surgeons informed about CIFeR who express interest in using CIFeR in their clinical practice and agree to participate in the implementation study). This will be recorded as the difference between the total number of oncologists or surgeons informed about the CIFeR implementation study (recorded by the study research assistant) and how many agree to participate. Sustainability: (proportion of participating oncologists or surgeons who report having used CIFeR with at least one patient within the last 3 months, 6-months after the CIFeR training. Costs (Oncologist or surgeon-estimated mean time taken to deliver CIFeR in minutes; costs of printing CIFeR leaflet (recorded by study staff); Oncologist or surgeon-estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR; CIFeR will be determined to be low cost if the time taken to deliver the intervention is < 10 min, printing costs are < $1 per leaflet, and proportion of patient referrals to psychosocial health professionals does not increase. Barriers and facilitators to implementation generated from semi-structured qualitative interviews with oncologists or surgeons 6 months after the CIFeR training. Self-efficacy: (change in oncologists’ or surgeons’ self-efficacy to manage FCR scores from baseline to post-training, 3 and 6 months after the CIFeR training). This will be measured by the 12-item Self-efficacy Questionnaire (SE-12) , adapted to target self-efficacy in managing FCR in patients, with 4-point response scales. Data on demographics and professional characteristics will be collected at baseline. Participants will also be asked to report estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR, and the number of patients to whom they have delivered CIFeR at baseline and 3 and 6-month assessment points. The primary outcome of the CIFeR_2 study is adoption (percentage of participating oncologists or surgeons who report offering the CIFeR intervention to at least one early stage/curable breast cancer patient attending a follow-up appointment 3 months after receiving the CIFeR training). The CIFeR intervention will be defined as adopted if ≥ 50% of oncologists or surgeons deliver the CIFeR intervention to at least one early stage/curable breast cancer patient in that timeframe, whereas the intervention will be deemed not adopted if < 50% of oncologists deliver CIFeR in that timeframe. Secondary outcome measures are. Acceptability (percentage of participating oncologists or surgeons who report CIFeR to be acceptable or very acceptable) 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not acceptable, 2 = moderately acceptable, 3 = acceptable and 4 = very acceptable. Appropriateness (percentage of participating oncologists or surgeons who report CIFeR to be appropriate or very appropriate to their patients) 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not appropriate, 2 = slightly appropriate, 3 = appropriate and 4 = very appropriate. Feasibility (proportion of participating oncologists who report CIFeR to be feasible or very feasible in their practice 3 and 6 months after the CIFeR training). This is measured on a 4-point Likert scale where 1 = not feasible, 2 = slightly feasible, 3 = feasible and 4 = very feasible. Fidelity (proportion of the first three patients receiving CIFeR to whom all 5 components of CIFeR are delivered, as assessed by oncologist-completed checklist self-reported 3 months after the CIFeR training). The checklist comprises 5 items assessing fidelity to each of 5 CIFeR intervention components, with yes/no response options and open-ended questions eliciting reasons for not delivering components if that occurs. Penetration: (percentage of oncologists or surgeons informed about CIFeR who express interest in using CIFeR in their clinical practice and agree to participate in the implementation study). This will be recorded as the difference between the total number of oncologists or surgeons informed about the CIFeR implementation study (recorded by the study research assistant) and how many agree to participate. Sustainability: (proportion of participating oncologists or surgeons who report having used CIFeR with at least one patient within the last 3 months, 6-months after the CIFeR training. Costs (Oncologist or surgeon-estimated mean time taken to deliver CIFeR in minutes; costs of printing CIFeR leaflet (recorded by study staff); Oncologist or surgeon-estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR; CIFeR will be determined to be low cost if the time taken to deliver the intervention is < 10 min, printing costs are < $1 per leaflet, and proportion of patient referrals to psychosocial health professionals does not increase. Barriers and facilitators to implementation generated from semi-structured qualitative interviews with oncologists or surgeons 6 months after the CIFeR training. Self-efficacy: (change in oncologists’ or surgeons’ self-efficacy to manage FCR scores from baseline to post-training, 3 and 6 months after the CIFeR training). This will be measured by the 12-item Self-efficacy Questionnaire (SE-12) , adapted to target self-efficacy in managing FCR in patients, with 4-point response scales. Data on demographics and professional characteristics will be collected at baseline. Participants will also be asked to report estimated proportion of patients referred to psychologists or other psychosocial health professionals for help with FCR, and the number of patients to whom they have delivered CIFeR at baseline and 3 and 6-month assessment points. At 6 months post CIFeR training, participants will be contacted to arrange a semi-structured telephone interview at a time convenient to them, to elicit their feedback about barriers and facilitators to using CIFeR in routine clinical practice, conducted by a trained qualitative researcher. Open questions will elicit discussion about the CIFeR training, the CIFeR intervention as a whole, specific components of the intervention, barriers and facilitators to implementing CIFeR, and ideas to improve translation of CIFeR into routine practice (Supplementary File 5). Probing questions will be used to deepen and extend responses. The recorded interviews will be transcribed verbatim. The sample size was determined using the power-based approach for the primary endpoint, assuming that the intervention may be adopted if H1: p > 50% (greater than 50% of consenting clinicians offer CIFeR to at least 1 patient at 3-months follow-up), and that H0: p < 30%, a level below which the intervention will not be regarded as adopted. If 30% is assumed for the participation rate under the null and 50% under the alternative, then based on a one-sided alpha = 5% and a power of 80%, the estimated sample size is n = 39. The intervention would be regarded as adopted, if at least 17 out of 39 clinicians offer CIFeR to at least 1 early stage/curable breast cancer patient by 3 months. Assuming a drop-out rate of 25% then 50 clinicians will need to be recruited to meet the primary endpoint of the study. Study data will be recorded on the password protected Qualtrics server. All required data entry fields will be completed. All completed questionnaires, audio-recordings and transcripts will be managed centrally at the University of Sydney. Electronic data will be collected securely by the Qualtrics database, and only the chief investigator or site principal investigator will have access to the study data. All information will be stored securely for seven years as per NHMRC and will only be available to staff directly involved with the study. Non-identified data will be analysed by the core research team (chief investigator, principle investigator and sub-investigators), with the possible assistance of additional research staff or research students. A Clinical Study Report will be issued which may form the basis of a manuscript intended for publication. All data collected for, used in, or generated by this project will be disposed by secure methods after seven years from the completion of the study. Paper files will be shredded and computer files will be deleted. Any major changes to the protocol will be updated to participants and relevant ethics committee. Measures of Proctor outcomes (e.g. intervention adoption, acceptability) will be reported using descriptive statistics including proportions, means and standard deviations. Baseline demographics will be summarized in table format. Repeated measure t-tests will be used to examine changes in oncologist self-efficacy scores pre- and post-intervention. Proctor outcomes will be summarised using descriptive statistics. Exploratory predictors of higher adoption rates will be examined using linear models. Kolmogorov-Smirnov and Shapiro-Wilk tests of normality (tests statistic, degrees of freedom, p-value) will be performed. As there are no pre-specified instructions available for handling missing data for the SE-12, the averages for the remaining items for the scale in question will be calculated with the adjusted denominator. Missing data from the questionnaires will be descriptively reported and all available data will be included for analysis. Qualitative data will be analysed using Framework analysis . Line-by-line coding will be conducted on three transcripts by the research team to develop the preliminary coding framework, which will be iteratively refined following review of subsequent transcripts. Over-arching themes and sub-themes will be developed to summarize the data. Differences in researcher interpretation of the data will be resolved through discussion. Themes arising from medical and radiation oncologists and surgeons will be compared. We will use the consolidated criteria for reporting qualitative research (COREQ) to guide reporting . Recruitment has started as of March 2022 and recruitment is in process of finishing with final follow-up to be sent in May 2023. Theoretical significance Given there are currently no clinician-delivered interventions to address FCR, the CIFeR_2 implementation study advances the field by representing and solidifying the evidence for the beneficial and appropriate use of a clinician-lead educational intervention for patient FCR within the context of follow-up clinics. Additionally, the current study will provide a guide of implementation efforts, as well as provide a greater understanding of the barriers and facilitators of implementing CIFeR in routine care. Implementation studies are increasingly being used to identify and address barriers early in implementation efforts, to ensure successful integration of evidence into practice . Clinical significance Successful completion of the CIFeR_2 Study will provide proof-of-principal that doctors can address worries regarding FCR with their patients, and that CIFeR can be feasibly introduced into routine care. CIFeR_2 addresses psycho-oncology workforce shortages through increased training of oncologists and oncology surgeons to deliver care to patients with mild-moderate FCR. Providing a brief intervention that incorporates self-management has the potential to decrease health service utilisation by patients with untreated FCR. If this project is successful, we will have a user-tested online training module that can be delivered to oncologists and oncology surgeons across Australia. We will also have data on the use and perceived utility of this training module that can guide further refinement of the training. We will have a rich set of quantitative and qualitative data on the factors required for success in implementing a clinician-delivered intervention for FCR (CIFeR). These data will allow us to further refine the CIFeR intervention and the system, clinician and patient focused strategies that will optimise the likelihood that it will be effective and implemented in routine care in the long-term. Finally, the five key components of the intervention are tumour site agnostic and thus the CIFeR intervention could be readily adapted to other tumour streams where FCR is a common and problematic phenomenon (E.g., childhood haematological malignancies, adolescent sarcomas, testicular cancers and ovarian cancers). Future research is needed to explore if CIFeR can be as effectively used by diverse clinical and allied health specialists after adequate training. Given there are currently no clinician-delivered interventions to address FCR, the CIFeR_2 implementation study advances the field by representing and solidifying the evidence for the beneficial and appropriate use of a clinician-lead educational intervention for patient FCR within the context of follow-up clinics. Additionally, the current study will provide a guide of implementation efforts, as well as provide a greater understanding of the barriers and facilitators of implementing CIFeR in routine care. Implementation studies are increasingly being used to identify and address barriers early in implementation efforts, to ensure successful integration of evidence into practice . Successful completion of the CIFeR_2 Study will provide proof-of-principal that doctors can address worries regarding FCR with their patients, and that CIFeR can be feasibly introduced into routine care. CIFeR_2 addresses psycho-oncology workforce shortages through increased training of oncologists and oncology surgeons to deliver care to patients with mild-moderate FCR. Providing a brief intervention that incorporates self-management has the potential to decrease health service utilisation by patients with untreated FCR. If this project is successful, we will have a user-tested online training module that can be delivered to oncologists and oncology surgeons across Australia. We will also have data on the use and perceived utility of this training module that can guide further refinement of the training. We will have a rich set of quantitative and qualitative data on the factors required for success in implementing a clinician-delivered intervention for FCR (CIFeR). These data will allow us to further refine the CIFeR intervention and the system, clinician and patient focused strategies that will optimise the likelihood that it will be effective and implemented in routine care in the long-term. Finally, the five key components of the intervention are tumour site agnostic and thus the CIFeR intervention could be readily adapted to other tumour streams where FCR is a common and problematic phenomenon (E.g., childhood haematological malignancies, adolescent sarcomas, testicular cancers and ovarian cancers). Future research is needed to explore if CIFeR can be as effectively used by diverse clinical and allied health specialists after adequate training. FCR continues to impact a large proportion of cancer survivors, and increasingly stepped care interventions are required to address this issue in patients depending on their level of worry. The CIFeR intervention is a brief and low-cost intervention that has been shown to be feasible, acceptable and potentially effective in preventing FCR, as well as reducing its severity and duration in patients who develop FCR. This research will explore whether oncologists and oncology surgeons more widely will take up CIFeR after formal education and training on how to manage FCR in routine clinical practice, providing valuable information about the practicalities of implementing this beneficial intervention into routine clinical care and telehealth. Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5
Panorama de la ectasia coronaria en el Instituto Nacional de Cardiología Ignacio Chávez: un estudio transversal	d5330fd5-688a-4300-bb48-7376a9d9ca90	10161814	Internal Medicine[mh]	La ectasia coronaria (EC) es una remodelación patológica con una prevalencia mundial baja. Se define como una dilatación difusa mayor a 1.5 veces el diámetro de los segmentos adyacentes de esta o diferentes arterias coronarias, se encuentra como hallazgo en un 0.3-5.3% en las coronariografías, tiene una prevalencia variable de 0.5-5% por año. Es secundaria a una enfermedad como la aterosclerosis, aorta bivalva, enfermedad de Kawasaki, arteritis de Takayasu, síndrome de Marfan o síndrome de Ehlers-Danlos, siendo más frecuente en hombres que en mujeres con una proporción 3:1 - . La EC consiste en un remodelado excéntrico excesivo de la pared arterial coronaria causado por adelgazamiento y debilidad de la capa media debido a una degradación enzimática incrementada de la matriz extracelular por metaloproteasas. Esta modificación estructural de la pared arterial coronaria puede causar alteración en el flujo y la perfusión coronaria, mayor activación y agregación plaquetaria y desencadenar isquemia miocárdica aguda y crónica con sus consecuencias , . La EC se diagnostica con mayor frecuencia de manera incidental con una angiografía coronaria , se utiliza la clasificación de Markis para categorizar la gravedad de la enfermedad según el grado de afectación de las arterias coronarias: tipo 1, ectasia difusa de 2-3 arterias; tipo 2, ectasia difusa en una arteria y localizado en otra; tipo 3, ectasia difusa arterial única, y tipo 4, ectasia localizada o segmentaria de una arteria , . Aún no se ha determinado la terapia más adecuada para prevenir el proceso trombogénico que puede acompañar a la EC, pero se ha observado que la terapia antiplaquetaria dual con anticoagulante (APS + ACO) ha sido eficaz para prevenir desenlaces trombóticos subsecuentes, mientras que la triple terapia (APD + ACO) reduce más eventos isquémicos, pero se asocia a un mayor riesgo de complicaciones hemorrágicas a largo plazo . El objetivo del estudio es documentar las características clínicas, angiográficas y el tratamiento farmacológico que recibieron los pacientes con EC incluidos en el estudio «Panorama de la ectasia coronaria en el Instituto Nacional de Cardiología Ignacio Chávez: un estudio transversal». Se desarrolló un estudio de tipo transversal con diseño no experimental descriptivo, con un muestreo por conveniencia no probabilístico. Se reclutaron 69 pacientes que se atendieron en el tercer piso de hospitalización de adultos y la unidad coronaria del instituto, entre el periodo de octubre del 2019 a octubre del 2021. Se seleccionaron aquellos que cumplían con los criterios de inclusión: edad entre 30-90 años, antecedentes de hipertensión arterial sistémica, diabetes mellitus tipo 2, tabaquismo, obesidad, dislipidemia o cardiopatía isquémica, cuadro clínico inicial de angina típico, atípico o no anginoso, estudio de coronariografía con reporte de EC en al menos un vaso arterial coronario, de acuerdo con la clasificación de Markis, así como el tratamiento antitrombótico al egreso. Se excluyeron los pacientes que requerían procedimiento quirúrgico, con alguna valvulopatía, arritmias, hipertensión pulmonar y que no cumplían con algún parámetro antes comentado y aquellos pacientes que no se concluyeron con un egreso hospitalario. Variables de estudio Las variables documentadas en el presente estudio se definieron conceptualmente de la siguiente manera: la EC es una dilatación difusa mayor a 1.5 veces el diámetro de los segmentos adyacentes de esta o diferentes arterias , ; hipertensión si la presión arterial era igual o mayor a 140/90 mm Hg ; dislipidemia mixta si el colesterol total era igual o mayor a 200 mg/dl y los triglicéridos iguales o mayores a 150 mg/dl ; respecto al tabaquismo, fue considerado para aquellos pacientes que habiendo sido fumadores se hubiesen mantenido en abstinencia al menos por los últimos seis meses, fumador activo que hubiese fumado por lo menos un cigarrillo en los últimos seis meses y fumador pasivo aquel que no fuma, pero que respira el humo de tabaco ajeno o humo de tabaco ambiental ; la diabetes (DM) se conceptualizó como niveles de glucosa en sangre en ayuno iguales o mayores a 126 mg/dl ; obesidad como un índice de masa corporal (IMC) igual o mayor a 30 y un perímetro abdominal igual o mayor a 80 cm mujeres e igual o mayor a 90 cm en hombres . Cardiopatía isquémica se define como aquellos pacientes que hayan padecido un episodio de angina o infarto previo con o sin elevación del segmento ST ; el síndrome coronario agudo (SCA) es un dolor opresivo persistente, irradiado, con síntomas asociados y de duración > 20 minutos . La clasificación Markis se refiere al grado de afectación de las arterias coronarias: tipo 1, ectasia difusa de 2-3 arterias; tipo 2, ectasia difusa en una arteria y localizado en otro; tipo 3, ectasia difusa arterial única, y tipo 4, ectasia localizada o segmentaria de una arteria , . Análisis de datos Considerando el objetivo de esta investigación y las particularidades metodológicas del tipo de estudio que se desarrolló (transversal), para el análisis de los datos se hizo uso de estadística descriptiva, que permitió documentar los porcentajes y frecuencias de los datos clínicos. Las variables documentadas en el presente estudio se definieron conceptualmente de la siguiente manera: la EC es una dilatación difusa mayor a 1.5 veces el diámetro de los segmentos adyacentes de esta o diferentes arterias , ; hipertensión si la presión arterial era igual o mayor a 140/90 mm Hg ; dislipidemia mixta si el colesterol total era igual o mayor a 200 mg/dl y los triglicéridos iguales o mayores a 150 mg/dl ; respecto al tabaquismo, fue considerado para aquellos pacientes que habiendo sido fumadores se hubiesen mantenido en abstinencia al menos por los últimos seis meses, fumador activo que hubiese fumado por lo menos un cigarrillo en los últimos seis meses y fumador pasivo aquel que no fuma, pero que respira el humo de tabaco ajeno o humo de tabaco ambiental ; la diabetes (DM) se conceptualizó como niveles de glucosa en sangre en ayuno iguales o mayores a 126 mg/dl ; obesidad como un índice de masa corporal (IMC) igual o mayor a 30 y un perímetro abdominal igual o mayor a 80 cm mujeres e igual o mayor a 90 cm en hombres . Cardiopatía isquémica se define como aquellos pacientes que hayan padecido un episodio de angina o infarto previo con o sin elevación del segmento ST ; el síndrome coronario agudo (SCA) es un dolor opresivo persistente, irradiado, con síntomas asociados y de duración > 20 minutos . La clasificación Markis se refiere al grado de afectación de las arterias coronarias: tipo 1, ectasia difusa de 2-3 arterias; tipo 2, ectasia difusa en una arteria y localizado en otro; tipo 3, ectasia difusa arterial única, y tipo 4, ectasia localizada o segmentaria de una arteria , . Considerando el objetivo de esta investigación y las particularidades metodológicas del tipo de estudio que se desarrolló (transversal), para el análisis de los datos se hizo uso de estadística descriptiva, que permitió documentar los porcentajes y frecuencias de los datos clínicos. Características clínicas Durante el periodo de dos años se analizó un total de 69 pacientes del INC con diagnóstico de EC, de los cuales el 84.4% fueron hombres y el 11.6% mujeres, con una media de edad de 56 ± 11 años. Dentro de los factores de riesgo identificados para hombres destacan el tabaquismo, presente en 36 (59%), seguido de hipertensión arterial sistémica (HAS) en 23 (37.7%) y DM en 23 (37.7%). Para el caso de las mujeres tabaquismo en 4 (50%), dislipidemia en 4 (50%) y HAS en 3 (37.5%) fueron los antecedentes clínicos más frecuentes . Evidenciando como factor de riesgo la aterosclerosis y alteraciones en la microvasculatura. Al momento del diagnóstico, el 100% de los pacientes fueron sintomáticos, manifestándose como síntoma cardinal el dolor precordial en 64 (92.8%), con una irradiación característica hacía brazo izquierdo en 15 (21.7%), cuello en 11 (15.9%) y en 21 (30.4%) fue localizado en el precordio, con una intensidad en escala analógica visual (EVA) de entre 6-10 en 66 (91.3%), de carácter opresivo en 57 (82.6%). Los síntomas acompañantes fueron diaforesis en 46 (66.7%), náuseas/vómito en 32 (46.4%) y disnea en 31 (44.9%). En el electrocardiograma 4 (34.8%) presentaron infarto agudo de miocardio sin elevación del segmento ST (IAMSEST) y 45 (65.2%) presentaron infarto agudo de miocardio con elevación del segmento ST (IAMCEST), siendo la localización más frecuente la cara inferior, en 18 (40%), seguido de la cara anterior en 15 (33.33%), posterior en 10 (22.22%) y otra localización en 2 (4.44%) . Como estudio complementario se realizó un ecocardiograma transtorácico a cada paciente para valorar la fracción de eyección del ventrículo izquierdo (FEVI), identificando que 40 (57.97%) tenía una FEVI normal (> 50%), 17 (24.63%) FEVI reducida (< 40%) y 12 (17.39%) una FEVI ligeramente reducida (40-50%); obteniendo una media de FEVI del 47 ± 9.72%. Características angiográficas Durante su hospitalización a los 69 pacientes se les realizó una coronariografía diagnóstica, evidenciando que la arteria más afectada es la coronaria derecha (CD) 48 (69.6%), seguida de la coronaria circunfleja (Cx) 39 (56.5%),arteria descendente anterior (DA) en 36 (52.2%) y por último el tronco coronario (TC) en 10 (14.5%) ; de acuerdo con la clasificación, el 35% presentó un Markis tipo 3, seguido del Markis tipo 4 en un 32% . Dentro del manejo farmacológico durante el cateterismo, se empleó heparina no fraccionada intraarterial en 67 (97.1%), verapamilo en 63 (91.3%), nitroglicerina en 33 (47.8%) debido a la presencia de vasoespasmo y tirofibán en 19 (27.5%) por el comportamiento trombótico. Solo 25 (36.2%) de los pacientes requirieron stent coronario, de los cuales 17 (24.6%) fueron de tipo everolimús y 7 (10.1%) stent metálico. De estos solo en 13 (52%) la arteria ectasia fue la causa del evento isquémico. Terapia antitrombótica Al egreso hospitalario de cada paciente se observó que en el INC la terapia APD+ACO es la más utilizada, en 40 (58%) por un mes, seguida de la terapia APS + ACO con 29 (42%). Predominó el uso de ACO en 33 (47.8%) con respecto a los nuevos anticoagulantes (NACO) en 25 (36.2%) de los casos, mientras que 11 (16%) no recibieron terapia anticoagulante, siendo acenocumarol en 16 (23.2%) el más empleado en la mayoría de los pacientes. Durante el periodo de dos años se analizó un total de 69 pacientes del INC con diagnóstico de EC, de los cuales el 84.4% fueron hombres y el 11.6% mujeres, con una media de edad de 56 ± 11 años. Dentro de los factores de riesgo identificados para hombres destacan el tabaquismo, presente en 36 (59%), seguido de hipertensión arterial sistémica (HAS) en 23 (37.7%) y DM en 23 (37.7%). Para el caso de las mujeres tabaquismo en 4 (50%), dislipidemia en 4 (50%) y HAS en 3 (37.5%) fueron los antecedentes clínicos más frecuentes . Evidenciando como factor de riesgo la aterosclerosis y alteraciones en la microvasculatura. Al momento del diagnóstico, el 100% de los pacientes fueron sintomáticos, manifestándose como síntoma cardinal el dolor precordial en 64 (92.8%), con una irradiación característica hacía brazo izquierdo en 15 (21.7%), cuello en 11 (15.9%) y en 21 (30.4%) fue localizado en el precordio, con una intensidad en escala analógica visual (EVA) de entre 6-10 en 66 (91.3%), de carácter opresivo en 57 (82.6%). Los síntomas acompañantes fueron diaforesis en 46 (66.7%), náuseas/vómito en 32 (46.4%) y disnea en 31 (44.9%). En el electrocardiograma 4 (34.8%) presentaron infarto agudo de miocardio sin elevación del segmento ST (IAMSEST) y 45 (65.2%) presentaron infarto agudo de miocardio con elevación del segmento ST (IAMCEST), siendo la localización más frecuente la cara inferior, en 18 (40%), seguido de la cara anterior en 15 (33.33%), posterior en 10 (22.22%) y otra localización en 2 (4.44%) . Como estudio complementario se realizó un ecocardiograma transtorácico a cada paciente para valorar la fracción de eyección del ventrículo izquierdo (FEVI), identificando que 40 (57.97%) tenía una FEVI normal (> 50%), 17 (24.63%) FEVI reducida (< 40%) y 12 (17.39%) una FEVI ligeramente reducida (40-50%); obteniendo una media de FEVI del 47 ± 9.72%. Durante su hospitalización a los 69 pacientes se les realizó una coronariografía diagnóstica, evidenciando que la arteria más afectada es la coronaria derecha (CD) 48 (69.6%), seguida de la coronaria circunfleja (Cx) 39 (56.5%),arteria descendente anterior (DA) en 36 (52.2%) y por último el tronco coronario (TC) en 10 (14.5%) ; de acuerdo con la clasificación, el 35% presentó un Markis tipo 3, seguido del Markis tipo 4 en un 32% . Dentro del manejo farmacológico durante el cateterismo, se empleó heparina no fraccionada intraarterial en 67 (97.1%), verapamilo en 63 (91.3%), nitroglicerina en 33 (47.8%) debido a la presencia de vasoespasmo y tirofibán en 19 (27.5%) por el comportamiento trombótico. Solo 25 (36.2%) de los pacientes requirieron stent coronario, de los cuales 17 (24.6%) fueron de tipo everolimús y 7 (10.1%) stent metálico. De estos solo en 13 (52%) la arteria ectasia fue la causa del evento isquémico. Al egreso hospitalario de cada paciente se observó que en el INC la terapia APD+ACO es la más utilizada, en 40 (58%) por un mes, seguida de la terapia APS + ACO con 29 (42%). Predominó el uso de ACO en 33 (47.8%) con respecto a los nuevos anticoagulantes (NACO) en 25 (36.2%) de los casos, mientras que 11 (16%) no recibieron terapia anticoagulante, siendo acenocumarol en 16 (23.2%) el más empleado en la mayoría de los pacientes. La EC es una entidad poco descrita, algunas veces agrupada junto a la patología aneurismática de las arterias coronarias, lo que suele confundir el diagnóstico, además de un espectro clínico variado que va desde una presentación aislada, como hallazgo incidental, o como consecuencia de un SCA , ; sin embargo la etiología suele ser de naturaleza diversa, que incluye enfermedades inflamatorias (10-20%), autoinmunes, infecciosas (< 5%), congénitas (20-30%) y enfermedad aterosclerótica (50-60%), esta última es la causa más común en la población adulta . Este artículo permitió demostrar la importancia que tiene el diagnóstico de EC en un centro de concentración de casos como el INC y conocer mejor las características demográficas, clínicas y angiográficas de dicha patología. La población en riesgo son adultos jóvenes con antecedente de tabaquismo, el cual ocasiona un estado procoagulante y protrombótico en el paciente, que aunado a la sobreexpresión de metaloproteinasas, la disminución de los inhibidores de estas y el remodelado arterial excéntrico de la íntima de los vasos coronarios, causan su dilatación , . Debido a microembolia generada hacia los segmentos distales u oclusión trombótica del vaso afectado, causa síntomas comunes de infarto como el dolor precordial de características opresivo (síntoma cardinal de la patología) con irradiación a brazo izquierdo asociada en la mayoría de los casos a disnea, náuseas y vómito, como se observó en los pacientes que ingresaron al INC, por la obstrucción parcial o total de la luz del vaso, disminuyendo el aporte de oxígeno al miocardio. Con ayuda del electrocardiograma se observó que el infarto más frecuente es con elevación del segmento ST y la zona más afectada fue la inferior, ya que este sitio corresponde a la irrigación de la CD, que está estrechamente relacionada al ser la arteria más afectada por EC, fenómeno concordante con las diferentes series consultadas, seguido de la CX, DA correspondiente a la cara anterior y raramente afectado el TC, en cambio solo un pequeño porcentaje de las arterias con ectasia fueron candidatas a un procedimiento percutáneo, siendo el stent farmacológico ( everolimus promus primer ) más utilizado en esta entidad. La clasificación Markis permite definir la cantidad de vasos afectados y qué proporción del vaso se encuentra dilatado, además de ser usada como un predictor importante de sobrevida y mortalidad a corto plazo y largo plazo, con una tasa de mortalidad del 2% , incluso su utilidad para definir el tipo de terapia antitrombótica que podría recibir el paciente. En nuestro estudio se reporta que la clasificación Markis tipo 3 es la más frecuente, esperando que en estos pacientes la mortalidad sea menor comparada con el tipo 4, que fue poco reportada. A pesar del conocimiento de esta patología, ha resultado difícil elegir un plan terapéutico adecuado para los pacientes y de acuerdo con los estudios WOEST e ISAR-TRIPLE se recomienda el uso de la terapia doble en algunos pacientes debido a menor riesgo de sangrado e incluso disminución de la mortalidad, sin embargo aún no se demuestra con certeza el grado de efectividad entre ambas terapias para la EC. De acuerdo con esta revisión, se prefiere el uso de terapia APD + ACO, en un periodo de un mes, ya que la combinación de estos fármacos tendría una efectividad significativa para la prevención de recurrencias en eventos isquémicos, con un posterior cambio a doble terapia a largo plazo, debido a que en la EC hay una constante activación plaquetaria y procoagulante, así como un flujo turbulento por daño al endotelio vascular. Se observó el uso preferente de ACO clásicos, aunque se ha propuesto que el uso de los NACO puede tener un beneficio mayor que los ACO clásicos en el manejo de la EC . En el ensayo COMPASS, el uso de rivaroxabán más aspirina en pacientes con enfermedad coronaria estable tuvo un 3.4% de muertes, en comparación con aspirina sola, un 4.1%; esto refleja que el rivaroxabán aminora la progresión de la placa aterosclerótica, aunque se necesitan más estudios que comparen ACO vs. NACO . La EC es una remodelación patológica no infrecuente el INC. Se observó un ingreso equivalente de 35 pacientes por año en el área de hospitalización de adultos, siendo un IAMCEST la manifestación más típica de la EC; los pacientes presentaron una FEVI dentro de valores limítrofes, una coronariografía diagnóstica con un Markis tipo 3, por lo que se esperaría una tasa baja de mortalidad y recurrencia de eventos cardiovasculares a largo plazo. Aún no existe un consenso sobre la terapia ideal, sin embargo en el INC se prefiere el tratamiento individualizado, recomendando modificación en el estilo de vida y empleando como tratamiento médico el uso de terapia APD+ACO solo al momento de egreso del paciente, con su respectivo ajuste de tratamiento al mes a una terapia APS+ACO y con el objetivo de prevenir recurrencias en eventos coronarios.
Efecto de la pandemia por COVID-19 en la formación de los residentes de cardiología: más allá del efecto clínico	2e3f6751-1f93-4fd4-87d1-a8074943ba4f	10161867	Internal Medicine[mh]	La COVID-19, causada por el coronavirus SARS-CoV-2, ha provocado una pandemia con más de 28.4 millones de casos y 906,000 muertes a nivel mundial hasta el 12 de septiembre del 2020 . España ha sido uno de los países más afectados en Europa y la adaptación del sistema sanitario español ha jugado un papel clave en la atención de la pandemia. Tal y como ha ocurrido en los hospitales de otros países del mundo , se ha desplazado a los médicos de distintas especialidades de su departamento habitual de trabajo para atender a pacientes afectados por la COVID-19. Las investigaciones previas han señalado que la adaptación de los sistemas sanitarios a la pandemia por la COVID-19 (suspensión de rotaciones formativas intrahospitalarias, sesiones de docencia y procedimientos no urgentes) podría afectar a la formación de los médicos residentes en formación . En España, la formación sanitaria especializada se organiza a través del programa médico interno residente (MIR). Después de terminar la escuela de medicina, los médicos graduados deben pasar por una prueba nacional que permite ingresar a un programa de formación, una prueba que es la misma para las especialidades clínicas, quirúrgicas y de medicina familiar. El MIR de cardiología en España es un programa de cinco años de duración, en cuyo primer año se realizan rotaciones clínicas externas no cardiológicas (como medicina interna, nefrología, neumología, endocrinología y otros departamentos afines). El segundo año se destina sobre todo a la cardiología clínica y el tercer año a los cuidados agudos cardiológicos e imagen cardíaca. El cuarto año se compone de modo predominante de hemodinámica y electrofisiología, mientras que los MIR del quinto año rotan principalmente por insuficiencia cardíaca avanzada (incluido el trasplante de corazón y dispositivos de asistencia mecánica) y rotaciones externas específicas de cardiología. Dentro del ámbito de la cardiología, la actividad asistencial sufrió cambios drásticos durante esos meses . Los médicos internos residentes (MIR) no quedaron exentos de ellos y sus calendarios formativos se han adaptado como consecuencia de esta situación. Sin embargo, se desconoce la transcendencia a largo plazo de estos ajustes, así como la relevancia concebida por los propios residentes. Esta investigación tiene como objetivo conocer la opinión de los MIR de cardiología de España sobre el efecto de la pandemia por COVID-19 en su formación, así como también los principales cambios organizativos de sus servicios que podrían influir en su formación como cardiólogos. Diseño del estudio Se realizó un estudio de corte transversal mediante el empleo de una encuesta digital, voluntaria y anónima, difundida a través de los correos electrónicos de las distintas comisiones de docencia o secretarías de cardiología de todos los hospitales de España que cuentan con programa de formación sanitaria especializada en cardiología. En el correo electrónico se solicitaba la difusión a los MIR de cardiología de su respectivo hospital, por lo que los investigadores nunca obtuvieron información personal que permitiera identificar o contactar directamente a los participantes. Para llevar a cabo este estudio se obtuvo aprobación del Comité de ética del Principado de Asturias (número de referencia 2020.363). Se solicitó consentimiento informado de manera implícita dentro de la encuesta por la naturaleza del estudio: sin contactar de modo directo con los participantes, anónimo y voluntario. En la encuesta se indagaba sobre cambios adaptativos en los servicios de cardiología, cambios en rotaciones formativas estipuladas (p. ej., suspensión de rotaciones clínicas, de imagen, intervencionistas, etc.), traslados a otros servicios o centros sanitarios y sobre la percepción en el efecto formativo y emocional. Las preguntas fueron en la modalidad de selección múltiple, con oportunidad de seleccionar sólo una opción por pregunta. El período de encuesta fue del 12 al 21 de mayo del 2020. Datos epidemiológicos Se utilizaron los datos de la Dirección General de Salud Pública, Calidad e Innovación del Ministerio de Sanidad de España, en su actualización número 103 con fecha del 12 de mayo de 2020 y datos demográficos del Instituto Nacional de Estadística, que coincidieron con la fecha de inicio del período de encuesta. De estas fuentes se obtuvieron el número de casos totales, la prevalencia (número de casos por cada 1,000 habitantes) y la incidencia en cada una de las 17 comunidades autónomas (CA) de España (en este país, las distintas regiones reciben el nombre de comunidades autónomas). Las CA se clasificaron con posterioridad con base en una prevalencia mayor o menor de 5 casos/1,000 habitantes. Las ciudades autónomas de Ceuta y Melilla no se incluyeron ya que no cuentan con programa formativo de MIR en cardiología. Análisis estadístico El análisis estadístico se realizó con el software STATA 15.2 (Stata Corp. LP, EE.UU.). De manera inicial se realizó un análisis descriptivo de las distintas variables (todas ellas categóricas) para conocer la opinión global de los MIR de cardiología y el porcentaje que consideraba afectada su formación. Después se realizaron comparaciones entre los distintos grupos mediante la prueba estadística ji cuadrada de Pearson. Se realizó un modelo de regresión logística para analizar los factores estudiados que se relacionaron con una percepción negativa en su formación. Por último, se analizaron los cambios organizativos que efectuaron los servicios de cardiología de cada centro y por CA de acuerdo con su prevalencia de casos de COVID-19. Se realizó un estudio de corte transversal mediante el empleo de una encuesta digital, voluntaria y anónima, difundida a través de los correos electrónicos de las distintas comisiones de docencia o secretarías de cardiología de todos los hospitales de España que cuentan con programa de formación sanitaria especializada en cardiología. En el correo electrónico se solicitaba la difusión a los MIR de cardiología de su respectivo hospital, por lo que los investigadores nunca obtuvieron información personal que permitiera identificar o contactar directamente a los participantes. Para llevar a cabo este estudio se obtuvo aprobación del Comité de ética del Principado de Asturias (número de referencia 2020.363). Se solicitó consentimiento informado de manera implícita dentro de la encuesta por la naturaleza del estudio: sin contactar de modo directo con los participantes, anónimo y voluntario. En la encuesta se indagaba sobre cambios adaptativos en los servicios de cardiología, cambios en rotaciones formativas estipuladas (p. ej., suspensión de rotaciones clínicas, de imagen, intervencionistas, etc.), traslados a otros servicios o centros sanitarios y sobre la percepción en el efecto formativo y emocional. Las preguntas fueron en la modalidad de selección múltiple, con oportunidad de seleccionar sólo una opción por pregunta. El período de encuesta fue del 12 al 21 de mayo del 2020. Se utilizaron los datos de la Dirección General de Salud Pública, Calidad e Innovación del Ministerio de Sanidad de España, en su actualización número 103 con fecha del 12 de mayo de 2020 y datos demográficos del Instituto Nacional de Estadística, que coincidieron con la fecha de inicio del período de encuesta. De estas fuentes se obtuvieron el número de casos totales, la prevalencia (número de casos por cada 1,000 habitantes) y la incidencia en cada una de las 17 comunidades autónomas (CA) de España (en este país, las distintas regiones reciben el nombre de comunidades autónomas). Las CA se clasificaron con posterioridad con base en una prevalencia mayor o menor de 5 casos/1,000 habitantes. Las ciudades autónomas de Ceuta y Melilla no se incluyeron ya que no cuentan con programa formativo de MIR en cardiología. El análisis estadístico se realizó con el software STATA 15.2 (Stata Corp. LP, EE.UU.). De manera inicial se realizó un análisis descriptivo de las distintas variables (todas ellas categóricas) para conocer la opinión global de los MIR de cardiología y el porcentaje que consideraba afectada su formación. Después se realizaron comparaciones entre los distintos grupos mediante la prueba estadística ji cuadrada de Pearson. Se realizó un modelo de regresión logística para analizar los factores estudiados que se relacionaron con una percepción negativa en su formación. Por último, se analizaron los cambios organizativos que efectuaron los servicios de cardiología de cada centro y por CA de acuerdo con su prevalencia de casos de COVID-19. Población estudiada Participó un total de 180 MIR de cardiología, con una distribución equitativa de acuerdo con el año de residencia. Se obtuvo la participación de al menos un residente de cada CA, y las CA con mayor representación fueron Andalucía, Madrid y Cataluña. Las características generales de los participantes se reúnen en la . Adaptación de los programas formativos de los MIR de cardiología La rotación formativa del 84% (n = 151) de los encuestados se suspendió, con el 49% (n = 84) que fue desplazado a otro servicio distinto de cardiología. Sólo el 3% (n = 5) se desplazó a otro centro sanitario. El 65% (n = 118) interrumpió alguna rotación por más de un mes y en más de la mitad de los casos (54%, n = 101) la rotación se dio por completada y no recuperable. Los residentes de tercer año son los que podrán recuperar con más frecuencia las rotaciones afectadas, con una diferencia estadísticamente significativa (p = 0.011). Factores relacionados con la percepción de la influencia negativa en su formación Del total de los 180 participantes, el 52% (n = 94) consideró que su formación como especialista se había afectado de manera negativa, aunque el 82% (n = 77) de éstos refiere que fue de forma leve y potencialmente recuperable ( ). La percepción de los participantes según el año de residencia se observa en la . Por el contrario, un 36% (n = 64) de residentes considera que su formación como especialista se ha enriquecido por la pandemia, ya que obligó a trabajar en nuevas áreas en las que no se habría entrenado si no fuera por la situación especial de la pandemia por la COVID-19. En la regresión logística, los factores relacionados con percibir su formación como afectada negativamente fueron: encontrarse en el tercer año de residencia (OR: 3.66; IC 95%, 1.013-13.217; p = 0.048, al compararse con los del primer año de residencia), desplazamiento por más de un mes de su rotación formativa (OR: 7.01; IC 95%, 1.54-31.99; p = 0.012) y encontrarse en la rotación formativa de imagen cardíaca (p = 0.001) (Tabla y ). Hallarse en el quinto (último) año de residencia se reveló como un factor protector para percibir su formación como afectada (OR: 0.11; IC 95%, 0.03-0.48; p = 0.003, al compararse con los del primer año de residencia). Con respecto al efecto emocional, el 41% (n = 74) de los participantes considera que, si bien ha vivido momentos estresantes, éstos no influirán en su futuro profesional y sólo el 9% (n = 17) sí considera que tendrá transcendencia a largo plazo. Cambios organizativos en los servicios de cardiología como resultado de la pandemia por la COVID-19 En los hospitales, de un 7.22% (n = 13) de los participantes, las guardias de cardiología se suspendieron de manera parcial o total y un 32.22% (n = 58) redujo el número de integrantes de la guardia respecto de lo usual. Si se analiza por año de residencia, no hubo diferencia estadísticamente significativa para ser desplazados del servicio (p = 0.272) o del hospital (p = 0.067). Si se tiene en cuenta la prevalencia de la enfermedad en cada CA, y se compara a los residentes de aquéllas con más de 5 casos/1,000 habitantes (Castilla la Mancha, Castilla León, Cataluña, Comunidad de Madrid, Navarra, País Vasco y La Rioja) en relación con las de menor prevalencia, se identifican diferencias estadísticamente significativas en el desplazamiento de servicio ( , p = 0.024), pero no en el desplazamiento del hospital (p = 0.064) ni en la pérdida/disminución de personal de guardia (p = 0.638). Participó un total de 180 MIR de cardiología, con una distribución equitativa de acuerdo con el año de residencia. Se obtuvo la participación de al menos un residente de cada CA, y las CA con mayor representación fueron Andalucía, Madrid y Cataluña. Las características generales de los participantes se reúnen en la . La rotación formativa del 84% (n = 151) de los encuestados se suspendió, con el 49% (n = 84) que fue desplazado a otro servicio distinto de cardiología. Sólo el 3% (n = 5) se desplazó a otro centro sanitario. El 65% (n = 118) interrumpió alguna rotación por más de un mes y en más de la mitad de los casos (54%, n = 101) la rotación se dio por completada y no recuperable. Los residentes de tercer año son los que podrán recuperar con más frecuencia las rotaciones afectadas, con una diferencia estadísticamente significativa (p = 0.011). Del total de los 180 participantes, el 52% (n = 94) consideró que su formación como especialista se había afectado de manera negativa, aunque el 82% (n = 77) de éstos refiere que fue de forma leve y potencialmente recuperable ( ). La percepción de los participantes según el año de residencia se observa en la . Por el contrario, un 36% (n = 64) de residentes considera que su formación como especialista se ha enriquecido por la pandemia, ya que obligó a trabajar en nuevas áreas en las que no se habría entrenado si no fuera por la situación especial de la pandemia por la COVID-19. En la regresión logística, los factores relacionados con percibir su formación como afectada negativamente fueron: encontrarse en el tercer año de residencia (OR: 3.66; IC 95%, 1.013-13.217; p = 0.048, al compararse con los del primer año de residencia), desplazamiento por más de un mes de su rotación formativa (OR: 7.01; IC 95%, 1.54-31.99; p = 0.012) y encontrarse en la rotación formativa de imagen cardíaca (p = 0.001) (Tabla y ). Hallarse en el quinto (último) año de residencia se reveló como un factor protector para percibir su formación como afectada (OR: 0.11; IC 95%, 0.03-0.48; p = 0.003, al compararse con los del primer año de residencia). Con respecto al efecto emocional, el 41% (n = 74) de los participantes considera que, si bien ha vivido momentos estresantes, éstos no influirán en su futuro profesional y sólo el 9% (n = 17) sí considera que tendrá transcendencia a largo plazo. En los hospitales, de un 7.22% (n = 13) de los participantes, las guardias de cardiología se suspendieron de manera parcial o total y un 32.22% (n = 58) redujo el número de integrantes de la guardia respecto de lo usual. Si se analiza por año de residencia, no hubo diferencia estadísticamente significativa para ser desplazados del servicio (p = 0.272) o del hospital (p = 0.067). Si se tiene en cuenta la prevalencia de la enfermedad en cada CA, y se compara a los residentes de aquéllas con más de 5 casos/1,000 habitantes (Castilla la Mancha, Castilla León, Cataluña, Comunidad de Madrid, Navarra, País Vasco y La Rioja) en relación con las de menor prevalencia, se identifican diferencias estadísticamente significativas en el desplazamiento de servicio ( , p = 0.024), pero no en el desplazamiento del hospital (p = 0.064) ni en la pérdida/disminución de personal de guardia (p = 0.638). El estudio presentado es el primero realizado en España cuyo objetivo es conocer el efecto de la pandemia por COVID-19 en la formación de los MIR. En opinión de los residentes de cardiología que participaron, la pandemia ha supuesto cambios sustanciales en su programa formativo y la organización asistencial de sus servicios. Ha representado la pérdida de rotaciones específicas, desplazamientos a otros servicios dentro de su hospital y cambios en la organización de sus servicios. El porcentaje de MIR en cardiología de esta investigación que percibieron su formación como afectada (52%, n = 94) resultó similar a la opinión de los directores de programa de centros sanitarios en la ciudad de Nueva York con médicos en formación en cardiología intervencionista, quienes opinaron que la formación de sus médicos en formación se vería moderadamente afectada (57%) e incluso muy afectada (14%) . La percepción de los MIR, según lo mostrado en los resultados, ha estado condicionada al menos en parte por la situación particular del residente, sobre todo por el año de formación en que cursaba y por la rotación formativa en la que se encontraba. Los residentes que se hallaban a mitad del programa formativo (residentes de tercer año) circulan por rotaciones básicas para la formación de los cardiólogos (como imagen cardíaca y cuidados agudos cardiológicos), lo cual podría ser una de las razones por lo que este grupo de residentes percibió, con una diferencia estadísticamente significativa, que su formación podría afectarse por los meses de rotación que se alteraron. Por otro lado, los residentes de quinto año, quienes ya han superados las rotaciones formativas básicas de la cardiología y se encuentran en rotaciones más específicas (tanto externas como en la parte de insuficiencia cardíaca avanzada) consideraron, también con una diferencia estadísticamente significativa, que su formación como cardiólogos no se afectó por la pandemia, por lo que desde el punto de vista de los autores, cursar el quinto año de residencia podría considerarse entonces como un factor “protector”. Se identificaron diferencias de acuerdo con la CA y el mayor desplazamiento del servicio, siendo mayor el desplazamiento en las CA con prevalencia de casos de COVID-19 > 5 casos/1,000 habitantes. Esto puede explicarse porque las CA con mayor prevalencia de casos fueron forzadas quizá a desplazar a sus residentes de sus rotaciones formativas en cardiología a otros lugares de mayor presión asistencial durante la pandemia, similar a lo que ocurrió en otros continentes . No obstante, en la mayor parte de los casos, la percepción del residente es que esta situación de pandemia por COVID-19 no implicará un detrimento considerable en su formación global, e incluso algunos consideran que ha propiciado un enriquecimiento personal y laboral. El efecto psicológico en los encuestados ha sido menor, de modo semejante a lo observado en médicos residentes de otros países . Este estudio posee algunas fortalezas, como ser el primer estudio realizado en España cuyo fin es conocer la percepción de los MIR sobre el efecto de la pandemia por la COVID-19 en su formación como médicos especialistas en cardiología. De igual manera, el estudio se condujo de una manera conveniente para la situación epidemiológica crítica vivida en España durante la pandemia, ya que se realizó a través de una plataforma digital (para evitar el contacto personal). Por otro lado, este estudio tiene como limitaciones que no obtuvo la participación de más del 50% de los MIR en cardiología en el momento de la encuesta, lo que podría limitar la extrapolación de los resultados. El sesgo de selección pudo ser una de las limitantes del estudio, si se considera que la recepción final del correo con el enlace a la encuesta dependió de que el mensaje original se reenviara por las comisiones de docencia/secretaría a los MIR en cardiología para su participación. En definitiva, este estudio es de interés para conocer cómo la pandemia por la COVID-19 ha influido en la formación de los MIR en cardiología de España. Como perspectiva futura sería de interés desarrollar formas de docencia o simulación que se acoplaran a los programas formativos actuales y sirvieran para situaciones en las que los médicos residentes de distintas especialidades no puedan llevar a cabo un número adecuado de intervenciones o procedimientos por situaciones ajenas al sistema de salud, como otra pandemia. Sería conveniente también conocer la percepción sobre el efecto de la pandemia en la formación de otros médicos en formación (tanto de grado como de especialización en alguna rama de la medicina) en el resto de Europa, América y el mundo. La pandemia por la COVID-19 ha llevado a la reorganización de la actividad asistencial, en casi todos los casos con afectación directa del programa formativo de los MIR en cardiología de España. Si bien la mayoría de los MIR en cardiología suspendió de manera transitoria su programa formativo, la percepción negativa en su formación fue mayor en los residentes que se encontraban a mitad del período formativo y quienes rotaban en imagen cardíaca. Ninguno. Los autores declaran que no tienen conflicto de intereses con la autoría o publicación de este artículo. Se obtuvo aprobación por parte del Comité de ética del Principado de Asturias, con número de referencia 2020.363. Protección de sujetos humanos y animales. Los autores declaran que no se realizaron experimentos en seres humanos o animales para este estudio. Confidencialidad de los datos. Los autores declaran haber seguido los protocolos de su centro de trabajo sobre la publicación de datos de pacientes. Derecho a la privacidad y consentimiento informado. Los autores han obtenido el consentimiento informado por escrito de los pacientes o sujetos mencionados en el artículo. El autor de correspondencia se halla en posesión de este documento.
Morphological variations of the human spleen: no evidence for a multifocal or lobulated developmental origin	91a189b7-4a71-4e16-aa77-b95a876d46ad	10161904	Forensic Medicine[mh]	The spleen is often considered a “forgotten organ” among clinicians and radiologist. Still, the spleen may be involved in a variety of congenital and acquired conditions and can be well visualised on abdominal CT and MRI. As the spleen is a highly variable organ, radiologists frequently encounter normal variants which might be misinterpreted. For example, deep splenic clefts may be misinterpreted as splenic lacerations in patients with abdominal trauma and there have been several case reports of accessory spleens being mistaken for lymphadenopathy or intra-abdominal tumours, most notably in the liver and pancreatic tail. Central to increasing our understanding of normal spleen morphology, is accurate knowledge about normal spleen development and clearly defined terms to describe anatomical variants. Natural incisions of the splenic surface may be referred to as clefts, notches or fissures. These terms are used heterogeneously in the literature and a clear definition is lacking ( ). Therefore, we use exclusively the term cleft to describe a natural incision of the splenic surface for the purpose of this paper. Splenic clefts have a reported prevalence of 40–98% on the superior border. Four to five clefts on the superior border are considered normal ( ). An excessive number of clefts, or clefts situated on locations other than the superior border—including on the diaphragmatic and visceral surfaces of the spleen—is regarded abnormal. Another relatively frequent finding is the presence of an accessory spleen; a small nodule of splenic tissue that is completely separate from the main body, which is found in 10–30% of patients at autopsy. Variations in adult spleen morphology are often attributed to the common assumption that multiple spleen primordia fuse to form the adult spleen during early embryonic development. It is believed that when one or more splenic buds fail to fuse to the main body, this results in an accessory spleen. During the foetal stage, the spleen is thought to be lobulated, where the natural incisions forming the lobules demonstrate incomplete fusion of the embryonic spleen primordia. In turn, splenic clefts are considered remnants of these foetal lobulations in adults. Although these notions are commonly accepted, in this paper, we address some major issues with the evidence in support of these concepts. The first issue we address is the fact that the term “foetal lobulation” with regard to splenic surface morphology is poorly defined and little original research papers that examined normal spleen morphology in human foetuses exist. To the best of our knowledge, Ungör and colleagues’ assessment of 141 dissected foetal spleens, is the only study using foetal specimens for studying splenic surface morphology. They found a prevalence of 95% clefts on the superior border and 7.8 and 3.5% clefts on respectively the diaphragmatic and visceral surface. These numbers show a remarkable similarity to the numbers found in adult spleens which raises the question whether the surface morphology differs between foetuses and adults. The second issue is that the assertion that both the occurrence of (persistent) foetal splenic lobulation and accessory spleens can be explained through a multifocal origin of the spleen, is not substantiated by observations in human or animal embryos. To the best of our knowledge, a multifocal origin of the spleen has never been described based on primary observations in human, nor animal embryos and can only be found in secondary sources. We hypothesise that the spleen originates from a single primordium during the embryonic period and that variations in spleen morphology form during development. Furthermore, we hypothesise that splenic morphology is highly variable in both adults and foetuses. In this study, we first provide an overview of embryonic spleen development from the first appearance of the splenic primordium until the end of the embryonic period based on histological sections. Second, we provide quantitative data on the number of splenic clefts in adults and foetuses based on post-mortem micro-CT scans of human foetuses and post-mortem CT scans of adults and statistically test for differences between these groups. Embryonic spleen development For the first part of this study, 22 human embryos from the Carnegie and Boyd collections ( ), ranging from 4 to 8.5 weeks post-conception, corresponding to Carnegie stages (CS) 13–23, were used to study the development of the spleen. Histological sections of two specimens per stage were analysed with special attention to the occurrence of single or multiple splenic primordia and surface anatomy of the developing spleen. Specimen characteristics and staining methods are presented in . Methods for image acquisition have been described previously by de Bakker et al. Foetal spleen morphology Seventeen human foetal specimens, aged 11–21 weeks post-conception ( ), were obtained from the Dutch Fetal Biobank (DFB), located at the Amsterdam University Medical Centers (Amsterdam UMC), location AMC, the Netherlands. The DFB collects high quality human foetuses in toto following medical termination of pregnancy for foetal or maternal indication (including social indications). Written informed consent was obtained after the decision to terminate the pregnancy was made and prior to delivery. All specimens were fixed in 4.0% paraformaldehyde (PFA)) in 10 mM Na 2 HPO 4 / NaH 2 PO 4 , 150 mM NaCl pH 7.4 (PBS) for 2–7 days at 4°C, depending on foetal size, and stored in 0.2% PFA in PBS. Submersion of the specimens in a 3.75% Lugol’s Iodine solution was used to enhance soft-tissue contrast. Prior to scanning, each foetus was stabilised in 1.5% agarose gel to prevent movement artefacts during scanning. Micro-CT scans were performed using a GE Phoenix v\|tome\|x tomographer (General Electric, Wunstorf, Germany). The voltage (180–210 kV) and current (180–210 µA) and resolution were adjusted according to foetal size. Adult spleen morphology We retrospectively screened the human cadaver body donation database of the Department of Medical Biology, Section Clinical Anatomy and Embryology of the Amsterdam UMC, location AMC and identified 90 adults with available post-mortem CT scans between 1 February 2017 and 1 March 2021. The age, cause of death and medical history could not be obtained from the anonymised records. Post-mortem CT-scans were made without contrast using a SOMATOM clinical CT scanner (Siemens Healthcare GmbH, Erlangen, Germany). Scans were made using dual energy with a voltage of 100 and 150 kV and a current of respectively 750 and 375 mAs. Ethical approval All protocols concerning the Dutch Fetal Biobank were approved by the medical ethics committee of Amsterdam UMC, location AMC (2016; 285, #B2017369). The adult bodies were donated to science in accordance with Dutch legislation and the regulations of the medical ethics committee of the Amsterdam UMC, location AMC. Due to the anonymous and retrospective nature of the study, no further approval was needed. Measurements and additional data analysis Foetal post-mortem micro-CT and adult post-mortem CT scans were assessed for the presence of splenic clefts using multiplanar view ( ). We classified the location of splenic clefts as extending from the superior border, inferior border, anterior extremity, posterior extremity, diaphragmatic surface and visceral surface, according to Netter’s classical anatomical subdivision ( ). Clefts were only counted if the cleft could be identified in more than one imaging plane and if visible in multiple consecutive slices in at least one imaging plane. Clefts that extended from the location where blood vessels enter or leave the splenic hilum were not counted ( ). The investigator performing the analyses was blinded to the identity and age of the foetuses. Statistics Data on embryonic spleen development are presented descriptively. The data of foetal and adult CT scans were analysed using SPSS (v. 26.0 for Windows; IBM Corp. Armonk, NY). We used a scatter plot and linear regression to determine whether the number of clefts was associated with the post-conception age. We performed an independent samples Kolmogorov–Smirnov (KS) test to compare distributions of clefts between groups of foetal and adult spleens. To examine whether the presence of genetic, chromosomal or other structural abnormalities were associated with the number of splenic clefts in the foetal spleens, we performed two subgroup analyses using the independent samples KS test. For the first subgroup analysis, we used a very strict definition for normal foetuses to test for any possible association between extrasplenic abnormalities and the number of splenic clefts observed. We defined following subgroups: Normal foetuses: foetuses donated following termination of pregnancy on a social indication who had no known abnormalities. Abnormal foetuses: foetuses with any abnormality, including minor monogenetic defects that have no known association with structural abnormalities. We performed a second subgroup analysis using a broader definition of normal. For this analysis, we expanded the group of foetuses considered normal to also include those with minor monogenetic defects that have no known association with structural abnormalities, which we considered to be structurally normal. This definition allowed for more equal group sizes. We considered a p -value < 0.05 statistically significant. For the first part of this study, 22 human embryos from the Carnegie and Boyd collections ( ), ranging from 4 to 8.5 weeks post-conception, corresponding to Carnegie stages (CS) 13–23, were used to study the development of the spleen. Histological sections of two specimens per stage were analysed with special attention to the occurrence of single or multiple splenic primordia and surface anatomy of the developing spleen. Specimen characteristics and staining methods are presented in . Methods for image acquisition have been described previously by de Bakker et al. Seventeen human foetal specimens, aged 11–21 weeks post-conception ( ), were obtained from the Dutch Fetal Biobank (DFB), located at the Amsterdam University Medical Centers (Amsterdam UMC), location AMC, the Netherlands. The DFB collects high quality human foetuses in toto following medical termination of pregnancy for foetal or maternal indication (including social indications). Written informed consent was obtained after the decision to terminate the pregnancy was made and prior to delivery. All specimens were fixed in 4.0% paraformaldehyde (PFA)) in 10 mM Na 2 HPO 4 / NaH 2 PO 4 , 150 mM NaCl pH 7.4 (PBS) for 2–7 days at 4°C, depending on foetal size, and stored in 0.2% PFA in PBS. Submersion of the specimens in a 3.75% Lugol’s Iodine solution was used to enhance soft-tissue contrast. Prior to scanning, each foetus was stabilised in 1.5% agarose gel to prevent movement artefacts during scanning. Micro-CT scans were performed using a GE Phoenix v\|tome\|x tomographer (General Electric, Wunstorf, Germany). The voltage (180–210 kV) and current (180–210 µA) and resolution were adjusted according to foetal size. We retrospectively screened the human cadaver body donation database of the Department of Medical Biology, Section Clinical Anatomy and Embryology of the Amsterdam UMC, location AMC and identified 90 adults with available post-mortem CT scans between 1 February 2017 and 1 March 2021. The age, cause of death and medical history could not be obtained from the anonymised records. Post-mortem CT-scans were made without contrast using a SOMATOM clinical CT scanner (Siemens Healthcare GmbH, Erlangen, Germany). Scans were made using dual energy with a voltage of 100 and 150 kV and a current of respectively 750 and 375 mAs. All protocols concerning the Dutch Fetal Biobank were approved by the medical ethics committee of Amsterdam UMC, location AMC (2016; 285, #B2017369). The adult bodies were donated to science in accordance with Dutch legislation and the regulations of the medical ethics committee of the Amsterdam UMC, location AMC. Due to the anonymous and retrospective nature of the study, no further approval was needed. Foetal post-mortem micro-CT and adult post-mortem CT scans were assessed for the presence of splenic clefts using multiplanar view ( ). We classified the location of splenic clefts as extending from the superior border, inferior border, anterior extremity, posterior extremity, diaphragmatic surface and visceral surface, according to Netter’s classical anatomical subdivision ( ). Clefts were only counted if the cleft could be identified in more than one imaging plane and if visible in multiple consecutive slices in at least one imaging plane. Clefts that extended from the location where blood vessels enter or leave the splenic hilum were not counted ( ). The investigator performing the analyses was blinded to the identity and age of the foetuses. Data on embryonic spleen development are presented descriptively. The data of foetal and adult CT scans were analysed using SPSS (v. 26.0 for Windows; IBM Corp. Armonk, NY). We used a scatter plot and linear regression to determine whether the number of clefts was associated with the post-conception age. We performed an independent samples Kolmogorov–Smirnov (KS) test to compare distributions of clefts between groups of foetal and adult spleens. To examine whether the presence of genetic, chromosomal or other structural abnormalities were associated with the number of splenic clefts in the foetal spleens, we performed two subgroup analyses using the independent samples KS test. For the first subgroup analysis, we used a very strict definition for normal foetuses to test for any possible association between extrasplenic abnormalities and the number of splenic clefts observed. We defined following subgroups: Normal foetuses: foetuses donated following termination of pregnancy on a social indication who had no known abnormalities. Abnormal foetuses: foetuses with any abnormality, including minor monogenetic defects that have no known association with structural abnormalities. We performed a second subgroup analysis using a broader definition of normal. For this analysis, we expanded the group of foetuses considered normal to also include those with minor monogenetic defects that have no known association with structural abnormalities, which we considered to be structurally normal. This definition allowed for more equal group sizes. We considered a p -value < 0.05 statistically significant. Embryonic spleen development An overview of embryonic spleen development is given in . Comparing CS13 (4–4.5 post-conception weeks; PCW) and CS14 (4.5–5 PCW), the first sign of splenic development was observed in CS14, where the spleen primordium was identified as single bulge in the left side of the dorsal mesogastrium. The spleen primordium only became a clearly distinguishable structure within the dorsal mesogastrium at CS17 (6.5 PCW) because of the higher density of the staining compared to the surrounding tissue. Evaluating the entire series of sections of each embryo, the high-density area forming the spleen primordium was observed as a single circumscript structure in both CS17 specimens and all later staged embryos. The first signs of separation of the spleen from the dorsal mesogastrium were observed at CS20 (7.5 PCW) and became clearly evident at CS21-23 (7.5–8.5 PCW), forming the splenic hilum at CS23 (8–8.5 PCW). No evidence for the presence of multiple mesenchymal condensations within the dorsal mesogastrium could be identified in macroscopic or microscopic views in any of the sections of any of the studied embryos. Interestingly, from CS16 (5.5–6 PCW) onwards, clefts occurred in some, but not all embryos. The occurrence of clefts was unpredictable and no correlation to the developmental stage was observed ( ). Comparison between foetal and adult spleen morphology In both the foetal and adult spleen, a wide variation in splenic surface morphology was observed ( ). shows the number of spleens in both groups with clefts on the different borders and surfaces of the spleen. No major differences in the occurrence of clefts at different borders and surfaces were observed between groups. Spleens without any clefts at any location occurred in 3 out of 17 foetuses (18%), compared to 25 out of 90 adults (28%) ( ). In foetal spleens where clefts did occur, the total number of clefts ranged between 1 and 6 (median 3), compared to a range of 1 and 5 in adults (median 1) ( ). Statistical testing showed no significant difference in distributions of total number of splenic clefts between the foetal and adult sample ( p = 0.068). Moreover, no correlation was observed between foetal age and the total number of clefts of the splenic surface (R 2 = 0.004) ( ). The subgroup analyses comparing foetuses with and without known genetic, chromosomal or extrasplenic structural abnormalities showed no statistically significant differences. In the first subgroup analysis, we compared 5 foetuses considered normal according to the strict definition to 12 considered abnormal and found no difference ( p = 0.680). In the second subgroup analysis, we compared nine foetuses considered normal using the broader definition to eight considered abnormal and again found no difference ( p = 0.997). An overview of embryonic spleen development is given in . Comparing CS13 (4–4.5 post-conception weeks; PCW) and CS14 (4.5–5 PCW), the first sign of splenic development was observed in CS14, where the spleen primordium was identified as single bulge in the left side of the dorsal mesogastrium. The spleen primordium only became a clearly distinguishable structure within the dorsal mesogastrium at CS17 (6.5 PCW) because of the higher density of the staining compared to the surrounding tissue. Evaluating the entire series of sections of each embryo, the high-density area forming the spleen primordium was observed as a single circumscript structure in both CS17 specimens and all later staged embryos. The first signs of separation of the spleen from the dorsal mesogastrium were observed at CS20 (7.5 PCW) and became clearly evident at CS21-23 (7.5–8.5 PCW), forming the splenic hilum at CS23 (8–8.5 PCW). No evidence for the presence of multiple mesenchymal condensations within the dorsal mesogastrium could be identified in macroscopic or microscopic views in any of the sections of any of the studied embryos. Interestingly, from CS16 (5.5–6 PCW) onwards, clefts occurred in some, but not all embryos. The occurrence of clefts was unpredictable and no correlation to the developmental stage was observed ( ). In both the foetal and adult spleen, a wide variation in splenic surface morphology was observed ( ). shows the number of spleens in both groups with clefts on the different borders and surfaces of the spleen. No major differences in the occurrence of clefts at different borders and surfaces were observed between groups. Spleens without any clefts at any location occurred in 3 out of 17 foetuses (18%), compared to 25 out of 90 adults (28%) ( ). In foetal spleens where clefts did occur, the total number of clefts ranged between 1 and 6 (median 3), compared to a range of 1 and 5 in adults (median 1) ( ). Statistical testing showed no significant difference in distributions of total number of splenic clefts between the foetal and adult sample ( p = 0.068). Moreover, no correlation was observed between foetal age and the total number of clefts of the splenic surface (R 2 = 0.004) ( ). The subgroup analyses comparing foetuses with and without known genetic, chromosomal or extrasplenic structural abnormalities showed no statistically significant differences. In the first subgroup analysis, we compared 5 foetuses considered normal according to the strict definition to 12 considered abnormal and found no difference ( p = 0.680). In the second subgroup analysis, we compared nine foetuses considered normal using the broader definition to eight considered abnormal and again found no difference ( p = 0.997). In this study, we evaluated spleen development from just before the first emergence of the primordium (CS13; 4–4.5 PCW) until the end of the second trimester of pregnancy (22 PCW) and compared the results to adult spleens. We provide morphological evidence that the spleen develops from a single primordium, which was first identified at CS14 (4.5–5 PCW). We found the earliest signs of variation of splenic surface morphology at CS16 (5.5–6 PCW). From this stage onwards, splenic clefts occurred in some, but not all studied embryos, suggesting that variants of splenic surface anatomy are already established during the embryonic period. This suspicion is further supported by the finding that variation in the number of clefts in foetuses and adults is similar. Taken together, our findings contradict the frequently mentioned hypothesis that spleen normal variants can be explained by a multifocal embryonic origin and a lobulated foetal stage which may persist after birth. Although it could be argued that absence of evidence does not mean evidence of absence, findings from animal studies provide further support for the suspicion that the spleen has a single, rather than a multifocal origin. Historically, embryological studies have been limited to observing changes in cellular and macroscopic morphology. However, modern molecular and immunolabelling technologies allow for identification of splenic precursor cells, even before morphological changes become distinguishable using traditional methods. Early markers of spleen morphology, such as expression of the homeobox-containing transcription factor Nkx2-5, have been used to investigate the origin of putative splenic mesenchyme in mouse embryos. These studies have revealed that the pre-splenic tissue is initially located in symmetric domains on both sides of the embryonic gut, but during subsequent development, only the left side goes on to form the mature spleen. This is in keeping with the developmental breaking of morphological left–right symmetry of most visceral organs during early developmental stages in vertebrates. Failure to cease right-sided pre-splenic tissue to develop into the mature spleen, results in the formation of two (or more) spleens, as is the case in left isomerism (polysplenia), or may cause the spleen to be formed at the contralateral side of the abdomen, as seen in situs inversus. Both findings occur exclusively in the course of laterality defects and cannot explain the formation of accessory spleens or splenic notches and fissures in a normally developed left-sided spleen. Although the splenic marker Nkx2-5 is expressed in two distinct domains on the left side of the mesogastrium, which could be interpreted as a potential sign of a multifocal origin of the spleen, only the dorsal domain overlaps with other splenic markers, whereas the ventral domain overlaps with expression of the pancreas-specific homebox-containing transcription factor PDX1. The dorsal domain is therefore considered to be the sole source of splenic tissue. These findings provide molecular support for our morphological observations in human embryos. Therefore, the common occurrence of accessory spleens and irregularities of the splenic surface likely arise from a different developmental mechanism than the commonly believed multifocal origin hypothesis. One hypothesis is that splenic clefts arise secondary to pressure from adjacent organs and simultaneous growth of the spleen itself. This theory is in line with our observation that clefts were observed occasionally from CS16 (5.5–6 PCW) onwards, but not in earlier staged embryos. Moreover, in previously published quantitative volumetric growth analyses of the same embryonic specimens, we observed rapid growth of the neighbouring organs such as the liver and adrenal glands, whereas the volume of the abdominal cavity decreased. Morphologically, we observed that the spleen increasingly takes on a shape moulded by the surrounding structures at CS18-21 (6.5–8 PCW), suggesting that there is pressure from these organs on the spleen ( ). At stages 22 and 23 (8–8.5 PCW), the spleen no longer adapts its shape relative to the surrounding structures, but appears to be completely wedged between the liver, adrenal gland and lateral body wall ( ). It does not seem unlikely that a part of the spleen may become completely separated during this process when a deep cleft is formed, or when the tissue is pushed completely apart due to the rapidly increasing volume of surrounding organs and limited space in the abdominal cavity. Alternatively, accessory spleens may develop during formation of the splenic hilum, which occurs after CS20 through separation of the spleen from the dorsal mesogastrium. Our findings, together with the fact that primary sources on splenic development in human embryos as well as experimental animal studies all describe a unifocal origin of the organ, a multifocal origin causing a transient lobulated stage during foetal spleen development becomes highly unlikely. It is very well possible that the idea of a multifocal origin has originated from an urge to explain the presence of multiple accessory spleens from a developmental point of view. The sample sizes of the embryonic and foetal groups were relatively small. In order to maximise the foetal group, we chose to include normal foetuses, as well as foetuses with known abnormalities which are not considered to be associated with splenic malformations. However, as the subgroup analyses showed no difference in the observed number of clefts between normal foetuses and foetuses with abnormalities, it is unlikely that this has impacted our results. Another important limitation to our study is the use of a historical collection of human embryos, which did not allow for staining of splenic precursor cells in these specimens. Consequently, our observations have been limited to descriptive investigations of morphological changes in the splenic and presplenic mesenchyme in the studied specimens. A final limitation is the retrospective and anonymous nature of the human cadaver body donation database. Because of this, we were unable to obtain medical history and cause of death of the adult group and identify potential confounding factors. However, as conditions associated with acquired splenic abnormalities are rare, we expect that chance that this has affected our results is small. It has long been considered common knowledge among radiologists that splenic clefts (also called notches, intrasplenic folds or fissures) are remnants of the lobulated phase of foetal splenic morphology. Our study on the development of the human spleen revealed that splenic morphology is highly variable, independent of age and developmental stage. Therefore, we suggest that splenic clefts, regardless of their number and location, are part of the wide variation in splenic surface morphology.
Illuminating the druggable genome through patent bioactivity data	cb229e91-9eab-40b7-9996-354703620454	10162037	Pharmacology[mh]	One of the most useful and compelling pieces of evidence for the druggability of a new biological target is the existence of molecules that bind with sufficient affinity to modulate the biological activity of the target. However, only about 11% of the proteome has either an approved drug or a compound known to modulate it . Chemical probes represent a special type of small molecule for use in target validation studies, not only having good activity against the target but also selectivity, cellular activity, and potentially other relevant criteria and are subjected to a peer review process to ensure the quality of any conclusions when used by the wider community. The availability of open-access, public databases such as ChEMBL has greatly simplified the task of identifying potential molecules by providing easy access to more than 19 million bioactivity data points on almost 2 million compounds. The source of these data is primarily the peer-reviewed scientific literature manually extracted by curators; but some of the data has been integrated from other databases including PubChem BioAssay and BindingDB , and data are also deposited directly from experimental groups. An additional and potentially valuable source of information and data on bioactive molecules is the patent literature. In drug discovery, patents are routinely filed to protect novel inventions, by both industrial organisations and academic institutions. The relationship between the patent and the “traditional” (academic journal, peer-reviewed) literature has been examined in various published studies, mostly focussed on key questions relating to overlap/duplication and publication date. For example, in a 2009 article the authors found that just 6% of compounds in patents also appeared in the scientific literature in one of the commercial sources included in their study (GVKBIO) . A later study examined 130 drug-target pairs and on average found them published in patents 3.7 years earlier than in scientific articles . A 2017 study concluded that the first molecules for a novel target are more likely to be published first in the literature, whereas novel small molecules more frequently appear first in patents than in literature, regardless of which targets they modulate . Finally, a more recent study selected medicinal patents published between 2014 and 2019 and identified patents with information on small molecules, antibodies and vaccines that could potentially be repurposed for cancer related therapies. Some of the drug-disease links found were not present in scientific literature, while others were found in the articles; in some cases these were published before the patents and in others afterwards . These and similar studies suggest that the patent corpus potentially represents a wealth of information that is not available elsewhere and/or may appear in the scientific literature only after a significant time delay. Previous studies that have attempted to search or annotate pharmaceutical patents with target-compound information include who produced a set of 198 patents manually annotated with chemical compounds, diseases, targets and modes of action by four different groups of curators; who tried to identify targets in titles, abstracts and claims of patents that contained bioactive compound information, combining the search for target names with the search for some keywords that were related to bioactivity data; who compared different methods to extract the key compound from a given patent, and then applied one of the methods to inform the design of AXL kinase inhibitors; who searched full-text patents using keywords related to diabetic nephropathy and further narrowed the search by rules related to frequency and/or patent section of the keywords found, and subsequently extracted the genes mentioned in the claims section of these patents; who performed a search in the SureChEMBL database using keyword and/or chemical structure searches, with the goal of identifying patents with compounds that could activate the BK Ca channel; and who developed a tool (PEMT) to identify patents using genes as a starting point, searching for compounds in ChEMBL with activity data for each gene and then searching SureChEMBL using the compounds found in ChEMBL. These, and similar studies, also confirm that extracting information from patents poses many challenges, given the length and complexity of these documents. Patent data are currently freely available from a number of resources, including Google Patents ( https://patents.google.com/ ), The Lens ( https://www.lens.org/ ), Espacenet ( https://www.epo.org/searching-for-patents/technical/espacenet.html ), Patentscope ( https://patentscope.wipo.int/search/en/search.jsf ) and Free Patents Online ( https://www.freepatentsonline.com/ ). All of these systems allow searching for patents using various criteria. Pubchem provides links to the Patentscope database from the World Intellectual Property Organization (WIPO) for more than 16 million compounds, which allows users to find the patents associated with each of these chemical structures . BindingDB includes a curated set of US granted patents, from which protein-compound activity data is extracted. In the work reported here, we use SureChEMBL ( https://www.surechembl.org/ ), which is a fully automated, chemical-structure-enabled database providing the research community with open and free access to the patent literature. Currently, SureChEMBL sources data from both patent applications and granted patents via full text patents from the United States Patent and Trademark Office (USPTO), the European Patent Office (EPO) and the World Intellectual Property Organization (WIPO), and titles and abstracts from the Japanese patent office (JPO). SureChEMBL currently contains ~140 million patents with ~50,000 added monthly. Of these, ~25 million patents are chemically annotated. Approximately 80,000 new compounds are extracted and added each month to the SureChEMBL chemistry database which currently contains more than 25 million unique structures. The pipeline for the extraction of chemical compounds from patents has been described in detail . In summary, chemical entity names, images and molfiles associated with each patent are converted into chemical structures and then registered into a structure-searchable database. This process is fully automated, without requiring any manual step or curation. The data in SureChEMBL can be accessed via a web interface that enables users to perform text and chemical structure queries, filter the output and then display the results. The complete set of chemical structures and patent associations is also available for download. The US National Institutes of Health established the Illuminating the Druggable Genome (henceforth IDG) project in 2014 ( https://commonfund.nih.gov/idg ), with the goal of increasing the knowledge about understudied proteins that belong to well-studied protein families, such as ion channels, G-protein coupled receptors (GPCR) and protein kinases. One of the key deliverables of the IDG project is an informatics platform, Pharos ( https://pharos.nih.gov/ ), that provides researchers with free access to relevant data on targets. An important aspect of the IDG project (and the data in Pharos) is the classification of human proteins into four target development level (TDL) families, based on how well studied these proteins are. In the Tclin category are targets of at least one approved drug; Tchem targets do not have approved drugs but are modulated by at least one small molecule with a potency above the cut-off specified for the target protein family ( [12pt]{minimal} }{}$$ ≤ 30 nM for kinases, [12pt]{minimal} }{}$$ ≤ 100 nM for GPCRs and nuclear receptors, [12pt]{minimal} }{}$$ ≤ 10 μM for ion channels and [12pt]{minimal} }{}$$ ≤ 1 μM for other target families); Tbio targets do not have chemistry qualifying for the Tclin/Tchem categories but satisfy the criteria described at http://juniper.health.unm.edu/tcrd/ ; while Tdark targets are understudied proteins with little annotation . Of particular relevance to the work here is the availability of small molecule modulators for new targets, consistent with other work suggesting that the lack of high-quality chemical probes for understudied targets is an important cause for lack of interest . The default IDG process uses bioactivity data from ChEMBL , as well as from Guide to Pharmacology, which contains manually curated information on ligands and drug targets , and DrugCentral , which contains bioactivity data annotated from a variety of sources, including the scientific literature. DrugCentral has also information on drugs that have been approved by the United States Food and Drug Administration, the European Medicines Agency, and the Pharmaceuticals and Medical Devices Agency in Japan. The DrugCentral drug information is used in the IDG workflow to identify the targets that belong to the Tclin category (TCRD Home Page, http://juniper.health.unm.edu/tcrd/ ). At the time of writing, Tclin proteins represent ~3% of the human proteome; Tchem proteins represent ~8%; Tbio proteins represent ~58%; and Tdark proteins represent ~31% . In this article, we describe methods to systematically mine the SureChEMBL patent corpus to identify new bioactivity data for Tdark/Tbio targets, with the aim of (1) including the bioactivity data in the ChEMBL database and (2) promoting some of these targets to IDG Tchem status. Patents were processed using perl scripts written for this project, accessible via a GitHub repository ( https://github.com/chembl/idg_patents_paper ). The starting point for our work was the set of patents extracted from SureChEMBL covering the years 2012 to 2018, flagged as life-science relevant according to the International Patent Classification (IPC) ( https://www.wipo.int/classifications/ipc/en/ ) or the Cooperative Patent Classification (CPC) codes ( https://worldwide.espacenet.com/classification ) present in the patents. These codes classify the patents into different areas of technology. The specific codes taken into account by the life science flag are: A01, A23, A24, A61, A62B, C05, C06, C07, C08, C09, C10, C11, C12, C13, C14, G01N, which cover a broader set of patents than required but is still useful to filter out many patents that would not be relevant. This resulted in a set of 3.7 million patents. The goal was to find patents with bioactivity data on small molecules against understudied targets (Tdark or Tbio categories according to the IDG classification). Firstly, in order to determine which patents were likely to have bioactivity data, the files corresponding to the patents were processed to identify tables containing the following keywords: IC50, XC50, EC50, AC50, Ki, Kd, pIC50, pXC50, pEC50, pAC50, −log(IC50), −log(XC50), −log(EC50), −log(AC50), concentration to inhibit, IC-50, XC-50, EC-50, AC-50, IC 50, XC 50, EC 50, AC 50. Out of the 3.7 million patents, 69,289 patents (2%) were thus flagged as potentially containing bioactivity data in tables (for simplicity called “patents with bioactivity tables”). Separately, we identified patents that might contain information about IDG Tbio and Tdark targets. A list of Tdark and Tbio IDG target names and gene symbols was obtained from the Target Central Resource Database (36,044 target names/symbols) (TCRD Home Page, http://juniper.health.unm.edu/tcrd/ ). We searched for these target names and their gene symbols in the patent titles, abstracts, descriptions and claims sections, in the context of specific phrases that could indicate bioactivity data of small molecules against them: X inhibitors Inhibitors of X X inhibitor Modulators of X Modulation of X Targeting X X modulators Binding specifically to X X mutants Inhibit X Antibodies recognis\|zing X Modulating the X Selective X inhibitors X antagonists X agonist X selective binding compounds Activity of X X antibodies X activity Inhibitor of X X binding Antibodies directed against X Treatment of X related Antibody for X Anti-X antibody Human anti-X Antibodies to X High X affinity Inhibiting X Blocks\|block X Blocking X Ligand\|ligands for X Compounds that interact with X Modulating the function of X X ligand\|ligands The combination of these two procedures allowed us to classify the patents into six groups: patents with bioactivity tables, and targets mentioned in titles or abstracts; patents with bioactivity tables, and targets mentioned in descriptions or claims sections; patents without bioactivity tables, and targets mentioned in titles or abstracts; patents without bioactivity tables, and targets mentioned in descriptions and claims; patents with bioactivity tables but no targets; and patents without bioactivity tables and without targets . This was done with the goal of prioritising the patents, with the expectation that most data would be found in Group 1, followed by Group 2; we expected Group 3 and Group 4 to contain fewer patents with bioactivity data (given that they were not flagged as containing bioactivity tables). The patents in Group 5 and Group 6 did not have target matches and for this reason were not expected to have bioactivity data against the understudied targets of interest to us. Following this automated annotation/filtering process, a number of patents from each group were manually examined to confirm the presence of the correct Tbio/Tdark target, the presence of quantitative bioactivity measurements, and that the Tbio/Tdark target was the molecular target to which these bioactivity measurements applied. This final check is required because some of the patents were found to have data only on targets that did not belong to the IDG list of understudied targets; other patents did contain data exclusively on the targets of particular interest to us. Some patents fell into both categories. For patents with confirmed bioactivity data, details of compounds synthesised, biological assays performed, and bioactivity measurements were manually extracted according to the standard ChEMBL curation procedure described previously and loaded into the ChEMBL database. Briefly, structures and names of all tested compounds were extracted, together with a description of the assays performed, name of the targets, species, and measurement values and units. Compound structures were standardised and integrated into ChEMBL, mapping them to an existing structure or creating a new entry in the database as appropriate. In addition to registering the reported measurement values, the bioactivity data obtained was standardised to facilitate comparison of results for common activity types. Bioactivity data was also mapped to existing ChEMBL targets according to species and sequence or accession. When this was not possible a new target was created and then mapped to the corresponding assay. All bioactivities against all the targets present in these patents (irrespective of their inclusion or not in the IDG Tbio/Tdark categories) were extracted by the curators. As a result of this work, bioactivity data from 225 patents were loaded into ChEMBL, corresponding to 657 targets (including single proteins, protein families, protein complexes, organisms, cell lines and protein-protein interactions) and 18,319 compounds. For 145 of these targets, this represents the only source of information of bioactivity data in ChEMBL. A patent family is a set of patents of identical content (European Patent Office, https://www.epo.org/searching-for-patents/helpful-resources/first-time-here/patent-families/docdb.html ). The scripts described here were run against every patent in SureChEMBL that belongs to the set of 2012–2018 patents flagged as life science related. In order to avoid duplication of effort, patents were grouped by patent family. For this reason, the patent counts in the sections below are given as number of patent families rather than number of patents. We examined the distribution of positive and negative patent families among the different groups shown in , to identify which group or groups were more or less likely to contain useful information, as this might facilitate the task of identifying the most useful patents for future analyses. The group that had the highest percentage of positive patents was Group 1: 49 positive patent families in the 291 families examined (16.8%), followed by Group 3: 88 positive families in the 1,912 examined (4.6%). There was one patent family in common between the positive patents of these two groups. Group 2 had 92 positive patent families, but 46 of them were already present in Group 1. Group 4 had 96 positive patent families, but 86 of them were already present in Group 3. There were very few patents with data in Group 5 or Group 6 (0.1% and 0.4% patent families of the examined ones, respectively) . A full list of patents and the targets they contain can be found in . Note that one patent is omitted from this list (US-8409550-B2) because it contains data against a target from Bos taurus , whereas IDG is focussed solely on human targets. A total of 76 of these targets had at least one compound with bioactivity data values within the cut-off for its target family, as defined by the target class-specific IDG criteria outlined earlier. shows which targets had bioactivity data within the cut-offs for its target family, and shows how many patents, total compounds and compounds within the cut-off were found for each IDG target class. As BindingDB also extracts data from patents, we were interested in examining the overlap between the two data sets. For all the targets found, we performed a search by target name in BindingDB with the goal of comparing the results from the two different databases. Because BindingDB extracts only US granted patents, we used the patent family identifier to do the comparison. We found 33 targets in both BindingDB and the dataset from our method. Of the 70 patent families found by our method for these targets, 20 were also found in BindingDB. A total of 50 families were found exclusively with our method, and 34 families were found exclusively by BindingDB. In most cases, the patents that were missed had targets mentioned using a name that was not on our list of targets to find (for example, “CH24H” instead of “Cholesterol 24-hydroxylase”). In other cases the patents belonged to Group 3 or Group 4 and were not part of the set of patents that we selected to read. Examples of understudied targets with bioactivities found in patents In this section we briefly describe three specific examples of targets for which we were able to identify and curate bioactivity data from the patent workflow described above. Some of the compounds found for each target are shown in . LATS1 LATS1 is a Ser/Thr kinase that belongs to the LATS (large tumor suppressor) family . It is a component of the Hipo pathway which is involved in cancer, organ development, growth and regeneration , and cell contact inhibition . This kinase is conserved among several organisms, such as yeast, nematodes, flies, and mammals. In humans, LATS1 can be found in high levels in most tissues, and it has a role in regulation of mitosis and apoptosis. In some types of cancer there is evidence of mutations in LATS1, and of LATS1 inactivation through promoter hypermethylation in others . It was found that the expression of this protein is elevated in some cancers, but decreased in others . Without considering data from patents, there are currently 430 molecules in ChEMBL associated with this target, extracted from 44 different articles, but only four molecules satisfy the IDG cut-off criteria for kinases. These molecules were extracted from four different articles: PMID 19654408 , PMID 22037378 , PMID 29191878 and PMID 30384048 . As a result of the current work, there are now 289 additional molecules associated with this target, with 184 molecules satisfying the IDG cut-off criteria to promote the target to the Tchem category. The source of these compounds is the patent US-20180344738-A1 , which describes molecules designed to promote cell proliferation, with applications such as chronic wound healing, promoting liver regrowth, or treating limbal stem cell deficiency. Histone-lysine N-methyltransferase SUV39H2 SUV39H2 is a lysine methyltransferase, first identified in Drosophila , which methylates histone 3 on lysine 9 (H3K9). Di- and trimethylation of H3K9 results in gene expression repression . This protein is present only in embryogenesis and adult testis of healthy individuals , but overexpressed in several cancers, for example lung adenocarcinoma, colorectal carcinoma and gastric carcinoma . There are no selective inhibitors for this target . At the start of this work there were 19 molecules in ChEMBL with bioactivity data against SUV39H2, all of them from scientific literature, but none of these molecules were within the IDG cut-off. Our patent workflow identified 460 molecules from just a single patent (US-20180273529-A1) , all within the corresponding IDG cut-off. G protein-coupled receptor 6 GPR6 is a G-protein coupled receptor, still classified as orphan by the International Union of Basic and Clinical Pharmacology (IUPHAR) due to lack of consistency among reports related to endogenous ligands . It is expressed mainly in neurons in mammalian striatum and hypothalamus. There is evidence that it could have a role in several processes and diseases such as neurite outgrowth, instrumental learning, Alzheimer’s disease, Parkinson’s disease, Huntington’s disease , and schizophrenia . At the start of this work, there were 227 molecules in ChEMBL with bioactivity data values within the IDG cut-off. These molecules were obtained from patents, either from BindingDB, or our own curation efforts previous to the work described here. As a result of this search, 100 additional molecules with bioactivity against GPR6 with values within the cut-off of ≤100 nM were identified, from patents WO-2018183145-A1 and EP-2882722-A1 . shows a timeline with patent and scientific literature numbers by year for GPR6, showing that in this particular case, significantly more data were reported via patent disclosures than in the scientific literature. Interestingly, a new clinical candidate (currently in phase 2 clinical trials) for Parkinson’s disease, CVN424 (a GPR6 inverse agonist), has been disclosed . This molecule can be found in patents as early as 2015 in patent US-9181249-B2 . In this section we briefly describe three specific examples of targets for which we were able to identify and curate bioactivity data from the patent workflow described above. Some of the compounds found for each target are shown in . LATS1 is a Ser/Thr kinase that belongs to the LATS (large tumor suppressor) family . It is a component of the Hipo pathway which is involved in cancer, organ development, growth and regeneration , and cell contact inhibition . This kinase is conserved among several organisms, such as yeast, nematodes, flies, and mammals. In humans, LATS1 can be found in high levels in most tissues, and it has a role in regulation of mitosis and apoptosis. In some types of cancer there is evidence of mutations in LATS1, and of LATS1 inactivation through promoter hypermethylation in others . It was found that the expression of this protein is elevated in some cancers, but decreased in others . Without considering data from patents, there are currently 430 molecules in ChEMBL associated with this target, extracted from 44 different articles, but only four molecules satisfy the IDG cut-off criteria for kinases. These molecules were extracted from four different articles: PMID 19654408 , PMID 22037378 , PMID 29191878 and PMID 30384048 . As a result of the current work, there are now 289 additional molecules associated with this target, with 184 molecules satisfying the IDG cut-off criteria to promote the target to the Tchem category. The source of these compounds is the patent US-20180344738-A1 , which describes molecules designed to promote cell proliferation, with applications such as chronic wound healing, promoting liver regrowth, or treating limbal stem cell deficiency. SUV39H2 is a lysine methyltransferase, first identified in Drosophila , which methylates histone 3 on lysine 9 (H3K9). Di- and trimethylation of H3K9 results in gene expression repression . This protein is present only in embryogenesis and adult testis of healthy individuals , but overexpressed in several cancers, for example lung adenocarcinoma, colorectal carcinoma and gastric carcinoma . There are no selective inhibitors for this target . At the start of this work there were 19 molecules in ChEMBL with bioactivity data against SUV39H2, all of them from scientific literature, but none of these molecules were within the IDG cut-off. Our patent workflow identified 460 molecules from just a single patent (US-20180273529-A1) , all within the corresponding IDG cut-off. GPR6 is a G-protein coupled receptor, still classified as orphan by the International Union of Basic and Clinical Pharmacology (IUPHAR) due to lack of consistency among reports related to endogenous ligands . It is expressed mainly in neurons in mammalian striatum and hypothalamus. There is evidence that it could have a role in several processes and diseases such as neurite outgrowth, instrumental learning, Alzheimer’s disease, Parkinson’s disease, Huntington’s disease , and schizophrenia . At the start of this work, there were 227 molecules in ChEMBL with bioactivity data values within the IDG cut-off. These molecules were obtained from patents, either from BindingDB, or our own curation efforts previous to the work described here. As a result of this search, 100 additional molecules with bioactivity against GPR6 with values within the cut-off of ≤100 nM were identified, from patents WO-2018183145-A1 and EP-2882722-A1 . shows a timeline with patent and scientific literature numbers by year for GPR6, showing that in this particular case, significantly more data were reported via patent disclosures than in the scientific literature. Interestingly, a new clinical candidate (currently in phase 2 clinical trials) for Parkinson’s disease, CVN424 (a GPR6 inverse agonist), has been disclosed . This molecule can be found in patents as early as 2015 in patent US-9181249-B2 . The overall goal of this work was to identify bioactivity data on understudied targets from the patent literature, which could allow us to promote targets to the IDG Tchem category. We focused on small molecules only, but our workflow also identified several patents concerning antibodies or RNA as therapeutic agents. For the purposes of this work, we did not progress these patents further, but clearly they could also be useful in the context of “illuminating” new targets. It should be noted that the work described here was conducted over a period of time, during which complementary data from other sources was added to the various resources concerned. This reflects the natural evolution of the underlying databases, each with their own update mechanism and release schedule procedures. Thus, for example, targets designated as Tdark or Tbio at the time when the research was initiated may have been promoted to a higher category based on separate evidence whilst the patent bioactivity work described here was underway. In the narrative below, the data reflects the particular snapshot corresponding to the time on which the work was initiated or completed, as appropriate. Two of the proteins included in the list of targets that we searched for (sclerostin and exportin-1) were originally classified as Tbio at the start of this work, but later promoted to Tclin after the approval of the first-in-class drugs romosozumab and selinexor respectively. Coincidentally, over the same period of time, bioactivity data for 30 Tbio/Tdark targets were added into ChEMBL from the scientific literature. There was an overlap of 21 targets between the two sets. This shows the value of using patents as an additional source of bioactivity data. The work described here involved manually reviewing many patents from the six groups described in . As expected, the group with higher percentage of positive patents was Group 1, unexpectedly followed by the patents in Group 3, which even though they could not be flagged as containing bioactivity tables, still contained bioactivity data that was not detected automatically with the method used here, and were only found when reviewing the patents manually. Groups 2 and 4 had lower percentages but still delivered useful and relevant patents. Even though classifying the patents in this way provided a starting point for prioritisation, clearly there are some limitations to this approach as shown in the comparison with BindingDB and the reasons for missing some patents, and for future work it would be advantageous to develop methods that can reduce the level of manual review that is required. This is the focus of currently ongoing work to develop machine-learning models able to predict which patents should be prioritised for human examination and potential curation. A total of 74 Tdark/Tbio targets were promoted to the Tchem category on the basis of the bioactivity data identified from our patent analysis. Using relatively simple annotation and filtering pipelines, we have been able to identify a substantial number of patents containing quantitative bioactivity data for understudied targets that had not previously been reported in the peer-reviewed medicinal chemistry literature. This underlines the potential value in searching the patent corpus in addition to the more traditional peer-reviewed literature. The small molecules found in these patents, together with their measured activity against the targets, are now accessible via the ChEMBL database and Pharos, and have contributed to the “illumination” of previously dark targets. 10.7717/peerj.15153/supp-1 Supplemental Information 1 Full list of targets and patents found in the search described in the article. *TDL: target development level at the start of this work. Click here for additional data file. 10.7717/peerj.15153/supp-2 Supplemental Information 2 List of targets with at least one compound within the cut-off for its target family, from patents found in this work. Click here for additional data file. 10.7717/peerj.15153/supp-3 Supplemental Information 3 Number of targets, patents and compounds per IDG target class. Click here for additional data file.
A practical guide to expert learner skills in the research environment	cfd51cd8-7fda-43c7-aa09-a3d932fba523	10162410	Anatomy[mh]	We all know researchers with so-called good (or golden) hands: those enviable people who can start with a new technique and begin making major discoveries almost immediately . Of course, there are many others who, at least initially, seem to lack good hands and struggle through countless failures for every one success. Many people who initially struggle in the lab will eventually develop into highly talented experimentalists, but the process can be arduous and can prove an insurmountable barrier to some. Entering graduate school, I lacked good hands. It seemed that I needed to make every mistake possible before I could get even the simplest experiment to work. The pain of each failed experiment was made all the worse by the seeming ease with which my colleagues could produce publication-quality findings on the first or second try. Thinking that I wasn’t cut out for science, I nearly left graduate school in my fourth year. Fortunately for me, I stayed, and with the help of several patient and caring lab mates, I slowly developed what many describe as good hands. In the years since I graduated, I’ve seen this same scenario repeat itself with new cohorts of researchers. I often noticed that the struggling researchers made the same types of mistakes that I had. It puzzled me that some researchers seemed to be able to foresee and avoid common missteps while others seemed destined to walk in my painful footsteps. Why were the potential missteps obvious to some people and not others? More importantly, was there anything I could do to help a struggling researcher beyond pointing out every possible misstep in each new experiment? Was there some way I could teach people to exit the struggling phase faster? A few years ago, I stumbled into educational psychology literature that suggested a different interpretation of the phenomenon of “good hands” and a path toward learning and teaching the skills needed to master new experimental techniques quickly. Reading it, I realized that the difference between those that we think possess good hands and those without was not some native and untrainable talent in those described as having good hands, but rather their learning mindset. Those researchers who pick up new skills seemingly effortlessly are “expert learners,” whereas those who experience repeated failures are “novice learners” . Expert learners identify potential roadblocks before they occur, and more importantly, they apply the insight they gain from one misstep to prevent missteps in the future and even in different spheres of experience. Fortunately, educational psychologists have developed tools to teach people how to become expert learners in the classroom . I have adapted these concepts to a practical approach to develop and teach expert learner skills in the research environment. Here, I share this approach, with the hope that learners and mentors can use it to bypass the painful process I and others have endured while developing expert learner skills. The approach focuses on three stages of learning a new skill—self-awareness, planning, and evaluation—in an iterative loop that builds the learner’s skills with each technique or experiment attempted . The stages are loosely based on concepts from the educational psychology topics of self-regulated learning and metacognition . Stage 1 is self-awareness: assessing your own strengths and, more importantly, weaknesses. This stage builds on the learner’s prior experience and, for those with little research experience, can draw from activities outside of research such as coursework, hobbies, sports, and jobs. The learner asks themself: When I was successful at learning a new technique or skill in the past, how did I do that? When I struggled in the past, what missteps contributed to those struggles? Are there trends in those successes and struggles that tell me about my strengths and weaknesses? For me, drawing from experiences learning to cook from cookbooks and TV shows, my self-awareness list would include that I learn well from watching, I tend to lose track of time, my handwriting can be illegible, and I get flustered if I have to work fast. I can apply this knowledge to an experiment, regardless of my prior exposure to the techniques needed for the experiment. Stage 2 is planning: planning out how you are going to learn and perform a new technique. Here I separate planning into two steps: how to learn a new technique and how to perform the actual experiment. First, the learner plans how they will learn a new technique. The learner asks themself: Can I shadow someone? Are there several protocols that I can compare to understand which steps are essential and which might be attuned to a specific situation? If there is a step that is technically very difficult, is there a way to practice it until I master it? Knowing myself, which of these approaches is best for me? Here the learner’s decisions are guided by the self-awareness stage’s reflection on past successes. For example, for me, based on my self-awareness, I will try to watch an expert at work and keep notes about how they do each step. Next, the learner plans the detailed steps of the experiment. The learner reads through the protocol and asks themself: Which steps could go wrong? How would I know if it went wrong? How much would it impact the final outcome? Knowing myself and my environment, how likely is it that this will happen? Are there mitigation strategies that I could add to the protocol? This planning stage is guided by the self-awareness stage’s reflection on prior struggles. For example, because I tend to lose track of time, I might ruin my sample by leaving it in an ionic detergent for too long when performing the permeabilization step during immunofluorescence. For me, the best mitigation strategy is to switch to a nonionic detergent that can be left on overnight without impacting the final result. Because my handwriting is illegible, I might confuse sample 1332 with 1337 and swap the tubes. Therefore, I try to avoid long sample IDs and use alternate short IDs such as 2 and 7 that I can write large and clearly. Of course, the mitigation strategies that I use are attuned to each protocol as well as the sorts of missteps I am likely to make. Each learner will need to develop their own self-oriented mitigation strategies, which rely on knowledge from the self-awareness stage. Stage 3 is evaluation: after completion, reflecting on the experiment. Here the learner seeks to convert their experience into knowledge about themself and the protocol. The learner asks themself: Did I really learn well from watching? Should I have spent more time reading about the technique? Did I encounter problems that I did not anticipate? Seeing the outcome, are there additional experimental samples or controls that I needed to include to detect failed steps? Did my mitigation strategies work? Did I need them, or did they add unnecessary steps, or, worse, did they add to the potential for error? Importantly, if something went wrong, are there new items to add to my self-awareness list or to consider in future planning stages? The evaluation stage is the key to developing expert learner skills. It turns each experiment into a mechanism to improve the learner’s approach to science. As a novice learner, I was unable to identify what might go wrong because I lacked experience in what could go wrong, and I lacked self-awareness about how I personally was most likely to make a misstep. However, most importantly, I lacked a framework for learning from prior experiments to predict future problems. In other words, I failed to include an evaluation stage in my approach. As a graduate student, I stumbled into one style of the evaluation stage. During my fourth year, when I was thinking that I was not cut out for graduate school, Keith Kozminski, a postdoc in the lab, suggested that I keep an “attempts” journal. Unlike a notebook, the attempts journal listed tasks that I hoped to accomplish in a week. Each week, I listed what I wanted to accomplish; then at the end of each week, I checked off what was and was not accomplished. Most importantly, Keith pushed me to figure out why I didn’t complete or was not successful with the experiments I had planned. This did two things. First, it helped me see common missteps in individual experiments, such as overincubation or switching tubes, which let me identify mitigation strategies that I could incorporate into individual protocols. Second, it helped me see workflow problems. For example, one way my experiments commonly failed was that I dropped my sample on the floor, ruining it. After some reflection, I realized this often occurred because experiments were running into one another. When two experiments hit critical steps at the same time, I would hurry, which was when I dropped things, or I would crowd things on my bench so that important samples were perched precariously and were easily knocked onto the floor. This realization led me to space out experiments on different days or with enough time to complete both without hurrying. Unsurprisingly, spacing out my experiments let me get more done in fewer hours. A novice learner could use the same journal approach that I used to identify their own mistakes and learn from them. This is a common tool in teaching expert learning skills in the classroom . However, the learner would still need to make those common initial mistakes for themselves. My approach provides the framework for a mentor to share their experience with common mistakes, perhaps preventing them, while still providing the learner the lifetime skills of learning from their own mistakes and predicting roadblocks before they occur. In my approach, when new people join the lab, I talk them through these three stages. We discuss common things that might be on a self-awareness list. I tell them about my common missteps and how I mitigate them. I then ask them to review a planned protocol on their own. I instruct them to think about how this experiment may be similar to a task they performed in the past, such as another experiment or, for people very new to the lab, a task in a lab class, hobby, or job. I ask them to reflect on how their experience with that activity informs how they should plan for this new experiment. I also ask them to think about each step and what might make that step go wrong. After they review the protocol independently, I meet with them to discuss what they think could go wrong. Here, I often point out potential missteps that the learner did not identify; this way they can learn from my mistakes or ones I’ve seen others make. For example, I might point out that at a centrifugation step, a sample is particularly sensitive to prolonged incubation and needs to be spun immediately. Therefore, if the centrifuge is in use by another user for several hours, the experiment will fail. I use this discussion to guide the novice learner to realize that it is important to identify time-sensitive steps and make sure all resources and equipment will be available. The goal here is that the learner will incorporate this into their knowledge base so that they will be prepared to look out for and plan time-sensitive steps without my guidance with future protocols. After the experiment, I ask the learner to reevaluate the protocol and their experience. I ask them to reflect on whether anything unexpected happened or if they saw new areas where potential missteps could occur. If so, are there mitigation methods they might incorporate in the future? After this independent evaluation, we meet to discuss their reevaluation. After the first couple of new skills, we skip the prestep and just focus on the postanalysis. When they are comfortable, we stop the postanalysis. The immediate impact of this approach is that it helps a learner skip wasteful rounds of failure by providing a framework for the mentor to share their experience about potential pitfalls. However, the main advantage of this approach is that it provides the learner with the framework to become an expert learner in any field. Importantly, it normalizes the healthy habit of constructive self-critique, which is useful in many spheres of personal and professional life. I feel that this approach is best when presented to each new member of the lab, regardless of their prior experience. This establishes the process as part of the common lab culture so that no learner feels singled out. In addition, it serves as a mechanism for training more expert learners on how to teach expert learning skills themselves. Importantly, adopting this approach and dispelling the myth of good hands may be particularly impactful for learners from groups historically underrepresented or marginalized in cell biology. When seen through the prism of good hands, an experiment’s failure or success can appear to a learner as an indication of their innate scientific talent regardless of their identity . Similarly, mentors who lack the framework of expert and novice learners may begin to see a learner who encounters frequent experimental failures as unteachable or poorly matched to the field. This scenario with both learner and mentor viewing a failed experiment as a referendum on the learner’s innate talents can easily initiate a downward spiral in the mentoring relationship for any learner. For learners from groups historically underrepresented or marginalized in cell biology, this can be even more harmful, because it may amplify imposter syndrome and dissuade the learner from seeking help at the moments when help could be most impactful . At the same time, mentors may find it harder to connect with these same learners to stop the spiral than they would with a learner with whom they have more in common. In contrast, when viewed through the prism of expert learning, both learner and mentor should come to see each failed experiment as a step toward mastery of an approach and an opportunity to engage in productive mentoring. Moreover, this approach should establish comfort with and a framework for constructive conversations that the learner and mentor can apply to strengthen the mentoring relationship across many spheres of training. Finally, I think it would be useful for cell biology programs to adopt curricula that apply this approach to teaching in undergraduate laboratory courses and first-year graduate courses. These expert learning skills are often called self-reflective or self-regulated learning skills. Self-reflective learning skills are correlated with success in undergraduate and graduate studies; however, less information is available about how to teach self-reflective learning skills in the laboratory setting . This is in contrast to undergraduate science lecture classes, for which ample literature reveals effective strategies for teaching self-reflective learning skills . Thus, I see great value for our field in encouraging studies to explore how to develop curricula that are effective in teaching self-reflective skills in the laboratory environment. Notably, although the field is still in its early days, educators across the country are working to develop and test tools for self-reflective learning for undergraduate researchers. I point readers to the EvaluateUR method ( https://serc.carleton.edu/evaluateur ) developed by a team at SUNY Buffalo with support from the NSF WIDER and ATE programs . A fully supported implementation of the method is available to institutions and programs through a subscription-based online service. It provides a platform for parallel student and mentor assessment on several student outcome categories reflecting desired student knowledge and skills. The method stresses the importance of assessing a learner’s academic strengths and weaknesses and emphasizes to both mentors and learners that the assessments are a starting point for building self-reflective learning skills. Although EvaluateUR was developed and is available now as a tool for independent research for undergraduates, the team is currently in the final stages of testing a similar tool for research experiences in undergraduate laboratory courses . If it is successfully adapted to undergraduate laboratory courses, the development of such expert learning skills could normalize constructive self-critique among learners before they enter the independent or graduate research environment and thus empower an entire generation of scientists. I am an associate professor in the Department of Cell and Developmental Biology at the University of Michigan. I graduated from the University of Washington with a B.S. in cell and molecular biology, and I earned a Ph.D. from the University of California at Berkeley in the same field. I trained in the Biological Chemistry Department at the University of California at Los Angeles and then was an assistant professor of biology at the University of North Carolina at Chapel Hill before moving to my current position in 2013. My research focuses on how membrane traffic contributes to cell and tissue organization. I am passionate about improving the way we teach how to do science and enjoy both informal and formal mentoring opportunities I encounter in an academic environment.
Etiology and outcome of penetrating keratoplasty in bullous keratopathy post-cataract surgery vs post-glaucoma surgery	fdb26a3e-029b-4d63-bbbc-d12af646fe16	10162548	Ophthalmology[mh]	Bullous keratopathy (BK), caused by corneal endothelial dysfunction, is characterized by corneal stromal edema and often associated with bullae of the corneal epithelium. The edema reduces the visual acuity (VA) and corneal epithelial bullae cause ocular pain. Thus, advanced cases of BK require treatment, and the definitive treatment is corneal transplantation to replace the dysfunctional corneal endothelium with healthy endothelium from a donor cornea. In fact, BK is one of the most common indications for corneal transplantation worldwide . The etiologies that cause corneal endothelial dysfunction are diverse, including surgical/laser trauma, endothelial dystrophy, infection and immune-mediated damage . Importantly, the causes of corneal endothelial dysfunction exhibit regional and chronological differences as indicated by a recent systematic review showing that reported indications for penetrating keratoplasty (PK) significantly vary . For example, whereas Fuchs’ endothelial corneal dystrophy (FECD) is the leading cause of visually significant corneal endothelial decompensation in western countries , it is rare in Asian countries, and an injury to the corneal endothelium during intraocular surgery is a more common cause of BK in an Asian eye having shallow anterior chamber (AC) and narrow angle . In this study, we investigated the predisposing conditions of BK in the Korean population between 2010 and 2020 and compared clinical characteristics of BK among different etiologies. Also, we comparatively analyzed the long-term results of PK in BK eyes associated with the top two etiologies: cataract surgery-associated BK (i.e. pseudophakic BK, PBK) and glaucoma surgery-associated BK (GBK). The study was approved by the Institutional Review Board (IRB) of Seoul National University Hospital (IRB No. 2020-122-924), and conducted according to the principles expressed in the Declaration of Helsinki. Medical records were retrospectively reviewed of patients diagnosed with BK at our tertiary referral center (Seoul National University Hospital) between 2010 and 2020. Due to the retrospective nature of the study, the IRB waived the requirement for obtaining informed consent from the patients. Patients who had previously received PK before BK (i.e. BK due to corneal graft failure) and those with a follow-up period of < 3 months were excluded from analysis. As a result, a total of 340 BK eyes from 326 patients were included and analyzed in the study. The following data were collected from medical charts: demographic information including age, gender and ethnicity/race, general medical history, ocular medical and surgical histories, the interval from a causative event to BK onset, ophthalmic findings including VA, intraocular pressure (IOP), lens status, endothelial cell density (ECD) and central corneal thickness (CCT), graft clarity at last follow-up (in cases with PK), and the follow-up period. The onset of BK was designated as the first date when the signs of BK were first observed in the medical record. The cause of BK was determined by two corneal specialists (Y.J. and J.Y.O.). Specifically, the diagnosis of PBK was made when BK developed after cataract extraction and intraocular lens (IOL) implantation in an eye with no evidence of corneal dystrophy or congenital anomaly and no history of glaucoma surgery/laser or ocular trauma. GBK was defined as BK that occurred following glaucoma surgery, such as glaucoma drainage device (GDD) implantation, trabeculectomy and trabeculotomy, irrespective of a history of cataract surgery. BK cases which received both cataract and glaucoma surgeries were designated as GBK. For analysis of PK outcome, patients with the postoperative follow-up of ≥ 6 months were included. As a result, 5 out of total 114 eyes with PK were excluded from the outcome analysis due to insufficient follow-up duration (two PBK eyes, two GBK eyes and one uveitis-induced BK). Graft failure was defined as an irreversible loss of corneal graft clarity despite intensive medical treatment according to the cause of failure. Immunologic rejection of corneal grafts was defined as the sudden onset of corneal edema in the presence of ocular inflammation. VA measured initially by Snellen charts was converted to logarithm of the minimum angle of resolution (logMAR) value for analysis. The improvement in VA was decided when best-corrected VA (BCVA) was improved by ≥ 0.3 logMAR values . Patients with amblyopia, advanced glaucoma, or retinal disease (macular edema, retinal detachment and exudative age-related macular degeneration) significantly affecting VA were excluded from the visual outcome analysis. Increased IOP (IIOP) was defined as IOP > 21 mmHg as measured by Goldmann applanation tonometry, rebound tonometry or non-contact tonometry, and then was adjusted for corneal factors, including CCT and corneal curvature, using correction formulas such as the Ehlers and Doughty formulas . Statistical analysis was performed using SPSS Statistics 20.0 (IBM, Armonk, NY), and graphs were made using GraphPad Prism 8.4.2 (GraphPad Software, San Diego, CA). The chi-square test or Fisher’s exact test was used for comparison of two categorical variables. Student’s t-test was applied for comparison of mean values of continuous variables between two groups. The Kaplan-Meier curves were used to estimate the median time to BK onset or corneal graft failure, and the log-rank test was conducted to compare the differences between Kaplan-Meier curves. Multivariate Cox proportional hazards regression was used to assess multiple potential risk factors associated with graft survival. Data were presented as mean ± SD. Differences were considered significant at p < 0.05. Causes of BK Totally, 340 BK eyes of 326 patients were included in the study. All were Koreans by ethnicity, including 126 women (38.7%) and 200 men (61.3%). The mean age at the time of BK diagnosis was 63.9 ± 14.9 years (1 ─ 87 years). Right eye was involved in 162 patients (49.7%), left eye in 150 patients (46.0%), and both eyes in 14 patients (4.3%). The etiology of BK, clinical characteristics and PK outcomes according to each etiology are summarized in and . The major predisposing condition leading to BK was intraocular surgery or laser. Totally, 238 eyes (70%) of 340 BK eyes were associated with surgery or laser; cataract surgery (n = 162), glaucoma surgery (n = 48), vitrectomy (n = 6) and laser iridotomy (n = 22). Cataract surgery was the most common cause of BK, responsible for 47.6% (n = 162) out of a total of 340 cases; the types of cataract surgeries included phacoemulsification (PE) (n = 109), extracapsular cataract extraction (ECCE) (n = 17), intracapsular cataract extraction (ICCE) (n = 22) and unknown (n = 14). Specifically, PBK accounted for 40.3% (n = 137), and aphakic BK (ABK) for 7.4% (n = 25). Among the 137 PBK eyes, 119 (86.9%) had a posterior chamber (PC) IOL, while 18 eyes (13.1%) had an AC IOL including implantable collamer lens (n = 2) and iris-claw IOL (n = 1). Additionally, 103 out of 137 PBK eyes occurred after PE, 13 after ECCE, and 13 after ICCE, while information on the type of cataract surgery was not available for the remaining 8 eyes. The second leading cause of BK was glaucoma surgery/laser, representing 20.6% (n = 70) of total BK cases (n = 340); 14.1% (n = 48) occurred following glaucoma surgery (GDD implantation in 43 eyes, trabeculectomy in 3 and trabeculotomy in 2), and the other 6.5% (n = 22) followed laser iridotomy. The third most common cause of BK was sterile or infectious inflammation, which accounted for 12.7% (n = 43) of total BK cases (n = 340); herpes simplex virus keratitis (n = 19), idiopathic anterior uveitis (n = 13), endophthalmitis (n = 4), cytomegalovirus endotheliitis (n = 3), toxic anterior segment syndrome (n = 2), varicella zoster virus endotheliitis (n = 1) and bacterial corneal ulcer (n = 1). FECD was found to be the cause of BK in 7.1% (n = 24) of total BK cases (n = 340), followed by trauma (4.1%, n = 14), iridocorneal endothelial (ICE) syndrome (3.2%, n = 11) and vitrectomy (1.7%, n = 6, 5 of which had silicone oil injection). Among 14 BK eyes secondary to trauma, 9 were inflicted by blunt trauma, and 5 by penetrating injury. Of 10 eyes (3.0%) classified as others, 5 were associated with high IOP, one with posterior polymorphous corneal dystrophy, one with retinopathy of prematurity, one with congenital glaucoma and one with chronic retinal detachment. Onset of BK Further, we investigated the time to BK onset since the causative incident except in eyes where the disease had insidious course as in FECD and ICE syndrome. The results are presented in and . The mean interval from the causative event to corneal edema was shortest in BK occurring after infection (1.8 ± 0.8 months) or vitrectomy (7.8 ± 4.4 months), followed by BK occurring after inflammation (herpes simplex virus keratitis, cytomegalovirus or varicella zoster virus endotheliitis, idiopathic anterior uveitis or toxic anterior segment syndrome) (37.0 ± 58.8 months). The onset time was longest in BK eyes associated with cataract surgery (160.7 ± 138.0 months); 148.4 ± 125.1 months for PBK and 226.0 ± 184.7 months for ABK. BK developed 85.0 ± 94.6 months after glaucoma surgery, 107.7 ± 96.9 months after laser iridotomy, and 111.0 ± 172.3 months after ocular trauma. Specifically, the mean time of BK onset was significantly shorter in GBK than in PBK (85.0 ± 94.6 months vs 148.4 ± 125.1 months, p < 0.001). Similarly, the median time to BK onset, as analyzed by Kaplan-Meier survival curves, was significantly shorter in GBK than in PBK (60 months vs 120 months, p = 0.003) . Outcome of PK Totally, PK was performed in 114 eyes (33.5%) among 340 BK eyes. Among 114 eyes, 5 eyes patients with < 6 months of the post-PK follow-up were excluded (two PBK eyes, two GBK eyes and one uveitis-induced BK), and the remaining 109 eyes were included for further outcome analysis. The overall graft success rate was 43.1% (47 out of 109 eyes) for 53.9 ± 29.9 months of post-PK follow-up as determined by graft clarity at last follow-up. The mean and median survival times to graft failure were 49.1 ± 4.1 months and 42 months, respectively. The graft survival rates and times according to BK etiologies are presented in . Comparison of PK outcome between PBK and GBK Our results indicated that cataract surgery and glaucoma surgery/laser were the leading causes of BK in our population, representing 47.7% and 20.6% of total 340 BK eyes, respectively . Similarly, PBK and GBK were identified to be the top two indications for PK in BK eyes, accounting for 44.7% and 12.3% of total 114 PK cases, respectively . Thus, we next analyzed and compared the outcomes of PK between PBK and GBK. displays demographical data, pre-PK clinical characteristics and post-PK outcomes in eyes with PBK (49 eyes of 49 patients) and GBK (12 eyes of 12 patients) who received PK for the treatment of BK and were followed-up for > 6 months after PK. Among the 49 PBK eyes, 35 had undergone PE, 7 had undergone ECCE, and 6 had undergone ICCE, while the type of cataract surgery was unknown for one eye. Also, 43 out of 49 PBK eyes had PC IOL implantation, while 6 eyes had AC IOL. Among the 12 GBK eyes, 11 had undergone GDD surgery, and one had undergone trabeculectomy. There were no significant differences in age, gender, laterality of an involved eye, general medical history, pre-PK VA, pre-PK IOP and follow-up duration between PBK and GBK patients . All eyes in both GBK and PBK groups had normal, well-controlled IOP before PK. The time from glaucoma surgery to GBK onset (60.8 ± 31.9 months) was significantly shorter than the time from cataract surgery to PBK (155.9 ± 128.3 in PBK, p < 0.001). The interval between surgery to PK was markedly shorter in GBK than in PBK (74.2 ± 33.7 months vs 162.0 ± 131.4 months, p < 0.001). After uneventful PK in all eyes, the mean survival time was shorter in GBK eyes (28.3 ± 5.9 months) than in PBK eyes (56.4 ± 6.6 months, p = 0.020) . Similarly, the median survival time of corneal allografts, as analyzed by Kaplan-Meier survival curves, was significantly shorter in GBK group (24.0 months), compared with PBK group (51.0 months, p = 0.020) ( and ). Specifically, the graft survival rates in GBK and PBK groups were 83.3% and 93.9% at 1 year ( p = 0.252), 36.4% and 80.0% at 2 years ( p = 0.008), and 30.0% and 57.6% at 3 years after PK ( p = 0.162), respectively . The most common cause of graft failure in GBK group was an immunologic endothelial rejection . The rejection occurred more frequently in GBK eyes (6 of 12 eyes, 50%) than in PBK eyes (7 of 49 eyes, 14.3%) ( p = 0.014). Chronic endothelial decompensation (in the absence of clinically evident rejection) was observed in in 2 of 12 (16.7%) GBK eyes and in 22 of 49 (44.9%) PBK eyes ( p = 0.003). Graft infection developed in 2 eyes (4.1%) of PBK group at 14 and 24 months after PK, respectively, while no infection was observed after PK in GBK group. Secondary glaucoma (IIOP after PK) occurred in 14 of 49 (28.6%) PBK eyes, of which 6 eyes required GDD implantation. IOP was well-controlled after PK in 11 of 12 (91.7%) GBK eyes except in one eye experiencing IIOP after PK. Otherwise, there were no complications such as wound leak, hypotony, choroidal detachment, endophthalmitis or retinal detachment in either GBK or PBK eyes. The improvement in BCVA was achieved after PK in most of GBK and PBK eyes . In line with annual graft survival rates, however, BCVA (logMAR) was significantly lower in GBK eyes than in PBK eyes at 1 and 3 years after PK (1.4 ± 0.7 vs 0.9 ± 0.6, p = 0.017 at 1 year; 1.8 ± 0.7 vs 1.1 ± 0.8, p = 0.043 at 3 years). ECD, as measured in corneal grafts by specular microscopy, were consistently lower during follow-up, while CCT thickness measured by pachymetry was consistently thicker, in GBK eyes compared to PBK eyes, but the differences did not reach statistical significance . Factors affecting PK outcome in PBK and GBK We went on to seek for risk factors affecting PK outcome in GBK and PBK eyes. Multivariate Cox proportional hazards regression models confirmed that pseudophakic eyes had lower relative risk of graft failure (0.20, 0.05 ─ 0.75, p = 0.017) . These results of multivariate analysis are consistent with the results of direct comparisons of graft survival times and rates between GBK and PBK . Among the postoperative factors, immunologic endothelial rejection (2.72, 1.22 ─ 6.06, p = 0.014) and graft infection (10.29, 2.14 ─ 49.33, p = 0.004) were significantly associated with higher relative risk of graft failure . Other factors including age, gender, the presence of GDD tube in the AC or post-PK IIOP did not significantly affect the graft survival. Totally, 340 BK eyes of 326 patients were included in the study. All were Koreans by ethnicity, including 126 women (38.7%) and 200 men (61.3%). The mean age at the time of BK diagnosis was 63.9 ± 14.9 years (1 ─ 87 years). Right eye was involved in 162 patients (49.7%), left eye in 150 patients (46.0%), and both eyes in 14 patients (4.3%). The etiology of BK, clinical characteristics and PK outcomes according to each etiology are summarized in and . The major predisposing condition leading to BK was intraocular surgery or laser. Totally, 238 eyes (70%) of 340 BK eyes were associated with surgery or laser; cataract surgery (n = 162), glaucoma surgery (n = 48), vitrectomy (n = 6) and laser iridotomy (n = 22). Cataract surgery was the most common cause of BK, responsible for 47.6% (n = 162) out of a total of 340 cases; the types of cataract surgeries included phacoemulsification (PE) (n = 109), extracapsular cataract extraction (ECCE) (n = 17), intracapsular cataract extraction (ICCE) (n = 22) and unknown (n = 14). Specifically, PBK accounted for 40.3% (n = 137), and aphakic BK (ABK) for 7.4% (n = 25). Among the 137 PBK eyes, 119 (86.9%) had a posterior chamber (PC) IOL, while 18 eyes (13.1%) had an AC IOL including implantable collamer lens (n = 2) and iris-claw IOL (n = 1). Additionally, 103 out of 137 PBK eyes occurred after PE, 13 after ECCE, and 13 after ICCE, while information on the type of cataract surgery was not available for the remaining 8 eyes. The second leading cause of BK was glaucoma surgery/laser, representing 20.6% (n = 70) of total BK cases (n = 340); 14.1% (n = 48) occurred following glaucoma surgery (GDD implantation in 43 eyes, trabeculectomy in 3 and trabeculotomy in 2), and the other 6.5% (n = 22) followed laser iridotomy. The third most common cause of BK was sterile or infectious inflammation, which accounted for 12.7% (n = 43) of total BK cases (n = 340); herpes simplex virus keratitis (n = 19), idiopathic anterior uveitis (n = 13), endophthalmitis (n = 4), cytomegalovirus endotheliitis (n = 3), toxic anterior segment syndrome (n = 2), varicella zoster virus endotheliitis (n = 1) and bacterial corneal ulcer (n = 1). FECD was found to be the cause of BK in 7.1% (n = 24) of total BK cases (n = 340), followed by trauma (4.1%, n = 14), iridocorneal endothelial (ICE) syndrome (3.2%, n = 11) and vitrectomy (1.7%, n = 6, 5 of which had silicone oil injection). Among 14 BK eyes secondary to trauma, 9 were inflicted by blunt trauma, and 5 by penetrating injury. Of 10 eyes (3.0%) classified as others, 5 were associated with high IOP, one with posterior polymorphous corneal dystrophy, one with retinopathy of prematurity, one with congenital glaucoma and one with chronic retinal detachment. Further, we investigated the time to BK onset since the causative incident except in eyes where the disease had insidious course as in FECD and ICE syndrome. The results are presented in and . The mean interval from the causative event to corneal edema was shortest in BK occurring after infection (1.8 ± 0.8 months) or vitrectomy (7.8 ± 4.4 months), followed by BK occurring after inflammation (herpes simplex virus keratitis, cytomegalovirus or varicella zoster virus endotheliitis, idiopathic anterior uveitis or toxic anterior segment syndrome) (37.0 ± 58.8 months). The onset time was longest in BK eyes associated with cataract surgery (160.7 ± 138.0 months); 148.4 ± 125.1 months for PBK and 226.0 ± 184.7 months for ABK. BK developed 85.0 ± 94.6 months after glaucoma surgery, 107.7 ± 96.9 months after laser iridotomy, and 111.0 ± 172.3 months after ocular trauma. Specifically, the mean time of BK onset was significantly shorter in GBK than in PBK (85.0 ± 94.6 months vs 148.4 ± 125.1 months, p < 0.001). Similarly, the median time to BK onset, as analyzed by Kaplan-Meier survival curves, was significantly shorter in GBK than in PBK (60 months vs 120 months, p = 0.003) . Totally, PK was performed in 114 eyes (33.5%) among 340 BK eyes. Among 114 eyes, 5 eyes patients with < 6 months of the post-PK follow-up were excluded (two PBK eyes, two GBK eyes and one uveitis-induced BK), and the remaining 109 eyes were included for further outcome analysis. The overall graft success rate was 43.1% (47 out of 109 eyes) for 53.9 ± 29.9 months of post-PK follow-up as determined by graft clarity at last follow-up. The mean and median survival times to graft failure were 49.1 ± 4.1 months and 42 months, respectively. The graft survival rates and times according to BK etiologies are presented in . Our results indicated that cataract surgery and glaucoma surgery/laser were the leading causes of BK in our population, representing 47.7% and 20.6% of total 340 BK eyes, respectively . Similarly, PBK and GBK were identified to be the top two indications for PK in BK eyes, accounting for 44.7% and 12.3% of total 114 PK cases, respectively . Thus, we next analyzed and compared the outcomes of PK between PBK and GBK. displays demographical data, pre-PK clinical characteristics and post-PK outcomes in eyes with PBK (49 eyes of 49 patients) and GBK (12 eyes of 12 patients) who received PK for the treatment of BK and were followed-up for > 6 months after PK. Among the 49 PBK eyes, 35 had undergone PE, 7 had undergone ECCE, and 6 had undergone ICCE, while the type of cataract surgery was unknown for one eye. Also, 43 out of 49 PBK eyes had PC IOL implantation, while 6 eyes had AC IOL. Among the 12 GBK eyes, 11 had undergone GDD surgery, and one had undergone trabeculectomy. There were no significant differences in age, gender, laterality of an involved eye, general medical history, pre-PK VA, pre-PK IOP and follow-up duration between PBK and GBK patients . All eyes in both GBK and PBK groups had normal, well-controlled IOP before PK. The time from glaucoma surgery to GBK onset (60.8 ± 31.9 months) was significantly shorter than the time from cataract surgery to PBK (155.9 ± 128.3 in PBK, p < 0.001). The interval between surgery to PK was markedly shorter in GBK than in PBK (74.2 ± 33.7 months vs 162.0 ± 131.4 months, p < 0.001). After uneventful PK in all eyes, the mean survival time was shorter in GBK eyes (28.3 ± 5.9 months) than in PBK eyes (56.4 ± 6.6 months, p = 0.020) . Similarly, the median survival time of corneal allografts, as analyzed by Kaplan-Meier survival curves, was significantly shorter in GBK group (24.0 months), compared with PBK group (51.0 months, p = 0.020) ( and ). Specifically, the graft survival rates in GBK and PBK groups were 83.3% and 93.9% at 1 year ( p = 0.252), 36.4% and 80.0% at 2 years ( p = 0.008), and 30.0% and 57.6% at 3 years after PK ( p = 0.162), respectively . The most common cause of graft failure in GBK group was an immunologic endothelial rejection . The rejection occurred more frequently in GBK eyes (6 of 12 eyes, 50%) than in PBK eyes (7 of 49 eyes, 14.3%) ( p = 0.014). Chronic endothelial decompensation (in the absence of clinically evident rejection) was observed in in 2 of 12 (16.7%) GBK eyes and in 22 of 49 (44.9%) PBK eyes ( p = 0.003). Graft infection developed in 2 eyes (4.1%) of PBK group at 14 and 24 months after PK, respectively, while no infection was observed after PK in GBK group. Secondary glaucoma (IIOP after PK) occurred in 14 of 49 (28.6%) PBK eyes, of which 6 eyes required GDD implantation. IOP was well-controlled after PK in 11 of 12 (91.7%) GBK eyes except in one eye experiencing IIOP after PK. Otherwise, there were no complications such as wound leak, hypotony, choroidal detachment, endophthalmitis or retinal detachment in either GBK or PBK eyes. The improvement in BCVA was achieved after PK in most of GBK and PBK eyes . In line with annual graft survival rates, however, BCVA (logMAR) was significantly lower in GBK eyes than in PBK eyes at 1 and 3 years after PK (1.4 ± 0.7 vs 0.9 ± 0.6, p = 0.017 at 1 year; 1.8 ± 0.7 vs 1.1 ± 0.8, p = 0.043 at 3 years). ECD, as measured in corneal grafts by specular microscopy, were consistently lower during follow-up, while CCT thickness measured by pachymetry was consistently thicker, in GBK eyes compared to PBK eyes, but the differences did not reach statistical significance . We went on to seek for risk factors affecting PK outcome in GBK and PBK eyes. Multivariate Cox proportional hazards regression models confirmed that pseudophakic eyes had lower relative risk of graft failure (0.20, 0.05 ─ 0.75, p = 0.017) . These results of multivariate analysis are consistent with the results of direct comparisons of graft survival times and rates between GBK and PBK . Among the postoperative factors, immunologic endothelial rejection (2.72, 1.22 ─ 6.06, p = 0.014) and graft infection (10.29, 2.14 ─ 49.33, p = 0.004) were significantly associated with higher relative risk of graft failure . Other factors including age, gender, the presence of GDD tube in the AC or post-PK IIOP did not significantly affect the graft survival. This study documented the causes of BK in the Korean population over the past 10 years and analyzed the outcomes of PK in patients with BK according to its etiology. In our series, cataract surgery and glaucoma surgery/laser were the most common conditions leading to BK, accounting for 47.7% and 20.6% of total BK cases, respectively. These results reflect the importance of intraocular surgeries in the development of BK in our population, and are in accordance with reports from other Asian countries such as China and Japan . The incidence of BK after cataract surgery is estimated to be 0.6% to 2% in patients undergoing the surgery , while the incidence of BK following glaucoma surgery/laser has not been reported. As the population ages, the incidence of cataract and glaucoma, which require surgical intervention, is increasing. Moreover, the number of glaucoma surgery is rising every year, and GDD implantation has become an important method for lowering of IOP in glaucoma patients . It can be presumed that BK associated with cataract and glaucoma surgeries will become more prevalent in the future. Therefore, it is important to understand the clinical characteristics and therapeutic outcomes of BK following cataract and glaucoma sugeries. In this study, we found that BK following cataract surgery is characterized by late onset as compared to BK associated with other etiologies. For example, PBK developed at the mean 148.4 months (the median 120 months) after surgery, whereas GBK developed earlier at the mean 85.0 months (the median 60 months) following glaucoma surgery. BK developed even faster at the mean 7.8 months following vitrectomy. During cataract surgery, the most common cause of post-operative corneal edema is surgical injury to the corneal endothelium induced by ultrasound energy, turbulence of the irrigating solution, ricocheting of nuclear fragments and contact with surgical instruments . Therefore, it should be considered that surgical trauma and IOL might be causes of corneal decompensation occurring many years after cataract surgery, especially in eyes with shallow chamber and narrow angle. Moreover, in GBK eyes, corneal endothelial cells were damaged by the long-standing glaucoma and additionally injured by glaucoma surgery, which might be a reason for faster onset of BK in GBK eyes compared to PBK only eyes. Given these, surgical technique modification, alongside intraoperative and postoperative care, would help to reduce the risk of corneal endothelial decompensation in patients following surgery. In the same vein, a recent introduction of minimally invasive glaucoma surgery, called MIGS , may contribute to further reduction of the risk of BK in patients following glaucoma surgery. Both univariate and multivariate analyses in our study revealed that GBK, compared to PBK, was associated with lower graft survival and poorer visual outcome after PK. These results are consistent with previous reports showing that the risk of corneal graft failure is significantly increased in glaucoma eyes with prior glaucoma surgery or using preoperative glaucoma medications . Another intriguing finding of our study is that graft failure was more prevalent in GBK despite well-controlled IOP; IOP was increased in 8% of GBK eyes after PK, whereas 29% of PBK eyes had IIOP after PK. Importantly, in GBK eyes, immunologic endothelial rejection was the most common cause of graft failure; 50.0% of GBK eyes had an immune rejection after PK, while 14.3% of PBK eyes had the rejection. This finding suggests that GDD-implanted eyes might be vulnerable to an immunologic endothelial rejection presumably through exposure of intraocular antigens to systemic immune system . Hence, our data emphasize the notion that immune reaction should be monitored cautiously and controlled effectively in GBK eyes during the post-PK follow-up for prevention of graft failure. There are several limitations in this study. First, there is a difference in the number of PBK and GBK eyes due to such a large difference in initial surgery volumes between cataract and glaucoma surgeries, rendering a direct comparison between the two groups difficult. Regardless, we believe that our data (incidence of BK accorging to etiology, onset time of BK from insulting incident, and corneal graft failure rates and their survival times according to BK etiologies) would provide useful information on the importance of intraocular surgery-associated BK and its outcome in Asian eyes. Second, due to the retrospective nature of the study, detailed information on clinical and biological factors such as ECD and IOL position was often missing in medical charts, posing a challenge to further analysis of various individual risk factors. Third, the GBK group included the eyes that underwent cataract surgery as we defined GBK as BK following glaucoma surgery regardless of a history of cataract surgery, while we designated PBK as cases developing BK after cataract extraction and IOL implantation without history of glaucoma surgery or laser. This was inevitable because the majority of GBK patients had undergone cataract surgery in the real world, as was the case in our own case series and in other studies . Thus, our results should be interpreted with caution, considering the possibility that glaucoma surgery may act as an aggravating factor in BK development rather than its isolated cause. Nevertheless, our data supports the notion that glaucoma surgery plays a critical role in BK development and PK failure. In conclusion, we herein present the recent trend in BK etiologies and PK indications in the Korean population, highlighting the importance of BK associated with intraocular surgery. Developing surgical techniques that minimally affect the corneal endothelium during and after surgeries would be beneficial in preventing BK development. Our data also revealed an earlier onset of BK following glaucoma surgery and poorer PK outcomes in GBK patients. This warrants future investigation into the pathogenesis of BK and corneal graft failure in eyes with glaucoma and glaucoma surgery. Additionally, improving corneal transplant surgical techniques and postoperative management can help enhance graft outcomes in BK eyes. With the new era of minimally invasive intraocular surgeries and endothelial keratoplasties, we can expect favorable changes in the incidence of BK and its therapeutic outcomes.
Crises information dissemination through social media in the UK and Saudi Arabia: A linguistic perspective	b71a65fa-3a50-4a24-8b90-c3bb22eddf38	10162563	Health Communication[mh]	Although social media offers a great deal of content exposure and engagement with users, managing health-related crises through Twitter, for example, can be rather difficult. In particular, it can be difficult for Health Officials (HOs) to use this social-media service to change or influence the public’s perceptions . Many communication researchers view social media as having both benefits and disadvantages in times of crisis . Regarding disadvantages, for many officials managing crises through Twitter is relatively unfamiliar, as there are few available guidelines for handling crises through social media . Further, another challenge concerning effectively disseminating health-related messages through Twitter is widespread health misinformation; this was particularly notable in the context of Coronavirus Disease 2019 (COVID-19) and its vaccines. During the COVID-19 Pandemic, an unprecedented amount of misinformation flooded social media, resulting in an infordemic . This included theories concerning the origin of the virus and claims about home remedies that could cure the disease; such theories may have contributed to vaccine hesitancy and opposition to vaccines. Officials are often viewed as key figures for not only providing the most accurate information, but also for changing misperceptions among the public. Thus, it was imperative for officials to use their social media accounts to debunk false information and advocate for informed health decisions; for example, by confronting anti-vaccine campaigns. To achieve desirable outcomes, HOs and governments sometimes apply strategies such as coercion, persuasion, and directing emotional responses . The COVID-19 Pandemic created a new global crisis, which led to a coalition between politicians and HOs to enforce novel health measures. Consequently, a previously under-investigated behavior became prominent: political speeches on health topics. There have been many studies of this behavior since the beginning of the pandemic. For example, transparency and legitimacy are useful strategies to help preserve people’s health and well-being , and some research argues that strategies employed by the Pakistani government helped in reducing the number of people who contracted the disease . Additionally, Semino investigated the metaphors used by politicians and the media when discussing COVID-19. She consequently found war metaphors to be common, and suggested that substitute metaphors should be used in their place. Meanwhile, Alyeksyeyeva et al. investigated the rhetoric of the Australian prime minister regarding COVID-19. They argued that: [r]hetoric plays an influential part in managing crisis, therefore political leaders frame the disease outbreak in war imagery and ‘fear-language’ to get the public involved in what they call ‘national interest’ and make the audience accept the leaders’ decisions, which enhances their authority (p.92). Rhetoric, therefore, plays an important role in directing public opinion and reducing crisis effects on the public and private sectors as well . In addition to rhetoric, the framing of ideas can represent “a matter of life and death” . Maani et al. argue that the manner by which officials frame a problem to the public can lead to support for certain therapies in place of other, far more urgent, therapies. Some tweets can be explicitly identified as formal requests or appeals to the public to perform specific actions. However, such tweets do not include any consideration of people’s agency regarding the actions in question. Recipients of the tweets are presented with a single solution, and are continuously directed to adhere to this recommendation or appeal (in the context of COVID-19, such recommendations/appeals are generally from health organizations or HOs). Therefore, we felt that consideration of speech acts would represent a good starting point for the present study, which focused on HOs’ COVID-19-related discourse on Twitter; this was because through analyzing such acts we could determine how HOs communicate to the public and seek to direct them towards desirable outcomes. The specific objective of the present study is to reveal how HOs from different cultural contexts differ linguistically in terms of their use of lexical items and rhetorical devices when tweeting crisis communication regarding the COVID-19 Pandemic. Specifically, this study sets its main research question as follows: what crisis-response communication strategies have been employed by HOs during unprecedented health crises? To achieve this objective, we investigated how HOs in Britain and Saudi Arabia used social media to assist their management of the pandemic, and how they urged the public to conform to the guidelines of the World Health Organization (WHO). We gathered tweets by both health officials within the time frame from January 2020 till March 2021. The Arabic tweets were analyzed in Arabic and later translated to English for the readers’ convenience. We analyzed their tweets from three perspectives–keyness, rhetoric, and metaphor. We begin by conducting a keyword analysis; we performed this in order to identify common topics discussed by the two HOs and to investigate similarities and differences in how they communicated these topics to the public. The second perspective involved a speech act analysis. The dataset used for this analysis comprised tweets that aimed to convince people to follow certain procedures. The officials’ argumentation style was investigated in this phase. The third perspective concerned the metaphors used by both HOs. Metaphor represents a major aspect of health communication and, thus, we felt it was necessary to include it. Metaphors influence how concepts are framed, and can affect how people respond to the content being communicated to them. The present paper is structured as follows: the next section presents a review of related literature; section two presents the data, along with the methodology employed in their collection and analysis; section three features a discussion of findings; and section four contains concluding remarks. The first angle from which this paper considers the data is the rhetoric use of speech acts by HOs to persuade the public to follow recommended steps. Speech Act Theory, which was first introduced by Austin and further developed by Searle , pertains to how utterances can comprise locutionary acts, illocutionary acts, and perlocutionary acts. Locutioanry acts refer to what is meant, illocutionary acts refer to what is done, and prelocutionary acts refer to what happens as a result of what is said . According to Searle , there are five types of illocutionary speech acts: assertive, such as informing and claiming; commissive, such as promising and swearing; directive, such as ordering, requesting and advice; declaratory, such as naming and dismissing; and expressive, such as complaining and congratulating. With later developments in Speech Act Theory, however, further acts were added, such as apology, wishes, and gratitude . Searle also differentiated between direct (literal) and indirect speech acts. In indirect speech acts a sentence form is used to convey a function for which it is not conventionally used . For example, an interrogative form is conventionally used for asking questions, but can be used indirectly for requests as in the sentence Could you turn out the lights ? (example is from ). Speech Act Theory focuses on how speakers use words to perform certain actions, and on the effect these words have on the listener. We follow the taxonomy of speech acts as presented by Searle to categorize the speech acts we find in the tweets corpus. We also adopt further classifications found in . shows the speech act taxonomy adopted here. We chose to consider the speech acts used by the two HOs in their tweets in order to determine the argumentation styles they used to lead their respective audiences towards adherence to the WHO guidelines. Argumentation on Twitter has previously been investigated by Elliott-Maksymowicz et al. , who found that Twitter-based arguments are often of a very simple nature; such arguments generally comprise a specific form of qualitative content that is defined in a series of logically connected assertions that result deductively in a conclusion. They were also able to show that several different types of speech acts (not just assertions) can be utilized to make argumentative points. Our investigation sheds light on the HOs’ use of social media, especially Twitter, in order to communicate health advice and directions for the public to guide them in time of crises. Therefore, we predict that the speech act of directives is used frequently in the tweets of HOs. However, leaders do not only direct people to what to do, they also try to lift their spirits (expressive speech acts), pass laws (declaration speech acts), and announce future plans (commissive speech acts). This study investigates the different speech acts found in the HOs’ tweets, and attempts an explanation for the ones not found there. The second angle we investigate is metaphors used in the data. Metaphors shape thinking and, in turn, language shapes how people think of a certain subject . Metaphor is defined as the use of language to discuss a certain thing in terms of something else. The two things in question are mostly different, but some similarities can be identified between them. These similarities are highlighted through metaphor. For example, when someone says “this is heaven” while eating a piece of cake, they highlight a similarity between the two, possibly pleasure . Metaphors appear frequently in language. Some research suggests that metaphors occur 3–18 times per 100 words . In addition to their frequency, metaphors are said to have a crucial role as a communication tool. Conceptional metaphor theory posits that conventional metaphoric expressions can be used to propose the existence of conceptional metaphors, where conceptional metaphors refer to systematic mappings between a target domain and a source domain. For example, the expression h e was filled with anger contains the conceptual metaphor that “the body is a container.” Lakoff and Johnson also state that conceptional metaphors are used to facilitate certain inferences and evaluations, which makes them important communication tools for explanation or persuasion. In terms of health-related communication, studies, particularly those focusing on AIDS, have shown the potency of language to shape the impact of epidemic disease . Wallis and Nerlich found that war metaphors were absent from UK media coverage of the 2003 SARS epidemic, but prevalent in discussions of cancer. However, the use of military metaphors in discussions of illnesses has been widely criticized. For example, Sontag states that such metaphors promote shame and guilt among patients, in addition to an excessive desire among policy-makers and politicians to exercise control over the disease. Therefore, there have been wide calls to remove all metaphors from health communication , or to replace them with “liberating” metaphors . Semino , for example, suggests the use of “firefighting” metaphors for COVID-19 prevention and cure. This extensive body of literature on metaphor in English is counteracted by a shortage of research on metaphor in Arabic. In general, research on Arabic health communication is scarce . The present study, therefore, represents a step towards bridging this gap. Corpus compilation This section comprises a description of the data used and of the methodology followed in the present study’s analysis. The data investigated in this study were taken from the Twitter accounts of a HO in Saudi Arabia and a HO in the UK. The collection and analysis method followed in this paper complies with the terms and conditions of Twitter. Twitter provides a platform for users to write short blogs and incorporate visual media with their text. Both officials added visuals to some of their tweets. Although we did not conduct a multimodal analysis of the visual data contained in the officials’ tweets, we nevertheless used the visual data to guide our interpretation of the tweets. To address the research question, a specialized corpus was compiled: the “Arabic and English health official online corpus” (AEHOO) which is used specifically for the purposes of this paper. The tweets were collected from the British and Saudi officials’ Twitter accounts for the period from January 2020 to the end of March 2021. The British HO at that time was Nadine Dorries (female) and the Saudi HO was Tawfiq Alrabiah (male). We did not observe any major differences regarding gender, so this variable was ignored. This is in line with previous literature who analyzed leaders’ speeches during COVID-19 and have not noticed gender as a crisis communication strategy variable (See ). During the period of data collection, we gathered 86 tweets from the Saudi HO (hereafter, “SHO”; six tweets were excluded because they were in English), and 200 from the British HO (hereafter, “BHO”). The English sub-corpus of the AEHOO contained 6,207 words and several hashtags, such as “#NHS,” “#covid,” “#Patientsafety,” “#CovidVaccination,” “#StayHome,” and “#Lockdown,” while the Arabic sub-corpus contained 3,677 words and hashtags such as “#weareallresponsible,” “#takethestep,” and “#coronavirus”. These hashtags were tweeted in Arabic. They have been translated here by the authors for the readers’ convenience. There is a notable difference in the number of words between the two sub-corpora. However, we do not have reason to believe that this can affect the results. We chose to constrain our data collection using a time frame rather than by number of words because this helps to uncover the different way both HOs employ their twitter accounts which serves the objective of this paper. The Twitter data were gathered manually, and were cleaned and converted into plaintext before the corpus was uploaded to Sketch Engine ; Sketch Engine is a user-friendly online platform that allows searches for linguistic patterns using corpus-based tools such as keywords, collocation, and concordance . By integrating the two sub-corpora, the AEHOO included tweets from both the BHO and SHO. Sketch engine: An online platform for processing multilingual corpora This study featured a bottom-up investigation of keywords in the AEHOO; this differs from top-down analysis, which requires a priori assumptions about a discourse’s lexical components . In our bottom-up analysis, we performed corpus-assisted discourse study, which involves comparing a corpus to a reference corpus to highlight relevant terms . In particular, keyword analysis was used in addition to the creation of conventional frequency lists . The keyword function of Sketch Engine was used to identify the keywords in the two sub-corpora. Furthermore, as both English web 2020 (enTenTen20) and Arabic web 2012 (arTenTen12), two standardized web text corpora, are available in Sketch Engine and have fundamental genre-related qualities, we used both as general reference corpora to yield keyword lists. The statistical significance of the identified keywords was determined using the logDice test, which ensured that their keyness was not due to random chance . These keywords were then semantically classified and tabulated. A keyword score of 100 was used as a cut-off point, and all words of that score and above were copied to an Excel sheet. Thirty-five keywords were identified in the English corpus and 138 in the Arabic one. There are two reasons for the difference in the number of keywords. First, the reference corpus used for the Arabic data was an excerpt of the arTenTen, which is not continuously updated, while the reference corpus for the English corpus is continuously updated; this increased the likelihood of the two English corpora being more similar. Second, linguistic idiosyncrasies of Arabic mean the same word might be used in numerous different forms. An example is the word “infected,” which appears in the keyword list for the Arabic corpus in three word forms; another example is the word “precautionary,” which appears in four word forms. After this stage, insights obtained from linguistic analytical approaches such as rhetorical analysis and metaphor analysis were applied based on the framework presented in the literature review. This section comprises a description of the data used and of the methodology followed in the present study’s analysis. The data investigated in this study were taken from the Twitter accounts of a HO in Saudi Arabia and a HO in the UK. The collection and analysis method followed in this paper complies with the terms and conditions of Twitter. Twitter provides a platform for users to write short blogs and incorporate visual media with their text. Both officials added visuals to some of their tweets. Although we did not conduct a multimodal analysis of the visual data contained in the officials’ tweets, we nevertheless used the visual data to guide our interpretation of the tweets. To address the research question, a specialized corpus was compiled: the “Arabic and English health official online corpus” (AEHOO) which is used specifically for the purposes of this paper. The tweets were collected from the British and Saudi officials’ Twitter accounts for the period from January 2020 to the end of March 2021. The British HO at that time was Nadine Dorries (female) and the Saudi HO was Tawfiq Alrabiah (male). We did not observe any major differences regarding gender, so this variable was ignored. This is in line with previous literature who analyzed leaders’ speeches during COVID-19 and have not noticed gender as a crisis communication strategy variable (See ). During the period of data collection, we gathered 86 tweets from the Saudi HO (hereafter, “SHO”; six tweets were excluded because they were in English), and 200 from the British HO (hereafter, “BHO”). The English sub-corpus of the AEHOO contained 6,207 words and several hashtags, such as “#NHS,” “#covid,” “#Patientsafety,” “#CovidVaccination,” “#StayHome,” and “#Lockdown,” while the Arabic sub-corpus contained 3,677 words and hashtags such as “#weareallresponsible,” “#takethestep,” and “#coronavirus”. These hashtags were tweeted in Arabic. They have been translated here by the authors for the readers’ convenience. There is a notable difference in the number of words between the two sub-corpora. However, we do not have reason to believe that this can affect the results. We chose to constrain our data collection using a time frame rather than by number of words because this helps to uncover the different way both HOs employ their twitter accounts which serves the objective of this paper. The Twitter data were gathered manually, and were cleaned and converted into plaintext before the corpus was uploaded to Sketch Engine ; Sketch Engine is a user-friendly online platform that allows searches for linguistic patterns using corpus-based tools such as keywords, collocation, and concordance . By integrating the two sub-corpora, the AEHOO included tweets from both the BHO and SHO. This study featured a bottom-up investigation of keywords in the AEHOO; this differs from top-down analysis, which requires a priori assumptions about a discourse’s lexical components . In our bottom-up analysis, we performed corpus-assisted discourse study, which involves comparing a corpus to a reference corpus to highlight relevant terms . In particular, keyword analysis was used in addition to the creation of conventional frequency lists . The keyword function of Sketch Engine was used to identify the keywords in the two sub-corpora. Furthermore, as both English web 2020 (enTenTen20) and Arabic web 2012 (arTenTen12), two standardized web text corpora, are available in Sketch Engine and have fundamental genre-related qualities, we used both as general reference corpora to yield keyword lists. The statistical significance of the identified keywords was determined using the logDice test, which ensured that their keyness was not due to random chance . These keywords were then semantically classified and tabulated. A keyword score of 100 was used as a cut-off point, and all words of that score and above were copied to an Excel sheet. Thirty-five keywords were identified in the English corpus and 138 in the Arabic one. There are two reasons for the difference in the number of keywords. First, the reference corpus used for the Arabic data was an excerpt of the arTenTen, which is not continuously updated, while the reference corpus for the English corpus is continuously updated; this increased the likelihood of the two English corpora being more similar. Second, linguistic idiosyncrasies of Arabic mean the same word might be used in numerous different forms. An example is the word “infected,” which appears in the keyword list for the Arabic corpus in three word forms; another example is the word “precautionary,” which appears in four word forms. After this stage, insights obtained from linguistic analytical approaches such as rhetorical analysis and metaphor analysis were applied based on the framework presented in the literature review. To obtain an adequate understanding of how the two HOs managed the COVID-19 pandemic, how they discussed it and presented it to their respective publics, and how they guided the people to perform desired behaviors, we applied a three-step process: we first conducted a lexical analysis to identify the common topics discussed by both HOs; the lexical analysis is presented in the following section. Second, we investigated the speech acts used in both corpora. The rationale behind exploring the speech acts was that these tweets were intended to educate the public about a new disease. The HOs wanted the public to perform certain behaviors in order to prevent the spread of this new disease. They wanted the public to act, so they communicated to them how they themselves were acting to combat the virus, and explicitly directed them through their language use. Therefore, Speech Act Theory lends itself well to this type of discourse. Finally, we discuss the metaphors used by both HOs when discussing the pandemic. Keyword analysis By classifying nouns semantically, we obtained the following results. First, because both officials talk about the same topic, both the BHO and the SHO communicated the same general messages regarding managing COVID-19, such as naming the virus and advocating the public adoption of necessary measures to minimize its spread. However, we did observe some divergences in focus between the two, which might reflect the different management strategies implemented in the two countries (see Tables and ). The BHO addressed specific issues relating to certain events, people, or places, while the SHO spoke more generally about health care and medical support for tackling the virus. Further, the BHO mentioned specific medical equipment and medical conditions and gave specific numbers regarding the financial requirements of projects developed to combat the pandemic, while the SHO did not tweet any such specificities, with the exception of his mentioning of Nujood Alkhaibari, a Saudi nurse who passed away as a result of COVID-19. Finally, attention to specific COVID-19-related strategies was more evident in the BHO’s tweets, with the BHO highlighting the emotional support that should be implemented to help others manage losses caused by COVID-19. Rhetorical analysis This section discusses the speech acts used by both HOs. This is considered a suitable analysis approach for the current study because Speech Act Theory has been previously used to investigate both health-care communication and political discourse . The corpus analyzed in the present paper comprises tweets by HOs that are directed at the public. It is true that the tweets did not comprise two-way conversations; however, there were some features that caused these tweets to represent conversation. For example, the data included interrogative forms and imperative forms, which can be used in conversation. Also, people can retweet and comment on tweets which can be regarded as two-way conversation. Further, contextualization of ideas was also present. When the HOs discussed a topic, they expected their audience to already be familiar with content they had mentioned previously. Examples of this contextualization appeared on a number of occasions, and was mainly represented through emojis and photos. The following tweet by the BHO is a good example of this: (1) That’s me done. (syringe emoji) Tweet (1) above was accompanied by a photo of the BHO receiving a COVID-19 vaccine. Both the emoji and the photo served as contextualization cues that helped the audience understand the text of the tweet. However, contextualization is beyond the scope of this paper and, therefore, will not be discussed further. In contrast, questions and imperatives are within our speech act analysis because they can be used (directly or indirectly) to preform directive and/or interrogative speech acts. shows the list of speech acts used in both HOs’ tweets. The assertion speech act was prominent in both data sets, as shows. The frequencies of the other speech acts varied across the two corpora. Speech acts are counted based on sentences not tweets. Therefore, the number of speech acts in this table is greater than the number of tweets. The different speech acts are discussed below in detail. Assertion Most of the tweets in our corpus were assertions. The strength of assertions is that the content being communicated is not only intended to inform the recipient, but also features an expectation that they will believe this information as true since it comes from a health official. While the “social” aspect of assertions has been challenged by Pagin , we follow the theory of Marsili and Green , who state that assertions can be social speech acts. Specifically, Marsili and Green argue that assertions are similar to other speech acts because they represent a social phenomenon that can only be understood by knowing the situation in which they are uttered and who the speaker and recipient are. Thus, the present work considers assertions by the HOs to be social acts because, by knowing the authors of the tweets (HOs), the audience was inclined to take them more seriously. Health information should be obtained from reliable sources, and HOs represent reliable sources for the public. Therefore, in our context, assertions represent a social act in which the HOs communicate information to the public and expect the public to believe this information and reject contrasting information from other sources. Examples of assertions are listed below. (2) My responsibility towards my country is important; therefore, I will keep adhering–God willing–to the precautionary measures to safeguard others from infection. (SHO) (3) All vaccines used in the Kingdom show high protection against the virus two weeks after administration, in addition to high safety. (SHO) Tweet (2) is accompanied by a drawing of a Saudi person to show that it is not the HO speaking. In this tweet, the SHO presents an argument intended to persuade the public. This argument contains an assertion featuring a first-person subject, which is intended to have the reader consciously or unconsciously adopt the argument as their own. The SHO asserts that one’s responsibility towards their country requires him/her to adhere to the precautionary measures recommended by their country’s government. This assertion asks the public to believe in this motto and adopt it as a measure to prevent infection. Tweet (3) is also an assertion, and is intended to reassure the public that vaccines are effective. This tweet features an expectation that the public will obtain vaccines as a result of this reassurance. The argumentative property of this tweet is more visible than that contained in Tweet (2). No quantitative estimate of safety evidence, such as exact numbers, was presented by the SHO. Instead, the SHO focused on health literacy as a communication strategy, as he only reported a qualitative estimate of safety; in this case, simple frequency (see for the difference between quantitative versus qalitative estimates of risks and safety). Similar tweets can also be found in BHO dataset. Consider the following: (4) Those who cite low daily diagnosed #COVID19 cases as a reason to exit lockdown now, miss the point. They are low, because of lockdown. (BHO) (5) My 84yo mum couldn’t have been happier to be called up for her #CovidVaccine today. (BHO) In tweet (4), the BHO corrects what she views as a misconception by people who are advocating an easing of lockdown measures. She explains that they were “miss[ing] the point” and, therefore, that the public should not believe them. In tweet (5), she asserts her mother’s state regarding taking the vaccine, which is part of an effort to encourage others to do the same. By mentioning a personal anecdote in a public discourse regarding health promotion (5), the BHO uses empathy as a communication strategy; this strategy is generally used to enhance credibility and foster effective communication . Expressive Expressive content reflects feelings and epistemic knowledge . According to Searle , expressive speech acts include apologies, thanks, whishes, and greetings. However, because expressive speech acts tend to be highly routinized , genuine ones are distinguished from ritual phrases. However, Weigand considered both as speech acts. In her taxonomy, sincere thanks and apologies are included under emotives and when they are used as formulaic expressions, they are included under declaratives . This dual role of expressive speech acts is supported by the use of thanks and apologies in Covid-19 signage . In this category, we include tweets containing verbs such as “know” and “love” in such content. These expressives constituted 7.3% of the tweets by the BHO and 0% of the tweets by the SHO. It is not surprising that this category represented such a small proportion of the tweets, as the HOs’ tweets had a general aim of educating the public about a new disease, and personal feelings were of secondary importance. The tweets that expressed epistemic knowledge generally followed the theme of “we do not know much about this disease.” Sentence (6) provides an example from this category in which an act performed by the BHO that fosters a certain emotion is expressed. (6) We learn more about this virus every day. (BHO) (7) It is such a huge privilege and pleasure to work with local authorities in the development of their local outbreak management plans. (BHO) On the other hand, we did not find any expressive tweets from the SHO; he did not mention any mental or emotional acts at all. Thus, there was a divergence between the BHO and the SHO in this regard; the BHO seemed to show more emotions than the SHO. This may partly reflect gender differences (the BHO was female while the SHO was male), and also indicate that the BHO applied a communication strategy of using emotion to show empathy (see ). Expressive speech acts also include thanks, apologies, greetings and wishes. In this data, there are no apology speech acts. Both ministers emphasize that they are following the best guidelines the present situation requires. The other speech acts, however, appear in varying frequencies. Greetings Greetings appear only three times in the BHO tweets. One difference between the two HOs is that the BHO interacts with people who praise or criticize her work. She quotes their tweets, greets them, and replies to what they say. Thanks Both HOs regularly made “thank you” remarks or similar expressions. Thanking expressions are reactive speech acts, which means they are a response to a past act by another . These expressions can generate solidarity between interlocutors or, as in our case, between HOs and the public or other members of the health-care system. (8) I thank all government sectors that acted quickly to elevate levels of readiness and complementarity, that secured information for travelers and flights, and that activated precautionary measures at land borders, seaports, and airports to protect our country from this virus. (SHO) (9) Thank you Sean, and your team for your collaborative approach and your determination to beat #Covid in Oldham. (BHO) (10) Never all those years ago could I have imagined what our nurses today have to deal with on a day to day basis. We are so grateful, thank you. (BHO) The above tweets show expressive thanking speech acts in which the HOs thank certain people for performing acts that helped slow the spread of the virus. However, thanking can also occur in advance, before an act is performed, such as when thanking people for their compliance with precautionary measures; this form of thanking can be seen in tweet (11). (11) Thank you for your commitment to wearing masks when out of your homes. (SHO) In this tweet, the SHO thanks the people for their commitment to wearing face coverings; this represents an encouragement to the public to continue following this advice. This, and similar instances, however, are counted as indirect directive speech acts because their purpose is to direct the people to do something rather than thanking them for something they did. Wishes The final category of expressive speech acts found in the two corpora is wishes. The optative mood appeared regularly in the SHO’s tweets (in 9.4% of the total tweets in the corpus). These tweets expressed a prayer or a wish. Meanwhile, there were only four tweets of this type in the BHO’s sub-corpus; two of these tweets involved the HO saying “good luck” to someone. The other two are listed below: (12) I wish @realDonaldTrump a good recovery from this ghastly #Covid virus. (13) Best wishes to all Alevel students in MidBeds and everywhere. In tweet (12), the BHO expresses a wish that the President of the United States of America will have a good recovery from COVID-19. In tweet (13), she provides encouragement to students who are taking exams during the crisis. The Arabic data included thirteen tweets that could be classified as wishes. These tweets included the following: (14) May God keep us all safe from any harm. (15) I wish you lasting health. Batanova suggests that well-wishing is culturally determined and that it plays a pivotal role in some cultures. Commissive Tweets labeled as commissives featured promises, vows, or intentions. This category comprised 7.2% of the SHO’s tweets and 5.6% of the BHO’s tweets. In tweets (16) and (17) below, the HOs promise the people that they will act in the public’s best interests. They vow to help the people end this pandemic. (16) Your safety and health are a priority to us. (SHO) (17) We must do all we can to prevent our ICUs #NHS from becoming overwhelmed. (BHO) Both the social aspect and argumentative aspect of speech acts are present in commissives. The HOs were addressing the public directly, inviting them to become active players in the attempt to overcome the crisis. The HOs were also assuring the public that they were important and that they could help. We observe here that both HOs used openness, frankness, and honesty as a communication strategy. This approach accords with recent research findings regarding the most effective communication strategies government officials can use to encourage people to perform certain behaviors during the COVID-19 Pandemic . Declaration Declaration speech acts bring about a change in the world. There are no instances of declaration speech acts in the data. The HOs do not use any performative verbs expressing illocutionary acts that change the state of affairs. We find the lack of declaratory acts surprising as the HO are expected to declare procedures to be followed by the public. This behavior by the HOs can be explained by their adherence to the suggestions announced by WHO. The strategy of openness and frankness as represented in commissive and declaratory speech acts appears in a small number of tweets. Declaration was not found in the tweets and commissives are used less than other speech acts such as assertions, thanking, directives, and even interrogatives (in BHO tweets). This might suggest that the HOs assumed their responsibility as leaders in persuading their respective audiences, directing them, and being empathetic with them. As for passing laws, they both followed the recommendation of WHO. Directive In a directive speech acts, orders are expressed. However, the imperative form of the verb is not the only verb type included in this category. Directives can vary in strength from mere invitations to direct orders and, thus, verbs such as “call for” and “encourage” are included in this category because, similar to the imperative, they can be used to ask the listener to act. Examples include the following: (14) Dear brothers and sisters, be very mindful of the preventive measures, we do not want you, your parents, or loved ones to be the next casualty. (SHO) (15) Do not forget our health-care heroes in your prayers. (SHO) (16) I recommend for everyone to use a face covering if they need to leave the house. (SHO) (17) …don’t touch the mask to lower it to speak over and wash after every use. (BHO) (18) Please follow the guidance. (BHO) (19) Be proud (British flag emoji). (BHO) Tweet (19) was accompanied by a map that showed that Britain was the country with the highest number of vaccines administered at that point in the pandemic. The speech act of directive is used repeatedly in the corpus. This is understandable as, to guide the public through the pandemic, the HOs were telling the people what they should do. This category comprised 14.5% of the BHO’s tweets and 38.9% of the SHO’s tweets. In these tweets, both HOs applied clear communication as their main communication strategy; this was in order to maintain order and reduce anxiety by emphasizing concrete action . However, there is a large difference between the way the SHO uses directives and the way the BHO uses them. First, the SHO uses directives more than the BHO. Even though the number of words in the Arabic tweets was smaller, the SHO used more directives than the BHO both directly and indirectly. This indicates a divergence in their communication strategies. The SHO used 34 direct directive speech acts while the BHO used 20 speech acts. This might indicate that the SHO views guidance of the public as his main goal while the BHO focuses more on supporting the public and values empathy. There are also instances of indirect directive speech acts. In these speech acts, the HOs use declarative statements as directives Examples include the following: (20) Important and urgent: If you suspect that you have Covid symptoms, you can easily check your health status using “personal assessment” in the app Mawid. (SHO) (21) We have the capacity and the ability, but we need people to come forward for testing in order to do the tests In these tweets, the HOs reduce the force of the directive by not using direct imperative forms to mitigate the effect they might have on the public. Interrogative An interrogative speech act demands a reply to a question. However, interrogatives can be used in various ways; for example, as rhetorical questions or discourse markers. There was a large difference between the two HOs regarding their use of interrogative content in their tweets. That is, they used questions differently. The SHO only tweeted one question, and this was a genuine request for information. (20) As a result of the precautionary measures, the number of seasonal flu cases has decreased by over 98% in the past three months when compared to the same period last year. What do you suggest we change in our customs to keep infection rates low once the pandemic has ended? (SHO) Here, the SHO is inviting people to suggest ways of lowering infection rates of seasonal flu in the future; the public could respond using Twitter’s “reply” functionality. Here, the HO was again encouraging people to be active participants during the crisis; he was inviting people to think and produce solutions. Meanwhile, the BHO did not use the interrogative structure to ask direct questions. Instead, she used it rhetorically. In political discourse, rhetorical questions are used in attempts to persuade the audience by appealing to their emotions . (21) …in fight against #COVID19 Remember when we were told that [because] we had left the EU we would be at the back of the queue for vaccines? We were at the front and have secured 40m of the Pfizer vac (BHO). (22) It’s over 60s who are at risk. How do you shield 13 m people? (BHO) In tweet (21), the interrogative is not used to question, but rather to remind the reader of an incident that occurred sometime before. This question is used as a discourse marker. The BHO wants to refute an argument and, therefore, foregrounds it to ensure that the reader knows what she is talking about. The interrogative in tweet (22) is used to appeal to the public’s emotions. She wants them to appreciate the gravity of the situation by asking them to think about a difficulty that she faces as a HO. The use of interrogatives as a rhetorical tactic serves an important function as an effective persuasive device (see, for example, ). The speech acts discussed in this section reflect the role each HO assumed. By conceptualizing the HOs’ language as actions, we were able to identify the patterns of persuasion and argumentation used by both. On some occasions, the HOs used direct imperative acts to tell the public what to do, on other occasions they appealed to their emotions by asking rhetorical questions, and on still more occasions they built solidarity with their respective audiences by using wishing and thanking acts. The speech act analysis also reflected the modes of communication employed by the two HOs. Although they applied generally similar acts, they utilized these acts differently. The BHO communicated her emotions more frequently than did the SHO. She also appealed to the emotions of her audience more frequently than he did. This clearly shows how the BHO used empathy as a communication strategy. On the other hand, the SHO acted as a channel for promoting health literacy among the public. Metaphor This section focuses on metaphorical references to the COVID-19 Pandemic in the HOs’ tweets. Analysis of metaphor use can identify how certain issues are viewed and are, in turn, framed. Metaphor use in health communication has been discussed by many researchers , who found that war metaphors are used frequently in this type of discourse. In war metaphors, diseases are perceived as an enemy, and patients are said to be fighting against the enemy. Thus, people talk of battles in which they/patients/the public must fight/defend against an enemy. This metaphor type was found in the BHO’s tweets, albeit rarely. Tweets with metaphoric references to the pandemic represented just 4% (eight tweets) of her entire corpus. Examples of these eight metaphoric references are listed below. (27) In fight against #COVID19 Remember when we were told that [because] we had left the EU we would be at the back of the queue for vaccines? We were at the front and have secured 40m of the Pfizer vac [syringe emoji]. (28) Thank you Sean, and your team for your collaborative approach and your determination to beat #Covid in Oldham. (29) Just finished a zoom call with our ten beacon councils and new advisory group who will be leading the way in the battle against #Covid19 at a local level. #WhackTheMole £300 million additional funding for local authorities to support new test and trace (30) Funding charities and helping those most affected by lockdown -BEAT helping YP eating disorders/bereavement/Every Mind Matters. We’ve also been fighting #COVID19 (31) Time to turn tables on #COVID19 If you have symptoms, you will be tested. If + we will trace your recent contacts who will be asked to self isolate for 14days. Local outbreaks will be handled by LAs to prevent a further national lockdown. We’re coming after you, #coronavirus The use of words such as “fight,” “soldiers,” “enemy,” “beat,” and “battle” in these tweets reflect the metaphor that the nation was in a war against the disease. While this metaphor gives the impression that the nation is united against a common enemy, which might represent strength, the use of war metaphors when referring to health issues has been widely criticized in the literature. For example, Hauser and Schwarz state that war metaphors can have a negative impact on patients and their families. One of the above tweets uses a seemingly different metaphor; namely, tweet (31), which includes the phrase “turn tables on” COVID-19. While this metaphor is not a war metaphor, it is similar to such metaphors in many ways. First, in both metaphor types the disease is viewed as a strong opponent; one that is formidable and capable of launching an attack. Another similarity between this metaphor and the war metaphor is the personification of the disease. The disease is treated almost like a human being who has power, influence, and is able to sit at a table just like us and, thus, we must turn the table on it. Another, more important similarity between the two metaphors is that they are both conflict-based metaphors. The phrase “turn tables on” originates from gaming discourse which, like war, is a conflict-rich discourse. Strong metaphors such as these conflict-based ones can be suitable for the topic in question. In political discourse, speakers might seek to utilize a range of tactics, including coercion, to achieve their goals ; the HOs’ goal was to induce certain reactions from the public, hence the use of conflict-based language. The Arabic sub-corpus does not contain the same personification of the disease. On the contrary, the disease is perceived as a hurdle; an inconvenience that we must eliminate or surpass. See the following examples: (32) To every person: take the vaccine; take the step, please. (33) We’ve spent a year in this pandemic, and this is our plan to get out of it. Take the step and begin your journey; take the vaccine. (34) I urge everyone to cooperate by following the health precautions. We are all in one boat; negligence from some affects all. (35) When communicating with others, being able to see your eyes is enough. We thank you for your commitment to wearing the face mask outside your home, because with everyone’s cooperation we will overcome the pandemic, God willing. In the above tweets, the SHO frames a situation in which all people are on a journey through life together, and the disease is presented as a hurdle that must be cleared, as an obstacle that has narrowed a path but will soon be negotiated, or as a rough sea through which the people must navigate their way. In other words, overcoming the pandemic is represented as a continuation of an original journey; the pandemic has slowed progress, and people must “take a step” to recommence the journey. The SHO’s tweets contain eight instances of metaphoric reference to the pandemic; they represent 13.5% of his entire corpus. Six of these tweets feature the journey-through-life metaphor discussed above. The remaining two tweets, however, feature a war metaphor. Both of these tweets featured the text translated below. (36) Staying at home is our strongest weapon–God willing–for overcoming COVID-19. On further inspection of this tweet, we found that the same text had been used by many Saudi government accounts across various social media networks and on official government websites. The same text was repeated in tweets by the Saudi Minister of Education, Minister of Foreign Affairs, and Minister of Finance. An English version of this tweet is also present on Twitter, meaning the Arabic version may have been a translation of an original English tweet. As this text may not have been composed by the SHO himself, we cannot count these two tweets as part of SHO’s use of metaphor; however, we can say that this metaphor may have arrived in the Arabic corpus through translation. The SHO may have used this metaphor as a strategy to convince people to adhere to regulations imposed to combat the pandemic. Politicians use various measures, such as strong metaphor, direct imperatives, and even coercion, to achieve certain reactions from the public . The introduction of the war metaphor to Arabic might be a manifestation of these measures. Notably, this metaphor has been introduced in Arabic at a time when it appears to be in decline in English, as highlighted by . By classifying nouns semantically, we obtained the following results. First, because both officials talk about the same topic, both the BHO and the SHO communicated the same general messages regarding managing COVID-19, such as naming the virus and advocating the public adoption of necessary measures to minimize its spread. However, we did observe some divergences in focus between the two, which might reflect the different management strategies implemented in the two countries (see Tables and ). The BHO addressed specific issues relating to certain events, people, or places, while the SHO spoke more generally about health care and medical support for tackling the virus. Further, the BHO mentioned specific medical equipment and medical conditions and gave specific numbers regarding the financial requirements of projects developed to combat the pandemic, while the SHO did not tweet any such specificities, with the exception of his mentioning of Nujood Alkhaibari, a Saudi nurse who passed away as a result of COVID-19. Finally, attention to specific COVID-19-related strategies was more evident in the BHO’s tweets, with the BHO highlighting the emotional support that should be implemented to help others manage losses caused by COVID-19. This section discusses the speech acts used by both HOs. This is considered a suitable analysis approach for the current study because Speech Act Theory has been previously used to investigate both health-care communication and political discourse . The corpus analyzed in the present paper comprises tweets by HOs that are directed at the public. It is true that the tweets did not comprise two-way conversations; however, there were some features that caused these tweets to represent conversation. For example, the data included interrogative forms and imperative forms, which can be used in conversation. Also, people can retweet and comment on tweets which can be regarded as two-way conversation. Further, contextualization of ideas was also present. When the HOs discussed a topic, they expected their audience to already be familiar with content they had mentioned previously. Examples of this contextualization appeared on a number of occasions, and was mainly represented through emojis and photos. The following tweet by the BHO is a good example of this: (1) That’s me done. (syringe emoji) Tweet (1) above was accompanied by a photo of the BHO receiving a COVID-19 vaccine. Both the emoji and the photo served as contextualization cues that helped the audience understand the text of the tweet. However, contextualization is beyond the scope of this paper and, therefore, will not be discussed further. In contrast, questions and imperatives are within our speech act analysis because they can be used (directly or indirectly) to preform directive and/or interrogative speech acts. shows the list of speech acts used in both HOs’ tweets. The assertion speech act was prominent in both data sets, as shows. The frequencies of the other speech acts varied across the two corpora. Speech acts are counted based on sentences not tweets. Therefore, the number of speech acts in this table is greater than the number of tweets. The different speech acts are discussed below in detail. Assertion Most of the tweets in our corpus were assertions. The strength of assertions is that the content being communicated is not only intended to inform the recipient, but also features an expectation that they will believe this information as true since it comes from a health official. While the “social” aspect of assertions has been challenged by Pagin , we follow the theory of Marsili and Green , who state that assertions can be social speech acts. Specifically, Marsili and Green argue that assertions are similar to other speech acts because they represent a social phenomenon that can only be understood by knowing the situation in which they are uttered and who the speaker and recipient are. Thus, the present work considers assertions by the HOs to be social acts because, by knowing the authors of the tweets (HOs), the audience was inclined to take them more seriously. Health information should be obtained from reliable sources, and HOs represent reliable sources for the public. Therefore, in our context, assertions represent a social act in which the HOs communicate information to the public and expect the public to believe this information and reject contrasting information from other sources. Examples of assertions are listed below. (2) My responsibility towards my country is important; therefore, I will keep adhering–God willing–to the precautionary measures to safeguard others from infection. (SHO) (3) All vaccines used in the Kingdom show high protection against the virus two weeks after administration, in addition to high safety. (SHO) Tweet (2) is accompanied by a drawing of a Saudi person to show that it is not the HO speaking. In this tweet, the SHO presents an argument intended to persuade the public. This argument contains an assertion featuring a first-person subject, which is intended to have the reader consciously or unconsciously adopt the argument as their own. The SHO asserts that one’s responsibility towards their country requires him/her to adhere to the precautionary measures recommended by their country’s government. This assertion asks the public to believe in this motto and adopt it as a measure to prevent infection. Tweet (3) is also an assertion, and is intended to reassure the public that vaccines are effective. This tweet features an expectation that the public will obtain vaccines as a result of this reassurance. The argumentative property of this tweet is more visible than that contained in Tweet (2). No quantitative estimate of safety evidence, such as exact numbers, was presented by the SHO. Instead, the SHO focused on health literacy as a communication strategy, as he only reported a qualitative estimate of safety; in this case, simple frequency (see for the difference between quantitative versus qalitative estimates of risks and safety). Similar tweets can also be found in BHO dataset. Consider the following: (4) Those who cite low daily diagnosed #COVID19 cases as a reason to exit lockdown now, miss the point. They are low, because of lockdown. (BHO) (5) My 84yo mum couldn’t have been happier to be called up for her #CovidVaccine today. (BHO) In tweet (4), the BHO corrects what she views as a misconception by people who are advocating an easing of lockdown measures. She explains that they were “miss[ing] the point” and, therefore, that the public should not believe them. In tweet (5), she asserts her mother’s state regarding taking the vaccine, which is part of an effort to encourage others to do the same. By mentioning a personal anecdote in a public discourse regarding health promotion (5), the BHO uses empathy as a communication strategy; this strategy is generally used to enhance credibility and foster effective communication . Expressive Expressive content reflects feelings and epistemic knowledge . According to Searle , expressive speech acts include apologies, thanks, whishes, and greetings. However, because expressive speech acts tend to be highly routinized , genuine ones are distinguished from ritual phrases. However, Weigand considered both as speech acts. In her taxonomy, sincere thanks and apologies are included under emotives and when they are used as formulaic expressions, they are included under declaratives . This dual role of expressive speech acts is supported by the use of thanks and apologies in Covid-19 signage . In this category, we include tweets containing verbs such as “know” and “love” in such content. These expressives constituted 7.3% of the tweets by the BHO and 0% of the tweets by the SHO. It is not surprising that this category represented such a small proportion of the tweets, as the HOs’ tweets had a general aim of educating the public about a new disease, and personal feelings were of secondary importance. The tweets that expressed epistemic knowledge generally followed the theme of “we do not know much about this disease.” Sentence (6) provides an example from this category in which an act performed by the BHO that fosters a certain emotion is expressed. (6) We learn more about this virus every day. (BHO) (7) It is such a huge privilege and pleasure to work with local authorities in the development of their local outbreak management plans. (BHO) On the other hand, we did not find any expressive tweets from the SHO; he did not mention any mental or emotional acts at all. Thus, there was a divergence between the BHO and the SHO in this regard; the BHO seemed to show more emotions than the SHO. This may partly reflect gender differences (the BHO was female while the SHO was male), and also indicate that the BHO applied a communication strategy of using emotion to show empathy (see ). Expressive speech acts also include thanks, apologies, greetings and wishes. In this data, there are no apology speech acts. Both ministers emphasize that they are following the best guidelines the present situation requires. The other speech acts, however, appear in varying frequencies. Greetings Greetings appear only three times in the BHO tweets. One difference between the two HOs is that the BHO interacts with people who praise or criticize her work. She quotes their tweets, greets them, and replies to what they say. Thanks Both HOs regularly made “thank you” remarks or similar expressions. Thanking expressions are reactive speech acts, which means they are a response to a past act by another . These expressions can generate solidarity between interlocutors or, as in our case, between HOs and the public or other members of the health-care system. (8) I thank all government sectors that acted quickly to elevate levels of readiness and complementarity, that secured information for travelers and flights, and that activated precautionary measures at land borders, seaports, and airports to protect our country from this virus. (SHO) (9) Thank you Sean, and your team for your collaborative approach and your determination to beat #Covid in Oldham. (BHO) (10) Never all those years ago could I have imagined what our nurses today have to deal with on a day to day basis. We are so grateful, thank you. (BHO) The above tweets show expressive thanking speech acts in which the HOs thank certain people for performing acts that helped slow the spread of the virus. However, thanking can also occur in advance, before an act is performed, such as when thanking people for their compliance with precautionary measures; this form of thanking can be seen in tweet (11). (11) Thank you for your commitment to wearing masks when out of your homes. (SHO) In this tweet, the SHO thanks the people for their commitment to wearing face coverings; this represents an encouragement to the public to continue following this advice. This, and similar instances, however, are counted as indirect directive speech acts because their purpose is to direct the people to do something rather than thanking them for something they did. Wishes The final category of expressive speech acts found in the two corpora is wishes. The optative mood appeared regularly in the SHO’s tweets (in 9.4% of the total tweets in the corpus). These tweets expressed a prayer or a wish. Meanwhile, there were only four tweets of this type in the BHO’s sub-corpus; two of these tweets involved the HO saying “good luck” to someone. The other two are listed below: (12) I wish @realDonaldTrump a good recovery from this ghastly #Covid virus. (13) Best wishes to all Alevel students in MidBeds and everywhere. In tweet (12), the BHO expresses a wish that the President of the United States of America will have a good recovery from COVID-19. In tweet (13), she provides encouragement to students who are taking exams during the crisis. The Arabic data included thirteen tweets that could be classified as wishes. These tweets included the following: (14) May God keep us all safe from any harm. (15) I wish you lasting health. Batanova suggests that well-wishing is culturally determined and that it plays a pivotal role in some cultures. Commissive Tweets labeled as commissives featured promises, vows, or intentions. This category comprised 7.2% of the SHO’s tweets and 5.6% of the BHO’s tweets. In tweets (16) and (17) below, the HOs promise the people that they will act in the public’s best interests. They vow to help the people end this pandemic. (16) Your safety and health are a priority to us. (SHO) (17) We must do all we can to prevent our ICUs #NHS from becoming overwhelmed. (BHO) Both the social aspect and argumentative aspect of speech acts are present in commissives. The HOs were addressing the public directly, inviting them to become active players in the attempt to overcome the crisis. The HOs were also assuring the public that they were important and that they could help. We observe here that both HOs used openness, frankness, and honesty as a communication strategy. This approach accords with recent research findings regarding the most effective communication strategies government officials can use to encourage people to perform certain behaviors during the COVID-19 Pandemic . Declaration Declaration speech acts bring about a change in the world. There are no instances of declaration speech acts in the data. The HOs do not use any performative verbs expressing illocutionary acts that change the state of affairs. We find the lack of declaratory acts surprising as the HO are expected to declare procedures to be followed by the public. This behavior by the HOs can be explained by their adherence to the suggestions announced by WHO. The strategy of openness and frankness as represented in commissive and declaratory speech acts appears in a small number of tweets. Declaration was not found in the tweets and commissives are used less than other speech acts such as assertions, thanking, directives, and even interrogatives (in BHO tweets). This might suggest that the HOs assumed their responsibility as leaders in persuading their respective audiences, directing them, and being empathetic with them. As for passing laws, they both followed the recommendation of WHO. Directive In a directive speech acts, orders are expressed. However, the imperative form of the verb is not the only verb type included in this category. Directives can vary in strength from mere invitations to direct orders and, thus, verbs such as “call for” and “encourage” are included in this category because, similar to the imperative, they can be used to ask the listener to act. Examples include the following: (14) Dear brothers and sisters, be very mindful of the preventive measures, we do not want you, your parents, or loved ones to be the next casualty. (SHO) (15) Do not forget our health-care heroes in your prayers. (SHO) (16) I recommend for everyone to use a face covering if they need to leave the house. (SHO) (17) …don’t touch the mask to lower it to speak over and wash after every use. (BHO) (18) Please follow the guidance. (BHO) (19) Be proud (British flag emoji). (BHO) Tweet (19) was accompanied by a map that showed that Britain was the country with the highest number of vaccines administered at that point in the pandemic. The speech act of directive is used repeatedly in the corpus. This is understandable as, to guide the public through the pandemic, the HOs were telling the people what they should do. This category comprised 14.5% of the BHO’s tweets and 38.9% of the SHO’s tweets. In these tweets, both HOs applied clear communication as their main communication strategy; this was in order to maintain order and reduce anxiety by emphasizing concrete action . However, there is a large difference between the way the SHO uses directives and the way the BHO uses them. First, the SHO uses directives more than the BHO. Even though the number of words in the Arabic tweets was smaller, the SHO used more directives than the BHO both directly and indirectly. This indicates a divergence in their communication strategies. The SHO used 34 direct directive speech acts while the BHO used 20 speech acts. This might indicate that the SHO views guidance of the public as his main goal while the BHO focuses more on supporting the public and values empathy. There are also instances of indirect directive speech acts. In these speech acts, the HOs use declarative statements as directives Examples include the following: (20) Important and urgent: If you suspect that you have Covid symptoms, you can easily check your health status using “personal assessment” in the app Mawid. (SHO) (21) We have the capacity and the ability, but we need people to come forward for testing in order to do the tests In these tweets, the HOs reduce the force of the directive by not using direct imperative forms to mitigate the effect they might have on the public. Interrogative An interrogative speech act demands a reply to a question. However, interrogatives can be used in various ways; for example, as rhetorical questions or discourse markers. There was a large difference between the two HOs regarding their use of interrogative content in their tweets. That is, they used questions differently. The SHO only tweeted one question, and this was a genuine request for information. (20) As a result of the precautionary measures, the number of seasonal flu cases has decreased by over 98% in the past three months when compared to the same period last year. What do you suggest we change in our customs to keep infection rates low once the pandemic has ended? (SHO) Here, the SHO is inviting people to suggest ways of lowering infection rates of seasonal flu in the future; the public could respond using Twitter’s “reply” functionality. Here, the HO was again encouraging people to be active participants during the crisis; he was inviting people to think and produce solutions. Meanwhile, the BHO did not use the interrogative structure to ask direct questions. Instead, she used it rhetorically. In political discourse, rhetorical questions are used in attempts to persuade the audience by appealing to their emotions . (21) …in fight against #COVID19 Remember when we were told that [because] we had left the EU we would be at the back of the queue for vaccines? We were at the front and have secured 40m of the Pfizer vac (BHO). (22) It’s over 60s who are at risk. How do you shield 13 m people? (BHO) In tweet (21), the interrogative is not used to question, but rather to remind the reader of an incident that occurred sometime before. This question is used as a discourse marker. The BHO wants to refute an argument and, therefore, foregrounds it to ensure that the reader knows what she is talking about. The interrogative in tweet (22) is used to appeal to the public’s emotions. She wants them to appreciate the gravity of the situation by asking them to think about a difficulty that she faces as a HO. The use of interrogatives as a rhetorical tactic serves an important function as an effective persuasive device (see, for example, ). The speech acts discussed in this section reflect the role each HO assumed. By conceptualizing the HOs’ language as actions, we were able to identify the patterns of persuasion and argumentation used by both. On some occasions, the HOs used direct imperative acts to tell the public what to do, on other occasions they appealed to their emotions by asking rhetorical questions, and on still more occasions they built solidarity with their respective audiences by using wishing and thanking acts. The speech act analysis also reflected the modes of communication employed by the two HOs. Although they applied generally similar acts, they utilized these acts differently. The BHO communicated her emotions more frequently than did the SHO. She also appealed to the emotions of her audience more frequently than he did. This clearly shows how the BHO used empathy as a communication strategy. On the other hand, the SHO acted as a channel for promoting health literacy among the public. Most of the tweets in our corpus were assertions. The strength of assertions is that the content being communicated is not only intended to inform the recipient, but also features an expectation that they will believe this information as true since it comes from a health official. While the “social” aspect of assertions has been challenged by Pagin , we follow the theory of Marsili and Green , who state that assertions can be social speech acts. Specifically, Marsili and Green argue that assertions are similar to other speech acts because they represent a social phenomenon that can only be understood by knowing the situation in which they are uttered and who the speaker and recipient are. Thus, the present work considers assertions by the HOs to be social acts because, by knowing the authors of the tweets (HOs), the audience was inclined to take them more seriously. Health information should be obtained from reliable sources, and HOs represent reliable sources for the public. Therefore, in our context, assertions represent a social act in which the HOs communicate information to the public and expect the public to believe this information and reject contrasting information from other sources. Examples of assertions are listed below. (2) My responsibility towards my country is important; therefore, I will keep adhering–God willing–to the precautionary measures to safeguard others from infection. (SHO) (3) All vaccines used in the Kingdom show high protection against the virus two weeks after administration, in addition to high safety. (SHO) Tweet (2) is accompanied by a drawing of a Saudi person to show that it is not the HO speaking. In this tweet, the SHO presents an argument intended to persuade the public. This argument contains an assertion featuring a first-person subject, which is intended to have the reader consciously or unconsciously adopt the argument as their own. The SHO asserts that one’s responsibility towards their country requires him/her to adhere to the precautionary measures recommended by their country’s government. This assertion asks the public to believe in this motto and adopt it as a measure to prevent infection. Tweet (3) is also an assertion, and is intended to reassure the public that vaccines are effective. This tweet features an expectation that the public will obtain vaccines as a result of this reassurance. The argumentative property of this tweet is more visible than that contained in Tweet (2). No quantitative estimate of safety evidence, such as exact numbers, was presented by the SHO. Instead, the SHO focused on health literacy as a communication strategy, as he only reported a qualitative estimate of safety; in this case, simple frequency (see for the difference between quantitative versus qalitative estimates of risks and safety). Similar tweets can also be found in BHO dataset. Consider the following: (4) Those who cite low daily diagnosed #COVID19 cases as a reason to exit lockdown now, miss the point. They are low, because of lockdown. (BHO) (5) My 84yo mum couldn’t have been happier to be called up for her #CovidVaccine today. (BHO) In tweet (4), the BHO corrects what she views as a misconception by people who are advocating an easing of lockdown measures. She explains that they were “miss[ing] the point” and, therefore, that the public should not believe them. In tweet (5), she asserts her mother’s state regarding taking the vaccine, which is part of an effort to encourage others to do the same. By mentioning a personal anecdote in a public discourse regarding health promotion (5), the BHO uses empathy as a communication strategy; this strategy is generally used to enhance credibility and foster effective communication . Expressive content reflects feelings and epistemic knowledge . According to Searle , expressive speech acts include apologies, thanks, whishes, and greetings. However, because expressive speech acts tend to be highly routinized , genuine ones are distinguished from ritual phrases. However, Weigand considered both as speech acts. In her taxonomy, sincere thanks and apologies are included under emotives and when they are used as formulaic expressions, they are included under declaratives . This dual role of expressive speech acts is supported by the use of thanks and apologies in Covid-19 signage . In this category, we include tweets containing verbs such as “know” and “love” in such content. These expressives constituted 7.3% of the tweets by the BHO and 0% of the tweets by the SHO. It is not surprising that this category represented such a small proportion of the tweets, as the HOs’ tweets had a general aim of educating the public about a new disease, and personal feelings were of secondary importance. The tweets that expressed epistemic knowledge generally followed the theme of “we do not know much about this disease.” Sentence (6) provides an example from this category in which an act performed by the BHO that fosters a certain emotion is expressed. (6) We learn more about this virus every day. (BHO) (7) It is such a huge privilege and pleasure to work with local authorities in the development of their local outbreak management plans. (BHO) On the other hand, we did not find any expressive tweets from the SHO; he did not mention any mental or emotional acts at all. Thus, there was a divergence between the BHO and the SHO in this regard; the BHO seemed to show more emotions than the SHO. This may partly reflect gender differences (the BHO was female while the SHO was male), and also indicate that the BHO applied a communication strategy of using emotion to show empathy (see ). Expressive speech acts also include thanks, apologies, greetings and wishes. In this data, there are no apology speech acts. Both ministers emphasize that they are following the best guidelines the present situation requires. The other speech acts, however, appear in varying frequencies. Greetings appear only three times in the BHO tweets. One difference between the two HOs is that the BHO interacts with people who praise or criticize her work. She quotes their tweets, greets them, and replies to what they say. Both HOs regularly made “thank you” remarks or similar expressions. Thanking expressions are reactive speech acts, which means they are a response to a past act by another . These expressions can generate solidarity between interlocutors or, as in our case, between HOs and the public or other members of the health-care system. (8) I thank all government sectors that acted quickly to elevate levels of readiness and complementarity, that secured information for travelers and flights, and that activated precautionary measures at land borders, seaports, and airports to protect our country from this virus. (SHO) (9) Thank you Sean, and your team for your collaborative approach and your determination to beat #Covid in Oldham. (BHO) (10) Never all those years ago could I have imagined what our nurses today have to deal with on a day to day basis. We are so grateful, thank you. (BHO) The above tweets show expressive thanking speech acts in which the HOs thank certain people for performing acts that helped slow the spread of the virus. However, thanking can also occur in advance, before an act is performed, such as when thanking people for their compliance with precautionary measures; this form of thanking can be seen in tweet (11). (11) Thank you for your commitment to wearing masks when out of your homes. (SHO) In this tweet, the SHO thanks the people for their commitment to wearing face coverings; this represents an encouragement to the public to continue following this advice. This, and similar instances, however, are counted as indirect directive speech acts because their purpose is to direct the people to do something rather than thanking them for something they did. The final category of expressive speech acts found in the two corpora is wishes. The optative mood appeared regularly in the SHO’s tweets (in 9.4% of the total tweets in the corpus). These tweets expressed a prayer or a wish. Meanwhile, there were only four tweets of this type in the BHO’s sub-corpus; two of these tweets involved the HO saying “good luck” to someone. The other two are listed below: (12) I wish @realDonaldTrump a good recovery from this ghastly #Covid virus. (13) Best wishes to all Alevel students in MidBeds and everywhere. In tweet (12), the BHO expresses a wish that the President of the United States of America will have a good recovery from COVID-19. In tweet (13), she provides encouragement to students who are taking exams during the crisis. The Arabic data included thirteen tweets that could be classified as wishes. These tweets included the following: (14) May God keep us all safe from any harm. (15) I wish you lasting health. Batanova suggests that well-wishing is culturally determined and that it plays a pivotal role in some cultures. Tweets labeled as commissives featured promises, vows, or intentions. This category comprised 7.2% of the SHO’s tweets and 5.6% of the BHO’s tweets. In tweets (16) and (17) below, the HOs promise the people that they will act in the public’s best interests. They vow to help the people end this pandemic. (16) Your safety and health are a priority to us. (SHO) (17) We must do all we can to prevent our ICUs #NHS from becoming overwhelmed. (BHO) Both the social aspect and argumentative aspect of speech acts are present in commissives. The HOs were addressing the public directly, inviting them to become active players in the attempt to overcome the crisis. The HOs were also assuring the public that they were important and that they could help. We observe here that both HOs used openness, frankness, and honesty as a communication strategy. This approach accords with recent research findings regarding the most effective communication strategies government officials can use to encourage people to perform certain behaviors during the COVID-19 Pandemic . Declaration speech acts bring about a change in the world. There are no instances of declaration speech acts in the data. The HOs do not use any performative verbs expressing illocutionary acts that change the state of affairs. We find the lack of declaratory acts surprising as the HO are expected to declare procedures to be followed by the public. This behavior by the HOs can be explained by their adherence to the suggestions announced by WHO. The strategy of openness and frankness as represented in commissive and declaratory speech acts appears in a small number of tweets. Declaration was not found in the tweets and commissives are used less than other speech acts such as assertions, thanking, directives, and even interrogatives (in BHO tweets). This might suggest that the HOs assumed their responsibility as leaders in persuading their respective audiences, directing them, and being empathetic with them. As for passing laws, they both followed the recommendation of WHO. In a directive speech acts, orders are expressed. However, the imperative form of the verb is not the only verb type included in this category. Directives can vary in strength from mere invitations to direct orders and, thus, verbs such as “call for” and “encourage” are included in this category because, similar to the imperative, they can be used to ask the listener to act. Examples include the following: (14) Dear brothers and sisters, be very mindful of the preventive measures, we do not want you, your parents, or loved ones to be the next casualty. (SHO) (15) Do not forget our health-care heroes in your prayers. (SHO) (16) I recommend for everyone to use a face covering if they need to leave the house. (SHO) (17) …don’t touch the mask to lower it to speak over and wash after every use. (BHO) (18) Please follow the guidance. (BHO) (19) Be proud (British flag emoji). (BHO) Tweet (19) was accompanied by a map that showed that Britain was the country with the highest number of vaccines administered at that point in the pandemic. The speech act of directive is used repeatedly in the corpus. This is understandable as, to guide the public through the pandemic, the HOs were telling the people what they should do. This category comprised 14.5% of the BHO’s tweets and 38.9% of the SHO’s tweets. In these tweets, both HOs applied clear communication as their main communication strategy; this was in order to maintain order and reduce anxiety by emphasizing concrete action . However, there is a large difference between the way the SHO uses directives and the way the BHO uses them. First, the SHO uses directives more than the BHO. Even though the number of words in the Arabic tweets was smaller, the SHO used more directives than the BHO both directly and indirectly. This indicates a divergence in their communication strategies. The SHO used 34 direct directive speech acts while the BHO used 20 speech acts. This might indicate that the SHO views guidance of the public as his main goal while the BHO focuses more on supporting the public and values empathy. There are also instances of indirect directive speech acts. In these speech acts, the HOs use declarative statements as directives Examples include the following: (20) Important and urgent: If you suspect that you have Covid symptoms, you can easily check your health status using “personal assessment” in the app Mawid. (SHO) (21) We have the capacity and the ability, but we need people to come forward for testing in order to do the tests In these tweets, the HOs reduce the force of the directive by not using direct imperative forms to mitigate the effect they might have on the public. An interrogative speech act demands a reply to a question. However, interrogatives can be used in various ways; for example, as rhetorical questions or discourse markers. There was a large difference between the two HOs regarding their use of interrogative content in their tweets. That is, they used questions differently. The SHO only tweeted one question, and this was a genuine request for information. (20) As a result of the precautionary measures, the number of seasonal flu cases has decreased by over 98% in the past three months when compared to the same period last year. What do you suggest we change in our customs to keep infection rates low once the pandemic has ended? (SHO) Here, the SHO is inviting people to suggest ways of lowering infection rates of seasonal flu in the future; the public could respond using Twitter’s “reply” functionality. Here, the HO was again encouraging people to be active participants during the crisis; he was inviting people to think and produce solutions. Meanwhile, the BHO did not use the interrogative structure to ask direct questions. Instead, she used it rhetorically. In political discourse, rhetorical questions are used in attempts to persuade the audience by appealing to their emotions . (21) …in fight against #COVID19 Remember when we were told that [because] we had left the EU we would be at the back of the queue for vaccines? We were at the front and have secured 40m of the Pfizer vac (BHO). (22) It’s over 60s who are at risk. How do you shield 13 m people? (BHO) In tweet (21), the interrogative is not used to question, but rather to remind the reader of an incident that occurred sometime before. This question is used as a discourse marker. The BHO wants to refute an argument and, therefore, foregrounds it to ensure that the reader knows what she is talking about. The interrogative in tweet (22) is used to appeal to the public’s emotions. She wants them to appreciate the gravity of the situation by asking them to think about a difficulty that she faces as a HO. The use of interrogatives as a rhetorical tactic serves an important function as an effective persuasive device (see, for example, ). The speech acts discussed in this section reflect the role each HO assumed. By conceptualizing the HOs’ language as actions, we were able to identify the patterns of persuasion and argumentation used by both. On some occasions, the HOs used direct imperative acts to tell the public what to do, on other occasions they appealed to their emotions by asking rhetorical questions, and on still more occasions they built solidarity with their respective audiences by using wishing and thanking acts. The speech act analysis also reflected the modes of communication employed by the two HOs. Although they applied generally similar acts, they utilized these acts differently. The BHO communicated her emotions more frequently than did the SHO. She also appealed to the emotions of her audience more frequently than he did. This clearly shows how the BHO used empathy as a communication strategy. On the other hand, the SHO acted as a channel for promoting health literacy among the public. This section focuses on metaphorical references to the COVID-19 Pandemic in the HOs’ tweets. Analysis of metaphor use can identify how certain issues are viewed and are, in turn, framed. Metaphor use in health communication has been discussed by many researchers , who found that war metaphors are used frequently in this type of discourse. In war metaphors, diseases are perceived as an enemy, and patients are said to be fighting against the enemy. Thus, people talk of battles in which they/patients/the public must fight/defend against an enemy. This metaphor type was found in the BHO’s tweets, albeit rarely. Tweets with metaphoric references to the pandemic represented just 4% (eight tweets) of her entire corpus. Examples of these eight metaphoric references are listed below. (27) In fight against #COVID19 Remember when we were told that [because] we had left the EU we would be at the back of the queue for vaccines? We were at the front and have secured 40m of the Pfizer vac [syringe emoji]. (28) Thank you Sean, and your team for your collaborative approach and your determination to beat #Covid in Oldham. (29) Just finished a zoom call with our ten beacon councils and new advisory group who will be leading the way in the battle against #Covid19 at a local level. #WhackTheMole £300 million additional funding for local authorities to support new test and trace (30) Funding charities and helping those most affected by lockdown -BEAT helping YP eating disorders/bereavement/Every Mind Matters. We’ve also been fighting #COVID19 (31) Time to turn tables on #COVID19 If you have symptoms, you will be tested. If + we will trace your recent contacts who will be asked to self isolate for 14days. Local outbreaks will be handled by LAs to prevent a further national lockdown. We’re coming after you, #coronavirus The use of words such as “fight,” “soldiers,” “enemy,” “beat,” and “battle” in these tweets reflect the metaphor that the nation was in a war against the disease. While this metaphor gives the impression that the nation is united against a common enemy, which might represent strength, the use of war metaphors when referring to health issues has been widely criticized in the literature. For example, Hauser and Schwarz state that war metaphors can have a negative impact on patients and their families. One of the above tweets uses a seemingly different metaphor; namely, tweet (31), which includes the phrase “turn tables on” COVID-19. While this metaphor is not a war metaphor, it is similar to such metaphors in many ways. First, in both metaphor types the disease is viewed as a strong opponent; one that is formidable and capable of launching an attack. Another similarity between this metaphor and the war metaphor is the personification of the disease. The disease is treated almost like a human being who has power, influence, and is able to sit at a table just like us and, thus, we must turn the table on it. Another, more important similarity between the two metaphors is that they are both conflict-based metaphors. The phrase “turn tables on” originates from gaming discourse which, like war, is a conflict-rich discourse. Strong metaphors such as these conflict-based ones can be suitable for the topic in question. In political discourse, speakers might seek to utilize a range of tactics, including coercion, to achieve their goals ; the HOs’ goal was to induce certain reactions from the public, hence the use of conflict-based language. The Arabic sub-corpus does not contain the same personification of the disease. On the contrary, the disease is perceived as a hurdle; an inconvenience that we must eliminate or surpass. See the following examples: (32) To every person: take the vaccine; take the step, please. (33) We’ve spent a year in this pandemic, and this is our plan to get out of it. Take the step and begin your journey; take the vaccine. (34) I urge everyone to cooperate by following the health precautions. We are all in one boat; negligence from some affects all. (35) When communicating with others, being able to see your eyes is enough. We thank you for your commitment to wearing the face mask outside your home, because with everyone’s cooperation we will overcome the pandemic, God willing. In the above tweets, the SHO frames a situation in which all people are on a journey through life together, and the disease is presented as a hurdle that must be cleared, as an obstacle that has narrowed a path but will soon be negotiated, or as a rough sea through which the people must navigate their way. In other words, overcoming the pandemic is represented as a continuation of an original journey; the pandemic has slowed progress, and people must “take a step” to recommence the journey. The SHO’s tweets contain eight instances of metaphoric reference to the pandemic; they represent 13.5% of his entire corpus. Six of these tweets feature the journey-through-life metaphor discussed above. The remaining two tweets, however, feature a war metaphor. Both of these tweets featured the text translated below. (36) Staying at home is our strongest weapon–God willing–for overcoming COVID-19. On further inspection of this tweet, we found that the same text had been used by many Saudi government accounts across various social media networks and on official government websites. The same text was repeated in tweets by the Saudi Minister of Education, Minister of Foreign Affairs, and Minister of Finance. An English version of this tweet is also present on Twitter, meaning the Arabic version may have been a translation of an original English tweet. As this text may not have been composed by the SHO himself, we cannot count these two tweets as part of SHO’s use of metaphor; however, we can say that this metaphor may have arrived in the Arabic corpus through translation. The SHO may have used this metaphor as a strategy to convince people to adhere to regulations imposed to combat the pandemic. Politicians use various measures, such as strong metaphor, direct imperatives, and even coercion, to achieve certain reactions from the public . The introduction of the war metaphor to Arabic might be a manifestation of these measures. Notably, this metaphor has been introduced in Arabic at a time when it appears to be in decline in English, as highlighted by . This study comprised an investigation of the crisis-response communication strategies applied by two HOs–specifically, how both sought to manage the COVID-19 crisis and communicate to the public the procedures that should be followed during the pandemic to mitigate the effects of this novel health hazard. It is important to compare the strategies of health officials in different countries especially countries that differ culturally and linguistically in order to identify the best procedures for information dissemination during a crisis. A global pandemic required global co-operation and this study provides an in-depth view of how speech acts and metaphor are used in Arabic and how this use compares to how they are used in English. However, the study is limited by the number of tweets each HO chose to write. The SHO did not produce many tweets as the SHO did, so another study may compliment the findings of this one. Also, Investigation of other cultures might reveal different results, which suggests the importance of conducting similar studies based on data from different languages. We found that both HOs in Saudi Arabia and Britain utilized communication strategies that are suggested by previous research for good crisis information dissemination. For example, Su et al. suggest that health officials need to “develop fact-based, transparent, and accountable messaging.” Both BHO and SHO used their Twitter accounts to reach out to the public. Their accounts served as a method to send messages to the public that are fact-based and accountable. They used persuasive tactics using a variety of speech acts. Also, both HOs communicated to their respective audiences the procedures suggested by the WHO which shows that they followed another strategy suggested by Su et al. : to “leverage international collaboration for consistent messaging and comprehensive crisis communication.” The key word analysis shows that the messages they both delivered were consistent even though they differed in tackling different issues surrounding these messages. Contrastively, there was some divergence between the HOs communication styles and previously suggested procedures in the literature. For example, Hyland-Wood et al. suggest openness and frankness as the most effective strategy government officials must follow to encourage people to follow certain procedures. However, the strategies followed by the two HOs showed some indirectness. This appears in the lack of declaratory speech acts, the small number of commissive speech acts they both employed, and the prevalence of indirect speech acts, especially in the SHO data. There were also some further similarities in how they used speech acts; for example, the prevalence of the use of assertions. Elliott-Maksymowicz et al. found assertions predominate in tweets, which accords with our findings. Further, they reported that political arguments on Twitter are often of a very simple nature, featuring assertions in simple sentences; simple language increases the possibility of one’s intended meaning being understood . This explains why assertions appear extensively in our data. However, we did find some differences between the two HOs’ use of speech acts. The BHO used rhetorical questions, while the SHO did not. In political discourse, rhetorical questions are “used to make a point rather than to elicit an answer” , and this is exactly how they were used by the BHO. Also, the both HOs used empathy as a strategy for communication. However, they differed in the type of empathetic procedures they utilized. For example, the BHO used expressive speech acts to express her feelings and build solidarity with her audience. On the other hand, the SHO used indirect directive speech acts through the use of declarative sentences and thanking procedures. People-centered and empathetic persuasion is another strategy suggested by Su et al. . The aspect for which we observed the greatest divergence between the two HOs’ tweets was metaphor. In general, we can say that the BHO used conflict-based metaphors; namely, war and gaming metaphors. The SHO, on the other hand, used metaphors that reflected life as a journey that had been interrupted by the pandemic. However, the SHO’s (and other government officials in Saudi Arabia) use of a conflict-based metaphor (weapon) may indicate that, when seeking to implement extreme health measures, such as quarantine, it is necessary to use extreme conflict-based metaphors to convince people. This reminds us of Semino et al.’s work , in which they proved that the effectiveness of conflict-based metaphors depends on the context in which they are used. Violence metaphors can be disempowering, but here they serve a purpose of convincing people to follow procedures in the same way that soldiers are expected to follow procedures in wars. We also found that the discourse the HOs used when referring to the pandemic featured characteristics of both health communication and political discourse. The similarity to health communication relates to the fact that the HOs utilized directive speech acts to tell patients/audiences the procedures they should follow to achieve the desired conclusion: healing of patients and ending of the pandemic. On the other hand, this discourse is similar to political discourse in that it utilized rhetorical questions and assertions to convince people to perform certain behaviors favored by the officials. A similarity between health communication and political discourse in English is the use of war metaphors. Politicians tend to frequently use war metaphors , and such metaphors are also common in health-care discourse . While war metaphors were not prevalent in the Arabic corpus, we saw that such metaphors might be arriving in Arabic through translation, as has happened previously in Malaysia and Singapore . The tweet data included a number of images such as photographs, info graphs, emojis, and maps. Although we did not include these images in the present paper except to help in understanding the intended meaning of some tweets, images can be used to enforce and complement speech acts. For example, in the BHO’s assertion that her mother had received the vaccine, she strongly verified this by including with the tweet a photo of her mother. How speech acts are reinforced by imagery is a subject for further study. Does the type of speech act correlate with the type of imagery used? For example, are assertions accompanied by info graphs and thanks by emojis of hearts and flowers? Are assertions not enough to persuade people and we need to show them photographic and video evidence so that they believe what we say to them even if the information is from a reliable source like a health official? This study identifies different strategies used by HOs, and shows that these strategies are compatible with the ones suggested by previous research. However, it would be beneficial to examine the impact of these different strategies on the public using experimental studies that expose participants to different conditions and measure the effect of those communication strategies (See ). Other questions could be explored: Was the directness of the BHO more efficient than the indirectness of the SHO? Or vice versa? We leave this for future investigation. Addressing emergencies to the public, particularly for a crisis such as COVID-19, needs leaders to adopt the risk communication approach and develop a consensual narrative that does not rely on a coercion strategy. A narrative of this nature should include factual and open information about uncertainty. During the COVID-19 crisis, those leaders who were able to gain public trust were those who advocated participatory public communication and fact-checked information. S1 File English data: The tweets of the British health official analyzed in this paper. (DOCX) Click here for additional data file. S2 File Arabic data: The tweets of the Saudi health official analyzed in this paper. (TXT) Click here for additional data file.
Patient education interventions for the management of inflammatory bowel disease	8930c2ae-6195-4c26-84e0-67a95a165102	10162698	Patient Education as Topic[mh]	Description of the condition Inflammatory bowel disease (IBD) is an umbrella term for a range of conditions that cause inflammation to the human gastrointestinal tract, with the most prominent ones being ulcerative colitis (UC) and Crohn's disease. Symptoms can include pain, cramping, swelling, diarrhoea, weight loss and tiredness. The aetiology of IBD is still undetermined, but it is thought to be caused via a complex interaction of genetic and environmental factors ( ). More specifically, it is thought that IBD is due to an aberrant immune response to the gut commensal flora in a genetically susceptible individual ( ). IBD is a life‐long condition for which currently there is no cure. Treatment options include medications, lifestyle and diet changes, and surgery with the aim of inducing and maintaining remission of the disease. It is estimated that more than 6.8 million people are living with IBD globally, with incidences of the disease rising especially in regions that are newly adopting western lifestyles ( ; ). Apart from its physical manifestations, IBD can have a serious impact on patients' psychological and social well‐being by limiting the patient's ability to take part in social activities and engagements. It also places a significant burden on healthcare systems, with an estimated EUR 4.6 billion to EUR 5.6 billion of annual healthcare costs attributed to IBD in Europe and USD 7.2 billion in the USA ( ; ). Description of the intervention Patient educational interventions aim to deliver structured information to the recipient of the intervention and there is evidence to suggest patient education can have positive effects in other chronic diseases on specific clinical and quality of life outcomes ( ; ; ). However, the content, delivery method, duration and specific purposes of any given intervention can vary considerably and there are no set standards for any of these parameters. Local resources and healthcare systems, as well as individual patient factors, can have a major impact on patient education. Therefore, there is a need to understand whether such interventions can affect patient outcomes, and how and why they affect patient outcomes. How the intervention might work Education will enhance patient knowledge surrounding IBD. However, the question of how this may impact on their disease outcomes is complex. One point of focus has been about advising patients how to determine when their disease is deteriorating, so they can contact their healthcare provider. Improving medication adherence, recognising adverse effects and when to report them, and improving compliance might be some ways patient education interventions might work. IBD can affect patients' daily lives in several ways and can lead to a lower health‐related quality of life (HRQoL). Together with provider‐led management, self‐management and knowledge about their disease can play an important role in giving patients control over their condition. IBD educational interventions can provide patients with important information and advice towards that end. Why it is important to do this review More clarity about the types of educational interventions targeting people with IBD that have been researched at a randomised controlled trial (RCT) level; what they entail and to what extent they are effective is vital for people with IBD to make better informed decisions for the self‐management of their condition. It is important to review the evidence that has sought to address deficits identified in education systematically ( ), and to assess the attributes of training packages, so they can be applied effectively ( ). The extent to which we can answer 'how' training can be designed, 'why' it is effective and 'for whom and when' will depend on descriptive data within primary studies, but it is important to highlight this information to help professionals understand and deliver health education in a reliable and reproducible manner ( ; ). Inflammatory bowel disease (IBD) is an umbrella term for a range of conditions that cause inflammation to the human gastrointestinal tract, with the most prominent ones being ulcerative colitis (UC) and Crohn's disease. Symptoms can include pain, cramping, swelling, diarrhoea, weight loss and tiredness. The aetiology of IBD is still undetermined, but it is thought to be caused via a complex interaction of genetic and environmental factors ( ). More specifically, it is thought that IBD is due to an aberrant immune response to the gut commensal flora in a genetically susceptible individual ( ). IBD is a life‐long condition for which currently there is no cure. Treatment options include medications, lifestyle and diet changes, and surgery with the aim of inducing and maintaining remission of the disease. It is estimated that more than 6.8 million people are living with IBD globally, with incidences of the disease rising especially in regions that are newly adopting western lifestyles ( ; ). Apart from its physical manifestations, IBD can have a serious impact on patients' psychological and social well‐being by limiting the patient's ability to take part in social activities and engagements. It also places a significant burden on healthcare systems, with an estimated EUR 4.6 billion to EUR 5.6 billion of annual healthcare costs attributed to IBD in Europe and USD 7.2 billion in the USA ( ; ). Patient educational interventions aim to deliver structured information to the recipient of the intervention and there is evidence to suggest patient education can have positive effects in other chronic diseases on specific clinical and quality of life outcomes ( ; ; ). However, the content, delivery method, duration and specific purposes of any given intervention can vary considerably and there are no set standards for any of these parameters. Local resources and healthcare systems, as well as individual patient factors, can have a major impact on patient education. Therefore, there is a need to understand whether such interventions can affect patient outcomes, and how and why they affect patient outcomes. Education will enhance patient knowledge surrounding IBD. However, the question of how this may impact on their disease outcomes is complex. One point of focus has been about advising patients how to determine when their disease is deteriorating, so they can contact their healthcare provider. Improving medication adherence, recognising adverse effects and when to report them, and improving compliance might be some ways patient education interventions might work. IBD can affect patients' daily lives in several ways and can lead to a lower health‐related quality of life (HRQoL). Together with provider‐led management, self‐management and knowledge about their disease can play an important role in giving patients control over their condition. IBD educational interventions can provide patients with important information and advice towards that end. More clarity about the types of educational interventions targeting people with IBD that have been researched at a randomised controlled trial (RCT) level; what they entail and to what extent they are effective is vital for people with IBD to make better informed decisions for the self‐management of their condition. It is important to review the evidence that has sought to address deficits identified in education systematically ( ), and to assess the attributes of training packages, so they can be applied effectively ( ). The extent to which we can answer 'how' training can be designed, 'why' it is effective and 'for whom and when' will depend on descriptive data within primary studies, but it is important to highlight this information to help professionals understand and deliver health education in a reliable and reproducible manner ( ; ). To identify the different types of educational interventions, how they are delivered, and to determine their effectiveness and safety in people with inflammatory bowel disease (IBD). Criteria for considering studies for this review Types of studies All published, unpublished and ongoing RCTs that compare educational interventions targeted at people with IBD to any other type of intervention or no intervention. Cluster‐randomised and cross‐over trials that met our criteria were included. Types of participants People with IBD of all ages. Types of interventions Any type of formal or informal educational intervention, lasting for any time, that has content focused directly on knowledge about IBD or skills needed for direct management of IBD or its symptoms. Interventions that use education to deliver a different set of skills or outcomes that may by proxy enhance patients outcomes were not included (e.g. cognitive behavioural therapy (CBT) training, hypnotherapy training, relaxation therapy training, training on how to use a remote or other health tool for monitoring disease, training on diagnostic tools). Delivery methods can include face‐to‐face or remote educational sessions or workshops, guided study via the use of printed or online materials, the use of mobile applications or any other method that delivers information to patients. It became clear through data extraction that many papers did not mention details about standard therapies. Our team discussed this, and decided that it was highly unlikely that patients would be denied treatment in lieu of patient education or the control therapies. In addition, we could not assume the use of placebo if it was not mentioned by the authors. We considered terms such as “standard care”, “usual care”, “treatment as usual”, “routine follow‐up”, as interchangeable. We recognise this is a source of clinical heterogeneity, as these terms can refer to different approaches of standard care which are not identical in every way, however, we agreed they were probably similar enough for the meta‐analysis purposes of this review. We have listed all intervention and comparator groups in the ' ' table. Types of outcome measures We considered both dichotomous and continuous outcomes for this review. These were not used as criteria for considering inclusion. Primary outcomes Disease activity at study end, using a recognised disease activity scoring system as described by the study authors. Flare‐ups or relapse measured clinically, endoscopically or histologically, during the study period. Quality of life at study end using validated scales or tools. Secondary outcomes Number of episodes of accessing health care (outpatient, remote or inpatient) during the study follow‐up. Change in disease activity using a recognised score at study end. Change in quality of life using a validated tool at study end. Medication adherence. Patient knowledge or skill (or both) as measured by a study, at study end. Adverse effects Total adverse effects (serious and minor) at study end (e.g. functional bowel symptoms, worsening disease state symptoms, hospitalisation). Adverse events leading to withdrawal during the study (as per examples above). Search methods for identification of studies Electronic searches On 27 November 2022, the information Specialist searched the following sources: Cochrane Central Register of Controlled Trials (CENTRAL via Cochrane Library, from inception to issue 11, November 2022) ( ); MEDLINE (via Ovid SP, 1946 to 27 November 2022) ( ); Embase (via Ovid SP, 1974 to 27 November 2022) ( ); ClinicalTrials.gov ( www.clinicaltrials.gov ; ); World Health Organization International Clinical Trials Registry Platform (ICTRP, www.who.int/trialsearch/ , ). We followed the latest guidelines from Cochrane in designing and running the searches ( ). We also used the Cochrane highly sensitive search strategy for identifying randomised trials in MEDLINE (sensitivity‐maximising version, 2008 revision, Ovid format) and Cochrane's RCT search filter for Embase ( ) for identifying the randomised controlled trials. The MEDLINE search strategy was adapted and translated into the syntax of other sources. We did not apply any date, language, document type, or publication status limitations to this search. Searching other resources As complementary search methods, we carefully checked relevant systematic reviews for studies for potential inclusion in our review. In addition, we scrutinised the references of included studies in our review. We sought unpublished trials by contacting experts in the field. We attempted to obtain translations of papers when necessary. If this was needed, translation was completed first and then the study managed for screening and extraction as other papers. Data collection and analysis We carried out data collection and analysis according to the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions ( ). Selection of studies Two review authors (UI and MA) independently screened the titles and abstracts identified from the literature search. We discarded studies that did not meet the inclusion criteria. We then obtained the full report of studies that appeared to meet our inclusion criteria, or for which there was insufficient information to make a final decision. Two review authors independently assessed the reports to establish whether the studies met the inclusion criteria. We resolved disagreements by discussion, and consulted a third review author if resolution was not possible. We entered studies rejected at this or subsequent stages in the ' ' tables and recorded the main reason for exclusion. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram. Where studies had multiple publications, we identified and excluded duplicates, and collated the reports of the same study so that each study, rather than each report, is the unit of interest for the review, and such studies have a single identifier with multiple references. Data extraction and management Two review authors independently carried out data extraction using piloted data extraction forms. We extracted relevant data from full‐text articles that met the inclusion criteria including: trial setting: country and number of trial centres; methods: study design, total study duration and date; participant characteristics: age, socio‐demographics, ethnicity, diagnostic criteria and total number; eligibility criteria: inclusion and exclusion criteria; intervention and comparator — this included description of the learning outcomes planned for the intervention by the teacher or designer, methods of education used, target audience and any resources required; patient outcomes: patient outcome definition, unit of measurement and time of collection; outcomes from education: educational outcomes, if described, reported and classified as either satisfaction/reaction, attitudes or knowledge and skills; results: number of participants allocated to each group, missing participants, sample size; funding source. Assessment of risk of bias in included studies During data extraction, two review authors independently assessed all studies that met the inclusion criteria for their risk of bias using criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions ( ). The domains that we assessed are as follows. Sequence generation (selection bias). Allocation concealment (selection bias). Blinding of participants and personnel (performance bias). Blinding of outcome assessment (detection bias). Incomplete outcome data (attrition bias). Selective reporting (reporting bias). Other bias. We judged the studies to be at low, high or unclear risk of bias for each domain assessed, based on guidance in the Cochrane Handbook for Systematic Reviews of Interventions ( ). After data extraction, two review authors compared the extracted data to discuss and resolve discrepancies before the data were transferred into the ' ' table. For cluster‐RCTs, we judged risk of bias as prescribed in section 16.3.2 “Assessing risk of bias in cluster‐randomized trials” of the Cochrane Handbook for Systematic Reviews of Interventions ( ). Measures of treatment effect For dichotomous outcomes, we expressed treatment effect as risk ratios (RRs) with corresponding 95% confidence intervals (CIs). For continuous outcomes, we expressed the treatment effect as mean difference (MD) with 95% CI if studies used the same scales and methods. However, if studies assessed the same continuous outcome using different methods, we estimated the treatment effect using the standardised mean difference (SMD) with 95% CIs. SMD was used when a continuous outcome was measured on two or more different scales by the studies included in the meta‐analysis. We presented SMDs as standard deviation (SD) units and interpreted them as follows: 0.2 represents a small effect, 0.5 a moderate effect and 0.8 a large effect. Unit of analysis issues The participant is the unit of analysis. For studies comparing more than two intervention groups, we made multiple pair‐wise comparisons between all possible pairs of intervention groups. To avoid double counting, we divided out shared intervention groups evenly among the comparisons. For dichotomous outcomes, we divided up both the number of events and the total number of participants. For continuous outcomes, we divided up the total number of participants and left the means and standard deviations unchanged (this occurred for ). We included cross‐over studies if data were reported separately before and after cross over and we only used data from the first phase for our analysis. In the case of cluster RCTs, we used study data only if the authors used appropriate statistical methods that took the clustering effect into account. We also excluded cluster‐RCTs from a sensitivity analysis to assess their impact on the results. If studies reported dichotomous event data per episode instead of per patient, given the risk of unit of analysis issues, we contacted the authors for further data. If papers reported outcomes at several time points, we used the longest follow‐up. Dealing with missing data We contacted authors where there were missing data or where studies had not reported data in sufficient detail. We attempted to estimate missing standard deviations using relevant statistical tools and calculators when studies reported standard errors. We judged studies that failed to report measures of variance as being at high risk of selective reporting bias. For negative outcomes we used the plausible worst‐case scenario and added the numbers of dropouts to the numerator, as is normal practice for reviews for IBD given the chronic nature of the condition and the high rates of adverse events and treatment failures across a patient's journey. For withdrawals that were specifically due to adverse events, we considered all unspecified reasons and all reasons that did not automatically preclude the possibility of an adverse event, as adverse events. For analyses using continuous outcomes, we used the sample numbers as reported by the authors for each particular continuous outcome. If the sample numbers were not reported, we estimated the sample number based on the attrition percentages reported. For cluster‐trial data we estimated effective sample sizes based on Chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions ( ). Assessment of heterogeneity We scrutinised studies to ensure that they were clinically homogeneous in terms of participants, interventions, comparators and outcomes. To test for statistical heterogeneity, we used a Chi 2 test. A P value of less than 0.1 gives an indication of the presence of heterogeneity. Inconsistency was quantified and represented by the I 2 statistic. We interpreted the thresholds as follows ( ): 0% to 40%: might not be important; 30% to 60%: may represent moderate heterogeneity; 50% to 90%; may represent substantial heterogeneity; 75% to 100%: considerable heterogeneity. Assessment of reporting biases Most reporting biases were minimised by using an inclusive search strategy. We intended to investigate publication bias using a funnel plot if there were 10 or more studies that contributed to a meta‐analysis. We would determine the magnitude of publication bias by visual inspection of the asymmetry of the funnel plot. In addition, we would test funnel plot asymmetry by performing a linear regression of intervention effect estimate against its standard error, weighted by the inverse of the variance of the intervention effect estimate ( ). Data synthesis To summarise the study characteristics, we conducted a narrative synthesis of all the included studies. We then carried out a meta‐analysis if there were two or more studies that assessed similar populations, interventions and outcomes. We synthesised data using the random‐effects model in RevMan Web ( ). We combined effect estimates of studies which report data in a similar way, in the meta‐analysis. We pooled RRs for dichotomous outcomes and MDs or SMDs for continuous outcomes with 95% CIs. Where we were unable to carry out a meta‐analysis (e.g. due to lack of uniformity in data reporting), we presented a narrative summary of the included studies. We recorded and synthesised the following to characterise educational interventions. Educational content (primary material, learning outcomes, theoretical underpinning). Teaching attributes of training programmes used (staff and resource requirements, length of course, methods including whether e‐learning, asynchronous or synchronous, any follow‐up service or session). Any knowledge assessment, including method used and reported pre‐ and post‐test scores. Subgroup analysis and investigation of heterogeneity In case of heterogeneity, we planned to investigate possible causes and address them using methods described in the Cochrane Handbook for Systematic Reviews of Interventions ( ). We planned to undertake subgroup analyses of potential effect modifiers if there were 10 studies or more. If enough data were available, we planned to perform subgroup analyses by age, gender and disease type for all primary outcomes, as these are the most likely to impact the pedagogical methods ( ) and content of education ( ). There were not sufficient studies included and so these analyses did not take place. Sensitivity analysis Where enough data were available, we planned to undertake sensitivity analyses on the primary outcomes, to assess whether the findings of the review were robust to the decisions made during the review process. In particular, we excluded studies at high or unclear risk of bias in any field except for performance bias from analyses that had a mix of studies with different risk of bias judgements. Where data analyses included studies with reported and estimated standard deviations, we planned to exclude those with estimated standard deviations to assess whether this affected the findings of the review. We investigated whether the choice of model (fixed‐effect versus random‐effects) impacted the results to explore heterogeneity. For quality of life, when a mixture of validated and unvalidated measures were used, we performed a sensitivity analysis with only validated measures (e.g. Inflammatory Bowel Disease Questionnaire (IBDQ). Summary of findings and assessment of the certainty of the evidence We presented the main results in a summary of findings table. Each comparison and primary outcome was exported to GRADEprofiler software (developed by the GRADE Working Group) for quality assessment ( ). We included all primary outcomes. Based on risk of bias, inconsistency, imprecision, indirectness and publication bias, we rated the certainty of the evidence for each outcome as high, moderate, low or very low. These ratings have been defined as follows. High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect. We justified all decisions to downgrade the quality of studies using footnotes and we made comments to aid reader's understanding of the review where necessary. Types of studies All published, unpublished and ongoing RCTs that compare educational interventions targeted at people with IBD to any other type of intervention or no intervention. Cluster‐randomised and cross‐over trials that met our criteria were included. Types of participants People with IBD of all ages. Types of interventions Any type of formal or informal educational intervention, lasting for any time, that has content focused directly on knowledge about IBD or skills needed for direct management of IBD or its symptoms. Interventions that use education to deliver a different set of skills or outcomes that may by proxy enhance patients outcomes were not included (e.g. cognitive behavioural therapy (CBT) training, hypnotherapy training, relaxation therapy training, training on how to use a remote or other health tool for monitoring disease, training on diagnostic tools). Delivery methods can include face‐to‐face or remote educational sessions or workshops, guided study via the use of printed or online materials, the use of mobile applications or any other method that delivers information to patients. It became clear through data extraction that many papers did not mention details about standard therapies. Our team discussed this, and decided that it was highly unlikely that patients would be denied treatment in lieu of patient education or the control therapies. In addition, we could not assume the use of placebo if it was not mentioned by the authors. We considered terms such as “standard care”, “usual care”, “treatment as usual”, “routine follow‐up”, as interchangeable. We recognise this is a source of clinical heterogeneity, as these terms can refer to different approaches of standard care which are not identical in every way, however, we agreed they were probably similar enough for the meta‐analysis purposes of this review. We have listed all intervention and comparator groups in the ' ' table. Types of outcome measures We considered both dichotomous and continuous outcomes for this review. These were not used as criteria for considering inclusion. Primary outcomes Disease activity at study end, using a recognised disease activity scoring system as described by the study authors. Flare‐ups or relapse measured clinically, endoscopically or histologically, during the study period. Quality of life at study end using validated scales or tools. Secondary outcomes Number of episodes of accessing health care (outpatient, remote or inpatient) during the study follow‐up. Change in disease activity using a recognised score at study end. Change in quality of life using a validated tool at study end. Medication adherence. Patient knowledge or skill (or both) as measured by a study, at study end. Adverse effects Total adverse effects (serious and minor) at study end (e.g. functional bowel symptoms, worsening disease state symptoms, hospitalisation). Adverse events leading to withdrawal during the study (as per examples above). All published, unpublished and ongoing RCTs that compare educational interventions targeted at people with IBD to any other type of intervention or no intervention. Cluster‐randomised and cross‐over trials that met our criteria were included. People with IBD of all ages. Any type of formal or informal educational intervention, lasting for any time, that has content focused directly on knowledge about IBD or skills needed for direct management of IBD or its symptoms. Interventions that use education to deliver a different set of skills or outcomes that may by proxy enhance patients outcomes were not included (e.g. cognitive behavioural therapy (CBT) training, hypnotherapy training, relaxation therapy training, training on how to use a remote or other health tool for monitoring disease, training on diagnostic tools). Delivery methods can include face‐to‐face or remote educational sessions or workshops, guided study via the use of printed or online materials, the use of mobile applications or any other method that delivers information to patients. It became clear through data extraction that many papers did not mention details about standard therapies. Our team discussed this, and decided that it was highly unlikely that patients would be denied treatment in lieu of patient education or the control therapies. In addition, we could not assume the use of placebo if it was not mentioned by the authors. We considered terms such as “standard care”, “usual care”, “treatment as usual”, “routine follow‐up”, as interchangeable. We recognise this is a source of clinical heterogeneity, as these terms can refer to different approaches of standard care which are not identical in every way, however, we agreed they were probably similar enough for the meta‐analysis purposes of this review. We have listed all intervention and comparator groups in the ' ' table. We considered both dichotomous and continuous outcomes for this review. These were not used as criteria for considering inclusion. Primary outcomes Disease activity at study end, using a recognised disease activity scoring system as described by the study authors. Flare‐ups or relapse measured clinically, endoscopically or histologically, during the study period. Quality of life at study end using validated scales or tools. Secondary outcomes Number of episodes of accessing health care (outpatient, remote or inpatient) during the study follow‐up. Change in disease activity using a recognised score at study end. Change in quality of life using a validated tool at study end. Medication adherence. Patient knowledge or skill (or both) as measured by a study, at study end. Adverse effects Total adverse effects (serious and minor) at study end (e.g. functional bowel symptoms, worsening disease state symptoms, hospitalisation). Adverse events leading to withdrawal during the study (as per examples above). Disease activity at study end, using a recognised disease activity scoring system as described by the study authors. Flare‐ups or relapse measured clinically, endoscopically or histologically, during the study period. Quality of life at study end using validated scales or tools. Number of episodes of accessing health care (outpatient, remote or inpatient) during the study follow‐up. Change in disease activity using a recognised score at study end. Change in quality of life using a validated tool at study end. Medication adherence. Patient knowledge or skill (or both) as measured by a study, at study end. Adverse effects Total adverse effects (serious and minor) at study end (e.g. functional bowel symptoms, worsening disease state symptoms, hospitalisation). Adverse events leading to withdrawal during the study (as per examples above). Total adverse effects (serious and minor) at study end (e.g. functional bowel symptoms, worsening disease state symptoms, hospitalisation). Adverse events leading to withdrawal during the study (as per examples above). Electronic searches On 27 November 2022, the information Specialist searched the following sources: Cochrane Central Register of Controlled Trials (CENTRAL via Cochrane Library, from inception to issue 11, November 2022) ( ); MEDLINE (via Ovid SP, 1946 to 27 November 2022) ( ); Embase (via Ovid SP, 1974 to 27 November 2022) ( ); ClinicalTrials.gov ( www.clinicaltrials.gov ; ); World Health Organization International Clinical Trials Registry Platform (ICTRP, www.who.int/trialsearch/ , ). We followed the latest guidelines from Cochrane in designing and running the searches ( ). We also used the Cochrane highly sensitive search strategy for identifying randomised trials in MEDLINE (sensitivity‐maximising version, 2008 revision, Ovid format) and Cochrane's RCT search filter for Embase ( ) for identifying the randomised controlled trials. The MEDLINE search strategy was adapted and translated into the syntax of other sources. We did not apply any date, language, document type, or publication status limitations to this search. Searching other resources As complementary search methods, we carefully checked relevant systematic reviews for studies for potential inclusion in our review. In addition, we scrutinised the references of included studies in our review. We sought unpublished trials by contacting experts in the field. We attempted to obtain translations of papers when necessary. If this was needed, translation was completed first and then the study managed for screening and extraction as other papers. On 27 November 2022, the information Specialist searched the following sources: Cochrane Central Register of Controlled Trials (CENTRAL via Cochrane Library, from inception to issue 11, November 2022) ( ); MEDLINE (via Ovid SP, 1946 to 27 November 2022) ( ); Embase (via Ovid SP, 1974 to 27 November 2022) ( ); ClinicalTrials.gov ( www.clinicaltrials.gov ; ); World Health Organization International Clinical Trials Registry Platform (ICTRP, www.who.int/trialsearch/ , ). We followed the latest guidelines from Cochrane in designing and running the searches ( ). We also used the Cochrane highly sensitive search strategy for identifying randomised trials in MEDLINE (sensitivity‐maximising version, 2008 revision, Ovid format) and Cochrane's RCT search filter for Embase ( ) for identifying the randomised controlled trials. The MEDLINE search strategy was adapted and translated into the syntax of other sources. We did not apply any date, language, document type, or publication status limitations to this search. As complementary search methods, we carefully checked relevant systematic reviews for studies for potential inclusion in our review. In addition, we scrutinised the references of included studies in our review. We sought unpublished trials by contacting experts in the field. We attempted to obtain translations of papers when necessary. If this was needed, translation was completed first and then the study managed for screening and extraction as other papers. We carried out data collection and analysis according to the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions ( ). Selection of studies Two review authors (UI and MA) independently screened the titles and abstracts identified from the literature search. We discarded studies that did not meet the inclusion criteria. We then obtained the full report of studies that appeared to meet our inclusion criteria, or for which there was insufficient information to make a final decision. Two review authors independently assessed the reports to establish whether the studies met the inclusion criteria. We resolved disagreements by discussion, and consulted a third review author if resolution was not possible. We entered studies rejected at this or subsequent stages in the ' ' tables and recorded the main reason for exclusion. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram. Where studies had multiple publications, we identified and excluded duplicates, and collated the reports of the same study so that each study, rather than each report, is the unit of interest for the review, and such studies have a single identifier with multiple references. Data extraction and management Two review authors independently carried out data extraction using piloted data extraction forms. We extracted relevant data from full‐text articles that met the inclusion criteria including: trial setting: country and number of trial centres; methods: study design, total study duration and date; participant characteristics: age, socio‐demographics, ethnicity, diagnostic criteria and total number; eligibility criteria: inclusion and exclusion criteria; intervention and comparator — this included description of the learning outcomes planned for the intervention by the teacher or designer, methods of education used, target audience and any resources required; patient outcomes: patient outcome definition, unit of measurement and time of collection; outcomes from education: educational outcomes, if described, reported and classified as either satisfaction/reaction, attitudes or knowledge and skills; results: number of participants allocated to each group, missing participants, sample size; funding source. Assessment of risk of bias in included studies During data extraction, two review authors independently assessed all studies that met the inclusion criteria for their risk of bias using criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions ( ). The domains that we assessed are as follows. Sequence generation (selection bias). Allocation concealment (selection bias). Blinding of participants and personnel (performance bias). Blinding of outcome assessment (detection bias). Incomplete outcome data (attrition bias). Selective reporting (reporting bias). Other bias. We judged the studies to be at low, high or unclear risk of bias for each domain assessed, based on guidance in the Cochrane Handbook for Systematic Reviews of Interventions ( ). After data extraction, two review authors compared the extracted data to discuss and resolve discrepancies before the data were transferred into the ' ' table. For cluster‐RCTs, we judged risk of bias as prescribed in section 16.3.2 “Assessing risk of bias in cluster‐randomized trials” of the Cochrane Handbook for Systematic Reviews of Interventions ( ). Measures of treatment effect For dichotomous outcomes, we expressed treatment effect as risk ratios (RRs) with corresponding 95% confidence intervals (CIs). For continuous outcomes, we expressed the treatment effect as mean difference (MD) with 95% CI if studies used the same scales and methods. However, if studies assessed the same continuous outcome using different methods, we estimated the treatment effect using the standardised mean difference (SMD) with 95% CIs. SMD was used when a continuous outcome was measured on two or more different scales by the studies included in the meta‐analysis. We presented SMDs as standard deviation (SD) units and interpreted them as follows: 0.2 represents a small effect, 0.5 a moderate effect and 0.8 a large effect. Unit of analysis issues The participant is the unit of analysis. For studies comparing more than two intervention groups, we made multiple pair‐wise comparisons between all possible pairs of intervention groups. To avoid double counting, we divided out shared intervention groups evenly among the comparisons. For dichotomous outcomes, we divided up both the number of events and the total number of participants. For continuous outcomes, we divided up the total number of participants and left the means and standard deviations unchanged (this occurred for ). We included cross‐over studies if data were reported separately before and after cross over and we only used data from the first phase for our analysis. In the case of cluster RCTs, we used study data only if the authors used appropriate statistical methods that took the clustering effect into account. We also excluded cluster‐RCTs from a sensitivity analysis to assess their impact on the results. If studies reported dichotomous event data per episode instead of per patient, given the risk of unit of analysis issues, we contacted the authors for further data. If papers reported outcomes at several time points, we used the longest follow‐up. Dealing with missing data We contacted authors where there were missing data or where studies had not reported data in sufficient detail. We attempted to estimate missing standard deviations using relevant statistical tools and calculators when studies reported standard errors. We judged studies that failed to report measures of variance as being at high risk of selective reporting bias. For negative outcomes we used the plausible worst‐case scenario and added the numbers of dropouts to the numerator, as is normal practice for reviews for IBD given the chronic nature of the condition and the high rates of adverse events and treatment failures across a patient's journey. For withdrawals that were specifically due to adverse events, we considered all unspecified reasons and all reasons that did not automatically preclude the possibility of an adverse event, as adverse events. For analyses using continuous outcomes, we used the sample numbers as reported by the authors for each particular continuous outcome. If the sample numbers were not reported, we estimated the sample number based on the attrition percentages reported. For cluster‐trial data we estimated effective sample sizes based on Chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions ( ). Assessment of heterogeneity We scrutinised studies to ensure that they were clinically homogeneous in terms of participants, interventions, comparators and outcomes. To test for statistical heterogeneity, we used a Chi 2 test. A P value of less than 0.1 gives an indication of the presence of heterogeneity. Inconsistency was quantified and represented by the I 2 statistic. We interpreted the thresholds as follows ( ): 0% to 40%: might not be important; 30% to 60%: may represent moderate heterogeneity; 50% to 90%; may represent substantial heterogeneity; 75% to 100%: considerable heterogeneity. Assessment of reporting biases Most reporting biases were minimised by using an inclusive search strategy. We intended to investigate publication bias using a funnel plot if there were 10 or more studies that contributed to a meta‐analysis. We would determine the magnitude of publication bias by visual inspection of the asymmetry of the funnel plot. In addition, we would test funnel plot asymmetry by performing a linear regression of intervention effect estimate against its standard error, weighted by the inverse of the variance of the intervention effect estimate ( ). Data synthesis To summarise the study characteristics, we conducted a narrative synthesis of all the included studies. We then carried out a meta‐analysis if there were two or more studies that assessed similar populations, interventions and outcomes. We synthesised data using the random‐effects model in RevMan Web ( ). We combined effect estimates of studies which report data in a similar way, in the meta‐analysis. We pooled RRs for dichotomous outcomes and MDs or SMDs for continuous outcomes with 95% CIs. Where we were unable to carry out a meta‐analysis (e.g. due to lack of uniformity in data reporting), we presented a narrative summary of the included studies. We recorded and synthesised the following to characterise educational interventions. Educational content (primary material, learning outcomes, theoretical underpinning). Teaching attributes of training programmes used (staff and resource requirements, length of course, methods including whether e‐learning, asynchronous or synchronous, any follow‐up service or session). Any knowledge assessment, including method used and reported pre‐ and post‐test scores. Subgroup analysis and investigation of heterogeneity In case of heterogeneity, we planned to investigate possible causes and address them using methods described in the Cochrane Handbook for Systematic Reviews of Interventions ( ). We planned to undertake subgroup analyses of potential effect modifiers if there were 10 studies or more. If enough data were available, we planned to perform subgroup analyses by age, gender and disease type for all primary outcomes, as these are the most likely to impact the pedagogical methods ( ) and content of education ( ). There were not sufficient studies included and so these analyses did not take place. Sensitivity analysis Where enough data were available, we planned to undertake sensitivity analyses on the primary outcomes, to assess whether the findings of the review were robust to the decisions made during the review process. In particular, we excluded studies at high or unclear risk of bias in any field except for performance bias from analyses that had a mix of studies with different risk of bias judgements. Where data analyses included studies with reported and estimated standard deviations, we planned to exclude those with estimated standard deviations to assess whether this affected the findings of the review. We investigated whether the choice of model (fixed‐effect versus random‐effects) impacted the results to explore heterogeneity. For quality of life, when a mixture of validated and unvalidated measures were used, we performed a sensitivity analysis with only validated measures (e.g. Inflammatory Bowel Disease Questionnaire (IBDQ). Summary of findings and assessment of the certainty of the evidence We presented the main results in a summary of findings table. Each comparison and primary outcome was exported to GRADEprofiler software (developed by the GRADE Working Group) for quality assessment ( ). We included all primary outcomes. Based on risk of bias, inconsistency, imprecision, indirectness and publication bias, we rated the certainty of the evidence for each outcome as high, moderate, low or very low. These ratings have been defined as follows. High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect. We justified all decisions to downgrade the quality of studies using footnotes and we made comments to aid reader's understanding of the review where necessary. Two review authors (UI and MA) independently screened the titles and abstracts identified from the literature search. We discarded studies that did not meet the inclusion criteria. We then obtained the full report of studies that appeared to meet our inclusion criteria, or for which there was insufficient information to make a final decision. Two review authors independently assessed the reports to establish whether the studies met the inclusion criteria. We resolved disagreements by discussion, and consulted a third review author if resolution was not possible. We entered studies rejected at this or subsequent stages in the ' ' tables and recorded the main reason for exclusion. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram. Where studies had multiple publications, we identified and excluded duplicates, and collated the reports of the same study so that each study, rather than each report, is the unit of interest for the review, and such studies have a single identifier with multiple references. Two review authors independently carried out data extraction using piloted data extraction forms. We extracted relevant data from full‐text articles that met the inclusion criteria including: trial setting: country and number of trial centres; methods: study design, total study duration and date; participant characteristics: age, socio‐demographics, ethnicity, diagnostic criteria and total number; eligibility criteria: inclusion and exclusion criteria; intervention and comparator — this included description of the learning outcomes planned for the intervention by the teacher or designer, methods of education used, target audience and any resources required; patient outcomes: patient outcome definition, unit of measurement and time of collection; outcomes from education: educational outcomes, if described, reported and classified as either satisfaction/reaction, attitudes or knowledge and skills; results: number of participants allocated to each group, missing participants, sample size; funding source. During data extraction, two review authors independently assessed all studies that met the inclusion criteria for their risk of bias using criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions ( ). The domains that we assessed are as follows. Sequence generation (selection bias). Allocation concealment (selection bias). Blinding of participants and personnel (performance bias). Blinding of outcome assessment (detection bias). Incomplete outcome data (attrition bias). Selective reporting (reporting bias). Other bias. We judged the studies to be at low, high or unclear risk of bias for each domain assessed, based on guidance in the Cochrane Handbook for Systematic Reviews of Interventions ( ). After data extraction, two review authors compared the extracted data to discuss and resolve discrepancies before the data were transferred into the ' ' table. For cluster‐RCTs, we judged risk of bias as prescribed in section 16.3.2 “Assessing risk of bias in cluster‐randomized trials” of the Cochrane Handbook for Systematic Reviews of Interventions ( ). For dichotomous outcomes, we expressed treatment effect as risk ratios (RRs) with corresponding 95% confidence intervals (CIs). For continuous outcomes, we expressed the treatment effect as mean difference (MD) with 95% CI if studies used the same scales and methods. However, if studies assessed the same continuous outcome using different methods, we estimated the treatment effect using the standardised mean difference (SMD) with 95% CIs. SMD was used when a continuous outcome was measured on two or more different scales by the studies included in the meta‐analysis. We presented SMDs as standard deviation (SD) units and interpreted them as follows: 0.2 represents a small effect, 0.5 a moderate effect and 0.8 a large effect. The participant is the unit of analysis. For studies comparing more than two intervention groups, we made multiple pair‐wise comparisons between all possible pairs of intervention groups. To avoid double counting, we divided out shared intervention groups evenly among the comparisons. For dichotomous outcomes, we divided up both the number of events and the total number of participants. For continuous outcomes, we divided up the total number of participants and left the means and standard deviations unchanged (this occurred for ). We included cross‐over studies if data were reported separately before and after cross over and we only used data from the first phase for our analysis. In the case of cluster RCTs, we used study data only if the authors used appropriate statistical methods that took the clustering effect into account. We also excluded cluster‐RCTs from a sensitivity analysis to assess their impact on the results. If studies reported dichotomous event data per episode instead of per patient, given the risk of unit of analysis issues, we contacted the authors for further data. If papers reported outcomes at several time points, we used the longest follow‐up. We contacted authors where there were missing data or where studies had not reported data in sufficient detail. We attempted to estimate missing standard deviations using relevant statistical tools and calculators when studies reported standard errors. We judged studies that failed to report measures of variance as being at high risk of selective reporting bias. For negative outcomes we used the plausible worst‐case scenario and added the numbers of dropouts to the numerator, as is normal practice for reviews for IBD given the chronic nature of the condition and the high rates of adverse events and treatment failures across a patient's journey. For withdrawals that were specifically due to adverse events, we considered all unspecified reasons and all reasons that did not automatically preclude the possibility of an adverse event, as adverse events. For analyses using continuous outcomes, we used the sample numbers as reported by the authors for each particular continuous outcome. If the sample numbers were not reported, we estimated the sample number based on the attrition percentages reported. For cluster‐trial data we estimated effective sample sizes based on Chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions ( ). We scrutinised studies to ensure that they were clinically homogeneous in terms of participants, interventions, comparators and outcomes. To test for statistical heterogeneity, we used a Chi 2 test. A P value of less than 0.1 gives an indication of the presence of heterogeneity. Inconsistency was quantified and represented by the I 2 statistic. We interpreted the thresholds as follows ( ): 0% to 40%: might not be important; 30% to 60%: may represent moderate heterogeneity; 50% to 90%; may represent substantial heterogeneity; 75% to 100%: considerable heterogeneity. Most reporting biases were minimised by using an inclusive search strategy. We intended to investigate publication bias using a funnel plot if there were 10 or more studies that contributed to a meta‐analysis. We would determine the magnitude of publication bias by visual inspection of the asymmetry of the funnel plot. In addition, we would test funnel plot asymmetry by performing a linear regression of intervention effect estimate against its standard error, weighted by the inverse of the variance of the intervention effect estimate ( ). To summarise the study characteristics, we conducted a narrative synthesis of all the included studies. We then carried out a meta‐analysis if there were two or more studies that assessed similar populations, interventions and outcomes. We synthesised data using the random‐effects model in RevMan Web ( ). We combined effect estimates of studies which report data in a similar way, in the meta‐analysis. We pooled RRs for dichotomous outcomes and MDs or SMDs for continuous outcomes with 95% CIs. Where we were unable to carry out a meta‐analysis (e.g. due to lack of uniformity in data reporting), we presented a narrative summary of the included studies. We recorded and synthesised the following to characterise educational interventions. Educational content (primary material, learning outcomes, theoretical underpinning). Teaching attributes of training programmes used (staff and resource requirements, length of course, methods including whether e‐learning, asynchronous or synchronous, any follow‐up service or session). Any knowledge assessment, including method used and reported pre‐ and post‐test scores. In case of heterogeneity, we planned to investigate possible causes and address them using methods described in the Cochrane Handbook for Systematic Reviews of Interventions ( ). We planned to undertake subgroup analyses of potential effect modifiers if there were 10 studies or more. If enough data were available, we planned to perform subgroup analyses by age, gender and disease type for all primary outcomes, as these are the most likely to impact the pedagogical methods ( ) and content of education ( ). There were not sufficient studies included and so these analyses did not take place. Where enough data were available, we planned to undertake sensitivity analyses on the primary outcomes, to assess whether the findings of the review were robust to the decisions made during the review process. In particular, we excluded studies at high or unclear risk of bias in any field except for performance bias from analyses that had a mix of studies with different risk of bias judgements. Where data analyses included studies with reported and estimated standard deviations, we planned to exclude those with estimated standard deviations to assess whether this affected the findings of the review. We investigated whether the choice of model (fixed‐effect versus random‐effects) impacted the results to explore heterogeneity. For quality of life, when a mixture of validated and unvalidated measures were used, we performed a sensitivity analysis with only validated measures (e.g. Inflammatory Bowel Disease Questionnaire (IBDQ). We presented the main results in a summary of findings table. Each comparison and primary outcome was exported to GRADEprofiler software (developed by the GRADE Working Group) for quality assessment ( ). We included all primary outcomes. Based on risk of bias, inconsistency, imprecision, indirectness and publication bias, we rated the certainty of the evidence for each outcome as high, moderate, low or very low. These ratings have been defined as follows. High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect. We justified all decisions to downgrade the quality of studies using footnotes and we made comments to aid reader's understanding of the review where necessary. Description of studies Information on the results of the search, included and excluded studies, and risk of bias assessment is provided below. Results of the search We completed our literature search on 27 November 2022, identifying a total of 4046 records through database searching. After removal of duplicates, 3334 unique records remained. Title and abstract screening revealed 112 records for full‐text review. After assessing all 112 records, we identified 34 records of 14 studies that met the inclusion criteria and were included in the review. We also identified seven records of six ongoing studies, and 27 records of 20 studies awaiting classification (five of the studies awaiting classification were identified during the update search for this review and will be included in the analysis when this review is updated). We excluded 44 records of 37 studies for various reasons (see ). The results of the search are presented in a PRISMA flow diagram ( ). Included studies Additional details on the studies, participants, and interventions can be found in , , and . Setting Fourteen RCTs involving a total of 2708 participants met our inclusion criteria. Three studies were conducted in the USA ( ; ; ), three in Canada ( ; ; ), two in Germany ( ; ), two in Sweden ( ; ), one in the UK ( ), one in France ( ), one in the Netherlands ( ), and one in Turkey ( ). All the included studies were conducted in hospitals and tertiary centres . Seven studies were single‐centre ( ; ; ; ; ; ; ), and seven were multi‐centre ( ; ; ; ; ; ; ). Two studies were cluster‐RCTs ( ; ). Participants Age ranged from 11 years in to 75 years in . There were two studies in paediatric populations ( ; ). included adolescents between 11 and 18 years of age, and participants between 11 and 21 years of age. Both interventions were targeted towards the participating adolescents and not towards their caregivers. Two studies examined exclusively ulcerative colitis (UC) populations ( ; ), whilst the remaining studies examined a mix of IBD patients ( ; ; ; ; ; ; ; ; ; ; ; ). Six studies examined participants in both active and inactive states of the disease ( ; ; ; ; ; ); two studies examined participants in an inactive state of the disease ( ; ); one study examined participants in an active state of the disease ( ); two studies examined participants in remission or low disease activity ( ; ). One study reported the disease activity of its participants as a mean value using the Crohn's Disease Activity Index (CDAI) and the Activity Index (AI) ( ). Two studies did not report on activity of the disease ( ; ). Four of the studies had trial registrations ( ; ; ; ). Interventions The following interventions were assessed in the included trials. A 2‐part patient education seminar versus “ treatment as usual” ( ). Information booklets available from the Crohn’s and Colitis Foundation of Canada versus “ usual care” ( ). Weekly educational text messages versus once every other week educational text messages versus routine clinic visits ( ). E‐learning module accessible via telemedicine system (myIBDcoach) versus routine follow‐up visits ( ). Multi professional group‐based education programme versus regular information during visits to the IBD clinic ( ). Guidebooks for Crohn's Disease (CD) and UC versus “ standard care” ( ). Education programme delivered by a dedicated staff using an illustrated book versus no intervention ( ). A standardised education programme, followed by a group session versus standard care ( ). Nine sessions of lectures alternating with group therapy versus conventional “on demand” medical and psychosocial/psychological treatment ( ). Web‐based education versus education which presented information via easy‐to‐read, illustrated, colour‐printed books (educational content was exactly the same for both groups) ( ). A 30‐minute educational session using the IBD Pocket Guide versus usual care ( ). Internet blog access versus the receipt of text messaging versus Internet blog access and receipt of text messaging versus standard care ( ). Structured education programme and standard care versus standard care consisting of physician visits, at the discretion of the physicians and patients, with physician‐directed ad hoc teaching during visits and the presentation of printed educational literature ( ). Original, interactive video that provided a summary of the 2012 Canadian consensus statements on the treatment of hospitalised adult patients with severe UC versus standard care ( ). Outcomes The length of the interventions ranged from 30 minutes, in , to 12 months in . Primary outcome: Disease activity Only four studies mentioned disease activity as an outcome. measured IBD disease activity as a continuous outcome using the Bowel Disease Activity Index (GIBDI), and used the Crohn's Disease Harvey‐Bradshaw Index (HBI) for CD participants and the Simple Clinical Colitis Activity Index (SCCAI) for patients with UC/indeterminate colitis. In the authors stated disease activity as an outcome, and that they measured it using the Colitis Activity Index (CAI), however the data were not presented. reported the numbers of participants with mild and severe disease at each stage of the study. Primary outcome: Flare‐ups or relapse Five studies measured flare‐ups or relapse. and evaluated mean number of flare‐ups (SD) during the study as continuous data. reported numbers with acute relapse per group with relapse defined as clinical activity index ≥ 9. and also reported numbers of patients with relapse during the study. Primary outcome: Quality of life Ten studies reported quality of life ( ; ; ; ; ; ; ; ; ; ). The Inflammatory Bowel Disease Questionnaire (IBDQ) was used in seven studies, ( ; ; ; ; ; ; ). The short Inflammatory Bowel Disease Questionnaire (SIBDQ) was used by and . The SF‐12 short form health survey was used by . also used the Quality Index in Crohn’s and Colitis (QuICC) questionnaire, and the Rating Form of IBD Patient Concerns (RFIPC). Secondary outcome: Number of episodes accessing health care Four studies stated the number of episodes of accessing health care ( ; ; ; ). reported total encounters, IBD‐related hospitalisations, non‐IBD‐related hospitalisations, non‐invasive diagnostic tests, electronic encounters and telephone encounters, all as rates, adjusted for 100 participants per year. reported hospital admissions and emergency visits, reported kept hospital appointments and numbers of patients who did not attend. measured hospitalisations, and rate of healthcare use. Secondary outcome: Change in disease activity No studies reported this outcome. Secondary outcome: Change in quality of life Only one study reported the change in quality of life in its participants ( ). The study used the IBDQ (the questionniare has 32 questions and the score ranges from a minimum of 32 to a maximum of 224, but the authors presented results as mean scores for each question with a range; high score = better result) and the QuICC (range 1 = excellent to 5 = poor) at the start and after two weeks of the intervention to report the mean values (SD) on its sample. Secondary outcome: Medication adherence Five studies measured medication adherence ( ; ; ; ; ). , , and used the Morisky Medication Adherence Scale. reported adherence rates based on recordings with the MedMinder system. reported incidents and rates of missed medications, and rate of non‐adherence as measured by the Patient Satisfaction Questionnaire and participant self‐report. Secondary outcome: Patient knowledge and/or skill Patient knowledge/skills was reported in seven studies ( ; ; ; ; ; ; ). measured knowledge using the Crohn’s and Colitis Knowledge questionnaire, while used the IBD knowledge Inventory Device (IBD‐KID) and a modified version of the Crohn's & Colitis Foundation of America (CCFA) Knowledge Score (I‐M‐AWARE). used both the Chron's and Colitis Knowledge (CCKNOW) questionnaire and the Knowledge questionnaire (KQ), while it also assessed self‐perceived knowledge on a visual analogue scale (VAS). used the ECIPE (Étude randomisée et contrôlée évaluant l'impact du programme d'éducation (Controlled multicentre study of the impact of a programme of therapeutic Education in IBD)) score they developed for their education programme and defined success as a dichotomous outcome of improvement in patients' skills by an increase of the ECIPE score of more than 20%, from baseline to six months. In medical and psychological knowledge was self‐reported by the participants on a Likert scale, while in IBD knowledge and medication knowledge were self‐reported on a VAS. Secondary outcome: Total adverse events (serious and minor) Only two studies reported total adverse events ( ; ). Secondary outcome: Withdrawals due to adverse events Only three studies reported this outcome ( ; ; ). There were no withdrawals due to adverse events in these studies as no participant reported any adverse events related to use of the telemedicine intervention. Qualitative synthesis: Educational content The details on the contents of each intervention can be found in . Five studies relied on face‐to‐face workshops, seminars or teaching session for delivering their educational content ( ; ; ; ; ). Five used e‐learning or distance learning via mobile phones ( ; ; ; ; ). Three studies used written material as their primary material ( ; ; ). One study used mixed methods of lectures and group therapy for delivering information on IBD and psychological coping methods for IBD, respectively ( ). The educational learning outcomes were not clearly stated in any of the studies. Some studies mentioned generic aims such as empowering patients ( ), enhancing the sense of control and skills in coping with relapses ( ), and a greater sense of control in management, engagement in the care process and understanding of the overall management plan ( ). None of the studies described the educational theoretical underpinning of their interventions. Qualitative synthesis: Teaching attributes of training programmes used (staff and resource requirements, length of course, any follow‐up service or session) Six studies employed synchronous interventions ( ; ; ; ; ; ), and six asynchronous interventions ( ; ; ; ; ; ). Two studies were a mix of synchronous and asynchronous ( ; ). Three interventions were part of a package of measures that contained other elements as well ( ; ; ). Staff delivering the interventions included nurses, gastroenterologists and other physicians, psychologists, dietitians, medical social workers and educators. Resources included computers, tablets, smartphones, booklets and other written materials, as well as physical space and equipment for delivering workshops or lectures. Access issues included participants with insufficient language skills, severe vision or hearing impairments, serious physical or psychological comorbidities, people without access to computers, tablets, or smartphones and non‐access to transport ( ). Qualitative synthesis: Knowledge assessments (formative or summative assessment) Four of the five studies that assessed patient knowledge used summative assessment ( ; ; ; ); we did not have enough information to judge the type of assessment in . The pre‐ and post‐knowledge scores, or changes in knowledge scores from baseline, are presented in . Funding sources and conflicts of interest Nine studies reported their sources of funding ( ; ; ; ; ; ; ; ). Four studies were funded via government grants ( ; ; ; ), three studies by private sources ( ; ; ), one study by a non‐profit research association ( ), and one study declared that it received no financial support ( ). Five studies did not report anything about their source of funding ( ; ; ; ; ). Eight studies made declarations about conflicts of interest ( ; ; ; ; ; ; ; ), and five of these declared no conflicts of interest ( ; ; ; ; ). One study declared that one of the authors was an employee of the industrial partner that provided funding ( ), one study declared that several authors received honoraria from private industrial partners ( ), and one study declared that several authors had connections to healthcare companies unrelated to the study ( ). Six studies did not make any declarations about conflicts of interest ( ; ; ; ; ; ). Excluded studies We excluded 37 full‐text studies (44 records) for various reasons. The reasons for exclusion of each study are presented in the table and are summarised below. Wrong intervention (23 studies) ( ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ). Not RCTs (12 studies) ( ; ; ; ; ; ; ; ; ; ; ; ). Wrong population: (2 studies) ( ; ). There are 20 studies awaiting classification ( ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ). There are six ongoing RCTs ( ; ; ; ; ; ; ). Risk of bias in included studies Below we present the results of our risk of bias assessment ( ; ). Further details can be found in the risk of bias tables (beneath tables). Allocation Randomisation was described clearly in seven studies ( ; ; ; ; ; ; ), which we rated at low for risk of bias, and was not sufficiently described in the other seven studies ( ; ; ; ; ; ; ; ), which we rated unclear for risk of bias. We rated two studies at low risk from allocation concealment ( ; ), as the method of random allocation of participants to intervention and control groups and allocation concealment was described or the risk was low due to cluster randomisation. We rated nine studies at unclear risk of allocation concealment ( ; ; ; ; ; ; ; ; ), as they did not provide enough information about their selection and allocation concealment process. Three studies had no allocation concealment and were judged to be at high risk ( ; ; ). Blinding All studies were rated as high in performance bias, as the interventions they studied could not be blinded for both participants and personnel. Detection bias was rated as low in two studies that mentioned assessors being blinded ( ; ), unclear in six studies that did not provide enough information for a judgement ( ; ; ; ; ; ), and high in six that confirmed or mentioned that the assessors were not blinded ( ; ; ; ; ; ). Incomplete outcome data We judged attrition bias as low in nine studies that provided enough information for judgement ( ; ; ; ; ; ; ; ; ). The rest of the studies we rated at unclear risk ( ; ; ; ; ). Selective reporting We rated reporting bias as low in three studies that reported all outcomes they had set out to report either in their protocols or trial registrations ( ; ; ). We rated nine studies at unclear risk ( ; ; ; ; ; ; ; ; ), and two at high risk ( ; ). Other potential sources of bias We rated 12 studies as low risk for other potential sources of bias ( ; ; ; ; ; ; ; ; ; ; ; ). We rated two studies at unclear risk due to lack of information ( ; ). Effects of interventions See: ; ; A summary of primary and secondary outcome data can be found in and respectively. Any planned subgroup and sensitivity analyses that were not carried out because of a lack of data are mentioned in . 1. Patient education and standard care versus standard care Thirteen studies compared patient education interventions against no intervention ( , ; ; ; ; ; ; ; ; ; ; ; ). Primary outcomes Disease activity at study end Two of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ). There was no clear difference in disease activity when patient education (n = 277) combined with standard care was compared to standard care (n = 202). Patient education combined with standard care is probably equivalent to standard care in reducing disease activity in patients with IBD (standardised mean difference (SMD ‐0.03, 95% confidence interval (CI) ‐0.25 to 0.20). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned disease activity as an outcome, but did not present any results. Flare‐ups or relapse Two of the studies that reported this outcome reported it as a continuous outcome ( ; ), and three reported it as a dichotomous outcome ( ; ; ). For the continuous data meta‐analysis, there was no clear difference for flare‐ups or relapse when patient education (n = 515) combined with standard care was compared to standard care (n = 507), as a continuous outcome. Patient education combined with standard care is probably equivalent to standard care in reducing flare‐ups or relapse in patients with IBD (mean difference (MD) ‐0.00, 95% CI ‐0.06 to 0.05). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). From the dichotomous data, 10 participants experienced relapse in the patient education combined with standard care group (n = 157) and 10 participants experienced relapse in the standard care group (n = 150). The evidence is very uncertain on whether patient education combined with standard care is different to standard care in reducing flare‐ups or relapse in patients with IBD (RR 0.94, 95% CI 0.41 to 2.18). The certainty of the evidence was very low due to serious concerns with risk of bias and imprecision ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned that one participant relapsed during their study but did not clarify to which group they belonged. Quality of life at study end Six of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 721) was compared to standard care (n = 643). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.08, 95% CI ‐0.03 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). We conducted a sensitivity analysis excluding five studies at high risk of bias ( ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 193) was compared to standard care (n = 107). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (MD 1.11, 95% CI ‐5.74 to 7.97). The certainty of the evidence was moderate due to imprecision ( ). We conducted a sensitivity analysis excluding one cluster RCT ( ). There was no clear difference in quality of life when patient education combined with standard care (n = 667) was compared to standard care (n = 571). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.07, 95% CI ‐0.04 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ). We conducted a further sensitivity analysis including only the studies that used the full IBDQ (high score = better result) and as such allowed the use of the mean difference (MD) ( ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 297) was compared to standard care (n = 217). Patient education combined with standard care may be equivalent to standard care in improving quality of life in patients with IBD (MD 1.82, 95% CI ‐3.72 to 7.36). The certainty of the evidence was low due to concerns with risk of bias and imprecision ( ). also measured mental quality of life, in addition to the physical quality of life that was included in the meta‐analysis. The intervention group had a reported score of 46.41 (11.00) and the control group a score of 42.70 (10.89) at 3 months from study end (high score = better result). measured quality of life using the QuICC (low score = better result), in addition to the IBDQ that was used in the above meta‐analysis. The intervention group had a reported score of 87.0 (20.61) and the control group a score of 85.7 (19.83) at study end. used the IBDQ and provided mean scores with variance values. The intervention group had a reported score of 57.85 and the control group a score of 55.58 at study end (high score = better result). and did not provide the raw mean and variance scores per group at study end, only presenting the results of their own analysis. Secondary outcomes Number of episodes of accessing health care In , hospitalisations, surgery, emergency department and office visits, procedures, intravenous therapeutics, and telephone and electronic encounters were extracted from the electronic medical record (EMR) for one year before and after randomisation, and encounters were reported as rates, adjusted for 100 participants per year. The intervention group that received a telemedicine message every other week (IG1 (TELE‐IBD EOW)) had 2235 total encounters, the intervention group that received a telemedicine message every week (IG2 (TELE‐IBD W)) had 1935, and the control group had 2099 (the data on the specific types of encounters are presented in ). reported mean numbers of hospital admissions, which were 0.05 (SD 0.28) for the intervention group and 0.10 (SD 0.43) for the control group; and mean numbers of emergency visits, which were 0.07 (SD 0.35) for the intervention group and 0.10 (SD 0.54) for the control group. reported mean number of kept hospital appointments as 1.9 (SD 2.2) for the intervention group and 3.0 (SD 2.5) for the control group, as well as number of participants who did not attend appointments as 22/279 for the intervention group and 44/403 for the control group. reported rate of health care use as a mean of 0.63 for the intervention group and 0.95 for the control group without providing variance values. mentioned it as an outcome, but did not report data. Change in disease activity This outcome was not reported in any of the studies. Change in quality of life This outcome was only reported in . The mean difference in the intervention group was −0.17 (SD 0.49) and in the control group 0.28 (SD 0.62) for the IBDQ and −0.05 (SD 0.28) and −0.01 (SD 0.25), respectively, for the QuICC. Medication adherence reported medication adherence as a mean of 7.01 (SD 1.40) for the intervention group and 6.77 (SD 1.61) for the control group. reported 66/126 and 64/122 as non‐adherent in the intervention and control groups, respectively. In , difference in average adherence rates pre‐ and post‐randomisation was +0.36 (SD 10.28) for the intervention group and −15.3 (SD 25.34) for the control group. reported 166 incidents of missed medications, with a mean of 2.31 incidents per participant, and calculated the mean number of missed medications during the study as 0.91 for the intervention group and 3.43 for the control group. did not provide the raw mean and variance scores per group at study end, instead the authors presented the results of their own analysis. Patient knowledge or skill at study end In , the mean difference from baseline (no variance provided) was +2.4 in the TELE‐IBD EOW intervention group +2.0 in the TELE‐IBD W intervention group and +1.8 in the control group. reported that mean rank scores (no variance provided; high score = better result) at end of study were: 5.8 for the intervention group and 4.0 for the control group for gastrointestinal anatomy; 5.6 and 4.3 for general IBD knowledge; 6.1 and 3.6 for medications; and 4.2 and 6.0 for nutrition. reported that post‐intervention the mean score on the assessment was 55.6% (range 35.0% to 95.6%), but did not report results per intervention group. reported CCKNOW scores of 19.52 (SD 2.55) for the intervention group and 13.84 (SD 4.86) for the control group, and KQ scores of 27.19 (SD 3.03) and 21.47 (SD 6.81) respectively, at study end. In , an improvement in patients' skills was defined by an increase of the ECIPE score of more than 20%, from baseline to six months. In the intervention group 61 patients achieved that and 31 in the control. Per protocol median ECIPE scores were reported as 26 (range 22‐30) in the intervention group (n = 105) and 20 (range 16‐25) (n = 117) in the control group. In each of the results in this paragraph, higher scores indicate improvement. Self‐reported medical knowledge was reported in three studies as 4.05 (SD 0.41) for the intervention group and 3.42 (SD 0.71) for the control and psychological knowledge as 3.65 (SD 0.67) and 2.98 (SD 0.74), respectively in . Knowledge of IBD was reported as 8.17 (SD 1.16) for the intervention group and 7.84 (SD 1.47) for the control group, and knowledge of medication as 7.75 (SD 1.58) and 7.58 (SD 1.51), respectively in . Self‐perceived knowledge was reported as 7.6 for the intervention group and 6.2 for the control group at study end in . Total adverse effects , , and reported zero total adverse effects in their studies. Withdrawals due to adverse events The only study that reported withdrawals due to adverse effects was , which reported that in the TELE‐IBD EOW intervention group one participant withdrew due to breast cancer and in the TELE‐IBD intervention group two participants withdrew because they needed surgery. No participants withdrew due to adverse effects from the control group. 2. Web‐based patient education versus other delivery of patient education Two studies compared delivery methods of patient education in the form of web‐based interventions against other delivery methods ( ; ). Primary outcomes Only reported any of our primary outcomes. Disease activity at study end reported numbers of UC and CD participants in remission, or with mild, severe, or very severe disease at study end. For UC participants, 8/16 in the web‐based group and 10/16 in the control education group were in remission, 6/16 and 4/16 had mild disease, 2/16 and 1/16 had severe disease, and 0/16 and 0/16 had very severe disease. For CD participants, 5/14 and 10/14 were in remission, 7/14 and 3/14 had mild disease, 2/14 and 1/14 had severe disease, and 0/14 and 0/14 had very severe disease. Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life score on the IBDQ for the web‐based group was 156.53 (SD 30.97) and 155.63 (SD 34.30) for the control group (high score = better result). Secondary outcomes No secondary outcomes were reported except for the limited knowledge score data in , which we reported above. 3. Weekly educational texts messages versus once every other week educational text messages compared frequency of patient education in the form of weekly educational text messages versus once every other week educational text messages (in addition to comparing these interventions to standard care, the results of which we included in the patient education and standard care versus standard care comparison above). Primary outcomes Disease activity at study end Mean disease activity for the TELE‐EOW CD participants was 4.2 (SD 3.9) and for the TELE‐W CD participants 3.2 (SD 3.4). Mean disease activity for the TELE‐EOW UC participants was 1.7 (SD 1.9) and for the TELE‐W UC participants was 2.0 (SD 1.8). Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life scores for the TELE‐EOW participants was 181.5 (SD 28.2) and for the TELE‐W participants was 179.2 (SD 32.8) Secondary outcomes These have been reported in Comparison 1, patient education and standard care versus standard care. Information on the results of the search, included and excluded studies, and risk of bias assessment is provided below. Results of the search We completed our literature search on 27 November 2022, identifying a total of 4046 records through database searching. After removal of duplicates, 3334 unique records remained. Title and abstract screening revealed 112 records for full‐text review. After assessing all 112 records, we identified 34 records of 14 studies that met the inclusion criteria and were included in the review. We also identified seven records of six ongoing studies, and 27 records of 20 studies awaiting classification (five of the studies awaiting classification were identified during the update search for this review and will be included in the analysis when this review is updated). We excluded 44 records of 37 studies for various reasons (see ). The results of the search are presented in a PRISMA flow diagram ( ). Included studies Additional details on the studies, participants, and interventions can be found in , , and . Setting Fourteen RCTs involving a total of 2708 participants met our inclusion criteria. Three studies were conducted in the USA ( ; ; ), three in Canada ( ; ; ), two in Germany ( ; ), two in Sweden ( ; ), one in the UK ( ), one in France ( ), one in the Netherlands ( ), and one in Turkey ( ). All the included studies were conducted in hospitals and tertiary centres . Seven studies were single‐centre ( ; ; ; ; ; ; ), and seven were multi‐centre ( ; ; ; ; ; ; ). Two studies were cluster‐RCTs ( ; ). Participants Age ranged from 11 years in to 75 years in . There were two studies in paediatric populations ( ; ). included adolescents between 11 and 18 years of age, and participants between 11 and 21 years of age. Both interventions were targeted towards the participating adolescents and not towards their caregivers. Two studies examined exclusively ulcerative colitis (UC) populations ( ; ), whilst the remaining studies examined a mix of IBD patients ( ; ; ; ; ; ; ; ; ; ; ; ). Six studies examined participants in both active and inactive states of the disease ( ; ; ; ; ; ); two studies examined participants in an inactive state of the disease ( ; ); one study examined participants in an active state of the disease ( ); two studies examined participants in remission or low disease activity ( ; ). One study reported the disease activity of its participants as a mean value using the Crohn's Disease Activity Index (CDAI) and the Activity Index (AI) ( ). Two studies did not report on activity of the disease ( ; ). Four of the studies had trial registrations ( ; ; ; ). Interventions The following interventions were assessed in the included trials. A 2‐part patient education seminar versus “ treatment as usual” ( ). Information booklets available from the Crohn’s and Colitis Foundation of Canada versus “ usual care” ( ). Weekly educational text messages versus once every other week educational text messages versus routine clinic visits ( ). E‐learning module accessible via telemedicine system (myIBDcoach) versus routine follow‐up visits ( ). Multi professional group‐based education programme versus regular information during visits to the IBD clinic ( ). Guidebooks for Crohn's Disease (CD) and UC versus “ standard care” ( ). Education programme delivered by a dedicated staff using an illustrated book versus no intervention ( ). A standardised education programme, followed by a group session versus standard care ( ). Nine sessions of lectures alternating with group therapy versus conventional “on demand” medical and psychosocial/psychological treatment ( ). Web‐based education versus education which presented information via easy‐to‐read, illustrated, colour‐printed books (educational content was exactly the same for both groups) ( ). A 30‐minute educational session using the IBD Pocket Guide versus usual care ( ). Internet blog access versus the receipt of text messaging versus Internet blog access and receipt of text messaging versus standard care ( ). Structured education programme and standard care versus standard care consisting of physician visits, at the discretion of the physicians and patients, with physician‐directed ad hoc teaching during visits and the presentation of printed educational literature ( ). Original, interactive video that provided a summary of the 2012 Canadian consensus statements on the treatment of hospitalised adult patients with severe UC versus standard care ( ). Outcomes The length of the interventions ranged from 30 minutes, in , to 12 months in . Primary outcome: Disease activity Only four studies mentioned disease activity as an outcome. measured IBD disease activity as a continuous outcome using the Bowel Disease Activity Index (GIBDI), and used the Crohn's Disease Harvey‐Bradshaw Index (HBI) for CD participants and the Simple Clinical Colitis Activity Index (SCCAI) for patients with UC/indeterminate colitis. In the authors stated disease activity as an outcome, and that they measured it using the Colitis Activity Index (CAI), however the data were not presented. reported the numbers of participants with mild and severe disease at each stage of the study. Primary outcome: Flare‐ups or relapse Five studies measured flare‐ups or relapse. and evaluated mean number of flare‐ups (SD) during the study as continuous data. reported numbers with acute relapse per group with relapse defined as clinical activity index ≥ 9. and also reported numbers of patients with relapse during the study. Primary outcome: Quality of life Ten studies reported quality of life ( ; ; ; ; ; ; ; ; ; ). The Inflammatory Bowel Disease Questionnaire (IBDQ) was used in seven studies, ( ; ; ; ; ; ; ). The short Inflammatory Bowel Disease Questionnaire (SIBDQ) was used by and . The SF‐12 short form health survey was used by . also used the Quality Index in Crohn’s and Colitis (QuICC) questionnaire, and the Rating Form of IBD Patient Concerns (RFIPC). Secondary outcome: Number of episodes accessing health care Four studies stated the number of episodes of accessing health care ( ; ; ; ). reported total encounters, IBD‐related hospitalisations, non‐IBD‐related hospitalisations, non‐invasive diagnostic tests, electronic encounters and telephone encounters, all as rates, adjusted for 100 participants per year. reported hospital admissions and emergency visits, reported kept hospital appointments and numbers of patients who did not attend. measured hospitalisations, and rate of healthcare use. Secondary outcome: Change in disease activity No studies reported this outcome. Secondary outcome: Change in quality of life Only one study reported the change in quality of life in its participants ( ). The study used the IBDQ (the questionniare has 32 questions and the score ranges from a minimum of 32 to a maximum of 224, but the authors presented results as mean scores for each question with a range; high score = better result) and the QuICC (range 1 = excellent to 5 = poor) at the start and after two weeks of the intervention to report the mean values (SD) on its sample. Secondary outcome: Medication adherence Five studies measured medication adherence ( ; ; ; ; ). , , and used the Morisky Medication Adherence Scale. reported adherence rates based on recordings with the MedMinder system. reported incidents and rates of missed medications, and rate of non‐adherence as measured by the Patient Satisfaction Questionnaire and participant self‐report. Secondary outcome: Patient knowledge and/or skill Patient knowledge/skills was reported in seven studies ( ; ; ; ; ; ; ). measured knowledge using the Crohn’s and Colitis Knowledge questionnaire, while used the IBD knowledge Inventory Device (IBD‐KID) and a modified version of the Crohn's & Colitis Foundation of America (CCFA) Knowledge Score (I‐M‐AWARE). used both the Chron's and Colitis Knowledge (CCKNOW) questionnaire and the Knowledge questionnaire (KQ), while it also assessed self‐perceived knowledge on a visual analogue scale (VAS). used the ECIPE (Étude randomisée et contrôlée évaluant l'impact du programme d'éducation (Controlled multicentre study of the impact of a programme of therapeutic Education in IBD)) score they developed for their education programme and defined success as a dichotomous outcome of improvement in patients' skills by an increase of the ECIPE score of more than 20%, from baseline to six months. In medical and psychological knowledge was self‐reported by the participants on a Likert scale, while in IBD knowledge and medication knowledge were self‐reported on a VAS. Secondary outcome: Total adverse events (serious and minor) Only two studies reported total adverse events ( ; ). Secondary outcome: Withdrawals due to adverse events Only three studies reported this outcome ( ; ; ). There were no withdrawals due to adverse events in these studies as no participant reported any adverse events related to use of the telemedicine intervention. Qualitative synthesis: Educational content The details on the contents of each intervention can be found in . Five studies relied on face‐to‐face workshops, seminars or teaching session for delivering their educational content ( ; ; ; ; ). Five used e‐learning or distance learning via mobile phones ( ; ; ; ; ). Three studies used written material as their primary material ( ; ; ). One study used mixed methods of lectures and group therapy for delivering information on IBD and psychological coping methods for IBD, respectively ( ). The educational learning outcomes were not clearly stated in any of the studies. Some studies mentioned generic aims such as empowering patients ( ), enhancing the sense of control and skills in coping with relapses ( ), and a greater sense of control in management, engagement in the care process and understanding of the overall management plan ( ). None of the studies described the educational theoretical underpinning of their interventions. Qualitative synthesis: Teaching attributes of training programmes used (staff and resource requirements, length of course, any follow‐up service or session) Six studies employed synchronous interventions ( ; ; ; ; ; ), and six asynchronous interventions ( ; ; ; ; ; ). Two studies were a mix of synchronous and asynchronous ( ; ). Three interventions were part of a package of measures that contained other elements as well ( ; ; ). Staff delivering the interventions included nurses, gastroenterologists and other physicians, psychologists, dietitians, medical social workers and educators. Resources included computers, tablets, smartphones, booklets and other written materials, as well as physical space and equipment for delivering workshops or lectures. Access issues included participants with insufficient language skills, severe vision or hearing impairments, serious physical or psychological comorbidities, people without access to computers, tablets, or smartphones and non‐access to transport ( ). Qualitative synthesis: Knowledge assessments (formative or summative assessment) Four of the five studies that assessed patient knowledge used summative assessment ( ; ; ; ); we did not have enough information to judge the type of assessment in . The pre‐ and post‐knowledge scores, or changes in knowledge scores from baseline, are presented in . Funding sources and conflicts of interest Nine studies reported their sources of funding ( ; ; ; ; ; ; ; ). Four studies were funded via government grants ( ; ; ; ), three studies by private sources ( ; ; ), one study by a non‐profit research association ( ), and one study declared that it received no financial support ( ). Five studies did not report anything about their source of funding ( ; ; ; ; ). Eight studies made declarations about conflicts of interest ( ; ; ; ; ; ; ; ), and five of these declared no conflicts of interest ( ; ; ; ; ). One study declared that one of the authors was an employee of the industrial partner that provided funding ( ), one study declared that several authors received honoraria from private industrial partners ( ), and one study declared that several authors had connections to healthcare companies unrelated to the study ( ). Six studies did not make any declarations about conflicts of interest ( ; ; ; ; ; ). Excluded studies We excluded 37 full‐text studies (44 records) for various reasons. The reasons for exclusion of each study are presented in the table and are summarised below. Wrong intervention (23 studies) ( ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ). Not RCTs (12 studies) ( ; ; ; ; ; ; ; ; ; ; ; ). Wrong population: (2 studies) ( ; ). There are 20 studies awaiting classification ( ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ). There are six ongoing RCTs ( ; ; ; ; ; ; ). We completed our literature search on 27 November 2022, identifying a total of 4046 records through database searching. After removal of duplicates, 3334 unique records remained. Title and abstract screening revealed 112 records for full‐text review. After assessing all 112 records, we identified 34 records of 14 studies that met the inclusion criteria and were included in the review. We also identified seven records of six ongoing studies, and 27 records of 20 studies awaiting classification (five of the studies awaiting classification were identified during the update search for this review and will be included in the analysis when this review is updated). We excluded 44 records of 37 studies for various reasons (see ). The results of the search are presented in a PRISMA flow diagram ( ). Additional details on the studies, participants, and interventions can be found in , , and . Setting Fourteen RCTs involving a total of 2708 participants met our inclusion criteria. Three studies were conducted in the USA ( ; ; ), three in Canada ( ; ; ), two in Germany ( ; ), two in Sweden ( ; ), one in the UK ( ), one in France ( ), one in the Netherlands ( ), and one in Turkey ( ). All the included studies were conducted in hospitals and tertiary centres . Seven studies were single‐centre ( ; ; ; ; ; ; ), and seven were multi‐centre ( ; ; ; ; ; ; ). Two studies were cluster‐RCTs ( ; ). Participants Age ranged from 11 years in to 75 years in . There were two studies in paediatric populations ( ; ). included adolescents between 11 and 18 years of age, and participants between 11 and 21 years of age. Both interventions were targeted towards the participating adolescents and not towards their caregivers. Two studies examined exclusively ulcerative colitis (UC) populations ( ; ), whilst the remaining studies examined a mix of IBD patients ( ; ; ; ; ; ; ; ; ; ; ; ). Six studies examined participants in both active and inactive states of the disease ( ; ; ; ; ; ); two studies examined participants in an inactive state of the disease ( ; ); one study examined participants in an active state of the disease ( ); two studies examined participants in remission or low disease activity ( ; ). One study reported the disease activity of its participants as a mean value using the Crohn's Disease Activity Index (CDAI) and the Activity Index (AI) ( ). Two studies did not report on activity of the disease ( ; ). Four of the studies had trial registrations ( ; ; ; ). Interventions The following interventions were assessed in the included trials. A 2‐part patient education seminar versus “ treatment as usual” ( ). Information booklets available from the Crohn’s and Colitis Foundation of Canada versus “ usual care” ( ). Weekly educational text messages versus once every other week educational text messages versus routine clinic visits ( ). E‐learning module accessible via telemedicine system (myIBDcoach) versus routine follow‐up visits ( ). Multi professional group‐based education programme versus regular information during visits to the IBD clinic ( ). Guidebooks for Crohn's Disease (CD) and UC versus “ standard care” ( ). Education programme delivered by a dedicated staff using an illustrated book versus no intervention ( ). A standardised education programme, followed by a group session versus standard care ( ). Nine sessions of lectures alternating with group therapy versus conventional “on demand” medical and psychosocial/psychological treatment ( ). Web‐based education versus education which presented information via easy‐to‐read, illustrated, colour‐printed books (educational content was exactly the same for both groups) ( ). A 30‐minute educational session using the IBD Pocket Guide versus usual care ( ). Internet blog access versus the receipt of text messaging versus Internet blog access and receipt of text messaging versus standard care ( ). Structured education programme and standard care versus standard care consisting of physician visits, at the discretion of the physicians and patients, with physician‐directed ad hoc teaching during visits and the presentation of printed educational literature ( ). Original, interactive video that provided a summary of the 2012 Canadian consensus statements on the treatment of hospitalised adult patients with severe UC versus standard care ( ). Outcomes The length of the interventions ranged from 30 minutes, in , to 12 months in . Primary outcome: Disease activity Only four studies mentioned disease activity as an outcome. measured IBD disease activity as a continuous outcome using the Bowel Disease Activity Index (GIBDI), and used the Crohn's Disease Harvey‐Bradshaw Index (HBI) for CD participants and the Simple Clinical Colitis Activity Index (SCCAI) for patients with UC/indeterminate colitis. In the authors stated disease activity as an outcome, and that they measured it using the Colitis Activity Index (CAI), however the data were not presented. reported the numbers of participants with mild and severe disease at each stage of the study. Primary outcome: Flare‐ups or relapse Five studies measured flare‐ups or relapse. and evaluated mean number of flare‐ups (SD) during the study as continuous data. reported numbers with acute relapse per group with relapse defined as clinical activity index ≥ 9. and also reported numbers of patients with relapse during the study. Primary outcome: Quality of life Ten studies reported quality of life ( ; ; ; ; ; ; ; ; ; ). The Inflammatory Bowel Disease Questionnaire (IBDQ) was used in seven studies, ( ; ; ; ; ; ; ). The short Inflammatory Bowel Disease Questionnaire (SIBDQ) was used by and . The SF‐12 short form health survey was used by . also used the Quality Index in Crohn’s and Colitis (QuICC) questionnaire, and the Rating Form of IBD Patient Concerns (RFIPC). Secondary outcome: Number of episodes accessing health care Four studies stated the number of episodes of accessing health care ( ; ; ; ). reported total encounters, IBD‐related hospitalisations, non‐IBD‐related hospitalisations, non‐invasive diagnostic tests, electronic encounters and telephone encounters, all as rates, adjusted for 100 participants per year. reported hospital admissions and emergency visits, reported kept hospital appointments and numbers of patients who did not attend. measured hospitalisations, and rate of healthcare use. Secondary outcome: Change in disease activity No studies reported this outcome. Secondary outcome: Change in quality of life Only one study reported the change in quality of life in its participants ( ). The study used the IBDQ (the questionniare has 32 questions and the score ranges from a minimum of 32 to a maximum of 224, but the authors presented results as mean scores for each question with a range; high score = better result) and the QuICC (range 1 = excellent to 5 = poor) at the start and after two weeks of the intervention to report the mean values (SD) on its sample. Secondary outcome: Medication adherence Five studies measured medication adherence ( ; ; ; ; ). , , and used the Morisky Medication Adherence Scale. reported adherence rates based on recordings with the MedMinder system. reported incidents and rates of missed medications, and rate of non‐adherence as measured by the Patient Satisfaction Questionnaire and participant self‐report. Secondary outcome: Patient knowledge and/or skill Patient knowledge/skills was reported in seven studies ( ; ; ; ; ; ; ). measured knowledge using the Crohn’s and Colitis Knowledge questionnaire, while used the IBD knowledge Inventory Device (IBD‐KID) and a modified version of the Crohn's & Colitis Foundation of America (CCFA) Knowledge Score (I‐M‐AWARE). used both the Chron's and Colitis Knowledge (CCKNOW) questionnaire and the Knowledge questionnaire (KQ), while it also assessed self‐perceived knowledge on a visual analogue scale (VAS). used the ECIPE (Étude randomisée et contrôlée évaluant l'impact du programme d'éducation (Controlled multicentre study of the impact of a programme of therapeutic Education in IBD)) score they developed for their education programme and defined success as a dichotomous outcome of improvement in patients' skills by an increase of the ECIPE score of more than 20%, from baseline to six months. In medical and psychological knowledge was self‐reported by the participants on a Likert scale, while in IBD knowledge and medication knowledge were self‐reported on a VAS. Secondary outcome: Total adverse events (serious and minor) Only two studies reported total adverse events ( ; ). Secondary outcome: Withdrawals due to adverse events Only three studies reported this outcome ( ; ; ). There were no withdrawals due to adverse events in these studies as no participant reported any adverse events related to use of the telemedicine intervention. Qualitative synthesis: Educational content The details on the contents of each intervention can be found in . Five studies relied on face‐to‐face workshops, seminars or teaching session for delivering their educational content ( ; ; ; ; ). Five used e‐learning or distance learning via mobile phones ( ; ; ; ; ). Three studies used written material as their primary material ( ; ; ). One study used mixed methods of lectures and group therapy for delivering information on IBD and psychological coping methods for IBD, respectively ( ). The educational learning outcomes were not clearly stated in any of the studies. Some studies mentioned generic aims such as empowering patients ( ), enhancing the sense of control and skills in coping with relapses ( ), and a greater sense of control in management, engagement in the care process and understanding of the overall management plan ( ). None of the studies described the educational theoretical underpinning of their interventions. Qualitative synthesis: Teaching attributes of training programmes used (staff and resource requirements, length of course, any follow‐up service or session) Six studies employed synchronous interventions ( ; ; ; ; ; ), and six asynchronous interventions ( ; ; ; ; ; ). Two studies were a mix of synchronous and asynchronous ( ; ). Three interventions were part of a package of measures that contained other elements as well ( ; ; ). Staff delivering the interventions included nurses, gastroenterologists and other physicians, psychologists, dietitians, medical social workers and educators. Resources included computers, tablets, smartphones, booklets and other written materials, as well as physical space and equipment for delivering workshops or lectures. Access issues included participants with insufficient language skills, severe vision or hearing impairments, serious physical or psychological comorbidities, people without access to computers, tablets, or smartphones and non‐access to transport ( ). Qualitative synthesis: Knowledge assessments (formative or summative assessment) Four of the five studies that assessed patient knowledge used summative assessment ( ; ; ; ); we did not have enough information to judge the type of assessment in . The pre‐ and post‐knowledge scores, or changes in knowledge scores from baseline, are presented in . Funding sources and conflicts of interest Nine studies reported their sources of funding ( ; ; ; ; ; ; ; ). Four studies were funded via government grants ( ; ; ; ), three studies by private sources ( ; ; ), one study by a non‐profit research association ( ), and one study declared that it received no financial support ( ). Five studies did not report anything about their source of funding ( ; ; ; ; ). Eight studies made declarations about conflicts of interest ( ; ; ; ; ; ; ; ), and five of these declared no conflicts of interest ( ; ; ; ; ). One study declared that one of the authors was an employee of the industrial partner that provided funding ( ), one study declared that several authors received honoraria from private industrial partners ( ), and one study declared that several authors had connections to healthcare companies unrelated to the study ( ). Six studies did not make any declarations about conflicts of interest ( ; ; ; ; ; ). Fourteen RCTs involving a total of 2708 participants met our inclusion criteria. Three studies were conducted in the USA ( ; ; ), three in Canada ( ; ; ), two in Germany ( ; ), two in Sweden ( ; ), one in the UK ( ), one in France ( ), one in the Netherlands ( ), and one in Turkey ( ). All the included studies were conducted in hospitals and tertiary centres . Seven studies were single‐centre ( ; ; ; ; ; ; ), and seven were multi‐centre ( ; ; ; ; ; ; ). Two studies were cluster‐RCTs ( ; ). Age ranged from 11 years in to 75 years in . There were two studies in paediatric populations ( ; ). included adolescents between 11 and 18 years of age, and participants between 11 and 21 years of age. Both interventions were targeted towards the participating adolescents and not towards their caregivers. Two studies examined exclusively ulcerative colitis (UC) populations ( ; ), whilst the remaining studies examined a mix of IBD patients ( ; ; ; ; ; ; ; ; ; ; ; ). Six studies examined participants in both active and inactive states of the disease ( ; ; ; ; ; ); two studies examined participants in an inactive state of the disease ( ; ); one study examined participants in an active state of the disease ( ); two studies examined participants in remission or low disease activity ( ; ). One study reported the disease activity of its participants as a mean value using the Crohn's Disease Activity Index (CDAI) and the Activity Index (AI) ( ). Two studies did not report on activity of the disease ( ; ). Four of the studies had trial registrations ( ; ; ; ). The following interventions were assessed in the included trials. A 2‐part patient education seminar versus “ treatment as usual” ( ). Information booklets available from the Crohn’s and Colitis Foundation of Canada versus “ usual care” ( ). Weekly educational text messages versus once every other week educational text messages versus routine clinic visits ( ). E‐learning module accessible via telemedicine system (myIBDcoach) versus routine follow‐up visits ( ). Multi professional group‐based education programme versus regular information during visits to the IBD clinic ( ). Guidebooks for Crohn's Disease (CD) and UC versus “ standard care” ( ). Education programme delivered by a dedicated staff using an illustrated book versus no intervention ( ). A standardised education programme, followed by a group session versus standard care ( ). Nine sessions of lectures alternating with group therapy versus conventional “on demand” medical and psychosocial/psychological treatment ( ). Web‐based education versus education which presented information via easy‐to‐read, illustrated, colour‐printed books (educational content was exactly the same for both groups) ( ). A 30‐minute educational session using the IBD Pocket Guide versus usual care ( ). Internet blog access versus the receipt of text messaging versus Internet blog access and receipt of text messaging versus standard care ( ). Structured education programme and standard care versus standard care consisting of physician visits, at the discretion of the physicians and patients, with physician‐directed ad hoc teaching during visits and the presentation of printed educational literature ( ). Original, interactive video that provided a summary of the 2012 Canadian consensus statements on the treatment of hospitalised adult patients with severe UC versus standard care ( ). The length of the interventions ranged from 30 minutes, in , to 12 months in . Primary outcome: Disease activity Only four studies mentioned disease activity as an outcome. measured IBD disease activity as a continuous outcome using the Bowel Disease Activity Index (GIBDI), and used the Crohn's Disease Harvey‐Bradshaw Index (HBI) for CD participants and the Simple Clinical Colitis Activity Index (SCCAI) for patients with UC/indeterminate colitis. In the authors stated disease activity as an outcome, and that they measured it using the Colitis Activity Index (CAI), however the data were not presented. reported the numbers of participants with mild and severe disease at each stage of the study. Primary outcome: Flare‐ups or relapse Five studies measured flare‐ups or relapse. and evaluated mean number of flare‐ups (SD) during the study as continuous data. reported numbers with acute relapse per group with relapse defined as clinical activity index ≥ 9. and also reported numbers of patients with relapse during the study. Primary outcome: Quality of life Ten studies reported quality of life ( ; ; ; ; ; ; ; ; ; ). The Inflammatory Bowel Disease Questionnaire (IBDQ) was used in seven studies, ( ; ; ; ; ; ; ). The short Inflammatory Bowel Disease Questionnaire (SIBDQ) was used by and . The SF‐12 short form health survey was used by . also used the Quality Index in Crohn’s and Colitis (QuICC) questionnaire, and the Rating Form of IBD Patient Concerns (RFIPC). Secondary outcome: Number of episodes accessing health care Four studies stated the number of episodes of accessing health care ( ; ; ; ). reported total encounters, IBD‐related hospitalisations, non‐IBD‐related hospitalisations, non‐invasive diagnostic tests, electronic encounters and telephone encounters, all as rates, adjusted for 100 participants per year. reported hospital admissions and emergency visits, reported kept hospital appointments and numbers of patients who did not attend. measured hospitalisations, and rate of healthcare use. Secondary outcome: Change in disease activity No studies reported this outcome. Secondary outcome: Change in quality of life Only one study reported the change in quality of life in its participants ( ). The study used the IBDQ (the questionniare has 32 questions and the score ranges from a minimum of 32 to a maximum of 224, but the authors presented results as mean scores for each question with a range; high score = better result) and the QuICC (range 1 = excellent to 5 = poor) at the start and after two weeks of the intervention to report the mean values (SD) on its sample. Secondary outcome: Medication adherence Five studies measured medication adherence ( ; ; ; ; ). , , and used the Morisky Medication Adherence Scale. reported adherence rates based on recordings with the MedMinder system. reported incidents and rates of missed medications, and rate of non‐adherence as measured by the Patient Satisfaction Questionnaire and participant self‐report. Secondary outcome: Patient knowledge and/or skill Patient knowledge/skills was reported in seven studies ( ; ; ; ; ; ; ). measured knowledge using the Crohn’s and Colitis Knowledge questionnaire, while used the IBD knowledge Inventory Device (IBD‐KID) and a modified version of the Crohn's & Colitis Foundation of America (CCFA) Knowledge Score (I‐M‐AWARE). used both the Chron's and Colitis Knowledge (CCKNOW) questionnaire and the Knowledge questionnaire (KQ), while it also assessed self‐perceived knowledge on a visual analogue scale (VAS). used the ECIPE (Étude randomisée et contrôlée évaluant l'impact du programme d'éducation (Controlled multicentre study of the impact of a programme of therapeutic Education in IBD)) score they developed for their education programme and defined success as a dichotomous outcome of improvement in patients' skills by an increase of the ECIPE score of more than 20%, from baseline to six months. In medical and psychological knowledge was self‐reported by the participants on a Likert scale, while in IBD knowledge and medication knowledge were self‐reported on a VAS. Secondary outcome: Total adverse events (serious and minor) Only two studies reported total adverse events ( ; ). Secondary outcome: Withdrawals due to adverse events Only three studies reported this outcome ( ; ; ). There were no withdrawals due to adverse events in these studies as no participant reported any adverse events related to use of the telemedicine intervention. Qualitative synthesis: Educational content The details on the contents of each intervention can be found in . Five studies relied on face‐to‐face workshops, seminars or teaching session for delivering their educational content ( ; ; ; ; ). Five used e‐learning or distance learning via mobile phones ( ; ; ; ; ). Three studies used written material as their primary material ( ; ; ). One study used mixed methods of lectures and group therapy for delivering information on IBD and psychological coping methods for IBD, respectively ( ). The educational learning outcomes were not clearly stated in any of the studies. Some studies mentioned generic aims such as empowering patients ( ), enhancing the sense of control and skills in coping with relapses ( ), and a greater sense of control in management, engagement in the care process and understanding of the overall management plan ( ). None of the studies described the educational theoretical underpinning of their interventions. Qualitative synthesis: Teaching attributes of training programmes used (staff and resource requirements, length of course, any follow‐up service or session) Six studies employed synchronous interventions ( ; ; ; ; ; ), and six asynchronous interventions ( ; ; ; ; ; ). Two studies were a mix of synchronous and asynchronous ( ; ). Three interventions were part of a package of measures that contained other elements as well ( ; ; ). Staff delivering the interventions included nurses, gastroenterologists and other physicians, psychologists, dietitians, medical social workers and educators. Resources included computers, tablets, smartphones, booklets and other written materials, as well as physical space and equipment for delivering workshops or lectures. Access issues included participants with insufficient language skills, severe vision or hearing impairments, serious physical or psychological comorbidities, people without access to computers, tablets, or smartphones and non‐access to transport ( ). Qualitative synthesis: Knowledge assessments (formative or summative assessment) Four of the five studies that assessed patient knowledge used summative assessment ( ; ; ; ); we did not have enough information to judge the type of assessment in . The pre‐ and post‐knowledge scores, or changes in knowledge scores from baseline, are presented in . Only four studies mentioned disease activity as an outcome. measured IBD disease activity as a continuous outcome using the Bowel Disease Activity Index (GIBDI), and used the Crohn's Disease Harvey‐Bradshaw Index (HBI) for CD participants and the Simple Clinical Colitis Activity Index (SCCAI) for patients with UC/indeterminate colitis. In the authors stated disease activity as an outcome, and that they measured it using the Colitis Activity Index (CAI), however the data were not presented. reported the numbers of participants with mild and severe disease at each stage of the study. Five studies measured flare‐ups or relapse. and evaluated mean number of flare‐ups (SD) during the study as continuous data. reported numbers with acute relapse per group with relapse defined as clinical activity index ≥ 9. and also reported numbers of patients with relapse during the study. Ten studies reported quality of life ( ; ; ; ; ; ; ; ; ; ). The Inflammatory Bowel Disease Questionnaire (IBDQ) was used in seven studies, ( ; ; ; ; ; ; ). The short Inflammatory Bowel Disease Questionnaire (SIBDQ) was used by and . The SF‐12 short form health survey was used by . also used the Quality Index in Crohn’s and Colitis (QuICC) questionnaire, and the Rating Form of IBD Patient Concerns (RFIPC). Four studies stated the number of episodes of accessing health care ( ; ; ; ). reported total encounters, IBD‐related hospitalisations, non‐IBD‐related hospitalisations, non‐invasive diagnostic tests, electronic encounters and telephone encounters, all as rates, adjusted for 100 participants per year. reported hospital admissions and emergency visits, reported kept hospital appointments and numbers of patients who did not attend. measured hospitalisations, and rate of healthcare use. No studies reported this outcome. Only one study reported the change in quality of life in its participants ( ). The study used the IBDQ (the questionniare has 32 questions and the score ranges from a minimum of 32 to a maximum of 224, but the authors presented results as mean scores for each question with a range; high score = better result) and the QuICC (range 1 = excellent to 5 = poor) at the start and after two weeks of the intervention to report the mean values (SD) on its sample. Five studies measured medication adherence ( ; ; ; ; ). , , and used the Morisky Medication Adherence Scale. reported adherence rates based on recordings with the MedMinder system. reported incidents and rates of missed medications, and rate of non‐adherence as measured by the Patient Satisfaction Questionnaire and participant self‐report. Patient knowledge/skills was reported in seven studies ( ; ; ; ; ; ; ). measured knowledge using the Crohn’s and Colitis Knowledge questionnaire, while used the IBD knowledge Inventory Device (IBD‐KID) and a modified version of the Crohn's & Colitis Foundation of America (CCFA) Knowledge Score (I‐M‐AWARE). used both the Chron's and Colitis Knowledge (CCKNOW) questionnaire and the Knowledge questionnaire (KQ), while it also assessed self‐perceived knowledge on a visual analogue scale (VAS). used the ECIPE (Étude randomisée et contrôlée évaluant l'impact du programme d'éducation (Controlled multicentre study of the impact of a programme of therapeutic Education in IBD)) score they developed for their education programme and defined success as a dichotomous outcome of improvement in patients' skills by an increase of the ECIPE score of more than 20%, from baseline to six months. In medical and psychological knowledge was self‐reported by the participants on a Likert scale, while in IBD knowledge and medication knowledge were self‐reported on a VAS. Only two studies reported total adverse events ( ; ). Only three studies reported this outcome ( ; ; ). There were no withdrawals due to adverse events in these studies as no participant reported any adverse events related to use of the telemedicine intervention. The details on the contents of each intervention can be found in . Five studies relied on face‐to‐face workshops, seminars or teaching session for delivering their educational content ( ; ; ; ; ). Five used e‐learning or distance learning via mobile phones ( ; ; ; ; ). Three studies used written material as their primary material ( ; ; ). One study used mixed methods of lectures and group therapy for delivering information on IBD and psychological coping methods for IBD, respectively ( ). The educational learning outcomes were not clearly stated in any of the studies. Some studies mentioned generic aims such as empowering patients ( ), enhancing the sense of control and skills in coping with relapses ( ), and a greater sense of control in management, engagement in the care process and understanding of the overall management plan ( ). None of the studies described the educational theoretical underpinning of their interventions. Six studies employed synchronous interventions ( ; ; ; ; ; ), and six asynchronous interventions ( ; ; ; ; ; ). Two studies were a mix of synchronous and asynchronous ( ; ). Three interventions were part of a package of measures that contained other elements as well ( ; ; ). Staff delivering the interventions included nurses, gastroenterologists and other physicians, psychologists, dietitians, medical social workers and educators. Resources included computers, tablets, smartphones, booklets and other written materials, as well as physical space and equipment for delivering workshops or lectures. Access issues included participants with insufficient language skills, severe vision or hearing impairments, serious physical or psychological comorbidities, people without access to computers, tablets, or smartphones and non‐access to transport ( ). Four of the five studies that assessed patient knowledge used summative assessment ( ; ; ; ); we did not have enough information to judge the type of assessment in . The pre‐ and post‐knowledge scores, or changes in knowledge scores from baseline, are presented in . Nine studies reported their sources of funding ( ; ; ; ; ; ; ; ). Four studies were funded via government grants ( ; ; ; ), three studies by private sources ( ; ; ), one study by a non‐profit research association ( ), and one study declared that it received no financial support ( ). Five studies did not report anything about their source of funding ( ; ; ; ; ). Eight studies made declarations about conflicts of interest ( ; ; ; ; ; ; ; ), and five of these declared no conflicts of interest ( ; ; ; ; ). One study declared that one of the authors was an employee of the industrial partner that provided funding ( ), one study declared that several authors received honoraria from private industrial partners ( ), and one study declared that several authors had connections to healthcare companies unrelated to the study ( ). Six studies did not make any declarations about conflicts of interest ( ; ; ; ; ; ). We excluded 37 full‐text studies (44 records) for various reasons. The reasons for exclusion of each study are presented in the table and are summarised below. Wrong intervention (23 studies) ( ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ). Not RCTs (12 studies) ( ; ; ; ; ; ; ; ; ; ; ; ). Wrong population: (2 studies) ( ; ). There are 20 studies awaiting classification ( ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ). There are six ongoing RCTs ( ; ; ; ; ; ; ). Below we present the results of our risk of bias assessment ( ; ). Further details can be found in the risk of bias tables (beneath tables). Allocation Randomisation was described clearly in seven studies ( ; ; ; ; ; ; ), which we rated at low for risk of bias, and was not sufficiently described in the other seven studies ( ; ; ; ; ; ; ; ), which we rated unclear for risk of bias. We rated two studies at low risk from allocation concealment ( ; ), as the method of random allocation of participants to intervention and control groups and allocation concealment was described or the risk was low due to cluster randomisation. We rated nine studies at unclear risk of allocation concealment ( ; ; ; ; ; ; ; ; ), as they did not provide enough information about their selection and allocation concealment process. Three studies had no allocation concealment and were judged to be at high risk ( ; ; ). Blinding All studies were rated as high in performance bias, as the interventions they studied could not be blinded for both participants and personnel. Detection bias was rated as low in two studies that mentioned assessors being blinded ( ; ), unclear in six studies that did not provide enough information for a judgement ( ; ; ; ; ; ), and high in six that confirmed or mentioned that the assessors were not blinded ( ; ; ; ; ; ). Incomplete outcome data We judged attrition bias as low in nine studies that provided enough information for judgement ( ; ; ; ; ; ; ; ; ). The rest of the studies we rated at unclear risk ( ; ; ; ; ). Selective reporting We rated reporting bias as low in three studies that reported all outcomes they had set out to report either in their protocols or trial registrations ( ; ; ). We rated nine studies at unclear risk ( ; ; ; ; ; ; ; ; ), and two at high risk ( ; ). Other potential sources of bias We rated 12 studies as low risk for other potential sources of bias ( ; ; ; ; ; ; ; ; ; ; ; ). We rated two studies at unclear risk due to lack of information ( ; ). Randomisation was described clearly in seven studies ( ; ; ; ; ; ; ), which we rated at low for risk of bias, and was not sufficiently described in the other seven studies ( ; ; ; ; ; ; ; ), which we rated unclear for risk of bias. We rated two studies at low risk from allocation concealment ( ; ), as the method of random allocation of participants to intervention and control groups and allocation concealment was described or the risk was low due to cluster randomisation. We rated nine studies at unclear risk of allocation concealment ( ; ; ; ; ; ; ; ; ), as they did not provide enough information about their selection and allocation concealment process. Three studies had no allocation concealment and were judged to be at high risk ( ; ; ). All studies were rated as high in performance bias, as the interventions they studied could not be blinded for both participants and personnel. Detection bias was rated as low in two studies that mentioned assessors being blinded ( ; ), unclear in six studies that did not provide enough information for a judgement ( ; ; ; ; ; ), and high in six that confirmed or mentioned that the assessors were not blinded ( ; ; ; ; ; ). We judged attrition bias as low in nine studies that provided enough information for judgement ( ; ; ; ; ; ; ; ; ). The rest of the studies we rated at unclear risk ( ; ; ; ; ). We rated reporting bias as low in three studies that reported all outcomes they had set out to report either in their protocols or trial registrations ( ; ; ). We rated nine studies at unclear risk ( ; ; ; ; ; ; ; ; ), and two at high risk ( ; ). We rated 12 studies as low risk for other potential sources of bias ( ; ; ; ; ; ; ; ; ; ; ; ). We rated two studies at unclear risk due to lack of information ( ; ). See: ; ; A summary of primary and secondary outcome data can be found in and respectively. Any planned subgroup and sensitivity analyses that were not carried out because of a lack of data are mentioned in . 1. Patient education and standard care versus standard care Thirteen studies compared patient education interventions against no intervention ( , ; ; ; ; ; ; ; ; ; ; ; ). Primary outcomes Disease activity at study end Two of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ). There was no clear difference in disease activity when patient education (n = 277) combined with standard care was compared to standard care (n = 202). Patient education combined with standard care is probably equivalent to standard care in reducing disease activity in patients with IBD (standardised mean difference (SMD ‐0.03, 95% confidence interval (CI) ‐0.25 to 0.20). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned disease activity as an outcome, but did not present any results. Flare‐ups or relapse Two of the studies that reported this outcome reported it as a continuous outcome ( ; ), and three reported it as a dichotomous outcome ( ; ; ). For the continuous data meta‐analysis, there was no clear difference for flare‐ups or relapse when patient education (n = 515) combined with standard care was compared to standard care (n = 507), as a continuous outcome. Patient education combined with standard care is probably equivalent to standard care in reducing flare‐ups or relapse in patients with IBD (mean difference (MD) ‐0.00, 95% CI ‐0.06 to 0.05). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). From the dichotomous data, 10 participants experienced relapse in the patient education combined with standard care group (n = 157) and 10 participants experienced relapse in the standard care group (n = 150). The evidence is very uncertain on whether patient education combined with standard care is different to standard care in reducing flare‐ups or relapse in patients with IBD (RR 0.94, 95% CI 0.41 to 2.18). The certainty of the evidence was very low due to serious concerns with risk of bias and imprecision ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned that one participant relapsed during their study but did not clarify to which group they belonged. Quality of life at study end Six of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 721) was compared to standard care (n = 643). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.08, 95% CI ‐0.03 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). We conducted a sensitivity analysis excluding five studies at high risk of bias ( ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 193) was compared to standard care (n = 107). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (MD 1.11, 95% CI ‐5.74 to 7.97). The certainty of the evidence was moderate due to imprecision ( ). We conducted a sensitivity analysis excluding one cluster RCT ( ). There was no clear difference in quality of life when patient education combined with standard care (n = 667) was compared to standard care (n = 571). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.07, 95% CI ‐0.04 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ). We conducted a further sensitivity analysis including only the studies that used the full IBDQ (high score = better result) and as such allowed the use of the mean difference (MD) ( ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 297) was compared to standard care (n = 217). Patient education combined with standard care may be equivalent to standard care in improving quality of life in patients with IBD (MD 1.82, 95% CI ‐3.72 to 7.36). The certainty of the evidence was low due to concerns with risk of bias and imprecision ( ). also measured mental quality of life, in addition to the physical quality of life that was included in the meta‐analysis. The intervention group had a reported score of 46.41 (11.00) and the control group a score of 42.70 (10.89) at 3 months from study end (high score = better result). measured quality of life using the QuICC (low score = better result), in addition to the IBDQ that was used in the above meta‐analysis. The intervention group had a reported score of 87.0 (20.61) and the control group a score of 85.7 (19.83) at study end. used the IBDQ and provided mean scores with variance values. The intervention group had a reported score of 57.85 and the control group a score of 55.58 at study end (high score = better result). and did not provide the raw mean and variance scores per group at study end, only presenting the results of their own analysis. Secondary outcomes Number of episodes of accessing health care In , hospitalisations, surgery, emergency department and office visits, procedures, intravenous therapeutics, and telephone and electronic encounters were extracted from the electronic medical record (EMR) for one year before and after randomisation, and encounters were reported as rates, adjusted for 100 participants per year. The intervention group that received a telemedicine message every other week (IG1 (TELE‐IBD EOW)) had 2235 total encounters, the intervention group that received a telemedicine message every week (IG2 (TELE‐IBD W)) had 1935, and the control group had 2099 (the data on the specific types of encounters are presented in ). reported mean numbers of hospital admissions, which were 0.05 (SD 0.28) for the intervention group and 0.10 (SD 0.43) for the control group; and mean numbers of emergency visits, which were 0.07 (SD 0.35) for the intervention group and 0.10 (SD 0.54) for the control group. reported mean number of kept hospital appointments as 1.9 (SD 2.2) for the intervention group and 3.0 (SD 2.5) for the control group, as well as number of participants who did not attend appointments as 22/279 for the intervention group and 44/403 for the control group. reported rate of health care use as a mean of 0.63 for the intervention group and 0.95 for the control group without providing variance values. mentioned it as an outcome, but did not report data. Change in disease activity This outcome was not reported in any of the studies. Change in quality of life This outcome was only reported in . The mean difference in the intervention group was −0.17 (SD 0.49) and in the control group 0.28 (SD 0.62) for the IBDQ and −0.05 (SD 0.28) and −0.01 (SD 0.25), respectively, for the QuICC. Medication adherence reported medication adherence as a mean of 7.01 (SD 1.40) for the intervention group and 6.77 (SD 1.61) for the control group. reported 66/126 and 64/122 as non‐adherent in the intervention and control groups, respectively. In , difference in average adherence rates pre‐ and post‐randomisation was +0.36 (SD 10.28) for the intervention group and −15.3 (SD 25.34) for the control group. reported 166 incidents of missed medications, with a mean of 2.31 incidents per participant, and calculated the mean number of missed medications during the study as 0.91 for the intervention group and 3.43 for the control group. did not provide the raw mean and variance scores per group at study end, instead the authors presented the results of their own analysis. Patient knowledge or skill at study end In , the mean difference from baseline (no variance provided) was +2.4 in the TELE‐IBD EOW intervention group +2.0 in the TELE‐IBD W intervention group and +1.8 in the control group. reported that mean rank scores (no variance provided; high score = better result) at end of study were: 5.8 for the intervention group and 4.0 for the control group for gastrointestinal anatomy; 5.6 and 4.3 for general IBD knowledge; 6.1 and 3.6 for medications; and 4.2 and 6.0 for nutrition. reported that post‐intervention the mean score on the assessment was 55.6% (range 35.0% to 95.6%), but did not report results per intervention group. reported CCKNOW scores of 19.52 (SD 2.55) for the intervention group and 13.84 (SD 4.86) for the control group, and KQ scores of 27.19 (SD 3.03) and 21.47 (SD 6.81) respectively, at study end. In , an improvement in patients' skills was defined by an increase of the ECIPE score of more than 20%, from baseline to six months. In the intervention group 61 patients achieved that and 31 in the control. Per protocol median ECIPE scores were reported as 26 (range 22‐30) in the intervention group (n = 105) and 20 (range 16‐25) (n = 117) in the control group. In each of the results in this paragraph, higher scores indicate improvement. Self‐reported medical knowledge was reported in three studies as 4.05 (SD 0.41) for the intervention group and 3.42 (SD 0.71) for the control and psychological knowledge as 3.65 (SD 0.67) and 2.98 (SD 0.74), respectively in . Knowledge of IBD was reported as 8.17 (SD 1.16) for the intervention group and 7.84 (SD 1.47) for the control group, and knowledge of medication as 7.75 (SD 1.58) and 7.58 (SD 1.51), respectively in . Self‐perceived knowledge was reported as 7.6 for the intervention group and 6.2 for the control group at study end in . Total adverse effects , , and reported zero total adverse effects in their studies. Withdrawals due to adverse events The only study that reported withdrawals due to adverse effects was , which reported that in the TELE‐IBD EOW intervention group one participant withdrew due to breast cancer and in the TELE‐IBD intervention group two participants withdrew because they needed surgery. No participants withdrew due to adverse effects from the control group. 2. Web‐based patient education versus other delivery of patient education Two studies compared delivery methods of patient education in the form of web‐based interventions against other delivery methods ( ; ). Primary outcomes Only reported any of our primary outcomes. Disease activity at study end reported numbers of UC and CD participants in remission, or with mild, severe, or very severe disease at study end. For UC participants, 8/16 in the web‐based group and 10/16 in the control education group were in remission, 6/16 and 4/16 had mild disease, 2/16 and 1/16 had severe disease, and 0/16 and 0/16 had very severe disease. For CD participants, 5/14 and 10/14 were in remission, 7/14 and 3/14 had mild disease, 2/14 and 1/14 had severe disease, and 0/14 and 0/14 had very severe disease. Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life score on the IBDQ for the web‐based group was 156.53 (SD 30.97) and 155.63 (SD 34.30) for the control group (high score = better result). Secondary outcomes No secondary outcomes were reported except for the limited knowledge score data in , which we reported above. 3. Weekly educational texts messages versus once every other week educational text messages compared frequency of patient education in the form of weekly educational text messages versus once every other week educational text messages (in addition to comparing these interventions to standard care, the results of which we included in the patient education and standard care versus standard care comparison above). Primary outcomes Disease activity at study end Mean disease activity for the TELE‐EOW CD participants was 4.2 (SD 3.9) and for the TELE‐W CD participants 3.2 (SD 3.4). Mean disease activity for the TELE‐EOW UC participants was 1.7 (SD 1.9) and for the TELE‐W UC participants was 2.0 (SD 1.8). Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life scores for the TELE‐EOW participants was 181.5 (SD 28.2) and for the TELE‐W participants was 179.2 (SD 32.8) Secondary outcomes These have been reported in Comparison 1, patient education and standard care versus standard care. Thirteen studies compared patient education interventions against no intervention ( , ; ; ; ; ; ; ; ; ; ; ; ). Primary outcomes Disease activity at study end Two of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ). There was no clear difference in disease activity when patient education (n = 277) combined with standard care was compared to standard care (n = 202). Patient education combined with standard care is probably equivalent to standard care in reducing disease activity in patients with IBD (standardised mean difference (SMD ‐0.03, 95% confidence interval (CI) ‐0.25 to 0.20). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned disease activity as an outcome, but did not present any results. Flare‐ups or relapse Two of the studies that reported this outcome reported it as a continuous outcome ( ; ), and three reported it as a dichotomous outcome ( ; ; ). For the continuous data meta‐analysis, there was no clear difference for flare‐ups or relapse when patient education (n = 515) combined with standard care was compared to standard care (n = 507), as a continuous outcome. Patient education combined with standard care is probably equivalent to standard care in reducing flare‐ups or relapse in patients with IBD (mean difference (MD) ‐0.00, 95% CI ‐0.06 to 0.05). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). From the dichotomous data, 10 participants experienced relapse in the patient education combined with standard care group (n = 157) and 10 participants experienced relapse in the standard care group (n = 150). The evidence is very uncertain on whether patient education combined with standard care is different to standard care in reducing flare‐ups or relapse in patients with IBD (RR 0.94, 95% CI 0.41 to 2.18). The certainty of the evidence was very low due to serious concerns with risk of bias and imprecision ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned that one participant relapsed during their study but did not clarify to which group they belonged. Quality of life at study end Six of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 721) was compared to standard care (n = 643). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.08, 95% CI ‐0.03 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). We conducted a sensitivity analysis excluding five studies at high risk of bias ( ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 193) was compared to standard care (n = 107). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (MD 1.11, 95% CI ‐5.74 to 7.97). The certainty of the evidence was moderate due to imprecision ( ). We conducted a sensitivity analysis excluding one cluster RCT ( ). There was no clear difference in quality of life when patient education combined with standard care (n = 667) was compared to standard care (n = 571). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.07, 95% CI ‐0.04 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ). We conducted a further sensitivity analysis including only the studies that used the full IBDQ (high score = better result) and as such allowed the use of the mean difference (MD) ( ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 297) was compared to standard care (n = 217). Patient education combined with standard care may be equivalent to standard care in improving quality of life in patients with IBD (MD 1.82, 95% CI ‐3.72 to 7.36). The certainty of the evidence was low due to concerns with risk of bias and imprecision ( ). also measured mental quality of life, in addition to the physical quality of life that was included in the meta‐analysis. The intervention group had a reported score of 46.41 (11.00) and the control group a score of 42.70 (10.89) at 3 months from study end (high score = better result). measured quality of life using the QuICC (low score = better result), in addition to the IBDQ that was used in the above meta‐analysis. The intervention group had a reported score of 87.0 (20.61) and the control group a score of 85.7 (19.83) at study end. used the IBDQ and provided mean scores with variance values. The intervention group had a reported score of 57.85 and the control group a score of 55.58 at study end (high score = better result). and did not provide the raw mean and variance scores per group at study end, only presenting the results of their own analysis. Secondary outcomes Number of episodes of accessing health care In , hospitalisations, surgery, emergency department and office visits, procedures, intravenous therapeutics, and telephone and electronic encounters were extracted from the electronic medical record (EMR) for one year before and after randomisation, and encounters were reported as rates, adjusted for 100 participants per year. The intervention group that received a telemedicine message every other week (IG1 (TELE‐IBD EOW)) had 2235 total encounters, the intervention group that received a telemedicine message every week (IG2 (TELE‐IBD W)) had 1935, and the control group had 2099 (the data on the specific types of encounters are presented in ). reported mean numbers of hospital admissions, which were 0.05 (SD 0.28) for the intervention group and 0.10 (SD 0.43) for the control group; and mean numbers of emergency visits, which were 0.07 (SD 0.35) for the intervention group and 0.10 (SD 0.54) for the control group. reported mean number of kept hospital appointments as 1.9 (SD 2.2) for the intervention group and 3.0 (SD 2.5) for the control group, as well as number of participants who did not attend appointments as 22/279 for the intervention group and 44/403 for the control group. reported rate of health care use as a mean of 0.63 for the intervention group and 0.95 for the control group without providing variance values. mentioned it as an outcome, but did not report data. Change in disease activity This outcome was not reported in any of the studies. Change in quality of life This outcome was only reported in . The mean difference in the intervention group was −0.17 (SD 0.49) and in the control group 0.28 (SD 0.62) for the IBDQ and −0.05 (SD 0.28) and −0.01 (SD 0.25), respectively, for the QuICC. Medication adherence reported medication adherence as a mean of 7.01 (SD 1.40) for the intervention group and 6.77 (SD 1.61) for the control group. reported 66/126 and 64/122 as non‐adherent in the intervention and control groups, respectively. In , difference in average adherence rates pre‐ and post‐randomisation was +0.36 (SD 10.28) for the intervention group and −15.3 (SD 25.34) for the control group. reported 166 incidents of missed medications, with a mean of 2.31 incidents per participant, and calculated the mean number of missed medications during the study as 0.91 for the intervention group and 3.43 for the control group. did not provide the raw mean and variance scores per group at study end, instead the authors presented the results of their own analysis. Patient knowledge or skill at study end In , the mean difference from baseline (no variance provided) was +2.4 in the TELE‐IBD EOW intervention group +2.0 in the TELE‐IBD W intervention group and +1.8 in the control group. reported that mean rank scores (no variance provided; high score = better result) at end of study were: 5.8 for the intervention group and 4.0 for the control group for gastrointestinal anatomy; 5.6 and 4.3 for general IBD knowledge; 6.1 and 3.6 for medications; and 4.2 and 6.0 for nutrition. reported that post‐intervention the mean score on the assessment was 55.6% (range 35.0% to 95.6%), but did not report results per intervention group. reported CCKNOW scores of 19.52 (SD 2.55) for the intervention group and 13.84 (SD 4.86) for the control group, and KQ scores of 27.19 (SD 3.03) and 21.47 (SD 6.81) respectively, at study end. In , an improvement in patients' skills was defined by an increase of the ECIPE score of more than 20%, from baseline to six months. In the intervention group 61 patients achieved that and 31 in the control. Per protocol median ECIPE scores were reported as 26 (range 22‐30) in the intervention group (n = 105) and 20 (range 16‐25) (n = 117) in the control group. In each of the results in this paragraph, higher scores indicate improvement. Self‐reported medical knowledge was reported in three studies as 4.05 (SD 0.41) for the intervention group and 3.42 (SD 0.71) for the control and psychological knowledge as 3.65 (SD 0.67) and 2.98 (SD 0.74), respectively in . Knowledge of IBD was reported as 8.17 (SD 1.16) for the intervention group and 7.84 (SD 1.47) for the control group, and knowledge of medication as 7.75 (SD 1.58) and 7.58 (SD 1.51), respectively in . Self‐perceived knowledge was reported as 7.6 for the intervention group and 6.2 for the control group at study end in . Total adverse effects , , and reported zero total adverse effects in their studies. Withdrawals due to adverse events The only study that reported withdrawals due to adverse effects was , which reported that in the TELE‐IBD EOW intervention group one participant withdrew due to breast cancer and in the TELE‐IBD intervention group two participants withdrew because they needed surgery. No participants withdrew due to adverse effects from the control group. Disease activity at study end Two of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ). There was no clear difference in disease activity when patient education (n = 277) combined with standard care was compared to standard care (n = 202). Patient education combined with standard care is probably equivalent to standard care in reducing disease activity in patients with IBD (standardised mean difference (SMD ‐0.03, 95% confidence interval (CI) ‐0.25 to 0.20). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned disease activity as an outcome, but did not present any results. Flare‐ups or relapse Two of the studies that reported this outcome reported it as a continuous outcome ( ; ), and three reported it as a dichotomous outcome ( ; ; ). For the continuous data meta‐analysis, there was no clear difference for flare‐ups or relapse when patient education (n = 515) combined with standard care was compared to standard care (n = 507), as a continuous outcome. Patient education combined with standard care is probably equivalent to standard care in reducing flare‐ups or relapse in patients with IBD (mean difference (MD) ‐0.00, 95% CI ‐0.06 to 0.05). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). From the dichotomous data, 10 participants experienced relapse in the patient education combined with standard care group (n = 157) and 10 participants experienced relapse in the standard care group (n = 150). The evidence is very uncertain on whether patient education combined with standard care is different to standard care in reducing flare‐ups or relapse in patients with IBD (RR 0.94, 95% CI 0.41 to 2.18). The certainty of the evidence was very low due to serious concerns with risk of bias and imprecision ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned that one participant relapsed during their study but did not clarify to which group they belonged. Quality of life at study end Six of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 721) was compared to standard care (n = 643). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.08, 95% CI ‐0.03 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). We conducted a sensitivity analysis excluding five studies at high risk of bias ( ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 193) was compared to standard care (n = 107). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (MD 1.11, 95% CI ‐5.74 to 7.97). The certainty of the evidence was moderate due to imprecision ( ). We conducted a sensitivity analysis excluding one cluster RCT ( ). There was no clear difference in quality of life when patient education combined with standard care (n = 667) was compared to standard care (n = 571). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.07, 95% CI ‐0.04 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ). We conducted a further sensitivity analysis including only the studies that used the full IBDQ (high score = better result) and as such allowed the use of the mean difference (MD) ( ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 297) was compared to standard care (n = 217). Patient education combined with standard care may be equivalent to standard care in improving quality of life in patients with IBD (MD 1.82, 95% CI ‐3.72 to 7.36). The certainty of the evidence was low due to concerns with risk of bias and imprecision ( ). also measured mental quality of life, in addition to the physical quality of life that was included in the meta‐analysis. The intervention group had a reported score of 46.41 (11.00) and the control group a score of 42.70 (10.89) at 3 months from study end (high score = better result). measured quality of life using the QuICC (low score = better result), in addition to the IBDQ that was used in the above meta‐analysis. The intervention group had a reported score of 87.0 (20.61) and the control group a score of 85.7 (19.83) at study end. used the IBDQ and provided mean scores with variance values. The intervention group had a reported score of 57.85 and the control group a score of 55.58 at study end (high score = better result). and did not provide the raw mean and variance scores per group at study end, only presenting the results of their own analysis. Two of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ). There was no clear difference in disease activity when patient education (n = 277) combined with standard care was compared to standard care (n = 202). Patient education combined with standard care is probably equivalent to standard care in reducing disease activity in patients with IBD (standardised mean difference (SMD ‐0.03, 95% confidence interval (CI) ‐0.25 to 0.20). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned disease activity as an outcome, but did not present any results. Two of the studies that reported this outcome reported it as a continuous outcome ( ; ), and three reported it as a dichotomous outcome ( ; ; ). For the continuous data meta‐analysis, there was no clear difference for flare‐ups or relapse when patient education (n = 515) combined with standard care was compared to standard care (n = 507), as a continuous outcome. Patient education combined with standard care is probably equivalent to standard care in reducing flare‐ups or relapse in patients with IBD (mean difference (MD) ‐0.00, 95% CI ‐0.06 to 0.05). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). From the dichotomous data, 10 participants experienced relapse in the patient education combined with standard care group (n = 157) and 10 participants experienced relapse in the standard care group (n = 150). The evidence is very uncertain on whether patient education combined with standard care is different to standard care in reducing flare‐ups or relapse in patients with IBD (RR 0.94, 95% CI 0.41 to 2.18). The certainty of the evidence was very low due to serious concerns with risk of bias and imprecision ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). mentioned that one participant relapsed during their study but did not clarify to which group they belonged. Six of the studies that reported this outcome provided continuous data that we could use for a meta‐analysis ( ; ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 721) was compared to standard care (n = 643). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.08, 95% CI ‐0.03 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ; ). A fixed‐effect sensitivity analysis had similar results ( ). We conducted a sensitivity analysis excluding five studies at high risk of bias ( ; ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 193) was compared to standard care (n = 107). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (MD 1.11, 95% CI ‐5.74 to 7.97). The certainty of the evidence was moderate due to imprecision ( ). We conducted a sensitivity analysis excluding one cluster RCT ( ). There was no clear difference in quality of life when patient education combined with standard care (n = 667) was compared to standard care (n = 571). Patient education combined with standard care is probably equivalent to standard care in improving quality of life in patients with IBD (SMD 0.07, 95% CI ‐0.04 to 0.18). The certainty of the evidence was moderate due to concerns with risk of bias ( ). We conducted a further sensitivity analysis including only the studies that used the full IBDQ (high score = better result) and as such allowed the use of the mean difference (MD) ( ; ; ; ). There was no clear difference in quality of life when patient education combined with standard care (n = 297) was compared to standard care (n = 217). Patient education combined with standard care may be equivalent to standard care in improving quality of life in patients with IBD (MD 1.82, 95% CI ‐3.72 to 7.36). The certainty of the evidence was low due to concerns with risk of bias and imprecision ( ). also measured mental quality of life, in addition to the physical quality of life that was included in the meta‐analysis. The intervention group had a reported score of 46.41 (11.00) and the control group a score of 42.70 (10.89) at 3 months from study end (high score = better result). measured quality of life using the QuICC (low score = better result), in addition to the IBDQ that was used in the above meta‐analysis. The intervention group had a reported score of 87.0 (20.61) and the control group a score of 85.7 (19.83) at study end. used the IBDQ and provided mean scores with variance values. The intervention group had a reported score of 57.85 and the control group a score of 55.58 at study end (high score = better result). and did not provide the raw mean and variance scores per group at study end, only presenting the results of their own analysis. Number of episodes of accessing health care In , hospitalisations, surgery, emergency department and office visits, procedures, intravenous therapeutics, and telephone and electronic encounters were extracted from the electronic medical record (EMR) for one year before and after randomisation, and encounters were reported as rates, adjusted for 100 participants per year. The intervention group that received a telemedicine message every other week (IG1 (TELE‐IBD EOW)) had 2235 total encounters, the intervention group that received a telemedicine message every week (IG2 (TELE‐IBD W)) had 1935, and the control group had 2099 (the data on the specific types of encounters are presented in ). reported mean numbers of hospital admissions, which were 0.05 (SD 0.28) for the intervention group and 0.10 (SD 0.43) for the control group; and mean numbers of emergency visits, which were 0.07 (SD 0.35) for the intervention group and 0.10 (SD 0.54) for the control group. reported mean number of kept hospital appointments as 1.9 (SD 2.2) for the intervention group and 3.0 (SD 2.5) for the control group, as well as number of participants who did not attend appointments as 22/279 for the intervention group and 44/403 for the control group. reported rate of health care use as a mean of 0.63 for the intervention group and 0.95 for the control group without providing variance values. mentioned it as an outcome, but did not report data. Change in disease activity This outcome was not reported in any of the studies. Change in quality of life This outcome was only reported in . The mean difference in the intervention group was −0.17 (SD 0.49) and in the control group 0.28 (SD 0.62) for the IBDQ and −0.05 (SD 0.28) and −0.01 (SD 0.25), respectively, for the QuICC. Medication adherence reported medication adherence as a mean of 7.01 (SD 1.40) for the intervention group and 6.77 (SD 1.61) for the control group. reported 66/126 and 64/122 as non‐adherent in the intervention and control groups, respectively. In , difference in average adherence rates pre‐ and post‐randomisation was +0.36 (SD 10.28) for the intervention group and −15.3 (SD 25.34) for the control group. reported 166 incidents of missed medications, with a mean of 2.31 incidents per participant, and calculated the mean number of missed medications during the study as 0.91 for the intervention group and 3.43 for the control group. did not provide the raw mean and variance scores per group at study end, instead the authors presented the results of their own analysis. Patient knowledge or skill at study end In , the mean difference from baseline (no variance provided) was +2.4 in the TELE‐IBD EOW intervention group +2.0 in the TELE‐IBD W intervention group and +1.8 in the control group. reported that mean rank scores (no variance provided; high score = better result) at end of study were: 5.8 for the intervention group and 4.0 for the control group for gastrointestinal anatomy; 5.6 and 4.3 for general IBD knowledge; 6.1 and 3.6 for medications; and 4.2 and 6.0 for nutrition. reported that post‐intervention the mean score on the assessment was 55.6% (range 35.0% to 95.6%), but did not report results per intervention group. reported CCKNOW scores of 19.52 (SD 2.55) for the intervention group and 13.84 (SD 4.86) for the control group, and KQ scores of 27.19 (SD 3.03) and 21.47 (SD 6.81) respectively, at study end. In , an improvement in patients' skills was defined by an increase of the ECIPE score of more than 20%, from baseline to six months. In the intervention group 61 patients achieved that and 31 in the control. Per protocol median ECIPE scores were reported as 26 (range 22‐30) in the intervention group (n = 105) and 20 (range 16‐25) (n = 117) in the control group. In each of the results in this paragraph, higher scores indicate improvement. Self‐reported medical knowledge was reported in three studies as 4.05 (SD 0.41) for the intervention group and 3.42 (SD 0.71) for the control and psychological knowledge as 3.65 (SD 0.67) and 2.98 (SD 0.74), respectively in . Knowledge of IBD was reported as 8.17 (SD 1.16) for the intervention group and 7.84 (SD 1.47) for the control group, and knowledge of medication as 7.75 (SD 1.58) and 7.58 (SD 1.51), respectively in . Self‐perceived knowledge was reported as 7.6 for the intervention group and 6.2 for the control group at study end in . Total adverse effects , , and reported zero total adverse effects in their studies. Withdrawals due to adverse events The only study that reported withdrawals due to adverse effects was , which reported that in the TELE‐IBD EOW intervention group one participant withdrew due to breast cancer and in the TELE‐IBD intervention group two participants withdrew because they needed surgery. No participants withdrew due to adverse effects from the control group. In , hospitalisations, surgery, emergency department and office visits, procedures, intravenous therapeutics, and telephone and electronic encounters were extracted from the electronic medical record (EMR) for one year before and after randomisation, and encounters were reported as rates, adjusted for 100 participants per year. The intervention group that received a telemedicine message every other week (IG1 (TELE‐IBD EOW)) had 2235 total encounters, the intervention group that received a telemedicine message every week (IG2 (TELE‐IBD W)) had 1935, and the control group had 2099 (the data on the specific types of encounters are presented in ). reported mean numbers of hospital admissions, which were 0.05 (SD 0.28) for the intervention group and 0.10 (SD 0.43) for the control group; and mean numbers of emergency visits, which were 0.07 (SD 0.35) for the intervention group and 0.10 (SD 0.54) for the control group. reported mean number of kept hospital appointments as 1.9 (SD 2.2) for the intervention group and 3.0 (SD 2.5) for the control group, as well as number of participants who did not attend appointments as 22/279 for the intervention group and 44/403 for the control group. reported rate of health care use as a mean of 0.63 for the intervention group and 0.95 for the control group without providing variance values. mentioned it as an outcome, but did not report data. This outcome was not reported in any of the studies. This outcome was only reported in . The mean difference in the intervention group was −0.17 (SD 0.49) and in the control group 0.28 (SD 0.62) for the IBDQ and −0.05 (SD 0.28) and −0.01 (SD 0.25), respectively, for the QuICC. reported medication adherence as a mean of 7.01 (SD 1.40) for the intervention group and 6.77 (SD 1.61) for the control group. reported 66/126 and 64/122 as non‐adherent in the intervention and control groups, respectively. In , difference in average adherence rates pre‐ and post‐randomisation was +0.36 (SD 10.28) for the intervention group and −15.3 (SD 25.34) for the control group. reported 166 incidents of missed medications, with a mean of 2.31 incidents per participant, and calculated the mean number of missed medications during the study as 0.91 for the intervention group and 3.43 for the control group. did not provide the raw mean and variance scores per group at study end, instead the authors presented the results of their own analysis. In , the mean difference from baseline (no variance provided) was +2.4 in the TELE‐IBD EOW intervention group +2.0 in the TELE‐IBD W intervention group and +1.8 in the control group. reported that mean rank scores (no variance provided; high score = better result) at end of study were: 5.8 for the intervention group and 4.0 for the control group for gastrointestinal anatomy; 5.6 and 4.3 for general IBD knowledge; 6.1 and 3.6 for medications; and 4.2 and 6.0 for nutrition. reported that post‐intervention the mean score on the assessment was 55.6% (range 35.0% to 95.6%), but did not report results per intervention group. reported CCKNOW scores of 19.52 (SD 2.55) for the intervention group and 13.84 (SD 4.86) for the control group, and KQ scores of 27.19 (SD 3.03) and 21.47 (SD 6.81) respectively, at study end. In , an improvement in patients' skills was defined by an increase of the ECIPE score of more than 20%, from baseline to six months. In the intervention group 61 patients achieved that and 31 in the control. Per protocol median ECIPE scores were reported as 26 (range 22‐30) in the intervention group (n = 105) and 20 (range 16‐25) (n = 117) in the control group. In each of the results in this paragraph, higher scores indicate improvement. Self‐reported medical knowledge was reported in three studies as 4.05 (SD 0.41) for the intervention group and 3.42 (SD 0.71) for the control and psychological knowledge as 3.65 (SD 0.67) and 2.98 (SD 0.74), respectively in . Knowledge of IBD was reported as 8.17 (SD 1.16) for the intervention group and 7.84 (SD 1.47) for the control group, and knowledge of medication as 7.75 (SD 1.58) and 7.58 (SD 1.51), respectively in . Self‐perceived knowledge was reported as 7.6 for the intervention group and 6.2 for the control group at study end in . , , and reported zero total adverse effects in their studies. The only study that reported withdrawals due to adverse effects was , which reported that in the TELE‐IBD EOW intervention group one participant withdrew due to breast cancer and in the TELE‐IBD intervention group two participants withdrew because they needed surgery. No participants withdrew due to adverse effects from the control group. Two studies compared delivery methods of patient education in the form of web‐based interventions against other delivery methods ( ; ). Primary outcomes Only reported any of our primary outcomes. Disease activity at study end reported numbers of UC and CD participants in remission, or with mild, severe, or very severe disease at study end. For UC participants, 8/16 in the web‐based group and 10/16 in the control education group were in remission, 6/16 and 4/16 had mild disease, 2/16 and 1/16 had severe disease, and 0/16 and 0/16 had very severe disease. For CD participants, 5/14 and 10/14 were in remission, 7/14 and 3/14 had mild disease, 2/14 and 1/14 had severe disease, and 0/14 and 0/14 had very severe disease. Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life score on the IBDQ for the web‐based group was 156.53 (SD 30.97) and 155.63 (SD 34.30) for the control group (high score = better result). Secondary outcomes No secondary outcomes were reported except for the limited knowledge score data in , which we reported above. Only reported any of our primary outcomes. Disease activity at study end reported numbers of UC and CD participants in remission, or with mild, severe, or very severe disease at study end. For UC participants, 8/16 in the web‐based group and 10/16 in the control education group were in remission, 6/16 and 4/16 had mild disease, 2/16 and 1/16 had severe disease, and 0/16 and 0/16 had very severe disease. For CD participants, 5/14 and 10/14 were in remission, 7/14 and 3/14 had mild disease, 2/14 and 1/14 had severe disease, and 0/14 and 0/14 had very severe disease. Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life score on the IBDQ for the web‐based group was 156.53 (SD 30.97) and 155.63 (SD 34.30) for the control group (high score = better result). reported numbers of UC and CD participants in remission, or with mild, severe, or very severe disease at study end. For UC participants, 8/16 in the web‐based group and 10/16 in the control education group were in remission, 6/16 and 4/16 had mild disease, 2/16 and 1/16 had severe disease, and 0/16 and 0/16 had very severe disease. For CD participants, 5/14 and 10/14 were in remission, 7/14 and 3/14 had mild disease, 2/14 and 1/14 had severe disease, and 0/14 and 0/14 had very severe disease. This outcome was not reported. Mean quality of life score on the IBDQ for the web‐based group was 156.53 (SD 30.97) and 155.63 (SD 34.30) for the control group (high score = better result). No secondary outcomes were reported except for the limited knowledge score data in , which we reported above. compared frequency of patient education in the form of weekly educational text messages versus once every other week educational text messages (in addition to comparing these interventions to standard care, the results of which we included in the patient education and standard care versus standard care comparison above). Primary outcomes Disease activity at study end Mean disease activity for the TELE‐EOW CD participants was 4.2 (SD 3.9) and for the TELE‐W CD participants 3.2 (SD 3.4). Mean disease activity for the TELE‐EOW UC participants was 1.7 (SD 1.9) and for the TELE‐W UC participants was 2.0 (SD 1.8). Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life scores for the TELE‐EOW participants was 181.5 (SD 28.2) and for the TELE‐W participants was 179.2 (SD 32.8) Secondary outcomes These have been reported in Comparison 1, patient education and standard care versus standard care. Disease activity at study end Mean disease activity for the TELE‐EOW CD participants was 4.2 (SD 3.9) and for the TELE‐W CD participants 3.2 (SD 3.4). Mean disease activity for the TELE‐EOW UC participants was 1.7 (SD 1.9) and for the TELE‐W UC participants was 2.0 (SD 1.8). Flare‐ups or relapse This outcome was not reported. Quality of life at study end Mean quality of life scores for the TELE‐EOW participants was 181.5 (SD 28.2) and for the TELE‐W participants was 179.2 (SD 32.8) Mean disease activity for the TELE‐EOW CD participants was 4.2 (SD 3.9) and for the TELE‐W CD participants 3.2 (SD 3.4). Mean disease activity for the TELE‐EOW UC participants was 1.7 (SD 1.9) and for the TELE‐W UC participants was 2.0 (SD 1.8). This outcome was not reported. Mean quality of life scores for the TELE‐EOW participants was 181.5 (SD 28.2) and for the TELE‐W participants was 179.2 (SD 32.8) These have been reported in Comparison 1, patient education and standard care versus standard care. Summary of main results Education is clearly of vital importance within any chronic disease and almost certainly offered to all people affected by the condition in some form. However, this review has investigated the use of education as a specific intervention to enhance outcomes for patients. Given the complexity of educational interventions, there are several ways in which this eclectic mix of packages could be categorised. There were synchronous learning sessions which offered live teaching through a number of methods ( ; ; ; ; ; ; ) versus those which offered asynchronous access to learning materials ( ; ; ; ; ; ; ). There were also materials in either digital forms ( ; ; ; ), or traditional printed educational materials ( ; ; ). Most studies compared one of these forms of education to normal care, but descriptions of normal care were limited to a few words and no study defined how much education, whether formally or informally, was offered in these standard care groups. Reporting of most outcomes in a homogeneous fashion was limited, with quality of life at study end reported most commonly in six of the 14 studies which allowed for meta‐analysis, with all other outcomes reported in a more heterogeneous manner that limited analysis. The analysis found that there was no difference in quality of life in the education group ( ). The poor reporting of other outcome measures severely limited the scope for meta‐analysis and also significantly impacted the certainty of evidence due to the imprecision in other results, and may have contributed to inconsistency. Whilst these judgements are objective and in line with guidance, it is possible that further studies could impact the results. Since no studies reported knowledge or skill assessments in a manner that allowed meta‐analysis, conclusions cannot be drawn about whether the body of evidence for education in inflammatory bowel disease (IBD) shows that such education can educate people in a measurable way. Similarly, medication adherence was discussed in just five studies and was not reported in a manner that allowed meta‐analysis in any of these studies. Safety was also not reported in most studies, but this may reflect the primary authors' inference that education is unlikely to lead to harm. However, in those that did mention this outcome, no adverse events were reported. Overall completeness and applicability of evidence Despite the issues with heterogeneity of reporting discussed above, efficacy outcomes demonstrate with moderate certainty that there is no benefit to quality of life or disease state from patient education interventions. In these areas, it is questionable whether further research would be beneficial. There are, however, a number of areas where the evidence remains incomplete. The reporting of the educational interventions themselves is a concern. As shown in there was capricious reporting of the details of the education. Only those that used standard educational resources, such as booklets or guidebooks) could be considered reproducible ( ; . For the other interventions it was unclear what content was delivered to achieve which learning outcomes, which pedagogical techniques were deployed in detail to support dissemination, and with what resources. No details of any underpinning theoretical or conceptual frameworks and not much detail of the resources used were reported. Unlike pharmacological intervention reviews, readers of this review will not just require information about whether something is effective or safe, but about which specific interventions are effective ( ) to offer utility in clinical practice ( ). This information is not available for most studies in this review. This is a recognised problem in non‐pharmacological trial reporting, even though there is published guidance for primary study authors to help rectify the issue ( ); this guidance clearly was not employed in the primary studies included in this review. In a recent study, 65% of authors within non‐pharmacological intervention trials forwarded the required information on request ( ). This was not the case in this review, with no authors returning further educational details on request, mirroring our previous experience in Cochrane reviewing ( ). Future studies must rectify this gap and provide details about interventions and utility, for a more complete evidence base. The choice of outcomes that were used by primary researchers was also a concern. The primary outcomes in many of these studies, which are mirrored in this review, focused on clinically common and important outcomes within IBD research. Disease activity, change of disease state and quality of life are all vital outcomes. As the evidence from this review suggests that for two of these outcomes there is probably no benefit to education, this clearly challenges the initial assumption that led to a focus on these outcomes. On the surface it appears an entirely appropriate hypothesis that these outcomes should be the focus for educational studies. However, on reflection, if education were to have such an impact, it would raise deep questions about the level of basic medical discussion, consent and information sharing of professionals in standard care. Rather, it is the secondary outcomes of this review that have not been fully addressed by the evidence, and it would appear that in many ways these are not only more likely to be impacted by such interventions, but they would seem to have more utility and relevance to the people and professionals investigating such education effects ( ). Medication adherence is a common issue and enhancing education to improve this by empowering patients to make their own choices proactively would seem a suitable outcome for such interventions, but these data were poorly reported in a heterogenous fashion that did not facilitate any meta‐analysis ( ). Whilst, in the long run, medication adherence may also impact the previously discussed primary outcomes, this in many ways is indirect and would probably require a far longer follow‐up than any of the included studies had. Attendance at, or need for interventions from primary or secondary care sources also seems a useful focus ( ). It may not be as simple as reducing these, but rather changing patterns of behaviour. As such, investigators may want to consider not just whether attendance changed, but in what way, and ‐ most importantly ‐ why. Empowering patients to seek support at the times that are most vital to enhance their care is as important as reducing attendance, and so simple quantitative comparisons may not be sufficient for such studies ( ). Similarly, quality of life measures overall may not be the best to consider for such studies. The Inflammatory Bowel Disease Questionnaire (IBDQ) was the most reported measure ( ), but most of the items included are clinically and symptom focused, with only two subsets that are potentially relevant (emotional and social activity sets). As data from these subsets were rarely reported, this once again represents a gap in the synthesised evidence, and future researchers may wish to consider separate subset reporting ( ). Standard care was commonly used as the comparison, and was poorly reported in all of these studies, with no study providing clear and concise descriptions of what specific education, in what forms, by which people and at what intervals were offered routinely within it. This information is vital, as it is possible that there are huge differences between this and the interventions. The reverse could also be true, with the same education being offered to both study groups, just in different forms. Without clarity about this issue, the completeness and utility of the evidence is limited. For our analyses we used study end outcome data and we recognise the variability in the timing of outcome assessment as a limitation. Follow‐ups in IBD interventional studies can vary widely, as this is a chronic remitting and relapsing non‐curable condition, which makes it different to other areas of health care. We identified six ongoing studies, which appear to have the potential to add to the evidence base. However, it is not clear if these studies will be presented in a way that will address the pervasive issues discussed above. Quality of the evidence There were significant issues related to risk of bias in the studies included in this review. Despite our requests emailed to authors of all included studies, we received little data to change our judgements in these key areas. Whilst most studies were not blinded for performance or detection bias, this can be seen as acceptable given the context of the review. However, there were issues in all other areas that cannot be similarly accepted. The reporting of the interventions themselves is a source of potential bias, as it is difficult for readers of the studies to understand what specific intervention was delivered, and this limits consideration in all other areas. As already discussed, this is recognised as a problem within health intervention reporting ( ), and within health education systematic review ( ), although it is not explicitly identified when applying GRADE to evidence ( ). This is the biggest issue with the evidence base, and it limits the utility of any outcomes, as these interventions cannot be replicated or disseminated. We downgraded certainty for the outcome of disease activity one level due to issues with risk of bias related to blinding, allocation concealment and randomisation in the two studies that provided data for this outcome. Flare‐ups as a continuous outcome had the same issues with risk of bias, for which we downgraded the certainty by one level. We downgraded flare‐ups as a dichotomous outcome by a total of three levels; two levels due to serious issues with risk of bias for the three studies that provided data related to blinding, allocation concealment, randomisation, selective reporting and other sources of bias, as well as one level for imprecision due to limited event numbers. We downgraded quality of life one level due to concerns with risk of bias related to blinding and allocation concealment. Potential biases in the review process Clinical heterogeneity is a key area of concern in this review. Most studies included patients with both Crohn's Disease (CD) and ulcerative colitis (UC) and at different disease states. It would not have been possible to exclude studies that did not differentiate between CD and UC, as this would have affected the vast majority of studies. Exclusion of these studies would exclude a key source of evidence in this area, but their inclusion clearly introduces a source of bias. We decided that in order for a study to be included in the review, the educational component had to be the primary focus of the study and not part of a larger package. Our decisions were clearly systematic, but it is possible that we missed relevant studies. It is also possible that education may have been part of a package, but again this was not included in the review. Missing data or unclear outcome data were ongoing issues we encountered for many studies, which represent ways in which the evidence base is lacking. To deal with this, we made a number of methodological choices which have in turn influenced the findings of the review. We contacted authors for missing data and we used the data for analysis, when provided to us. For analyses using dichotomous data, we used the numbers randomised as denominators. As numerators we used the numbers as reported by the authors for positive outcomes. For negative outcomes we used the plausible worst‐case scenario and added the numbers of dropouts to the numerator, as is normal practice for reviews for IBD, given the chronic nature of the condition and the high rates of adverse events and treatment failures across a patient's journey. For withdrawals due to adverse events specifically, we considered as adverse events all unspecified reasons and all reasons that did not automatically preclude the possibility of an adverse event. For analyses using continuous outcomes, we used the sample numbers as reported by the authors, for each particular continuous outcome. If the sample numbers were not reported, we estimated the sample number based on the attrition percentages reported. For cluster‐trial data we calculated effective sample sizes based on chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions ( ). Finally, there are 20 studies awaiting classification. These represent a mix of studies that are potential inclusions, but that have either not produced an output after trial registration, or published an abstract only that would not allow the study to be included. This large number of studies must be considered as another source of bias. Agreements and disagreements with other studies or reviews This is the first Cochrane Review on this topic, and as far as we can tell no other systematic reviews on the topic exist. None of the international guidelines for IBD mentions the evidence base in support of, or to propose, any specific educational interventions for people with IBD. Education is clearly of vital importance within any chronic disease and almost certainly offered to all people affected by the condition in some form. However, this review has investigated the use of education as a specific intervention to enhance outcomes for patients. Given the complexity of educational interventions, there are several ways in which this eclectic mix of packages could be categorised. There were synchronous learning sessions which offered live teaching through a number of methods ( ; ; ; ; ; ; ) versus those which offered asynchronous access to learning materials ( ; ; ; ; ; ; ). There were also materials in either digital forms ( ; ; ; ), or traditional printed educational materials ( ; ; ). Most studies compared one of these forms of education to normal care, but descriptions of normal care were limited to a few words and no study defined how much education, whether formally or informally, was offered in these standard care groups. Reporting of most outcomes in a homogeneous fashion was limited, with quality of life at study end reported most commonly in six of the 14 studies which allowed for meta‐analysis, with all other outcomes reported in a more heterogeneous manner that limited analysis. The analysis found that there was no difference in quality of life in the education group ( ). The poor reporting of other outcome measures severely limited the scope for meta‐analysis and also significantly impacted the certainty of evidence due to the imprecision in other results, and may have contributed to inconsistency. Whilst these judgements are objective and in line with guidance, it is possible that further studies could impact the results. Since no studies reported knowledge or skill assessments in a manner that allowed meta‐analysis, conclusions cannot be drawn about whether the body of evidence for education in inflammatory bowel disease (IBD) shows that such education can educate people in a measurable way. Similarly, medication adherence was discussed in just five studies and was not reported in a manner that allowed meta‐analysis in any of these studies. Safety was also not reported in most studies, but this may reflect the primary authors' inference that education is unlikely to lead to harm. However, in those that did mention this outcome, no adverse events were reported. Despite the issues with heterogeneity of reporting discussed above, efficacy outcomes demonstrate with moderate certainty that there is no benefit to quality of life or disease state from patient education interventions. In these areas, it is questionable whether further research would be beneficial. There are, however, a number of areas where the evidence remains incomplete. The reporting of the educational interventions themselves is a concern. As shown in there was capricious reporting of the details of the education. Only those that used standard educational resources, such as booklets or guidebooks) could be considered reproducible ( ; . For the other interventions it was unclear what content was delivered to achieve which learning outcomes, which pedagogical techniques were deployed in detail to support dissemination, and with what resources. No details of any underpinning theoretical or conceptual frameworks and not much detail of the resources used were reported. Unlike pharmacological intervention reviews, readers of this review will not just require information about whether something is effective or safe, but about which specific interventions are effective ( ) to offer utility in clinical practice ( ). This information is not available for most studies in this review. This is a recognised problem in non‐pharmacological trial reporting, even though there is published guidance for primary study authors to help rectify the issue ( ); this guidance clearly was not employed in the primary studies included in this review. In a recent study, 65% of authors within non‐pharmacological intervention trials forwarded the required information on request ( ). This was not the case in this review, with no authors returning further educational details on request, mirroring our previous experience in Cochrane reviewing ( ). Future studies must rectify this gap and provide details about interventions and utility, for a more complete evidence base. The choice of outcomes that were used by primary researchers was also a concern. The primary outcomes in many of these studies, which are mirrored in this review, focused on clinically common and important outcomes within IBD research. Disease activity, change of disease state and quality of life are all vital outcomes. As the evidence from this review suggests that for two of these outcomes there is probably no benefit to education, this clearly challenges the initial assumption that led to a focus on these outcomes. On the surface it appears an entirely appropriate hypothesis that these outcomes should be the focus for educational studies. However, on reflection, if education were to have such an impact, it would raise deep questions about the level of basic medical discussion, consent and information sharing of professionals in standard care. Rather, it is the secondary outcomes of this review that have not been fully addressed by the evidence, and it would appear that in many ways these are not only more likely to be impacted by such interventions, but they would seem to have more utility and relevance to the people and professionals investigating such education effects ( ). Medication adherence is a common issue and enhancing education to improve this by empowering patients to make their own choices proactively would seem a suitable outcome for such interventions, but these data were poorly reported in a heterogenous fashion that did not facilitate any meta‐analysis ( ). Whilst, in the long run, medication adherence may also impact the previously discussed primary outcomes, this in many ways is indirect and would probably require a far longer follow‐up than any of the included studies had. Attendance at, or need for interventions from primary or secondary care sources also seems a useful focus ( ). It may not be as simple as reducing these, but rather changing patterns of behaviour. As such, investigators may want to consider not just whether attendance changed, but in what way, and ‐ most importantly ‐ why. Empowering patients to seek support at the times that are most vital to enhance their care is as important as reducing attendance, and so simple quantitative comparisons may not be sufficient for such studies ( ). Similarly, quality of life measures overall may not be the best to consider for such studies. The Inflammatory Bowel Disease Questionnaire (IBDQ) was the most reported measure ( ), but most of the items included are clinically and symptom focused, with only two subsets that are potentially relevant (emotional and social activity sets). As data from these subsets were rarely reported, this once again represents a gap in the synthesised evidence, and future researchers may wish to consider separate subset reporting ( ). Standard care was commonly used as the comparison, and was poorly reported in all of these studies, with no study providing clear and concise descriptions of what specific education, in what forms, by which people and at what intervals were offered routinely within it. This information is vital, as it is possible that there are huge differences between this and the interventions. The reverse could also be true, with the same education being offered to both study groups, just in different forms. Without clarity about this issue, the completeness and utility of the evidence is limited. For our analyses we used study end outcome data and we recognise the variability in the timing of outcome assessment as a limitation. Follow‐ups in IBD interventional studies can vary widely, as this is a chronic remitting and relapsing non‐curable condition, which makes it different to other areas of health care. We identified six ongoing studies, which appear to have the potential to add to the evidence base. However, it is not clear if these studies will be presented in a way that will address the pervasive issues discussed above. There were significant issues related to risk of bias in the studies included in this review. Despite our requests emailed to authors of all included studies, we received little data to change our judgements in these key areas. Whilst most studies were not blinded for performance or detection bias, this can be seen as acceptable given the context of the review. However, there were issues in all other areas that cannot be similarly accepted. The reporting of the interventions themselves is a source of potential bias, as it is difficult for readers of the studies to understand what specific intervention was delivered, and this limits consideration in all other areas. As already discussed, this is recognised as a problem within health intervention reporting ( ), and within health education systematic review ( ), although it is not explicitly identified when applying GRADE to evidence ( ). This is the biggest issue with the evidence base, and it limits the utility of any outcomes, as these interventions cannot be replicated or disseminated. We downgraded certainty for the outcome of disease activity one level due to issues with risk of bias related to blinding, allocation concealment and randomisation in the two studies that provided data for this outcome. Flare‐ups as a continuous outcome had the same issues with risk of bias, for which we downgraded the certainty by one level. We downgraded flare‐ups as a dichotomous outcome by a total of three levels; two levels due to serious issues with risk of bias for the three studies that provided data related to blinding, allocation concealment, randomisation, selective reporting and other sources of bias, as well as one level for imprecision due to limited event numbers. We downgraded quality of life one level due to concerns with risk of bias related to blinding and allocation concealment. Clinical heterogeneity is a key area of concern in this review. Most studies included patients with both Crohn's Disease (CD) and ulcerative colitis (UC) and at different disease states. It would not have been possible to exclude studies that did not differentiate between CD and UC, as this would have affected the vast majority of studies. Exclusion of these studies would exclude a key source of evidence in this area, but their inclusion clearly introduces a source of bias. We decided that in order for a study to be included in the review, the educational component had to be the primary focus of the study and not part of a larger package. Our decisions were clearly systematic, but it is possible that we missed relevant studies. It is also possible that education may have been part of a package, but again this was not included in the review. Missing data or unclear outcome data were ongoing issues we encountered for many studies, which represent ways in which the evidence base is lacking. To deal with this, we made a number of methodological choices which have in turn influenced the findings of the review. We contacted authors for missing data and we used the data for analysis, when provided to us. For analyses using dichotomous data, we used the numbers randomised as denominators. As numerators we used the numbers as reported by the authors for positive outcomes. For negative outcomes we used the plausible worst‐case scenario and added the numbers of dropouts to the numerator, as is normal practice for reviews for IBD, given the chronic nature of the condition and the high rates of adverse events and treatment failures across a patient's journey. For withdrawals due to adverse events specifically, we considered as adverse events all unspecified reasons and all reasons that did not automatically preclude the possibility of an adverse event. For analyses using continuous outcomes, we used the sample numbers as reported by the authors, for each particular continuous outcome. If the sample numbers were not reported, we estimated the sample number based on the attrition percentages reported. For cluster‐trial data we calculated effective sample sizes based on chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions ( ). Finally, there are 20 studies awaiting classification. These represent a mix of studies that are potential inclusions, but that have either not produced an output after trial registration, or published an abstract only that would not allow the study to be included. This large number of studies must be considered as another source of bias. This is the first Cochrane Review on this topic, and as far as we can tell no other systematic reviews on the topic exist. None of the international guidelines for IBD mentions the evidence base in support of, or to propose, any specific educational interventions for people with IBD. Implications for practice There is evidence that education is probably of no benefit to disease activity or quality of life when compared with standard care, and may be of no benefit to occurrence of relapse when compared with standard care. However, as there was a paucity of specific information regarding the components included in either education or standard care, the utility of these findings is questionable. Implications for research Further research to investigate the impact of education on our primary outcomes of disease activity, disease state and quality of life is probably not indicated. This conclusion is not based on the outcomes of the analyses in this review alone, but on consideration of the likely mechanism of action of extra or bespoke inflammatory bowel disease (IBD) education, and indeed the goals of educational interventions for the stakeholders they are likely to impact. Further research should focus on two key areas. The first is to report details of the educational interventions in a manner that supports transparency, dissemination and replication using existing guidance. The second is to focus on outcomes that educational interventions can be directly targeted to address. These should be informed by direct engagement with stakeholders and people affected by Crohn's disease and ulcerative colitis. Medication adherence and quality of life subsets would be good targets for further work. Further research on subsets of patients ‐ such as the newly diagnosed, or socially and financially disadvantaged ‐ who may be in greater need of educational support, should also be encouraged. Within all such studies, reporting in a manner that is consistent with clarity for risk of bias judgements is vital. Protocol first published: Issue 1, 2021
Challenges in oncology career: are we closing the gender gap? Results of the new ESMO Women for Oncology Committee survey	6d23c3ef-6d3a-4a77-bc87-795617006661	10163010	Internal Medicine[mh]	A lack of gender balance in the workplace has far-reaching adverse educational, health, economic and societal consequences, while gender equity and diversity bring opportunities for greater innovation, increased productivity and better decision making. In medicine, there is evidence that a gender diverse workforce can result in improved outcomes for patients, , , and may foster more relevant research applicable to a broader population. , Between 2000 and 2012, there was a greater increase in female than in male oncologists, and female membership of European Society for Medical Oncology (ESMO) rose from 24.9% in 2004 to 35.2% in 2012, 40.5% in 2016 and 49% today. Women for Oncology (W4O) was established in 2013 to explore the challenges facing female oncologists and promote equal access to career development opportunities and access to leadership roles, addressing the needs of the rapidly rising female membership of ESMO. In 2016, a W4O survey of female and male oncologists established that women were under-represented in managerial and leadership roles and identified work and family balance as the most important challenge to career progression. Regular monitoring studies have been carried out to provide further objective information on gender equity in the oncology workplace. , However, research in oncology has consistently shown similar findings, , , , and progress towards gender equity in career development in oncology is slow. Research findings have created the basis for informing the decisions of the ESMO W4O Committee in setting priorities and developing projects to address obstacles to gender equity, and rebalancing gender representation in leadership positions in oncology. A broad range of gender-focused career development initiatives has been launched and W4O has become a hub for facilitating regional and local activities and tackling gender equity issues at grassroot level including raising awareness, leadership and mentorship programmes as well as roundtable social media discussions to shed light on challenges that oncologists face at all career levels. In the framework of those initiatives, the ESMO W4O Committee continuously monitors the evolution of oncology professionals throughout their career, aiming to identify diverse or changing needs and challenging areas, such as equal access to leadership positions, competing demands of time for clinical, academic and research work, salary reductions and pay gaps. Not only do such efforts raise awareness but also provide evidence for interventions. In such context, in 2021 the ESMO W4O Committee decided to carry out a new survey to gather comparative and new data and explore the possible impact of W4O and other interventions in bridging the gender gap identified in 2016. In recent years, ESMO has expanded its membership beyond Europe and now includes a substantial number of oncologists across Asia, South America and Africa and so the new survey, carried out in October 2021, was designed to capture the full impact of diversity on career development in today’s multinational oncology workforce. The W4O 2021 survey questionnaire was based on the one previously used for the survey carried out in 2016, with additional questions related to the impact of ethnicity, sexual orientation and religion on career development, as well as gender. Results were analysed according to the gender and age (≤40 versus >40 years) of respondents. This age comparison was chosen to correspond with ESMO membership categories, one of which is specifically for oncologists ≤40 years of age. The questionnaire consisted of seven sections: (i) demographics; (ii) household duties; (iii) place of work; (iv) challenges for career progression; (v) diversity’s impact on career development and barriers for equality; (vi) inappropriate behaviour experienced in the professional career; (vii) closing the gender gap . In October 2021, the survey was sent to ESMO members and other oncology organisations, made available online and disseminated through national oncology societies as well as representatives of national initiatives of women in oncology. It was accessible to female and male oncology professionals of all ages, working in a range of clinical and academic environments internationally. It was promoted on the ESMO website, ESMO W4O and ESMO Facebook pages and through ESMO’s digital newsletters. The responses were anonymous. Results are presented overall and by respondent’s gender and age (≤40 versus >40 years). The differences between women and men or young and older respondents are tested using chi-square tests (for categorical variables) or Mann–Whitney U test (for continuous variables). The Cochran–Mantel–Haenszel statistic was also used to address the differences between women and men after controlling for age. All tests were considered significant if P value was <0.05. All statistical analysis was carried out with the software in SAS Version 9.4 (SAS Institute Inc., Cary, NC). Demographics and professional environment of respondents A total of 1473 responses were received (69.4% women, 30.3% men). Of respondents, 47.3% were 40 years and under. The majority were based in Europe (64.4%) though there was also a significant number of responses from Asia (16.4%). Most responses were from ESMO members (74.9%). Nearly three-quarters (73.9%) were medical oncologists, with 13.6% clinical/radiation oncologists, and a significant proportion were at early stages of their career (17.8% trainees, 39.9% practising oncology <10 years). Of total respondents, 16.5% lived alone, 34.1% had pre-school children, 27.9% had children at primary school and 27.6% had children at secondary school. Respondents were almost equally divided between working in a cancer centre (43.6%) and a general hospital (44.1%). Sixty percent of working time was dedicated to clinical care, 10% to research, 5% to teaching, 2% to management and 5% to administration. In only 39.1% of cases, respondents worked in a team led by a woman. There was a similar gender breakdown for heads of department (32.9% female, 67.1% male). Challenges to career progression Nearly 90% of all 1473 respondents to the survey said that career progression was important, but 27.8% ( n = 409) were slightly or not at all satisfied with how their career was progressing; this was more frequent among female than male respondents (30.6% n = 313 versus 21.1% n = 94, respectively) and those aged ≤40 years compared to those >40 years (33.1% n = 231 versus 22.9% n = 178, respectively) . The main obstacles were finding a balance between work and family life (59.5%), managing and organising family commitments (23.9%) and lack of mentors/role models (38.9%). The impact of ethnicity, sexual orientation, religion and gender on professional career is summarised by gender and age in . Ethnicity had little or no impact on professional career, career opportunities, salary setting or pay gap. Of respondents, 11% said that ethnicity had a significant or major impact on their professional career, 15.4% said it resulted in fewer career opportunities and 8.8% said that it affected salary setting. Few respondents said that ethnicity contributed to the pay gap in the workplace (7.5%), the pay gap in oncology in their country (12.4%) or the pay gap in general in oncology (18.2%). A similar lack of effect was reported for all these parameters for sexual orientation (4.8%, 6.3%, 2.7%, 3.8%, 6.3% and 7.8%, respectively) and religion (2.6%, 6.6%, 2.3%, 3.0%, 3.9% and 4.8%, respectively). In contrast, gender did have a significant or major impact on career development (25.5%), resulted in fewer career opportunities (45.8%) and affected salary setting (27.6%). Gender contributed to the pay gap in the workplace (29.6%), the pay gap in oncology in the respondent’s country (36.5%) and the pay gap in general in oncology (36.4%). Significantly more respondents ≤40 years of age said that gender had a major or significant impact on their professional career than those >40 years of age (51.9% versus 44.1%, P = 0.0002); similarly, the proportion was higher in female respondents versus male respondents (60.4% versus 19.1%, P < 0.0001) . Little or no progress had been made in closing the gender gap since they started working (48.8% versus 32.6%) . The differences were strongest in female respondents in these age groups. Among those <40 years, 51.9% of women and 39.1% of men ( P = 0.0004) felt that no progress had been made. In those >40 years, the gender difference was particularly marked, with 42.5% of women reporting no progress compared to only 14.4% of men. Gender and discrimination Women were more likely than men to report discrimination from a senior colleague (56.9% women versus 8.3% men) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ) or discrimination from another colleague in general due to gender (31.8% women versus 29.7% men) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ). This was independent of age. In addition, gender was also a discriminatory factor in interactions with patients (54.2% women versus 8.6% men) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ), independently of age. Over a third of respondents (38.4%) had at any time experienced harassment in their workplace and 43.2% had witnessed harassment. Women were more likely to have experienced or witnessed harassment than men (49.0% versus 13.9% and 48.7% versus 30.3%, respectively) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ). This was independent of age. Experiences or witnesses of inappropriate behaviours were mainly inappropriate sexual advances (67.8%) and generalised sexist remarks (21.9%). Over three-quarters of respondents (77.9%) had not reported the harassment they experienced or witnessed, mainly because they did not think anything would be done (36%), they did not think it was important enough (27.6%) or they feared reprisal (20.2%). Gender and family life Among the respondents who answered that they have children, females were more likely than males to be the primary child caretaker in both those aged ≤40 and >40 years (25.1% versus 7.2% and 31.0% versus 6.6%, respectively). In both age groups, women were also more likely than men to do housekeeping (41.3% versus 24.3% and 36.6% versus 10.1%, respectively) and prepare meals (48.1% versus 33.7% and 51.4% versus 14.1%, respectively). There were non-significant differences between female and male respondents aged ≤40 years for doing household administration (47.2% versus 49.7%) but, in those >40 years, women were significantly more likely than men to do household administration (47.6% versus 36.5%). Significantly more female than male respondents reported that parental leave and difficulties in coming back to work were obstacles to career progression (5.2% versus 1.6%), especially in the age group >40 years. Social pressure related to cultural gender prejudice about family and domestic responsibilities was also significantly more of an obstacle for women than men (12.6% versus 2.0%), as were lack of support from family (4.6% versus 2.0%) and not being perceived adequate to cover a leadership position (18.7% versus 13.1%). In contrast, significantly more male than female respondents said that financial constraints were an obstacle to career progression (20.7% versus 14.2%). Significant differences in the impact of career progression on family life for women and men were also reported. Significantly more women than men said that career progression very much or extremely impacted their parental leave (27.4% versus 11.3%, P < 0.0001), time dedicated to childcare (48.3% versus 21.3%, P < 0.0001) and leisure activities (42.4% versus 31.4%, P < 0.0005). Significantly more women than men said that having children very much or extremely impacted career progression (33.6% versus 13.8%, P < 0.0001). Gender and pay gap Significantly more female respondents reported a gender pay gap than men, especially in those >40 years: (≤40 years—35.8% versus 13.0%; >40 years—40.5% versus 8.7%). Significantly more women than men thought that gender had a major or significant impact on their professional career (33.7% versus 6.5%, P < 0.0001), and significantly more women than men said that gender, ethnicity and sexual orientation had an impact on pay gap in oncology in their country (45.5% versus 15.9%, 14.1% versus 8.3% and 7.0% versus 4.7%, respectively). Barriers to gender parity The main barriers for gender parity were lack of work–life balance (54.3%), lack of leadership development for women (32.7%), unconscious bias (34.5%), societal pressures (31.9%) and lack of role models (20.6%). Compared with male respondents and irrespective of age (≤40 versus >40 years), female respondents were significantly more likely to report lack of female professionals’ self-confidence and lack of leadership development for women as barriers for gender parity . Fostering gender equity Approaches for fostering gender equity in the workplace supported by respondents included promoting work–life balance (56.8%), development and leadership training (38.6%), offering and supporting flexible working (36.3%) and transparent career paths and salary structure (30.1%). Female respondents were significantly more likely than male respondents to support development and leadership training (42.2% versus 30.5%) and the promotion of education on culture and gender equality at work for all workers (22.5% versus 15.5%) . In terms of progress in closing the gender gap in oncology, 40.3% (47.4% women, 23.7% men) said that minor or no progress had been made since they started working . Among the proposed ESMO interventions to foster gender equity in oncology, respondents preferred the implementation of programmes including mentorship for female oncologists (47.5% overall, 53.1% women, 35.0% men, P < 0.0001), flexible education/fellowship programmes (41.0% overall, 40.6% women, 41.7% men), family-friendly facilities at oncology events (28.7% overall, 26.3% women, 34.3% men), online professional career development tools (25.4% overall, 27.1% women, 21.1% men) and scholarships to learn from leaders in the field (23.5% overall, 24.6% women, 21.1% men) . Overall, 19.4% of respondents favoured quotas for women in ESMO committees, faculties and events, significantly more frequently in female than male respondents (23.1% versus 11.0%). The difference between women and men for the need for development and leadership training was irrespective of age. In respondents ≤40 years, the need for promotion of education on culture and gender equity at work for all workers was recognised equally by female and male respondents while, in those >40 years, significantly more women than men reported this need (23.7% versus 13.7%). Comparison between 2021 and 2016 In the 2021 survey, there was a larger overall population of respondents ( n = 1473 versus n = 482) and a significantly higher proportion were male ( n = 446/1473; 30.4% versus n = 103/482; 23.0%) or worked in non-European countries (33.6% versus 28.4%) . Significantly fewer respondents worked in an academic/research field (2.1% versus 49.7%, P < 0.0001), and significantly more in a cancer centre (42.7% versus 17.0%) or in a general hospital (43.2% versus 21.6%). Significantly fewer respondents in 2021 had managerial or leadership roles than in 2016 ( n = 258/147; 31.8% versus 249/482; 51.7%) . Some of the challenges to career progression in 2021 were less of a barrier than in 2016 , including travel to attend meetings (10.6% versus 18.8%, P = 0.0001), difficulty spending time abroad/another institute for research fellowships (24.2% versus 31.0%, P = 0.0177), maternity leave and difficulties coming back to work (5.2% versus 13.7%, P < 0.0001), financial constraints (14.2% versus 26.1%, P < 0.0001), social pressure related to cultural gender prejudice about family and domestic responsibilities (12.6% versus 22.5%, P < 0.0001) and not perceived adequate to cover a leadership position (18.7% versus 39.2%, P < 0.0001). Among the main barriers for gender parity, only lack of leadership development for women (32.7% versus 16.9%, P < 0.0001) and unconscious bias (34.5% versus 27.5%, P = 0.0068) were perceived as more important in 2021 than in 2016. Though not statistically significantly different, there appears to have been less impact of gender on professional career in 2021 than in 2016 (66.1% versus 82.1%). There has been a significant increase in people reporting harassment in the workplace since 2016 (50.3% versus 41.0%) . In terms of approaches needed in oncology to foster gender equity in the workplace, the highest ranked steps remained promotion of work–life balance, development and leadership training and offering and supporting flexible working, though there was a statistically significant difference in responses for 2021 compared to 2016 only for promotion of work–life balance ( P = 0.012) . Significantly more respondents in 2021 than in 2016 supported promotion of education on culture and gender equity at work for all workers (20.5% versus 10.0%), and this increase was seen in both female (22.7% versus 9.8%) and male (15.5% versus 10.6%) respondents, though it was not significant in the latter. Seeking ways to remove unconscious bias in decision making also gained support in 2021 (28.7% versus 23.1%), though this was mainly due to a significant increase among female respondents (29.5% versus 22.7%). The increase in male respondents was not significant (26.9% versus 24.5%). There was a significant reduction in support for visible leadership commitment towards diversity (e.g. symbolic actions by top management) in 2021 (15.1% versus 24.3%), and this was reflected across responses from both female and male respondents. A total of 1473 responses were received (69.4% women, 30.3% men). Of respondents, 47.3% were 40 years and under. The majority were based in Europe (64.4%) though there was also a significant number of responses from Asia (16.4%). Most responses were from ESMO members (74.9%). Nearly three-quarters (73.9%) were medical oncologists, with 13.6% clinical/radiation oncologists, and a significant proportion were at early stages of their career (17.8% trainees, 39.9% practising oncology <10 years). Of total respondents, 16.5% lived alone, 34.1% had pre-school children, 27.9% had children at primary school and 27.6% had children at secondary school. Respondents were almost equally divided between working in a cancer centre (43.6%) and a general hospital (44.1%). Sixty percent of working time was dedicated to clinical care, 10% to research, 5% to teaching, 2% to management and 5% to administration. In only 39.1% of cases, respondents worked in a team led by a woman. There was a similar gender breakdown for heads of department (32.9% female, 67.1% male). Nearly 90% of all 1473 respondents to the survey said that career progression was important, but 27.8% ( n = 409) were slightly or not at all satisfied with how their career was progressing; this was more frequent among female than male respondents (30.6% n = 313 versus 21.1% n = 94, respectively) and those aged ≤40 years compared to those >40 years (33.1% n = 231 versus 22.9% n = 178, respectively) . The main obstacles were finding a balance between work and family life (59.5%), managing and organising family commitments (23.9%) and lack of mentors/role models (38.9%). The impact of ethnicity, sexual orientation, religion and gender on professional career is summarised by gender and age in . Ethnicity had little or no impact on professional career, career opportunities, salary setting or pay gap. Of respondents, 11% said that ethnicity had a significant or major impact on their professional career, 15.4% said it resulted in fewer career opportunities and 8.8% said that it affected salary setting. Few respondents said that ethnicity contributed to the pay gap in the workplace (7.5%), the pay gap in oncology in their country (12.4%) or the pay gap in general in oncology (18.2%). A similar lack of effect was reported for all these parameters for sexual orientation (4.8%, 6.3%, 2.7%, 3.8%, 6.3% and 7.8%, respectively) and religion (2.6%, 6.6%, 2.3%, 3.0%, 3.9% and 4.8%, respectively). In contrast, gender did have a significant or major impact on career development (25.5%), resulted in fewer career opportunities (45.8%) and affected salary setting (27.6%). Gender contributed to the pay gap in the workplace (29.6%), the pay gap in oncology in the respondent’s country (36.5%) and the pay gap in general in oncology (36.4%). Significantly more respondents ≤40 years of age said that gender had a major or significant impact on their professional career than those >40 years of age (51.9% versus 44.1%, P = 0.0002); similarly, the proportion was higher in female respondents versus male respondents (60.4% versus 19.1%, P < 0.0001) . Little or no progress had been made in closing the gender gap since they started working (48.8% versus 32.6%) . The differences were strongest in female respondents in these age groups. Among those <40 years, 51.9% of women and 39.1% of men ( P = 0.0004) felt that no progress had been made. In those >40 years, the gender difference was particularly marked, with 42.5% of women reporting no progress compared to only 14.4% of men. Women were more likely than men to report discrimination from a senior colleague (56.9% women versus 8.3% men) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ) or discrimination from another colleague in general due to gender (31.8% women versus 29.7% men) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ). This was independent of age. In addition, gender was also a discriminatory factor in interactions with patients (54.2% women versus 8.6% men) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ), independently of age. Over a third of respondents (38.4%) had at any time experienced harassment in their workplace and 43.2% had witnessed harassment. Women were more likely to have experienced or witnessed harassment than men (49.0% versus 13.9% and 48.7% versus 30.3%, respectively) ( , available at https://doi.org/10.1016/j.esmoop.2023.100781 ). This was independent of age. Experiences or witnesses of inappropriate behaviours were mainly inappropriate sexual advances (67.8%) and generalised sexist remarks (21.9%). Over three-quarters of respondents (77.9%) had not reported the harassment they experienced or witnessed, mainly because they did not think anything would be done (36%), they did not think it was important enough (27.6%) or they feared reprisal (20.2%). Among the respondents who answered that they have children, females were more likely than males to be the primary child caretaker in both those aged ≤40 and >40 years (25.1% versus 7.2% and 31.0% versus 6.6%, respectively). In both age groups, women were also more likely than men to do housekeeping (41.3% versus 24.3% and 36.6% versus 10.1%, respectively) and prepare meals (48.1% versus 33.7% and 51.4% versus 14.1%, respectively). There were non-significant differences between female and male respondents aged ≤40 years for doing household administration (47.2% versus 49.7%) but, in those >40 years, women were significantly more likely than men to do household administration (47.6% versus 36.5%). Significantly more female than male respondents reported that parental leave and difficulties in coming back to work were obstacles to career progression (5.2% versus 1.6%), especially in the age group >40 years. Social pressure related to cultural gender prejudice about family and domestic responsibilities was also significantly more of an obstacle for women than men (12.6% versus 2.0%), as were lack of support from family (4.6% versus 2.0%) and not being perceived adequate to cover a leadership position (18.7% versus 13.1%). In contrast, significantly more male than female respondents said that financial constraints were an obstacle to career progression (20.7% versus 14.2%). Significant differences in the impact of career progression on family life for women and men were also reported. Significantly more women than men said that career progression very much or extremely impacted their parental leave (27.4% versus 11.3%, P < 0.0001), time dedicated to childcare (48.3% versus 21.3%, P < 0.0001) and leisure activities (42.4% versus 31.4%, P < 0.0005). Significantly more women than men said that having children very much or extremely impacted career progression (33.6% versus 13.8%, P < 0.0001). Significantly more female respondents reported a gender pay gap than men, especially in those >40 years: (≤40 years—35.8% versus 13.0%; >40 years—40.5% versus 8.7%). Significantly more women than men thought that gender had a major or significant impact on their professional career (33.7% versus 6.5%, P < 0.0001), and significantly more women than men said that gender, ethnicity and sexual orientation had an impact on pay gap in oncology in their country (45.5% versus 15.9%, 14.1% versus 8.3% and 7.0% versus 4.7%, respectively). The main barriers for gender parity were lack of work–life balance (54.3%), lack of leadership development for women (32.7%), unconscious bias (34.5%), societal pressures (31.9%) and lack of role models (20.6%). Compared with male respondents and irrespective of age (≤40 versus >40 years), female respondents were significantly more likely to report lack of female professionals’ self-confidence and lack of leadership development for women as barriers for gender parity . Approaches for fostering gender equity in the workplace supported by respondents included promoting work–life balance (56.8%), development and leadership training (38.6%), offering and supporting flexible working (36.3%) and transparent career paths and salary structure (30.1%). Female respondents were significantly more likely than male respondents to support development and leadership training (42.2% versus 30.5%) and the promotion of education on culture and gender equality at work for all workers (22.5% versus 15.5%) . In terms of progress in closing the gender gap in oncology, 40.3% (47.4% women, 23.7% men) said that minor or no progress had been made since they started working . Among the proposed ESMO interventions to foster gender equity in oncology, respondents preferred the implementation of programmes including mentorship for female oncologists (47.5% overall, 53.1% women, 35.0% men, P < 0.0001), flexible education/fellowship programmes (41.0% overall, 40.6% women, 41.7% men), family-friendly facilities at oncology events (28.7% overall, 26.3% women, 34.3% men), online professional career development tools (25.4% overall, 27.1% women, 21.1% men) and scholarships to learn from leaders in the field (23.5% overall, 24.6% women, 21.1% men) . Overall, 19.4% of respondents favoured quotas for women in ESMO committees, faculties and events, significantly more frequently in female than male respondents (23.1% versus 11.0%). The difference between women and men for the need for development and leadership training was irrespective of age. In respondents ≤40 years, the need for promotion of education on culture and gender equity at work for all workers was recognised equally by female and male respondents while, in those >40 years, significantly more women than men reported this need (23.7% versus 13.7%). In the 2021 survey, there was a larger overall population of respondents ( n = 1473 versus n = 482) and a significantly higher proportion were male ( n = 446/1473; 30.4% versus n = 103/482; 23.0%) or worked in non-European countries (33.6% versus 28.4%) . Significantly fewer respondents worked in an academic/research field (2.1% versus 49.7%, P < 0.0001), and significantly more in a cancer centre (42.7% versus 17.0%) or in a general hospital (43.2% versus 21.6%). Significantly fewer respondents in 2021 had managerial or leadership roles than in 2016 ( n = 258/147; 31.8% versus 249/482; 51.7%) . Some of the challenges to career progression in 2021 were less of a barrier than in 2016 , including travel to attend meetings (10.6% versus 18.8%, P = 0.0001), difficulty spending time abroad/another institute for research fellowships (24.2% versus 31.0%, P = 0.0177), maternity leave and difficulties coming back to work (5.2% versus 13.7%, P < 0.0001), financial constraints (14.2% versus 26.1%, P < 0.0001), social pressure related to cultural gender prejudice about family and domestic responsibilities (12.6% versus 22.5%, P < 0.0001) and not perceived adequate to cover a leadership position (18.7% versus 39.2%, P < 0.0001). Among the main barriers for gender parity, only lack of leadership development for women (32.7% versus 16.9%, P < 0.0001) and unconscious bias (34.5% versus 27.5%, P = 0.0068) were perceived as more important in 2021 than in 2016. Though not statistically significantly different, there appears to have been less impact of gender on professional career in 2021 than in 2016 (66.1% versus 82.1%). There has been a significant increase in people reporting harassment in the workplace since 2016 (50.3% versus 41.0%) . In terms of approaches needed in oncology to foster gender equity in the workplace, the highest ranked steps remained promotion of work–life balance, development and leadership training and offering and supporting flexible working, though there was a statistically significant difference in responses for 2021 compared to 2016 only for promotion of work–life balance ( P = 0.012) . Significantly more respondents in 2021 than in 2016 supported promotion of education on culture and gender equity at work for all workers (20.5% versus 10.0%), and this increase was seen in both female (22.7% versus 9.8%) and male (15.5% versus 10.6%) respondents, though it was not significant in the latter. Seeking ways to remove unconscious bias in decision making also gained support in 2021 (28.7% versus 23.1%), though this was mainly due to a significant increase among female respondents (29.5% versus 22.7%). The increase in male respondents was not significant (26.9% versus 24.5%). There was a significant reduction in support for visible leadership commitment towards diversity (e.g. symbolic actions by top management) in 2021 (15.1% versus 24.3%), and this was reflected across responses from both female and male respondents. The 2021 survey on the challenges facing oncology professionals in their career development prompted a much larger response, especially among male oncologists, than in 2016. We believe this reflects ESMO’s extensive activities in recent years to raise awareness of the gender gap in oncology and its strategies for change. However, the results of the survey show that gender remains the main obstacle to career development, as highlighted by responses from both female and male oncology professionals, and we are still a long way from closing the gender gap. The significant reduction in the proportion of respondents in managerial or leadership roles between 2016 and 2021 is surprising but may be related to the self-selection of those who chose to take the survey and may not be representative of the oncology workforce in general. Among other important findings of the 2021 survey is the fact that over a quarter of respondents were dissatisfied with the way in which their career was progressing, especially women aged 40 years or under. As also reported in 2016, the main challenge for career progression was the difficulty in balancing work and family life. However, among female respondents, nearly all obstacles reported in 2016 were significantly less relevant in 2021, including managing family commitments, travel to attend international meetings, difficulty spending time abroad/another research institute for a research fellowship, maternity leave and difficulties coming back to work and financial constraints. These are gratifying improvements in the light of ESMO initiatives to address these obstacles, not just for female members but in response to all member needs in an inclusive way which values diversity. Such initiatives include: (i) W4O activities to increase awareness of gender-related issues through publications and dissemination of objective information at the annual W4O Forum, the W4O web page, social media, virtual networking opportunities; (ii) W4O acting as a hub for local initiatives ( https://www.esmo.org/career-development/women-for-oncology/w4o-hub )—providing advice for setting them up and facilitating collaboration among them (in 2021, W4O held the first meeting of representatives of national/regional initiatives for women working in oncology); (iii) expansion of the ESMO Oncology fellowship offer ( https://www.esmo.org/career-development/oncology-fellowships ), with an increase in number and variety of opportunities to better reflect the evolution of the oncology field (e.g. introduction of a clinician-scientist fellowship); (iv) establishment of the ESMO Resilience Task Force ( https://www.esmo.org/career-development/resilience-task-force ); (v) ESMO educational events both in person and virtually; (vi) travel grants to make access to meetings easier for many oncologists, together with free childcare services at the ESMO Congress 2019. In due consideration of the variety of backgrounds and needs of the professionals who make up the ESMO membership base (from 160 countries), ESMO felt it was important to expand the range of potential diversity issues addressed in the 2021 survey, including additional aspects, such as ethnicity and others, which could have an impact on the professional career of oncologists. It is interesting to see that ethnicity, sexual orientation and religion were not considered to have an impact on professional career or salary. The finding that gender remains the most important factor, especially among female respondents, with little change in perceptions of its significance between 2016 and 2021, underlines the importance of the W4O Committee and supports the necessity for ESMO to continue focusing on gender in devising career development strategies. ESMO, together with many other professional societies in medicine, is striving for gender equity, and our results together with those of other medical specialties/societies could form the basis for interventions. , , , , , It is disappointing to see that almost twice as many respondents in 2021 reported that a lack of leadership development for female oncologists was a barrier to equity, compared to 2016. This is reinforced in the 2021 survey by the fact that significantly more women than men report the need for development and leadership training, whatever their age. Possible changes in the proportion of respondents from certain regions in 2021 compared to 2016 may have contributed to these findings. ESMO recognises the importance of role models and leadership development programmes to address this issue, and has introduced (i) the ESMO Leaders Generation Programme ( https://www.esmo.org/career-development/leaders-generation-programme ); (ii) the ESMO Virtual Mentorship Programme whose format makes it easier for participants to balance work and family life; (iii) ESMO Young Oncologists Committee (YOC) and the W4O mentorships on specific topics such as work–life balance and leadership skills, and more generally about career opportunities; (iv) mentorship sessions at ESMO congresses. In addition to reinforcing the existing programmes, ESMO intends to foster initiatives and training at local and regional level as a way to facilitate leadership development. We noted that among respondents in 2021 very limited time seemed to be spent in research and teaching, 10% and 5%, respectively, possibly related to lack of advancement in academic career or even lack of opportunities. However, significantly fewer respondents in the current survey worked in an academic/research field compared to 2016 (2.1% versus 49.7%, P < 0.0001), and this is most likely reflected in the responses regarding how working time was spent. ESMO will also continue to reinforce initiatives to promote work–life balance and education on culture and gender equity—two approaches for which there was increased support in the 2021 survey, compared to 2016. The issue of discrimination and inappropriate behaviour is of grave concern, particularly in view of the significant increase in the level of harassment reported in 2021 compared to 2016, with a much larger survey sample of oncologists in 2021 than in the previous survey. As in 2016, respondents had encountered unwanted sexual comments and behaviour in the workplace. They had also experienced gender-related discrimination from patients, senior colleagues and other colleagues. These findings reflect a concerning reality of today’s oncology workplace. The fact that less than one in five people who experienced or witnessed harassment reported it, mainly because they did not think anything would be done or because of fear of reprisal, is a serious indictment of our systems of governance which needs to be urgently addressed. The W4O Committee aims to monitor all aspects of gender equity and diversity in oncology and the survey results shape our activities, our plans for action and our proposals for relevant changes within the society. The finding about the increase in the number of reports of harassment will be further investigated in a dedicated study going forward. Limitations of the survey include the fact that the sampling method was not systematic which could result in sampling bias, decreasing the generalizability of results. This was addressed by the large sample size involving almost three times more respondents than in 2016. ESMO is not alone in recognising and endeavouring to address lack of gender equity in medicine. In its 2021 policy action paper on closing the leadership gap in health care, the World Health Organization (WHO) outlined four action areas in its framework to support female leadership: build the foundation for equality, address social norms and stereotypes, address workplace systems and cultures and enable women to achieve. In March 2022, ASCO and City Cancer Challenge Foundation (C/Can) announced their Leadership Program for Women in Oncology. The programme seeks to address the specific challenges and barriers faced by women leaders in oncology and aims to strengthen leadership mindsets and skills of women. In the UK, the Athena Scientific Women’s Academic Network (SWAN) charter is widely used to advance the careers of women in science, technology, engineering, mathematics and medicine and address gender challenges in higher education. Similar programmes are in use in other countries, including Science in Australia Gender Equity (SAGE) in Australia and the Dimensions Charter in Canada. The importance of equity and diversity in the global health workplace is not in doubt, and nowhere more so than in oncology. Gender equity at all levels of the oncology career pathway, and in medicine overall, will bring better patient care and outcomes, greater productivity, less risk of discrimination and harassment and a more satisfied and sustainable workforce. , , Together with the other W4O research on the representation of women in leadership roles, the findings of the W4O 2021 survey provide new evidence that could serve as a basis for ESMO strategies and appropriate interventions to support career development for all oncologists, whatever their gender, and ensure equal access to leadership roles, thus shaping the future direction of the profession and optimisation of patient care. This work was supported by the 10.13039/501100007075 European Society for Medical Oncology (no grant number). AAA reports personal and/or institutional financial interests in the form of local or coordinating PI for: iMAB, MERCK AG, MSD, Regeneron, Tolero, Zai Labs; reports non-financial interests in the form of advisor for: Cagent Pharmaceuticals, Johnson and Johnson, Merck AG, Swiss Rockets, Zai Labs; reports other roles in the International Association for the Study of Lung Cancer, Editor-in-Chief of Journal of Thoracic Oncology and JTO CRR. JB reports personal and/or institutional financial interests in the form of local PI for: Eli Lilly, Novartis, Paxman Coolers Ltd, Roche, Samsung Bioepis co. Ltd, Sun Pharma; reports non-financial interests in the form of advisor for Novartis and in the form of ESMO investigational immunotherapy faculty member, ESMO Women for Oncology Committee member, Immuno-Oncology Society of India (IOSI) founder general secretary, Indian Cooperative Oncology Network (ICON) managing committee member, Indian Society of Medical and Paediatric Oncology (ISMPO) executive committee member, Teenage and Young Adult Cancer Foundation (TYAcan) founder member and joint secretary. SB reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Amgen, AstraZeneca, Clovis, Epsilogen, Genmab, GSK, Immunogen, Medscape, Merck Serono, Mersana, MSD, Oncxerna, Peerview, Pfizer, Research to Practice, Roche, Shattuck Labs, Takeda; research grants for: AstraZeneca, GSK. ASB reports research support from Daiichi Sankyo (≤10 000€), Roche (>10 000€) and honoraria for lectures, consultation or advisory board participation from Roche Bristol-Meyers Squibb, Merck, Daiichi Sankyo, AstraZeneca, CeCaVa (all <5000€) as well as travel support from Roche, Amgen and AbbVie. SPC reports personal and/or institutional financial interests in the form of advisor or invited speaker for: AstraZeneca, Bristol Myers Squibb, Ipsen, DKSH, Amgen, MSD, Merck, Roche. RD reports personal and/or institutional financial interests in the form of advisor or invited speaker for: AstraZeneca, Eisai, Lilly, Merck, Pfizer, Roche; reports personal and institutional financial interests in the form of research grant for AstraZeneca; reports non-financial interests: AstraZeneca Steering Committee member and Roche trial chair. EF reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Amgen, AstraZeneca, Bayer, Bristol Myers Squibb, DAICHII SANKYO, Eli Lilly, F. Hoffmann-La Roche, Glaxo Smith Kline, Janssen, Medical Trends, Medscape, Merck Serono, Merck Sharp & Dohme, NOVARTIS, Peervoice, Peptomyc, Pfizer, Sanofi, Takeda, TouchONCOLOGY; reports personal and/or institutional financial interests in the form of local PI for: Abbvie Deutschland GmbH & Co KG, Amgen Inc, AstraZeneca AB, Bayer Consumer Care AG, Boehringer Ingelheim International GmbH, Bristol-Myers Squibb International Corporation (BMS), Daiichi Sankyo Inc, Exelixis Inc, F. Hoffmann-La Roche Ltd, GlaxoSmithKline Research & Development Limited, Janssen Cilag International NV, Merck KGAA, Merck Sharp & Dohme Corp, Mirati Therapeutics Inc, Novartis Farmaceutica SA, Pfizer S.L.U., Takeda Pharmaceuticals International; reports other personal and/or institutional financial interests for FUNDACIÓN MERCK SALUD, in the form of grant for oncology innovation for MERCK HEALTHCARE KGAa, in the form of independent member of the Board of Directors for Grifols; reports non-financial interests: Member of ESMO Nominating Committee and Compliance Committee, ETOP (European Thoracic Oncology Platform) Member of Scientific Committee, SEOM (Sociedad Espanola de Oncologia Medica) President 2021-2023. AF reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Bristol Myers Squibb, Eisai, Immuncore, Ipsen; reports non-financial interest in the form of advisor for Erase Mesothelioma. MG reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Astrazenca/MedImmune, AstraZeneca and Daiichi Sankyo Oncology Teams, AstraZeneca Poland, AstraZeneca Spain, Bayer Healthcare Pharmaceuticals, Blueprint Medicines, Daiichi Sankyo, Daiichi Sankyo, Inc., Daiichi Sankyo/AstraZeneca, ecancer, Eli Lilly, GlaxoSmithKline, GrupoPacifico-Secretaria Técnica ICAPEM/AstraZeneca, Incyte, Medscape, Mirati Therapeutics, Inc., MSD, MSD Italia, Regeneron Pharmaceuticals, Roche, S.O.S S.r.l, Sanofi/Prex, Sanofi Genzyme corporation, SeaGen International GmbH, Takeda, WebMD, WebMD Oncology/Takeda; reports other personal and/or institutional financial interests for: Astra Zeneca UK (PACIFIC-R Scientific Committee), AstraZeneca (PACIFIC-R Global Scientific Committee), AstraZeneca (Pacific 6 International Coordinating Investigator), AstraZeneca UK (Expert Testimony), GSK, (GSK-Garassino-ZEAL Steering Committee 2020-23), GSK (GSK Lung Cancer Global Council), Janssen (Jannesen Scientific Advisory Board and Therapeutic Area Steering Committee Meeting on Lung Cancer), Pfizer (Pfizer Global Lung Cancer Educational Programme—Steering Committee Pfizer, Advisory Board), Seattle Genetics (Seattle Genetics Lung Cancer Platform Study); reports personal and/or institutional financial interests in the form of local or coordinating PI or steering committee member for: Amgen, Astrazeneca, AstraZeneca AB, AstraZeneca S.p.A., Bayer, Bluprint, BMS, Celgene Corporation, Daiichi Sankyo Development Ltd., Exelixis Inc., GlaxoSmithkline Research & Development Ltd., Incyte Corporation, IPSEN Bioscience Inc., Janssen, MedImmune LCC, Merk, Merk Serono, MSD, Novartis, Otsuka Pharmaceutical Italy S.r.l., Pfizer, Roche, Roche Sanofi, Spectrum Pharmaceuticals, Turning Point Therapeutics, Inc.; reports non-financial interests in the form of PI for: AO Spedali Civili Brescia, Eli Lilly, European Thoracic Oncology Platform (ETOP), GOIRC, GUSTAVE-ROUSSY PARIGI LIPI TRIAL, Istituto dei Tumori Pascale - Napoli, MSD, Pfizer, Sant'Orsola Malpighi—Bologna (Alma Mater Studiorum Università Bologna); reports non-financial interests: AACR (Scientific Programme Committee), AIOM Member, AIOT (Associazione Italiana Oncologia Toracica Member and Board Member), ASCO (Member of ASCO Scientific Committee 2018-2021), EMA Scientific Advisory Group (SAG) Member, ESMO National Societies Committee Chair and ESMO Council Member, IPOP (Italian lung cancer charity) Scientific Committee member, TUTOR (Italian thymic malignancies charity) Scientific Committee member, WCLC member and WCLC annual congress Lung Cancer Track, Women for Oncology Italy President and Founder. EG reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Alkermes, Anaveon, Boehringer Ingelheim, Bristol-Mayers Squibb, Ellipses Pharma, F-Star Therapeutics, F.Hoffmann/La Roche, Genentech, Hengrui, Janssen Global Services, Lilly, MabDiscovery, MSD, Neomed Therapeutics1 Inc, Seattle Genetics, Sotio,Thermo Fisher; reports institutional financial interests in the form of funding for: AstraZeneca, Novartis, Roche Taiho, Thermo Fisher. PGL reports personal financial interests as follows: advisory role: Abbvie, Amgen, AstraZeneca, Bayer, BMS, GSK, Janssen, Lilly, MSD, Novartis, Pfizer, Roche, Takeda, Sanofi; speaker: Janssen, MSD, Novartis, Medscape, Takeda, TouchTime; reports other personal and/or institutional financial interests for: AstraZeneca (PACIFIC-R Global Scientific Committee), Novartis (CACZ885V2201C; CINC280I12), IO Biotech (IO102-012/KN-764), Janssen (JNJ-73841937). Janssen Medical Education Steering Committee; reports non-financial interests: ASCO (Member of ASCO Scientific Committee 2021-2023), EMA Scientific Advisory Group (SAG) Member, ESMO Women for Oncology Committee Chair and ESMO Council Member, IASLC (education committee member, Women In Thoracic Onoclogy Committee Member and Academy member), Lung Cancer Policy Network Committee Member. Annals of Oncology associate editor. AKK reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Amgen, Astellas, Astra Zeneca, Elli-Lilly, Merck, MSD, Novartis, Pfizer, Roche, Sanofi, Teva, Yanssen. AL reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Astra Zeneca, Bayer, BMS, Incyte, Lilly, Merck, MSD, Pfizer, Pierre Fabre Pharma GmbH, Sanofi, Servier, Takeda, Tesaro; reports personal and/or institutional non-financial interests for: Böhringer Ingelheim; other non-financial interests: German Cancer Society (DKG) Member of several Certification Commissions in the German Certification Program, German Society for Palliative Care (DGP) Member of Board of Directors, German Society Hematology and Oncology (DGHO) Working-Group Co-Leader: Tumor-Associated Fatigue, Individualized Medicine. HL reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Amgen, Astra Zeneca, Bristol Myers Squibb, Lilly, Merck, MSD, Novartis, Pfizer, Roche, Takeda; reports personal and/or institutional financial interests in the form of local PI for: Abbvie, Amgen, Astra Zeneca, Boehringer Ingelheim, Bristol Myers Squibb, Health Data Specialist, Lilly, Novartis, Parexel ILR, PPD, Qualitis, Local PI, Roche; reports non-financial interests: Fairlife Lung Cancer Care Member of Board of Directors, Hellenic Cooperative Oncology Group Principal Investigator, Hellenic Cooperative Oncology Group Elected President of the Scientific Committee Women 4 Oncology—Hellas, Member of Board of Directors. AM reports personal financial interests in the form of advisor for: BMS, MSD, Novartis, Pierre-Fabre, QBiotics, Roche. DM reports personal financial interests in the form of advisor or invited speaker for: Astellas, Astra Zeneca, Bayer, BMS, Janssen, MSD, Pfizer; reports institutional non-financial interests in the form of research grant for: Astellas, BMS, Janssen, Novartis. SP reports institutional financial interests in the form of advisor or invited speaker for: AbbVie, Amgen, Arcus, AstraZeneca, Bayer, BeiGene, Bio Invent, Biocartis, Blueprint Medicines, BMS, Boehringer Ingelheim, Daiichi Sankyo, Debiopharm, Ecancer, Eli Lilly, Fishawack, Foundation Medicine, Genzyme, Gilhead, GlaxoSmithKline, Illumina, Imedex, Incyte, IQVIA, iTheos, Janssen, Medscape, Merck Serono, Mirati, MSD, Novartis, Novocure, OncologyEducation, PER, Pfizer, PharmaMar, Phosplatin Therapeutics, PRIME, Regeneron, RMEI, Roche/Genentech, RTP, Sanofi, Seattle Genetics, Takeda, Advisory Board, Vaccibody; reports financial interests for Elsevier (Associate Editor Annals of Oncology F-Star); reports institutional non-financial interests in the form of coordinating PI or steering committee or trial chair member for: AstraZeneca, Beigene, BMS, GSK, iTeos, Mirati, MSD, Pharma Mar, Phosplatin Therapeutics, Roche/Genentech; reports non-financial interests: Ballet Béjart Lausanne Foundation (President and Council Member), ESMO Officer and President 2020-2022, ETOP/EORTC/SAKK (Principal Investigator, involved in academic trials), ETOP/IBCSG Partners (Council Member & Scientific Committee Chair), SAKK Vice-President Lung Group, Involved in Swiss politics, AACR Member, ASCO Member, ASMAC/VSAO Member, Association of Swiss interns and residents; Association of Swiss Physicians FMH (CH) Member, IASLC Member, SAKK Member, Vice-President Lung Group SAMO. CS reports non-financial interests: ESO (Coordinator Gynecological Programme). JWHT reports non-financial interests: Anglican Health & Community Network (AHCN) Co-convenor, Hon. Advisor for the Global Chinese Breast Cancer Organizations Alliance, Hong Kong Breast Cancer Foundation (HKBCF), Steering Committee Members of the Hong Kong Breast Cancer Registry under the HKBCF, Founding Convenor of the Hong Kong Breast Oncology Group (HKBOG), Executive Committee Member and Medical Advisor for the Hong Kong Cancer Fund (HKCF), Hon. Secretary of the Hong Kong Cancer Therapy Society (HKCTS), Founding Hon. Advisor for the Hong Kong Christian Cancer Care Association (HKCCCA), Founding Convenor of the Hong Kong Women for Oncology (HKW4O), International Society of Geriatric Oncology (SIOG) Member of the Publication Committee. JCHY reports personal and/or institutional financial interests in the form of advisor or invited speaker for: Amgen, Astrazeneca, Bayer, Boehringer Ingelheim, Daiichi Sankyo, Eli Lilly, GSK, Janssen, Merck Serono, MSD, Novartis, Novartis, Ono Pharmaceuticals, Pfizer, Puma Pharmaceuticals, Roche/Genentech, Takeda Oncology, Yuhan Pharmaceuticals; reports institutional financial interests in the form of coordinating PI for Astrazeneca; reports non-financial interests in the form of coordinating PI or steering committee member for: Bayer, Daiichi Sankyo, Dizal Pharmaceutical, Eli Lilly, Ipsen, Janssen, Merck, MSD, Novartis, Numab, Takeda Oncology; reports non-financial interests: ASCO member. CMDCM has declared no conflicts of interest.
Sex and gender perspectives in colorectal cancer	3f60bbf7-df36-45fb-a1e1-813ba1b9331d	10163160	Internal Medicine[mh]	With the exception of thyroid cancer, incidence of non-reproductive tumours is higher in males than in females, while mortality rates in males doubling those in females. Sex does not only influence cancer incidence, but also clinicopathological features of disease, differences in treatments, therapeutic outcomes, and tolerability. These sex-associated differences are known as sexual dimorphisms. However, it was not until recently that sex disparities in oncology have been acknowledged. Although interrelated and often used interchangeably, sex and gender are not equivalent concepts. Sex is a biological attribute that defines species, including humans, as male, female, and/or intersex according to their chromosomal makeup and reproductive organs. Gender, on the other hand, is a chosen sexual identity and represents a social construct that refers to the norms, identities, and relations that structure our societies and organizations, and shape behaviours, products, technologies, environments, and knowledge. Gender is a dynamic concept that varies from society to society and can change throughout an individual’s lifetime. Although appropriate reporting of sex and gender in oncology practices is vital, the incorporation of gender variables in clinical research and patients’ medical histories remains limited, restricting us from adequately addressing the impact of gender-influenced behaviours on health outcomes, which are different from those influenced by biological sex. Colorectal cancer (CRC) is the third most common cancer worldwide and the fourth leading cause of cancer death in the world. As in other tumours, there are differences in incidence according to sex. Worldwide estimation of incident cases in 2020 was 547 619 and 288 852 cases of colon and rectum cancer in females, respectively, and 600 896 and 443 358 cases of colon and rectum cancer in males, respectively. This review provides context for understanding sex and gender differences in cancer and summarizes the growing body of literature illustrating the sex and gender perspective in CRC and their impact in relation to tumour biology and treatment efficacy and toxicity. Women were historically largely excluded as subjects of investigations in non-reproductive clinical research, resulting in the extrapolation of data from male-based investigations to women. , However, underlying and fundamental differences in biology are likely to affect disease development and pharmacokinetics, and impact treatment efficacy and toxicity, which have been widely described. Following the scandals resulting from the use of thalidomide in women during pregnancy, warnings about fetal risks led to the labelling of pregnant women by the National Commission for the Protection of Human Subjects and Biomedical and Behavioural Research as vulnerable research subjects. In 1977, the United States Food and Drug Administration (FDA) issued a guidance document entitled “General Considerations for the Clinical Evaluation of Drugs” advising that women of childbearing potential should be excluded from early-phase clinical research, with the exception of trials testing drugs for life-threatening illnesses. Women could be included in later phase II and III trials for drugs with a favourable risk–benefit ratio, as long as studies about teratogenicity and fertility had been accomplished. However, the term ‘woman capable of becoming pregnant’ covered a broad range of women, since it could include premenopausal single abstinent women, women using contraceptives, and women with sterile partners, whereas it did not account for the reproductive desires of women and their partners. Thus, advocacy groups criticized the 1977 FDA guideline by arguing that it deprived women of opportunities, and did not focus on women’s independence to make decisions. They also voiced that women were capable of endorsing drug development about sex differences through clinical research participation; furthermore, this policy had an unintended consequence of causing underrepresentation of women in clinical research. In 1986, the National Institutes of Health (NIH) reinforced the movement by establishing a policy that urged the inclusion of women in clinical trials and finally, in 1993, the FDA reversed the 1977 guidance and lifted the ban that prevented women of childbearing potential from being enrolled in early-phase trials and promulgated “Guidelines for the Study and Evaluation of Gender Differences in the Clinical Evaluation of Drugs”. The document pointed out that clinical trial subjects should be representative of the patient population to which the drug would likely be prescribed after approval, and highlighted the importance of exploring differences in terms of safety, efficacy, pharmacokinetics, and pharmacodynamics among subpopulations. In 1994, the FDA furthered its commitment with the creation of an Office of Women’s Health to address the health of women through policy, science, and outreach and to promote the inclusion of women in clinical studies as well as subanalyses of sex, gender, and subpopulations. In 1998, the FDA amended its regulations pertaining to new drug applications (NDAs) and announced the publication of a new document “Presentation of Safety and Effectiveness Data for Certain Subgroups of the Population in Investigational New Drug Application Reports and New Drug Applications” that specifically stated that safety and efficacy data for important populations, including sex, age, and racial subgroups, were mandatory for NDAs. In 2000, the FDA issued the “Investigational New Drug Applications: Amendment to Clinical Hold Regulations for Products Intended for Life-Threatening Disease and Conditions” that permitted placing on hold any trial for a life-threatening condition that excluded patients only because of their reproductive potential. Although today women are systematically included in clinical trials, inadequate analysis of differences in outcomes according to sex or gender remains widespread and reflects an unmet need deserving of attention and deeper knowledge. The powerhouse of clinical cancer research springs from in vitro workhorse models using cell lines and validation in in vivo animal models, including patient-derived models. However, historical ignorance of the role that sex plays across this process has widely jeopardized bench-to-bedside cancer research, obscuring fundamental sex differences that may guide clinical studies. Male cell lines stocked in repositories of human non-reproductive cancer cell lines outnumber female cell lines, leading to single-sex analyses. Furthermore, few in vitro cell-based experiments report the sex of the cell line, posing a potential risk to analysis and interpretation and, even when investigators do acknowledge the sex origin of a cell line, the original sexual identity may transform over the course of routine cell culture passaging. , Added to this, the sex of cultured cells and that of cell culture media are rarely matched. In fact, the effect of culture media components on the hormonal environment of cultured cells is seldom considered. Hormone concentrations in calf fetus sera may differ depending on whether the serum is sourced from a single sex or a mixture, whereas hormone levels are not routinely measured and sex identity is usually not reported. Finally, the estrogenic contribution of plastic labware commonly used for cell culture or that of the common pH dye indicator phenol red is also rarely considered. , , Phenol red, present in most commercial media, turns yellow in response to acidification of the medium during cell growth; however, binding and activation of estrogen receptors in multiple cell lines in a dose-dependent manner has been reported. In the case of in vivo experiments using animal models, bias concerning the sex parameter also exists. Species-specific sex differences, and the fact that these vary between species [e.g. that fewer genes escape X chromosome inactivation (XCI) in female mice than in humans, and X-linked gene regulation] should be taken into consideration. , However, research invariably fails to investigate sex disparity by the differential use of models of male and female animals and furthermore, when using xenografts, investigators seldom account for matching the sex of transplanted cells and their animal recipients, even though this is not consistent with reliable modelling. Plus, as males present aggressive behaviour that requires cage separation, cost concerns often result in favouring a single-sex study in co-caged female mice of premenopausal age, even though human cancers are predominantly diagnosed at a late age. , In light of this and striving for the inclusion of both sexes in preclinical research, in 2014 the NIH unveiled a new policy requiring federally funded scientists to include both males and females in cell and animal studies. Exposome The exposome, defined as the repertoire of exposures and associated interactions of a given individual during their lifetime, may differ among individuals according to their gender. The exposome is known to have an impact on cancer development, with obesity, smoking habits, and inflammation having a direct and strong correlation with carcinogenesis in multiple tumour types , , . Concerning CRC, external exposures such as dietary habits, exercise, smoking, drinking, socioeconomic status, social support, and industrial pollution are known risk factors for colorectal carcinogenesis. , , , , , , , , Exposure can differ according to gender, while strength and direction of the correlation might be dependent on sex and/or gender. As such, stronger associations between obesity and increased risk of overall CRC have been found in men, compared with women. Patterns also differ: weight gain later in life seems to be an important risk factor for CRC in men, while in women, early life obesity is a known risk factor. , , However, to date, sex-specific differences have not been sufficiently investigated for CRC. Findings about dietary patterns, which are related to obesity, and their contribution to CRC have been reported mainly from female-specific prospective cohort studies such as the Nurses’ Health. However, unfortunately, large cohort studies that included both women and men mostly have not reported sex-specific estimates. To date, investigations have focused on tumour location and analysed sex differences in terms of risk by colorectal subsite according to high carbohydrate intake, concluding that it might increase right-sided colon cancer in women, while increasing rectal cancer in men. A high inflammatory profile (proinflammatory diet plus sedentarism plus obesity) has been associated with higher risk for colon cancer in men and no significant association in women in the European Prospective Investigation into Cancer and nutrition study (EPIC), while in another study, soy consumption (a known phytoestrogen) was not associated with risk of CRC in males, but risk reduction in females was reported. , However, even after adjusting for external exposures, incidence of non-reproductive cancers is lower in females than in males, suggesting sexual dimorphisms. , In the following sections, we focus on the complex role that sex chromosomes and sex hormones appear to have in this context. Sex chromosomes and cancer Male and female development diverges under the influence of both X and Y allosomes (sexual chromosomes) and autosomes (non-sexual chromosomes) and the contribution of sex steroid hormones. However, these sexual dimorphisms might also contribute to sex disparities in cancer development. Since female cells harbour two entire X chromosomes, a critical XCI to avoid simultaneous expression of two entire X chromosomes must occur at early cell division in implanted embryos. This phenomenon is mediated by the long non-coding RNA X-inactive specific transcript (XIST), which silences one X pair at a random. As a consequence, XX cells present a silenced and inactive X chromosome (Xi) and an expressed and active one (Xa). As XX cells express either the maternal or the paternal X arbitrarily, females and males with Klinefelter syndrome present distinct X gene repertoires, resulting in mosaicism and phenotypic diversity. This contrasts with the exclusive expression of the only X chromosome in XY cells from males. However, some genes escape and are expressed from both the Xa and the Xi, conferring protective benefits from cancer and other diseases, but also potential risks. X-linked oncogene activation or loss of a tumour suppressor would not be expressed in all cells due to mosaicism, while in males it would lead to obligatory expression of the same alteration in the maternal X-linked gene. Tumour suppressor genes that escape from X-inactivation, termed EXITS, are ATRX , CNKSR2 , DDX3X , KDM5C , KDM6A , and MAGEC3 , and might be responsible for the lower propensity of cancer in females, in comparison with males. , On the other hand, the loss of XIST expression may promote tumour development, as the deletion of XIST expression causes the reactivation of the Xi chromosome, triggering unfavourable genome-wide changes, including involvement in DNA replication, chromosome segregation, cell cycle checkpoints, and haematopoietic genetic disorders. , , In males, loss of Y chromosome expression might also constitute a sex-specific biomarker, as it is related to six Y-linked genes known to serve as tumour suppression genes ( KDM6C , KDM5D, DDX3Y, EIF1AY , RPS4Y1 , and ZFY ). Reduced transcription levels of these genes have been found in 12 non-reproductive tumour types to date. , Immunity and cancer Influenced by sex hormones, sex chromosomes also have an impact on cancer immune defences. In general, females are able to mount stronger innate and adaptive immune responses than males, which results in faster clearance of pathogens and greater vaccine efficacy in females than in males, but also explains the greater incidence of inflammatory and autoimmune diseases in females. , Approximately 50 X-linked genes are involved in adaptive and innate immunity. These X-linked genes code for proteins including pattern recognition receptors [such as Toll-like receptor 7 (TLR7) and TLR8], cytokine receptors [e.g. Interleukin 2 receptor subunit gamma (IL2RG) and Interleukin 13 receptor subunit alpha 2 (IL13RA2)], and transcriptional factors (e.g. FOXP3). The X-linked Forkhead box P3 (FOXP3), expressed in regulatory T (Treg) cells, is critical for immune homeostasis, as it limits the adaptive immune responses. In comparison with female visceral adipose tissue (VAT), male VAT is enriched in FOXP3+ Treg cells and presents a distinct molecular profile that is enforced by a sex hormone-dependent niche. The X-linked TLR8 also performs a central role in Treg cell biology. Activation of TLR8 in Treg cells triggers selective inhibition of glucose uptake leading to their senescence, relieving Treg cell inhibition of effector T cells. In the case of the TLR7 gene, TLR7 may escape XCI in immune cells, leading to enhanced TLR7 expression owing to biallelism in XX cells, which contributes to the higher risk of developing autoimmune disorders in women and in men with Klinefelter syndrome. Concerning lymphoid cell subsets, adult females present a higher frequency of B cells during adulthood, whereas higher CD4+ T-cell counts, higher CD4/CD8 ratios, and lower presence of CD8+ T cells are reported throughout life, compared with age-matched males. , , , Immune defence against cancer declines with age and shows sexual dimorphisms. Reduction in T-cell numbers occurs in both sexes with age, but a disproportionate decrease in T-cell and B-cell populations is more evident in older males. Two waves of epigenetic regulation depleting immune cell functions have been identified, the first one in the late 30s, with a similar impact across the sexes. Genomic differences between sexes increase after the age of 65 years, with a second wave in males in their early 60s that results in increased proinflammatory activity and innate immunity and lower adaptive immunity. This is delayed by 5-6 years in females, who exhibit greater adaptive immunity. A pan-cancer analysis to evaluate the sex-based variance of different genomic immune-related factors using The Cancer Genome Atlas showed differences in tumour mutation burden, neoantigen burden, tumour purity, cytolytic activity, CD8+ T cell, and expressions of immune checkpoint genes according to sex. Sex hormones and cancer Sex hormones exert pleiotropic functions on multiple tissues. The estrogen effect depends on the activation of the estrogen receptor-α (ERα) and -β (ERβ). Activation of ERα promotes expansion and mobilization of haematopoietic stem cells, favours skin wound healing, promotes angiogenesis and endothelial cell precursor mobilization, reduces hepatocyte proliferation, and restrains the inflammatory role of macrophages. ERβ also blocks macrophage activation and, contrary to ERα, negatively regulates vessel formation. The androgen receptor promotes angiogenesis, liver cell proliferation, and macrophage activation through the stimulation of tumour necrosis factor and CC-chemokine ligand 4, while it also suppresses wound healing. Hormone pathways are interrelated, since androgenic hormones are converted into estrogens through the action of aromatase, resulting in indirect control of pathways affected by ERα. In the tumour microenvironment, cancer cells secrete vascular endothelial growth factor (VEGF) A and platelet-activating factor in response to estrogens, enhancing proliferation and migration. , In addition, estrogens promote mobilization of bone marrow-derived precursors to cancer stroma in breast tumours. Sex hormones are partially responsible for the sex-related differences in immune response as ERα and ERβ are expressed by many types of immune cells, including T cells, B cells, dendritic cells, macrophages, neutrophils, and natural killer (NK) cells and these receptors present differential expression among immune cell subsets, as ERα is up-regulated in T cells while ERβ is highly expressed in B cells. Finally, age-related changes in sex steroid concentrations and sex steroid receptor signalling might subsequently contribute to age-associated changes of immune function and populations. The exposome, defined as the repertoire of exposures and associated interactions of a given individual during their lifetime, may differ among individuals according to their gender. The exposome is known to have an impact on cancer development, with obesity, smoking habits, and inflammation having a direct and strong correlation with carcinogenesis in multiple tumour types , , . Concerning CRC, external exposures such as dietary habits, exercise, smoking, drinking, socioeconomic status, social support, and industrial pollution are known risk factors for colorectal carcinogenesis. , , , , , , , , Exposure can differ according to gender, while strength and direction of the correlation might be dependent on sex and/or gender. As such, stronger associations between obesity and increased risk of overall CRC have been found in men, compared with women. Patterns also differ: weight gain later in life seems to be an important risk factor for CRC in men, while in women, early life obesity is a known risk factor. , , However, to date, sex-specific differences have not been sufficiently investigated for CRC. Findings about dietary patterns, which are related to obesity, and their contribution to CRC have been reported mainly from female-specific prospective cohort studies such as the Nurses’ Health. However, unfortunately, large cohort studies that included both women and men mostly have not reported sex-specific estimates. To date, investigations have focused on tumour location and analysed sex differences in terms of risk by colorectal subsite according to high carbohydrate intake, concluding that it might increase right-sided colon cancer in women, while increasing rectal cancer in men. A high inflammatory profile (proinflammatory diet plus sedentarism plus obesity) has been associated with higher risk for colon cancer in men and no significant association in women in the European Prospective Investigation into Cancer and nutrition study (EPIC), while in another study, soy consumption (a known phytoestrogen) was not associated with risk of CRC in males, but risk reduction in females was reported. , However, even after adjusting for external exposures, incidence of non-reproductive cancers is lower in females than in males, suggesting sexual dimorphisms. , In the following sections, we focus on the complex role that sex chromosomes and sex hormones appear to have in this context. Male and female development diverges under the influence of both X and Y allosomes (sexual chromosomes) and autosomes (non-sexual chromosomes) and the contribution of sex steroid hormones. However, these sexual dimorphisms might also contribute to sex disparities in cancer development. Since female cells harbour two entire X chromosomes, a critical XCI to avoid simultaneous expression of two entire X chromosomes must occur at early cell division in implanted embryos. This phenomenon is mediated by the long non-coding RNA X-inactive specific transcript (XIST), which silences one X pair at a random. As a consequence, XX cells present a silenced and inactive X chromosome (Xi) and an expressed and active one (Xa). As XX cells express either the maternal or the paternal X arbitrarily, females and males with Klinefelter syndrome present distinct X gene repertoires, resulting in mosaicism and phenotypic diversity. This contrasts with the exclusive expression of the only X chromosome in XY cells from males. However, some genes escape and are expressed from both the Xa and the Xi, conferring protective benefits from cancer and other diseases, but also potential risks. X-linked oncogene activation or loss of a tumour suppressor would not be expressed in all cells due to mosaicism, while in males it would lead to obligatory expression of the same alteration in the maternal X-linked gene. Tumour suppressor genes that escape from X-inactivation, termed EXITS, are ATRX , CNKSR2 , DDX3X , KDM5C , KDM6A , and MAGEC3 , and might be responsible for the lower propensity of cancer in females, in comparison with males. , On the other hand, the loss of XIST expression may promote tumour development, as the deletion of XIST expression causes the reactivation of the Xi chromosome, triggering unfavourable genome-wide changes, including involvement in DNA replication, chromosome segregation, cell cycle checkpoints, and haematopoietic genetic disorders. , , In males, loss of Y chromosome expression might also constitute a sex-specific biomarker, as it is related to six Y-linked genes known to serve as tumour suppression genes ( KDM6C , KDM5D, DDX3Y, EIF1AY , RPS4Y1 , and ZFY ). Reduced transcription levels of these genes have been found in 12 non-reproductive tumour types to date. , Influenced by sex hormones, sex chromosomes also have an impact on cancer immune defences. In general, females are able to mount stronger innate and adaptive immune responses than males, which results in faster clearance of pathogens and greater vaccine efficacy in females than in males, but also explains the greater incidence of inflammatory and autoimmune diseases in females. , Approximately 50 X-linked genes are involved in adaptive and innate immunity. These X-linked genes code for proteins including pattern recognition receptors [such as Toll-like receptor 7 (TLR7) and TLR8], cytokine receptors [e.g. Interleukin 2 receptor subunit gamma (IL2RG) and Interleukin 13 receptor subunit alpha 2 (IL13RA2)], and transcriptional factors (e.g. FOXP3). The X-linked Forkhead box P3 (FOXP3), expressed in regulatory T (Treg) cells, is critical for immune homeostasis, as it limits the adaptive immune responses. In comparison with female visceral adipose tissue (VAT), male VAT is enriched in FOXP3+ Treg cells and presents a distinct molecular profile that is enforced by a sex hormone-dependent niche. The X-linked TLR8 also performs a central role in Treg cell biology. Activation of TLR8 in Treg cells triggers selective inhibition of glucose uptake leading to their senescence, relieving Treg cell inhibition of effector T cells. In the case of the TLR7 gene, TLR7 may escape XCI in immune cells, leading to enhanced TLR7 expression owing to biallelism in XX cells, which contributes to the higher risk of developing autoimmune disorders in women and in men with Klinefelter syndrome. Concerning lymphoid cell subsets, adult females present a higher frequency of B cells during adulthood, whereas higher CD4+ T-cell counts, higher CD4/CD8 ratios, and lower presence of CD8+ T cells are reported throughout life, compared with age-matched males. , , , Immune defence against cancer declines with age and shows sexual dimorphisms. Reduction in T-cell numbers occurs in both sexes with age, but a disproportionate decrease in T-cell and B-cell populations is more evident in older males. Two waves of epigenetic regulation depleting immune cell functions have been identified, the first one in the late 30s, with a similar impact across the sexes. Genomic differences between sexes increase after the age of 65 years, with a second wave in males in their early 60s that results in increased proinflammatory activity and innate immunity and lower adaptive immunity. This is delayed by 5-6 years in females, who exhibit greater adaptive immunity. A pan-cancer analysis to evaluate the sex-based variance of different genomic immune-related factors using The Cancer Genome Atlas showed differences in tumour mutation burden, neoantigen burden, tumour purity, cytolytic activity, CD8+ T cell, and expressions of immune checkpoint genes according to sex. Sex hormones exert pleiotropic functions on multiple tissues. The estrogen effect depends on the activation of the estrogen receptor-α (ERα) and -β (ERβ). Activation of ERα promotes expansion and mobilization of haematopoietic stem cells, favours skin wound healing, promotes angiogenesis and endothelial cell precursor mobilization, reduces hepatocyte proliferation, and restrains the inflammatory role of macrophages. ERβ also blocks macrophage activation and, contrary to ERα, negatively regulates vessel formation. The androgen receptor promotes angiogenesis, liver cell proliferation, and macrophage activation through the stimulation of tumour necrosis factor and CC-chemokine ligand 4, while it also suppresses wound healing. Hormone pathways are interrelated, since androgenic hormones are converted into estrogens through the action of aromatase, resulting in indirect control of pathways affected by ERα. In the tumour microenvironment, cancer cells secrete vascular endothelial growth factor (VEGF) A and platelet-activating factor in response to estrogens, enhancing proliferation and migration. , In addition, estrogens promote mobilization of bone marrow-derived precursors to cancer stroma in breast tumours. Sex hormones are partially responsible for the sex-related differences in immune response as ERα and ERβ are expressed by many types of immune cells, including T cells, B cells, dendritic cells, macrophages, neutrophils, and natural killer (NK) cells and these receptors present differential expression among immune cell subsets, as ERα is up-regulated in T cells while ERβ is highly expressed in B cells. Finally, age-related changes in sex steroid concentrations and sex steroid receptor signalling might subsequently contribute to age-associated changes of immune function and populations. The tumour biology of CRC has been proven to be different in males and females as a result of the sex hormones and sex chromosomes that can influence immunity. In addition, key proliferative pathways in CRC tumourigenesis are regulated through estrogens. Estrogens have been reported to control the activity of a class of Kv channels (KCNQ1 : KCNE3), which regulate fundamental ion transport functions of the colon and ultimately promote cell proliferation and epithelial–mesenchymal transition through bi-directional interactions with the Wnt/β-catenin signalling pathway. At the same time, estrogen modulates proliferative responses to hypoxia via the novel membrane estrogen receptor G protein-coupled estrogen receptor (GPER), as well as by Hypoxia-inducible factor 1-alpha (HIF1A) and VEGF signalling. Differences in oncogene expression, such as a higher frequency of mutations of STK11 in males, and sexual dimorphisms in proteomes of CRC cells may contribute to the disparities in tumour biology according to sex in these tumours. , , Lastly, sexual dimorphisms in the tumour microenvironment of colorectal tumours have been investigated using tissue microarrays comprising primary tumour, tumour-infiltrated lymph nodes, and uninvolved colon. Differential gene expression was observed in pathways related to Treg function, T-cell activity, and T-cell exhaustion, amongst several others, in females compared to males. Whilst globally females diagnosed with CRC have better overall survival compared with males, , in some countries the 5-year survival rate among women has been reported to be lower than among men, especially after the age of 70 years. The proportion of patients presenting right-sided tumours is higher among females than males, as it is the proportion of BRAF -mutated tumours. Right-sided colon tumours are often at a more advanced stage at diagnosis and are less differentiated, which might partially explain this lower 5-year survival rate in females. Sex hormones may explain the higher frequency of right-sided CRC in females. It has been proposed that exposure to estrogen is protective against the development of tumours with microsatellites instability (MSI), while the lack of estrogen in older females might increase the risk of MSI-high CRC. PIK3CA mutations, associated with poorer prognosis, and methylation of CpG island in the 5ʹ region of the p 16 INKa tumour suppressor also occur at a higher frequency in females. , Sex-associated differences in the microbiome have also been reported in healthy individuals during their life span, some of which are mediated by sex hormones and conditioning the estrobolome (a gene repertoire of intestinal microbiota able to carry out estrogen metabolism). , Plus, the microbiome is highly conditioned by the exposome, which might also vary according to gender. In patients with CRC, sexual dimorphism of microbiome has also been reported. The microbiome might be more stable in the male gut than in the female gut. The male gut shows an enrichment of rare species that may contribute to the stability of microbial communities, whereas in the female gut there is a loss of species that could be responsible for the vulnerable microbial communities with the development of CRC. Chemotherapy agents Fluoropyrimidines are the backbone of chemotherapy in CRC management. Dosage is based on body surface area but does not take into account sex differences. However, a substantial body of literature indicates sex-associated differences in pharmacokinetics of 5-fluorouracil (5-FU). Plasma clearance, plasma clearance per kilogram, and dose have been found to be lower in females compared with males, whereas plasma life and area under the plasma concentration–time curve might be higher in females. On the contrary, volume of distribution, volume of distribution per kilogram, and dose per kilogram do not differ significantly between sexes. These differences may have an impact on outcomes and toxicity. Toxicity associated with 5-FU has been reported to be more extensive for females than for males in terms of the number of different types of toxicity, maximum toxicity grade, and incidence of severe toxicities, including haematologic toxicities such as leukopaenia, neutropaenia, and thrombocytopaenia, and self-reported toxicity such as stomatitis, diarrhoea, nausea, vomiting, or hand-foot syndrome. , , This increased toxicity might be due to lower clearance of 5-FU leading to higher plasma levels in females compared to males, as previously reported. Concerning patient-reported outcomes, an investigation exploring the occurrence and severity of self-reported physical and psychological co-occurring symptoms in patients with stage IV CRC receiving different 5-FU-based chemotherapy schemes reported more severe worrying, lack of energy, and nausea in women. Similar data about toxicity have been reported with the use of capecitabine. In a cohort of 299 patients (163 males, 136 females) receiving capecitabine in the adjuvant setting, females had significantly higher dose-limiting toxicity than men. Incidence of all common toxicities was higher in females and required significantly more dose reductions than males, leading to statistically significant lower relative dose intensities in females. Consistent with these results, an analysis of 36 640 patients with colon cancer receiving adjuvant fluoropyrimidine-based chemotherapy in the ACCENT database showed that incidence of grade 3 or grade 4 non-haematological toxicity (nausea, vomiting, stomatitis, diarrhoea, peripheral neuropathy, and transaminitis) and haematological toxicity (neutropaenia and leukopaenia) was statistically significant and clinically relevant higher in females. Differential efficacy according to sex has only recently been studied. The SOLSTICE trial, comparing standard capecitabine and bevacizumab versus TAS102 and bevacizumab as frontline therapy for patients with metastatic CRC (mCRC) not candidates for intensive chemotherapy, did not meet its primary point in terms of progression-free survival (PFS). But interestingly, statistical differences for PFS were found in the subgroup analyses for males treated with the experimental treatment in comparison with those treated with capecitabine and bevacizumab. A subgroup analysis of the VALENTINO trial, a multicentre, randomized, phase II trial, investigating two panitumumab-based maintenance strategies following first-line panitumumab plus FOLFOX in RAS wild-type mCRC patients, showed no significant differences in PFS, overall survival, overall response rate, or clinical benefit rate according to sex, but a significantly higher rate of grade 3-4 toxicity in females, compared to males. Similarly, in the phase III trials TRIBE and TRIBE2 investigating first-line FOLFOXIRI/bevacizumab or a doublet (FOLFIRI or FOLFOX)/bevacizumab, no statistical differences were reported depending on age or sex, although an increased risk of grade 3-4 toxicity was found in elderly females. However, it must be noted that these studies aimed to explore overall differences between sexes but were not specifically designed to compare efficacy of each arm between males and females. These results suggest that patient sex should be taken into account and that conventional methods of using body surface area for dosing may be inaccurate. Therefore, therapeutic drug monitoring has been proposed as an alternative dosage of 5-FU as it might lead to therapeutic plasma levels with a maximized risk/benefit ratio. Targeted therapies The use of targeted therapies for patients with CRC has been limited to antiangiogenic agents or epidermal growth factor receptor inhibitors in combination with chemotherapy for years. The impact of sex and gender on efficacy and toxicity of these therapies might therefore be obscured by the concomitant use of chemotherapy. Regorafenib, a multi-kinase inhibitor, obtained approval for refractory mCRC in monotherapy. More frequent toxicity in terms of fatigue, anorexia, hypertension, and rash, as well as severe toxicity have been reported in females compared with males, which might lead to lesser adherence to treatment. , Only recently, the use of BRAF inhibitors without chemotherapy has proved efficient in patients with mCRC harbouring BRAF V600 mutations. Influence of sex on efficacy and toxicity with the use of BRAF targeted therapy has been reported with the use of encorafenib plus cetuximab, with or without binimetinib. Among patients with mCRC harbouring BRAF V600E mutations treated with these combinations, a trend for superior clinical benefit in females, particularly with the triplet combination but with a higher toxicity cost, was observed. Ongoing trials testing the efficacy of targeted therapies, such as KRAS G12C inhibitors or anti-human epidermal growth factor receptor 2 (HER2) therapies, should specifically address these sex-associated differences. Immunotherapy Efficacy of immunotherapy according to sex across different tumour types has been investigated with inconsistent conclusions. , Given that the use of immunotherapy in CRC is limited to mCRC presenting MSI, analyses exploring this phenomenon usually include only a small number of patients. Better overall survival has been reported in males compared with females in patients with mCRC treated with immune checkpoint inhibitors. However, this analysis included only 99 patients, 45 of whom were females and with an uncertain number of patients with MSI tumours. In an analysis exploring symptomatic and objective toxicities across multiple cancer types, including gastrointestinal tumours, and patients treated with immunotherapy, the risk of symptomatic and haematologic adverse events was higher for females receiving immunotherapy compared with males, while the risk of objective non-haematologic toxicity was similar for both groups. Concerning the mechanism of action of immunotherapy, females receiving immune checkpoint inhibitors and immune system modulators had a higher risk of symptomatic toxicity, but this association was not observed for asymptomatic toxicity. Taken together, these data support designing trials addressing specifically sex and gender, as the balance between efficacy and toxicity may be improved by the development of sex-specific dosing strategies . Fluoropyrimidines are the backbone of chemotherapy in CRC management. Dosage is based on body surface area but does not take into account sex differences. However, a substantial body of literature indicates sex-associated differences in pharmacokinetics of 5-fluorouracil (5-FU). Plasma clearance, plasma clearance per kilogram, and dose have been found to be lower in females compared with males, whereas plasma life and area under the plasma concentration–time curve might be higher in females. On the contrary, volume of distribution, volume of distribution per kilogram, and dose per kilogram do not differ significantly between sexes. These differences may have an impact on outcomes and toxicity. Toxicity associated with 5-FU has been reported to be more extensive for females than for males in terms of the number of different types of toxicity, maximum toxicity grade, and incidence of severe toxicities, including haematologic toxicities such as leukopaenia, neutropaenia, and thrombocytopaenia, and self-reported toxicity such as stomatitis, diarrhoea, nausea, vomiting, or hand-foot syndrome. , , This increased toxicity might be due to lower clearance of 5-FU leading to higher plasma levels in females compared to males, as previously reported. Concerning patient-reported outcomes, an investigation exploring the occurrence and severity of self-reported physical and psychological co-occurring symptoms in patients with stage IV CRC receiving different 5-FU-based chemotherapy schemes reported more severe worrying, lack of energy, and nausea in women. Similar data about toxicity have been reported with the use of capecitabine. In a cohort of 299 patients (163 males, 136 females) receiving capecitabine in the adjuvant setting, females had significantly higher dose-limiting toxicity than men. Incidence of all common toxicities was higher in females and required significantly more dose reductions than males, leading to statistically significant lower relative dose intensities in females. Consistent with these results, an analysis of 36 640 patients with colon cancer receiving adjuvant fluoropyrimidine-based chemotherapy in the ACCENT database showed that incidence of grade 3 or grade 4 non-haematological toxicity (nausea, vomiting, stomatitis, diarrhoea, peripheral neuropathy, and transaminitis) and haematological toxicity (neutropaenia and leukopaenia) was statistically significant and clinically relevant higher in females. Differential efficacy according to sex has only recently been studied. The SOLSTICE trial, comparing standard capecitabine and bevacizumab versus TAS102 and bevacizumab as frontline therapy for patients with metastatic CRC (mCRC) not candidates for intensive chemotherapy, did not meet its primary point in terms of progression-free survival (PFS). But interestingly, statistical differences for PFS were found in the subgroup analyses for males treated with the experimental treatment in comparison with those treated with capecitabine and bevacizumab. A subgroup analysis of the VALENTINO trial, a multicentre, randomized, phase II trial, investigating two panitumumab-based maintenance strategies following first-line panitumumab plus FOLFOX in RAS wild-type mCRC patients, showed no significant differences in PFS, overall survival, overall response rate, or clinical benefit rate according to sex, but a significantly higher rate of grade 3-4 toxicity in females, compared to males. Similarly, in the phase III trials TRIBE and TRIBE2 investigating first-line FOLFOXIRI/bevacizumab or a doublet (FOLFIRI or FOLFOX)/bevacizumab, no statistical differences were reported depending on age or sex, although an increased risk of grade 3-4 toxicity was found in elderly females. However, it must be noted that these studies aimed to explore overall differences between sexes but were not specifically designed to compare efficacy of each arm between males and females. These results suggest that patient sex should be taken into account and that conventional methods of using body surface area for dosing may be inaccurate. Therefore, therapeutic drug monitoring has been proposed as an alternative dosage of 5-FU as it might lead to therapeutic plasma levels with a maximized risk/benefit ratio. The use of targeted therapies for patients with CRC has been limited to antiangiogenic agents or epidermal growth factor receptor inhibitors in combination with chemotherapy for years. The impact of sex and gender on efficacy and toxicity of these therapies might therefore be obscured by the concomitant use of chemotherapy. Regorafenib, a multi-kinase inhibitor, obtained approval for refractory mCRC in monotherapy. More frequent toxicity in terms of fatigue, anorexia, hypertension, and rash, as well as severe toxicity have been reported in females compared with males, which might lead to lesser adherence to treatment. , Only recently, the use of BRAF inhibitors without chemotherapy has proved efficient in patients with mCRC harbouring BRAF V600 mutations. Influence of sex on efficacy and toxicity with the use of BRAF targeted therapy has been reported with the use of encorafenib plus cetuximab, with or without binimetinib. Among patients with mCRC harbouring BRAF V600E mutations treated with these combinations, a trend for superior clinical benefit in females, particularly with the triplet combination but with a higher toxicity cost, was observed. Ongoing trials testing the efficacy of targeted therapies, such as KRAS G12C inhibitors or anti-human epidermal growth factor receptor 2 (HER2) therapies, should specifically address these sex-associated differences. Efficacy of immunotherapy according to sex across different tumour types has been investigated with inconsistent conclusions. , Given that the use of immunotherapy in CRC is limited to mCRC presenting MSI, analyses exploring this phenomenon usually include only a small number of patients. Better overall survival has been reported in males compared with females in patients with mCRC treated with immune checkpoint inhibitors. However, this analysis included only 99 patients, 45 of whom were females and with an uncertain number of patients with MSI tumours. In an analysis exploring symptomatic and objective toxicities across multiple cancer types, including gastrointestinal tumours, and patients treated with immunotherapy, the risk of symptomatic and haematologic adverse events was higher for females receiving immunotherapy compared with males, while the risk of objective non-haematologic toxicity was similar for both groups. Concerning the mechanism of action of immunotherapy, females receiving immune checkpoint inhibitors and immune system modulators had a higher risk of symptomatic toxicity, but this association was not observed for asymptomatic toxicity. Taken together, these data support designing trials addressing specifically sex and gender, as the balance between efficacy and toxicity may be improved by the development of sex-specific dosing strategies . The consequence of failing to include sex-based differences in study design and analyses has repeatedly led to ‘one-drug’ treatment regimens for both males and females.The impact of gender on health outcomes, which are different from those influenced by biological sex, is even more unknown, since gender variables are seldom included in patients’ medical histories. In the literature, these terms are often used interchangeably, but the results of the investigations that have been achieved so far exploring the sexual differences are mainly focused on sex indeed. This is due to lack of information about gender, that, because of its nature, must be specifically obtained by directly questioning the patient. In this sense, professionals should be trained in the importance of data collection about gender identity as a first step to be able to move forward in this field. Precision oncology is not limited to exploring molecular biomarkers and it requires deeper understanding of biological sex and gender differences. If the long-term goal of personalizing treatments for patients with CRC is effective treatment for all individuals, then the sex and gender must be accounted for. Despite the calls from the NIH, FDA, industry, and advocacy, the path ahead is long. Since basic science and translational research serves as the cornerstone for clinical research and medical decision making, sex disparities in physiology and pathophysiology cannot be neglected. In our perspective, barriers to enrolment of females in clinical trials in oncology should be addressed through partnership with all stakeholders, including patients, investigators, referring clinicians, health care systems, and social community. Interventional clinical trials with the focus on investigating sex-specific differences in efficacy and toxicity (including both objective toxicity and measurement of patient-reported outcomes) as a primary endpoint and the evaluation of specific dosing regimens according to sex are needed to improve outcomes. In endorsing research on how biological sex and gender influence CRC, we have the opportunity for greater precision, as it might usher in the development of sex-specific therapeutics with greater efficacy and safer toxicity profile for our patients.
Single-arm trials supporting the approval of anticancer medicinal products in the European Union: contextualization of trial results and observed clinical benefit	3a2f59cc-5276-4c06-a7d5-f6939c94aa89	10163162	Internal Medicine[mh]	Randomized controlled trials (RCTs) are referred to as the ‘gold standard’ in testing medicinal products. These trials have several advantages over clinical trials with other designs due to their design features. For example, randomization facilitates subjects in the experimental and control groups being comparable at baseline. Randomization and blinding are useful techniques to determine whether there is a cause–effect relation between treatment and outcome. , RCTs are the preferred trials to be included in applications for marketing authorization, as laid down in Directive 2001/83/EC. In this directive, it is stated that clinical trials relevant to the indication “shall be done as ‘controlled clinical trials’ if possible, randomised; any other design shall be justified”. Yet, it is not always possible to conduct an RCT, and, consequently, clinical trials with other designs need to be considered for registrational purposes. The latter includes the use of single-arm trials (SATs). Tenhunen et al . identified that, between 2010 and 2019, the European Commission (EC) approved 22 medicinal products for the treatment of solid tumors or hematological malignancies on the basis of SAT results. Many of the medicinal products included in their study received ‘conditional marketing authorization’ (CMA). This type of approval was introduced in the past to address an unmet medical need, and is based on less complete data than are usually required for standard approval. It should be mentioned, however, that SATs can also support standard approvals—albeit less common. Examples are the approvals of engineered autologous T-cell immunotherapies. , However, demonstrating that an investigational medicinal product provides clinical benefit can be challenging when it is tested solely in an SAT. Trials like these are associated with different forms of bias, including selection bias. , Besides, surrogate endpoints such as objective response rate (ORR) are commonly used in SATs, at least when focusing on cancer research. , ORR is not a direct measure of clinical benefit. Yet, it is a measure of (antitumor) activity, as spontaneous regression occurs infrequently in cancer. Some guidance exists on the use of SATs for regulatory purposes. It is stated in the “Guideline on the clinical evaluation of anticancer medicinal products” of the European Medicines Agency (EMA) that resorting to a non-randomized design should be justified by, among others, a large treatment effect on ORR and duration of response (DoR), effects that will likely translate into clinical benefit. Moreover, in the same guideline, it is stated that contextualization of results is an important topic for SATs, particularly for less evident cases. Indirect comparisons with available therapies are often made for these purposes. , While it is not the task of regulatory agencies to ensure comparative efficacy, there is a general need to ensure that new medicinal products are not worse—in terms of efficacy and/or safety—than standard of care. Importantly, the aspects described above, such as the size and durability of the treatment effect and context, will help to determine the clinical relevance of trial results. The aim of this study was to provide details on how clinical benefit of anticancer medicinal products tested in SATs was determined, including the methods used to contextualize the trial results. In addition, we were interested in how many of the authorized medicinal products based on SATs showed ‘substantial’ benefit. We started with investigating whether a threshold for the relevant treatment effect was (pre)specified in the pivotal trials—for example, in a power calculation. Subsequently, we determined if applicants submitted additional evidence to contextualize the SAT results. Finally, by limiting this study to medicinal products for the treatment of solid tumors, we evaluated the magnitude of benefit of the medicinal products included in our study via a validated tool, the European Society for Medical Oncology (ESMO)-Magnitude of Clinical Benefit Scale (MCBS). Medicinal products An overview of all human medicines that were granted approval by the EC was retrieved from the EMA database ( https://www.ema.europa.eu/en/medicines ). Products were identified on the basis of their Anatomic Therapeutic Chemical (ATC) codes, that is, L01-04 for antineoplastic and immunomodulating agents. We focused on medicinal products for the treatment of solid tumors authorized between 2012 and 2021—a 10-year period. The inclusion criterion for our analysis was initial approvals based on an SAT(s). Approvals based on RCTs were excluded. Approvals of generic and biosimilar products were also excluded. Data sources The main data source was the European public assessment reports (EPARs). These reports were obtained from the EMA database ( https://www.ema.europa.eu/en/medicines ). EPARs contain information on the scientific evaluation conducted by the Committee for Medicinal Products for Human Use (CHMP)—a committee of the EMA. The scientific evaluation forms the basis for the EC decision on approval. Another data source was published literature on pivotal clinical trials. Relevant publications were identified via PubMed and/or ClinicalTrials.gov . Data collection Data were retrieved from EPARs and/or scientific publications. We focused on pivotal trials, meaning that clinical pharmacology and dose-finding studies were not included. We collected the following information on the pivotal trials: the study design, dosing regimen, study population, planned sample size, statistical methods, primary/secondary endpoints, clinical outcomes, and type of authorization. It was also determined whether applicants made additional efforts to contextualize the results of the SAT(s), i.e. the use of external evidence to facilitate the interpretation of trial results. This concerned analyses (e.g. within-patient analysis) and/or evidence such as publications and additional studies that were included in the EPAR as supportive evidence. In addition to EPARs, scientific publications, including publicly available protocols that were supplementary to these publications, were used to complement information on the statistical methods. Determining clinical benefit The ESMO created the ESMO-MCBS, a validated tool to evaluate the magnitude of clinical benefit. The ESMO-MCBS scores already assigned to clinical trials (i.e. ESMO publications or EMSO-MCBS scorecards) were identified. The remaining SATs included in our analysis were assigned an ESMO-MCBS score independently by two researchers (VSB and JM). This was done according to EMSO instructions. Scientific publications were used for this purpose. In case a CMA was converted to standard marketing authorization (SMA) at the time of data analysis, an ESMO-MCBS score was assigned to the confirmatory trial. For non-curative therapies, ESMO-MCBS scores ≥4 represent substantial benefit. An overview of all human medicines that were granted approval by the EC was retrieved from the EMA database ( https://www.ema.europa.eu/en/medicines ). Products were identified on the basis of their Anatomic Therapeutic Chemical (ATC) codes, that is, L01-04 for antineoplastic and immunomodulating agents. We focused on medicinal products for the treatment of solid tumors authorized between 2012 and 2021—a 10-year period. The inclusion criterion for our analysis was initial approvals based on an SAT(s). Approvals based on RCTs were excluded. Approvals of generic and biosimilar products were also excluded. The main data source was the European public assessment reports (EPARs). These reports were obtained from the EMA database ( https://www.ema.europa.eu/en/medicines ). EPARs contain information on the scientific evaluation conducted by the Committee for Medicinal Products for Human Use (CHMP)—a committee of the EMA. The scientific evaluation forms the basis for the EC decision on approval. Another data source was published literature on pivotal clinical trials. Relevant publications were identified via PubMed and/or ClinicalTrials.gov . Data were retrieved from EPARs and/or scientific publications. We focused on pivotal trials, meaning that clinical pharmacology and dose-finding studies were not included. We collected the following information on the pivotal trials: the study design, dosing regimen, study population, planned sample size, statistical methods, primary/secondary endpoints, clinical outcomes, and type of authorization. It was also determined whether applicants made additional efforts to contextualize the results of the SAT(s), i.e. the use of external evidence to facilitate the interpretation of trial results. This concerned analyses (e.g. within-patient analysis) and/or evidence such as publications and additional studies that were included in the EPAR as supportive evidence. In addition to EPARs, scientific publications, including publicly available protocols that were supplementary to these publications, were used to complement information on the statistical methods. The ESMO created the ESMO-MCBS, a validated tool to evaluate the magnitude of clinical benefit. The ESMO-MCBS scores already assigned to clinical trials (i.e. ESMO publications or EMSO-MCBS scorecards) were identified. The remaining SATs included in our analysis were assigned an ESMO-MCBS score independently by two researchers (VSB and JM). This was done according to EMSO instructions. Scientific publications were used for this purpose. In case a CMA was converted to standard marketing authorization (SMA) at the time of data analysis, an ESMO-MCBS score was assigned to the confirmatory trial. For non-curative therapies, ESMO-MCBS scores ≥4 represent substantial benefit. Approval of medicinal products for the treatment of solid tumors A total of 731 medicinal products received EC approval between 2012 and 2021. Of these, 66 (9.0%) were granted approval for the treatment of solid tumors—excluding generics or biosimilars ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). Over the recent years, the proportion of approvals for solid tumors based on SATs increased compared to prior years . In total, 18 (2.5%) medicinal products were approved based on 21 SATs ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). , available at https://doi.org/10.1016/j.esmoop.2023.101209 , shows the intended patient populations for which the medicinal products were approved. Half of the medicinal products were approved (also) for the treatment of advanced non-small-cell lung cancer (NSCLC). The approvals of alectinib, avapritinib, and crizotinib were on the basis of results from an SAT(s) with top-line results from an RCT—albeit not always in a similar treatment setting (e.g. different line of therapy). However, as the SATs remained the pivotal trial(s) supporting these applications, the three products were retained in our analyses. All 18 medicinal products approved based on an SAT(s) were granted CMA. At the time of data analysis, eight CMAs were converted to SMAs ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). For one of the CMAs, i.e. rucaparib, the benefit–risk balance was no longer considered favorable by the CHMP based on the confirmatory trial. The marketing authorization holder (MAH) requested to remove the indication. Single-arm trials and thresholds for clinically relevant treatment effect Most approvals were supported by one pivotal trial. The approvals of alectinib, osimertinib, and rucaparib were supported by two SATs. For the approvals of entrectinib and larotrectinib, integrated analyses by pooling data across clinical trials were used for the evaluation of efficacy (three trials each). For all trials or integrated analyses, the primary endpoint was ORR ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). For the majority of clinical trials or integrated analyses [15 out of 21 (71.4%)], a clinically relevant treatment effect was (pre)specified and most often an accompanying sample size calculation was provided . The test for a relevant effect was often defined as the lower bound of the 95% confidence interval (CI) for ORR exceeding a (predefined) value, which is equivalent to testing a null hypothesis corresponding to that value. Protocols confirmed the results, which were publicly available (i.e. supplementary to publication) for all SATs except for those studying alectinib (both trials), avelumab, osimertinib (AURA 2), and rucaparib (CO-338-010). For the trials investigating entrectinib, larotrectinib, pemigatinib, and selpercatinib, a clinically relevant lower boundary of the 95% CI for ORR was defined, but the null and alternative hypotheses were not explicitly mentioned in the EPARs/publications. For the trials testing ceritinib, crizotinib, lorlatinib, and rucaparib, no power calculations were carried out based on the information presented in the EPARs and/or publications. Regarding trial CO-338-017, one of the SATs testing rucaparib, some sample size assumptions were made for subgroup allocation (part 1 and 2) and comparison (part 1) of the trial, but no calculations were made based on expected treatment effects. For 10 out of 21 trials (47.6%), each testing a different medicinal product, justification for the threshold to the statistical test could be extracted from EPARs/publications/protocols . Mostly, the treatment effect of available therapies was used as a benchmark ( n = 5). Other justifications were ‘consistent with the response rates seen with approved targeted therapies in genetically defined patient populations who have progressed on prior therapies’ ( n = 2), ‘limited treatment options’ ( n = 1), and ‘absence of literature documenting treatment outcomes for second-line patients’ ( n = 1). Pralsetinib and selpercatinib were tested in trials that included patients with RET fusion-positive NSCLC who previously received platinum-based chemotherapy. The specified clinically relevant lower bound of the 95% CI for ORR was different between the two trials, namely 23% and 30%, respectively . Larotrectinib and entrectinib were tested in clinical trials that included patients with NTRK gene fusion-positive tumors. For both applications, the lower bound of the 95% CI for ORR was 30% for the integrated analysis across clinical trials . Contextualization The type and amount of information that was included for contextualization purposes varied between the applications for marketing authorization for the 18 medicinal products. At least 12 out of 18 applications (71.4%) included some additional information for contextualization purposes ( , , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). Six out of 18 applications included supportive studies ( , , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). One of these supportive studies concerned a bibliographic reference, namely the Dermatologic Cooperative Oncology Group (DeCOG) study. Other supportive studies were of a retrospective nature, and included real-world data from various sources. From the supportive studies included in the applications of trastuzumab deruxtecan and entrectinib, i.e. the Unicancer study and WO40977, respectively, matched populations were generated. In the latter study, a comparative analysis with a matched crizotinib arm derived from real-world data was conducted. Evaluating benefit and , available at https://doi.org/10.1016/j.esmoop.2023.101209 , show the ESMO-MCBS scores for the pivotal SATs ( n = 21), either assigned by us or already published by ESMO. For all the SATs included in our study, three SATs were assigned an ESMO-MCBS score of ‘4’. Fifteen SATs were assigned an EMSO-MCBS score of ‘3’, two SATs were assigned an ESMO-MCBS score of ‘2’, and one SAT was assigned an ESMO-MCBS score of ‘1’. ESMO-MCBS scores of ‘4’ were assigned as a result of the score upgrades for quality of life (QoL), meaning the investigators reported improvements in QoL. Five out of eight CMAs were converted to SMA based on an RCT, i.e. reaching a comprehensive level of evidence. Of these RCTs, four were assigned an ESMO-MCBS score of ‘4’ and one was assigned a score of ‘2’ ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). A total of 731 medicinal products received EC approval between 2012 and 2021. Of these, 66 (9.0%) were granted approval for the treatment of solid tumors—excluding generics or biosimilars ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). Over the recent years, the proportion of approvals for solid tumors based on SATs increased compared to prior years . In total, 18 (2.5%) medicinal products were approved based on 21 SATs ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). , available at https://doi.org/10.1016/j.esmoop.2023.101209 , shows the intended patient populations for which the medicinal products were approved. Half of the medicinal products were approved (also) for the treatment of advanced non-small-cell lung cancer (NSCLC). The approvals of alectinib, avapritinib, and crizotinib were on the basis of results from an SAT(s) with top-line results from an RCT—albeit not always in a similar treatment setting (e.g. different line of therapy). However, as the SATs remained the pivotal trial(s) supporting these applications, the three products were retained in our analyses. All 18 medicinal products approved based on an SAT(s) were granted CMA. At the time of data analysis, eight CMAs were converted to SMAs ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). For one of the CMAs, i.e. rucaparib, the benefit–risk balance was no longer considered favorable by the CHMP based on the confirmatory trial. The marketing authorization holder (MAH) requested to remove the indication. Most approvals were supported by one pivotal trial. The approvals of alectinib, osimertinib, and rucaparib were supported by two SATs. For the approvals of entrectinib and larotrectinib, integrated analyses by pooling data across clinical trials were used for the evaluation of efficacy (three trials each). For all trials or integrated analyses, the primary endpoint was ORR ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). For the majority of clinical trials or integrated analyses [15 out of 21 (71.4%)], a clinically relevant treatment effect was (pre)specified and most often an accompanying sample size calculation was provided . The test for a relevant effect was often defined as the lower bound of the 95% confidence interval (CI) for ORR exceeding a (predefined) value, which is equivalent to testing a null hypothesis corresponding to that value. Protocols confirmed the results, which were publicly available (i.e. supplementary to publication) for all SATs except for those studying alectinib (both trials), avelumab, osimertinib (AURA 2), and rucaparib (CO-338-010). For the trials investigating entrectinib, larotrectinib, pemigatinib, and selpercatinib, a clinically relevant lower boundary of the 95% CI for ORR was defined, but the null and alternative hypotheses were not explicitly mentioned in the EPARs/publications. For the trials testing ceritinib, crizotinib, lorlatinib, and rucaparib, no power calculations were carried out based on the information presented in the EPARs and/or publications. Regarding trial CO-338-017, one of the SATs testing rucaparib, some sample size assumptions were made for subgroup allocation (part 1 and 2) and comparison (part 1) of the trial, but no calculations were made based on expected treatment effects. For 10 out of 21 trials (47.6%), each testing a different medicinal product, justification for the threshold to the statistical test could be extracted from EPARs/publications/protocols . Mostly, the treatment effect of available therapies was used as a benchmark ( n = 5). Other justifications were ‘consistent with the response rates seen with approved targeted therapies in genetically defined patient populations who have progressed on prior therapies’ ( n = 2), ‘limited treatment options’ ( n = 1), and ‘absence of literature documenting treatment outcomes for second-line patients’ ( n = 1). Pralsetinib and selpercatinib were tested in trials that included patients with RET fusion-positive NSCLC who previously received platinum-based chemotherapy. The specified clinically relevant lower bound of the 95% CI for ORR was different between the two trials, namely 23% and 30%, respectively . Larotrectinib and entrectinib were tested in clinical trials that included patients with NTRK gene fusion-positive tumors. For both applications, the lower bound of the 95% CI for ORR was 30% for the integrated analysis across clinical trials . The type and amount of information that was included for contextualization purposes varied between the applications for marketing authorization for the 18 medicinal products. At least 12 out of 18 applications (71.4%) included some additional information for contextualization purposes ( , , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). Six out of 18 applications included supportive studies ( , , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). One of these supportive studies concerned a bibliographic reference, namely the Dermatologic Cooperative Oncology Group (DeCOG) study. Other supportive studies were of a retrospective nature, and included real-world data from various sources. From the supportive studies included in the applications of trastuzumab deruxtecan and entrectinib, i.e. the Unicancer study and WO40977, respectively, matched populations were generated. In the latter study, a comparative analysis with a matched crizotinib arm derived from real-world data was conducted. and , available at https://doi.org/10.1016/j.esmoop.2023.101209 , show the ESMO-MCBS scores for the pivotal SATs ( n = 21), either assigned by us or already published by ESMO. For all the SATs included in our study, three SATs were assigned an ESMO-MCBS score of ‘4’. Fifteen SATs were assigned an EMSO-MCBS score of ‘3’, two SATs were assigned an ESMO-MCBS score of ‘2’, and one SAT was assigned an ESMO-MCBS score of ‘1’. ESMO-MCBS scores of ‘4’ were assigned as a result of the score upgrades for quality of life (QoL), meaning the investigators reported improvements in QoL. Five out of eight CMAs were converted to SMA based on an RCT, i.e. reaching a comprehensive level of evidence. Of these RCTs, four were assigned an ESMO-MCBS score of ‘4’ and one was assigned a score of ‘2’ ( , available at https://doi.org/10.1016/j.esmoop.2023.101209 ). In specific situations, medicinal products may receive (expedited) regulatory approval on the basis of results from SATs. In this study, we analyzed pivotal SAT-based applications for anticancer medicinal products in the European Union between 2012 and 2021. In this period, 18 medicinal products for the treatment of solid tumors received an approval based on 21 SATs. At least 12 out of 18 applications included additional information to contextualize the results from the pivotal trials, which included supportive studies, external evidence, information on response to prior therapy, and/or a within-patient comparison. Of all the SATs or integrated analyses supporting the 18 EC approvals, three were assigned an ESMO-MCBS score of ‘4’, that is, a score indicating substantial benefit. SATs are generally initiated to determine whether an investigational product has sufficient activity to continue development. , Often statistical testing is used to determine whether the treatment effect is above a prespecified threshold, which is reflected in whether the null hypothesis related to the threshold is rejected. Our results indicate that a justification for this threshold was not always reported in the EPAR (or scientific publication). Tenhunen et al . reported that the threshold for ‘success’ in pivotal SATs is relatively uniform—20% ORR—and often not scientifically justified. Our study does not confirm their results, as thresholds varied—ranging from 10% to 50%. This might, however, be explained by the partial differences in datasets. The threshold for success is often based on historical data or clinical judgment, which reflects ORRs by available treatment or standard of care. However, determining this threshold can be challenging. For instance, historical data can be inconsistent with regard to the observed ORRs. Studies with doxorubicin plus ifosfamide in soft tissue sarcoma showed varied ORRs (i.e. 16%-35%). Also, historical data might be absent or derived from studies that differ in, but not limited to, design or study population in comparison to the SAT. The latter being particularly relevant for biomarker-driven SATs, which concerns the majority of SATs included in our study. Overall, it is important to select an appropriate threshold before conducting an SAT, but even more to provide argumentation why having a lower bound of the 95% CI above this threshold constitutes a clinically relevant outcome. Our results show that the ORR to be ruled out at a particular significance level is not always ambitious. For example, thresholds based on historical data were sometimes lower than thresholds in absence of treatment options. Also, similar historical data sometimes led to different thresholds. Another point to mention is that the ORR used for sample size/power calculations, i.e. the effect under the alternative hypothesis, is rarely justified to be clinically relevant or corresponding to an effect one would not want to miss (data not shown), the latter, for instance, in the context of a go/no-go decision for proceeding with drug development program. Simply rejecting the null hypothesis may not be sufficient for regulatory decision making. As already highlighted a few decades ago, the meaningfulness of ORR depends on whether this translates into ‘true’ benefit (e.g. improvement in survival). Oxnard et al . showed that an ORR statistically exceeding 30% (or higher) is associated with regulatory approval, at least for monotherapies tested in SATs. However, not only ORR but also DoR will be important for regulatory decision making. For example, the CHMP was of the opinion that the activity of pralsetinib, indicated by a high percentage of durable responses in the pivotal trial, would translate into clinical meaningful benefit. The observed ORR of entrectinib shown in the integrated analysis was below the assumed ORR used for the sample size calculation. However, the observed ORR, in combination with DoR, was considered of clinical relevance by the CHMP. In contrast, the ORR and DoR shown by retifanlimab in the pivotal SAT were not considered clinically relevant by the CHMP. In fact, the criterion for ‘success’ was not met in this SAT—i.e. ruling out an ORR of 13%—and the applicant withdrew the application for marketing authorization. If it is justified to use an SAT for regulatory purposes, it will be key to motivate which effect would constitute an (minimal) important effect from a clinical point of view, not merely ruling out a, sometimes unimpressive, historical ORR. During the approval process, context may be sought via indirect comparisons with (well-)documented outcomes for clinical trials testing available therapies. This is also relevant considering that new data may have become available after initiation of the SAT. We demonstrate that applications frequently include information for contextualization purposes, including results from supportive studies. There are, however, limitations associated with cross-trial comparisons, which necessitate caution when interpreting these results. For example, differences between study populations may lead to inappropriate comparisons. One approach to (partly) overcome these limitations is to use patient-level data to generate a matched external control. Interestingly, a recent study carried out by Schröder et al . demonstrated that external controls generated from electronic health record-derived databases were successful in replicating a control arm from an RCT in metastatic colorectal cancer. However, matched comparisons with external controls are rare—at least in our dataset. Only two comparative matched analyses with standard of care were carried out. External controls, however, cannot be corrected for confounders that are unknown or unmeasured. There is some regulatory guidance available to reduce potential bias with external controls. , However, after addressing all the limitations as much as possible, the issue remains that, if there is a high chance for residual bias, the outcome in an SAT has to be convincing to compensate for the potential bias. Importantly, the quality of data will likely determine the extent to which external controls can be used for regulatory decision making. Pignatti et al . highlighted that the definition of clinical value is different between stakeholders, which may lead to different conclusions. While the CHMP concluded that the benefit of the medicinal products included in our analysis was clinically relevant, stakeholders other than regulators might appreciate benefit differently. For instance, the ESMO considers benefit as ‘living longer and/or living better’, which resonates in the ESMO-MCBS form for SATs. , This is evident by our results, as the benefit of the majority of products was ‘modest’ on the basis of the ESMO-MCBS scores. Tibau et al. stated that large treatment effects in combination with an improvement in QoL (or data from post-marketing studies) are needed for SATs to be assigned a high ESMO-MCBS score. However, QoL is not always a secondary endpoint in clinical trials, and one of the shortcomings of the ESMO-MCBS is that it does not take into account delayed publications or publication bias for QoL. Besides, the CHMP repeatedly stated in assessment reports that no firm conclusion can be drawn from QoL data generated by SATs. , , , Thus, QoL is of lesser importance in regulatory decision making on SATs. There are other tools to evaluate the benefit of approved anticancer medicinal products. For instance, a committee of the Dutch Society of Medical Oncology created the PASKWIL criteria for non-randomized trials, for which the ESMO-MCBS was used as a basis. In comparison to the ESMO-MCBS, QoL and safety are not incorporated in this instrument, and benefit is based on predefined ORR and DoR thresholds. Other criteria are that the medicinal product is authorized by the EC, the disease is rare, the patient population is adequately selected, and there is a biological rationale for therapy. As tools are created on a national level that do not completely align with the EMSO-MCBS, there might be a need to fine-tune what can be considered benefit on an European level. Consistency among tools may warrant further discussion among stakeholders so as to prevent potential inequality in care. All medicinal products included in our study received a CMA. When the MAH intends to fulfill the specific obligation(s) associated with the CMA, the benefit–risk balance will be re-assessed on a more complete dataset, preferably results from an RCT. However, Tenhunen et al . showed that post-authorization measures associated with CMAs are not always to submit results from an RCT. Of course, the level of evidence to be generated in the post-marketing setting depends on, amongst others, feasibility to conduct large trials. Recently, Fashoyin-Aje et al . informed that a ‘comprehensive strategy’ for confirmatory trials is needed, which focusses on the so-called on-ramp (e.g. trial design, patient population, etc.) and off-ramp considerations (i.e. verify clinical benefit). The authors highlight that, for accelerated approvals, efforts should be made to timely and adequately address remaining uncertainties regarding the benefit–risk balance. Similarly, Bloem et al . highlight that RCTs should be ongoing when a CMA is granted, ensuring rapid access to a more complete dataset. Important to mention is that re-assessment of the ESMO-MCBS score is possible when results from confirmatory trials are published. This may lead to an improvement in ESMO-MCBS score—as also seen in our study. Furthermore, extended follow-up for the SATs themselves may also improve the EMSO-MCBS score. For example, we previously assigned an ESMO-MCBS score of ‘2’ to the SAT investigating cemiplimab. However, our current research shows a score of ‘4’ (from an ESMO-MCBS scorecard), which is based on a more recent publication. While this study provides insights into the contextualization process of SAT results, it is limited to SATs supporting initial approvals. While extensions of therapeutic indication(s) can in principle be based on SATs, this is rare and such applications are not included in our analysis. For an extension of indication, there is already existing knowledge on the benefits and risks of the concerned medicinal product due to the initial marketing authorization, which might impact decision making. In addition, we did not include withdrawals of SAT-based applications, as these numbers ( n = 4) were too limited for a meaningful analysis. It can also be considered a limitation that we restricted our research to publicly available documents. However, we assume that all information relevant to the benefit–risk assessment is incorporated in the EPARs, as it is a reflection of the core documents included in an application, as well as in literature and/or protocols, the latter being available for most SATs. Another limitation is that confirmatory trials were ongoing for some of the products included in our study. The ESMO-MCBS score could, therefore, not yet be re-assessed for these products. Finally, we focused only on SAT-based applications submitted to the EMA. It would be interesting to compare regulatory decision making between agencies, such as the Food and Drug Administration and EMA. In conclusion, we found that 18 medicinal products were approved for the treatment of solid tumors based on one or more SAT(s). For the majority of clinical trials or integrated analyses supporting these approvals, a threshold to be ruled out was (pre)specified, and most often accompanied by a sample size calculation based on an assumed ORR. However, a justification for the threshold and the assumed ORR could not be identified for all cases. The majority of applications included additional information for contextualization purposes. Determining the benefit–risk balance of medicinal products tested in SATs is challenging and benefit can be appreciated differently by various stakeholders. The clinical relevance of the treatment effects shown by medicinal products tested in SATs is dependent on the activity, its durability and context, especially if other therapies are available that provide benefit. As general recommendations, prespecifying and motivating a clinically relevant effect and aligning the sample size to that effect is of importance for regulatory decision making. External controls may facilitate in the contextualization process, but the limitations associated with such comparisons must be (adequately) addressed. Preferably, such comparisons should be preplanned. It is of relevance that information on these aspects is presented in the EPAR, as this provides transparency on regulatory decision making toward stakeholders. Finally, it is considered of value to further discuss among stakeholders what can be considered clinical benefit in the context of SATs and thus when approval on the basis of lower levels of evidence is justified. This is considered of importance, as SATs will likely continue to form the basis of authorization of part of the new medicinal products.
A giant orbital solitary fibrous tumor treated by surgical excision: a case report and literature review	0622aab7-7470-4863-aafe-7ac56c5bb344	10163734	Anatomy[mh]	Solitary fibrous tumors (SFTs), rare spindle cell tumors, were first described by Klemperer and Rabin in 1931 . They usually occur in the pleura, pericardium, respiratory tract, peritoneum or mesentery, orbit, breasts, other soft tissues, and visceral organs . Since Westra et al. reported the first orbital SFT in 1994, increasing orbital involvement has been reported. Although most cases typically present as a slow-growing orbital mass and behave in a benign fashion, a few exhibit malignant behavior, such as recurrence and local invasion . This report details a case of a giant orbit SFT compressing and engulfing the eyeball. To the best of our knowledge, this is the largest benign orbital SFT ever reported. In June 2018, a 57-year-old woman presented with a massive, painless mass in the right orbit which had developed over 19 years. As the tumor grew, vision in the right eye gradually deteriorated until it was completely lost approximately 15 years before presentation. Due to budgetary constraints, the patient avoided seeking appropriate medical care. Upon ophthalmological examination, she had a best-corrected visual acuity of no light perception and 20/20 of the right and left eyes, respectively. In the right orbit, a large rubbery mass stretched horizontally from the bridge of the nose to the lateral canthus, and vertically from the eyebrow to the center of the cheek. The upper eyelid covering the tumor was pushed forward due to the large size of the mass. The vessels of the eyelids were dilated and tortuous (Fig. A). The palpebral fissure was elongated and widened; the eyeball, covered with soft tissue, was displaced anteromedially (Fig. B). No abnormalities were observed in the left eye. There was no associated lymphadenopathy and a systemic examination revealed no abnormalities. Orbital CT revealed a huge mass in the right orbit measuring 10.6 × 9.5 × 11 cm. After administration of contrast material, inhomogeneous enhancement and vascular-like enhancing structures were observed. The mass was closely related to the internal rectus muscle. The eyeball appeared totally engulfed and compressed by the large mass and was displaced anteromedially. In the bone structures, no evident erosion or extension damage was found; however, the temporal orbital wall was observed to be slightly thinner than anticipated (Fig. C-1D). Orbital cavernous hemangioma was suspected. The patient underwent lid-sparing orbital exenteration under general anesthesia, a procedure that involved removal of the tumor, globe, orbital contents, and most of the upper eyelid. Using a preserved part of the normal eyelid skin, the orbital cavity was covered and sutured with a skin incision margin. On gross pathological examination, the orbital lesion consisted of a 15 × 13 × 9 cm well-circumscribed solid tumor covered with a 16 × 14 cm flap (Fig. A). The cut surface was composed of gray-white or gray-red soft tissue with foci of hemorrhage identified in some areas. Nerve tissue of approximately 2.5 cm in length and 0.2 cm in diameter was observed in the adipose tissue next to the tumor. On microscopic examination, H&E staining showed that the tumor tissue constituted spindle-shaped cells arranged in bundles or irregular shapes. Cell nuclei were oval or spindle-shaped. No necrosis, mitoses, or nuclear pleomorphisms were present. (Fig. B–2C). IHC testing demonstrated strong and diffuse spindle cells with antibodies against CD34, signal transducer and activator of transcription 6 (STAT6) (Fig. D-2E), B-cell lymphoma 2 (BCL-2), and progesterone receptors (PR). Approximately 2% of these cells were reactive to Ki-67. Staining for antibodies against P53, glial fibrillary acidic protein, and S-100 was negative. These findings were compatible with a diagnosis of a benign SFT. The patient had good cosmetic results and exhibited no symptoms of recurrence during the 4-year follow-up. Patients with orbital SFTs are predominantly middle-aged adults . There was no apparent sex predilection for SFT. Any orbital space, including the intraconal and extraconal spaces of the orbit, can be affected by orbital SFTs. Lesions occurring in other tissues, such as the conjunctivae, lacrimal gland fossa, lacrimal sacs, eyelid, and pigmented outer layer of the pars plana of the ciliary body, have also been reported . Symptoms and signs are related to tumor size and location. Patients with orbital SFTs often present with proptosis, eyelid swelling, blepharoptosis, diplopia, ocular motility restriction, and a slow-growing painless palpable mass in the periocular area . Headaches and epiphora are rare symptoms. Visual acuity is generally normal or mildly impaired; however, if the optic nerve is involved, patients may develop significant reduction in visual acuity and even blindness of the affected eye. Fundus examination is usually unremarkable, but some patients show dilated vessels, optic disc and macular edema, and optic nerve atrophy due to elevated intraorbital and intraocular pressure . In our case, the patient presented with a painless mass for more than 10 years without medical intervention, which resulted in significant disfiguration and blindness of the right eye. The reported radiological features of SFT are non-specific. Orbital SFTs are seen as well-circumscribed soft tissue tumors with moderate to intense enhancement on CT images, which are attributed to the high vascularity within the tumor . Although extremely rare, bony erosion should prompt suspicion of a malignant tumor. On MRI, tumors have been demonstrated to have T1-weighted signal isointensity and T2-weighted isointensity to hypointensity, reflecting differences in the amounts of cellular components, collagen, and fibroblasts among different tumors . They may be difficult to distinguish from tumors with high blood flow, such as fibrous histiocytomas, neurofibromas, hemangiomas, and schwannomas on CT and MRI. However, these imaging modalities may help with localization, tumor sizing, planning of surgical intervention, and postoperative monitoring. Complete en bloc excision is required to reduce the risk of recurrence . In this case, orbital CT revealed that the giant tumor occupied virtually the entire orbit and extended beyond the orbit, entirely disrupting the normal structure and function of the eyeball and the optic nerve, ultimately leading to blindness. Therefore, an orbital exenteration was required to thoroughly remove the tumor. Benigh SFTs typically exhibit low mitotic activity and lack nuclear pleomorphism and/or necrosis . Histomorphological features of malignancy include increased mitotic activity (≥ 4/10 HPFs or > 2 mitoses/2 mm2), nuclear pleomorphism, tumor necrosis, increased tumor size (≥ 5 cm), and infiltrative borders . Microscopic features alone are insufficient to confirm the diagnosis of SFT and further IHC analysis must be conducted. SFT cells typically stain positive for specific markers, such as CD34 and STAT6, and variably positive for vimentin, S-100 protein, progesterone receptors (PR), P53, BCL-2, and Ki-67 . CD34, an antigen expressed on endothelial cells and hematopoietic progenitor cells, stained strongly and diffusely, and it is believed to be the most diagnostic immunohistochemical biomarker for benign SFTs. CD34 negative or weakly expressed cells may be associated with malignant transformation . Nuclear STAT6 overexpression is a highly sensitive and specific biomarker for SFTs; thus, SFTs can be distinguished from other orbital fibroblastic tumors . Steroid hormone receptors and PRs are expressed in SFTs. Previous studies have reported both increased and decreased expression of PR in predicting high-risk behavior. Bongiovanni et al. reported that PR positivity is a feature of pleura SFTs, demonstrating increased proliferative activity and a propensity for recurrence, while Carretta et al. reported that lower expression of PR identifies pleura SFT with a higher risk of recurrence after surgery. Additional studies are still needed to confirm the effect of PR expression in orbital SFTs on predicting prognosis. In adult mammalian tissues, BCL-2 protein has a restricted pattern of expression which is limited to proliferating cells, stem cells, and hormone-responsive tissues. The presence of BCL-2 in this case may be related to the expression of PRs in neoplastic tissue . According to Sun et al., Ki-67, a protein linked to ribosomal RNA synthesis and cell proliferation, can be used as a prognostic marker of SFTs and is diagnostically relevant for the assessment of malignant SFTs . In benign SFTs, the Ki-67 index frequently reacts positively in 0–2% of spindle-cell nuclei; this proportion can increase to 40% in malignant tumors. A tumor suppressor gene, p53, plays a critical role in regulating cell proliferation. p53 is strongly expressed in SFTs with fatal outcomes, such as clinical recurrence, local invasion, and metastasis . In this case, CD34 and STAT6 immunoreactivity supported the diagnosis of SFT. Low expression of both p53 and Ki-67 was consistent with the tumor's histological features. Therefore, the features of microscopic features and IHC analysis in this case support the diagnosis of benign SFT. The clinical behavior of SFTs is variable . The majority of SFTs are slow-growing masses that pursue a nonaggressive course; however, clinical and radiological features are not necessarily associated with histological signs of malignancy . Even in cases of benign tumors, aggressive clinical behaviors, such as adjacent tissue invasion , recurrence , metastasis , and malignant transformation , have been demonstrated. Malignant SFT may occur de novo or by transformation within benign or low-grade tumors . Local recurrence is usually attributed to incomplete initial resection of the tumor, which then shows a tendency to spread into the orbital bone or extra-orbital soft tissue . The giant and slow-growing orbital SFT in our case showed an indolent course of growth and did not display any aggressive behavior during the follow-up period. However, Demicco EG et al. reported a risk prediction model for SFTs incorporating patient age, tumor size, and mitotic activity to predict risk of metastasis. Low-risk patients did not acquire any metastases, whereas the intermediate-risk group had a 7% 10-year metastatic risk and the high-risk group had a 49% 5-year metastatic risk. According to the criteria above, this case can be regarded as at intermediate risk. Regular long-term follow-up is necessary, even though our case did not show any evidence of tumor recurrence after a 4-year follow-up period. Complete resection is necessary for adequate local tumor control. Other adjuvant treatments, including preoperative transarterial embolization, radiotherapy, and chemotherapy for the treatment of recurrent orbital SFTs, have been reported in individual cases . The potential benefits of adjunctive therapy should be further evaluated in clinical trials. In conclusion, we present an extremely rare case of orbital SFT which was definitively treated with surgical excision. To the best of our knowledge, this is the largest orbital SFT reported thus far. The tumor exhibited an aggressive course of eyeball compression and had typical histomorphological and IHC features indicative of a benign SFT. Therefore, careful, long-term follow-up is necessary.
Close orthopedic surgery skin incision with combination of barbed sutures and running subcuticular suturing technique for less dermal tension concentration: a finite element analysis	d7e93d55-a143-4ed4-be99-480dbfcb6489	10163751	Suturing[mh]	Keloid and hypertrophic scars (HSs) are common complications of orthopedic surgical incisions. They occur frequently at particular sites, such as the scapular area and the suprapubic region . This is due to the pressure on the skin when the suture is left in place, which can be aggravated by tension on the wound, large bites of tissue, edema, and infection . These conditions are particularly common in patients with extremity fractures, where soft tissue edema results in high tension and long duration of the surgical incision. A murine model of hypertrophic scarring suggested that mechanical forces may be primarily responsible for such scars’ generation in wounded skin, not only strongly promotes their growth . The finite element method (FEM) also clarified that stretching tension is an important condition associated with keloid growth . To avoid incision complications or “railroad tracks” with the reduction of dermal tension, surgeons may choose a buried suture technique other than the traditional interrupted vertical mattress suture. The usage of the buried continuous suture techniques has been prompted by the unpleasant appearance of visible scars on the limbs, the cost of stitch removal and dressing change care, and the inconvenience of traveling to and from the hospital for the elderly. Running subcuticular suture and intradermal buried vertical mattress suture are two common buried suture techniques invented to close wounds with reduced surgical scars . These two suture techniques avoid the uneven distribution of tension throughout the wound associated with surgical knots, and the higher tension burden placed at the knots. It remains unclear which one of these two suturing techniques applied to orthopedic wounds has the better biomechanical advantage in reducing skin tension. Some other efforts to reduce scarring are improved sutures. Given the excessive relative wound tension and the reasonable concerns of surgeons for suture failure due to tissue-suture slippage using smooth sutures, there is a natural tendency toward overcoming these concerns by over-tightening every stitch in running suture. Excessive pressure within this tissue can produce enzymatic degradation and result in a loss of wound strength and a higher incidence of wound dehiscence . An inevitable partial stitch removal due to infection or “cheese-wiring” can cause the complete strength loss throughout the wound due to the breakage of smooth sutures, which high risky leading entire wound to disintegrate. Barbed suture technology was developed to allow wound closure using a self-anchoring suture that avoids the requirement for knot tying and the loops they entail. In orthopedic surgeries, absorbable barbed sutures are widely used due to their convenience and reducing wound tension . Different from smooth sutures, these barbs aid to maintain tensile strength by continuously gripping the sutured tissue . However, the mechanical interaction between the barbed suture and the tissue as well as the biomechanical advantages compared to conventional smooth sutures still remain unclear. A few 2D and 3D finite element (FE) models have been developed to study the biomechanics of wound closure and skin tension [ , , ]. It is more reasonable to simulate the interaction between the skin tissue and the sutures with 3D finite element models, as the sutures travel through various skin tissue layers, with each layer varying greatly in mechanical parameters. Therefore, the aim of this research is to compare and evaluate the biomechanical properties for surgical wound dehiscence using numerical simulation. In the FE simulation, four realistic 3D computational wound models (two different suturing techniques with two different sutures) were built to (a) evaluate the biomechanical differences between intradermal buried vertical mattress suture and running subcuticular suture, (b) compare tissue stress reduction on the tissues’ contact surfaces to the sutures with barbed sutures. Geometrical modeling As the aim of the FE model was to evaluate the biomechanical properties, including wound tension and dehiscence characters that related to sutures and suturing techniques, the geometry should be able to reflect the necessary and general suture techniques for surgical wound closure with sufficient mechanical information in details. The skin wound included two symmetrical layered skin sections. Each skin section consisted of three layers, including epidermis, dermis and fat with the thickness of 0.98 mm, 1.0 mm and 8.9 mm, respectively [ – ]. The single section of the skin tissue was modelled in the size of 15 mm × 10 mm × 10 mm. A single stitch of running subcuticular suture with skin was modelled to represent a generalized section of a wound, so that the computational resources can be mostly utilized to refine the mesh element to study the interaction between the sutures and the skin tissue (Fig. ). The suture diameters were set to 0.2 mm, which is the ideal size of USP 3–0 suture . Two different suture techniques were modelled as shown in Fig. . The techniques of intradermal buried vertical mattress suture and running subcuticular suture were idealized with constant radius curvature with R = 4 mm. The single stitch of subcuticular suture was idealized in a U-shaped tunnel in the dermis. The single stitch of intradermal buried vertical mattress suture was modelled in a spiral tunnel. In both suture techniques, the sutures did not pass through epidermises . FE modeling Meshing The mesh of the subcuticular suture model contained 445,036 elements with the minimal element size of 0.03 mm (Fig. a). The geometry of the intradermal buried vertical mattress suture model was more complicated than that of the subcuticular suture model as more elements were required for the spiral shaped sutures. It contained 673,313 elements with the minimal element size of 0.03 mm (Fig. b). Loads and boundary conditions The mechanical deformation and contact models were implemented for skin suture modelling. As the aim of this study is to compare and evaluate the tissue tension difference between the suturing techniques and sutures at relatively small deformation, it is reasonable to use the linear elastic material model for skin and suture models. The suture process was carried out at a steady and slow motion. The dynamic forces were neglected. The deformation equation of the materials can be expressed by 1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nabla \cdot \left( {{\text{FS}}} \right)^{{\text{T}}} + {\mathbf{F}}_{{\text{V}}} = 0$$\end{document} ∇ · FS T + F V = 0 where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{F}} = {\mathbf{I}} + \nabla {\mathbf{u}}$$\end{document} F = I + ∇ u is the deformation tensor, u is the displacement vector, I is the identity tensor, S is the second Piola–Kirchhoff stress tensor and F v is the body load. The constitutive equation can be expressed as 2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{S}} = {\text{JF}}^{ - 1} \left( {{\mathbf{C}}:\varepsilon } \right){\text{F}}^{{ - {\text{T}}}}$$\end{document} S = JF - 1 C : ε F - T where J is the volumetric deformation, C is the material constitutive tensor and ε is the strain tensor. The stress can be calculated by 3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma_{ij} = C_{ijkl} \varepsilon_{kl}$$\end{document} σ ij = C ijkl ε kl The interface between sutures and skin tissues was modelled by contacting models. The mechanical property difference between standard sutures and barbed sutures was modelled using different contact friction coefficient. The friction coefficients of the standard sutures and barbed sutures were zero and 0.3, respectively . The contact pressure T n was calculated using the penalty factor and contact gap by 4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{n} = - p_{n} g_{n} + p_{0}\,if\,g_{n} < p_{0} /p_{n}$$\end{document} T n = - p n g n + p 0 i f g n < p 0 / p n where p n is the penalty factor, g n is the gap and p 0 is the pressure at zero gap. The penalty factor can be interpreted as the stiffness of a spring inserted between the contact surfaces. Certain degree of overlapping could occur while p 0 was zero. In this study, the initial pressure between the skin tissue and the sutures was estimated based on the assumption that the sutures are able to expand the suture tunnels to the suture diameters. The simplified FE models were established to model a single stitch passing the tissue. The suture pressure was applied to the cross section of 0.2 mm width, which represents the USP 3–0 suture. The displacement was calculated under different pressure. The pressure that generated the 0.1 mm displacement was 100 kPa, which was considered to be the suture initial contact pressure at zero gap. The prescribed displacement was applied to the epidermis layer to simulate the skin stretch (Fig. ). Material modeling The soft tissues are close to volumetrically incompressible . The Poisson’s ratios were close to 0.5 for the tissues. The material properties are listed in Table . As the aim of the FE model was to evaluate the biomechanical properties, including wound tension and dehiscence characters that related to sutures and suturing techniques, the geometry should be able to reflect the necessary and general suture techniques for surgical wound closure with sufficient mechanical information in details. The skin wound included two symmetrical layered skin sections. Each skin section consisted of three layers, including epidermis, dermis and fat with the thickness of 0.98 mm, 1.0 mm and 8.9 mm, respectively [ – ]. The single section of the skin tissue was modelled in the size of 15 mm × 10 mm × 10 mm. A single stitch of running subcuticular suture with skin was modelled to represent a generalized section of a wound, so that the computational resources can be mostly utilized to refine the mesh element to study the interaction between the sutures and the skin tissue (Fig. ). The suture diameters were set to 0.2 mm, which is the ideal size of USP 3–0 suture . Two different suture techniques were modelled as shown in Fig. . The techniques of intradermal buried vertical mattress suture and running subcuticular suture were idealized with constant radius curvature with R = 4 mm. The single stitch of subcuticular suture was idealized in a U-shaped tunnel in the dermis. The single stitch of intradermal buried vertical mattress suture was modelled in a spiral tunnel. In both suture techniques, the sutures did not pass through epidermises . Meshing The mesh of the subcuticular suture model contained 445,036 elements with the minimal element size of 0.03 mm (Fig. a). The geometry of the intradermal buried vertical mattress suture model was more complicated than that of the subcuticular suture model as more elements were required for the spiral shaped sutures. It contained 673,313 elements with the minimal element size of 0.03 mm (Fig. b). Loads and boundary conditions The mechanical deformation and contact models were implemented for skin suture modelling. As the aim of this study is to compare and evaluate the tissue tension difference between the suturing techniques and sutures at relatively small deformation, it is reasonable to use the linear elastic material model for skin and suture models. The suture process was carried out at a steady and slow motion. The dynamic forces were neglected. The deformation equation of the materials can be expressed by 1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nabla \cdot \left( {{\text{FS}}} \right)^{{\text{T}}} + {\mathbf{F}}_{{\text{V}}} = 0$$\end{document} ∇ · FS T + F V = 0 where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{F}} = {\mathbf{I}} + \nabla {\mathbf{u}}$$\end{document} F = I + ∇ u is the deformation tensor, u is the displacement vector, I is the identity tensor, S is the second Piola–Kirchhoff stress tensor and F v is the body load. The constitutive equation can be expressed as 2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{S}} = {\text{JF}}^{ - 1} \left( {{\mathbf{C}}:\varepsilon } \right){\text{F}}^{{ - {\text{T}}}}$$\end{document} S = JF - 1 C : ε F - T where J is the volumetric deformation, C is the material constitutive tensor and ε is the strain tensor. The stress can be calculated by 3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma_{ij} = C_{ijkl} \varepsilon_{kl}$$\end{document} σ ij = C ijkl ε kl The interface between sutures and skin tissues was modelled by contacting models. The mechanical property difference between standard sutures and barbed sutures was modelled using different contact friction coefficient. The friction coefficients of the standard sutures and barbed sutures were zero and 0.3, respectively . The contact pressure T n was calculated using the penalty factor and contact gap by 4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{n} = - p_{n} g_{n} + p_{0}\,if\,g_{n} < p_{0} /p_{n}$$\end{document} T n = - p n g n + p 0 i f g n < p 0 / p n where p n is the penalty factor, g n is the gap and p 0 is the pressure at zero gap. The penalty factor can be interpreted as the stiffness of a spring inserted between the contact surfaces. Certain degree of overlapping could occur while p 0 was zero. In this study, the initial pressure between the skin tissue and the sutures was estimated based on the assumption that the sutures are able to expand the suture tunnels to the suture diameters. The simplified FE models were established to model a single stitch passing the tissue. The suture pressure was applied to the cross section of 0.2 mm width, which represents the USP 3–0 suture. The displacement was calculated under different pressure. The pressure that generated the 0.1 mm displacement was 100 kPa, which was considered to be the suture initial contact pressure at zero gap. The prescribed displacement was applied to the epidermis layer to simulate the skin stretch (Fig. ). Material modeling The soft tissues are close to volumetrically incompressible . The Poisson’s ratios were close to 0.5 for the tissues. The material properties are listed in Table . The mesh of the subcuticular suture model contained 445,036 elements with the minimal element size of 0.03 mm (Fig. a). The geometry of the intradermal buried vertical mattress suture model was more complicated than that of the subcuticular suture model as more elements were required for the spiral shaped sutures. It contained 673,313 elements with the minimal element size of 0.03 mm (Fig. b). The mechanical deformation and contact models were implemented for skin suture modelling. As the aim of this study is to compare and evaluate the tissue tension difference between the suturing techniques and sutures at relatively small deformation, it is reasonable to use the linear elastic material model for skin and suture models. The suture process was carried out at a steady and slow motion. The dynamic forces were neglected. The deformation equation of the materials can be expressed by 1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nabla \cdot \left( {{\text{FS}}} \right)^{{\text{T}}} + {\mathbf{F}}_{{\text{V}}} = 0$$\end{document} ∇ · FS T + F V = 0 where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{F}} = {\mathbf{I}} + \nabla {\mathbf{u}}$$\end{document} F = I + ∇ u is the deformation tensor, u is the displacement vector, I is the identity tensor, S is the second Piola–Kirchhoff stress tensor and F v is the body load. The constitutive equation can be expressed as 2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{S}} = {\text{JF}}^{ - 1} \left( {{\mathbf{C}}:\varepsilon } \right){\text{F}}^{{ - {\text{T}}}}$$\end{document} S = JF - 1 C : ε F - T where J is the volumetric deformation, C is the material constitutive tensor and ε is the strain tensor. The stress can be calculated by 3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma_{ij} = C_{ijkl} \varepsilon_{kl}$$\end{document} σ ij = C ijkl ε kl The interface between sutures and skin tissues was modelled by contacting models. The mechanical property difference between standard sutures and barbed sutures was modelled using different contact friction coefficient. The friction coefficients of the standard sutures and barbed sutures were zero and 0.3, respectively . The contact pressure T n was calculated using the penalty factor and contact gap by 4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{n} = - p_{n} g_{n} + p_{0}\,if\,g_{n} < p_{0} /p_{n}$$\end{document} T n = - p n g n + p 0 i f g n < p 0 / p n where p n is the penalty factor, g n is the gap and p 0 is the pressure at zero gap. The penalty factor can be interpreted as the stiffness of a spring inserted between the contact surfaces. Certain degree of overlapping could occur while p 0 was zero. In this study, the initial pressure between the skin tissue and the sutures was estimated based on the assumption that the sutures are able to expand the suture tunnels to the suture diameters. The simplified FE models were established to model a single stitch passing the tissue. The suture pressure was applied to the cross section of 0.2 mm width, which represents the USP 3–0 suture. The displacement was calculated under different pressure. The pressure that generated the 0.1 mm displacement was 100 kPa, which was considered to be the suture initial contact pressure at zero gap. The prescribed displacement was applied to the epidermis layer to simulate the skin stretch (Fig. ). The soft tissues are close to volumetrically incompressible . The Poisson’s ratios were close to 0.5 for the tissues. The material properties are listed in Table . Four suture models were developed with two different suture techniques and two different surgical sutures. The stresses of the tissues under stretch loads were simulated as shown in Fig. . For the subcuticular suture with smooth suture the maximum stress reached 1.21 kPa under 0.04 mm stretch, which located near the bending tunnel. For the intradermal buried vertical mattress suture the maximum stress reached 8.01 kPa under 0.04 mm stretch. The reason of the difference was that the contact area difference at dermis. As dermis has higher elastic modulus and strength, it was able to undertake more load at same deformation. For subcuticular suture technique, all the suture pass was in the dermis layers. For the intradermal buried vertical mattress suture technique, only partial of the suture pass through dermis layers. The results suggested that subcuticular sutures caused less stress concentration compared with intradermal buried vertical mattress. The maximum stress of the tissues using barbed suture was 0.62 kPa for subcuticular suture and 1.98 kPa for intradermal buried vertical mattress suture. The proportion of the stress reduction was more significant for subcuticular suture compared with for intradermal suture. The reason was that the stress concentration for intradermal buried vertical mattress was due to the limited contact area in dermis layers. The barbed sutures effectively increased the contact force for subepidermal layers, which led the less force variation between different layers. Regardless the proportional difference for intradermal buried vertical mattress, subcuticular sutures caused less stress concentration using both standard sutures and barbed sutures. Suturing practices vary widely depending on the surgeon and type of wound operated. Incorrect suture technique and material selection, as well as the nature of the wound, can lead to surgical incision complications such as cheese-wiring, infection, necrosis, dehiscence, and keloids. Mechanical forces have an important role in the initiation and progression of these complications . Tissue edema and the volume of the implant cause considerable dermal tension when the skin incision is closed in orthopedic surgery. This tension was increased by joint movement during early postoperative rehabilitation exercises. When sutures are put in an interrupted pattern, the whole wound strain is directed to a few important closure spots . Individual suture loops are subjected to excessive tension, which might result in localized ischemia . Microangiographic investigations have shown that wound edges closed with tight suture loops have limited blood flow, resulting in tissue necrosis . The main causes of wound dehiscence include pressure-induced ischemia and necrosis, which predispose the wound to infection . The tension concentration that happens with interrupted sutures is avoided with continuous sutures. The greater friction between the barbed sutures and the tissue helps surgeons overcome the impulse of over-tightening every stitch in continuous sutures. The axially placed escarpments spaced along the length of the suture allow diffuse distribution of tensions along the entire length of the wound, which may provide additional benefits with respect to safety and cosmesis. As indicated in Table , whether employing running subcuticular suture or intradermal buried vertical mattress suture, barbed sutures stretched the dermis more equally than smooth sutures. Wounds subject to excessive tension can also result in wider and more unsightly scars . Intrinsic mechanical forces, including as tension, shear force, osmotic pressure, and hydrostatic pressure, are vital in cutaneous wound healing, with cells converting mechanical inputs into electrical or chemical signals . Natural mechanical qualities vary by body location and have been recognized as key etiological variables in keloid development, with areas of more mechanical stimulation having a higher keloid incidence . Keloids develop and spread both vertically and horizontally, with the direction of their horizontal growth resulting in distinct morphologies that vary depending on where they are. The keloid growth patterns reflect the primary directions of skin tension, which is corroborated by finite element analysis . Skin tension manipulation has been shown to be useful in the prevention and treatment of scars in clinical applications. Because of the foregoing data and how keloids and HTS originate in the dermis, it is hypothesized that reducing the stress on the wound dermis might minimize the chance of postsurgery keloid and HTS development . As a result, Hans employed subcutaneous/fascial tensile reduction sutures, which put the stress on the deep fascia and superficial fascia layers, allowing the wound edges to be connected spontaneously under very tiny tension without the need of dermal sutures . According to early research, this method may help to prevent the formation of big scars . The stress on the dermis in the horizontal direction from the sutures is the crucial signal since the key to preventing scar creation using an optimum suture method is to minimize tension in the dermis. When utilizing the intradermal buried vertical mattress suture, the sutures travel through the fat layer and the dermis, whereas the running subcuticular suture travels entirely through the dermis. Because fat is softer and more sensitive to deformation than the dermis, it is ineffective at balancing tension when the tissue mass is torn apart. Intradermal buried vertical mattress suture has less dermal tissue in contact with the suture, resulting in greater tension. Scar-less sutures had previously been widely utilized in aesthetic surgery, and as patient demand grows, cosmetic sutures are increasingly being employed in orthopedic surgery. Fully buried sutures are often utilized following thoracolumbar spine and hip surgery with few complications, while interrupted suture is more advised for other locations such as wounds after heel internal fixation. To promote better wound healing, it is necessary to research and select appropriate suturing techniques and tools. Based on our findings, we advocate using barbed sutures in conjunction with running subcuticular suture to close orthopedic surgical skin incisions unless contraindicated. When compared to the combination of running subcuticular suture and intradermal buried vertical mattress suture, the use of the intradermal buried vertical mattress suture as a single suture in closing the surgical neck incision produced the same suture and better cosmetic effects . It should be observed that the neck skin is lax, and the dermis can be spontaneously aligned with low tension once the subcutaneous tissue is sutured. Dermal sutures are required during orthopedic surgery due to increased pressure in the dermis caused by tissue edema and implant volume. The modified buried vertical mattress suture (heart-shaped suture) is currently commonly used in obstetrics and other professions, and is gaining popularity in orthopedics, but its biomechanical benefits are rarely reported . Because of the larger bite of dermis in a single stitch, its decompression effect should be superior to that of the intradermal buried vertical mattress suture (Fig. ). With satisfactory clinical response, barbed sutures are frequently utilized in further aspects of orthopedic surgery, such as sealing the joint capsule . It is the first time that the effect of suturing techniques and sutures used to close orthopedic skin incisions has been studied using 3D finite elements method. This experiment is a preliminary exploration, the model only selects one stitch in the suture instead of simulating the whole wound, and does not consider the individualized mechanical characteristics of wounds in various physical regions. Hopefully, it will be the cornerstone of research on wound closure mechanics. In conclusion, our study indicated that running subcuticular suturing technique with absorbable barbed sutures for orthopedic surgical incisions closure results in more uniform stress distribution in the dermis. We recommend this combination as the preferred method of skin closure in orthopedic surgery unless contraindicated.
Generalising uncertainty improves accuracy and safety of deep learning analytics applied to oncology	c8606f4e-6c7e-40ed-923b-d193961650d1	10164181	Internal Medicine[mh]	Recent advances in deep learning (DL) have led to the rapid development of diagnostic and treatment support applications in various aspects of healthcare, including oncology – . The proposed applications of DL utilise a range of data modalities, including MRI scans , CT scans , histopathology slides , genomics , transcriptomics , , and most recently, integrated approaches with various data types , . In general, studies using DL show excellent predictive performance, providing hope for successful translation into clinical practice , . However, prediction accuracy in DL comes with potential pitfalls which need to be overcome before wider adoption can be eventuated . The lack of transparency over prediction reliability is one challenge for implementing DL . One approach to overcome this is by providing uncertainty estimates about a model’s prediction , , enabling better-informed decision making. Another obstacle relates to the assumptions made about data when transitioning from training to real-world applications. In standard DL practice, during the ‘development’ stage, models are trained and validated on data prepared to satisfy the assumption of independent and identically distributed (IID) data, meaning that model would be applied to make predictions on the data that are independently drawn and come from the same distribution as the training data. However, this assumption cannot be guaranteed and is, in fact, frequently violated when models are deployed in ‘production’ (i.e. real-world application). This is because confounding variables, which we cannot control for, cause distributional shifts that push data out-of-distribution (OOD) . For oncology applications, confounding variables can include technical differences in how the data are collected (e.g., batch effects, differences in sequencing depth or library choice for genomic and transcriptomic data; differences in instrumentation and imaging settings for medical imaging data), as well as biological differences (e.g., differences in patient demographics or a data class unseen during model development). The consequences from OOD data include inaccurate predictions coupled with underestimated uncertainties, which together result in the model’s overconfidence from distributional shifts, or what we call ‘shift-induced’ overconfidence – . Consequently, implementation of DL into clinical practice (i.e., production) requires that models are robust (i.e., generalise) to distributional shifts and provide correct predictions with calibrated uncertainties. Methods to address DL overconfidence in production exist, albeit with different limitations. Repeated retraining of deployed models on new production data is beneficial for accuracy, but introduces new risks such as over-computation or catastrophic forgetting, whereby DL models lose performance on original training/development data , . Using tracking metrics such as accuracy can help inform ML engineers about the DL reliability, although such metrics are only available retrospectively. A key pitfall for these methods is that they are reactive and not proactive. One proactive approach for managing risks in production is with ‘uncertainty thresholding’, whereby only predictions with uncertainties below a threshold are accepted (to increase accuracy). Unfortunately, a DL model’s uncertainty threshold is established with development (IID) data. Thus, when the model is deployed to production (OOD) data it runs a high risk of becoming overconfident. Therefore, the uncertainty threshold established in development corresponds to higher error-rate in production, which is a problem if expectations (between healthcare professionals and engineers) are set during the development phase of a project. To address this problem, post-hoc methods exist that calibrate uncertainty (e.g., with ‘Temperature scaling’ ). However, while post-hoc calibration effectively controls overconfidence in IID data , it fails to do so proactively in OOD data , . Despite the notable theoretical and empirical research towards generalising DL uncertainties from OOD data , , shift-induced overconfidence is yet to be sufficiently addressed in practice. In this study, we aim to address the generally under-appreciated shift-induced DL overconfidence in the context of oncology—the field that is particularly vulnerable to this pitfall due to frequent data distribution shifts. We conduct our experiments with a case study that predicts cancer of origin with transcriptomic data. Cancer of origin prediction has been an active application area for DL , – , since accurate diagnosis is critical for the treatment of cancers of unknown primary (CUP), i.e. metastatic cancers in which the primary cancer site cannot be reliably determined. We investigate multiple cancer datasets, including one newly introduced dataset, with simple, effective, and scalable approximate Bayesian DL techniques that improve generalisation. We examine if the techniques improve model robustness to shift-induced overconfidence and, therefore, improve the DL reliability. We introduce the prototypical ADP metric to measure model robustness to shift-induced overconfidence and to directly explain the “expected loss of accuracy during deployment in an uncertainty-thresholding regime”. Finally, we provide a brief discussion about how ADP supports model selection and how that can be helpful within a clinical setting. Bayesian model benchmarking approach to predict cancer of unknown primary The primary DL task was to predict the tissue of origin (primary cancer type) of cancer samples using transcriptomic data. We used transcriptomic data from TCGA of primary cancer samples corresponding to 32 primary cancer types as model ‘development’ data: training (n = 8202 ) and validation IID data (n = 1434; Supplementary Table ). The test data were OOD (representing ‘production’), providing a platform for benchmarking resilience to overconfidence, and included TCGA metastatic samples (n = 392 ), Met500 metastatic samples (n = 479 ), and a combination of primary and metastatic samples from our own independent internal custom dataset, i.e. ICD (n = 461 – ; Fig. a, Supplementary Fig. ). The distributional shifts in the test data were likely to be caused by several factors, including dataset batches, sample metastasis status (metastatic or primary) and whether the cancer type was absent during training (‘unseen’). We aimed to evaluate if three simple ‘distribution-wise’ Bayesian DL models improve performance and reduce shift-induced overconfidence compared to a pointwise baseline model (with identical Resnet architecture). To achieve this, we performed controlled benchmarking of the models over IID and OOD data (Fig. b). The experiment was controlled by enforcing consistency for factors affecting uncertainty within the validation/IID dataset. Specifically, all models had identical architecture, hyperparameter, and optimisation settings. Importantly, all models had identical (negative log likelihood) loss within the validation/IID dataset. We intentionally did not perform hyperparameter optimisation for each model, as it was important for our study design to control for accuracy. The Bayesian models were Monte Carlo Dropout approximation (‘MCD’) , MCD with smoothness and sensitivity constraints (‘Bilipschitz’) , , and an ensemble of Bilipschitz models (‘Ensemble’) . The ways in which models differed were canonical: MCD modified Resnet by keeping Dropout during prediction, Bilipschitz modified MCD with spectral normalisation, Ensemble modified Bilipschitz by combining multiple models. Approximate Bayesian inference reduces shift-induced overconfidence for ‘seen’ classes in a primary cancer site context The predictive performance of each model to predict primary tissue was assessed using micro-F1 (equivalent to Accuracy; abbreviated F1). For the IID validation data, the difference between the highest and lowest ranking models was 0.28% (97.07% for Resnet and 96.79% for Ensemble, respectively; Fig. a, Supplementary Fig. – ). This was anticipated, since the loss was controlled for within validation data. As expected, F1 scores dropped for the OOD test set across all four models, with a 1.74% difference between the highest and lowest ranking models (82.04% for Ensemble and 80.30% for Resnet, respectively; Fig. b, Supplementary Figs. – ). All models had higher predictive uncertainties (Shannon’s entropy II) for OOD, relative to IID data (Fig. b). Uncertainties were significantly higher for all approximate Bayesian models (MCD, Bilipschitz, and Ensemble) relative to (pointwise) Resnet ( p < 0.0001). Moreover, overconfidence in OOD data was evident for the Resnet and MCD models since their binned accuracies (i.e., the correct classification rates within bins delineated by the confidence scores) were consistently lower than corresponding confidence scores (Fig. c). The expected calibration errors (ECEs) for OOD data ranged between 5% for Ensemble and Bilipschitz and 16% for Resnet (Fig. c). Estimation of overconfidence as an absolute error was negligible across all models for IID data, with high amounts of overconfidence for OOD data, highlighting the shift-induced overconfidence when transitioning from IID to OOD data (Fig. d). Furthermore, Resnet had significantly higher overconfidence than MCD ( p value < 0.01), Bilipschitz ( p value < 0.001), and Ensemble ( p value < 0.001) for OOD data but not IID data. This shows that the shift-induced overconfidence in pointwise DL models can be reduced with simple (approximate) Bayesian inference. Prediction overconfidence for ‘unseen’ classes explained by related primary cancer types Classes absent from training (‘unseen’) cannot have correct predictions, and prediction uncertainties should be higher compared to ‘seen’ classes. As expected, mean total uncertainties were higher for ‘unseen’ classes for all models (Fig. a). Moreover, approximate Bayesian models were significantly more uncertain with ‘unseen’ classes compared to Resnet ( p value < 0.01; Fig. a). However, exceptions occurred across all models, where total uncertainty values were low, at both: class level, where predictions for a whole ‘unseen’ class consistently had low uncertainty; and sample level, where predictions for only some samples from a class had low uncertainty (Fig. b). We wanted to investigate whether any of the exceptions could be examples of ‘silent catastrophic failure’ (Supplementary Information— ), a phenomenon where data are far from the training data’s support, resulting in incorrect yet extremely confident predictions – . ‘Unseen’ classes (i.e., cancer types) with low levels of uncertainty (averaged within the class) corresponded to ‘seen’ classes that either (biologically) related to the predicted primary cancer type, or were from a similar tissue or cell of origin. For example, all acral melanoma (ACRM) samples (n = 40), a subtype of melanoma that occurs on soles, palms and nail beds, were predicted to be cutaneous melanoma (MEL) by all four models (Supplementary Figs. – ) with the smallest median total uncertainty for all four models (Fig. b). All three fibrolamellar carcinoma (FLC) samples, a rare type of liver cancer, were predicted to be hepatocellular carcinomas (HCC), although the median uncertainty was much higher for Bilipschitz and Ensemble models compared to Resnet and MCD (1.8, 1.5, 0.1 and 0.29 Shannon’s Entropy II, respectively). Two bladder squamous cell carcinomas (BLSC) showed different examples of class-level exceptions with one sample predicted as a bladder adenocarcinoma (BLCA), with the same primary tissue site as BLSC, or a lung squamous carcinoma (LUSC), with similar cell of origin. For the ‘unseen’ class pancreatic neuroendocrine tumours (PANET) we saw a wide spread of uncertainty values (Fig. b). Interestingly, only PANET samples that were predicted as another subtype of pancreatic cancer, pancreatic adenocarcinomas (PAAD), had low prediction uncertainty across all models compared to other incorrectly predicted PANET samples (Supplementary Fig. ). Overall, since most of the incorrect predictions with low uncertainties had a reasonable biological explanation for the prediction, we concluded that we did not find strong evidence of catastrophic silent failure in this case study. Robustness to shift-induced overconfidence is integral for production inference To evaluate the robustness of the models’ accuracy, as well as the uncertainty’s correlation with the error-rate (abbreviated “uncertainty’s error-rate correlation”) we used the F1-Retention Area Under the Curve (F1-AUC) . Evaluation was carried out on ‘seen’ and ‘unseen’ OOD data (i.e., ‘production data’). All models yielded similar results, with only a 0.45% percent decrease between the highest and lowest ranking models (F1-AUC of 93.67% for Bilipschitz and 93.25% for MCD, respectively; Fig. a). The performance difference between all models was marginal as F1-AUC doesn’t capture the lost calibration caused by the distributional shift when transitioning from IID to (‘seen’ and ‘unseen’) OOD. In other words, the F1-AUC metric did not detect effects caused by the shift-induced overconfidence. This was evident from the following observations: (1) inter-model accuracies were similar within IID, as well as OOD data (Fig. a); (2) calibration errors (i.e. overconfidence) were not different for IID ( p value > 0.05), but different for OOD ( p value < 0.01; Fig. d); and (3) F1-AUC scores were similar for all models, which implies ‘uncertainty’s error-rate correlation’ must have been similar (since F1-AUC encapsulates accuracy and ‘uncertainty’s error-rate correlation’ ). Thus, while we showed that F1-AUC encapsulated accuracy and ‘uncertainty’s error-rate correlation’, both of which are important components of robustness when deploying DL in production, we caution that F1-AUC does not encapsulate robustness to shift-induced overconfidence. Hence it is not sufficient for safe deployment in clinical practice. To overcome the limitation of the F1-AUC metric’s insensitivity to shift-induced overconfidence, we developed a new (prototypical) metric called the Area between the Development and Production curve (ADP), which depends on both IID (i.e., ‘development’) data, as well as the (‘seen’ and ‘unseen’) OOD (i.e., ‘production’) data. The ADP may be interpreted as “the expected decrease in accuracy when transitioning from development to production if uncertainty thresholding is utilised to boost reliability”. The ADP differs from ECE and Accuracy in two primary ways. First, ECE and accuracy relate to a single data set, whereas the ADP relates to two data sets, hence ADP explains the expected change in, for example, accuracy from one data set relative to the other. Second, the ADP complements and subsumes F1-AUC in the context of deploying models from training/development data (IID) to production test data (OOD). The ADP was calculated by averaging the set of decreases in F1, from development (IID) to production (OOD) datasets, at multiple different uncertainty thresholds (a single F1-decrease is demonstrated in Fig. b; refer to the “ ” section for details). The ADP metric detected effects from shift-induced overconfidence, with an inter-model percent decrease that was two orders of magnitude larger than F1-AUC (Fig. c). The percent decrease between the top and bottom ranking models was 53.68%. The top-ranking model was Bilipschitz with an ADP of 4.28%, and the bottom ranking model was Resnet with ADP of 9.24% (Fig. c). This highlights that ADP may be relevant when evaluating the performance of models that are deployed in production by encapsulating shift-induced overconfidence, which is inevitable in an oncological setting. To further illustrate the utility of ADP, we performed an additional experiment (Supplementary Fig. ). We used an independent classification task, the well-known CIFAR-10 (IID) dataset and its’ OOD variant—CIFAR-10-C, and compared a non-Bayesian CNN Resnet model and a Deep Kernel Learning Model (i.e., neural Gaussian process). The results were in line with our hypothesis that Bayesian deep learning improves robustness to distribution shift, demonstrated by a lower ADP for the Gaussian process model compared to the Resnet model. The primary DL task was to predict the tissue of origin (primary cancer type) of cancer samples using transcriptomic data. We used transcriptomic data from TCGA of primary cancer samples corresponding to 32 primary cancer types as model ‘development’ data: training (n = 8202 ) and validation IID data (n = 1434; Supplementary Table ). The test data were OOD (representing ‘production’), providing a platform for benchmarking resilience to overconfidence, and included TCGA metastatic samples (n = 392 ), Met500 metastatic samples (n = 479 ), and a combination of primary and metastatic samples from our own independent internal custom dataset, i.e. ICD (n = 461 – ; Fig. a, Supplementary Fig. ). The distributional shifts in the test data were likely to be caused by several factors, including dataset batches, sample metastasis status (metastatic or primary) and whether the cancer type was absent during training (‘unseen’). We aimed to evaluate if three simple ‘distribution-wise’ Bayesian DL models improve performance and reduce shift-induced overconfidence compared to a pointwise baseline model (with identical Resnet architecture). To achieve this, we performed controlled benchmarking of the models over IID and OOD data (Fig. b). The experiment was controlled by enforcing consistency for factors affecting uncertainty within the validation/IID dataset. Specifically, all models had identical architecture, hyperparameter, and optimisation settings. Importantly, all models had identical (negative log likelihood) loss within the validation/IID dataset. We intentionally did not perform hyperparameter optimisation for each model, as it was important for our study design to control for accuracy. The Bayesian models were Monte Carlo Dropout approximation (‘MCD’) , MCD with smoothness and sensitivity constraints (‘Bilipschitz’) , , and an ensemble of Bilipschitz models (‘Ensemble’) . The ways in which models differed were canonical: MCD modified Resnet by keeping Dropout during prediction, Bilipschitz modified MCD with spectral normalisation, Ensemble modified Bilipschitz by combining multiple models. The predictive performance of each model to predict primary tissue was assessed using micro-F1 (equivalent to Accuracy; abbreviated F1). For the IID validation data, the difference between the highest and lowest ranking models was 0.28% (97.07% for Resnet and 96.79% for Ensemble, respectively; Fig. a, Supplementary Fig. – ). This was anticipated, since the loss was controlled for within validation data. As expected, F1 scores dropped for the OOD test set across all four models, with a 1.74% difference between the highest and lowest ranking models (82.04% for Ensemble and 80.30% for Resnet, respectively; Fig. b, Supplementary Figs. – ). All models had higher predictive uncertainties (Shannon’s entropy II) for OOD, relative to IID data (Fig. b). Uncertainties were significantly higher for all approximate Bayesian models (MCD, Bilipschitz, and Ensemble) relative to (pointwise) Resnet ( p < 0.0001). Moreover, overconfidence in OOD data was evident for the Resnet and MCD models since their binned accuracies (i.e., the correct classification rates within bins delineated by the confidence scores) were consistently lower than corresponding confidence scores (Fig. c). The expected calibration errors (ECEs) for OOD data ranged between 5% for Ensemble and Bilipschitz and 16% for Resnet (Fig. c). Estimation of overconfidence as an absolute error was negligible across all models for IID data, with high amounts of overconfidence for OOD data, highlighting the shift-induced overconfidence when transitioning from IID to OOD data (Fig. d). Furthermore, Resnet had significantly higher overconfidence than MCD ( p value < 0.01), Bilipschitz ( p value < 0.001), and Ensemble ( p value < 0.001) for OOD data but not IID data. This shows that the shift-induced overconfidence in pointwise DL models can be reduced with simple (approximate) Bayesian inference. Classes absent from training (‘unseen’) cannot have correct predictions, and prediction uncertainties should be higher compared to ‘seen’ classes. As expected, mean total uncertainties were higher for ‘unseen’ classes for all models (Fig. a). Moreover, approximate Bayesian models were significantly more uncertain with ‘unseen’ classes compared to Resnet ( p value < 0.01; Fig. a). However, exceptions occurred across all models, where total uncertainty values were low, at both: class level, where predictions for a whole ‘unseen’ class consistently had low uncertainty; and sample level, where predictions for only some samples from a class had low uncertainty (Fig. b). We wanted to investigate whether any of the exceptions could be examples of ‘silent catastrophic failure’ (Supplementary Information— ), a phenomenon where data are far from the training data’s support, resulting in incorrect yet extremely confident predictions – . ‘Unseen’ classes (i.e., cancer types) with low levels of uncertainty (averaged within the class) corresponded to ‘seen’ classes that either (biologically) related to the predicted primary cancer type, or were from a similar tissue or cell of origin. For example, all acral melanoma (ACRM) samples (n = 40), a subtype of melanoma that occurs on soles, palms and nail beds, were predicted to be cutaneous melanoma (MEL) by all four models (Supplementary Figs. – ) with the smallest median total uncertainty for all four models (Fig. b). All three fibrolamellar carcinoma (FLC) samples, a rare type of liver cancer, were predicted to be hepatocellular carcinomas (HCC), although the median uncertainty was much higher for Bilipschitz and Ensemble models compared to Resnet and MCD (1.8, 1.5, 0.1 and 0.29 Shannon’s Entropy II, respectively). Two bladder squamous cell carcinomas (BLSC) showed different examples of class-level exceptions with one sample predicted as a bladder adenocarcinoma (BLCA), with the same primary tissue site as BLSC, or a lung squamous carcinoma (LUSC), with similar cell of origin. For the ‘unseen’ class pancreatic neuroendocrine tumours (PANET) we saw a wide spread of uncertainty values (Fig. b). Interestingly, only PANET samples that were predicted as another subtype of pancreatic cancer, pancreatic adenocarcinomas (PAAD), had low prediction uncertainty across all models compared to other incorrectly predicted PANET samples (Supplementary Fig. ). Overall, since most of the incorrect predictions with low uncertainties had a reasonable biological explanation for the prediction, we concluded that we did not find strong evidence of catastrophic silent failure in this case study. To evaluate the robustness of the models’ accuracy, as well as the uncertainty’s correlation with the error-rate (abbreviated “uncertainty’s error-rate correlation”) we used the F1-Retention Area Under the Curve (F1-AUC) . Evaluation was carried out on ‘seen’ and ‘unseen’ OOD data (i.e., ‘production data’). All models yielded similar results, with only a 0.45% percent decrease between the highest and lowest ranking models (F1-AUC of 93.67% for Bilipschitz and 93.25% for MCD, respectively; Fig. a). The performance difference between all models was marginal as F1-AUC doesn’t capture the lost calibration caused by the distributional shift when transitioning from IID to (‘seen’ and ‘unseen’) OOD. In other words, the F1-AUC metric did not detect effects caused by the shift-induced overconfidence. This was evident from the following observations: (1) inter-model accuracies were similar within IID, as well as OOD data (Fig. a); (2) calibration errors (i.e. overconfidence) were not different for IID ( p value > 0.05), but different for OOD ( p value < 0.01; Fig. d); and (3) F1-AUC scores were similar for all models, which implies ‘uncertainty’s error-rate correlation’ must have been similar (since F1-AUC encapsulates accuracy and ‘uncertainty’s error-rate correlation’ ). Thus, while we showed that F1-AUC encapsulated accuracy and ‘uncertainty’s error-rate correlation’, both of which are important components of robustness when deploying DL in production, we caution that F1-AUC does not encapsulate robustness to shift-induced overconfidence. Hence it is not sufficient for safe deployment in clinical practice. To overcome the limitation of the F1-AUC metric’s insensitivity to shift-induced overconfidence, we developed a new (prototypical) metric called the Area between the Development and Production curve (ADP), which depends on both IID (i.e., ‘development’) data, as well as the (‘seen’ and ‘unseen’) OOD (i.e., ‘production’) data. The ADP may be interpreted as “the expected decrease in accuracy when transitioning from development to production if uncertainty thresholding is utilised to boost reliability”. The ADP differs from ECE and Accuracy in two primary ways. First, ECE and accuracy relate to a single data set, whereas the ADP relates to two data sets, hence ADP explains the expected change in, for example, accuracy from one data set relative to the other. Second, the ADP complements and subsumes F1-AUC in the context of deploying models from training/development data (IID) to production test data (OOD). The ADP was calculated by averaging the set of decreases in F1, from development (IID) to production (OOD) datasets, at multiple different uncertainty thresholds (a single F1-decrease is demonstrated in Fig. b; refer to the “ ” section for details). The ADP metric detected effects from shift-induced overconfidence, with an inter-model percent decrease that was two orders of magnitude larger than F1-AUC (Fig. c). The percent decrease between the top and bottom ranking models was 53.68%. The top-ranking model was Bilipschitz with an ADP of 4.28%, and the bottom ranking model was Resnet with ADP of 9.24% (Fig. c). This highlights that ADP may be relevant when evaluating the performance of models that are deployed in production by encapsulating shift-induced overconfidence, which is inevitable in an oncological setting. To further illustrate the utility of ADP, we performed an additional experiment (Supplementary Fig. ). We used an independent classification task, the well-known CIFAR-10 (IID) dataset and its’ OOD variant—CIFAR-10-C, and compared a non-Bayesian CNN Resnet model and a Deep Kernel Learning Model (i.e., neural Gaussian process). The results were in line with our hypothesis that Bayesian deep learning improves robustness to distribution shift, demonstrated by a lower ADP for the Gaussian process model compared to the Resnet model. A major barrier to using DL in clinical practice is the shift-induced overconfidence encountered when deploying a DL model from development to production. Reducing and accounting for shift-induced overconfidence with appropriate models and relevant metrics should make the models more transparent and trustworthy for translation into practice. Our work herein shows that marked progress can be made with simple Bayesian DL models deployed in conjunction with uncertainty thresholding. However, the performance of models deployed in production can be difficult to evaluate without a suitable metric, therefore we developed ADP to directly measure shift-induced overconfidence. Three Bayesian models with canonical extensions, namely MCD, Bilipschitz, Ensemble, were chosen to test whether simple modifications applicable to any DL architecture can improve performance in production. The Bayesian models were selected according to criteria for which we believe would facilitate adoption: (1) simplicity, for wider accessibility; (2) ubiquity, to ensure models were accepted and tested methods; (3) already demonstrated as robust to shift-induced overconfidence , , ; and (4) computational scalability. Our prior expectations were that each canonical extension would further improve generalisation of both accuracy and uncertainty quality, albeit at the cost of increased complexity. Those expectations were mostly in line with our benchmarking results, since the most complex model (Ensemble) went from worst-performing in IID to best-performing model in OOD in terms of accuracy. Furthermore, while inspection into overconfidence presented no significant inter-model differences within IID data, the OOD overconfidence differences were significant, whereby added complexity corresponded to less shift-induced overconfidence. Using the ADP statistic, improvements in robustness to shift-induced overconfidence were shown to have a large impact on the accuracy in production when rejecting unreliable predictions above an acceptable uncertainty threshold. Hence, any DL architecture’s accuracy in production can be substantially improved with simple and scalable approximate Bayesian modifications. This phenomenon is sometimes referred to as “turning the Bayesian crank” . We restricted our uncertainty statistics to predictive (i.e., total) uncertainties, since it was not possible to estimate the sub-divisions of uncertainty with the baseline Resnet model, which only captures uncertainty about the data. The Bayesian models captured an additional component of uncertainty, the ‘epistemic’ uncertainty, hence they all had larger total uncertainty estimates when compared to the non-Bayesian baseline. Consequently, the Bayesian models filled the uncertainty gap caused by distribution shift (i.e., shift-induced overconfidence). In future work, a richer picture may be understood by focusing only on distribution-wise models to inspect the two sub-divisions of the predictive uncertainty: epistemic (model) uncertainty and aleatoric (inherent) uncertainty. Epistemic uncertainty is dependent on the model specification and may be reduced with more data or informative priors. Aleatoric uncertainty is dependent on data’s inherent noise and can be reduced with more data features that explain variance caused by confounding variables (e.g., patient age, cancer stage, batch effect). Epistemic and aleatoric uncertainties present the potential for further insights, including whether a data point’s predictive uncertainty will reduce with either more examples or by an altered model design (epistemic uncertainty), or more features (aleatoric uncertainty) – . This study addressed distributional shift effects on uncertainties with parametric models, which assume parameters are sufficient to represent all training data. Non-parametric models relax that assumption, which is arguably crucial to detect when data are outside the domain of training data (‘out-of-domain’) and for avoiding extreme overconfidence, i.e., ‘silent catastrophic failure’. In future work, non-parametric models, for example Gaussian Processes, capable of measuring uncertainties about ‘out-of-domain’ data, should also be explored – , . Our work suggests that considerations of robustness to distributional shifts must encapsulate uncertainty and prediction to improve performance in production. While this study focused on the quality of uncertainty, it is important to note that other DL components are worth consideration too. These include model architecture (i.e. inductive bias), which can be tailored to ignore redundant data-specific aspects of a problem via invariant or equivariant model representations , data-augmentation strategies , and/or structural causal models – . Such tailored models can further improve data efficiency , robustness to distributional shifts , and are central to an appropriate model specification that challenges DL deployment . The importance of tailored inductive biases is supported by the prolific advances in fields beyond clinical diagnostics in computer vision (e.g. CNN’s translational equivariance ), and biology (e.g. how Alpha Fold 2 solved the Critical Assessment of protein Structure Prediction (CASP ). These studies show that a wide array of DL components can improve generalisation and, thus, DL performance in production. Our study argues uncertainty calibration as an important element in that array; hence, improving the quality of uncertainty can lead to improved DL reliability in production. In practice, we hope the community considers utilising uncertainty thresholding as a proactive method to improve accuracy and safety of DL applications, deployed in the clinic. This may involve (iterative) consultation between ML engineer and medical professionals to agree on a ‘minimally acceptable accuracy’ for production (deem this [12pt]{minimal} $$(F{1}_{dev})$$ min F 1 dev ). The ML engineer may then use development data to train an approximate Bayesian DL model and produce Development F1-Uncertainty curves (with validation data). The engineer then, with another independent dataset, can proceed to develop an ADP estimate (as described in the “ ” section) to help communicate (in context of available dataset differences) what the expected accuracy decrease may be when the model is deployed to production, which helps manage expectations and facilitate trust. Importantly, with the (prototypical) ADP, the team may better judge which uncertainty quantification techniques are most effective for boosting accuracy under the ‘uncertainty thresholding’ risk-management regime. This procedure, as well as the ADP statistic, is of course prototypical and only suggestive. We leave improvement, and clarification of this for future work. In conclusion, our study highlighted approaches for quantifying and improving robustness to shift-induced overconfidence with simple and accessible DL methods in the context of oncology. We justified our approach with mathematical and empirical evidence, biological interpretation, and a new metric, the ADP designed to encapsulate shift-induced overconfidence—a crucial aspect that needs to be considered when deploying DL in real-world production. Moreover, the ADP is directly interpretable as a proxy to expected accuracy loss when deploying DL models from development to production. Although we have addressed the shift-induced overconfidence by utilising first-line solutions, work remains to bridge DL from theory to practice. We must account for data distributions, evaluation metrics, and modelling assumptions as all are equally important and necessary considerations to see safe translation of DL into clinical practice. Prediction task and datasets The task was to predict a patient's primary cancer type, which we cast under the supervised learning framework by learning the map [12pt]{minimal} $$\{ y\}$$ x → y , with [12pt]{minimal} $$y$$ y denoting the primary cancer category, and [12pt]{minimal} $$ {}^{D}$$ x ∈ R D denoting a patient’s sampled bulk gene expression signature. Three independent datasets were used: our own independent Internal Custom Dataset, ICD – , TCGA , and Met500 . All datasets were pre-processed and partitioned into groups (i.e., strata) that uniquely proxied different distribution shifts. Proxies of approximately unique shifts were assumed to be governed by their respective intervention (i.e. unique shift), as deemed by values of four presumed hidden variables influencing the modelled map [12pt]{minimal} $$\{ y\}$$ x → y . Those variables were ‘Batch’ (indicating source dataset label, e.g., ‘ TCGA ’), ‘ State-of-Metastases ’ (valued ‘ Primary ’, or ‘ Metastatic ’), and ‘ Seen ’ (indicating whether a target value y was seen during training) (Supplementary Table ). Training and validation data comprised of the Strata ID. [12pt]{minimal} $$ {{( {'Batch',\;'State\;of\;Metastases',\;'Seen'} )}}_{Strata \;ID\;key} = {{( {'TCGA',\;'Primary',\; True} )}}_{key \;value},$$ ′ B a t c h ′ , ′ S t a t e o f M e t a s t a s e s ′ , ′ S e e n ′ ⏟ S t r a t a I D k e y = ′ T C G A ′ , ′ P r i m a r y ′ , T r u e ⏟ k e y v a l u e , since we believed it to be approximately independent and identically distributed (IID) data. All other strata were assumed out-of-distribution (OOD) due to distribution shifts caused by confounding variables. As a result, the training and validation data were IID, while the test data were OOD. Benchmarked models Four models were benchmarked in this study—the baseline pointwise Resnet, MCD, Bilipschitz, and Ensemble. All models shared identical model architecture and hyperparameter settings (including early stopping), respectively controlling the inductive bias and accuracy from confounding overconfidence. Although we did not perform explicit hyperparameter optimisation, some manual intervention was used to adjust hyperparameters within the validation set. For example, the singular value bound hyperparameter (for spectral normalisation) was manually tuned to be as low as practically possible, while being capable of being flexible enough to learn the training task of predicting the primary site. Baseline resnet model Resnet architecture had four hidden layers, each with 1024-neurons, Mish activations , batch normalisation , and standard residual connections from the first hidden layer up to the final hidden ‘logit-space’ layer, which was then normalised using the SoftMax function to yield probability vector [12pt]{minimal} $$()= {[]}^{K}$$ p x = ∈ [ 0 , 1 ] K , where the prediction’s class index, [12pt]{minimal} $$c=\,}\{{[{{}_{1},}_{2}, , {}_{K}]}^{T}\}$$ c = arg max k p 1 , p 2 , ⋯ , p K T indicates the primary cancer site’s label [12pt]{minimal} $$y c$$ y ← c . Specifically, a batch [12pt]{minimal} $$ {}^{B D}$$ X ∈ R B × D with [12pt]{minimal} $$B$$ B individual samples is first transformed by the input layer [12pt]{minimal} $${}^{(0)}=g( , {}^{(0)} +{}^{(0)})$$ U 0 = g ( ⟨ X , W 0 ⟩ + b 0 ) , with affine transform parameters [12pt]{minimal} $$\{{}^{(0)}, {}^{(0)}\}$$ W 0 , b 0 , non-linear activations [12pt]{minimal} $$g$$ g , and output representation [12pt]{minimal} $${}^{(0)}$$ U 0 . Hidden layers have residual connections [12pt]{minimal} $${}^{(l)}=g( {}^{(l-1)},{}^{(l)} +{}^{(l)})+{}^{(l-1)}$$ U l = g ⟨ U l - 1 , W l ⟩ + b l + U ( l - 1 ) where [12pt]{minimal} $$l , ,L$$ l ∈ 1 , 2 , ⋯ , L denotes the hidden layer index ( [12pt]{minimal} $$L=3$$ L = 3 in this case). The final output layer is a pointwise (mean estimate) function in logit-space [12pt]{minimal} $$()= g( {}^{(L)},{}^{( )} +{}^{( )})$$ f X = g ⟨ U L , W μ ⟩ + b μ , where [12pt]{minimal} $$\{{}^{( )}, {}^{( )}\}$$ W μ , b μ are the final output (affine) transformation parameters. Finally, SoftMax normalisation yields a K-vector [12pt]{minimal} $$()= {}(())$$ p X = SoftMax f X . All other hyperparameter settings are defined in Supplementary Table . This baseline Resnet model architecture was inherited by all other models in this study to control inductive biases. Approximate Bayesian inference Bayesian inference may yield a predictive distribution about sample [12pt]{minimal} $${}^{}$$ x ∗ , [12pt]{minimal} $$p(\|{}^{},)$$ p ( p \| x ∗ , D ) , from the likelihood of an assumed parametric model [12pt]{minimal} $$p(\|{}^{},)$$ p ( p \| x ∗ , Θ ) , an (approximate) parametric posterior [12pt]{minimal} $$q( \|)$$ q Θ \| D , and potentially Monte Carlo Integration (MCI) technique, also referred to as Bayesian model averaging: [12pt]{minimal} $$p(\|{}^{},) {_{}}p(\|{}^{}, )q( \|)d _{t=1}^{T}p(\|{}^{},{ }_{t})$$ p p \| x ∗ , D ≈ ∫ Θ p p \| x ∗ , Θ q Θ \| D d Θ ≈ 1 T ∑ t = 1 T p ( p \| x ∗ , Θ t ) Most neural networks are parametric models, which assume [12pt]{minimal} $$$$ Θ can perfectly represent [12pt]{minimal} $$$$ D . As a result, the model likelihood [12pt]{minimal} $$p(\|{}^{},,)$$ p ( p \| x ∗ , D , Θ ) is often replaced with [12pt]{minimal} $$(\|{}^{}, )$$ p ( p \| x ∗ , Θ ) . The main differentiating factor among all Bayesian deep learning inference methods lies in how the parametric posterior [12pt]{minimal} $$q( \|)$$ q Θ \| D is approximated. Resnet extended with Monte Carlo Dropout The MCD model approximates the parametric posterior [12pt]{minimal} $$q( \|)$$ q ( Θ \| D ) by keeping dropout activated during inference . Dropout randomly ‘switches off’ a subset of neurons to zero-vectors at each iteration. Hence, a collection of dropout configurations [12pt]{minimal} $${\{{ }_{t}\}}_{t=1}^{T}$$ Θ t t = 1 T are samples from the (approximate) posterior [12pt]{minimal} $$q( \|)$$ q ( Θ \| D ) . For more information, refer to the Appendix of where an approximate dual connection between Monte Carlo Dropout neural networks and Deep Gaussian processes is established. The MCD also extends the Resnet model architecture by including an additional output layer to estimate a data-dependent variance function [12pt]{minimal} $${}_{t}^{2}()= g( {}^{(L)},{}_{t}^{( )} +{}_{t}^{( )})$$ s t 2 X = g ( ⟨ U ( L ) , W t ( Σ ) ⟩ + b t ( Σ ) ) in addition to the (now stochastic) mean function [12pt]{minimal} $${}_{t}()= g( {}^{(L)},{}_{t}^{( )} +{}_{t}^{( )})$$ f t X = g ⟨ U L , W t ( μ ) ⟩ + b t ( μ ) . Both final output layers had a shared input [12pt]{minimal} $${}^{(L)}$$ U ( L ) , but unique parameters [12pt]{minimal} $$\{{}_{t}^{( )},{}_{t}^{( )}\}$$ W t ( μ ) , b t ( μ ) and [12pt]{minimal} $$\{{}_{t}^{( )},{}_{t}^{( )}\}$$ W t ( Σ ) , b t ( Σ ) . Together, the stochastic mean [12pt]{minimal} $${}_{t}()$$ f t X and variance [12pt]{minimal} $${}_{t}^{2}()$$ s t 2 X specify a Gaussian distribution in the logit-space, which was then sampled once [12pt]{minimal} $${}_{t}() ( ={}_{t}(), ={}_{t}^{2}{()}^{T}$$ u t X ∼ N ( μ = f t X , Σ = s t 2 X T I and normalised with the SoftMax function [12pt]{minimal} $${}_{t}()= {}({}_{t}())$$ p t X = SoftMax u t X . [12pt]{minimal} $${}_{t}()$$ p t X represents a single sample from the model likelihood [12pt]{minimal} $$p(\|, )$$ p ( p \| x , Θ ) , from which [12pt]{minimal} $$T$$ T samples are averaged for Monte Carlo integration: [12pt]{minimal} $$()= _{t=1}^{T}{}_{t}().$$ p X = 1 T ∑ t = 1 T p t ( X ) . Finally, [12pt]{minimal} $$()$$ p X estimates the cancer primary site label [12pt]{minimal} $$y$$ y , the predictive uncertainties [12pt]{minimal} $$$$ Conf , and [12pt]{minimal} $$()$$ H . for each individual sample in data batch [12pt]{minimal} $$$$ x . MCD extended with a bi-Lipschitz constraint The Bilipschitz model shared all the properties of the MCD model with an additional bi-Lipschitz constraint: [12pt]{minimal} $${L}_{1}{ {}_{1}-{}_{2} }_{ } { ({}_{1})- ({}_{2}) }_{} {L}_{2}{ {}_{1}-{}_{2} }_{ }$$ L 1 ‖ x 1 - x 2 ‖ X ≤ ‖ f x 1 - f x 2 ‖ F ≤ L 2 ‖ x 1 - x 2 ‖ X where scalars [12pt]{minimal} $${L}_{1}$$ L 1 and [12pt]{minimal} $${L}_{2}$$ L 2 respectively control the tightness of the lower- and upper-bound. Norm operators [12pt]{minimal} $$\{{ }_{ },{ }_{}\}$$ ‖ . ‖ X , ‖ . ‖ F are over the data space [12pt]{minimal} $$$$ X and function space [12pt]{minimal} $$$$ F . The effect of the bi-Lipschitz constraint is such that the changes in input data [12pt]{minimal} $${ {}_{1}-{}_{2} }_{ }$$ ‖ x 1 - x 2 ‖ χ (e.g. distribution shifts) are proportional to the changes in the output, [12pt]{minimal} $${ ({}_{1})- ({}_{2}) }_{}$$ ‖ f x 1 - f x 2 ‖ F . These changes are within a bound determined by [12pt]{minimal} $${L}_{1}$$ L 1 (controlling sensitivity) and [12pt]{minimal} $${L}_{2}$$ L 2 (controlling smoothness). Interestingly, recent studies have established that bi-Lipschitz constraints are beneficial to the robustness of the neural network under distributional shifts , . Sensitivity (i.e. [12pt]{minimal} $${L}_{1}$$ L 1 ) is controlled with residual connections , , which allows [12pt]{minimal} $$()$$ f x to avoid arbitrarily small changes, especially in the presence of distributional shifts in those regions of [12pt]{minimal} $$$$ X with no (training data) support . Sensitivity (i.e. [12pt]{minimal} $${L}_{2}$$ L 2 ) is controlled with spectral normalisation on parameters [12pt]{minimal} $$$$ Θ , and batch-normalisation functions , which allow [12pt]{minimal} $$()$$ f x to avoid arbitrarily large changes (under shifts) that induce feature collapse and extreme overconfidence – . Deep ensemble of Bilipschitz models The Ensemble model was a collection of eight independently trained Bilipschitz models with unique initial parameter configurations. Each Bayesian model in the Ensemble model is sampled [12pt]{minimal} $$T/10(=25)$$ T / 10 ( = 25 ) times and then pooled to control for Monte Carlo integration between the ‘Ensemble’ and all other models. Models in deep ensembles yield similarly performant (low-loss) solutions, but are diverse and distant in parameter- and function-space . This allows the ensemble to have an (approximate) posterior [12pt]{minimal} $$q( \|)$$ q Θ \| D with multiple modes , which was not the case for the Resnet, MCD, and Bilipschitz models. We believe the ensemble modelled [12pt]{minimal} $$q( \|)$$ q Θ \| D with the highest fidelity to the true parametric posterior [12pt]{minimal} $$p( \|)$$ p Θ \| D due to empirical evidence from other studies' results , , , . Model efficacy assessment Model efficacy was assessed using several metrics with practical relevance in mind (justification provided in the Supplementary Information— ). Predictive performance, the predictive uncertainties and the total overconfidence were, respectively, measured with the micro-F1 score, Shannon’s Entropy II and Expected Calibration Error (ECE). F1-AUC was used to evaluate the robustness of the predictive performance and the uncertainty’s error-rate correlation. The Area between Development and Production (ADP) metric was designed to complement F1-AUC by evaluating robustness to shift-induced overconfidence . This may be interpreted as the expected predictive loss during a model’s transition from development inference (IID) to production inference (OOD) while controlling for the uncertainty threshold. Quantifying predictive uncertainty A predictive uncertainty (or total uncertainty) indicates the likelihood of an erroneous inference [12pt]{minimal} $$()= {}(())$$ p x = SoftMax f x , with a probability vector [12pt]{minimal} $$() {[]}^{K}$$ p x ∈ [ 0 , 1 ] K , normalising operator [12pt]{minimal} $${}(.)$$ SoftMax . , pointwise SoftMax function in logit-space, [12pt]{minimal} $$(.)$$ f . , and an gene expression vector [12pt]{minimal} $$ {}^{D}$$ x ∈ R D . The ideal predictive uncertainties depend on the combination of many factors including the training data [12pt]{minimal} $${}_{train}={\{({}_{i},{y}_{i})\}}_{i=1}^{n}$$ D train = x i , y i i = 1 n , model specification (e.g. model architecture, hyperparameters, etc.), inherent noise in data, model parameters [12pt]{minimal} $$$$ Θ , test data inputs [12pt]{minimal} $$ {}_{test}$$ x ∈ D test (if modelling heteroscedastic noise), and hidden confounding variables causing distribution shifts. Consequently, there are many statistics, each explaining different phenomena, which make up the predictive uncertainty. Given that some sub-divisions of uncertainty are exclusive to distribution-wise predictive models , we restricted ourselves to uncertainties that are accessible to both pointwise and distribution-wise models, namely, the confidence score, [12pt]{minimal} $$()$$ Conf ( x ) , and Shannon’s Entropy [12pt]{minimal} $$(())$$ H ( p x ) . A model’s confidence score with reference to sample x, is defined by the largest element from the SoftMax vector, [12pt]{minimal} $$()={ () }_{ },$$ Conf x = ‖ p x ‖ ∞ , where \|\|p(x)\|\| ∞ denotes the matrix-induced infinity norm of the vector p(x). Confidence scores approximately quantify the probability of being correct and thus they are often used for rejecting ‘untrustworthy’ predictions (recall ‘uncertainty thresholding’ from the Introduction). Moreover, an average conf(x) is comparable to the accuracy metric, which allows for evaluating the overconfidence via ECE, which we will shortly detail. Another notion of predictive uncertainty is that of Shannon’s Entropy, i.e., [12pt]{minimal} $$()=_{k=1}^{K}{}_{k}({}_{k})= - ,(),$$ H p = ∑ k = 1 K p k log p k = - ⟨ p , log p ⟩ , where [12pt]{minimal} $$ .,.$$ ⟨ . , . ⟩ is the dot product operator. Recall that [12pt]{minimal} $$()$$ H p is maximised when [12pt]{minimal} $$$$ p encodes a uniform distribution. Defining out-of-distribution data and the DL effects The IID assumption on data implies true causal mechanisms (i.e. structural causal model) where the underlying data generating process is immutable across observations, and hence the samples are independently generated from the same distribution . The OOD assumption, however, underpins a different setting where the underlying causal mechanisms are affected (e.g. via interventions), thus the distribution of data changes . There are many different types of distributional shifts, all of which negatively affect model performance. Deep learning models can degrade under distribution shifts as the IID assumption is necessary for most optimisation strategies (Supplementary Information— ). Furthermore, it is worth noting that the resulting overconfidence can be extreme, whereby arbitrary model predictions correspond with maximal confidence scores [12pt]{minimal} $${s}_{i} 1$$ s i → 1 (Supplementary Information— ). Evaluation in OOD using ECE The Expected Calibration Error was determined by binning each model’s confidence scores into M bins. The absolute difference between each bin’s accuracy and average maximum SoftMax score is averaged to weigh the bins proportionally with sample count. The ECE is defined as follows: [12pt]{minimal} $$={ }_{m=1}^{M}_{m}\|}{n}\|({B}_{m})-({B}_{m}))\|,$$ ECE = ∑ m = 1 M B m n acc ( B m ) - conf ( B m ) ) , where [12pt]{minimal} $${B}_{m}$$ B m is the number of predictions in bin [12pt]{minimal} $$m$$ m , [12pt]{minimal} $$n$$ n is the total number of samples, and [12pt]{minimal} $$({B}_{m})$$ acc ( B m ) and [12pt]{minimal} $$({B}_{m})$$ conf ( B m ) are the accuracy and confidence scores of bin [12pt]{minimal} $$m$$ m , respectively. Evaluation in OOD using the area under the F1-retention curve (F1-AUC) Area under the F1-Retention Curve (F1-AUC) was used to evaluate model performance in OOD, as it accounts for both predictive accuracy and an uncertainty’s error-rate correlation . High F1-AUC values result from high accuracy (reflected by vertical shifts in F1-Retention curves) and/or high uncertainty error-rate correlation (reflected by the gradient of the F1-Retention curves). An uncertainty’s error-rate correlation is important in the production (OOD) context as higher correlations imply more discarded erroneous predictions. F1-AUC was quantified according to the following method. Predictions were sorted by their descending order of uncertainty. All predictions were iterated over in order once, while at each iteration, F1 and retention (initially 100%) were calculated before replacing the current prediction with ground truth, hence decreasing the retention. The increasing F1 scores and the corresponding decreasing retention rates determined the F1-Retention curve. Approximate integration of the F1-Retention curve determined F1-AUC. F1-Retention curves and F1-AUC metrics were quantified for all models on OOD data, including samples with classes that were not seen during training. Using ADP for evaluating models in OOD data relative to IID data The Area between the Development and Production Curve (ADP) aimed to complement F1-AUC, especially in the context of deploying models from development inference (IID) to production inference (OOD). Thus, ADP was designed to capture (in OOD data, relative to IID) three aspects of a model’s robustness relating to the accuracy, uncertainty error-rate correlation, and shift-induced overconfidence. This is because benchmarked inter-model performance can reduce similarly in terms of robustness to accuracy and uncertainty’s error-rate correlation (as measured by F1-AUC), but significantly differ by their uncertainty calibration (as measured by ADP). ADP was calculated according to the following method: Development and Production F1-Uncertainty curves were produced by iteratively calculating F1 and discarding (not replacing) samples by their descending order of uncertainty. A nominal F1 target range of [12pt]{minimal} $$[(_{dev}),(_{dev})]=[0.975, 0.990]$$ min F 1 dev , max F 1 dev = 0.975 , 0.990 was selected, based on the Development F1-Uncertainty curve; with [12pt]{minimal} $$(_{dev}, {}_{accept})$$ F 1 dev , U accept denoting a point on the Development F1-Uncertainty curve at uncertainty threshold [12pt]{minimal} $${U}_{accept}$$ U accept . Nominal F1 target points, [12pt]{minimal} $$_{nominal}$$ F 1 nominal , were incremented at 1e-5 intervals from [12pt]{minimal} $$_{nom}=(_{dev})$$ F 1 nom = min ( F 1 dev ) to [12pt]{minimal} $$_{nom}=(_{dev})$$ F 1 nom = max ( F 1 dev ) , with the per cent decrease in F1, from development [12pt]{minimal} $$_{nom}$$ F 1 nom to production [12pt]{minimal} $$_{prod}$$ F 1 prod , recalculated at each step: [12pt]{minimal} $${}^{(dev prod)} (_{nom})=(_{nom}-_{prod}) 100\%.$$ Decrease d e v → p r o d ( F 1 nom ) = ( F 1 nom - F 1 prod ) × 100 % . The set of recalculated [12pt]{minimal} $${}^{(dev prod)} (_{nom})$$ Decrease d e v → p r o d ( F 1 nom ) values was averaged to approximate the Area between the Development and Production curves (ADP). The ADP may be interpreted as “the expected decrease in accuracy when transitioning from development to production if uncertainty thresholding is utilised to boost reliability”. It is important to note that our method for selecting the range [12pt]{minimal} $$[(_{}),(_{})]$$ min F 1 dev , max F 1 dev was not arbitrary and required two checks for each model’s Development F1-Uncertainty curve. The first check was to ensure the sample size corresponding to [12pt]{minimal} $$(_{})$$ max F 1 dev was sufficiently large (see Supplementary Table ). The second check was to ensure that [12pt]{minimal} $$(_{})$$ min F 1 dev was large enough to satisfy production needs. Failing to undertake these checks may result in the ADP statistic to mislead explanations about the expected loss when deploying models to production. ADP is practically relevant by relating to the uncertainty thresholding technique for improving reliability in production (recall introduction). This is because [12pt]{minimal} $${}^{(dev prod)} (_{nom})$$ Decrease d e v → p r o d ( F 1 nom ) first depends on a nominated target performance [12pt]{minimal} $$_{nom}$$ F 1 nom , which selects corresponding [12pt]{minimal} $${}_{accept}$$ U accept from the Development F1-Uncertainty Curve. Predictions with uncertainties below [12pt]{minimal} $${}_{accept}$$ U accept are accepted in production, with performance denoted by [12pt]{minimal} $$_{prod}$$ F 1 prod . As far as the authors are aware, no other metric monitors the three robustness components of accuracy, uncertainty’s error-rate correlation, and shift-induced overconfidence. Ethics approval and consent to participate This project used RNA-seq data which was previously published or is in the process of publication. The QIMR Berghofer Human Research Ethics Committee approved use of public data (P2095). The task was to predict a patient's primary cancer type, which we cast under the supervised learning framework by learning the map [12pt]{minimal} $$\{ y\}$$ x → y , with [12pt]{minimal} $$y$$ y denoting the primary cancer category, and [12pt]{minimal} $$ {}^{D}$$ x ∈ R D denoting a patient’s sampled bulk gene expression signature. Three independent datasets were used: our own independent Internal Custom Dataset, ICD – , TCGA , and Met500 . All datasets were pre-processed and partitioned into groups (i.e., strata) that uniquely proxied different distribution shifts. Proxies of approximately unique shifts were assumed to be governed by their respective intervention (i.e. unique shift), as deemed by values of four presumed hidden variables influencing the modelled map [12pt]{minimal} $$\{ y\}$$ x → y . Those variables were ‘Batch’ (indicating source dataset label, e.g., ‘ TCGA ’), ‘ State-of-Metastases ’ (valued ‘ Primary ’, or ‘ Metastatic ’), and ‘ Seen ’ (indicating whether a target value y was seen during training) (Supplementary Table ). Training and validation data comprised of the Strata ID. [12pt]{minimal} $$ {{( {'Batch',\;'State\;of\;Metastases',\;'Seen'} )}}_{Strata \;ID\;key} = {{( {'TCGA',\;'Primary',\; True} )}}_{key \;value},$$ ′ B a t c h ′ , ′ S t a t e o f M e t a s t a s e s ′ , ′ S e e n ′ ⏟ S t r a t a I D k e y = ′ T C G A ′ , ′ P r i m a r y ′ , T r u e ⏟ k e y v a l u e , since we believed it to be approximately independent and identically distributed (IID) data. All other strata were assumed out-of-distribution (OOD) due to distribution shifts caused by confounding variables. As a result, the training and validation data were IID, while the test data were OOD. Four models were benchmarked in this study—the baseline pointwise Resnet, MCD, Bilipschitz, and Ensemble. All models shared identical model architecture and hyperparameter settings (including early stopping), respectively controlling the inductive bias and accuracy from confounding overconfidence. Although we did not perform explicit hyperparameter optimisation, some manual intervention was used to adjust hyperparameters within the validation set. For example, the singular value bound hyperparameter (for spectral normalisation) was manually tuned to be as low as practically possible, while being capable of being flexible enough to learn the training task of predicting the primary site. Resnet architecture had four hidden layers, each with 1024-neurons, Mish activations , batch normalisation , and standard residual connections from the first hidden layer up to the final hidden ‘logit-space’ layer, which was then normalised using the SoftMax function to yield probability vector [12pt]{minimal} $$()= {[]}^{K}$$ p x = ∈ [ 0 , 1 ] K , where the prediction’s class index, [12pt]{minimal} $$c=\,}\{{[{{}_{1},}_{2}, , {}_{K}]}^{T}\}$$ c = arg max k p 1 , p 2 , ⋯ , p K T indicates the primary cancer site’s label [12pt]{minimal} $$y c$$ y ← c . Specifically, a batch [12pt]{minimal} $$ {}^{B D}$$ X ∈ R B × D with [12pt]{minimal} $$B$$ B individual samples is first transformed by the input layer [12pt]{minimal} $${}^{(0)}=g( , {}^{(0)} +{}^{(0)})$$ U 0 = g ( ⟨ X , W 0 ⟩ + b 0 ) , with affine transform parameters [12pt]{minimal} $$\{{}^{(0)}, {}^{(0)}\}$$ W 0 , b 0 , non-linear activations [12pt]{minimal} $$g$$ g , and output representation [12pt]{minimal} $${}^{(0)}$$ U 0 . Hidden layers have residual connections [12pt]{minimal} $${}^{(l)}=g( {}^{(l-1)},{}^{(l)} +{}^{(l)})+{}^{(l-1)}$$ U l = g ⟨ U l - 1 , W l ⟩ + b l + U ( l - 1 ) where [12pt]{minimal} $$l , ,L$$ l ∈ 1 , 2 , ⋯ , L denotes the hidden layer index ( [12pt]{minimal} $$L=3$$ L = 3 in this case). The final output layer is a pointwise (mean estimate) function in logit-space [12pt]{minimal} $$()= g( {}^{(L)},{}^{( )} +{}^{( )})$$ f X = g ⟨ U L , W μ ⟩ + b μ , where [12pt]{minimal} $$\{{}^{( )}, {}^{( )}\}$$ W μ , b μ are the final output (affine) transformation parameters. Finally, SoftMax normalisation yields a K-vector [12pt]{minimal} $$()= {}(())$$ p X = SoftMax f X . All other hyperparameter settings are defined in Supplementary Table . This baseline Resnet model architecture was inherited by all other models in this study to control inductive biases. Bayesian inference may yield a predictive distribution about sample [12pt]{minimal} $${}^{}$$ x ∗ , [12pt]{minimal} $$p(\|{}^{},)$$ p ( p \| x ∗ , D ) , from the likelihood of an assumed parametric model [12pt]{minimal} $$p(\|{}^{},)$$ p ( p \| x ∗ , Θ ) , an (approximate) parametric posterior [12pt]{minimal} $$q( \|)$$ q Θ \| D , and potentially Monte Carlo Integration (MCI) technique, also referred to as Bayesian model averaging: [12pt]{minimal} $$p(\|{}^{},) {_{}}p(\|{}^{}, )q( \|)d _{t=1}^{T}p(\|{}^{},{ }_{t})$$ p p \| x ∗ , D ≈ ∫ Θ p p \| x ∗ , Θ q Θ \| D d Θ ≈ 1 T ∑ t = 1 T p ( p \| x ∗ , Θ t ) Most neural networks are parametric models, which assume [12pt]{minimal} $$$$ Θ can perfectly represent [12pt]{minimal} $$$$ D . As a result, the model likelihood [12pt]{minimal} $$p(\|{}^{},,)$$ p ( p \| x ∗ , D , Θ ) is often replaced with [12pt]{minimal} $$(\|{}^{}, )$$ p ( p \| x ∗ , Θ ) . The main differentiating factor among all Bayesian deep learning inference methods lies in how the parametric posterior [12pt]{minimal} $$q( \|)$$ q Θ \| D is approximated. The MCD model approximates the parametric posterior [12pt]{minimal} $$q( \|)$$ q ( Θ \| D ) by keeping dropout activated during inference . Dropout randomly ‘switches off’ a subset of neurons to zero-vectors at each iteration. Hence, a collection of dropout configurations [12pt]{minimal} $${\{{ }_{t}\}}_{t=1}^{T}$$ Θ t t = 1 T are samples from the (approximate) posterior [12pt]{minimal} $$q( \|)$$ q ( Θ \| D ) . For more information, refer to the Appendix of where an approximate dual connection between Monte Carlo Dropout neural networks and Deep Gaussian processes is established. The MCD also extends the Resnet model architecture by including an additional output layer to estimate a data-dependent variance function [12pt]{minimal} $${}_{t}^{2}()= g( {}^{(L)},{}_{t}^{( )} +{}_{t}^{( )})$$ s t 2 X = g ( ⟨ U ( L ) , W t ( Σ ) ⟩ + b t ( Σ ) ) in addition to the (now stochastic) mean function [12pt]{minimal} $${}_{t}()= g( {}^{(L)},{}_{t}^{( )} +{}_{t}^{( )})$$ f t X = g ⟨ U L , W t ( μ ) ⟩ + b t ( μ ) . Both final output layers had a shared input [12pt]{minimal} $${}^{(L)}$$ U ( L ) , but unique parameters [12pt]{minimal} $$\{{}_{t}^{( )},{}_{t}^{( )}\}$$ W t ( μ ) , b t ( μ ) and [12pt]{minimal} $$\{{}_{t}^{( )},{}_{t}^{( )}\}$$ W t ( Σ ) , b t ( Σ ) . Together, the stochastic mean [12pt]{minimal} $${}_{t}()$$ f t X and variance [12pt]{minimal} $${}_{t}^{2}()$$ s t 2 X specify a Gaussian distribution in the logit-space, which was then sampled once [12pt]{minimal} $${}_{t}() ( ={}_{t}(), ={}_{t}^{2}{()}^{T}$$ u t X ∼ N ( μ = f t X , Σ = s t 2 X T I and normalised with the SoftMax function [12pt]{minimal} $${}_{t}()= {}({}_{t}())$$ p t X = SoftMax u t X . [12pt]{minimal} $${}_{t}()$$ p t X represents a single sample from the model likelihood [12pt]{minimal} $$p(\|, )$$ p ( p \| x , Θ ) , from which [12pt]{minimal} $$T$$ T samples are averaged for Monte Carlo integration: [12pt]{minimal} $$()= _{t=1}^{T}{}_{t}().$$ p X = 1 T ∑ t = 1 T p t ( X ) . Finally, [12pt]{minimal} $$()$$ p X estimates the cancer primary site label [12pt]{minimal} $$y$$ y , the predictive uncertainties [12pt]{minimal} $$$$ Conf , and [12pt]{minimal} $$()$$ H . for each individual sample in data batch [12pt]{minimal} $$$$ x . The Bilipschitz model shared all the properties of the MCD model with an additional bi-Lipschitz constraint: [12pt]{minimal} $${L}_{1}{ {}_{1}-{}_{2} }_{ } { ({}_{1})- ({}_{2}) }_{} {L}_{2}{ {}_{1}-{}_{2} }_{ }$$ L 1 ‖ x 1 - x 2 ‖ X ≤ ‖ f x 1 - f x 2 ‖ F ≤ L 2 ‖ x 1 - x 2 ‖ X where scalars [12pt]{minimal} $${L}_{1}$$ L 1 and [12pt]{minimal} $${L}_{2}$$ L 2 respectively control the tightness of the lower- and upper-bound. Norm operators [12pt]{minimal} $$\{{ }_{ },{ }_{}\}$$ ‖ . ‖ X , ‖ . ‖ F are over the data space [12pt]{minimal} $$$$ X and function space [12pt]{minimal} $$$$ F . The effect of the bi-Lipschitz constraint is such that the changes in input data [12pt]{minimal} $${ {}_{1}-{}_{2} }_{ }$$ ‖ x 1 - x 2 ‖ χ (e.g. distribution shifts) are proportional to the changes in the output, [12pt]{minimal} $${ ({}_{1})- ({}_{2}) }_{}$$ ‖ f x 1 - f x 2 ‖ F . These changes are within a bound determined by [12pt]{minimal} $${L}_{1}$$ L 1 (controlling sensitivity) and [12pt]{minimal} $${L}_{2}$$ L 2 (controlling smoothness). Interestingly, recent studies have established that bi-Lipschitz constraints are beneficial to the robustness of the neural network under distributional shifts , . Sensitivity (i.e. [12pt]{minimal} $${L}_{1}$$ L 1 ) is controlled with residual connections , , which allows [12pt]{minimal} $$()$$ f x to avoid arbitrarily small changes, especially in the presence of distributional shifts in those regions of [12pt]{minimal} $$$$ X with no (training data) support . Sensitivity (i.e. [12pt]{minimal} $${L}_{2}$$ L 2 ) is controlled with spectral normalisation on parameters [12pt]{minimal} $$$$ Θ , and batch-normalisation functions , which allow [12pt]{minimal} $$()$$ f x to avoid arbitrarily large changes (under shifts) that induce feature collapse and extreme overconfidence – . The Ensemble model was a collection of eight independently trained Bilipschitz models with unique initial parameter configurations. Each Bayesian model in the Ensemble model is sampled [12pt]{minimal} $$T/10(=25)$$ T / 10 ( = 25 ) times and then pooled to control for Monte Carlo integration between the ‘Ensemble’ and all other models. Models in deep ensembles yield similarly performant (low-loss) solutions, but are diverse and distant in parameter- and function-space . This allows the ensemble to have an (approximate) posterior [12pt]{minimal} $$q( \|)$$ q Θ \| D with multiple modes , which was not the case for the Resnet, MCD, and Bilipschitz models. We believe the ensemble modelled [12pt]{minimal} $$q( \|)$$ q Θ \| D with the highest fidelity to the true parametric posterior [12pt]{minimal} $$p( \|)$$ p Θ \| D due to empirical evidence from other studies' results , , , . Model efficacy was assessed using several metrics with practical relevance in mind (justification provided in the Supplementary Information— ). Predictive performance, the predictive uncertainties and the total overconfidence were, respectively, measured with the micro-F1 score, Shannon’s Entropy II and Expected Calibration Error (ECE). F1-AUC was used to evaluate the robustness of the predictive performance and the uncertainty’s error-rate correlation. The Area between Development and Production (ADP) metric was designed to complement F1-AUC by evaluating robustness to shift-induced overconfidence . This may be interpreted as the expected predictive loss during a model’s transition from development inference (IID) to production inference (OOD) while controlling for the uncertainty threshold. A predictive uncertainty (or total uncertainty) indicates the likelihood of an erroneous inference [12pt]{minimal} $$()= {}(())$$ p x = SoftMax f x , with a probability vector [12pt]{minimal} $$() {[]}^{K}$$ p x ∈ [ 0 , 1 ] K , normalising operator [12pt]{minimal} $${}(.)$$ SoftMax . , pointwise SoftMax function in logit-space, [12pt]{minimal} $$(.)$$ f . , and an gene expression vector [12pt]{minimal} $$ {}^{D}$$ x ∈ R D . The ideal predictive uncertainties depend on the combination of many factors including the training data [12pt]{minimal} $${}_{train}={\{({}_{i},{y}_{i})\}}_{i=1}^{n}$$ D train = x i , y i i = 1 n , model specification (e.g. model architecture, hyperparameters, etc.), inherent noise in data, model parameters [12pt]{minimal} $$$$ Θ , test data inputs [12pt]{minimal} $$ {}_{test}$$ x ∈ D test (if modelling heteroscedastic noise), and hidden confounding variables causing distribution shifts. Consequently, there are many statistics, each explaining different phenomena, which make up the predictive uncertainty. Given that some sub-divisions of uncertainty are exclusive to distribution-wise predictive models , we restricted ourselves to uncertainties that are accessible to both pointwise and distribution-wise models, namely, the confidence score, [12pt]{minimal} $$()$$ Conf ( x ) , and Shannon’s Entropy [12pt]{minimal} $$(())$$ H ( p x ) . A model’s confidence score with reference to sample x, is defined by the largest element from the SoftMax vector, [12pt]{minimal} $$()={ () }_{ },$$ Conf x = ‖ p x ‖ ∞ , where \|\|p(x)\|\| ∞ denotes the matrix-induced infinity norm of the vector p(x). Confidence scores approximately quantify the probability of being correct and thus they are often used for rejecting ‘untrustworthy’ predictions (recall ‘uncertainty thresholding’ from the Introduction). Moreover, an average conf(x) is comparable to the accuracy metric, which allows for evaluating the overconfidence via ECE, which we will shortly detail. Another notion of predictive uncertainty is that of Shannon’s Entropy, i.e., [12pt]{minimal} $$()=_{k=1}^{K}{}_{k}({}_{k})= - ,(),$$ H p = ∑ k = 1 K p k log p k = - ⟨ p , log p ⟩ , where [12pt]{minimal} $$ .,.$$ ⟨ . , . ⟩ is the dot product operator. Recall that [12pt]{minimal} $$()$$ H p is maximised when [12pt]{minimal} $$$$ p encodes a uniform distribution. The IID assumption on data implies true causal mechanisms (i.e. structural causal model) where the underlying data generating process is immutable across observations, and hence the samples are independently generated from the same distribution . The OOD assumption, however, underpins a different setting where the underlying causal mechanisms are affected (e.g. via interventions), thus the distribution of data changes . There are many different types of distributional shifts, all of which negatively affect model performance. Deep learning models can degrade under distribution shifts as the IID assumption is necessary for most optimisation strategies (Supplementary Information— ). Furthermore, it is worth noting that the resulting overconfidence can be extreme, whereby arbitrary model predictions correspond with maximal confidence scores [12pt]{minimal} $${s}_{i} 1$$ s i → 1 (Supplementary Information— ). The Expected Calibration Error was determined by binning each model’s confidence scores into M bins. The absolute difference between each bin’s accuracy and average maximum SoftMax score is averaged to weigh the bins proportionally with sample count. The ECE is defined as follows: [12pt]{minimal} $$={ }_{m=1}^{M}_{m}\|}{n}\|({B}_{m})-({B}_{m}))\|,$$ ECE = ∑ m = 1 M B m n acc ( B m ) - conf ( B m ) ) , where [12pt]{minimal} $${B}_{m}$$ B m is the number of predictions in bin [12pt]{minimal} $$m$$ m , [12pt]{minimal} $$n$$ n is the total number of samples, and [12pt]{minimal} $$({B}_{m})$$ acc ( B m ) and [12pt]{minimal} $$({B}_{m})$$ conf ( B m ) are the accuracy and confidence scores of bin [12pt]{minimal} $$m$$ m , respectively. Area under the F1-Retention Curve (F1-AUC) was used to evaluate model performance in OOD, as it accounts for both predictive accuracy and an uncertainty’s error-rate correlation . High F1-AUC values result from high accuracy (reflected by vertical shifts in F1-Retention curves) and/or high uncertainty error-rate correlation (reflected by the gradient of the F1-Retention curves). An uncertainty’s error-rate correlation is important in the production (OOD) context as higher correlations imply more discarded erroneous predictions. F1-AUC was quantified according to the following method. Predictions were sorted by their descending order of uncertainty. All predictions were iterated over in order once, while at each iteration, F1 and retention (initially 100%) were calculated before replacing the current prediction with ground truth, hence decreasing the retention. The increasing F1 scores and the corresponding decreasing retention rates determined the F1-Retention curve. Approximate integration of the F1-Retention curve determined F1-AUC. F1-Retention curves and F1-AUC metrics were quantified for all models on OOD data, including samples with classes that were not seen during training. The Area between the Development and Production Curve (ADP) aimed to complement F1-AUC, especially in the context of deploying models from development inference (IID) to production inference (OOD). Thus, ADP was designed to capture (in OOD data, relative to IID) three aspects of a model’s robustness relating to the accuracy, uncertainty error-rate correlation, and shift-induced overconfidence. This is because benchmarked inter-model performance can reduce similarly in terms of robustness to accuracy and uncertainty’s error-rate correlation (as measured by F1-AUC), but significantly differ by their uncertainty calibration (as measured by ADP). ADP was calculated according to the following method: Development and Production F1-Uncertainty curves were produced by iteratively calculating F1 and discarding (not replacing) samples by their descending order of uncertainty. A nominal F1 target range of [12pt]{minimal} $$[(_{dev}),(_{dev})]=[0.975, 0.990]$$ min F 1 dev , max F 1 dev = 0.975 , 0.990 was selected, based on the Development F1-Uncertainty curve; with [12pt]{minimal} $$(_{dev}, {}_{accept})$$ F 1 dev , U accept denoting a point on the Development F1-Uncertainty curve at uncertainty threshold [12pt]{minimal} $${U}_{accept}$$ U accept . Nominal F1 target points, [12pt]{minimal} $$_{nominal}$$ F 1 nominal , were incremented at 1e-5 intervals from [12pt]{minimal} $$_{nom}=(_{dev})$$ F 1 nom = min ( F 1 dev ) to [12pt]{minimal} $$_{nom}=(_{dev})$$ F 1 nom = max ( F 1 dev ) , with the per cent decrease in F1, from development [12pt]{minimal} $$_{nom}$$ F 1 nom to production [12pt]{minimal} $$_{prod}$$ F 1 prod , recalculated at each step: [12pt]{minimal} $${}^{(dev prod)} (_{nom})=(_{nom}-_{prod}) 100\%.$$ Decrease d e v → p r o d ( F 1 nom ) = ( F 1 nom - F 1 prod ) × 100 % . The set of recalculated [12pt]{minimal} $${}^{(dev prod)} (_{nom})$$ Decrease d e v → p r o d ( F 1 nom ) values was averaged to approximate the Area between the Development and Production curves (ADP). The ADP may be interpreted as “the expected decrease in accuracy when transitioning from development to production if uncertainty thresholding is utilised to boost reliability”. It is important to note that our method for selecting the range [12pt]{minimal} $$[(_{}),(_{})]$$ min F 1 dev , max F 1 dev was not arbitrary and required two checks for each model’s Development F1-Uncertainty curve. The first check was to ensure the sample size corresponding to [12pt]{minimal} $$(_{})$$ max F 1 dev was sufficiently large (see Supplementary Table ). The second check was to ensure that [12pt]{minimal} $$(_{})$$ min F 1 dev was large enough to satisfy production needs. Failing to undertake these checks may result in the ADP statistic to mislead explanations about the expected loss when deploying models to production. ADP is practically relevant by relating to the uncertainty thresholding technique for improving reliability in production (recall introduction). This is because [12pt]{minimal} $${}^{(dev prod)} (_{nom})$$ Decrease d e v → p r o d ( F 1 nom ) first depends on a nominated target performance [12pt]{minimal} $$_{nom}$$ F 1 nom , which selects corresponding [12pt]{minimal} $${}_{accept}$$ U accept from the Development F1-Uncertainty Curve. Predictions with uncertainties below [12pt]{minimal} $${}_{accept}$$ U accept are accepted in production, with performance denoted by [12pt]{minimal} $$_{prod}$$ F 1 prod . As far as the authors are aware, no other metric monitors the three robustness components of accuracy, uncertainty’s error-rate correlation, and shift-induced overconfidence. This project used RNA-seq data which was previously published or is in the process of publication. The QIMR Berghofer Human Research Ethics Committee approved use of public data (P2095). Supplementary Information. Supplementary Table 1. Supplementary Table 4.
The effect on incisional hernia of absorbable barbed suture for midline fascial closure in minimally invasive surgery for colorectal and gastric cancers: study protocol for a randomized controlled trial	9f7c7a89-bf65-4302-8d89-cb1a86642bfc	10164296	Suturing[mh]	Background and rationale {6a, 6b} Laparoscopic and robotic surgery has become a preferred treatment option for colorectal and gastric cancers rather than open surgery because of its minimally invasive nature and short-term operative outcomes [ – ]. During minimally invasive gastrointestinal surgery, mini-laparotomy is used for extracorporeal anastomosis, specimen extraction, or single-port surgery. Midline incision is widely used for mini-laparotomy , because it is easy to perform, results in minimal blood loss, provides a better surgical view, and is more convenient for closure than other types of incision . However, a midline incision carries a high incidence of incisional hernia . Incisional hernia is a commonly encountered complication after abdominal surgery, with a reported incidence of 10–32% . Incisional hernia diminishes a patient’s quality of life due to pain and discomfort . Although minimally invasive surgery with a smaller incision and limited abdominal wall trauma is expected to reduce incisional hernia, the incidence of incisional hernia of this approach did not differ from that of open surgery . Several methods of abdominal closure have been reported to prevent incisional hernia after abdominal surgery. A running technique with long-lasting monofilament suture material is recommended to prevent incisional hernia . However, monofilament suture has the disadvantage of being easily loosening and requiring a surgical knot, resulting in a potential risk of incisional hernia . Barbed suture is a new type of suture material, first patented in 1964. It consists of standard monofilament suture with tiny barbs along its length that are designed to anchor the suture without the need for knots . These new materials have become popular for skin and uterine closures during cesarean section, orthopedic surgery, and gastrointestinal surgery due to reduced tissue trauma and convenience [ – ]. In animal experiments, barbed sutures for fascia closure showed adequate tensile strength . However, they are yet to be clinically adopted owing to insufficient evidence of their safety and efficacy as compared to those of conventional suture materials. Therefore, we propose a randomized controlled clinical trial (Barbed trial) to compare the outcomes of absorbable barbed sutures and conventional absorbable monofilament sutures for midline fascia closure in minimally invasive surgery for colorectal and gastric cancers. Objective {7} This study aimed to evaluate the safety and efficacy of absorbable barbed sutures for midline fascia closure in minimally invasive surgery for colorectal and gastric cancers in comparison with conventional absorbable monofilament sutures. Study design {8} The Barbed trial is a prospective, superiority, single-center, randomized trial comparing the clinical outcomes of absorbable barbed sutures and monofilament sutures. A total of 312 patients will be enrolled and randomly allocated to use absorbable barbed or monofilament sutures for midline fascia closure in a 1:1 ratio. The surgical procedure used to close the midline fascia was standardized for both types of suture material. The investigator will examine the patients until discharge and at 6, 12, 18, 24, and 36 months postoperatively (Fig. ). This study is ongoing at Jeonbuk National University Hospital and will last 3 years for each patient. This trial was registered with the Clinical Research Information Service of Korea. Laparoscopic and robotic surgery has become a preferred treatment option for colorectal and gastric cancers rather than open surgery because of its minimally invasive nature and short-term operative outcomes [ – ]. During minimally invasive gastrointestinal surgery, mini-laparotomy is used for extracorporeal anastomosis, specimen extraction, or single-port surgery. Midline incision is widely used for mini-laparotomy , because it is easy to perform, results in minimal blood loss, provides a better surgical view, and is more convenient for closure than other types of incision . However, a midline incision carries a high incidence of incisional hernia . Incisional hernia is a commonly encountered complication after abdominal surgery, with a reported incidence of 10–32% . Incisional hernia diminishes a patient’s quality of life due to pain and discomfort . Although minimally invasive surgery with a smaller incision and limited abdominal wall trauma is expected to reduce incisional hernia, the incidence of incisional hernia of this approach did not differ from that of open surgery . Several methods of abdominal closure have been reported to prevent incisional hernia after abdominal surgery. A running technique with long-lasting monofilament suture material is recommended to prevent incisional hernia . However, monofilament suture has the disadvantage of being easily loosening and requiring a surgical knot, resulting in a potential risk of incisional hernia . Barbed suture is a new type of suture material, first patented in 1964. It consists of standard monofilament suture with tiny barbs along its length that are designed to anchor the suture without the need for knots . These new materials have become popular for skin and uterine closures during cesarean section, orthopedic surgery, and gastrointestinal surgery due to reduced tissue trauma and convenience [ – ]. In animal experiments, barbed sutures for fascia closure showed adequate tensile strength . However, they are yet to be clinically adopted owing to insufficient evidence of their safety and efficacy as compared to those of conventional suture materials. Therefore, we propose a randomized controlled clinical trial (Barbed trial) to compare the outcomes of absorbable barbed sutures and conventional absorbable monofilament sutures for midline fascia closure in minimally invasive surgery for colorectal and gastric cancers. This study aimed to evaluate the safety and efficacy of absorbable barbed sutures for midline fascia closure in minimally invasive surgery for colorectal and gastric cancers in comparison with conventional absorbable monofilament sutures. The Barbed trial is a prospective, superiority, single-center, randomized trial comparing the clinical outcomes of absorbable barbed sutures and monofilament sutures. A total of 312 patients will be enrolled and randomly allocated to use absorbable barbed or monofilament sutures for midline fascia closure in a 1:1 ratio. The surgical procedure used to close the midline fascia was standardized for both types of suture material. The investigator will examine the patients until discharge and at 6, 12, 18, 24, and 36 months postoperatively (Fig. ). This study is ongoing at Jeonbuk National University Hospital and will last 3 years for each patient. This trial was registered with the Clinical Research Information Service of Korea. Study setting {9} This study will enroll participants who are scheduled to undergo minimally invasive gastric or colorectal cancer surgery at Jeonbuk National University Hospital. This study was approved and supported by the Institutional Review Board of Jeonbuk National University Hospital (CUH 2021-09-040). Eligibility criteria {10} Inclusion criteria The inclusion criteria were as follows: (1) patients who were diagnosed with colorectal or gastric adenocarcinoma, (2) patients who underwent laparoscopic and robotic radical surgery, and (3) patients who were older than 20 years of age. Exclusion criteria Patients would be excluded if they fulfill any of the following exclusion criteria: Patients who underwent open surgery Patients with possible distant metastasis in preoperative studies Patients who underwent past abdominal surgery Patients who had midline incision of less than 1 cm or longer than 10 cm Patients who had incision which is not located at the midline Patients who underwent emergency operation, palliative surgery, or ileostomy or colostomy formation Patients with cancer-related complications (perforation or abscess) Patients who participated in another clinical trial within the past 6 months Who will take informed consent? {26a} The trial details will be explained in full to potential participants by investigators, and an informed consent form will be provided. Adequate time will be given to participants to consider their decision regarding trial participation. Subsequently, participants can sign the informed consent and they can withdraw at any time during the trial. Additional consent provisions for collection and use of participant data and biological specimens {26b} Not applicable as no participant data and biological specimens were collected in this study. Interventions Intervention description {11a} Surgical procedures are performed under general anesthesia and by an experienced surgeon. To achieve standardization of suture technique, we provide surgeons with formal video of fascial closure using two suture materials. The location of the midline incision, which is supra-, trans-, or infra-umbilical mini-laparotomy, is determined according to the surgeon’s preference. At the end of surgery, fascia closure is achieved with absorbable barbed sutures (Stratafix®, Ethicon Inc., USA) for the study group and absorbable monofilament sutures (Maxon®, Covidien Inc., USA) for the control group. In all cases, the suture-to-wound length ratio is at least 4:1 for closure of the midline incision, and the inter-suture spacing is less than 1 cm. We avoid mass closure, which is performed with a suture bite, including all layers of the abdominal wall, except the skin. A single aponeurotic closure is applied to the midline incision in both groups. Subcutaneous tissue closure is not mandatory in this study. The skin can be approximated by using staples. Criteria for discontinuing or modifying allocated interventions {11b} There are no predetermined criteria for discontinuing or modifying the intervention assigned to participants. All individuals are participating on a voluntary basis, and they have the option to withdraw from the study at any time for any reason without facing any negative consequences. Strategies to improve adherence to interventions {11c} We will recruit patients who are scheduled to undergo surgery for gastric or colorectal cancer. By enrolling participants who require regular cancer screening, we anticipate that adherence to the study protocol will improve. Furthermore, participants will be informed of the significance of completing follow-up assessments. Relevant concomitant care permitted or prohibited during the trial {11d} All participants will receive standard postoperative management after surgery for gastric or colorectal cancer. For patients with wound infection, commercial dressing preparations or negative pressure wound therapy will be permitted. Provisions for post-trial care {30} There will be no provision for patients who participate in the trial. Post-trial care will follow the standards of care after gastric and colorectal cancer surgery. Outcomes {12, 18a} Primary outcome The primary outcome of this trial is the frequency of incisional hernia 3 years after surgery. We use the definition of incisional hernia according to the European Hernia Society: “any abdominal wall gap with or without bulge in the area of a postoperative scar perceptible or palpable by clinical examination or imaging” . All patients will be examined using computed tomography to evaluate cancer recurrence during the study period. Therefore, the investigator will examine the occurrence of incisional hernia through physical examination and computed tomography scans at 6, 12, 18, 24, and 36 months after surgery in the outpatient clinic. Secondary outcomes Postoperative complications, including surgical site infection (SSI), pain, and quality of life, will be compared between the absorbable barbed suture and conventional absorbable suture groups. Postoperative complications is defined as any adverse events that required additional pharmacological, interventional, or surgical management within 30 days after surgery. All postoperative complications were graded according to the Clavien-Dindo classification . SSI is defined according to the criteria of the Centers for Disease Control and Prevention . We divided SSI into superficial, deep incisional, and organ/space-related types occurring within 30 days after surgery. We will evaluate postoperative pain on postoperative days 1, 2, and 3 using a numeral rating scale. The verbal numeral rating scale ranged from 0 to 10 (0 = no pain and 10 = worst pain). To analyze the quality of life, the SF-36 and body image questionnaire will be used and documented by the patient 12 months postoperatively (Fig. ). All outcomes will be evaluated by the investigators and they are specially trained for reliability. Participant timeline {13} The participant timeline is shown in Fig. . Sample size {14} We mainly have used absorbable monofilament sutures for fascia closure, and the overall incidence of incisional hernia was about 15%. Barbed suture began to be used for fascial closure in August 2020 during colorectal surgery. Before this randomized controlled trial, we conducted a pilot study of barbed suture for fascia closure. A total of 60 patients underwent fascia closure with barbed suture during laparoscopic colorectal surgery, and incisional hernia occurred in 2 patients (3.3%) within 1 year after surgery. Assuming additional patients with an incisional hernia during follow-up period, we determined possible incisional hernia rate to 5% in the barbed suture group for 3 years after surgery. Therefore, we hypothesized that the incidence of incisional hernia at 3 years after surgery will be lower in the barbed suture group than in the monofilament suture group. The sample size was calculated on the basis of this analysis plan, assuming that the incisional hernia rate of the barbed suture and monofilament groups at 3 years after surgery will be 5% and 15%, respectively. We calculated that 141 patients per group needed to achieve a two-sided α of 0.05 and statistical power of 80% using a chi-squared test. Assuming a 10% dropout rate, it is necessary to recruit 312 patients. Finally, 156 patients were assigned to the barbed suture and monofilament groups. The calculations were performed using GPower 3.1. Recruitment {15} Patients who are scheduled to undergo minimally invasive gastric or colorectal cancer surgery will be considered as candidates for trial enrolment in concordance with the inclusion and exclusion criteria. Informed consent will be obtained from all participants by investigators. We are confident that we can enroll a sufficient number of eligible patients, as our institution conducts over 150 surgeries for gastric cancer and 300 surgeries for colorectal cancer annually. This study will enroll participants who are scheduled to undergo minimally invasive gastric or colorectal cancer surgery at Jeonbuk National University Hospital. This study was approved and supported by the Institutional Review Board of Jeonbuk National University Hospital (CUH 2021-09-040). Inclusion criteria The inclusion criteria were as follows: (1) patients who were diagnosed with colorectal or gastric adenocarcinoma, (2) patients who underwent laparoscopic and robotic radical surgery, and (3) patients who were older than 20 years of age. Exclusion criteria Patients would be excluded if they fulfill any of the following exclusion criteria: Patients who underwent open surgery Patients with possible distant metastasis in preoperative studies Patients who underwent past abdominal surgery Patients who had midline incision of less than 1 cm or longer than 10 cm Patients who had incision which is not located at the midline Patients who underwent emergency operation, palliative surgery, or ileostomy or colostomy formation Patients with cancer-related complications (perforation or abscess) Patients who participated in another clinical trial within the past 6 months The inclusion criteria were as follows: (1) patients who were diagnosed with colorectal or gastric adenocarcinoma, (2) patients who underwent laparoscopic and robotic radical surgery, and (3) patients who were older than 20 years of age. Patients would be excluded if they fulfill any of the following exclusion criteria: Patients who underwent open surgery Patients with possible distant metastasis in preoperative studies Patients who underwent past abdominal surgery Patients who had midline incision of less than 1 cm or longer than 10 cm Patients who had incision which is not located at the midline Patients who underwent emergency operation, palliative surgery, or ileostomy or colostomy formation Patients with cancer-related complications (perforation or abscess) Patients who participated in another clinical trial within the past 6 months The trial details will be explained in full to potential participants by investigators, and an informed consent form will be provided. Adequate time will be given to participants to consider their decision regarding trial participation. Subsequently, participants can sign the informed consent and they can withdraw at any time during the trial. Not applicable as no participant data and biological specimens were collected in this study. Intervention description {11a} Surgical procedures are performed under general anesthesia and by an experienced surgeon. To achieve standardization of suture technique, we provide surgeons with formal video of fascial closure using two suture materials. The location of the midline incision, which is supra-, trans-, or infra-umbilical mini-laparotomy, is determined according to the surgeon’s preference. At the end of surgery, fascia closure is achieved with absorbable barbed sutures (Stratafix®, Ethicon Inc., USA) for the study group and absorbable monofilament sutures (Maxon®, Covidien Inc., USA) for the control group. In all cases, the suture-to-wound length ratio is at least 4:1 for closure of the midline incision, and the inter-suture spacing is less than 1 cm. We avoid mass closure, which is performed with a suture bite, including all layers of the abdominal wall, except the skin. A single aponeurotic closure is applied to the midline incision in both groups. Subcutaneous tissue closure is not mandatory in this study. The skin can be approximated by using staples. Criteria for discontinuing or modifying allocated interventions {11b} There are no predetermined criteria for discontinuing or modifying the intervention assigned to participants. All individuals are participating on a voluntary basis, and they have the option to withdraw from the study at any time for any reason without facing any negative consequences. Strategies to improve adherence to interventions {11c} We will recruit patients who are scheduled to undergo surgery for gastric or colorectal cancer. By enrolling participants who require regular cancer screening, we anticipate that adherence to the study protocol will improve. Furthermore, participants will be informed of the significance of completing follow-up assessments. Relevant concomitant care permitted or prohibited during the trial {11d} All participants will receive standard postoperative management after surgery for gastric or colorectal cancer. For patients with wound infection, commercial dressing preparations or negative pressure wound therapy will be permitted. Provisions for post-trial care {30} There will be no provision for patients who participate in the trial. Post-trial care will follow the standards of care after gastric and colorectal cancer surgery. Surgical procedures are performed under general anesthesia and by an experienced surgeon. To achieve standardization of suture technique, we provide surgeons with formal video of fascial closure using two suture materials. The location of the midline incision, which is supra-, trans-, or infra-umbilical mini-laparotomy, is determined according to the surgeon’s preference. At the end of surgery, fascia closure is achieved with absorbable barbed sutures (Stratafix®, Ethicon Inc., USA) for the study group and absorbable monofilament sutures (Maxon®, Covidien Inc., USA) for the control group. In all cases, the suture-to-wound length ratio is at least 4:1 for closure of the midline incision, and the inter-suture spacing is less than 1 cm. We avoid mass closure, which is performed with a suture bite, including all layers of the abdominal wall, except the skin. A single aponeurotic closure is applied to the midline incision in both groups. Subcutaneous tissue closure is not mandatory in this study. The skin can be approximated by using staples. There are no predetermined criteria for discontinuing or modifying the intervention assigned to participants. All individuals are participating on a voluntary basis, and they have the option to withdraw from the study at any time for any reason without facing any negative consequences. We will recruit patients who are scheduled to undergo surgery for gastric or colorectal cancer. By enrolling participants who require regular cancer screening, we anticipate that adherence to the study protocol will improve. Furthermore, participants will be informed of the significance of completing follow-up assessments. All participants will receive standard postoperative management after surgery for gastric or colorectal cancer. For patients with wound infection, commercial dressing preparations or negative pressure wound therapy will be permitted. There will be no provision for patients who participate in the trial. Post-trial care will follow the standards of care after gastric and colorectal cancer surgery. Primary outcome The primary outcome of this trial is the frequency of incisional hernia 3 years after surgery. We use the definition of incisional hernia according to the European Hernia Society: “any abdominal wall gap with or without bulge in the area of a postoperative scar perceptible or palpable by clinical examination or imaging” . All patients will be examined using computed tomography to evaluate cancer recurrence during the study period. Therefore, the investigator will examine the occurrence of incisional hernia through physical examination and computed tomography scans at 6, 12, 18, 24, and 36 months after surgery in the outpatient clinic. Secondary outcomes Postoperative complications, including surgical site infection (SSI), pain, and quality of life, will be compared between the absorbable barbed suture and conventional absorbable suture groups. Postoperative complications is defined as any adverse events that required additional pharmacological, interventional, or surgical management within 30 days after surgery. All postoperative complications were graded according to the Clavien-Dindo classification . SSI is defined according to the criteria of the Centers for Disease Control and Prevention . We divided SSI into superficial, deep incisional, and organ/space-related types occurring within 30 days after surgery. We will evaluate postoperative pain on postoperative days 1, 2, and 3 using a numeral rating scale. The verbal numeral rating scale ranged from 0 to 10 (0 = no pain and 10 = worst pain). To analyze the quality of life, the SF-36 and body image questionnaire will be used and documented by the patient 12 months postoperatively (Fig. ). All outcomes will be evaluated by the investigators and they are specially trained for reliability. The primary outcome of this trial is the frequency of incisional hernia 3 years after surgery. We use the definition of incisional hernia according to the European Hernia Society: “any abdominal wall gap with or without bulge in the area of a postoperative scar perceptible or palpable by clinical examination or imaging” . All patients will be examined using computed tomography to evaluate cancer recurrence during the study period. Therefore, the investigator will examine the occurrence of incisional hernia through physical examination and computed tomography scans at 6, 12, 18, 24, and 36 months after surgery in the outpatient clinic. Postoperative complications, including surgical site infection (SSI), pain, and quality of life, will be compared between the absorbable barbed suture and conventional absorbable suture groups. Postoperative complications is defined as any adverse events that required additional pharmacological, interventional, or surgical management within 30 days after surgery. All postoperative complications were graded according to the Clavien-Dindo classification . SSI is defined according to the criteria of the Centers for Disease Control and Prevention . We divided SSI into superficial, deep incisional, and organ/space-related types occurring within 30 days after surgery. We will evaluate postoperative pain on postoperative days 1, 2, and 3 using a numeral rating scale. The verbal numeral rating scale ranged from 0 to 10 (0 = no pain and 10 = worst pain). To analyze the quality of life, the SF-36 and body image questionnaire will be used and documented by the patient 12 months postoperatively (Fig. ). All outcomes will be evaluated by the investigators and they are specially trained for reliability. The participant timeline is shown in Fig. . We mainly have used absorbable monofilament sutures for fascia closure, and the overall incidence of incisional hernia was about 15%. Barbed suture began to be used for fascial closure in August 2020 during colorectal surgery. Before this randomized controlled trial, we conducted a pilot study of barbed suture for fascia closure. A total of 60 patients underwent fascia closure with barbed suture during laparoscopic colorectal surgery, and incisional hernia occurred in 2 patients (3.3%) within 1 year after surgery. Assuming additional patients with an incisional hernia during follow-up period, we determined possible incisional hernia rate to 5% in the barbed suture group for 3 years after surgery. Therefore, we hypothesized that the incidence of incisional hernia at 3 years after surgery will be lower in the barbed suture group than in the monofilament suture group. The sample size was calculated on the basis of this analysis plan, assuming that the incisional hernia rate of the barbed suture and monofilament groups at 3 years after surgery will be 5% and 15%, respectively. We calculated that 141 patients per group needed to achieve a two-sided α of 0.05 and statistical power of 80% using a chi-squared test. Assuming a 10% dropout rate, it is necessary to recruit 312 patients. Finally, 156 patients were assigned to the barbed suture and monofilament groups. The calculations were performed using GPower 3.1. Patients who are scheduled to undergo minimally invasive gastric or colorectal cancer surgery will be considered as candidates for trial enrolment in concordance with the inclusion and exclusion criteria. Informed consent will be obtained from all participants by investigators. We are confident that we can enroll a sufficient number of eligible patients, as our institution conducts over 150 surgeries for gastric cancer and 300 surgeries for colorectal cancer annually. Sequence generation {16a} Eligible patients will be randomly allocated to use absorbable barbed or monofilament sutures for midline fascia closure in a 1:1 ratio. Stratified randomization was performed by medical statisticians who do not participate in this study. This randomization is based on an independent, computer-based sequence generated from the implementation of the dynamic algorithm with the cancer types as the stratifying variables. Concealment mechanism {16b} To achieve randomization, a computer-generated sequence methodology will be used, which will ensure that both the randomization methodology and the allocation sequence are concealed from both the investigator and the participants. Implementation {16c} An independent medical statistician will generate a stratified randomization list for allocation sequence. Investigators will enroll and assign participants to interventions. Eligible patients will be randomly allocated to use absorbable barbed or monofilament sutures for midline fascia closure in a 1:1 ratio. Stratified randomization was performed by medical statisticians who do not participate in this study. This randomization is based on an independent, computer-based sequence generated from the implementation of the dynamic algorithm with the cancer types as the stratifying variables. To achieve randomization, a computer-generated sequence methodology will be used, which will ensure that both the randomization methodology and the allocation sequence are concealed from both the investigator and the participants. An independent medical statistician will generate a stratified randomization list for allocation sequence. Investigators will enroll and assign participants to interventions. Who will be blinded {17a} After the interventions are assigned, participants will be blinded. The results will be kept confidential from the participants throughout the entire trial. Surgeons will not be blinded because the intervention involves a surgical procedure. Procedure for unblinding if needed {17b} As investigators are aware of the participant allocation, unblinding would not be required. After the interventions are assigned, participants will be blinded. The results will be kept confidential from the participants throughout the entire trial. Surgeons will not be blinded because the intervention involves a surgical procedure. As investigators are aware of the participant allocation, unblinding would not be required. Plans to promote participant retention and complete follow-up {18b} Participants are patients with gastric or colorectal cancer who require regular follow-up to evaluate cancer recurrence. We expect that participants will be available for follow-up due to concerns of cancer recurrence. Data management {19} The study data will be collected and recorded electronically, and stored securely on computers and hard drives. Only the investigators will have access to the participants’ files, and they will be the only ones authorized to make changes to the information. This will ensure that the data remains accurate and validated throughout the study. Confidentiality {27} Datasets will be anonymized, and only summary data, which cannot identify individual participants, will be presented in the manuscript. Plans for collection, laboratory evaluation, and storage of biological specimens for genetic or molecular analysis in this trial/future use {33} Not applicable as no biological specimens were collected in this study. Participants are patients with gastric or colorectal cancer who require regular follow-up to evaluate cancer recurrence. We expect that participants will be available for follow-up due to concerns of cancer recurrence. The study data will be collected and recorded electronically, and stored securely on computers and hard drives. Only the investigators will have access to the participants’ files, and they will be the only ones authorized to make changes to the information. This will ensure that the data remains accurate and validated throughout the study. Datasets will be anonymized, and only summary data, which cannot identify individual participants, will be presented in the manuscript. Not applicable as no biological specimens were collected in this study. Statistical methods for primary and secondary outcomes {20a} A planned analysis will be performed after all patients have completed their 3-year follow-up. Categorical variables will be presented as numbers with percentages, and continuous variables will be presented as means with standard deviations or medians with interquartile ranges. The analysis will be performed using the intention-to-treat approach. The primary outcome will be analyzed using chi-squared and Kaplan-Meier tests. Risk factors for incisional hernia will be analyzed using the Cox regression analysis. Secondary outcomes will be tested using the chi-squared test for categorical variables and the t -test or Mann-Whitney test for continuous variables, as appropriate. The level of statistical significance will be set at p < 0.05. Interim analyses {21b} We have no plans to conduct any interim analyses, since both interventions have low risk. Method for additional analyses (e.g., subgroup analyses) {20b} Subgroup analyses will be performed on the type of cancer (gastric or colorectal cancer). Values of p < 0.05 will be considered statistically significant. Methods in the analysis to handle protocol nonadherence and any statistical methods to handle missing data {20c} Loss to follow-up will be minimized due to the characteristics of cancer patients. If missing data is exceed 5% of any variable, multiple imputations will be used to handle missing values. Plans to give access to the full protocol, participant-level data, and statistical code {31c} The full protocol and final datasets will be available from the corresponding author on reasonable request. A planned analysis will be performed after all patients have completed their 3-year follow-up. Categorical variables will be presented as numbers with percentages, and continuous variables will be presented as means with standard deviations or medians with interquartile ranges. The analysis will be performed using the intention-to-treat approach. The primary outcome will be analyzed using chi-squared and Kaplan-Meier tests. Risk factors for incisional hernia will be analyzed using the Cox regression analysis. Secondary outcomes will be tested using the chi-squared test for categorical variables and the t -test or Mann-Whitney test for continuous variables, as appropriate. The level of statistical significance will be set at p < 0.05. We have no plans to conduct any interim analyses, since both interventions have low risk. Subgroup analyses will be performed on the type of cancer (gastric or colorectal cancer). Values of p < 0.05 will be considered statistically significant. Loss to follow-up will be minimized due to the characteristics of cancer patients. If missing data is exceed 5% of any variable, multiple imputations will be used to handle missing values. The full protocol and final datasets will be available from the corresponding author on reasonable request. Composition of the coordinating center and trial steering committee {5d} The authors will coordinate and steer this study. Composition of the data monitoring committee, its role and reporting structure {21a} We do not have composition of the data monitoring committee. Adverse event reporting and harms {22} Any adverse events that required additional pharmacological, interventional, or surgical management within 30 days after surgery will be collected and monitored. We regarded adverse events as postoperative complications and graded according to the Clavien-Dindo classification . Adverse events will be treated according to the standards of care. Frequency and plans for auditing trial conduct {23} The study data will be maintained in accordance with Good Clinical Practice requirements by the investigators. The original study data and information will be securely stored for a minimum of 5 years following the completion of the study. Data monitoring reports will be submitted to the ethics committee every 3 months. Plans for communicating important protocol amendments to relevant parties (e.g., trial participants, ethical committees) {25} If the protocol needs to be modified, it will be reviewed again by the ethics committee, and, upon approval, the trial registry and protocol will be updated. Dissemination {31a} The findings will be published in peer-reviewed journals and disseminated through scientific and academic conferences. The authors will coordinate and steer this study. We do not have composition of the data monitoring committee. Any adverse events that required additional pharmacological, interventional, or surgical management within 30 days after surgery will be collected and monitored. We regarded adverse events as postoperative complications and graded according to the Clavien-Dindo classification . Adverse events will be treated according to the standards of care. The study data will be maintained in accordance with Good Clinical Practice requirements by the investigators. The original study data and information will be securely stored for a minimum of 5 years following the completion of the study. Data monitoring reports will be submitted to the ethics committee every 3 months. If the protocol needs to be modified, it will be reviewed again by the ethics committee, and, upon approval, the trial registry and protocol will be updated. The findings will be published in peer-reviewed journals and disseminated through scientific and academic conferences. This is the first randomized controlled trial to compare absorbable barbed sutures with absorbable monofilament sutures for midline fascia closure in minimally invasive surgery for colorectal and gastric cancers. The current guidelines recommend monofilament suture material for continuous closure of the midline incision . However, absorbable monofilaments can stretch up to 30% of their length and slip easily, loosening the suture material . At the end of the monofilament suture, it is necessary to create a surgical knot that can break and cause an incisional hernia. Barbed sutures were introduced several years ago. This novel suture material has a self-anchorage system that maintains tension and requires no surgical knots after the sutures are strained . Although barbed sutures have been applied in uterine closure, arthroscopic surgery, and gastrointestinal surgery, they are not widely used for fascia closure due to the lack of evidence of its safety and efficacy. In this trial, we aimed to determine which suture material is safer for patients who require mini-laparotomy in minimally invasive surgery. The strength of this study was its prospective and randomized design. If absorbable barbed sutures demonstrate comparable results to those of monofilament sutures, they might be suggested as an alternative suture material for abdominal fascia closure. The study protocol was registered on December 23, 2021. The first patient was recruited on December 24, 2021. Recruitment is expected to end in December 2023.
Dynamics of Functional Genes and Bacterial Community during Bioremediation of Diesel-Contaminated Soil Amended with Compost	2eb50299-a583-4833-b7d5-d1a385117a39	10164733	Microbiology[mh]	Total petroleum hydrocarbons (TPHs) are major components of petrochemicals and are used in various industrial fields . Upon leakage during storage, transportation, and use, TPHs become notable sources of soil and groundwater contamination [ - ]. Hence, bioremediation is attracting attention as an eco-friendly and economical technology for the remediation of TPH-contaminated soil [ - ]. This process has the advantage that secondary pollution due to by-products hardly occurs because the TPHs are decomposed into carbon dioxide and water by the microorganisms [ - ]. Nevertheless, the metabolic activities of the microorganisms in the TPH-contaminated soil tend to be low as the TPHs are mainly composed of recalcitrant carbon- and hydrogen-containing compounds, such as long-chain aliphatic and aromatic hydrocarbons . Bioremediation performance can therefore be improved by using chemical or organic additives to supply nitrogen, phosphorus, and growth factors for the soil microorganisms . Compost is widely used as an organic additive to improve the bioremediation performance of TPH-contaminated soil [ - ]. Compost can provide not only nutrients such as nitrogen, phosphorus, and other minerals, but also microbial sources [ - ]. Compost amendment changes the soil microbial activity and microbial community related to the metabolic pathways of carbon and nitrogen compounds. Therefore, compost amendment can influence CH 4 and N 2 O emissions as well as the removal of TPHs during bioremediation. In one study, the bioremediation efficiency of crude oil-contaminated soil was improved by more than 29 times by the addition of compost . Moreover, when compost was added to diesel-contaminated soil, the abundance of the alkB gene, a functional gene contributing to alkane decomposition, increased . In another study, the addition of compost to diesel-contaminated soil not only enhanced the dominance of the methane-oxidizing microorganism community, but also increased the abundance of the pmoA gene, a functional gene contributing to methane oxidation, relative to that of soil treated with chemical nutrients . Lee et al . reported that when compost was added to diesel-contaminated soil planted with maize, the abundances of the cnorB and nosZ genes, functional genes involved in NO and N 2 O reduction, increased compared to the control in the absence of compost. In spite of the above-mentioned studies, there remains a lack of information on the effects of compost amendment upon the dynamics of the microbial community, and the functional genes associated with the degradation of TPHs and the emission of CH 4 and N 2 O, during the bioremediation of TPH-contaminated soil. Therefore, the present study evaluates the TPH removal, soil enzyme activity, CH 4 -oxidizing, and N 2 O-reducing potentials of TPH-contaminated soils amended with various ratios (0–20%, w/w) of compost. During the bioremediation period, the functional genes associated with TPH degradation (namely, alkB and CYP153 ), CH 4 production ( mcrA ), CH 4 oxidation ( pmoA ), N 2 O production ( cnorB , qnorB ), and N 2 O reduction ( nosZI ) are determined via the quantitative polymerase chain reaction (qPCR). Moreover, the dynamics of the soil bacterial community are characterized via a high-throughput sequencing method (Miseq, Illumina Inc., USA). Finally, based on correlation and network analyses, the key parameters that are highly affected by compost amendment in the bioremediation of TPH-contaminated soil are clarified. Materials and Soil Preparation Coarse sand (2 mm in diameter, Kimhae Masato, Korea) and perlite (Kyungdong One Co. Ltd., Korea) were mixed in a 4:1 ratio (v/v) to generate barren soil. Various amounts of compost (0, 5, 10, and 20 wt.%) were added to the barren soil to produce mixtures designated as S-C0 (control), S-C5, S-C10, and S-C20, respectively. The compost was prepared by a commercial vendor (Seokgang Green Fertilizer Inc., Korea) via the fermentation of swine manure (40%), sawdust (29%), cow manure (10%), and bacterial inoculum (1%). After purchase, the soil samples were artificially contaminated with diesel (10,000 mg diesel/kg soil). The physicochemical properties of the barren soil and compost were analyzed by the National Instrumentation Center for Environmental Management (NICEM), Republic of Korea ( ). The textures of the barren soil and compost were classified as sandy loam and sand, respectively. The moisture and organic contents of the barren soil were 2.9 and 0.2%, respectively. The total nitrogen, ammonium nitrogen, and nitrate nitrogen concentrations of the compost were 26,000, 465.3, and 34.5 mg N/kg soil, respectively. The total phosphate content of the compost was significantly high (11,443.8 mg P/kg soil). The moisture and organic contents of the compost were 55.3 and 36.7%, respectively. Pot Experiment and Soil Sampling The pot-scale experiment was conducted for 103 days (from June 3 rd to September 14 th , 2021) in the rooftop garden of the new engineering building at Ewha Womans University (37° 56¢ 65¢¢ N, 126° 94¢ 85¢¢ E). The processes of soil sample preparation and pot setting are shown photographically in of the Supplementary Material. After setting a layer of coarse sand up to 1 cm from the bottom of each pot (18 cm in diameter × 15 cm in height), each soil mixture amended with 0–20 wt.% compost was added into its own pot so that the soil layer was 14 cm in height ( ). The lower parts of the pots were then buried in the garden to avoid the intense solar radiation ( ). Each pot experiment was conducted in duplicate. The pot soil was watered 3 times a week to maintain an average moisture content of 14.58%, and the soil mixture in the pot was manually mixed every other day using a trowel. Soil sampling was performed on days 0, 12, 33, 51, 76, and 103 according to the United States Environmental Protection Agency (US EPA) method, as modified by Hu et al . . The soil in the pot was mixed well before sampling, and 200 g of soil was randomly collected in a polyethylene bag. After further thorough mixing, 50 g of the collected soil was placed in each of two conical tubes, one of which was then stored at –23°C for subsequent DNA extraction, and the other stored at –80°C for measuring the residual TPH concentrations. The remaining soil sample in the polyethylene bag was stored at 4°C for use in analyzing the moisture content, organic content, pH, water holding capacity (WHC), and microbial enzyme activity. The analyses of moisture content, organic content, and pH were performed within 4 h of soil sampling, and the microbial enzyme activity was measured within a week. Analysis of Soil Properties and Residual Diesel Concentrations The moisture contents of 3-g soil samples were measured by drying each sample at 110°C for 4 h, and the organic contents of 3-g samples were measured by drying at 550°C for 2 h. The moisture and organic contents were then calculated from the weight differences before and after heating . To measure the pH of the soil, a 5-g sample was mixed with distilled water (25 ml), and then kept at room temperature for 1 h . The pH of the soil suspension was then measured using a pH meter (Orion 420A, Thermo Scientific Inc., Japan). The WHC of the soil was analyzed using the method described by Vengadaramana and Jashothan . All experiments were performed in triplicate. After freeze-drying the soil samples according to the method of Lee et al . , their residual diesel concentrations were measured by first thoroughly mixing the 3-g samples with 1:1 (v/v) hexane:acetone solution (5 ml) in a 15-ml test tube and shaking for 30 min at 30°C to extract the residual diesel. Then, an organic-phase aliquot (1.5 ml) was transferred to a 2-ml vial, and a 1-μl sample of this aliquot was analyzed using a gas chromatograph (6890N, Agilent Technologies, CA, USA) . The residual diesel analysis was performed in 5 replicates. The removal efficiency of diesel was calculated according to Eq. (1): RE (%) = ( C 0 − C i ) / C 0 × 100 , (1) where C 0 is the initial concentration, and C i is the concentration on the i th day. As a result of diesel extraction using the same method for the soil without the addition of diesel (negative control), the diesel concentration was below the detection limit. Soil Enzyme Activity To evaluate the effect of compost amendment on the soil enzyme activity during the bioremediation of diesel-contaminated soil, the dehydrogenase activity (DHA) and urease activity (UA) were measured. For measuring the DHA, 2 g of the soil was mixed with 2 ml of Tris-HCL buffer (pH 7.6) and 1 ml of 1% (w/v) triphenyl tetrazolium chloride (TTC) solution in a test tube, and incubated in the dark at 37°C for 24 h. Then, 96% ethanol (10 ml) was added, and the mixture was shaken and transferred to a 15-ml centrifuge tube for centrifugation at 3,000 g for 5 min. The concentration of the produced triphenyl formazan (TPF) in the supernatant was then measured using a spectrophotometer (Libra S22, Biochrom, UK) at 485 nm. The DHA was then determined as the amount of TPF (in mg) produced per 1 gram of dry soil per hour (mg TPF/g dry soil/h) . For measuring the UA, the soil sample (2 g) was mixed with 0.08 M urea solution (0.5 ml) and incubated for 2 h at 37°C in a standing incubator. Then, 1 M KCl/HCl solution (10 ml) was added, and the mixture was stirred at 180 rpm and 25°C for 30 min. The mixture was then transferred to a 15-ml centrifuge tube and centrifuged at 3,000 g for 5 min. Next, the supernatant (3 ml) was mixed gently with distilled water (1.2 ml), 12% sodium phenolate solution (0.5 ml), 0.15% sodium nitroprusside solution (50 μl), and 1% NaOCl solution (0.25 ml), and allowed to stand at room temperature for 30 min. After that, a 1 ml aliquot was examined using a Libra S22 spectrophotometer (Biochrom) at 630 nm, and the UA was determined as the amount of ammonium (in μg) produced per 1 gram of dry soil per hour (μg NH 4 + -N/g dry soil/h) . Evaluation of Soil CH 4 Oxidation and N 2 O Reduction Potentials The effects of compost amendment upon the CH 4 oxidation and N 2 O reduction potentials of diesel-contaminated soil during bioremediation were investigated using the same methods as in previous studies [ , , ]. Thus, the soil sample (2 g) was inoculated into nitrate minimal salt medium (NMS, 6 ml) in a 120-ml serum bottle . The bottle was then sealed with a butyl rubber cap, and 50,000 ppm (v/v) of CH 4 gas (99.99%, Dong-A Specialty Gases Co., Korea) was injected into the headspace of the bottle. While incubating the bottle at 30°C and 150 rpm, the CH 4 concentration in the headspace was measured using a gas chromatograph (7890A, Agilent Technologies, USA) . To evaluate the soil N 2 O reduction potential, the soil sample (5 g) was inoculated into mineral medium (30 ml) in a 600-ml serum bottle and sealed with a butyl rubber cap . After purging with N 2 gas (99%, Dong-A Specialty Gases), the glucose and acetate solution (100 mg COD/L) was added into the bottle, followed by the injection of 1,000 ppm (v/v) of N 2 O gas (99.99%, Dong-A Specialty Gases) into the headspace. While incubating the bottle at 30°C and 150 rpm, the N 2 O concentration in the headspace was measured using a gas chromatograph (7890B, Agilent Technologies) . Functional Gene Abundance Analysis The qPCR was performed to evaluate the change in functional gene abundances during the bioremediation of diesel-contaminated soil amended with various ratios of compost. For this procedure, DNA was extracted from the soil sample using a NucleoSpin Soil Kit (Macherey-Nagel GmbH, Germany) . The 16S rRNA gene was quantitatively evaluated for total bacteria abundances using 340F and 805R primer sets . The alkane monooxygenase ( alkB ) and CYP153 gene abundances associated with diesel degradation were quantified using alkB -1F/ alkB -1R and CYP153 _C4_F/ CYP153 _C4_R primer sets, respectively. The abundances of the particulate methane monooxygenase ( pmoA ) gene associated with CH 4 oxidation, and of the methane reductase subunit A ( mcrA ) gene involved in CH 4 production, were measured using A189f/mb661r and mlas/ mcrA -rev primer sets, respectively. The abundances of the nitric oxide reductase genes qnorB , cnorB , and nosZI , that are associated with N 2 O production and reduction, were quantified using the qnorB -2F/ qnorB -7R , norB1-F/norB6-R , and nosZ 1F/ nosZ 1R primer sets, respectively. The solution compositions and reaction conditions for the qPCR are listed in . Bacterial Community Analysis The extracted DNA was used as a PCR template to analyze the bacterial community with an Illumina Miseq sequencing platform (Macrogen Inc., Korea) by the method reported previously . The PCR was performed to amplify the 16S ribosomal RNA gene containing the V4 region using the 515F/806R primer set . The sequences were analyzed via the Illumina MiSeq sequencing platform purchased from Macrogen Inc. The sequence reads were analyzed using the QIIME software version 1.9 by Macrogen Inc. . Sequences that were too short to contain the target base pairs were removed by using the Fast Length Adjustment of Short reads (FLASH) software version 1.2.11 . Ambiguous and chimeric sequences were then removed, and the remaining sequences were classified into operational taxonomic units (OTUs) at 97% similarity using the CD-HIT-OTU program . The taxonomy for each OTU was assigned based on the National Center for Biotechnology Information (NCBI) 16S microbial database. The obtained sequencing reads were deposited to the NCBI Sequence Read Archive ( http://www.ncbi.nlm.nih.gov/ ) under accession number SRP339855. Finally, the Chao1, Shannon, and Simpson indices were calculated using the QIIME software version 1.9. Statistical Analysis The t -tests and multiple comparisons were conducted using the R software package ( www.rstudio.com ), with a p-value of 0.05 indicating a significant difference. The bacterial community structures were compared by correlation analysis and principal component analysis (PCA) using the R software package and the CANOCO 4.5 software (Microcomputer Power, Ithaca, NY, USA) . The heatmap for the bacterial community was visualized using the gplots tool in the R software package. The Pearson correlations between parameters were also calculated using the R software package. The extended local similarity analysis (eLSA) was performed at p < 0.05 to evaluate the correlation between parameters, and the results were visualized using the Cytoscape program version 3.4.0 (Institute for Systems Biology, USA). Coarse sand (2 mm in diameter, Kimhae Masato, Korea) and perlite (Kyungdong One Co. Ltd., Korea) were mixed in a 4:1 ratio (v/v) to generate barren soil. Various amounts of compost (0, 5, 10, and 20 wt.%) were added to the barren soil to produce mixtures designated as S-C0 (control), S-C5, S-C10, and S-C20, respectively. The compost was prepared by a commercial vendor (Seokgang Green Fertilizer Inc., Korea) via the fermentation of swine manure (40%), sawdust (29%), cow manure (10%), and bacterial inoculum (1%). After purchase, the soil samples were artificially contaminated with diesel (10,000 mg diesel/kg soil). The physicochemical properties of the barren soil and compost were analyzed by the National Instrumentation Center for Environmental Management (NICEM), Republic of Korea ( ). The textures of the barren soil and compost were classified as sandy loam and sand, respectively. The moisture and organic contents of the barren soil were 2.9 and 0.2%, respectively. The total nitrogen, ammonium nitrogen, and nitrate nitrogen concentrations of the compost were 26,000, 465.3, and 34.5 mg N/kg soil, respectively. The total phosphate content of the compost was significantly high (11,443.8 mg P/kg soil). The moisture and organic contents of the compost were 55.3 and 36.7%, respectively. The pot-scale experiment was conducted for 103 days (from June 3 rd to September 14 th , 2021) in the rooftop garden of the new engineering building at Ewha Womans University (37° 56¢ 65¢¢ N, 126° 94¢ 85¢¢ E). The processes of soil sample preparation and pot setting are shown photographically in of the Supplementary Material. After setting a layer of coarse sand up to 1 cm from the bottom of each pot (18 cm in diameter × 15 cm in height), each soil mixture amended with 0–20 wt.% compost was added into its own pot so that the soil layer was 14 cm in height ( ). The lower parts of the pots were then buried in the garden to avoid the intense solar radiation ( ). Each pot experiment was conducted in duplicate. The pot soil was watered 3 times a week to maintain an average moisture content of 14.58%, and the soil mixture in the pot was manually mixed every other day using a trowel. Soil sampling was performed on days 0, 12, 33, 51, 76, and 103 according to the United States Environmental Protection Agency (US EPA) method, as modified by Hu et al . . The soil in the pot was mixed well before sampling, and 200 g of soil was randomly collected in a polyethylene bag. After further thorough mixing, 50 g of the collected soil was placed in each of two conical tubes, one of which was then stored at –23°C for subsequent DNA extraction, and the other stored at –80°C for measuring the residual TPH concentrations. The remaining soil sample in the polyethylene bag was stored at 4°C for use in analyzing the moisture content, organic content, pH, water holding capacity (WHC), and microbial enzyme activity. The analyses of moisture content, organic content, and pH were performed within 4 h of soil sampling, and the microbial enzyme activity was measured within a week. The moisture contents of 3-g soil samples were measured by drying each sample at 110°C for 4 h, and the organic contents of 3-g samples were measured by drying at 550°C for 2 h. The moisture and organic contents were then calculated from the weight differences before and after heating . To measure the pH of the soil, a 5-g sample was mixed with distilled water (25 ml), and then kept at room temperature for 1 h . The pH of the soil suspension was then measured using a pH meter (Orion 420A, Thermo Scientific Inc., Japan). The WHC of the soil was analyzed using the method described by Vengadaramana and Jashothan . All experiments were performed in triplicate. After freeze-drying the soil samples according to the method of Lee et al . , their residual diesel concentrations were measured by first thoroughly mixing the 3-g samples with 1:1 (v/v) hexane:acetone solution (5 ml) in a 15-ml test tube and shaking for 30 min at 30°C to extract the residual diesel. Then, an organic-phase aliquot (1.5 ml) was transferred to a 2-ml vial, and a 1-μl sample of this aliquot was analyzed using a gas chromatograph (6890N, Agilent Technologies, CA, USA) . The residual diesel analysis was performed in 5 replicates. The removal efficiency of diesel was calculated according to Eq. (1): RE (%) = ( C 0 − C i ) / C 0 × 100 , (1) where C 0 is the initial concentration, and C i is the concentration on the i th day. As a result of diesel extraction using the same method for the soil without the addition of diesel (negative control), the diesel concentration was below the detection limit. To evaluate the effect of compost amendment on the soil enzyme activity during the bioremediation of diesel-contaminated soil, the dehydrogenase activity (DHA) and urease activity (UA) were measured. For measuring the DHA, 2 g of the soil was mixed with 2 ml of Tris-HCL buffer (pH 7.6) and 1 ml of 1% (w/v) triphenyl tetrazolium chloride (TTC) solution in a test tube, and incubated in the dark at 37°C for 24 h. Then, 96% ethanol (10 ml) was added, and the mixture was shaken and transferred to a 15-ml centrifuge tube for centrifugation at 3,000 g for 5 min. The concentration of the produced triphenyl formazan (TPF) in the supernatant was then measured using a spectrophotometer (Libra S22, Biochrom, UK) at 485 nm. The DHA was then determined as the amount of TPF (in mg) produced per 1 gram of dry soil per hour (mg TPF/g dry soil/h) . For measuring the UA, the soil sample (2 g) was mixed with 0.08 M urea solution (0.5 ml) and incubated for 2 h at 37°C in a standing incubator. Then, 1 M KCl/HCl solution (10 ml) was added, and the mixture was stirred at 180 rpm and 25°C for 30 min. The mixture was then transferred to a 15-ml centrifuge tube and centrifuged at 3,000 g for 5 min. Next, the supernatant (3 ml) was mixed gently with distilled water (1.2 ml), 12% sodium phenolate solution (0.5 ml), 0.15% sodium nitroprusside solution (50 μl), and 1% NaOCl solution (0.25 ml), and allowed to stand at room temperature for 30 min. After that, a 1 ml aliquot was examined using a Libra S22 spectrophotometer (Biochrom) at 630 nm, and the UA was determined as the amount of ammonium (in μg) produced per 1 gram of dry soil per hour (μg NH 4 + -N/g dry soil/h) . 4 Oxidation and N 2 O Reduction Potentials The effects of compost amendment upon the CH 4 oxidation and N 2 O reduction potentials of diesel-contaminated soil during bioremediation were investigated using the same methods as in previous studies [ , , ]. Thus, the soil sample (2 g) was inoculated into nitrate minimal salt medium (NMS, 6 ml) in a 120-ml serum bottle . The bottle was then sealed with a butyl rubber cap, and 50,000 ppm (v/v) of CH 4 gas (99.99%, Dong-A Specialty Gases Co., Korea) was injected into the headspace of the bottle. While incubating the bottle at 30°C and 150 rpm, the CH 4 concentration in the headspace was measured using a gas chromatograph (7890A, Agilent Technologies, USA) . To evaluate the soil N 2 O reduction potential, the soil sample (5 g) was inoculated into mineral medium (30 ml) in a 600-ml serum bottle and sealed with a butyl rubber cap . After purging with N 2 gas (99%, Dong-A Specialty Gases), the glucose and acetate solution (100 mg COD/L) was added into the bottle, followed by the injection of 1,000 ppm (v/v) of N 2 O gas (99.99%, Dong-A Specialty Gases) into the headspace. While incubating the bottle at 30°C and 150 rpm, the N 2 O concentration in the headspace was measured using a gas chromatograph (7890B, Agilent Technologies) . The qPCR was performed to evaluate the change in functional gene abundances during the bioremediation of diesel-contaminated soil amended with various ratios of compost. For this procedure, DNA was extracted from the soil sample using a NucleoSpin Soil Kit (Macherey-Nagel GmbH, Germany) . The 16S rRNA gene was quantitatively evaluated for total bacteria abundances using 340F and 805R primer sets . The alkane monooxygenase ( alkB ) and CYP153 gene abundances associated with diesel degradation were quantified using alkB -1F/ alkB -1R and CYP153 _C4_F/ CYP153 _C4_R primer sets, respectively. The abundances of the particulate methane monooxygenase ( pmoA ) gene associated with CH 4 oxidation, and of the methane reductase subunit A ( mcrA ) gene involved in CH 4 production, were measured using A189f/mb661r and mlas/ mcrA -rev primer sets, respectively. The abundances of the nitric oxide reductase genes qnorB , cnorB , and nosZI , that are associated with N 2 O production and reduction, were quantified using the qnorB -2F/ qnorB -7R , norB1-F/norB6-R , and nosZ 1F/ nosZ 1R primer sets, respectively. The solution compositions and reaction conditions for the qPCR are listed in . The extracted DNA was used as a PCR template to analyze the bacterial community with an Illumina Miseq sequencing platform (Macrogen Inc., Korea) by the method reported previously . The PCR was performed to amplify the 16S ribosomal RNA gene containing the V4 region using the 515F/806R primer set . The sequences were analyzed via the Illumina MiSeq sequencing platform purchased from Macrogen Inc. The sequence reads were analyzed using the QIIME software version 1.9 by Macrogen Inc. . Sequences that were too short to contain the target base pairs were removed by using the Fast Length Adjustment of Short reads (FLASH) software version 1.2.11 . Ambiguous and chimeric sequences were then removed, and the remaining sequences were classified into operational taxonomic units (OTUs) at 97% similarity using the CD-HIT-OTU program . The taxonomy for each OTU was assigned based on the National Center for Biotechnology Information (NCBI) 16S microbial database. The obtained sequencing reads were deposited to the NCBI Sequence Read Archive ( http://www.ncbi.nlm.nih.gov/ ) under accession number SRP339855. Finally, the Chao1, Shannon, and Simpson indices were calculated using the QIIME software version 1.9. The t -tests and multiple comparisons were conducted using the R software package ( www.rstudio.com ), with a p-value of 0.05 indicating a significant difference. The bacterial community structures were compared by correlation analysis and principal component analysis (PCA) using the R software package and the CANOCO 4.5 software (Microcomputer Power, Ithaca, NY, USA) . The heatmap for the bacterial community was visualized using the gplots tool in the R software package. The Pearson correlations between parameters were also calculated using the R software package. The extended local similarity analysis (eLSA) was performed at p < 0.05 to evaluate the correlation between parameters, and the results were visualized using the Cytoscape program version 3.4.0 (Institute for Systems Biology, USA). Physicochemical Parameters and Diesel Removal Efficiency The variations in ambient temperature, precipitation, soil organic matter content, and soil pH with time during the pot experiments are presented in . The ambient temperature and precipitation information were obtained from the Korea Meteorological Administration. The average ambient temperature ranged from 13.3 to 30.2°C, and the maximum and minimum temperatures were 35.4 and 8.5°C, respectively ( ). Among the total of 52 rainfall events, there were 3 intensive rainfalls (> 90 mm) between days 50 and 69 ( ). The average organic contents of the soil samples increased with increasing compost addition, and were 0.64, 2.50, 3.19, and 6.07% for the S-C0, S-C5, S-C10, and S-C20 samples, respectively. These values did not change significantly during the experimental period ( ). Meanwhile, the average pH values of the soil samples decreased with increasing amounts of added compost, being 9.01, 8.10, 7.80, and 7.76 for the S-C0, S-C5, S-C10, and S-C20 samples, respectively, and did not significantly vary during the experiment ( ). The changes in the residual diesel concentrations of the various samples with time are presented in . Thus, the initial TPH concentration was 9,432 mg TPH/kg dry soil, and did not change significantly until day 12 in all samples. Thereafter, the residual diesel concentration decreased significantly, and the diesel removal rate was proportional to the amount of compost added. On day 103, the diesel removal efficiencies of the S-C0, S-C5, S-C10, and S-C20 samples were 54.6, 77.5, 80.7, and 85.7%, respectively. Notably, the residual diesel concentration of the S-C20 sample on day 76 was below the 2,000 mg TPH/kg soil pollution risk criterion for oil-contaminated soil in Korea. Soil Enzyme Activity Dehydrogenase is known to be involved in the initial decomposition of soil organics, catalyzing the removal of hydrogen from organic molecules; hence, the dehydrogenase activity (DHA) is used as an index for evaluating the degradation activity of soil organics . The results in indicate that the DHA of the S-C0 did not significantly change during the initial 33 days, but increased slightly to 19.5 μg TPF/g dry soil/h on day 103. In the soils amended with compost, the initial DHA increased with increasing amounts of added compost, being 203.8, 333.6, and 462.2 μg TPF/g dry soil/h in the S-C5, S-C10, and S-C20, respectively ( ). During the experimental period, the DHAs of the amended soils decreased gradually as the residual diesel concentration decreased ( and ). Urease promotes the mineralization of organic nitrogen to hydrogen-bound nitrogen, thereby providing the soil microorganisms with ammonia as an available nitrogen source . Although the urease activity (UA) cannot explain all of the biological mechanisms, it can be used as a good indicator of TPH metabolism in the soil under various soil conditions [ - ]. As with the DHA, the UA of the S-C0 sample did not change significantly during the early stages of the experiment, but increased slightly during the mid-late period ( ). In the amended soils, the initial UA increased with increasing amounts of added compost, and further increased with time until the 33rd day, decreasing gradually thereafter ( ). CH 4 Oxidation and N 2 O Reduction Potentials The results in and indicate that there was no significant difference in the initial CH 4 oxidation potential of the various soil samples, which ranged from 1.40 to 1.95 μmol/g dry soil/h. During the experiment, however, the CH 4 oxidation potential of the S-C0 increased significantly until around day 33, and remained relatively constant thereafter. Except on day 12, the CH 4 oxidation potential of the soils amended with compost (7.00–9.83 μmol/g dry soil/h) were higher than that of the non-compost-amended soil (5.39–5.80 μmol/g dry soil/h), and continued to increase significantly with time up until at least day 51. Further, the CH 4 oxidation potential of the S-C20 sample (8.49–9.83 μmol/g dry soil/h) was slightly higher than those of the S-C5 and S-C10 samples (7.00–8.21 μmol/g dry soil/h). The results in and indicate that the initial N 2 O reduction potential of the non-compost amended soil (S-C0) was insignificant (< 56.82 nmol/g dry soil/h), while those of the S-C5, S-C10, and S-C20 were 868.03, 1,399.57 and 1,757.76 nmol/g dry soil/h, respectively. Moreover, while the N 2 O reduction potential gradually decreased with time during bioremediation, a relatively high activity was maintained when the amount of compost added was large. In the S-C20 sample, the N 2 O reduction potential decreased from 838.14 nmol/g dry soil/h on day 12, to 224.08 nmol/g dry soil/h on day 103. In the S-C5 sample, it decreased from 328.57 nmol/g dry soil/h on day 12, to 140.37 nmol/g dry soil/h on day 103. Functional Gene Dynamics The functional gene dynamics during bioremediation of various diesel-contaminated soil samples are indicated in . Thus, the 16S rRNA gene copy number of the S-C0 sample increased from 10 3 to 10 5 /g dry soil, while those of the S-C5, S-C10, and S-C20 were maintained at around 10 6 /g dry soil during bioremediation ( ). Meanwhile, the relative copy numbers of the alkB gene in the S-C0 increased with bioremediation time to match that of the compost-amended soils (gene copy number = 10 5 /g dry soil) at day 51, and remained constant thereafter ( ). The relative alkB gene copy numbers of the S-C5, S-C10, and S-C20 samples also varied during the initial period (0–33 days), but did not vary significantly after day 51. However, while the relative CYP153 gene copy number in the S-C0 increased from 10 2 to 10 4 /g dry soil during bioremediation, those of the S-C5, S-C10, and S-C20 samples decreased from 10 6 to 10 4 /g dry soil ( ). Ultimately, on day 103, the CYP153 gene copy numbers were similar in all soil samples regardless of compost addition. The pmoA /16S rRNA ratio of the S-C0 was always higher than that of the S-C5, S-C10, and S-C20 during bioremediation ( ). The mcrA /16S rRNA ratio of the S-C0 sample increased from 10 1 to 10 3 , whereas that of the S-C5, S-C10, and S-C20 samples increased from 10 2 to 10 6 during the initial 12 days, and then gradually decreased to 103 ( ). The nosZ I/16S rRNA and cnorB /16S rRNA ratios in the S-C0 increased until day 51, and remained constant thereafter, while those of the compost-amended soils increased slightly, with some exceptions ( and ). Bacterial Community Dynamics The dynamics of the bacterial communities during bioremediation of the diesel-contaminated soil are characterized by the Miseq analysis in and . All samples showed good coverages of 0.99 or higher, thereby indicating that the results explain the actual bacterial communities of diesel-contaminated soil effectively ( ). The richness and diversity indices of all samples were increased during the bioremediation process, with those of the compost-amended soils being slightly higher than those of the non-compost-amended soil. However, the indices of the compost-amended soil samples were largely identical, regardless of the amount of compost added. The genus-level analysis in indicates that the structure of the bacterial community in the compost-amended soil was different from that of the control sample. In the S-C0 sample, Sphingomonas remained predominant (10.4–6.7%) throughout the bioremediation, whereas the other major genera changed over time. Thus, Stenotrophobacter (5.2%) and Sphingohabdus (3.0%) were dominant during the initial period, but were superseded by Alkanindiges (11.9%), Ralstonia (5.1%), and Pseudomonas (3.9%) during the intermediate period, and by Rugosibacter (6.8%), Chthoniobacter (4.3%), and Parvibaculum (3.8%) during the late period. By contrast, the dominant genera in the various compost-amended soil samples were initially Membranicola (15.1–19.6%) and Truepera (6.5–7.6%), which were superseded by Immundisolibacter (7.3–16.9%), Dietzia (5.4–10.9%), and Paracoccus (2.3–4.8%) during the intermediate period, and by Sphingomonas (3.5–5.1%), Acidibacter (3.0–4.50%), Immundisolibacter (2.2–5.0%), Marinobacter (2.0–5.6%) and Terrimonas (2.0-4–0%) during the late period. The PCA results in indicate that the initial bacterial community was clearly divided into two groups according to compost addition, whereas the similarity between the two groups increased as the bioremediation progressed. These results suggest that bacterial community succession in the two groups progressed in a similar direction. Hence, to evaluate the effect of compost amendment upon the structure of the bacterial community, those genera having a close relationship with compost amendment were selected via network analysis ( ). The results indicate that compost amendment resulted in an increase in the relative abundances of Atopostipes , Halomona , Massilia , Membranicola , Paracoccus , Pseudogracilibacillus , Pusilimonas , Sphingorhabdus , and Truepera , and a decrease in those of Stenotrophobacter , Sphingomonas and Massilia . The variations in ambient temperature, precipitation, soil organic matter content, and soil pH with time during the pot experiments are presented in . The ambient temperature and precipitation information were obtained from the Korea Meteorological Administration. The average ambient temperature ranged from 13.3 to 30.2°C, and the maximum and minimum temperatures were 35.4 and 8.5°C, respectively ( ). Among the total of 52 rainfall events, there were 3 intensive rainfalls (> 90 mm) between days 50 and 69 ( ). The average organic contents of the soil samples increased with increasing compost addition, and were 0.64, 2.50, 3.19, and 6.07% for the S-C0, S-C5, S-C10, and S-C20 samples, respectively. These values did not change significantly during the experimental period ( ). Meanwhile, the average pH values of the soil samples decreased with increasing amounts of added compost, being 9.01, 8.10, 7.80, and 7.76 for the S-C0, S-C5, S-C10, and S-C20 samples, respectively, and did not significantly vary during the experiment ( ). The changes in the residual diesel concentrations of the various samples with time are presented in . Thus, the initial TPH concentration was 9,432 mg TPH/kg dry soil, and did not change significantly until day 12 in all samples. Thereafter, the residual diesel concentration decreased significantly, and the diesel removal rate was proportional to the amount of compost added. On day 103, the diesel removal efficiencies of the S-C0, S-C5, S-C10, and S-C20 samples were 54.6, 77.5, 80.7, and 85.7%, respectively. Notably, the residual diesel concentration of the S-C20 sample on day 76 was below the 2,000 mg TPH/kg soil pollution risk criterion for oil-contaminated soil in Korea. Dehydrogenase is known to be involved in the initial decomposition of soil organics, catalyzing the removal of hydrogen from organic molecules; hence, the dehydrogenase activity (DHA) is used as an index for evaluating the degradation activity of soil organics . The results in indicate that the DHA of the S-C0 did not significantly change during the initial 33 days, but increased slightly to 19.5 μg TPF/g dry soil/h on day 103. In the soils amended with compost, the initial DHA increased with increasing amounts of added compost, being 203.8, 333.6, and 462.2 μg TPF/g dry soil/h in the S-C5, S-C10, and S-C20, respectively ( ). During the experimental period, the DHAs of the amended soils decreased gradually as the residual diesel concentration decreased ( and ). Urease promotes the mineralization of organic nitrogen to hydrogen-bound nitrogen, thereby providing the soil microorganisms with ammonia as an available nitrogen source . Although the urease activity (UA) cannot explain all of the biological mechanisms, it can be used as a good indicator of TPH metabolism in the soil under various soil conditions [ - ]. As with the DHA, the UA of the S-C0 sample did not change significantly during the early stages of the experiment, but increased slightly during the mid-late period ( ). In the amended soils, the initial UA increased with increasing amounts of added compost, and further increased with time until the 33rd day, decreasing gradually thereafter ( ). 4 Oxidation and N 2 O Reduction Potentials The results in and indicate that there was no significant difference in the initial CH 4 oxidation potential of the various soil samples, which ranged from 1.40 to 1.95 μmol/g dry soil/h. During the experiment, however, the CH 4 oxidation potential of the S-C0 increased significantly until around day 33, and remained relatively constant thereafter. Except on day 12, the CH 4 oxidation potential of the soils amended with compost (7.00–9.83 μmol/g dry soil/h) were higher than that of the non-compost-amended soil (5.39–5.80 μmol/g dry soil/h), and continued to increase significantly with time up until at least day 51. Further, the CH 4 oxidation potential of the S-C20 sample (8.49–9.83 μmol/g dry soil/h) was slightly higher than those of the S-C5 and S-C10 samples (7.00–8.21 μmol/g dry soil/h). The results in and indicate that the initial N 2 O reduction potential of the non-compost amended soil (S-C0) was insignificant (< 56.82 nmol/g dry soil/h), while those of the S-C5, S-C10, and S-C20 were 868.03, 1,399.57 and 1,757.76 nmol/g dry soil/h, respectively. Moreover, while the N 2 O reduction potential gradually decreased with time during bioremediation, a relatively high activity was maintained when the amount of compost added was large. In the S-C20 sample, the N 2 O reduction potential decreased from 838.14 nmol/g dry soil/h on day 12, to 224.08 nmol/g dry soil/h on day 103. In the S-C5 sample, it decreased from 328.57 nmol/g dry soil/h on day 12, to 140.37 nmol/g dry soil/h on day 103. The functional gene dynamics during bioremediation of various diesel-contaminated soil samples are indicated in . Thus, the 16S rRNA gene copy number of the S-C0 sample increased from 10 3 to 10 5 /g dry soil, while those of the S-C5, S-C10, and S-C20 were maintained at around 10 6 /g dry soil during bioremediation ( ). Meanwhile, the relative copy numbers of the alkB gene in the S-C0 increased with bioremediation time to match that of the compost-amended soils (gene copy number = 10 5 /g dry soil) at day 51, and remained constant thereafter ( ). The relative alkB gene copy numbers of the S-C5, S-C10, and S-C20 samples also varied during the initial period (0–33 days), but did not vary significantly after day 51. However, while the relative CYP153 gene copy number in the S-C0 increased from 10 2 to 10 4 /g dry soil during bioremediation, those of the S-C5, S-C10, and S-C20 samples decreased from 10 6 to 10 4 /g dry soil ( ). Ultimately, on day 103, the CYP153 gene copy numbers were similar in all soil samples regardless of compost addition. The pmoA /16S rRNA ratio of the S-C0 was always higher than that of the S-C5, S-C10, and S-C20 during bioremediation ( ). The mcrA /16S rRNA ratio of the S-C0 sample increased from 10 1 to 10 3 , whereas that of the S-C5, S-C10, and S-C20 samples increased from 10 2 to 10 6 during the initial 12 days, and then gradually decreased to 103 ( ). The nosZ I/16S rRNA and cnorB /16S rRNA ratios in the S-C0 increased until day 51, and remained constant thereafter, while those of the compost-amended soils increased slightly, with some exceptions ( and ). The dynamics of the bacterial communities during bioremediation of the diesel-contaminated soil are characterized by the Miseq analysis in and . All samples showed good coverages of 0.99 or higher, thereby indicating that the results explain the actual bacterial communities of diesel-contaminated soil effectively ( ). The richness and diversity indices of all samples were increased during the bioremediation process, with those of the compost-amended soils being slightly higher than those of the non-compost-amended soil. However, the indices of the compost-amended soil samples were largely identical, regardless of the amount of compost added. The genus-level analysis in indicates that the structure of the bacterial community in the compost-amended soil was different from that of the control sample. In the S-C0 sample, Sphingomonas remained predominant (10.4–6.7%) throughout the bioremediation, whereas the other major genera changed over time. Thus, Stenotrophobacter (5.2%) and Sphingohabdus (3.0%) were dominant during the initial period, but were superseded by Alkanindiges (11.9%), Ralstonia (5.1%), and Pseudomonas (3.9%) during the intermediate period, and by Rugosibacter (6.8%), Chthoniobacter (4.3%), and Parvibaculum (3.8%) during the late period. By contrast, the dominant genera in the various compost-amended soil samples were initially Membranicola (15.1–19.6%) and Truepera (6.5–7.6%), which were superseded by Immundisolibacter (7.3–16.9%), Dietzia (5.4–10.9%), and Paracoccus (2.3–4.8%) during the intermediate period, and by Sphingomonas (3.5–5.1%), Acidibacter (3.0–4.50%), Immundisolibacter (2.2–5.0%), Marinobacter (2.0–5.6%) and Terrimonas (2.0-4–0%) during the late period. The PCA results in indicate that the initial bacterial community was clearly divided into two groups according to compost addition, whereas the similarity between the two groups increased as the bioremediation progressed. These results suggest that bacterial community succession in the two groups progressed in a similar direction. Hence, to evaluate the effect of compost amendment upon the structure of the bacterial community, those genera having a close relationship with compost amendment were selected via network analysis ( ). The results indicate that compost amendment resulted in an increase in the relative abundances of Atopostipes , Halomona , Massilia , Membranicola , Paracoccus , Pseudogracilibacillus , Pusilimonas , Sphingorhabdus , and Truepera , and a decrease in those of Stenotrophobacter , Sphingomonas and Massilia . Effects of Compost Amendment on Diesel Removal Efficiency and Soil Enzyme Activity The results in and reveal that the residual diesel concentration is negatively correlated with compost amendment (r = –0.23, p < 0.05), thereby indicating a positive correlation between the diesel removal efficiency and compost amendment. However, compost amendment shows an insignificant or weak negative correlation with the functional genes associated with diesel biodegradation ( i.e. , alkB and CYP153 ; ). Compost amendment for the bioremediation of diesel-contaminated soil has the effects of supplying nutrients along with a highly diverse microbial community with excellent metabolic potential for diesel degradation [ - ]. Moreover, the compost might have affected the diesel removal efficiency by adsorbing diesel and converting it into an available form for the microorganisms . Humic acids in the compost might act as natural surfactants in the soil, thereby further improving the diesel removal efficiency . However, the compost addition does not always enhance the bioremediation efficiency of oil-contaminated soil. Excessive compost addition can decrease the C/N ratio and thereby inhibit microbial activity . In addition, further study to identify the by-products of diesel during bioremediation is necessary. The results in also indicate positive correlations between compost amendment and the DHA (r = 0.78, p < 0.05) and UA (r = 0.44, p < 0.05) values. Taken together, the results in and suggest that the DHA and UA are improved because the addition of compost increased the amount of soil organic matter and nutrients, along with the microbial activity. These results are in agreement with previous work by Namkoong et al . , who noted that the DHA of diesel-contaminated soil increased with the addition of increasing amounts of compost. Other researchers have also reported that the increase in DHA during the bioremediation of oil-contaminated soil is temporary, with a gradual decrease being observed over time . It is well known that the UA has close positive correlations with organic carbon and total nitrogen [ - ]. Effects of Compost Amendment on the CH 4 Oxidation and N 2 O Reduction Potentials All soil samples exhibited CH 4 oxidation potential regardless of the amount of added compost, and the pmoA gene involved in CH 4 oxidation was also detected in all samples ( , ). Nevertheless, the addition of compost resulted in an enhanced CH 4 oxidation potential during the bioremediation period ( ), and there was a positive correlation between the soil CH 4 oxidation potential and the compost amendment ( ). However, the pmoA /16S rRNA ratio was higher in the non-compost-amended soil than in the compost-amended soil ( ), and the soil CH 4 oxidation potential showed a negative correlation with the pmoA gene ( ). These results suggest that there is a limit to explaining the CH 4 oxidation potential only in terms of the behavior of the pmoA gene. Bhardwaj and Dubey reported that the concentration of CH 4 -oxidizing bacteria in dry deciduous tropical forest soil had a significant positive correlation with the copy number of the pmoA gene (r = 0.9, p < 0.01), whereas Qin et al . reported no significant relationship between CH 4 -oxidizing bacteria and the pmoA gene copy number in acidic paddy soil. In a study by Seo and Cho , the compost amendment of diesel-contaminated soil increased the abundance of the pmoA gene, but Yang et al . noted that the levels of CH 4 emission were also enhanced by this treatment due to an increased abundance of CH 4 -producing bacteria. Hence, further research is needed to determine the reason for the improvement in the CH 4 oxidation potential of diesel-contaminated soil by compost addition. Although the soil CH 4 production potential was not evaluated in the present study, the dynamics of the CH 4 production gene, mcrA , were monitored. As shown in , the mcrA /16SRNA ratio was increased during bioremediation of the non-compost-amended soil, but decreased in the compost-amended soil. Further, the correlation matrix in reveals negative correlations between the mcrA gene abundance and both the compost amendment and soil CH 4 oxidation potential. This result suggests that the soil air permeability is improved by the addition of compost, thus making it unfavorable to the growth of anaerobic methanogenic bacteria. This is consistent with previous reports that the addition of exogenous organic matter such as compost can increase the air permeability by increasing the porosity of the soil, thereby improving the removal efficiency of petroleum pollutants . The results in and also reveal positive correlations between compost amendment and both the soil N 2 O reduction potential and the levels of denitrifying genes, such as nosZI , cnorB , and qnorB . In addition, the compost amendment and soil N 2 O reduction potential were each positively correlated with the organic matter content and the DHA. This can be explained by the requirement of carbon and nitrogen sources to act as electron donors and acceptors, respectively, for denitrification metabolism to occur . Moreover, the results in also reveal strong negative correlations between the residual TPH concentrations and the abundances of nosZI , cnorB , and qnorB . The N 2 O reduction potential of diesel-contaminated soil was higher during the initial stages of bioremediation, and decreased with time as the available carbon and nitrogen sources were consumed ( ). Bacterial Community Contributing to Diesel Degradation, CH 4 Oxidation, and N 2 O Reduction The correlations between the bacterial community, CH 4 oxidation, and N 2 O reduction are indicated in . Those genera exhibiting a negative correlation with the residual diesel concentration are associated with diesel degradation. These are Acidibacter , Blastochloris , Erythrobacter , Hyphomicrobium , Marinobacter , Parvibaculum , Pseudoxanthomonas , and Terrimonas . Interestingly, these bacteria also exhibit a strong positive correlation with the soil CH 4 oxidation potential. Previous studies have detected Acidibacter in soil contaminated with high concentrations (25,000–404,300 mg/kg soil) of petroleum . This genus has also been identified among the dominant bacteria during rhizoremediation of diesel-contaminated soil planted with tall fescue or maize . Meanwhile, Blastochloris has been reported as a phototrophic bacterium capable of growing by using aromatic hydrocarbons , and has been detected during the rhizoremediation of diesel-contaminated soil . Erythrobacter has been shown to degrade petroleum, and its relative abundance was shown to increase with time during the bioaugmentation of petroleum-contaminated seawater . Hypomicrobium , a methylotrophic bacterium, was one of the dominant species in a biocover used for the simultaneous removal of CH 4 and odor . In another study, Hypomicrobium was shown to oxidize a high concentration of CH 4 (100,000 ppm) in a batch reactor . Marinobacter has been shown to degrade alkane and polycyclic aromatic hydrocarbons and play a key role in oil degradation during bioremediation . This genus has also been found in frozen soil in the presence of high concentrations of CH 4 . Meanwhile, Xia et al . have reported an increase in the relative abundance of Parvibaculum during the remediation of petroleum-contaminated seawater, while Hou et al . identified Pseudoxanthomonas as one of the dominant rhizobacteria contributing to petroleum degradation during the phytoremediation of contaminated soil using tall fescue. The latter has been shown to degrade diesel in soil , and its relative abundance was found to increase during methane oxidation in an anaerobic methane oxidation system . In particular, Pseudoxanthomonas sp. Q3 has been isolated as a CH 4 degrader from a gasfield in China . Terrimonas has been identified as one of the active rhizobacteria in the rhizoremediation of diesel-contaminated soil using maize or tall fescue . Meanwhile, Bacosa et al . reported that aromatic and aliphatic petroleum compounds were degraded by a bacterial consortium that included Terrimonas and Pseudomonax. Taken together, these results of previous studies and those of the present work suggest that Acidibacter , Blastochloris , Erythrobacter , Hyphomicrobium , Marinobacter , Parvibaculum , Pseudoxanthomonas , and Terrimonas contributed to the diesel degradation and/or CH 4 oxidation during the bioremediation of the diesel-contaminated soil. In , the soil CH 4 oxidation potential shows a significant positive correlation with Brevundimonas and Ferruginibacter . Previous studies have described Brevundimonas and Ferruginibacter as gram-negative heterotrophs , but there are no studies on their relevance to CH 4 oxidation. Hence, future research is needed in order to explain the positive correlation between soil CH 4 oxidation potential and these bacteria. In , the genera exhibiting strong association with soil N 2 O reduction potential are Atopostipes , Bacillus , Halomonas , Oblitimonas , Pusillimonas , Truepera , and Wenahouziangella . In particular, Atopostipes is negatively correlated with NO 3 concentration and has been shown to contribute to the denitrification process during the composting of cattle manure . Bacillus has been shown to remove nitrate and nitrite by its denitrifying capacity . In particular, inoculation with Bacillus amyloliquefaciens has been shown to mitigate N 2 O emission from acidic soil . The aerobic and heterotrophic denitrification capacities of Halomonas have been identified and attributed to functional genes such as napA , nirS , norB , and nosZ . Halomonas has also been shown to contribute to the denitrifying process in an expanded granular sludge bioreactor . Meanwhile, Pusillimonas was isolated from nitrate and radionuclide-contaminated groundwater and shown to possess denitrifying functional genes . Truepera has been reported as one of the denitrifiers in a sequencing batch biofilm reactor used for landfill leachate treatment , and exhibited a high dominance of over 20% in a similar denitrification sequencing batch reactor . The complete denitrification ability of Wenzhouxiangella sp. AB-CW3 isolated from a hypersaline soda lake has also been reported . Based on these reports, Atopostipes , Bacillus , Halomonas , Pusillimonas , Truepera , and Wenahouziangella are presumed to have played an important role in the denitrification and/or N 2 O reduction during the bioremediation of diesel-contaminated soil in the present study. Bioremediation is a promising economical and environmentally friendly soil remediation technology that can be improved by using compost amendment. Herein, diesel-contaminated soil was amended with various weight ratios of compost (0–20%), and correlation and network analyses were used to examine the effects in terms of the dynamics of the bacterial community and functional genes associated with diesel degradation and CH 4 and N 2 O emission. Thus, compost amendment was positively correlated with the diesel removal efficiency, soil enzyme (dehydrogenase and urease) activity, and soil greenhouse gas (CH 4 and N 2 O) mitigation capacity via oxidation and reduction, respectively. However, a positive correlation between the compost amendment and functional gene abundance was only detected for the denitrifying genes ( nosZI , cnorB , and qnorB ) associated with N 2 O reduction. Compost amendment showed weak or insignificant negative correlations with the functional genes associated with diesel biodegradation ( i.e. , alkB and CYP153 ). In addition, compost amendment was negatively correlated with the CH 4 -oxidizing gene pmoA . Further detailed studies are needed to determine the reason for the observed mismatch between the activities (diesel degradation and soil CH 4 oxidation potential) and functional gene abundances ( alkB , CYP153 , and pmoA ). Network analysis showed that the relative abundances of Atopostipes , Halomona , Massilia , Membranicola , Paracoccus , Pseudogracilibacillus , Pusilimonas , Sphingorhabdus , and Truepera were significantly increased by the compost amendment. Among these genera, Atopostipes , Halomonas , Pusillimonas , and Truepera exhibited a strong positive correlation with the soil N 2 O reduction potential. However, the genera that are strongly associated with diesel degradation and soil CH 4 oxidation potential ( i.e. , Acidibacter , Blastochloris , Erythrobacter , Hyphomicrobium , Marinobacter , Parvibaculum , Pseudoxanthomonas and Terrimonas ) were not included among those that exhibited increased abundance upon compost amendment. These results suggest that it is necessary to consider the role of bacteria through an integrated interpretation of various data including bacterial abundance. The results in and reveal that the residual diesel concentration is negatively correlated with compost amendment (r = –0.23, p < 0.05), thereby indicating a positive correlation between the diesel removal efficiency and compost amendment. However, compost amendment shows an insignificant or weak negative correlation with the functional genes associated with diesel biodegradation ( i.e. , alkB and CYP153 ; ). Compost amendment for the bioremediation of diesel-contaminated soil has the effects of supplying nutrients along with a highly diverse microbial community with excellent metabolic potential for diesel degradation [ - ]. Moreover, the compost might have affected the diesel removal efficiency by adsorbing diesel and converting it into an available form for the microorganisms . Humic acids in the compost might act as natural surfactants in the soil, thereby further improving the diesel removal efficiency . However, the compost addition does not always enhance the bioremediation efficiency of oil-contaminated soil. Excessive compost addition can decrease the C/N ratio and thereby inhibit microbial activity . In addition, further study to identify the by-products of diesel during bioremediation is necessary. The results in also indicate positive correlations between compost amendment and the DHA (r = 0.78, p < 0.05) and UA (r = 0.44, p < 0.05) values. Taken together, the results in and suggest that the DHA and UA are improved because the addition of compost increased the amount of soil organic matter and nutrients, along with the microbial activity. These results are in agreement with previous work by Namkoong et al . , who noted that the DHA of diesel-contaminated soil increased with the addition of increasing amounts of compost. Other researchers have also reported that the increase in DHA during the bioremediation of oil-contaminated soil is temporary, with a gradual decrease being observed over time . It is well known that the UA has close positive correlations with organic carbon and total nitrogen [ - ]. 4 Oxidation and N 2 O Reduction Potentials All soil samples exhibited CH 4 oxidation potential regardless of the amount of added compost, and the pmoA gene involved in CH 4 oxidation was also detected in all samples ( , ). Nevertheless, the addition of compost resulted in an enhanced CH 4 oxidation potential during the bioremediation period ( ), and there was a positive correlation between the soil CH 4 oxidation potential and the compost amendment ( ). However, the pmoA /16S rRNA ratio was higher in the non-compost-amended soil than in the compost-amended soil ( ), and the soil CH 4 oxidation potential showed a negative correlation with the pmoA gene ( ). These results suggest that there is a limit to explaining the CH 4 oxidation potential only in terms of the behavior of the pmoA gene. Bhardwaj and Dubey reported that the concentration of CH 4 -oxidizing bacteria in dry deciduous tropical forest soil had a significant positive correlation with the copy number of the pmoA gene (r = 0.9, p < 0.01), whereas Qin et al . reported no significant relationship between CH 4 -oxidizing bacteria and the pmoA gene copy number in acidic paddy soil. In a study by Seo and Cho , the compost amendment of diesel-contaminated soil increased the abundance of the pmoA gene, but Yang et al . noted that the levels of CH 4 emission were also enhanced by this treatment due to an increased abundance of CH 4 -producing bacteria. Hence, further research is needed to determine the reason for the improvement in the CH 4 oxidation potential of diesel-contaminated soil by compost addition. Although the soil CH 4 production potential was not evaluated in the present study, the dynamics of the CH 4 production gene, mcrA , were monitored. As shown in , the mcrA /16SRNA ratio was increased during bioremediation of the non-compost-amended soil, but decreased in the compost-amended soil. Further, the correlation matrix in reveals negative correlations between the mcrA gene abundance and both the compost amendment and soil CH 4 oxidation potential. This result suggests that the soil air permeability is improved by the addition of compost, thus making it unfavorable to the growth of anaerobic methanogenic bacteria. This is consistent with previous reports that the addition of exogenous organic matter such as compost can increase the air permeability by increasing the porosity of the soil, thereby improving the removal efficiency of petroleum pollutants . The results in and also reveal positive correlations between compost amendment and both the soil N 2 O reduction potential and the levels of denitrifying genes, such as nosZI , cnorB , and qnorB . In addition, the compost amendment and soil N 2 O reduction potential were each positively correlated with the organic matter content and the DHA. This can be explained by the requirement of carbon and nitrogen sources to act as electron donors and acceptors, respectively, for denitrification metabolism to occur . Moreover, the results in also reveal strong negative correlations between the residual TPH concentrations and the abundances of nosZI , cnorB , and qnorB . The N 2 O reduction potential of diesel-contaminated soil was higher during the initial stages of bioremediation, and decreased with time as the available carbon and nitrogen sources were consumed ( ). 4 Oxidation, and N 2 O Reduction The correlations between the bacterial community, CH 4 oxidation, and N 2 O reduction are indicated in . Those genera exhibiting a negative correlation with the residual diesel concentration are associated with diesel degradation. These are Acidibacter , Blastochloris , Erythrobacter , Hyphomicrobium , Marinobacter , Parvibaculum , Pseudoxanthomonas , and Terrimonas . Interestingly, these bacteria also exhibit a strong positive correlation with the soil CH 4 oxidation potential. Previous studies have detected Acidibacter in soil contaminated with high concentrations (25,000–404,300 mg/kg soil) of petroleum . This genus has also been identified among the dominant bacteria during rhizoremediation of diesel-contaminated soil planted with tall fescue or maize . Meanwhile, Blastochloris has been reported as a phototrophic bacterium capable of growing by using aromatic hydrocarbons , and has been detected during the rhizoremediation of diesel-contaminated soil . Erythrobacter has been shown to degrade petroleum, and its relative abundance was shown to increase with time during the bioaugmentation of petroleum-contaminated seawater . Hypomicrobium , a methylotrophic bacterium, was one of the dominant species in a biocover used for the simultaneous removal of CH 4 and odor . In another study, Hypomicrobium was shown to oxidize a high concentration of CH 4 (100,000 ppm) in a batch reactor . Marinobacter has been shown to degrade alkane and polycyclic aromatic hydrocarbons and play a key role in oil degradation during bioremediation . This genus has also been found in frozen soil in the presence of high concentrations of CH 4 . Meanwhile, Xia et al . have reported an increase in the relative abundance of Parvibaculum during the remediation of petroleum-contaminated seawater, while Hou et al . identified Pseudoxanthomonas as one of the dominant rhizobacteria contributing to petroleum degradation during the phytoremediation of contaminated soil using tall fescue. The latter has been shown to degrade diesel in soil , and its relative abundance was found to increase during methane oxidation in an anaerobic methane oxidation system . In particular, Pseudoxanthomonas sp. Q3 has been isolated as a CH 4 degrader from a gasfield in China . Terrimonas has been identified as one of the active rhizobacteria in the rhizoremediation of diesel-contaminated soil using maize or tall fescue . Meanwhile, Bacosa et al . reported that aromatic and aliphatic petroleum compounds were degraded by a bacterial consortium that included Terrimonas and Pseudomonax. Taken together, these results of previous studies and those of the present work suggest that Acidibacter , Blastochloris , Erythrobacter , Hyphomicrobium , Marinobacter , Parvibaculum , Pseudoxanthomonas , and Terrimonas contributed to the diesel degradation and/or CH 4 oxidation during the bioremediation of the diesel-contaminated soil. In , the soil CH 4 oxidation potential shows a significant positive correlation with Brevundimonas and Ferruginibacter . Previous studies have described Brevundimonas and Ferruginibacter as gram-negative heterotrophs , but there are no studies on their relevance to CH 4 oxidation. Hence, future research is needed in order to explain the positive correlation between soil CH 4 oxidation potential and these bacteria. In , the genera exhibiting strong association with soil N 2 O reduction potential are Atopostipes , Bacillus , Halomonas , Oblitimonas , Pusillimonas , Truepera , and Wenahouziangella . In particular, Atopostipes is negatively correlated with NO 3 concentration and has been shown to contribute to the denitrification process during the composting of cattle manure . Bacillus has been shown to remove nitrate and nitrite by its denitrifying capacity . In particular, inoculation with Bacillus amyloliquefaciens has been shown to mitigate N 2 O emission from acidic soil . The aerobic and heterotrophic denitrification capacities of Halomonas have been identified and attributed to functional genes such as napA , nirS , norB , and nosZ . Halomonas has also been shown to contribute to the denitrifying process in an expanded granular sludge bioreactor . Meanwhile, Pusillimonas was isolated from nitrate and radionuclide-contaminated groundwater and shown to possess denitrifying functional genes . Truepera has been reported as one of the denitrifiers in a sequencing batch biofilm reactor used for landfill leachate treatment , and exhibited a high dominance of over 20% in a similar denitrification sequencing batch reactor . The complete denitrification ability of Wenzhouxiangella sp. AB-CW3 isolated from a hypersaline soda lake has also been reported . Based on these reports, Atopostipes , Bacillus , Halomonas , Pusillimonas , Truepera , and Wenahouziangella are presumed to have played an important role in the denitrification and/or N 2 O reduction during the bioremediation of diesel-contaminated soil in the present study. Bioremediation is a promising economical and environmentally friendly soil remediation technology that can be improved by using compost amendment. Herein, diesel-contaminated soil was amended with various weight ratios of compost (0–20%), and correlation and network analyses were used to examine the effects in terms of the dynamics of the bacterial community and functional genes associated with diesel degradation and CH 4 and N 2 O emission. Thus, compost amendment was positively correlated with the diesel removal efficiency, soil enzyme (dehydrogenase and urease) activity, and soil greenhouse gas (CH 4 and N 2 O) mitigation capacity via oxidation and reduction, respectively. However, a positive correlation between the compost amendment and functional gene abundance was only detected for the denitrifying genes ( nosZI , cnorB , and qnorB ) associated with N 2 O reduction. Compost amendment showed weak or insignificant negative correlations with the functional genes associated with diesel biodegradation ( i.e. , alkB and CYP153 ). In addition, compost amendment was negatively correlated with the CH 4 -oxidizing gene pmoA . Further detailed studies are needed to determine the reason for the observed mismatch between the activities (diesel degradation and soil CH 4 oxidation potential) and functional gene abundances ( alkB , CYP153 , and pmoA ). Network analysis showed that the relative abundances of Atopostipes , Halomona , Massilia , Membranicola , Paracoccus , Pseudogracilibacillus , Pusilimonas , Sphingorhabdus , and Truepera were significantly increased by the compost amendment. Among these genera, Atopostipes , Halomonas , Pusillimonas , and Truepera exhibited a strong positive correlation with the soil N 2 O reduction potential. However, the genera that are strongly associated with diesel degradation and soil CH 4 oxidation potential ( i.e. , Acidibacter , Blastochloris , Erythrobacter , Hyphomicrobium , Marinobacter , Parvibaculum , Pseudoxanthomonas and Terrimonas ) were not included among those that exhibited increased abundance upon compost amendment. These results suggest that it is necessary to consider the role of bacteria through an integrated interpretation of various data including bacterial abundance. Supplementary data for this paper are available on-line only at http://jmb.or.kr .
Designing highly potent compounds using a chemical language model	18b74ffb-0440-4877-8eed-b83dc518df0e	10164739	Pharmacology[mh]	Compound design is one of the major tasks for computational approaches in medicinal chemistry. The primary aim is the generation of compounds with desired properties, first and foremost, compounds with activity against individual pharmaceutical targets and high potency. For compound design and potency predictions, a variety of computational methods have been developed or adapted. Mainstays include quantitative structure–activity relationship (QSAR) analysis for the design of increasingly potent analogues of active compounds and methods for ligand- or structure-based virtual screening , to identify new hits. Ligand- and structure-based methods have different requirements. For example, for docking calculations , a variety of scoring functions have been developed to evaluate the quality and strength of receptor-ligand interactions and estimate binding energies , . For the structure-based prediction of relative potencies of congeneric compounds, free energy perturbation methods have been introduced , . At the ligand level, machine learning (ML) methods are widely used for hit identification and non-linear QSAR modeling . For potency prediction, support vector regression (SVR) has become a standard ML approach. Furthermore, for both computational compound screening and potency prediction, deep neural network (DNN) architectures are also increasingly investigated – . Recently, a methodological framework was developed for evaluating the performance of deep generative models and a recurrent neural network (RNN) was used to explore predictions based on sparse training data . However, the analysis mainly focused on physicochemical properties. For potency prediction, the assessment and comparison of different methods typically relies on the use of standard benchmark settings. Such benchmark calculations are required but not sufficient to evaluate potency prediction methods and their potential for practical applications. Moreover, such calculations should be considered with caution. Notably, in benchmark settings, nearest neighbor analysis and mean or median value regression often meet the accuracy of increasingly complex ML methods . The high performance of these simple reference methods is supported by potency value distributions in commonly used compound data sets . In addition, narrow error margins separating ML-based and randomized potency value predictions limit conclusions that can be drawn from conventional benchmarking . Such findings call for alternatives to conventional benchmarking such as focusing predictions on the most potent data set compounds, consistent with the final goal of compound optimization efforts. While potency predictions are mostly carried out for individual compounds, they can also be applied to assess potency differences in compound pairs such as activity cliffs (ACs), which are formed by structurally similar (analogous) active compounds with large differences in potency . In principle, ACs can be predicted by explicitly calculating potency differences between compounds in pairs or by distinguishing between ACs and other pairs of analogues using classification methods, which implicitly accounts for potency differences of varying magnitude. Previously, we have reported a deep learning approach for the prediction of ACs that further extended other ML classification methods by its ability to not only predict ACs, but also generate new AC compounds . Since ACs encode large potency differences, we have reasoned that this methodology might be adapted and further extended for the design of highly potent compounds. Therefore, in this work, we have devised and implemented a chemical language model (CLM) for the prediction of highly potent compounds from weakly potent ones used as input. These predictions do not depend on conventional benchmark settings and are thus not affected by their intrinsic limitations. Compounds, activity data, and analogue series From ChEMBL (release 29) , bioactive compounds with high-confidence activity data were assembled. Only compounds with reported direct interactions (assay relationship type: “D”) with human targets at the highest assay confidence level (assay confidence score 9) were considered. As potency measurements, only numerically specified equilibrium constants (K i values) were accepted and recorded as (negative logarithmic) pK i values. If multiple measurements were available for the same compound, the geometric mean was calculated as the final potency annotation, provided all values fell within the same order of magnitude; otherwise, the compound was disregarded. Qualifying compounds were organized into target-based activity classes. A total of 496 activity classes were obtained. For each activity class, a systematic search for analogue series (ASs) was conducted using the compound-core relationship (CCR) method , which uses a modified matched molecular pair (MMP) fragmentation procedure based on retrosynthetic rules to systematically identify ASs with single or multiple (maximally five) substitution sites. The core structure of an AS was required to consist of at least twice the number of non-hydrogen atoms of the combined substituents . Ultimately, 10 classes comprising ligands of different G protein coupled receptors were extracted as test cases for compound predictions that each contained more than 900 compounds and more than 100 analogue series. Table summarizes the targets and composition of these activity classes (first four columns from the left) and Fig. shows exemplary ASs with single or multiple substitution sites. For each of 10 activity classes, the number of compounds, ASs, CCR pairs, and AC-CCR pairs are provided. In addition, for each class, the ChEMBL target ID, target name, and abbreviation are given. AS, CCR, and AC stand for analogue series, compound-core relationship, and activity cliff, respectively. From each of the activity classes, all possible pairs of analogues (termed All_CCR pairs) were extracted, as illustrated in Fig. that shows All_CCR pairs for two different ASs. The 496 activity classes yielded a total of 881,990 All_CCR pairs. Tokenization For use by a CLM, compounds and potency differences must be tokenized. All compounds were represented as molecular-input line-entry system (SMILES) strings generated using RDKit and tokenized using a single chemical character with the exception of two-character tokens (i.e., “Cl” and “Br”) and tokens in brackets (e.g. “[nH]” and “[O-]”). For the conditional transformer, potency differences must also be transformed into input tokens. For tokenization of value ranges, different approaches have been introduced including binning , , and, more recently, numerical tokenization . Since human readability of token sequences supported by numerical approaches played no role for our analysis and encoding of drug discovery-relevant compound potency ranges via binning has yielded accurate predictions previously , we continued to use binned tokens herein. Accordingly, potency differences between source and target compounds, ranging from − 6.62 to 6.52 pK i units, were partitioned into 1314 binned tokens of a constant width of 0.01. This granularity (resolution) defines the limits of experimental potency measurements and was thus most appropriate for our analysis. Each bin was encoded by a single token and each potency difference was assigned to the token of the corresponding bin . Tokenization of compound SMILES strings and potency ranges yielded the chemical vocabulary for our model. In addition, the two special tokens “start” and “end” were added to the vocabulary indicating the start and end point of a sequence, respectively. Generative chemical language model Architecture For compound design, a CLM with the transformer architecture previously reported for the DeepAC approach for AC prediction was used. The transformer architecture consisted of multiple encoder-decoder neural modules with attention mechanism . In the model, a stack of encoding sub-layers including a multi-head self-attention sub-layer and a fully connected feed-forward network sub-layer constituted the encoder module. The encoder read an input sequence and compressed it into a context vector in its final hidden state. The context vector served as the input for the decoder block that interpreted the vector to predict an output sequence. Subsequently, the decoder module, which was composed of a feed-forward sub-layer and two multi-head attention sub-layers, re-converted the encodings into a sequence of tokens (one token at a time). Both encoder and decoder utilized the attention mechanism during training to comprehensively learn from feature space. During pre-training, the model was supposed to learn mappings of source to target compounds based on potency differences resulting from changes in substituent(s) (termed chemical transformations): [12pt]{minimal} $$(Source\;compound,\;Potency\;difference) ( {Target\;compound} ).$$ ( S o u r c e c o m p o u n d , P o t e n c y d i f f e r e n c e ) → T a r g e t c o m p o u n d . Then, given a new (Source compound, Potency difference) test instance, the model was applied to generate a set of candidate target compounds meeting the potency difference constraints, that is, having higher potency than the source compound (according to the given potency difference). During pre-training, distinguishing between different activity classes was not required because at this stage, the model should learn the syntax of textual molecular representations and, in addition, a variety of analogue pair-associated potency differences caused by chemical transformations. By contrast, during fine-tuning, activity class (target) information was required to focus the model on specific compound series or classes, as further discussed below. Model derivation The transformer model was implemented using Pytorch . Default hyperparameter settings were used together with a batch size of 64, learning rate of 0.0001, and encoding dimension of 256. The models were derived over 200 epochs on the basis of the general training set. During training, the transformer model minimized the cross-entropy loss between the ground-truth and output sequence. A checkpoint was saved at each epoch and for a validation set, minimal loss was determined for selecting the final model. Model pre-training A general data set for model pre-training was derived from the 881,990 All_CCR pairs of the 496 activity classes. From All_CCR pairs, All_CCR triples (Cpd A , Cpd B , Pot B -Pot A ) were generated by recording the potency difference for an All_CCR pair. Here, Cpd A represented the source compound that was concatenated with the potency difference ( Pot B -Pot A ) and Cpd B represented the target compound . For each All_CCR pair, two triples were obtained such that each All_CCR compound was used once as the source and target compound. To avoid data ambiguities, All_CCR pairs were eliminated if (1) a given source compound and potency difference was associated with multiple target compounds from different activity classes or (2) multiple potency values from different classes were available for a pair. On the basis of these criteria, a curated general data set of 522,331 qualifying All_CCR triples was obtained and used for pre-training. For each triple, the SMILES representation of the source compound concatenated with the binned token of the associated potency difference served as the input sequence for the encoder that was converted into a latent representation. Based on this representation, the decoder iteratively generated output SMILES sequences until the end token was detected. Model fine-tuning For model fine-tuning and evaluation, the 10 activity classes in Table were used. For fine-tuning, All_CCR pairs were extracted from each of the 10 activity classes and divided into subsets of so-called CCR pairs with a less than 100-fold potency difference and AC-CCR pairs capturing an at least 100-fold difference in potency. Accordingly, AC-CCR pairs represented analogue pairs forming ACs. Depending on the activity class, 8889–42,621 CCR pairs and 585–6219 AC-CCR pairs were obtained (Table , last two columns on the right). AC-CCR triples were ordered such that Cpd B was highly and Cpd A weakly potent. The pre-trained model was then separately fine-tuned and tested for each activity class. Therefore, AC-CCR pairs from each class were randomly divided into 80% fine-tuning and 20% test instances such that there was no overlap in core structures between these sets. Thus, the fine-tuning set exclusively consisted of AC-CCR pairs and was selected to train the model on activity class dependent analogue pairs with large potency differences. CCR pairs sharing core structures with the fine-tuning set were omitted from further consideration. The remaining CCR pairs were added to the test set. Hence, the fine-tuning and test sets were structurally distinct. Model evaluation is detailed below. From ChEMBL (release 29) , bioactive compounds with high-confidence activity data were assembled. Only compounds with reported direct interactions (assay relationship type: “D”) with human targets at the highest assay confidence level (assay confidence score 9) were considered. As potency measurements, only numerically specified equilibrium constants (K i values) were accepted and recorded as (negative logarithmic) pK i values. If multiple measurements were available for the same compound, the geometric mean was calculated as the final potency annotation, provided all values fell within the same order of magnitude; otherwise, the compound was disregarded. Qualifying compounds were organized into target-based activity classes. A total of 496 activity classes were obtained. For each activity class, a systematic search for analogue series (ASs) was conducted using the compound-core relationship (CCR) method , which uses a modified matched molecular pair (MMP) fragmentation procedure based on retrosynthetic rules to systematically identify ASs with single or multiple (maximally five) substitution sites. The core structure of an AS was required to consist of at least twice the number of non-hydrogen atoms of the combined substituents . Ultimately, 10 classes comprising ligands of different G protein coupled receptors were extracted as test cases for compound predictions that each contained more than 900 compounds and more than 100 analogue series. Table summarizes the targets and composition of these activity classes (first four columns from the left) and Fig. shows exemplary ASs with single or multiple substitution sites. For each of 10 activity classes, the number of compounds, ASs, CCR pairs, and AC-CCR pairs are provided. In addition, for each class, the ChEMBL target ID, target name, and abbreviation are given. AS, CCR, and AC stand for analogue series, compound-core relationship, and activity cliff, respectively. From each of the activity classes, all possible pairs of analogues (termed All_CCR pairs) were extracted, as illustrated in Fig. that shows All_CCR pairs for two different ASs. The 496 activity classes yielded a total of 881,990 All_CCR pairs. Tokenization For use by a CLM, compounds and potency differences must be tokenized. All compounds were represented as molecular-input line-entry system (SMILES) strings generated using RDKit and tokenized using a single chemical character with the exception of two-character tokens (i.e., “Cl” and “Br”) and tokens in brackets (e.g. “[nH]” and “[O-]”). For the conditional transformer, potency differences must also be transformed into input tokens. For tokenization of value ranges, different approaches have been introduced including binning , , and, more recently, numerical tokenization . Since human readability of token sequences supported by numerical approaches played no role for our analysis and encoding of drug discovery-relevant compound potency ranges via binning has yielded accurate predictions previously , we continued to use binned tokens herein. Accordingly, potency differences between source and target compounds, ranging from − 6.62 to 6.52 pK i units, were partitioned into 1314 binned tokens of a constant width of 0.01. This granularity (resolution) defines the limits of experimental potency measurements and was thus most appropriate for our analysis. Each bin was encoded by a single token and each potency difference was assigned to the token of the corresponding bin . Tokenization of compound SMILES strings and potency ranges yielded the chemical vocabulary for our model. In addition, the two special tokens “start” and “end” were added to the vocabulary indicating the start and end point of a sequence, respectively. For use by a CLM, compounds and potency differences must be tokenized. All compounds were represented as molecular-input line-entry system (SMILES) strings generated using RDKit and tokenized using a single chemical character with the exception of two-character tokens (i.e., “Cl” and “Br”) and tokens in brackets (e.g. “[nH]” and “[O-]”). For the conditional transformer, potency differences must also be transformed into input tokens. For tokenization of value ranges, different approaches have been introduced including binning , , and, more recently, numerical tokenization . Since human readability of token sequences supported by numerical approaches played no role for our analysis and encoding of drug discovery-relevant compound potency ranges via binning has yielded accurate predictions previously , we continued to use binned tokens herein. Accordingly, potency differences between source and target compounds, ranging from − 6.62 to 6.52 pK i units, were partitioned into 1314 binned tokens of a constant width of 0.01. This granularity (resolution) defines the limits of experimental potency measurements and was thus most appropriate for our analysis. Each bin was encoded by a single token and each potency difference was assigned to the token of the corresponding bin . Tokenization of compound SMILES strings and potency ranges yielded the chemical vocabulary for our model. In addition, the two special tokens “start” and “end” were added to the vocabulary indicating the start and end point of a sequence, respectively. Architecture For compound design, a CLM with the transformer architecture previously reported for the DeepAC approach for AC prediction was used. The transformer architecture consisted of multiple encoder-decoder neural modules with attention mechanism . In the model, a stack of encoding sub-layers including a multi-head self-attention sub-layer and a fully connected feed-forward network sub-layer constituted the encoder module. The encoder read an input sequence and compressed it into a context vector in its final hidden state. The context vector served as the input for the decoder block that interpreted the vector to predict an output sequence. Subsequently, the decoder module, which was composed of a feed-forward sub-layer and two multi-head attention sub-layers, re-converted the encodings into a sequence of tokens (one token at a time). Both encoder and decoder utilized the attention mechanism during training to comprehensively learn from feature space. During pre-training, the model was supposed to learn mappings of source to target compounds based on potency differences resulting from changes in substituent(s) (termed chemical transformations): [12pt]{minimal} $$(Source\;compound,\;Potency\;difference) ( {Target\;compound} ).$$ ( S o u r c e c o m p o u n d , P o t e n c y d i f f e r e n c e ) → T a r g e t c o m p o u n d . Then, given a new (Source compound, Potency difference) test instance, the model was applied to generate a set of candidate target compounds meeting the potency difference constraints, that is, having higher potency than the source compound (according to the given potency difference). During pre-training, distinguishing between different activity classes was not required because at this stage, the model should learn the syntax of textual molecular representations and, in addition, a variety of analogue pair-associated potency differences caused by chemical transformations. By contrast, during fine-tuning, activity class (target) information was required to focus the model on specific compound series or classes, as further discussed below. Model derivation The transformer model was implemented using Pytorch . Default hyperparameter settings were used together with a batch size of 64, learning rate of 0.0001, and encoding dimension of 256. The models were derived over 200 epochs on the basis of the general training set. During training, the transformer model minimized the cross-entropy loss between the ground-truth and output sequence. A checkpoint was saved at each epoch and for a validation set, minimal loss was determined for selecting the final model. For compound design, a CLM with the transformer architecture previously reported for the DeepAC approach for AC prediction was used. The transformer architecture consisted of multiple encoder-decoder neural modules with attention mechanism . In the model, a stack of encoding sub-layers including a multi-head self-attention sub-layer and a fully connected feed-forward network sub-layer constituted the encoder module. The encoder read an input sequence and compressed it into a context vector in its final hidden state. The context vector served as the input for the decoder block that interpreted the vector to predict an output sequence. Subsequently, the decoder module, which was composed of a feed-forward sub-layer and two multi-head attention sub-layers, re-converted the encodings into a sequence of tokens (one token at a time). Both encoder and decoder utilized the attention mechanism during training to comprehensively learn from feature space. During pre-training, the model was supposed to learn mappings of source to target compounds based on potency differences resulting from changes in substituent(s) (termed chemical transformations): [12pt]{minimal} $$(Source\;compound,\;Potency\;difference) ( {Target\;compound} ).$$ ( S o u r c e c o m p o u n d , P o t e n c y d i f f e r e n c e ) → T a r g e t c o m p o u n d . Then, given a new (Source compound, Potency difference) test instance, the model was applied to generate a set of candidate target compounds meeting the potency difference constraints, that is, having higher potency than the source compound (according to the given potency difference). During pre-training, distinguishing between different activity classes was not required because at this stage, the model should learn the syntax of textual molecular representations and, in addition, a variety of analogue pair-associated potency differences caused by chemical transformations. By contrast, during fine-tuning, activity class (target) information was required to focus the model on specific compound series or classes, as further discussed below. The transformer model was implemented using Pytorch . Default hyperparameter settings were used together with a batch size of 64, learning rate of 0.0001, and encoding dimension of 256. The models were derived over 200 epochs on the basis of the general training set. During training, the transformer model minimized the cross-entropy loss between the ground-truth and output sequence. A checkpoint was saved at each epoch and for a validation set, minimal loss was determined for selecting the final model. A general data set for model pre-training was derived from the 881,990 All_CCR pairs of the 496 activity classes. From All_CCR pairs, All_CCR triples (Cpd A , Cpd B , Pot B -Pot A ) were generated by recording the potency difference for an All_CCR pair. Here, Cpd A represented the source compound that was concatenated with the potency difference ( Pot B -Pot A ) and Cpd B represented the target compound . For each All_CCR pair, two triples were obtained such that each All_CCR compound was used once as the source and target compound. To avoid data ambiguities, All_CCR pairs were eliminated if (1) a given source compound and potency difference was associated with multiple target compounds from different activity classes or (2) multiple potency values from different classes were available for a pair. On the basis of these criteria, a curated general data set of 522,331 qualifying All_CCR triples was obtained and used for pre-training. For each triple, the SMILES representation of the source compound concatenated with the binned token of the associated potency difference served as the input sequence for the encoder that was converted into a latent representation. Based on this representation, the decoder iteratively generated output SMILES sequences until the end token was detected. For model fine-tuning and evaluation, the 10 activity classes in Table were used. For fine-tuning, All_CCR pairs were extracted from each of the 10 activity classes and divided into subsets of so-called CCR pairs with a less than 100-fold potency difference and AC-CCR pairs capturing an at least 100-fold difference in potency. Accordingly, AC-CCR pairs represented analogue pairs forming ACs. Depending on the activity class, 8889–42,621 CCR pairs and 585–6219 AC-CCR pairs were obtained (Table , last two columns on the right). AC-CCR triples were ordered such that Cpd B was highly and Cpd A weakly potent. The pre-trained model was then separately fine-tuned and tested for each activity class. Therefore, AC-CCR pairs from each class were randomly divided into 80% fine-tuning and 20% test instances such that there was no overlap in core structures between these sets. Thus, the fine-tuning set exclusively consisted of AC-CCR pairs and was selected to train the model on activity class dependent analogue pairs with large potency differences. CCR pairs sharing core structures with the fine-tuning set were omitted from further consideration. The remaining CCR pairs were added to the test set. Hence, the fine-tuning and test sets were structurally distinct. Model evaluation is detailed below. Study concept Our study had three primary goals. First, we aimed to devise a novel approach specifically for predicting highly potent compounds from weakly potent input molecules. Thus, rather than striving for prediction of potency values across large ranges, as is conventionally attempted using SVR or other machine learning methods, the primary focus was on potent compounds, in line with the practical relevance of potency predictions. Second, we aimed to generate a structural spectrum of output compounds, ranging from analogues of input molecules to structurally distinct compounds, thereby increasing medicinal chemistry novelty of predicted candidates. Third, it was intended to evaluate the methodology in a way that was not affected by limitations of conventional benchmarking of potency predictions, as discussed above, and enabled a non-ambiguous assessment of the ability to predict potent compounds. To meet the first two goals, which were central to our study, we implemented a CLM consisting of a chemical transformer architecture conditioned on compound potency differences. To meet the third goal, we designed a new compound test system. Compound pair-based test system For model evaluation, a compound pair-based test system was generated using the test set. By design, the fine-tuning and test sets were structurally distinct. Furthermore, in contrast to the fine-tuning set, the test set contained analogue pairs capturing small or large differences in potency (i.e., CCR and AC-CCR pairs, respectively). Table summarizes the composition of the test set. For each activity class, the test set contained varying numbers of CCR pairs and AC-CCR pairs yielding varying numbers of unique CCR and AC-CCR compounds. In the following, SC and TC are used as abbreviations for source (input) and target compound, respectively. For the evaluation of the fine-tuned CLM, test set compounds were divided into instances with maximally 1 μmol potency (corresponding to a pK i value of 6), which served as SCs, and candidate compounds with higher than 1 μmol potency (pK i > 6), which served as known candidate compounds ( KCCs ) for comparison with newly generated TCs. In addition, the model generated varying numbers of novel (hypothetical) TCs. For each activity class, smaller numbers of SCs than KCCs were available. With the exception of activity class 251 (3838 KCCs), the test set contained 366–824 KCCs for the activity classes (Table ), with on average 576 KCCs per class. Each CCR-SC (pK i ≤ 6) and AC-CCR-SC (pK i ≤ 6) was once used as an input compound for the model and in each case, 50 TCs were sampled, canonicalized, and compared to KCCs to search for exact matches, that is, fully reproduced compounds with known potency. Because the model generated novel TCs, probabilities for re-generating known TCs could not be derived in a meaningful way. Consequently, the main measure for establishing proof-of-principle for the ability of the model to predict potent compounds was the reproduction of any KCCs . For each activity class, compound statistics were derived over three independent sampling trials, as reported below. Table reports the possible predictions outcomes for the compound pair-based test system. For each SC, a TC could be a known CCR or AC-CCR compound or a novel (hypothetical) compound representing a TC not contained in the fine-tuning or test set. Taking core structure matches into consideration (that is, a TC either contained the same core structure as a SC or not), a total of 12 formally defined prediction outcomes were possible, including six each for CCR-SCs and AC-CCR-SCs, as identified by indices 1.1.–1.6. and 2.1.–2.6. in Table , respectively. Accordingly, a newly generated compound might be a structural analogue of a given SC (having the same core structure) or contain a different core structure. Furthermore, SCs and TCs might be distinguished by single or multiple substituents. On the basis of this classification scheme, CLM predictions were rigorously evaluated focusing on the reproduction of known active compounds, as explained above. This was the most relevant measure of model performance because it enabled the exact determination of potency differences between SCs and TCs and hence the ability of the CLM to predict highly potent compounds. For novel (hypothetical) compounds generated by the model, no assessment was possible (without subsequent experimental evaluation). Model performance For the SCs from all activity classes, systematic compound predictions were carried out using the CLM. The model only produced 0.5–2% invalid SMILES (assessed using RDKit) for all activity classes. With the exception of class 251 (1391 SCs), the test set contained 40–359 SCs for the activity classes, with on average 162 compounds per class (Table ). The predictions were then assessed on the basis of well-defined pair categories detailed above, as reported in Table . For each activity class and compound pair category indexed according to Table (top row), the number of unique TCs produced by the CLM is reported. With the exception of categories 1.5., 1.6., 2.5., and 2.6., which report novel (hypothetical) candidate compounds not contained in the fine-tuning or test set, the TCs represent KCCs, as defined in the text. Encouragingly, for all activity classes, the CLM successfully reproduced large numbers of KCCs for all SCs (categories 1.1.–1.4. and 2.1.–2.4., respectively). Frequently, multiple KCCs were obtained for the same SC. Furthermore, depending on the activity class, the model produced varying numbers of TCs with the same or different core structure, thus confirming its ability to generate frequent core structure transformations. In many cases, more structurally unique TCs were generated than analogues of SCs. Moreover, large numbers of hypothetical candidate compounds not contained in the training set were obtained (categories 1.5.–1.6. and 2.5.–2.6., respectively). The reproducibility of the limited numbers of available KCCs representing known ACs (12–84 unique compounds per activity class) was of particular interest (categories 2.1.–2.4.). AC-CCR KCCs were consistently reproduced and for five activity classes, the total count exceeded the number of unique AC-CCR KCCs per class (due to multiple reproductions of individual KCCs). Table reports statistics for reproduction of KCCs. Reported are statistics for the re-generation of KCCs including the mean number of KCCs over three independent sampling trials and the proportion of reproduced KCCs relative to all available KCCs with standard deviations (±). In addition, the mean number of non-KCCs over three independent trials is provided. The proportion of exactly reproduced KCCs over independent sampling trials ranged from ~ 7 to ~ 37%, depending on the activity class (with generally small standard deviations). For nine, six, and two classes, more than 10, 20, and 30% of all available KCCs were reproduced, respectively. Applying the most rigorous criterion of exact re-generation of known potent compounds as a performance measure (see above), the observed numbers and proportions represented unexpectedly good predictions, which clearly established proof-of-concept for the approach. For each activity class, ASs were also extracted from newly generated (predicted) compounds. Table reports the number of ASs (multiple compounds having the same core structure) and singletons (compounds with unique core structures not belonging to any AS). Depending on the activity class, 90–1414 ASs and 219–1762 singletons were obtained, respectively. Since each AS and singleton contained a unique core structure (scaffold), the core structure diversity of newly generated compounds was generally high. Between 4 and 18% of the core structures contained in the original activity classes (from ASs and singletons) were reproduced by the model, as also reported in Table . Having confirmed the ability of the CLM to generate structurally analogous and diverse TCs including KCCs, the key question then was whether or not the model would produce TCs that had much higher potency than the corresponding SCs. Figure shows the distributions of potency differences between pairs of known source and target compounds with experimental potency values involving compounds from ACs. For five activity classes, the median potency difference fell between one and two orders of magnitude (10–100-fold) and for the other five classes, the median value exceeded two orders of magnitude (100-fold). Furthermore, for all but one class, multiple compounds with at least 1000-fold higher potency than the corresponding SCs were generated (including highly potent statistical outliers). Thus, these observations unambiguously confirmed the ability of the CLM to generate highly potent compounds from weakly potent (micromolar) input molecules. Figure shows exemplary pairs of SCs and newly designed compounds (TCs) with different structural relationships. Given our design strategy, all SCs were known compounds with experimentally determined potency. The generated TCs included known potent analogues of SCs (Fig. a), structurally distinct known potent compounds (Fig. b), and novel (hypothetical) compounds (Fig. c). Taken together, these examples illustrate successful CLM predictions. Our study had three primary goals. First, we aimed to devise a novel approach specifically for predicting highly potent compounds from weakly potent input molecules. Thus, rather than striving for prediction of potency values across large ranges, as is conventionally attempted using SVR or other machine learning methods, the primary focus was on potent compounds, in line with the practical relevance of potency predictions. Second, we aimed to generate a structural spectrum of output compounds, ranging from analogues of input molecules to structurally distinct compounds, thereby increasing medicinal chemistry novelty of predicted candidates. Third, it was intended to evaluate the methodology in a way that was not affected by limitations of conventional benchmarking of potency predictions, as discussed above, and enabled a non-ambiguous assessment of the ability to predict potent compounds. To meet the first two goals, which were central to our study, we implemented a CLM consisting of a chemical transformer architecture conditioned on compound potency differences. To meet the third goal, we designed a new compound test system. For model evaluation, a compound pair-based test system was generated using the test set. By design, the fine-tuning and test sets were structurally distinct. Furthermore, in contrast to the fine-tuning set, the test set contained analogue pairs capturing small or large differences in potency (i.e., CCR and AC-CCR pairs, respectively). Table summarizes the composition of the test set. For each activity class, the test set contained varying numbers of CCR pairs and AC-CCR pairs yielding varying numbers of unique CCR and AC-CCR compounds. In the following, SC and TC are used as abbreviations for source (input) and target compound, respectively. For the evaluation of the fine-tuned CLM, test set compounds were divided into instances with maximally 1 μmol potency (corresponding to a pK i value of 6), which served as SCs, and candidate compounds with higher than 1 μmol potency (pK i > 6), which served as known candidate compounds ( KCCs ) for comparison with newly generated TCs. In addition, the model generated varying numbers of novel (hypothetical) TCs. For each activity class, smaller numbers of SCs than KCCs were available. With the exception of activity class 251 (3838 KCCs), the test set contained 366–824 KCCs for the activity classes (Table ), with on average 576 KCCs per class. Each CCR-SC (pK i ≤ 6) and AC-CCR-SC (pK i ≤ 6) was once used as an input compound for the model and in each case, 50 TCs were sampled, canonicalized, and compared to KCCs to search for exact matches, that is, fully reproduced compounds with known potency. Because the model generated novel TCs, probabilities for re-generating known TCs could not be derived in a meaningful way. Consequently, the main measure for establishing proof-of-principle for the ability of the model to predict potent compounds was the reproduction of any KCCs . For each activity class, compound statistics were derived over three independent sampling trials, as reported below. Table reports the possible predictions outcomes for the compound pair-based test system. For each SC, a TC could be a known CCR or AC-CCR compound or a novel (hypothetical) compound representing a TC not contained in the fine-tuning or test set. Taking core structure matches into consideration (that is, a TC either contained the same core structure as a SC or not), a total of 12 formally defined prediction outcomes were possible, including six each for CCR-SCs and AC-CCR-SCs, as identified by indices 1.1.–1.6. and 2.1.–2.6. in Table , respectively. Accordingly, a newly generated compound might be a structural analogue of a given SC (having the same core structure) or contain a different core structure. Furthermore, SCs and TCs might be distinguished by single or multiple substituents. On the basis of this classification scheme, CLM predictions were rigorously evaluated focusing on the reproduction of known active compounds, as explained above. This was the most relevant measure of model performance because it enabled the exact determination of potency differences between SCs and TCs and hence the ability of the CLM to predict highly potent compounds. For novel (hypothetical) compounds generated by the model, no assessment was possible (without subsequent experimental evaluation). For the SCs from all activity classes, systematic compound predictions were carried out using the CLM. The model only produced 0.5–2% invalid SMILES (assessed using RDKit) for all activity classes. With the exception of class 251 (1391 SCs), the test set contained 40–359 SCs for the activity classes, with on average 162 compounds per class (Table ). The predictions were then assessed on the basis of well-defined pair categories detailed above, as reported in Table . For each activity class and compound pair category indexed according to Table (top row), the number of unique TCs produced by the CLM is reported. With the exception of categories 1.5., 1.6., 2.5., and 2.6., which report novel (hypothetical) candidate compounds not contained in the fine-tuning or test set, the TCs represent KCCs, as defined in the text. Encouragingly, for all activity classes, the CLM successfully reproduced large numbers of KCCs for all SCs (categories 1.1.–1.4. and 2.1.–2.4., respectively). Frequently, multiple KCCs were obtained for the same SC. Furthermore, depending on the activity class, the model produced varying numbers of TCs with the same or different core structure, thus confirming its ability to generate frequent core structure transformations. In many cases, more structurally unique TCs were generated than analogues of SCs. Moreover, large numbers of hypothetical candidate compounds not contained in the training set were obtained (categories 1.5.–1.6. and 2.5.–2.6., respectively). The reproducibility of the limited numbers of available KCCs representing known ACs (12–84 unique compounds per activity class) was of particular interest (categories 2.1.–2.4.). AC-CCR KCCs were consistently reproduced and for five activity classes, the total count exceeded the number of unique AC-CCR KCCs per class (due to multiple reproductions of individual KCCs). Table reports statistics for reproduction of KCCs. Reported are statistics for the re-generation of KCCs including the mean number of KCCs over three independent sampling trials and the proportion of reproduced KCCs relative to all available KCCs with standard deviations (±). In addition, the mean number of non-KCCs over three independent trials is provided. The proportion of exactly reproduced KCCs over independent sampling trials ranged from ~ 7 to ~ 37%, depending on the activity class (with generally small standard deviations). For nine, six, and two classes, more than 10, 20, and 30% of all available KCCs were reproduced, respectively. Applying the most rigorous criterion of exact re-generation of known potent compounds as a performance measure (see above), the observed numbers and proportions represented unexpectedly good predictions, which clearly established proof-of-concept for the approach. For each activity class, ASs were also extracted from newly generated (predicted) compounds. Table reports the number of ASs (multiple compounds having the same core structure) and singletons (compounds with unique core structures not belonging to any AS). Depending on the activity class, 90–1414 ASs and 219–1762 singletons were obtained, respectively. Since each AS and singleton contained a unique core structure (scaffold), the core structure diversity of newly generated compounds was generally high. Between 4 and 18% of the core structures contained in the original activity classes (from ASs and singletons) were reproduced by the model, as also reported in Table . Having confirmed the ability of the CLM to generate structurally analogous and diverse TCs including KCCs, the key question then was whether or not the model would produce TCs that had much higher potency than the corresponding SCs. Figure shows the distributions of potency differences between pairs of known source and target compounds with experimental potency values involving compounds from ACs. For five activity classes, the median potency difference fell between one and two orders of magnitude (10–100-fold) and for the other five classes, the median value exceeded two orders of magnitude (100-fold). Furthermore, for all but one class, multiple compounds with at least 1000-fold higher potency than the corresponding SCs were generated (including highly potent statistical outliers). Thus, these observations unambiguously confirmed the ability of the CLM to generate highly potent compounds from weakly potent (micromolar) input molecules. Figure shows exemplary pairs of SCs and newly designed compounds (TCs) with different structural relationships. Given our design strategy, all SCs were known compounds with experimentally determined potency. The generated TCs included known potent analogues of SCs (Fig. a), structurally distinct known potent compounds (Fig. b), and novel (hypothetical) compounds (Fig. c). Taken together, these examples illustrate successful CLM predictions. The underlying idea for the development of the approach reported herein was to predict highly potent compounds from individual weakly potent input molecules. For all practical purposes, this represents an ultimate goal of potency prediction, especially for compound optimization in medicinal chemistry. This prediction task could not be addressed using conventional regression models. In addition, going beyond the applicability domain of standard QSAR modeling, we also aimed to design structurally diverse compounds, in addition to analogues. Therefore, a different methodological framework was required and we adapted a conditional transformer architecture previously used for AC predictions. These predictions established that compound generation could be conditioned on potency differences. However, since AC predictions were also confined to structurally analogous compounds, it remained unclear whether or not potency difference conditioning was transferable to the design of structurally diverse compounds with high potency. The CLM reported herein was fine-tuned on pairs of SCs and TCs with associated potency differences and we then examined its ability to predict structurally diverse compounds with large increases in potency relative to input molecules. Therefore, a compound pair-based test system was generated that covered all possible prediction outcomes and enabled a well-defined and rigorous assessment of model performance. Our analysis confirmed the ability of the model to reproduce known potent compounds not encountered during training at unexpectedly high rates, including both analogues of weakly potent SCs and structurally distinct compounds. With median potency increases close to or above 100-fold across activity classes and multiple predictions with more than 1000-fold increases in compound potency, model performance was generally high. In addition, the CLM also produced large numbers of novel compounds for the activity classes that were not contained in the fine-tuning or test set. Taken together, our findings indicate that the approach reported herein should have considerable potential for practical applications. In compound optimization, we envision that the CLM will be fine-tuned using sets of active compounds for a target of interest and that the predictions will then focus on input compounds prioritized by medicinal chemistry. For these and other applications, the CLM is made freely available as a part of our study.
Expression significance of Emi1, UBCH10 and CyclinB1 in esophageal squamous cell carcinoma	a73f9ac0-8916-4d4f-904c-13175c0d3d12	10164988	Anatomy[mh]	Esophageal cancer is characterized by insidious onset, and patients often seek medical treatment in the middle and late stages, leading to poor prognosis and short survival time . Esophageal squamous cell carcinoma (ESCC) is a highly common subtype of esophageal carcinoma, so it is urgent to further explore the molecular pathology and the mechanism of malignant progression of ESCC. The uncontrolled proliferation of tumor cells is the result of dysregulation of cell cycle . Early mitotic inhibitor 1 (Emi1) mainly plays a role in promoting endogenous inhibitor of anaphase-promoting complex/cyclosome (APC/C), promoting the accumulation of S and G2 related mitotic Cyclin . Ubiquitin-conjugating enzyme 10 (UBCH10) is involved in the degradation of target proteins by the ubiquitin-proteasome system and thus plays a role in regulating the cell cycle . CyclinB1 belongs to the Cyclin family, which plays an important role in regulating the whole process of cell division and proliferation. Abnormal expression of cyclinB1 will cause the cell cycle to stall or exit, and the cell division and proliferation cannot be completed . Currently, it is known that CyclinB1 is the target protein of ubiquitination protein degradation system, in which Emi1 and UBCH10 are participants. Studies have confirmed that Emi1, UBCH10 and CyclinB1 are abnormally expressed in a variety of tumors, and are closely related to the occurrence of tumors . However, the expressions of Emi1, UBCH10 and CyclinB1 in ESCC, as well as whether there is interaction among them to jointly regulate the cell cycle process, are still unknown. Therefore, by exploring the expression of three proteins in ESCC and their correlation with tumor growth, we deeply understand the cell cycle process and regulatory mechanism of ESCC, hoping to provide a new theoretical basis for improving the therapeutic effect and prognosis of ESCC. Tissue sample All histological specimens were collected from confirmed ESCC and paracancer normal mucosal tissues from the Department of Pathology, the First Affiliated Hospital of Zhengzhou University from June 2020 to June 2021. All patients had not received radiotherapy or chemotherapy before surgery, and clinicopathologic data were complete. Among the 50 patients with ESCC, 29 were males and 21 were females. They ranged in age from 48 to 83, with a mean age of 60; 16 cases were classified as grade I-II and 34 cases as grade III. There were 23 cases with lymph node metastasis and 27 cases without lymph node metastasis. There were 27 cases of stage I to stage II and 23 cases of stage III to stage IV. Pathological staging was determined according to the TNM staging criteria for esophageal cancer (8th Edition) jointly published by the International Union Against Cancer (UICC) and the American Cancer Federation (AJCC) in 2017. Reagents Rabbit anti-human Emi1 polyclonal antibody and rabbit anti-human UBCH10 polyclonal antibody were purchased from Proteintech, United States. Rabbit anti-human CyclinB1 monoclonal antibody was purchased from Shanghai Biyuntian Biotechnology Co., LTD. Rabbit anti-human Ki-67 monoclonal antibody was purchased from Shanghai Gene Technology Co., LTD. TUNEL test kit was purchased from Jiangsu Kaiji Biotechnology Co., LTD. In situ hybridization biotin labeled probe was designed and synthesized by Shanghai GenePharma Technology Co., LTD. Immunohistochemistry (IHC) The slices were baked in a 61°C drying oven for 150min, and then dewaxed and hydrated. High pressure heating was used for antigen repair. Slices were added with appropriate 3% H 2 O 2 drops, and then normal goat serum working solution was added. Add appropriate amount of primary antibody working solution and put the wet box in the refrigerator at 4°C overnight. On the second day, biotin labeled secondary antibody was added, then horseradish peroxidase labeled chain enzyme ovalbumin was added, and DAB working solution was added finally. ESCC tissue slices with known positive expressions of Emi1, UBCH10, CyclinB1 and Ki-67 were used as positive controls. PBS buffer will be used instead of primary antibody as the negative control. Score by number of positive cells/percentage of observed cells. <1% is 0 point, 1%–25% is 1 point, 26%–50% is 2 point, 51%–75% is 3 point, >76% is 4 point. Score according to cell staining intensity: no color development is 0 point, light yellow is 1 point, brown and yellow is 2 point, tan is 3 point. Finally, the percentage score of positive cells was multiplied by the score of cell staining intensity to obtain the final score. The score ≤4 is negative, and the score >4 is positive. In-situ hybridization (ISH) The in-situ hybridization biotin labeled probe sequence is shown below. Emi1 probe sequence: 5′-CAACTATCCGAGGGTCGAGG-3′. UBCH10 probe sequence: 5′-CAGGGCTCCTGGCTGGTGACCTGCTT-3′. CyclinB1 probe sequence: 5′-CAGTGACTTCCCGACCCAGTAGGTATTT-3′. The slices were baked in a 61°C drying oven for 150 min, and then dewaxed and hydrated. The slices were then dripped with an appropriate amount of 3% H 2 O 2 , followed by 0.3% Triton-X100. After pepsin was added, the pre-hybridization solution was added and incubated at 40°C for 3 h in a wet box. Biotin-labeled oligonucleotide probes were added and incubated overnight in a wet box at 42°C. On the second day, wash with SSC solution, add appropriate amount of sealing solution, and incubate in a wet box at 37°C for 30 min. Appropriate amount of SA-AP was added and incubated in a wet box at 37°C for 30 min. Appropriate amount of BCIP/NBT color developing working solution was added and incubated at 37°C in a wet box for 30–60 min. Appropriate amount of nuclear solid red dye solution was added to restain the nucleus for 5–15 min. ESCC slices with known positive expressions of Emi1, UBCH10 and CyclinB1 were used as positive controls. Hybridization solution without probe was used as negative control. Score by number of positive cells/percentage of observed cells. <10% is 1 point, 11%–30% is 2 point, 31%–70% is 3 point, >70% is 4 point. Score according to cell staining intensity: no color development is 0 point, light blue is 1 point, darker purple blue is 2 point, deep purple blue is 3 point. Finally, the percentage score of positive cells was multiplied by the score of cell staining intensity to obtain the final score. The score <1 is negative, and the score ≥1 is positive. In-situ end labeling (TUNEL) The slices were baked in a 61°C drying oven for 150 min, and then dewaxed and hydrated. Microwave heating was used to repair antigen. The slices was dripped with 3% H 2 O 2 , and then the sealer was dripped. Add TdT enzyme working solution and incubate at 37°C for 60 min in the dark. Streptavidin-HRP working solution was added and incubated at 37°C for 30 min in the dark. DAB working solution was added and color was developed for 3–7 min. The known positive tissue slices were treated with Dnase I as positive control. Reaction solution without TDT enzyme was used as negative control. The degree of apoptosis was suggested by the apoptosis index (AI), which was calculated according to the number of positive cells/percentage of observed cells: AI = number of positive cells/200100%. Statistical analysis SPSS21.0 (United States) software was used for statistical analysis. Measurement data were expressed as ‾X ± S, and t -test was used to compare the differences between the two groups. Comparison of count data were performed by χ 2 test or Fisher’s exact probability method, and correlation analysis was performed by Spearman method. Test level α = 0.05, p < 0.05 was considered statistically significant. All histological specimens were collected from confirmed ESCC and paracancer normal mucosal tissues from the Department of Pathology, the First Affiliated Hospital of Zhengzhou University from June 2020 to June 2021. All patients had not received radiotherapy or chemotherapy before surgery, and clinicopathologic data were complete. Among the 50 patients with ESCC, 29 were males and 21 were females. They ranged in age from 48 to 83, with a mean age of 60; 16 cases were classified as grade I-II and 34 cases as grade III. There were 23 cases with lymph node metastasis and 27 cases without lymph node metastasis. There were 27 cases of stage I to stage II and 23 cases of stage III to stage IV. Pathological staging was determined according to the TNM staging criteria for esophageal cancer (8th Edition) jointly published by the International Union Against Cancer (UICC) and the American Cancer Federation (AJCC) in 2017. Rabbit anti-human Emi1 polyclonal antibody and rabbit anti-human UBCH10 polyclonal antibody were purchased from Proteintech, United States. Rabbit anti-human CyclinB1 monoclonal antibody was purchased from Shanghai Biyuntian Biotechnology Co., LTD. Rabbit anti-human Ki-67 monoclonal antibody was purchased from Shanghai Gene Technology Co., LTD. TUNEL test kit was purchased from Jiangsu Kaiji Biotechnology Co., LTD. In situ hybridization biotin labeled probe was designed and synthesized by Shanghai GenePharma Technology Co., LTD. The slices were baked in a 61°C drying oven for 150min, and then dewaxed and hydrated. High pressure heating was used for antigen repair. Slices were added with appropriate 3% H 2 O 2 drops, and then normal goat serum working solution was added. Add appropriate amount of primary antibody working solution and put the wet box in the refrigerator at 4°C overnight. On the second day, biotin labeled secondary antibody was added, then horseradish peroxidase labeled chain enzyme ovalbumin was added, and DAB working solution was added finally. ESCC tissue slices with known positive expressions of Emi1, UBCH10, CyclinB1 and Ki-67 were used as positive controls. PBS buffer will be used instead of primary antibody as the negative control. Score by number of positive cells/percentage of observed cells. <1% is 0 point, 1%–25% is 1 point, 26%–50% is 2 point, 51%–75% is 3 point, >76% is 4 point. Score according to cell staining intensity: no color development is 0 point, light yellow is 1 point, brown and yellow is 2 point, tan is 3 point. Finally, the percentage score of positive cells was multiplied by the score of cell staining intensity to obtain the final score. The score ≤4 is negative, and the score >4 is positive. hybridization (ISH) The in-situ hybridization biotin labeled probe sequence is shown below. Emi1 probe sequence: 5′-CAACTATCCGAGGGTCGAGG-3′. UBCH10 probe sequence: 5′-CAGGGCTCCTGGCTGGTGACCTGCTT-3′. CyclinB1 probe sequence: 5′-CAGTGACTTCCCGACCCAGTAGGTATTT-3′. The slices were baked in a 61°C drying oven for 150 min, and then dewaxed and hydrated. The slices were then dripped with an appropriate amount of 3% H 2 O 2 , followed by 0.3% Triton-X100. After pepsin was added, the pre-hybridization solution was added and incubated at 40°C for 3 h in a wet box. Biotin-labeled oligonucleotide probes were added and incubated overnight in a wet box at 42°C. On the second day, wash with SSC solution, add appropriate amount of sealing solution, and incubate in a wet box at 37°C for 30 min. Appropriate amount of SA-AP was added and incubated in a wet box at 37°C for 30 min. Appropriate amount of BCIP/NBT color developing working solution was added and incubated at 37°C in a wet box for 30–60 min. Appropriate amount of nuclear solid red dye solution was added to restain the nucleus for 5–15 min. ESCC slices with known positive expressions of Emi1, UBCH10 and CyclinB1 were used as positive controls. Hybridization solution without probe was used as negative control. Score by number of positive cells/percentage of observed cells. <10% is 1 point, 11%–30% is 2 point, 31%–70% is 3 point, >70% is 4 point. Score according to cell staining intensity: no color development is 0 point, light blue is 1 point, darker purple blue is 2 point, deep purple blue is 3 point. Finally, the percentage score of positive cells was multiplied by the score of cell staining intensity to obtain the final score. The score <1 is negative, and the score ≥1 is positive. end labeling (TUNEL) The slices were baked in a 61°C drying oven for 150 min, and then dewaxed and hydrated. Microwave heating was used to repair antigen. The slices was dripped with 3% H 2 O 2 , and then the sealer was dripped. Add TdT enzyme working solution and incubate at 37°C for 60 min in the dark. Streptavidin-HRP working solution was added and incubated at 37°C for 30 min in the dark. DAB working solution was added and color was developed for 3–7 min. The known positive tissue slices were treated with Dnase I as positive control. Reaction solution without TDT enzyme was used as negative control. The degree of apoptosis was suggested by the apoptosis index (AI), which was calculated according to the number of positive cells/percentage of observed cells: AI = number of positive cells/200100%. SPSS21.0 (United States) software was used for statistical analysis. Measurement data were expressed as ‾X ± S, and t -test was used to compare the differences between the two groups. Comparison of count data were performed by χ 2 test or Fisher’s exact probability method, and correlation analysis was performed by Spearman method. Test level α = 0.05, p < 0.05 was considered statistically significant. The expression difference of Emi1, UBCH10 and CyclinB1 proteins in ESCC and paracancer tissues The expressions of Emi1, UBCH10 and CyclinB1 proteins in ESCC and paracancer tissues were detected by IHC. The results showed that Emi1, UBCH10 and CyclinB1 proteins were highly expressed in ESCC tissues, as shown in and . There were significant differences in protein expression between ESCC and paracancer tissues ( p < 0.05). The expression difference of Emi1, UBCH10 and CyclinB1 mRNA in ESCC and paracancer tissues The expressions of Emi1, UBCH10 and CyclinB1 mRNA in ESCC and paracancer tissues were detected by ISH. The results showed that Emi1, UBCH10 and CyclinB1 mRNA were highly expressed in ESCC tissues, as shown in and . mRNA expression was significantly different between ESCC and paracancer tissues ( p < 0.05). Correlation between the expressions of Emi1, UBCH10, CyclinB1 and clinicopathological indexes The correlation between Emi1, UBCH10, CyclinB1 and clinicopathological indexes was analyzed, including gender, age, tumor diameter, tissue grade, depth of invasion, lymph node metastasis, and pathological stage. The results showed that the expression of Emi1 protein and mRNA were correlated with tissue grade, lymph node metastasis and pathological stage ( p < 0.05), as shown in . UBCH10 protein and mRNA expression were correlated with tissue grade, lymph node metastasis and pathological stage ( p < 0.05), as shown in . CyclinB1 protein is correlated with tissue grade, lymph node metastasis, and pathological stage, while CyclinB1 mRNA is only correlated with tissue grade ( p < 0.05). The difference between CyclinB1 protein and mRNA may be related to the small tissue sample size, as shown in . Correlation of Emi1, UBCH10 and CyclinB1 expression The correlation of protein expression among Emi1, UBCH10 and CyclinB1 in ESCC tissues was analyzed, as shown in . The results showed that Emi1 was positively correlated with UBCH10 protein expression ( r = 0.5418, p < 0.0001). Emi1 was positively correlated with the expression of CyclinB1 ( r = 0.5539, p < 0.0001). UBCH10 was positively correlated with the expression of CyclinB1 ( r = 0.6020, p < 0.0001). The mRNA expression correlation among Emi1, UBCH10 and CyclinB1 in ESCC tissues was further analyzed, as shown in . The results showed that Emi1 was positively correlated with UBCH10 mRNA expression ( r = 0.4181, p = 0.0025). Emi1 was positively correlated with the expression of CyclinB1 mRNA ( r = 0.7357, p < 0.0001). UBCH10 was positively correlated with the expression of CyclinB1 mRNA ( r = 0.5997, p < 0.0001). Proliferation and apoptosis in ESCC and paracancer tissues The expression of Ki-67 protein in ESCC and paracancer tissues was detected by IHC. Ki-67 is a nuclear antigen closely related to cell proliferation. The proliferation index was determined by evaluating the expression of Ki-67 in ESCC and paracancer tissues, as shown in and . The results showed that the ESCC tissue presented a very obvious high proliferation index, with a mean of 60.40%, while the proliferation index of the paracancer tissue was significantly reduced, about 11.90%, compared with the tumor tissue ( p < 0.05). Apoptosis in ESCC and paracancer tissues was determined by TUNEL technique, as shown in and . The results showed that the apoptotic index was about 29.60% in ESCC tissue and 42.20% in paracarcinoma tissue, and the apoptotic index was lower in ESCC tissue. The proliferation and apoptosis indexes of ESCC and paracancer tissues were significantly different ( p < 0.05). Correlation between Emi1, UBCH10, CyclinB1 expression and tumor proliferation The correlation between Emi1, UBCH10 and CyclinB1 protein expression and proliferation index in ESCC tissues was analyzed, as shown in . The results showed that the protein expression of Emi1, UBCH10 and CyclinB1 was positively correlated with the proliferation index ( r = 0.4561, p = 0.0009) ( r = 0.4082, p = 0.0033) ( r = 0.4300, p = 0.0018). The correlation between Emi1, UBCH10 and CyclinB1 mRNA expression and proliferation index in ESCC tissues was analyzed, as shown in . The results showed that the mRNA expression of Emi1, UBCH10 and CyclinB1 was positively correlated with the proliferation index ( r = 0.5326, p < 0.0001) ( r = 0.5764, p < 0.0001) ( r = 0.6794, p < 0.0001). Correlation between expression of Emi1, UBCH10, CyclinB1 and tumor apoptosis The correlation between Emi1, UBCH10, CyclinB1 protein expression and apoptosis index in ESCC tissues was analyzed, as shown in . The results showed that the protein expressions of Emi1, UBCH10 and CyclinB1 were negatively correlated with the apoptosis index ( r = −0.5737, p < 0.0001) ( r = −0.4178, p = 0.0025) ( r = −0.4939, p = 0.0018). The correlation between Emi1, UBCH10 and CyclinB1 mRNA expression and apoptosis index in ESCC tissues was analyzed, as shown in . The results showed that the mRNA expressions of Emi1, UBCH10 and CyclinB1 were negatively correlated with the apoptosis index ( r = −0.4614, p = 0.0007) ( r = −0.3450, p = 0.0141) ( r = −0.4742, p = 0.0005). The expressions of Emi1, UBCH10 and CyclinB1 proteins in ESCC and paracancer tissues were detected by IHC. The results showed that Emi1, UBCH10 and CyclinB1 proteins were highly expressed in ESCC tissues, as shown in and . There were significant differences in protein expression between ESCC and paracancer tissues ( p < 0.05). The expressions of Emi1, UBCH10 and CyclinB1 mRNA in ESCC and paracancer tissues were detected by ISH. The results showed that Emi1, UBCH10 and CyclinB1 mRNA were highly expressed in ESCC tissues, as shown in and . mRNA expression was significantly different between ESCC and paracancer tissues ( p < 0.05). The correlation between Emi1, UBCH10, CyclinB1 and clinicopathological indexes was analyzed, including gender, age, tumor diameter, tissue grade, depth of invasion, lymph node metastasis, and pathological stage. The results showed that the expression of Emi1 protein and mRNA were correlated with tissue grade, lymph node metastasis and pathological stage ( p < 0.05), as shown in . UBCH10 protein and mRNA expression were correlated with tissue grade, lymph node metastasis and pathological stage ( p < 0.05), as shown in . CyclinB1 protein is correlated with tissue grade, lymph node metastasis, and pathological stage, while CyclinB1 mRNA is only correlated with tissue grade ( p < 0.05). The difference between CyclinB1 protein and mRNA may be related to the small tissue sample size, as shown in . The correlation of protein expression among Emi1, UBCH10 and CyclinB1 in ESCC tissues was analyzed, as shown in . The results showed that Emi1 was positively correlated with UBCH10 protein expression ( r = 0.5418, p < 0.0001). Emi1 was positively correlated with the expression of CyclinB1 ( r = 0.5539, p < 0.0001). UBCH10 was positively correlated with the expression of CyclinB1 ( r = 0.6020, p < 0.0001). The mRNA expression correlation among Emi1, UBCH10 and CyclinB1 in ESCC tissues was further analyzed, as shown in . The results showed that Emi1 was positively correlated with UBCH10 mRNA expression ( r = 0.4181, p = 0.0025). Emi1 was positively correlated with the expression of CyclinB1 mRNA ( r = 0.7357, p < 0.0001). UBCH10 was positively correlated with the expression of CyclinB1 mRNA ( r = 0.5997, p < 0.0001). The expression of Ki-67 protein in ESCC and paracancer tissues was detected by IHC. Ki-67 is a nuclear antigen closely related to cell proliferation. The proliferation index was determined by evaluating the expression of Ki-67 in ESCC and paracancer tissues, as shown in and . The results showed that the ESCC tissue presented a very obvious high proliferation index, with a mean of 60.40%, while the proliferation index of the paracancer tissue was significantly reduced, about 11.90%, compared with the tumor tissue ( p < 0.05). Apoptosis in ESCC and paracancer tissues was determined by TUNEL technique, as shown in and . The results showed that the apoptotic index was about 29.60% in ESCC tissue and 42.20% in paracarcinoma tissue, and the apoptotic index was lower in ESCC tissue. The proliferation and apoptosis indexes of ESCC and paracancer tissues were significantly different ( p < 0.05). The correlation between Emi1, UBCH10 and CyclinB1 protein expression and proliferation index in ESCC tissues was analyzed, as shown in . The results showed that the protein expression of Emi1, UBCH10 and CyclinB1 was positively correlated with the proliferation index ( r = 0.4561, p = 0.0009) ( r = 0.4082, p = 0.0033) ( r = 0.4300, p = 0.0018). The correlation between Emi1, UBCH10 and CyclinB1 mRNA expression and proliferation index in ESCC tissues was analyzed, as shown in . The results showed that the mRNA expression of Emi1, UBCH10 and CyclinB1 was positively correlated with the proliferation index ( r = 0.5326, p < 0.0001) ( r = 0.5764, p < 0.0001) ( r = 0.6794, p < 0.0001). The correlation between Emi1, UBCH10, CyclinB1 protein expression and apoptosis index in ESCC tissues was analyzed, as shown in . The results showed that the protein expressions of Emi1, UBCH10 and CyclinB1 were negatively correlated with the apoptosis index ( r = −0.5737, p < 0.0001) ( r = −0.4178, p = 0.0025) ( r = −0.4939, p = 0.0018). The correlation between Emi1, UBCH10 and CyclinB1 mRNA expression and apoptosis index in ESCC tissues was analyzed, as shown in . The results showed that the mRNA expressions of Emi1, UBCH10 and CyclinB1 were negatively correlated with the apoptosis index ( r = −0.4614, p = 0.0007) ( r = −0.3450, p = 0.0141) ( r = −0.4742, p = 0.0005). Although a series of early screening work has been carried out in areas with high incidence of esophageal cancer, showing preliminary tumor prevention and control effect, the tumor burden of esophageal cancer is still very large. Finding new genes and proteins that may play an important role in the occurrence and development of ESCC and elucidating their related molecular mechanisms will be of great significance in the diagnosis and treatment of ESCC. Normal cells proliferate through mitosis and are regulated by a variety of cyclins, Cyclin-dependent protein kinases (CDKs), cell cycle checkpoints and cell cycle signaling pathways . The level of cyclin determines the mitotic process of cells. When the level of cyclin decreases, the cell cycle will actively stop or even exit the mitotic process. Protein ubiquitination plays an important role in the stabilization of cyclin levels . Emi1 was initially thought to be a protein involved in the regulation of cell cycle. However, with the development of tumor research, researchers found that the “identity” of Emi1 in the development of tumor was actually oncogene . However, there are few reports on the expression of Emi1 in ESCC tissues, so this study started from exploring the expression of Emi1 gene in ESCC. We found that Emi1 mRNA and protein expressions in ESCC tumor tissues were higher than those in paracancer normal esophageal mucosa tissues, confirming the view that Emi1 is an oncogenic gene. Moreover, Emi1 expression was correlated with tumor differentiation, lymph node metastasis, and pathological stage, and was positively correlated with proliferation and negatively correlated with apoptosis, suggesting that Emi1 played an important role in the malignant process of ESCC. The findings of Emi1 in breast cancer are similar to our findings, Emi1 mediates the anti-apoptotic and pro-proliferative carcinogenicity of Skp2 through PI3K/Akt signaling pathway . UBCH10 is an active protein in ubiquitin-proteasome. However, in addition to regulating ubiquitination degradation of protein, UBCH10 is also closely related to tumor proliferation and metastasis, and may be a new tumor marker or therapeutic target . Our study found that UBCH10 was significantly highly expressed in ESCC, which was related to tumor differentiation, lymph node metastasis, and pathological stage, and was positively correlated with tumor proliferation index and negatively correlated with apoptosis index, suggesting that UBCH10 could promote tumor proliferation and inhibit tumor apoptosis in ESCC. Compared with ESCC, UBCH10 has been extensively studied in other tumors. In glioma, UBCH10 expression increases with the increase of tumor malignancy. After siRNA silencing UBCH10, glioma cells show growth inhibition, cell cycle arrest and increased apoptosis . UBCH10 in lung cancer is also related to tumor differentiation and patient survival, and UBCH10 can lead to P53 and EGFR gene mutations, resulting in loss of tumor inhibitory effect of P53 gene and enhancement of growth promoting effect of EGFR, while the proliferation of lung cancer cells and drug resistance of chemotherapy drugs are weakened after UBCH10 gene silencing . UBCH10 is associated with ER and Ki-67 in breast cancer. Silencing UBCH10 can inhibit the proliferation of tumor cells and increase the sensitivity to chemotherapy. UBCH10 has also been detected in circulating tumor cells, suggesting that it can be used as an indicator for early screening and diagnosis of breast cancer . CyclinB1 is a “star molecule” in the cyclin family and a core protein that regulates the G2 phase of the cell cycle. Similar to UBCH10, CyclinB1 not only has cell cycle regulation function, but also is an oncogenic gene associated with abnormal tumor proliferation. Through the detection of CyclinB1 mRNA and protein in ESCC tissues, it was found that the high expression of CyclinB1 in ESCC tissues was positively correlated with the proliferation index and negatively correlated with the apoptosis index. The expression of CyclinB1 protein was correlated with tumor grade, lymph node metastasis and pathological stage. Other researchers have found evidence that confirms our results, The expression of CyclinB1 was downregulated in cervical cancer cells, and it was found that tumor progression was inhibited and cell cycle was stagnated in G2/M phase, which ultimately delayed the process of tumor development . When the level of CyclinB1 was reduced in breast cancer cells, the proliferation and migration of tumor cells were inhibited, and the cell cycle was stagnated in G2/M phase . Emi1 is involved in cell cycle regulation by acting as an endogenous inhibitor of APC/C and hindering the degradation of its substrates by APC/C . UBCH10, as a member of ubiquitin binding enzyme E2 family, binds to ubiquitin ligase E3 to form a complex to initiate ubiquitin-proteasome degradation pathway, degrades APC/C substrate CyclinB, and finally plays a role in regulating cell cycle . CyclinB1 is mainly involved in the regulation of G2 phase of the cell cycle. In late mitosis, CyclinB1 combines with CDK1 to form a complex, Cyclinb1-CDK1, which phosphorylates APC/C and then promotes the degradation of CyclinB1 . Emi1, UBCH10 and CyclinB1 are all participants in the cell cycle regulation mechanism and are key points in the APC/C molecular mechanism network. By analyzing the correlation between the expressions of Emi1, UBCH10 and CyclinB1, the results confirmed that there was a positive correlation between Emi1, UBCH10 and CyclinB1 at both protein level and mRNA level. The degradation of cyclins is mainly dependent on ubiquitination, among which APC/C is an important ubiquitin-binding enzyme E3, and the degradation substrates of APC/C include a variety of proteins such as CyclinA and CyclinB. The silencing of Emi1 in cells will cause a large number of ubiquitination substrate proteins of APC/C, resulting in ubiquitination degradation of CyclinA and CyclinB, and insufficient accumulation of cyclin in cells, which cannot enter the next phase, and eventually lead to cell cycle arrest . However, overexpression of Emi1 in cells will block the catalytic site of APC/C and competitively prevent APC/C substrate proteins from binding to APC/C co-receptors, resulting in increased CyclinA and CyclinB levels and disorder of cell proliferation cycle . Normal cell cycle arrest and cell physiological apoptosis can not be carried out, which further leads to malignant cell proliferation, promotes cell proliferation and inhibits cell apoptosis. In-depth analysis of the inhibitory effect and mechanism of Emi1 revealed that Emi1 protein contains a variety of domains that block the substrate binding site of APC/C and inhibit the formation of ubiquitin chains . As ubiquitin binding enzyme E2, UBCH10 has the function of forming and extending ubiquitin chains, which can connect the substrate protein and ubiquitin ligase E3 through the ubiquitin chain, label ubiquitin on the substrate protein, and initiate the ubiquitination degradation process . In normal cells, Emi1 can effectively inhibit UBCH10 ubiquitin chain extension, thus effectively stabilizing the level of substrate protein . At present, the specific molecular mechanism of Emi1, UBCH10 and CyclinB1 genes in promoting tumor proliferation and inhibiting apoptosis has not been reported. Combined with our experimental results, we conjectured that Emi1 reduced UBCH10 consumption by inhibiting ubiquitin chain extension between UBCH10 and CyclinB1, and inhibited the ubiquitination degradation of CyclinB1 protein, resulting in high expression of Emi1, UBCH10 and CyclinB1 in tumor tissues. Cell cycle regulation is a complex and huge mechanism network. In addition to the influence of UBCH10 and CyclinB1 proteins, Emi1 may also have other molecular effects, which requires further discussion.
Learning from the COVID-19 pandemic: health care disturbances and telemedicine as an alternative rheumatology practice in Indonesia	a1ae45c5-4204-467e-afa2-bec9975ccd01	10165285	Internal Medicine[mh]	Coronavirus disease 2019 (COVID-19) was first identified in December 2019. Due to the highly pathogenic and rapid spread of SARS-CoV-2 worldwide, the World Health Organization (WHO) declared COVID-19 a global public health concern. To control the spread, countries, including Indonesia, implemented health protocols to regulate social distancing behaviors and restrict nonvital social mobility, such as public transportation and traveling across regions or borders . Besides, autoimmune rheumatic disease (ARD) patients are at risk of being infected and may have a poorer outcome of SARS-CoV-2 infection due to immune dysregulation and immunosuppressive effects of ARD treatment, which increases patients’ concerns about being infected with COVID-19 and promote preventive social distancing behavior . Moreover, the lockdown policy was a barrier to access for patients. These problems became a challenge for ARD patients in accessing continuing medication and disease monitoring to prevent the progression of the disease and worse outcomes . Thus, this study provided new insight into factors that affected health care disruptions and treatment interruptions. Alternatively, the WHO advised switching to telemedicine during the pandemic . Moreover, the Asia-Pacific League of Associations for Rheumatology (APLAR) published recommendations for telemedicine in rheumatology practice in early 2022, . in addition to the Indonesian government telemedicine guideline during the pandemic . Prior to the pandemic, telemedicine was used to provide teleconsultation across regions, nations, and even international space stations. In rheumatology, telemedicine has been used to monitor chronic autoimmune diseases and even deliver biological agents . Hence, we evaluated patients’ telemedicine reception as an alternative in health care practice in the pandemic situation in Indonesia. Currently, in 2023, the COVID-19 pandemic status shifted into an endemic status, and lockdown policies are gradually being rescinded. However, due to Indonesia’s unique geography and sociodemographic conditions, barriers to access remain a major challenge. Finally, learning from this pandemic, there should be considerable changes in rheumatology practice to meet patients’ needs. Study population A cross-sectional study was conducted in Indonesia using a national online survey (Google Forms). With the support of ARD communities and rheumatologists in several rheumatology centers in Indonesia, we recruited participants to fill out an online submission form to be collected consecutively from September to December 2021. The inclusion criteria were as follows: (1) respondents had to be at least 18 years of age; (2) respondents who lived in Indonesia during the first and second COVID-19 wave; and (3) respondents who were diagnosed with one or more autoimmune rheumatic disease(es) such as rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), psoriatic arthritis (PsA), ankylosing spondylitis (AS), Sjögren’s syndrome, and scleroderma/systemic sclerosis. The specific criterion for telemedicine was as follows: (1) Respondents who used conventional or hospital telemedicine services during the COVID-19 pandemic. The exclusion criterion was as follows: (1) respondents who did not agree to participate in this study. Data collection The study protocol was provided on the first page of the online questionnaire, and each participant gave consent to continue to participate or decline to participate voluntarily. All responses were anonymous, and each respondent could only complete the questionnaire once. The questionnaire was conducted in Bahasa, Indonesia. The questionnaire was a self-developed questionnaire that was built based on available previous studies using the health belief model and the health-seeking behavior theory (Additional File ) . The collected data were pooled in a Google spreadsheet. The questionnaire was divided into three sections: the first section evaluated demographics, medical history, and current treatment; the second section evaluated respondents’ concerns, behaviors, and external factors that contributed to health care disruptions; and the third section assessed respondents’ satisfaction with telemedicine as an alternative type of health care consultation during the COVID-19 pandemic. The questionnaire regarding patient concerns about COVID-19 infection, social distancing behavior, and satisfaction was rated on a five-point Likert scale as follows: “never (1),“ “rarely (2),“ “sometimes” (3), “often (4),“ and “always (5)”. We used the respondent’s district location to determine the level of travel restrictions by the government during the second wave of the COVID-19 pandemic based on the Ministry of Home Affairs’s instructions for the lockdown policy . For the outcome, we designed the questionnaire to assess health care disruptions and treatment interruptions as “yes” or “no” questions. The questionnaire regarding satisfaction with telemedicine was rated on a five-point Likert scale as follows: “very dissatisfied (1),“ “dissatisfied (2),“ “neutral (3),“ “satisfied (4),“ and “very satisfied (5).“ Statistical analysis Collected data were analyzed using SPSS 25 (SPSS Inc., Chicago, IL, USA) for Windows 11. Continuous variables are presented as the means (standard deviations) when normally distributed and medians (minimums-maximums) when not normally distributed. Categorical variables are presented as percentages. The chi-square test explored the association between patients’ concerns, social distancing behaviors, and external factors in health care disruptions such as avoid a hospital visit and treatment interruptions such as stopping the medication. All statistical tests were 2-sided, P values less than 0.05 were considered significant, and P values less than 0.001 were considered highly significant. A cross-sectional study was conducted in Indonesia using a national online survey (Google Forms). With the support of ARD communities and rheumatologists in several rheumatology centers in Indonesia, we recruited participants to fill out an online submission form to be collected consecutively from September to December 2021. The inclusion criteria were as follows: (1) respondents had to be at least 18 years of age; (2) respondents who lived in Indonesia during the first and second COVID-19 wave; and (3) respondents who were diagnosed with one or more autoimmune rheumatic disease(es) such as rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), psoriatic arthritis (PsA), ankylosing spondylitis (AS), Sjögren’s syndrome, and scleroderma/systemic sclerosis. The specific criterion for telemedicine was as follows: (1) Respondents who used conventional or hospital telemedicine services during the COVID-19 pandemic. The exclusion criterion was as follows: (1) respondents who did not agree to participate in this study. The study protocol was provided on the first page of the online questionnaire, and each participant gave consent to continue to participate or decline to participate voluntarily. All responses were anonymous, and each respondent could only complete the questionnaire once. The questionnaire was conducted in Bahasa, Indonesia. The questionnaire was a self-developed questionnaire that was built based on available previous studies using the health belief model and the health-seeking behavior theory (Additional File ) . The collected data were pooled in a Google spreadsheet. The questionnaire was divided into three sections: the first section evaluated demographics, medical history, and current treatment; the second section evaluated respondents’ concerns, behaviors, and external factors that contributed to health care disruptions; and the third section assessed respondents’ satisfaction with telemedicine as an alternative type of health care consultation during the COVID-19 pandemic. The questionnaire regarding patient concerns about COVID-19 infection, social distancing behavior, and satisfaction was rated on a five-point Likert scale as follows: “never (1),“ “rarely (2),“ “sometimes” (3), “often (4),“ and “always (5)”. We used the respondent’s district location to determine the level of travel restrictions by the government during the second wave of the COVID-19 pandemic based on the Ministry of Home Affairs’s instructions for the lockdown policy . For the outcome, we designed the questionnaire to assess health care disruptions and treatment interruptions as “yes” or “no” questions. The questionnaire regarding satisfaction with telemedicine was rated on a five-point Likert scale as follows: “very dissatisfied (1),“ “dissatisfied (2),“ “neutral (3),“ “satisfied (4),“ and “very satisfied (5).“ Collected data were analyzed using SPSS 25 (SPSS Inc., Chicago, IL, USA) for Windows 11. Continuous variables are presented as the means (standard deviations) when normally distributed and medians (minimums-maximums) when not normally distributed. Categorical variables are presented as percentages. The chi-square test explored the association between patients’ concerns, social distancing behaviors, and external factors in health care disruptions such as avoid a hospital visit and treatment interruptions such as stopping the medication. All statistical tests were 2-sided, P values less than 0.05 were considered significant, and P values less than 0.001 were considered highly significant. Demographics of the respondents We excluded three out of 314 recruited respondents because they did not meet the inclusion criteria. The remaining 311 respondents’ demographic and baseline characteristics are summarized in Table . The median age of respondents was 40 years (min 21-max 68), and the majority of the respondents were female (299, 96.1%). SLE was the most common ARD in this study (217, 70.1%), followed by RA (86, 27.3%). In addition, 54 (17.4%) respondents had more than one ARD diagnosis. Most of the respondents received nonbiologic disease-modifying antirheumatic drugs (DMARDs) (74.3%) and glucocorticoids (71.4%). Approximately 81 (26%) respondents reported avoiding hospital visits, and 76 (24.4%) respondents had ever stopped ARD medication without medical advice. Factors affecting health care and treatment disruptions during the COVID-19 pandemic First, regarding the respondents’ concerns about being infected with COVID-19, the majority were neutral (38,6%), followed by 27.7% who were always concerned and 24.4% who were often concerned about being infected with COVID-19. Regarding basic knowledge about COVID-19 infection, the majority of the respondents were always (30.2%) and often (30.2%) concerned that ARD conditions may increase the risk of COVID-19 infection, followed by those who were neutral (29.3%). Moreover, some respondents were always (37.6%) concerned that COVID-19 disease may worsen their ARD condition, and some were always (39.2%) concerned that COVID-19 symptoms among people with ARDs are more severe than those among healthy people. In addition, as seen in Fig. , the mean overall concern score was 3.9 ± 0.9 from the maximum score of 5. Second, regarding respondents’ social distancing behaviors during the COVID-19 pandemic, the majority of our respondents sometimes (38.9%) avoided leaving the house, followed by those who often (22.2%) avoided leaving the house. In contrast, 35.4% of our respondents sometimes (35.4%) avoided going to the grocery store, followed by those who never (19%) and rarely (18%) avoided going to the grocery store. Moreover, 38.6% sometimes avoided meeting in person with colleagues and families, followed by those who often (23.2%) avoided meeting in person. As shown in Fig. , the mean overall social distancing behavior score was 3.0 ± 1.0 from the maximum score of 5. In addition, as seen in Fig. , respondents’ concerns were positively correlated with their social distancing behaviors during the pandemic (p value 0.000, r 0.458). Third, regarding external factors that contributed to health care disruptions, most of our respondents lived on the Java and Bali islands, where 245 (78.8%) lived in level 4 PPKM regions (Pemberlakuan Pembatasan Kegiatan Masyarakat/Enforcement of Restrictions on Community Activities). PPKM level 4 was the highest level of this community activity restriction policy, and the implication of this was that 240 (77.2%) of our respondents had blocked access to the hospital during the implementation of PPKM. Moreover, the majority of our respondents used private transportation (64.3%) compared to public transportation (35.7%) and both modes of transportation were severely affected by PPKM. Furthermore, the bivariate analysis of all variables related is shown in Table . First, there was a significant association between sex and stopping medication without medical advice (p value 0.005). There was also an association between respondents’ concerns about infection, social distancing behaviors during the pandemic, and blocked access during PPKM (p value 0.014, 0.001, and 0.045). Finally, multivariate analysis of the respondents’ concerns, behaviors, and blocked access to the hospital is shown in Table . A strong association was found between blocked access to the hospital during the pandemic and the need to avoid hospital visits (OR 1.786; CI: 1.008–3.162) and between sex and stopping medication without medical advice (OR: 4.667; CI: 1.436–15.168). The use of telemedicine and the respondents’ reception As shown in Table , only 81 of 311 (26.0%) respondents reported that they used telemedicine as an alternative consultation method during the COVID-19 pandemic. Several reasons were recorded among nonusers of telemedicine, and the most common reasons were ‘I don’t know about telemedicine’ (126 (54.8%)), followed by ‘I do not need telemedicine services’ (48 (20.9%)), and ‘No telemedicine service is available’ (27 (11.7%)). The respondents’ characteristics regarding telemedicine use are illustrated in Table . Among telemedicine users, the mean age was 40 ± 12.8 years, and the majority were female (75, 92.6%). SLE was the most common ARD in this study (57, 70.4%), followed by RA (12, 14.8%), with 14 (17.3%) respondents having more than one ARD diagnosis. In addition, most of the respondents received nonbiologic DMARDs (80.2%), followed by glucocorticoids (77.8%). Furthermore, the most preferred modality among the respondents was text/chat consultations (n = 56, 69.1%), followed by video consultations (n = 44, 54.3%), and 27 (33.3%) preferred to fill out an online form for a direct appointment. Finally, approximately 44 (54.3%) respondents preferred telemedicine during the COVID-19 pandemic, and 33 (40.7%) preferred telemedicine after the COVID-19 pandemic was over. For the last question, the respondents were able to choose more than one answer. Respondent satisfaction with telemedicine services is shown in a bar graph in Fig. . The mean overall satisfaction score with telemedicine services was 3.8 ± 0.7, with a maximum score of 5. First, users were very satisfied (44.4%) with the convenience of telemedicine. Second, users were very satisfied (48.1%) with the ease of access to telemedicine. Third, users were very satisfied (42%) with the accuracy of the examination via telemedicine. Fourth, users were very satisfied (49.4%) with the therapy and/or advice given via telemedicine. Finally, users were very satisfied (43.2%) with the privacy provided by telemedicine consultation. We excluded three out of 314 recruited respondents because they did not meet the inclusion criteria. The remaining 311 respondents’ demographic and baseline characteristics are summarized in Table . The median age of respondents was 40 years (min 21-max 68), and the majority of the respondents were female (299, 96.1%). SLE was the most common ARD in this study (217, 70.1%), followed by RA (86, 27.3%). In addition, 54 (17.4%) respondents had more than one ARD diagnosis. Most of the respondents received nonbiologic disease-modifying antirheumatic drugs (DMARDs) (74.3%) and glucocorticoids (71.4%). Approximately 81 (26%) respondents reported avoiding hospital visits, and 76 (24.4%) respondents had ever stopped ARD medication without medical advice. First, regarding the respondents’ concerns about being infected with COVID-19, the majority were neutral (38,6%), followed by 27.7% who were always concerned and 24.4% who were often concerned about being infected with COVID-19. Regarding basic knowledge about COVID-19 infection, the majority of the respondents were always (30.2%) and often (30.2%) concerned that ARD conditions may increase the risk of COVID-19 infection, followed by those who were neutral (29.3%). Moreover, some respondents were always (37.6%) concerned that COVID-19 disease may worsen their ARD condition, and some were always (39.2%) concerned that COVID-19 symptoms among people with ARDs are more severe than those among healthy people. In addition, as seen in Fig. , the mean overall concern score was 3.9 ± 0.9 from the maximum score of 5. Second, regarding respondents’ social distancing behaviors during the COVID-19 pandemic, the majority of our respondents sometimes (38.9%) avoided leaving the house, followed by those who often (22.2%) avoided leaving the house. In contrast, 35.4% of our respondents sometimes (35.4%) avoided going to the grocery store, followed by those who never (19%) and rarely (18%) avoided going to the grocery store. Moreover, 38.6% sometimes avoided meeting in person with colleagues and families, followed by those who often (23.2%) avoided meeting in person. As shown in Fig. , the mean overall social distancing behavior score was 3.0 ± 1.0 from the maximum score of 5. In addition, as seen in Fig. , respondents’ concerns were positively correlated with their social distancing behaviors during the pandemic (p value 0.000, r 0.458). Third, regarding external factors that contributed to health care disruptions, most of our respondents lived on the Java and Bali islands, where 245 (78.8%) lived in level 4 PPKM regions (Pemberlakuan Pembatasan Kegiatan Masyarakat/Enforcement of Restrictions on Community Activities). PPKM level 4 was the highest level of this community activity restriction policy, and the implication of this was that 240 (77.2%) of our respondents had blocked access to the hospital during the implementation of PPKM. Moreover, the majority of our respondents used private transportation (64.3%) compared to public transportation (35.7%) and both modes of transportation were severely affected by PPKM. Furthermore, the bivariate analysis of all variables related is shown in Table . First, there was a significant association between sex and stopping medication without medical advice (p value 0.005). There was also an association between respondents’ concerns about infection, social distancing behaviors during the pandemic, and blocked access during PPKM (p value 0.014, 0.001, and 0.045). Finally, multivariate analysis of the respondents’ concerns, behaviors, and blocked access to the hospital is shown in Table . A strong association was found between blocked access to the hospital during the pandemic and the need to avoid hospital visits (OR 1.786; CI: 1.008–3.162) and between sex and stopping medication without medical advice (OR: 4.667; CI: 1.436–15.168). As shown in Table , only 81 of 311 (26.0%) respondents reported that they used telemedicine as an alternative consultation method during the COVID-19 pandemic. Several reasons were recorded among nonusers of telemedicine, and the most common reasons were ‘I don’t know about telemedicine’ (126 (54.8%)), followed by ‘I do not need telemedicine services’ (48 (20.9%)), and ‘No telemedicine service is available’ (27 (11.7%)). The respondents’ characteristics regarding telemedicine use are illustrated in Table . Among telemedicine users, the mean age was 40 ± 12.8 years, and the majority were female (75, 92.6%). SLE was the most common ARD in this study (57, 70.4%), followed by RA (12, 14.8%), with 14 (17.3%) respondents having more than one ARD diagnosis. In addition, most of the respondents received nonbiologic DMARDs (80.2%), followed by glucocorticoids (77.8%). Furthermore, the most preferred modality among the respondents was text/chat consultations (n = 56, 69.1%), followed by video consultations (n = 44, 54.3%), and 27 (33.3%) preferred to fill out an online form for a direct appointment. Finally, approximately 44 (54.3%) respondents preferred telemedicine during the COVID-19 pandemic, and 33 (40.7%) preferred telemedicine after the COVID-19 pandemic was over. For the last question, the respondents were able to choose more than one answer. Respondent satisfaction with telemedicine services is shown in a bar graph in Fig. . The mean overall satisfaction score with telemedicine services was 3.8 ± 0.7, with a maximum score of 5. First, users were very satisfied (44.4%) with the convenience of telemedicine. Second, users were very satisfied (48.1%) with the ease of access to telemedicine. Third, users were very satisfied (42%) with the accuracy of the examination via telemedicine. Fourth, users were very satisfied (49.4%) with the therapy and/or advice given via telemedicine. Finally, users were very satisfied (43.2%) with the privacy provided by telemedicine consultation. Overall, as a predisposing factor, the COVID-19 pandemic affected ARD patients’ perspectives about their concerns about infection and promoted their preventive behaviors. This is aligned with the health belief model that shows that factors that contribute to patient’s perceptions can guide their behavior. Moreover, the COVID-19 situation and lockdown policy was an enabling factor and reinforcing factor for patients in health-seeking behavior. These behavioral changes obviously needed an intervention, and we showed that patients’ reception of telemedicine as an alternative health care practice during the pandemic was positively received. In 2023, the Indonesian government ended COVID-19 restrictions and declared COVID-19 an endemic disease, which created another shift toward a new normal situation. However, some patients still faced treatment interruptions even before and after the pandemic due to Indonesia’s sociodemographic and geographical factors. Finally, from this pandemic, we learned that telemedicine could be a breakthrough for patients to access health care, especially for ARD patients who face barriers to access. Fear of COVID-19 The American College of Rheumatology (ACR) and APLAR declared that patients with ARDs are at risk and more susceptible to COVID-19 infection than the general population. However, it is still debated whether ARD medications may become a possible risk factor for ARD patients having poor outcomes, including hospitalization, severe infection, mortality, intensive care unit admission, and ventilator use in COVID-19. On the other hand, ARD medication should not be immediately stopped and should be maintained at a specific dose to prevent disease flares, and the prescription should be determined on an individualized basis according to the disease . As seen in this study, this mindset may affect patient concerns about being infected and drive avoidance behavior toward COVID-19. A similar phenomenon has been reported in previous studies: in the study by Fragoulis et al., patients who changed medication due to concerns about the immunosuppression effect increased their susceptibility to infection and worse outcomes; in the study by Michaud et al., patients added and removed their medication due to worries about COVID-19 infection; in the study by Pineda-Sic et al., patients in Latin America changed medications due to fear of contracting COVID-19; and in the study by Khabbazi et al., nonadherent patients in East Azerbaijan feared the immunosuppressive effect of medications . Thus, in the future, patients need an understanding of COVID-19 infection, the importance of taking medication, and routine follow-up checks. The continuity of communication, information, and education between doctor and patient are key in managing therapy adherence to prevent a worse outcome. Additionally, our study showed that sex was associated with stopping medication without medical advice. This may be because we had more female respondents than male respondents, and this might have been a selection bias during our analysis. In contrast, a previous study showed no significant difference in adherence to medication between sexes during the pandemic. However, the study examined nonspecific diseases, while our study examined rheumatic autoimmune patients who were predominantly females . Lockdown and social distancing policies during the COVID-19 pandemic During the second wave of COVID-19 (June to August 2021), the Indonesian government implemented PPKM. PPKM tightened social distancing, including community activities and travel regulations. According to Instructions of the Ministry of Home Affairs number 27 (for Java and Bali) and 28 (for outside Java and Bali region), each region was classified into levels based on epidemiological indicators, surveillance indicators, and health care services. The majority of our respondents were living in level 3 and 4 zones, which were classified as high-risk areas at the time. However, there might have been recall bias as they filled out the questionnaire to evaluate their experience after the second wave was over. In addition, PPKM levels 3 and 4 implemented (1) online teaching for all levels of education; (2) 100% work-from-home (WFH) activities for all nonessential activities; (3) 50% work from an office (WFO) for essential activities such as the economic sector and data centers; and (4) 100% WFO with strict health protocols for critical sectors such as health care facilities, disaster response teams, and governmental vital objects . The difference was in the maximum capacity of transportation, which was up to 50% at level 4 and up to 70% at level 3. However, in both level 3 and 4 zones, long-distance travelers required a vaccination card, polymerase chain reaction (PCR) test or rapid diagnostic test (RDT) for COVID-19 . Although there was no regulation inhibiting citizens from traveling for medical or emergency purposes, travel restrictions still became a barrier to access. A possible reason was that a COVID-19 certificate was one of the requirements needed for traveling across the region; however, it is possible that some ARD patients were not eligible to be vaccinated due to their disease activity and treatment, as evidenced by a letter from a rheumatologist. Moreover, PCR or RDT tests became an economical burden when patients needed to frequently travel for hospital visits. In particular, this regulation was implemented for long-distance travelers; however, most rheumatology centers are located in large cities, and some patients may have been referred from a distant hospital. In addition, patients were anxious about COVID-19 exposure while traveling. This becomes more evident in several studies conducted in Indonesia, there was a considerable decrease in the number of outpatient clinics other than ARD clinics during the lockdown, and a rebound was seen afterward due to these factors . This problem was also evident in our study, in which blocked access during PPKM was associated with the need to avoid hospital visits. This showed that travel restrictions and transportation regulations were factors for patients to consider the use of telemedicine, especially during the pandemic. On the other hand, telemedicine had the potential role of preventing COVID-19 spread during travel; it also connected patients who were not able to travel to the health care facility, and most importantly, the patient had access to receive information from the health care worker about their medical condition and COVID-19 . Telemedicine in rheumatology Practice In our study, respondents were very satisfied with telemedicine (3.8 ± 0.7 over 5), and more than half preferred telemedicine visits during the pandemic, but less than half preferred telemedicine visits after the pandemic. Similar results were evidenced in surveys by Tornero-Molina et al. in Spain, where patients showed higher overall satisfaction in tele-rheumatology (RTC) and scored 8.62 out of 10, and approximately 391/469 (84%) wanted to repeat RTC; in the study by Jones et al., in most UK rheumatology outpatient clinics, 239/297 (84%) patients were satisfied with their health assessment, and 60% wanted to have routine follow-up telephone consultations; in the study by Mortazavi et al. in the US, most of their rheumatology clinic’s patients (74%) were satisfied with their virtual visit; and in the study by Cliffe and Stevenson in the UK, the majority of their musculoskeletal patients (194/241 (80.5%)) were delighted with a virtual consultation . Alternatively, early in this pandemic, telemedicine was adopted into the Indonesian health care system and regulated by the Indonesian Ministry of Health and Indonesian Medical Council to tackle geographical and distance barriers between health care providers and patients . Indonesian national health insurance covered all medical bills with a well-documented medical record. Many commercial telemedicine providers could also be accessed on smart devices linked to private insurance. Nevertheless, most of our respondents, (126 (54.8%)) who had never used this service, did not know about telemedicine. In early 2022, APLAR released a recommendation on telemedicine practice in rheumatology. They recommended telemedicine for situations in which rheumatologists and/or patients have a communication gap or when there is a disruption in regular health services to prevent unsupervised medication. Telemedicine should be based on clinical effectiveness, safety issues, the patient’s perspective, economic, organizational, sociocultural, ethical, and legal aspects, and equitable health access according to local regulations. Patient data privacy, integrity, and security should be protected. The decision should be shared by rheumatologists and patients and should be made after the preconsultation triage system to assess whether the patient’s condition is suitable for telemedicine follow-up. However, telemedicine is not recommended for an initial appointment or a patient with an unconfirmed diagnosis. Telemedicine is also recommended to train nurses, physicians, and rheumatologists to provide better clinical practice in telemedicine . Postpandemic situations According to the Indonesian government, a rural area is defined by topography, access to urban facilities, agriculture, landscapes, or population density. Some rural areas are locations outside Java-Bali, which are the less developed areas, whereas others are classified according to population density and medical workforce supply . However, in our study, there was no significant difference in health care disruption or treatment interruption between Java-Bali and outside the Java-Bali region. In a study by Putri et al., there was an inequality in specialist distribution across Indonesia based on geography. Doctors working in less developed areas may face a lack of health infrastructure and accommodation, even though the government has created policies and programs to provide financial and non-financial benefits to tackle this problem . These are probably the reasons that advanced health care practices such as rheumatology centers are concentrated in large cities. On the other hand, not all patients can accommodate travel to access health care. Indeed, ARDs are chronic diseases that need to be monitored continuously and telemedicine may helps to overcome the shortage of rheumatologists or internists to provide care outside rheumatology centers. This helps patients who are unable to travel and prevents the discontinuation of medical follow-up . Despite this limitation, there should be a new normal situation in rheumatology practice in Indonesia to meet the patient’s needs as we learned from the pandemic. Limitations Despite the results, this study has several limitations. First, this was a cross-sectional study that could not explain causal or effective relationships. Second, the self-developed questionnaire might have led to information bias among the respondents. Third, respondents’ medical histories were not traced from their medical records, which might have led to selection bias. Because we do not have a national database for rheumatology patients. Last, ARD patients who completed the questionnaire might have been more concerned about COVID-19 and their diseases compared to those who did not. For example, our respondents are predominantly females, joining the autoimmune community, and those who filled out the questionnaire given by their internist or rheumatologist must have received more information about COVID-19 and the ARD than those who did not. In addition, respondents completed the questionnaire during the nonpeak of COVID-19 cases, which may have led to recall bias, even though the questionnaire was designed to evaluate the overall patient experience throughout the pandemic. The American College of Rheumatology (ACR) and APLAR declared that patients with ARDs are at risk and more susceptible to COVID-19 infection than the general population. However, it is still debated whether ARD medications may become a possible risk factor for ARD patients having poor outcomes, including hospitalization, severe infection, mortality, intensive care unit admission, and ventilator use in COVID-19. On the other hand, ARD medication should not be immediately stopped and should be maintained at a specific dose to prevent disease flares, and the prescription should be determined on an individualized basis according to the disease . As seen in this study, this mindset may affect patient concerns about being infected and drive avoidance behavior toward COVID-19. A similar phenomenon has been reported in previous studies: in the study by Fragoulis et al., patients who changed medication due to concerns about the immunosuppression effect increased their susceptibility to infection and worse outcomes; in the study by Michaud et al., patients added and removed their medication due to worries about COVID-19 infection; in the study by Pineda-Sic et al., patients in Latin America changed medications due to fear of contracting COVID-19; and in the study by Khabbazi et al., nonadherent patients in East Azerbaijan feared the immunosuppressive effect of medications . Thus, in the future, patients need an understanding of COVID-19 infection, the importance of taking medication, and routine follow-up checks. The continuity of communication, information, and education between doctor and patient are key in managing therapy adherence to prevent a worse outcome. Additionally, our study showed that sex was associated with stopping medication without medical advice. This may be because we had more female respondents than male respondents, and this might have been a selection bias during our analysis. In contrast, a previous study showed no significant difference in adherence to medication between sexes during the pandemic. However, the study examined nonspecific diseases, while our study examined rheumatic autoimmune patients who were predominantly females . During the second wave of COVID-19 (June to August 2021), the Indonesian government implemented PPKM. PPKM tightened social distancing, including community activities and travel regulations. According to Instructions of the Ministry of Home Affairs number 27 (for Java and Bali) and 28 (for outside Java and Bali region), each region was classified into levels based on epidemiological indicators, surveillance indicators, and health care services. The majority of our respondents were living in level 3 and 4 zones, which were classified as high-risk areas at the time. However, there might have been recall bias as they filled out the questionnaire to evaluate their experience after the second wave was over. In addition, PPKM levels 3 and 4 implemented (1) online teaching for all levels of education; (2) 100% work-from-home (WFH) activities for all nonessential activities; (3) 50% work from an office (WFO) for essential activities such as the economic sector and data centers; and (4) 100% WFO with strict health protocols for critical sectors such as health care facilities, disaster response teams, and governmental vital objects . The difference was in the maximum capacity of transportation, which was up to 50% at level 4 and up to 70% at level 3. However, in both level 3 and 4 zones, long-distance travelers required a vaccination card, polymerase chain reaction (PCR) test or rapid diagnostic test (RDT) for COVID-19 . Although there was no regulation inhibiting citizens from traveling for medical or emergency purposes, travel restrictions still became a barrier to access. A possible reason was that a COVID-19 certificate was one of the requirements needed for traveling across the region; however, it is possible that some ARD patients were not eligible to be vaccinated due to their disease activity and treatment, as evidenced by a letter from a rheumatologist. Moreover, PCR or RDT tests became an economical burden when patients needed to frequently travel for hospital visits. In particular, this regulation was implemented for long-distance travelers; however, most rheumatology centers are located in large cities, and some patients may have been referred from a distant hospital. In addition, patients were anxious about COVID-19 exposure while traveling. This becomes more evident in several studies conducted in Indonesia, there was a considerable decrease in the number of outpatient clinics other than ARD clinics during the lockdown, and a rebound was seen afterward due to these factors . This problem was also evident in our study, in which blocked access during PPKM was associated with the need to avoid hospital visits. This showed that travel restrictions and transportation regulations were factors for patients to consider the use of telemedicine, especially during the pandemic. On the other hand, telemedicine had the potential role of preventing COVID-19 spread during travel; it also connected patients who were not able to travel to the health care facility, and most importantly, the patient had access to receive information from the health care worker about their medical condition and COVID-19 . In our study, respondents were very satisfied with telemedicine (3.8 ± 0.7 over 5), and more than half preferred telemedicine visits during the pandemic, but less than half preferred telemedicine visits after the pandemic. Similar results were evidenced in surveys by Tornero-Molina et al. in Spain, where patients showed higher overall satisfaction in tele-rheumatology (RTC) and scored 8.62 out of 10, and approximately 391/469 (84%) wanted to repeat RTC; in the study by Jones et al., in most UK rheumatology outpatient clinics, 239/297 (84%) patients were satisfied with their health assessment, and 60% wanted to have routine follow-up telephone consultations; in the study by Mortazavi et al. in the US, most of their rheumatology clinic’s patients (74%) were satisfied with their virtual visit; and in the study by Cliffe and Stevenson in the UK, the majority of their musculoskeletal patients (194/241 (80.5%)) were delighted with a virtual consultation . Alternatively, early in this pandemic, telemedicine was adopted into the Indonesian health care system and regulated by the Indonesian Ministry of Health and Indonesian Medical Council to tackle geographical and distance barriers between health care providers and patients . Indonesian national health insurance covered all medical bills with a well-documented medical record. Many commercial telemedicine providers could also be accessed on smart devices linked to private insurance. Nevertheless, most of our respondents, (126 (54.8%)) who had never used this service, did not know about telemedicine. In early 2022, APLAR released a recommendation on telemedicine practice in rheumatology. They recommended telemedicine for situations in which rheumatologists and/or patients have a communication gap or when there is a disruption in regular health services to prevent unsupervised medication. Telemedicine should be based on clinical effectiveness, safety issues, the patient’s perspective, economic, organizational, sociocultural, ethical, and legal aspects, and equitable health access according to local regulations. Patient data privacy, integrity, and security should be protected. The decision should be shared by rheumatologists and patients and should be made after the preconsultation triage system to assess whether the patient’s condition is suitable for telemedicine follow-up. However, telemedicine is not recommended for an initial appointment or a patient with an unconfirmed diagnosis. Telemedicine is also recommended to train nurses, physicians, and rheumatologists to provide better clinical practice in telemedicine . According to the Indonesian government, a rural area is defined by topography, access to urban facilities, agriculture, landscapes, or population density. Some rural areas are locations outside Java-Bali, which are the less developed areas, whereas others are classified according to population density and medical workforce supply . However, in our study, there was no significant difference in health care disruption or treatment interruption between Java-Bali and outside the Java-Bali region. In a study by Putri et al., there was an inequality in specialist distribution across Indonesia based on geography. Doctors working in less developed areas may face a lack of health infrastructure and accommodation, even though the government has created policies and programs to provide financial and non-financial benefits to tackle this problem . These are probably the reasons that advanced health care practices such as rheumatology centers are concentrated in large cities. On the other hand, not all patients can accommodate travel to access health care. Indeed, ARDs are chronic diseases that need to be monitored continuously and telemedicine may helps to overcome the shortage of rheumatologists or internists to provide care outside rheumatology centers. This helps patients who are unable to travel and prevents the discontinuation of medical follow-up . Despite this limitation, there should be a new normal situation in rheumatology practice in Indonesia to meet the patient’s needs as we learned from the pandemic. Despite the results, this study has several limitations. First, this was a cross-sectional study that could not explain causal or effective relationships. Second, the self-developed questionnaire might have led to information bias among the respondents. Third, respondents’ medical histories were not traced from their medical records, which might have led to selection bias. Because we do not have a national database for rheumatology patients. Last, ARD patients who completed the questionnaire might have been more concerned about COVID-19 and their diseases compared to those who did not. For example, our respondents are predominantly females, joining the autoimmune community, and those who filled out the questionnaire given by their internist or rheumatologist must have received more information about COVID-19 and the ARD than those who did not. In addition, respondents completed the questionnaire during the nonpeak of COVID-19 cases, which may have led to recall bias, even though the questionnaire was designed to evaluate the overall patient experience throughout the pandemic. In short, internal and external factors affected health care disruptions and treatment interruptions during the COVID-19 pandemic, which affected treatment adherence among ARD patients. Moreover, during the pandemic, the use of telemedicine was significantly associated with respondents who avoided hospital visits, and users showed high satisfaction with telemedicine services. Thus, in the future, telemedicine may become an alternative in rheumatology practice in Indonesia to increase visitation and treatment adherence, which is interesting because Indonesia’s sociodemography and geographical situation are the biggest challenges in health care practice. Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2
Task force greener gastroenterology: Towards eco‐friendly practices in healthcare	03e91ca0-04b8-40b8-a65e-32ed76774498	10165319	Internal Medicine[mh]
Specific associations between plasma biomarkers and postmortem amyloid plaque and tau tangle loads	ab80b846-0b90-46ae-9a45-481349309a6a	10165361	Pathology[mh]	Alzheimer's disease is characterized by the deposition of amyloid plaques and neurofibrillary tau tangles in the brain. In recent years, several plasma biomarkers have been developed to assess these two pathologies, which have been validated against other fluid and neuroimaging biomarkers. However, their specific associations with each of these two pathologies are still not fully understood. In a set of participants with available blood samples and a neuropathological exam, we observed that some plasma biomarkers are specifically associated with only amyloid (Aβ42/40 and p‐tau231), some to only tau (GFAP) and, some to both pathologies (p‐tau217 and p‐tau181). Further, we showed that the combination of p‐tau217 and the Aβ42/40 ratio was optimal for assessing amyloid, while p‐tau217 alone was sufficient to assess tau pathology. Comparing head‐to‐head the associations between high‐performing assays of different plasma biomarkers and neuropathological correlates allowed us to determine which is the optimal single or combination of plasma biomarkers for assessing actual pathology. This has a direct impact on the design of clinical trials, as we showed that p‐tau217 may not only be useful as a prescreening tool for clinical trials but also may be a good surrogate endpoint, especially on those trials targeting tau pathology. Furthermore, combining it with the Aβ42/40 ratio would significantly improve the assessment of continuous amyloid pathology. The recent development of plasma biomarkers for Alzheimer's disease (AD) has revolutionized the field (Zetterberg & Bendlin, ; Hansson, ; Ossenkoppele et al , ), as these markers have the benefit of being significantly cheaper and less invasive than established markers (i.e., cerebrospinal fluid [CSF] and positron emission tomography [PET]), while showing the excellent diagnostic performance (Karikari et al , ; Palmqvist et al , ; Janelidze et al , ). Several plasma biomarkers are currently available, among which, the most studied include the amyloid‐β42/40 (Aβ42/40) ratio, glial fibrillary acidic protein (GFAP), neurofilament light (NfL), and, particularly, phosphorylated tau (p‐tau) measures. Previous studies have indicated the excellent diagnostic performance of some of these plasma biomarkers for distinguishing AD from non‐AD neurodegenerative disorders (Karikari et al , ; Palmqvist et al , ; Thijssen et al , ; Janelidze et al , ), with noninferior performance compared with CSF and PET markers (Palmqvist et al , ; Janelidze et al , , ; Ashton et al , ; Benedet et al , ; Mielke et al , ), as well as an important utility for predicting disease progression (Ashton et al , ; Cullen et al , ; Mielke et al , ; Janelidze et al , ). Nonetheless, there are still important topics to be addressed to optimize their usage in clinical practice, including improved interpretation of obtained plasma biomarker results and a fair head‐to‐head comparison, especially against gold standard neuropathological measures (Hansson et al , ). One of the most important knowledge voids of plasma biomarkers is the degree to which they specifically correlate with key neuropathological changes. Although previous studies have already investigated the association of some of these biomarkers with measures of neuropathology, whether these markers are primarily related to amyloid, tau, or to both pathologies is still under debate. For instance, p‐tau181 has shown strong associations with neuropathological measures of amyloid‐β and tau pathologies (Lantero Rodriguez et al , ; Thijssen et al , ; Grothe et al , ; Smirnov et al , ), but this has been shown in independent analyses for amyloid and tau or with scales combining these two pathologies, which did not allow for the interpretation of the specific ‐or independent‐ associations with these two pathological measures. Similarly, plasma p‐tau231 has also shown associations with neuropathologically defined plaque and tangle load, without exploring specific associations with these neuropathologic measures (Ashton et al , ; Smirnov et al , ). Only one study with plasma p‐tau217 has suggested that this biomarker may be independently associated with both plaques and tangles (Mattsson‐Carlgren et al , ), but no other biomarkers were investigated. On the contrary, the accuracy of the plasma Aβ42/40 ratio, GFAP, or NfL levels to predict AD pathology seems to be lower than that of p‐tau markers, although only few studies have investigated their association with neuropathologic measures of AD pathology (Thijssen et al , ; Smirnov et al , ; Winder et al , ). Another challenge when trying to optimize the use of plasma biomarkers in clinical practice is the lack of comparison among biomarkers in the same population. Differences in clinical performance for the same biomarkers can be observed depending on the characteristics of the study (e.g., diagnostic groups, patient characteristics, outcomes, and/or presence of co‐pathologies). Thus, head‐to‐head studies are crucial to allow a fair comparison and avoid bias due to population selection. Nonetheless, these studies are scarce, especially those including neuropathological measures (Smirnov et al , ; Winder et al , ). The use of neuropathological data would also allow investigating whether any of these biomarkers might be useful for detecting other common co‐pathologies observed in AD patients, such as Lewy bodies or TAR DNA‐binding protein 43 (TDP‐43; Hansson, ; Smirnov et al , ). When comparing multiple plasma biomarkers, it is equally important to consider the discriminative power of the assays. While there are many plasma biomarkers currently available, there are also many platforms by which to measure them, which can highly affect their performance. As it has been shown recently with the plasma Aβ42/40 ratio, different assays and/or platforms could lead to significantly different performances in detecting AD‐related pathology (Janelidze et al , ). Similarly, comparisons between multiple species and assays of p‐tau measures showed only a modest correlation, suggesting also significantly different diagnostic performance (Mielke et al , ; Janelidze et al , ; Ashton et al , ). Considering the differences in the clinical performance of different assays for the same biomarkers, the use of high‐performing assays is of utmost importance when comparing different biomarkers to avoid reporting differences that are related to the method rather than the biomarkers themselves. Therefore, the main objective of this study was to identify specific relationships between multiple plasma biomarkers and core AD‐related pathologies using high‐performing assays. To this end, we investigated associations between multiple plasma biomarkers and autopsy‐assessed measures of core AD pathologies (plaque and tangle loads) in the same participants. We focused on investigating whether these biomarkers primarily reflect amyloid, tau, or both pathologies. Further, we identified the best combination of biomarkers to predict each of these pathological measures, as well as the presence or absence of AD as a binary measure based on pathology (Montine et al , ). We also investigated associations between plasma biomarkers and the presence of co‐pathologies commonly observed in AD patients including cerebral amyloid angiopathy (CAA), Lewy body disease (LBD), TDP‐43, cerebral white matter rarefaction (CWMR), and argyrophilic grain disease (AGD). Finally, we examined whether longitudinal changes of the two plasma biomarkers longitudinally available (i.e., p‐tau217 and p‐tau181) were associated with presence of AD pathology. Our sample comprised a total of 105 participants from the Arizona Study of Aging and Neurodegenerative Disorders (AZSAND) including all participants with complete antemortem plasma samples and a postmortem neuropathological exam (Table ). These participants were categorized as having significant AD pathologic change ( n = 59) or not ( n = 46) based on the Alzheimer's disease neuropathologic change (ADNC) scale, in which both amyloid and tau pathologies are accounted for (Montine et al , ). Participants with significant AD pathology were those with intermediate or high scores in the ADNC scale, whereas those with none or low scores were classified as having nonsignificant AD pathology. No differences in age at death nor sex were observed between groups. APOE‐ε4 prevalence (49.2% vs. 10.9%, P < 0.001) and core AD pathology (plaques: 12.70 vs. 1.03, P < 0.001; tangles: 9.79 vs. 5.54, P < 0.001) measures were significantly higher in participants with intermediate/high ADNC. Among all the co‐pathologies under investigation (i.e., CAA, LBD, CWMR, and AGD), only the presence of CAA was significantly higher in participants with intermediate/high ADNC (86.4% vs. 34.8%, P < 0.001). Associations between plasma biomarkers and core AD pathologies Our first objective was to assess the associations between each plasma biomarker and the two neuropathological measures of AD pathology (i.e., total amount of plaques and tangles), independently. Plaque and tangle loads were measured on a semi‐quantitative scale that ranged from 0 to 3 in five different brain regions (Mirra et al , ), and we combined these regional measures into a total score (range: 0–15) for each pathology. We used partial Spearman's ρ to assess the association between each plasma biomarker and each total amount of pathology while adjusting for age, sex, and time between blood draw and death. We found that all plasma biomarkers except NfL (ρ = 0.10, P = 0.895) were significantly associated with the total amount of plaques (0.41 ≤ \|ρ\| ≤ 0.73, P < 0.001, Fig and Table ). Plasma p‐tau217 showed the highest correlation coefficient with plaques, which was significantly higher than all others (0.10 ≤ ρ diff ≤ 0.63, P ≤ 0.016) except plasma Aβ42/40 ratio (ρ diff = 0.19, P = 0.055). All plasma biomarkers except NfL (ρ = 0.19, P = 0.257) were also associated with the total amount of tangles in the independent models (0.26 ≤ \|ρ\| ≤ 0.66, P ≤ 0.016; Fig and Table ). Again, p‐tau217 had the highest correlation coefficient with the total amount of tangles, which was significantly higher than all other plasma correlation coefficients (0.11 ≤ ρ diff ≤ 0.47, P ≤ 0.006) except for that of plasma GFAP (ρ diff = 0.10, P = 0.206). As a sensitivity analysis, we also investigated these correlations separately for participants without (i.e., ADNC none or low) and with significant AD pathology (i.e., ADNC intermediate or high, Appendix Table ). In the group without significant AD pathology, only the Aβ42/40 ratio showed a significant correlation with amyloid (ρ = ‐0.33, P < 0.001). No plasma biomarkers showed a significant correlation with tau tangle load in this group. In the group of significant AD pathology, both p‐tau217 (ρ = 0.41, P = 0.049) and the Aβ42/40 ratio (ρ = ‐0.30, P < 0.001) presented a significant correlation with amyloid plaque load. Further, p‐tau217 (ρ = 0.56, P = 0.001), p‐tau181 (ρ = 0.49, P = 0.008), and GFAP (ρ = 0.47, P = 0.011) had a significant correlation with tau tangle load. Since plaque and tangle load were highly correlated (ρ[95%CI] = 0.63[0.48, 0.73], P < 0.001, Appendix Fig ), we performed an analysis to identify the specific (or independent) associations between each plasma biomarker and the two pathologies. For this, we used partial Spearman's ρ again, adjusting further for the other pathology as well (i.e., when looking at plaques adjusting for tangles and vice‐versa). In these models, p‐tau217 (ρ = 0.40, P = 0.003), p‐tau181 (ρ = 0.36, P = 0.009), and the Aβ42/40 ratio (ρ = ‐0.53, P < 0.001) were significantly associated with plaques (Fig and Table ). Plasma GFAP showed no significant association with plaque load when adjusted for tangle load (β = 0.09, P = 1.00) and p‐tau231 showed an association only at a trend level (ρ = 0.28, P = 0.084). In this analysis, the Aβ42/40 ratio correlation coefficient with plaques was the highest, being significantly higher than that of p‐tau231 (ρ diff = 0.25, P = 0.028) but not than that of p‐tau217 (ρ diff = 0.13, P = 0.246) nor p‐tau181 (ρ diff = 0.17, P = 0.146). In addition, we observed that only p‐tau217 (ρ = 0.52, P < 0.001), p‐tau181 (ρ = 0.36, P = 0.010), and GFAP (ρ = 0.39, P = 0.004) were associated with tangles. The correlation coefficient of p‐tau217 with tangles was significantly higher than that of p‐tau181 (ρ diff = 0.17, P = 0.004) when adjusting for the total amount of plaques but not than that of plasma GFAP (ρ diff = 0.13, P = 0.207). Comparing the correlation coefficients to each of these two pathologies for each biomarker, we observed three groups of biomarkers (Fig and Appendix Table ). We observed that plasma p‐tau231 (71.8%) and the Aβ42/40 ratio (83.1%) had a major proportion of variance explained by plaques than tangles. On the contrary, the opposite happened with plasma GFAP and NfL, with tangles explaining the major part of these biomarkers' variance (GFAP: 82.1%, NfL: 82.9%). Finally, in p‐tau217 (plaques: 43.4%, tangles: 56.6%) and p‐tau181 (plaques: 50.5%, tangles: 49.5%) both pathologies contributed similarly to explaining their variance. Finally, we investigated which combination of biomarkers better‐predicted plaques and tangles, independently. We found that the parsimonious model that better‐predicted load of amyloid plaques included both p‐tau217 and the Aβ42/40 ratio ( R 2 = 0.57, Table ), which was significantly better than the one only including p‐tau217 based on AICc (ΔAICc = 15.5). On the contrary, p‐tau217 alone was selected as the parsimonious model to predict the load of neurofibrillary tangles ( R 2 = 0.50, Table ). Prediction of neuropathological scales' classification Next, we investigated differences in plasma levels by ADNC groups (as a four‐level variable, i.e., none, low, intermediate, or high) using a Kruskal‐Wallis test and Wilcoxon test for post hoc comparisons. All three p‐tau plasma measures showed significant differences between intermediate and high levels of ADNC. Plasma p‐tau217 and the Aβ42/40 ratio levels were significantly different between intermediate and low ADNC. However, we only found significant differences between none and low ADNC in plasma p‐tau217 levels (Fig ), although this became only a trend when removing the highest plasma value of the low group. Notably, p‐tau217 also showed the highest fold‐change among all ADNC consecutive levels (Appendix Table ). Then, we also examined differences in pathological scales specific for amyloid (Consortium to establish a registry for Alzheimer's disease [CERAD]) (Mirra et al , ) and tau (Braak staging) (Braak & Braak, ) pathologies. Similarly, all biomarkers except NfL showed significant differences between sparse and moderate/frequent groups on CERAD's classification. Only plasma p‐tau217 showed differences between zero and sparse groups (Appendix Fig ). Regarding Braak staging, all biomarkers except NfL were significantly different when comparing 0–IV with V–VI groups (Appendix Fig ). We next investigated the accuracy of each plasma biomarker to predict the presence of AD pathology as measured with the dichotomized ADNC (none/low vs. intermediate/high) classification. For this, we used receiver‐operating characteristic (ROC) curves and calculated the area under the curve (AUC) for each biomarker independently adjusting for age, sex, and time between blood sampling and death. All biomarkers except NfL (AUC[95%CI] = 0.61 [0.50, 0.71], P = 0.698) were predictive of the presence of ADNC when assessed individually (0.88 ≥ AUC ≥ 0.72, Appendix Table and Fig ). Plasma p‐tau217 had the highest AUC (AUC[95%CI] = 0.88 [0.81–0.95]) of all individual biomarkers, which was significantly higher than all others except for the Aβ42/40 ratio (AUC[95%CI] = 0.80 [0.72–0.89], P = 0.099). We also repeated this analysis with CERAD (low/sparse vs. moderate/frequent) and Braak staging (0–IV vs. V–VI) classification. For CERAD, p‐tau217 was also the best individual biomarker as per classification accuracy (AUC[95%CI] = 0.89 [0.83–0.96], P < 0.001), comparable only to that of the Aβ42/40 ratio (AUC[95%CI] = 0.82 [0.74–0.90], P < 0.001, Fig and Appendix Table ). For Braak staging, p‐tau217 again showed the highest accuracy (AUC[95%CI] = 0.93 [0.87–0.98], P < 0.001) comparable only to that of GFAP (AUC[95%CI] = 0.86[0.79–0.94], P < 0.001, Fig and Appendix Table ). Next, we investigated whether combining different biomarkers would improve the models with only individual biomarkers. We found that plasma p‐tau217 and the Aβ42/40 ratio was the optimal combination to predict the presence of ADNC (AUC[95%CI] = 0.90 [0.84, 0.96], Table and Fig ), but the AUC was not significantly higher than that of p‐tau217 alone when using the DeLong's test ( P = 0.124). With CERAD we observed a similar behavior, with plasma p‐tau217 and the Aβ42/40 ratio being the best combination (AUC[95%CI] = 0.91 [0.86–0.97], Appendix Table ), although the AUC was not significantly better than that of p‐tau217 alone ( P = 0.173). In the case of Braak staging, there were no combinations of biomarkers that improved the accuracy compared with p‐tau217‐only models. Prediction of presence of co‐pathologies In this analysis, we investigated whether any of the available plasma biomarkers improved the basic models' accuracy (only covariates) on predicting the presence of co‐pathologies, using a similar approach as before but further adjusting for the presence of intermediate/high ADNC. We observed that only plasma NfL significantly improved the prediction of the presence of CWMR (Dugger et al , ) (AUC[95%CI] = 0.76 [0.66, 0.85]) compared with the basic model (AUC[95%CI] = 0.65 [0.54, 0.76], P = 0.028, Appendix Table and Appendix Fig A). In particular, participants with CWMR had significantly higher plasma NfL levels than those without (β = 0.88, P = 0.002, Appendix Fig B). No other biomarkers improved the prediction of the presence of this nor any other co‐pathology (Appendix Tables and Appendix Fig ). Raw distribution of plasma levels by the presence of each co‐pathology can be observed in Appendix Figs . As an additional analysis, we further checked whether there were differences between plasma levels in participants with only co‐pathologies (e.g., CAA only) and participants with AD pathology and co‐pathologies (e.g., CAA and ADNC) as it may have important clinical implications. We found that p‐tau217 was significantly higher in those participants having both AD pathology (as ADNC intermediate or high) and CAA compared to those with only AD pathology ( P = 0.037, Appendix Fig ). However, the group of AD‐only pathology was small ( n = 8). At the statistical trend level, we also observed differences in plasma Aβ42/40 levels in AD‐only versus AD and LBD groups ( P = 0.058, Appendix Fig ); in both p‐tau217 and Aβ42/40 levels in AD‐only versus AD and AGD groups ( P = 0.052 and P = 0.069, respectively; Appendix Fig ) and in NfL levels in AD‐only versus AD and CWMR ( P = 0.090, Appendix Fig ). No differences were observed in the case of TDP‐43 (Appendix Fig ). Finally, we also considered primary tauopathies (CBD, PSP, and AGD) as a unique group and compared the plasma levels of those participants to those with only AD pathology and those with AD pathology and other tauopathies. Plasma p‐tau217 ( P < 0.001), p‐tau181 ( P = 0.001), Aβ42/40 ratio ( P < 0.001), and GFAP ( P = 0.024) levels were significantly different when comparing participants with only AD pathology and participants with only primary tauopathies (Appendix Fig ). Only Aβ42/40 ratio levels were different between the AD group with CBD, PSP, or AGD pathology ( P = 0.038). Use of the p‐tau217/Aβ42 ratio Given that the CSF p‐tau/Aβ42 ratio is commonly used both in research and in clinical practice, we wanted to investigate whether a plasma p‐tau/Aβ42 ratio would also be useful for predicting AD pathology. For this, we selected p‐tau217 as it showed the highest associations in the previous analyses. We compared the accuracy of predicting plaques and tangles, independently, comparing parsimonious models including the plasma p‐tau217/Aβ42 ratio as a possible independent variable to those obtained in the previous sections including p‐tau217. We observed that the p‐tau217/Aβ42 ratio was preferentially selected over p‐tau217 in the models predicting plaques and tangles. Based on the AICc, we observed that models including the p‐tau217/Aβ42 ratio were slightly, but significantly, better than those previously presented (plaques: R 2 p‐tau217/Aβ42 ratio = 0.60, AICc = 210.1 vs. R 2 p‐tau217 = 0.57, AICc = 218.4; tangles: R 2 p‐tau217/Aβ42 ratio = 0.52, AICc = 228.7 vs. R 2 p‐tau217 = 0.50, AICc = 233.5; Appendix Table ). We also observed that p‐tau217/Aβ42 ratio levels were significantly different between the none versus low ADNC groups ( P = 0.029), even when removing the outlier in the low group. Longitudinal associations between p‐tau217 and p‐tau181 with AD pathology Finally, we investigated whether longitudinal changes in plasma p‐tau217 and p‐tau181 were associated with the presence of AD pathology at death (median[range] timepoints: 2 [2–5], mean (SD) time difference from first timepoint to death: 1378 (1357) days). Details of these participants can be found in Appendix Table . First, we observed that longitudinal increments of p‐tau217 but not p‐tau181 were associated with plaque burden (p‐tau217: β = 0.09, P = 0.005; p‐tau181: β = 0.05, P = 0.350, Table ). In independent models, we observed that p‐tau217 increments, but not those in p‐tau181, were also associated with tangle load (p‐tau217: β = 0.09, P = 0.004; p‐tau181: β = 0.08, P = 0.094, Table ). In the last analysis, we examined whether participants with intermediate/high ADNC pathology at death showed higher increments in p‐tau levels compared with those with none/low ADNC pathology. We observed that participants with intermediate/high ADNC had significantly higher p‐tau217, but not p‐tau181, longitudinal increases (p‐tau217: β = 0.13, P = 0.018; p‐tau181: β = 0.12, P = 0.152, Table and Appendix Fig ). These differences were observable up to 7 years before death, as defined by nonoverlapping 95%CIs. These results remained when removing two cases with very high plasma levels (p‐tau217: β = 0.21, P = 0.009; p‐tau181: β = 0.16, P = 0.118). Our first objective was to assess the associations between each plasma biomarker and the two neuropathological measures of AD pathology (i.e., total amount of plaques and tangles), independently. Plaque and tangle loads were measured on a semi‐quantitative scale that ranged from 0 to 3 in five different brain regions (Mirra et al , ), and we combined these regional measures into a total score (range: 0–15) for each pathology. We used partial Spearman's ρ to assess the association between each plasma biomarker and each total amount of pathology while adjusting for age, sex, and time between blood draw and death. We found that all plasma biomarkers except NfL (ρ = 0.10, P = 0.895) were significantly associated with the total amount of plaques (0.41 ≤ \|ρ\| ≤ 0.73, P < 0.001, Fig and Table ). Plasma p‐tau217 showed the highest correlation coefficient with plaques, which was significantly higher than all others (0.10 ≤ ρ diff ≤ 0.63, P ≤ 0.016) except plasma Aβ42/40 ratio (ρ diff = 0.19, P = 0.055). All plasma biomarkers except NfL (ρ = 0.19, P = 0.257) were also associated with the total amount of tangles in the independent models (0.26 ≤ \|ρ\| ≤ 0.66, P ≤ 0.016; Fig and Table ). Again, p‐tau217 had the highest correlation coefficient with the total amount of tangles, which was significantly higher than all other plasma correlation coefficients (0.11 ≤ ρ diff ≤ 0.47, P ≤ 0.006) except for that of plasma GFAP (ρ diff = 0.10, P = 0.206). As a sensitivity analysis, we also investigated these correlations separately for participants without (i.e., ADNC none or low) and with significant AD pathology (i.e., ADNC intermediate or high, Appendix Table ). In the group without significant AD pathology, only the Aβ42/40 ratio showed a significant correlation with amyloid (ρ = ‐0.33, P < 0.001). No plasma biomarkers showed a significant correlation with tau tangle load in this group. In the group of significant AD pathology, both p‐tau217 (ρ = 0.41, P = 0.049) and the Aβ42/40 ratio (ρ = ‐0.30, P < 0.001) presented a significant correlation with amyloid plaque load. Further, p‐tau217 (ρ = 0.56, P = 0.001), p‐tau181 (ρ = 0.49, P = 0.008), and GFAP (ρ = 0.47, P = 0.011) had a significant correlation with tau tangle load. Since plaque and tangle load were highly correlated (ρ[95%CI] = 0.63[0.48, 0.73], P < 0.001, Appendix Fig ), we performed an analysis to identify the specific (or independent) associations between each plasma biomarker and the two pathologies. For this, we used partial Spearman's ρ again, adjusting further for the other pathology as well (i.e., when looking at plaques adjusting for tangles and vice‐versa). In these models, p‐tau217 (ρ = 0.40, P = 0.003), p‐tau181 (ρ = 0.36, P = 0.009), and the Aβ42/40 ratio (ρ = ‐0.53, P < 0.001) were significantly associated with plaques (Fig and Table ). Plasma GFAP showed no significant association with plaque load when adjusted for tangle load (β = 0.09, P = 1.00) and p‐tau231 showed an association only at a trend level (ρ = 0.28, P = 0.084). In this analysis, the Aβ42/40 ratio correlation coefficient with plaques was the highest, being significantly higher than that of p‐tau231 (ρ diff = 0.25, P = 0.028) but not than that of p‐tau217 (ρ diff = 0.13, P = 0.246) nor p‐tau181 (ρ diff = 0.17, P = 0.146). In addition, we observed that only p‐tau217 (ρ = 0.52, P < 0.001), p‐tau181 (ρ = 0.36, P = 0.010), and GFAP (ρ = 0.39, P = 0.004) were associated with tangles. The correlation coefficient of p‐tau217 with tangles was significantly higher than that of p‐tau181 (ρ diff = 0.17, P = 0.004) when adjusting for the total amount of plaques but not than that of plasma GFAP (ρ diff = 0.13, P = 0.207). Comparing the correlation coefficients to each of these two pathologies for each biomarker, we observed three groups of biomarkers (Fig and Appendix Table ). We observed that plasma p‐tau231 (71.8%) and the Aβ42/40 ratio (83.1%) had a major proportion of variance explained by plaques than tangles. On the contrary, the opposite happened with plasma GFAP and NfL, with tangles explaining the major part of these biomarkers' variance (GFAP: 82.1%, NfL: 82.9%). Finally, in p‐tau217 (plaques: 43.4%, tangles: 56.6%) and p‐tau181 (plaques: 50.5%, tangles: 49.5%) both pathologies contributed similarly to explaining their variance. Finally, we investigated which combination of biomarkers better‐predicted plaques and tangles, independently. We found that the parsimonious model that better‐predicted load of amyloid plaques included both p‐tau217 and the Aβ42/40 ratio ( R 2 = 0.57, Table ), which was significantly better than the one only including p‐tau217 based on AICc (ΔAICc = 15.5). On the contrary, p‐tau217 alone was selected as the parsimonious model to predict the load of neurofibrillary tangles ( R 2 = 0.50, Table ). Next, we investigated differences in plasma levels by ADNC groups (as a four‐level variable, i.e., none, low, intermediate, or high) using a Kruskal‐Wallis test and Wilcoxon test for post hoc comparisons. All three p‐tau plasma measures showed significant differences between intermediate and high levels of ADNC. Plasma p‐tau217 and the Aβ42/40 ratio levels were significantly different between intermediate and low ADNC. However, we only found significant differences between none and low ADNC in plasma p‐tau217 levels (Fig ), although this became only a trend when removing the highest plasma value of the low group. Notably, p‐tau217 also showed the highest fold‐change among all ADNC consecutive levels (Appendix Table ). Then, we also examined differences in pathological scales specific for amyloid (Consortium to establish a registry for Alzheimer's disease [CERAD]) (Mirra et al , ) and tau (Braak staging) (Braak & Braak, ) pathologies. Similarly, all biomarkers except NfL showed significant differences between sparse and moderate/frequent groups on CERAD's classification. Only plasma p‐tau217 showed differences between zero and sparse groups (Appendix Fig ). Regarding Braak staging, all biomarkers except NfL were significantly different when comparing 0–IV with V–VI groups (Appendix Fig ). We next investigated the accuracy of each plasma biomarker to predict the presence of AD pathology as measured with the dichotomized ADNC (none/low vs. intermediate/high) classification. For this, we used receiver‐operating characteristic (ROC) curves and calculated the area under the curve (AUC) for each biomarker independently adjusting for age, sex, and time between blood sampling and death. All biomarkers except NfL (AUC[95%CI] = 0.61 [0.50, 0.71], P = 0.698) were predictive of the presence of ADNC when assessed individually (0.88 ≥ AUC ≥ 0.72, Appendix Table and Fig ). Plasma p‐tau217 had the highest AUC (AUC[95%CI] = 0.88 [0.81–0.95]) of all individual biomarkers, which was significantly higher than all others except for the Aβ42/40 ratio (AUC[95%CI] = 0.80 [0.72–0.89], P = 0.099). We also repeated this analysis with CERAD (low/sparse vs. moderate/frequent) and Braak staging (0–IV vs. V–VI) classification. For CERAD, p‐tau217 was also the best individual biomarker as per classification accuracy (AUC[95%CI] = 0.89 [0.83–0.96], P < 0.001), comparable only to that of the Aβ42/40 ratio (AUC[95%CI] = 0.82 [0.74–0.90], P < 0.001, Fig and Appendix Table ). For Braak staging, p‐tau217 again showed the highest accuracy (AUC[95%CI] = 0.93 [0.87–0.98], P < 0.001) comparable only to that of GFAP (AUC[95%CI] = 0.86[0.79–0.94], P < 0.001, Fig and Appendix Table ). Next, we investigated whether combining different biomarkers would improve the models with only individual biomarkers. We found that plasma p‐tau217 and the Aβ42/40 ratio was the optimal combination to predict the presence of ADNC (AUC[95%CI] = 0.90 [0.84, 0.96], Table and Fig ), but the AUC was not significantly higher than that of p‐tau217 alone when using the DeLong's test ( P = 0.124). With CERAD we observed a similar behavior, with plasma p‐tau217 and the Aβ42/40 ratio being the best combination (AUC[95%CI] = 0.91 [0.86–0.97], Appendix Table ), although the AUC was not significantly better than that of p‐tau217 alone ( P = 0.173). In the case of Braak staging, there were no combinations of biomarkers that improved the accuracy compared with p‐tau217‐only models. In this analysis, we investigated whether any of the available plasma biomarkers improved the basic models' accuracy (only covariates) on predicting the presence of co‐pathologies, using a similar approach as before but further adjusting for the presence of intermediate/high ADNC. We observed that only plasma NfL significantly improved the prediction of the presence of CWMR (Dugger et al , ) (AUC[95%CI] = 0.76 [0.66, 0.85]) compared with the basic model (AUC[95%CI] = 0.65 [0.54, 0.76], P = 0.028, Appendix Table and Appendix Fig A). In particular, participants with CWMR had significantly higher plasma NfL levels than those without (β = 0.88, P = 0.002, Appendix Fig B). No other biomarkers improved the prediction of the presence of this nor any other co‐pathology (Appendix Tables and Appendix Fig ). Raw distribution of plasma levels by the presence of each co‐pathology can be observed in Appendix Figs . As an additional analysis, we further checked whether there were differences between plasma levels in participants with only co‐pathologies (e.g., CAA only) and participants with AD pathology and co‐pathologies (e.g., CAA and ADNC) as it may have important clinical implications. We found that p‐tau217 was significantly higher in those participants having both AD pathology (as ADNC intermediate or high) and CAA compared to those with only AD pathology ( P = 0.037, Appendix Fig ). However, the group of AD‐only pathology was small ( n = 8). At the statistical trend level, we also observed differences in plasma Aβ42/40 levels in AD‐only versus AD and LBD groups ( P = 0.058, Appendix Fig ); in both p‐tau217 and Aβ42/40 levels in AD‐only versus AD and AGD groups ( P = 0.052 and P = 0.069, respectively; Appendix Fig ) and in NfL levels in AD‐only versus AD and CWMR ( P = 0.090, Appendix Fig ). No differences were observed in the case of TDP‐43 (Appendix Fig ). Finally, we also considered primary tauopathies (CBD, PSP, and AGD) as a unique group and compared the plasma levels of those participants to those with only AD pathology and those with AD pathology and other tauopathies. Plasma p‐tau217 ( P < 0.001), p‐tau181 ( P = 0.001), Aβ42/40 ratio ( P < 0.001), and GFAP ( P = 0.024) levels were significantly different when comparing participants with only AD pathology and participants with only primary tauopathies (Appendix Fig ). Only Aβ42/40 ratio levels were different between the AD group with CBD, PSP, or AGD pathology ( P = 0.038). Given that the CSF p‐tau/Aβ42 ratio is commonly used both in research and in clinical practice, we wanted to investigate whether a plasma p‐tau/Aβ42 ratio would also be useful for predicting AD pathology. For this, we selected p‐tau217 as it showed the highest associations in the previous analyses. We compared the accuracy of predicting plaques and tangles, independently, comparing parsimonious models including the plasma p‐tau217/Aβ42 ratio as a possible independent variable to those obtained in the previous sections including p‐tau217. We observed that the p‐tau217/Aβ42 ratio was preferentially selected over p‐tau217 in the models predicting plaques and tangles. Based on the AICc, we observed that models including the p‐tau217/Aβ42 ratio were slightly, but significantly, better than those previously presented (plaques: R 2 p‐tau217/Aβ42 ratio = 0.60, AICc = 210.1 vs. R 2 p‐tau217 = 0.57, AICc = 218.4; tangles: R 2 p‐tau217/Aβ42 ratio = 0.52, AICc = 228.7 vs. R 2 p‐tau217 = 0.50, AICc = 233.5; Appendix Table ). We also observed that p‐tau217/Aβ42 ratio levels were significantly different between the none versus low ADNC groups ( P = 0.029), even when removing the outlier in the low group. Finally, we investigated whether longitudinal changes in plasma p‐tau217 and p‐tau181 were associated with the presence of AD pathology at death (median[range] timepoints: 2 [2–5], mean (SD) time difference from first timepoint to death: 1378 (1357) days). Details of these participants can be found in Appendix Table . First, we observed that longitudinal increments of p‐tau217 but not p‐tau181 were associated with plaque burden (p‐tau217: β = 0.09, P = 0.005; p‐tau181: β = 0.05, P = 0.350, Table ). In independent models, we observed that p‐tau217 increments, but not those in p‐tau181, were also associated with tangle load (p‐tau217: β = 0.09, P = 0.004; p‐tau181: β = 0.08, P = 0.094, Table ). In the last analysis, we examined whether participants with intermediate/high ADNC pathology at death showed higher increments in p‐tau levels compared with those with none/low ADNC pathology. We observed that participants with intermediate/high ADNC had significantly higher p‐tau217, but not p‐tau181, longitudinal increases (p‐tau217: β = 0.13, P = 0.018; p‐tau181: β = 0.12, P = 0.152, Table and Appendix Fig ). These differences were observable up to 7 years before death, as defined by nonoverlapping 95%CIs. These results remained when removing two cases with very high plasma levels (p‐tau217: β = 0.21, P = 0.009; p‐tau181: β = 0.16, P = 0.118). In this study, we have investigated the specific associations between multiple plasma biomarkers, using high‐performing assays, and autopsy‐assessed measures of AD pathology in a single cohort. Our main result was that the plasma Aβ42/40 ratio and p‐tau231 were selectively associated with plaques, plasma GFAP only with tangles, whereas p‐tau181 and, most strongly, p‐tau217 were independently associated with both plaques and tangles. We also observed that p‐tau217 showed the highest accuracy to predict the presence of AD pathology. Regarding co‐pathologies, only the use of plasma NfL showed an improvement on predicting the presence of cerebral white matter rarefaction (CWMR), but no other biomarkers further improved this prediction nor any of any other co‐pathology. Notably, the use of the plasma p‐tau217/Aβ42 ratio showed slight, although significant, improvements compared with p‐tau217 alone when assessing semi‐quantitative measures of AD pathology. Finally, we observed that longitudinal increases in p‐tau217, but not those of p‐tau181, were significantly associated with the presence of AD pathology at death, especially with tangle burden. Taken altogether, this study supports the use of plasma p‐tau217, when assessed with high‐performing assays, as the best biomarker for measuring AD‐related pathology, supported by its independent associations with neuropathological measures of both plaques and tangles. The main result of this study was the observation that plasma p‐tau217 and plasma p‐tau181 were specific markers of both amyloid plaques and tau tangles. A previous study with a subsample of the individuals included here ( n = 88) already suggested an independent association between plasma p‐tau217 and the two main AD‐related pathologies (Mattsson‐Carlgren et al , ). The novelty of our study was to demonstrate that this dual association only occurred in p‐tau217 and p‐tau181. Further, we observed that p‐tau217 changed earlier along the ADNC scale (Fig ), and also that longitudinal changes in plasma p‐tau217, but not those of p‐tau181, were associated with AD‐related pathology. Although this analysis was exploratory, due to the limited sample size with longitudinal data, it is in agreement with a very recent study in which plasma p‐tau217 was the only biomarker with significantly different longitudinal increases based on amyloid status in both CU and MCI participants (Ashton et al , ). Altogether, our data suggest that plasma p‐tau217 is the best‐suited plasma biomarker among the ones studied here to assess the presence of AD‐related pathology across the whole continuum . Although p‐tau181 has shown very good performance as an AD biomarker (Karikari et al , , ; Thijssen et al , ; Janelidze et al , ; Grothe et al , ), multiple (plasma and CSF) studies support that p‐tau217 may be a more useful biomarker than p‐tau181, as it has stronger correlations with amyloid and tau pathology proxies, earlier change, and better diagnostic accuracy (Barthélemy et al , ; Hanes et al , ; Palmqvist et al , ; Janelidze et al , , ; Grothe et al , ; Leuzy et al , ). Further, our longitudinal results suggest that the utilization of plasma p‐tau217 in clinical trials may be useful not only as a prescreening method, but also for disease monitoring, especially for those drugs targeting tau pathology, but larger sample sizes are needed to confirm this finding. While plasma p‐tau217 and p‐tau181 were associated with both plaques and tangles, the other studied biomarkers showed more specific associations to only one pathology. For instance, when amyloid was not accounted for, plasma p‐tau231 showed a significant correlation with tangle counts; however, when taking into account the two pathologies, this biomarker only showed a significant association with amyloid plaques. Previous studies have suggested that p‐tau231, both as a CSF and a plasma biomarker, may be an early AD marker tightly associated with amyloid pathology (Suárez‐calvet et al , ; Ashton et al , ; Meyer et al , ; Milà‐Alomà et al , ; Smirnov et al , ). Contrary to previous studies, we observed significantly lower associations with amyloid than that of p‐tau217 and the Aβ42/40 ratio, and some elevated levels in subjects without neuropathological evidence of amyloid plaques (Fig ). One possible explanation for the early increases observed here and in previous studies may be that they are related to soluble amyloid, which cannot be detected in our study and is presumably an earlier event in the Alzheimer's continuum . We acknowledge that further research is needed to understand the relationship between this biomarker and actual pathology. Similarly, the plasma Aβ42/40 ratio was also only associated with plaques when both pathologies were included in a single model, supporting its tight relationship with amyloid pathology (Verberk et al , ; Janelidze et al , ). Nonetheless, the most important finding regarding the plasma Aβ42/40 ratio was that combining it with plasma p‐tau217 could slightly improve amyloid plaque assessment, replicating a previous result from our group when assessing amyloid positivity by CSF (Janelidze et al , ); however, in our case this improvement only reached significance when predicting the continuous variable. Thus, our results suggest that the combination of the plasma Aβ42/40 ratio and p‐tau217 may be useful in clinical trials targeting amyloid pathology as a prescreening method, but more powered studies are needed to confirm the additional value of the Aβ42/40 ratio. One surprising finding of our study was the specific association between plasma GFAP and tau tangles. Contrarily, previous studies have shown significant associations between this marker and amyloid pathology (as measured by CSF or PET), which were stronger than those with tau pathology (also measured with CSF or PET; Benedet et al , ; Pereira et al , ). Apart from the fact that GFAP levels were high in some cases with no or low amounts of plaques, two main points must be accounted when comparing these to our results. First, none of the aforementioned studies adjusted for tau pathology when assessing associations with amyloid. Following this approach (i.e., adjusting only for covariates), we also observed an association between plasma GFAP and plaques. And second, that tau PET is known to not be sensitive to early tau pathology, which may have decreased the power to detect these associations in previous studies (Mattsson et al , ; Leuzy et al , ; Soleimani‐meigooni et al , ). On the contrary, recent studies have shown the association between higher levels of plasma GFAP and increased risk of clinical progression and steeper rates of cognitive decline (even after adjusting for amyloid) (Rajan et al , ; Verberk et al , ; Ebenau et al , ), which supports a link with tau pathology given the known strong association between tau and clinical symptoms. Another plausible hypothesis is that plasma GFAP levels are not directly related to either plaque or tangle deposition but to the astrocytic reactivity in response to these processes. Actually, GFAP as a protein is overexpressed in reactive astrocytes, and its measures in CSF GFAP have been widely accepted as a marker of reactive astrogliosis. Unfortunately, no measures neuropathological measures of astrocytic reactivity were available in this sample, which prevented us to investigate this important issue. Future studies should investigate whether plasma GFAP is related to astrocytic reactivity and up to what level this is also indirectly related to amyloid and/or tau pathologies. The CSF p‐tau/Aβ42 ratio has received a lot of attention in recent years, both as a research and a clinical tool (Hansson et al , ; Fagan et al , ; Li et al , ; Snider et al , ; Milà‐Alomà et al , ; Salvadó & Larsson, ). Thus, we wanted to investigate whether a similar ratio would be also useful using plasma biomarkers. We observed that p‐tau217/Aβ42 ratio slightly, but significantly, improved prediction accuracy to detect AD‐related pathology, and seem to be able to detect earlier changes. Although further investigation is needed, we suggest that this ratio may be useful to track AD pathology across the continuum due to its relationship with both main pathological hallmarks of AD, as well as better statistical characteristics of ratios, which can account for production/clearance participants' inter‐variability (Janelidze et al , ; Hansson et al , ). Finally, we also investigated whether the levels of these biomarkers could be used to predict the presence of common AD co‐pathologies. Only plasma NfL significantly predicted the presence of CWMR (i.e., significantly improved the model with only covariates), with those subjects with the presence of CWMR having higher plasma NfL levels. This is in agreement with NfL being a biomarker of nonspecific axonal degeneration (Zetterberg et al , ; Bridel et al , ), but these results should be confirmed in an independent cohort. None of the biomarkers investigated could predict any of the other co‐pathologies investigated (i.e., CAA, TDP‐43, LBD, and AGD), which replicates some of the results from a recent study in a different cohort with a subsample of the biomarkers described here (Smirnov et al , ). Interestingly, we found that participants with AD and CAA pathologies had significantly higher levels of p‐tau217 than those with only CAA or only AD pathology. However, due to the low number of subjects with only AD pathology, we consider this as a hypothesis‐generating result that needs confirmation in a larger sample. Given the results from our study and previous studies, we emphasize the urgent need of developing new biomarkers capable of measuring the presence of these and other common co‐pathologies in vivo for a better diagnosis and prognosis for AD patients. The main strength of this study was the availability of high‐performing assays of multiple plasma biomarkers, including the three main p‐tau biomarkers phosphorylated at different sites, in a relatively large neuropathological cohort. Thus, we were able to directly compare specific associations between all of these biomarkers to gold standard measures of pathology in the same participants. Further, the use of semi‐quantitative scores for measuring the burden of AD pathology, compared with the typical dichotomous scales used, allowed us to perform more complex analyses. However, some limitations must be acknowledged. First, we recognize the small number of participants with intermediate levels of pathology and those with or without certain co‐pathologies. Another limitation is the restricted number of participants in the longitudinal subsample, which may have reduced the power to find significant time interactions with plasma p‐tau181. In this analysis, the big difference in the number of blood draws and in their time lags may have also affected our results. Thus, our results in this regard should be taken with caution. Also, we could only analyze p‐tau217 and p‐tau181 in this longitudinal sample, which did not allow a complete comparison among biomarkers. Finally, we acknowledge that replication in an independent is needed to establish the robustness of our results. In conclusion, our results support that plasma p‐tau217 and plasma p‐tau181 are specific markers of both amyloid plaques and tau tangles, whereas the Aβ42/40 ratio and p‐tau231 levels are markers strictly associated with plaques and GFAP with tangles. This is important when interpreting p‐tau measures in the A/T/N (Jack et al , ) context, as they may be more related to A (amyloid) than previously thought, as recently suggested (Groot et al , ; Moscoso et al , ; Therriault et al , ). Further, the combination of high‐performing assays of plasma p‐tau217 and the Aβ42/40 ratio gives the highest accuracy for predicting amyloid plaque load, while p‐tau217 alone may be sufficient to predict the load of tangles. These results may be useful to design prescreening strategies for clinical trials targeting amyloid and tau pathologies. Participants All samples were obtained through autopsies of subjects enrolled in the Arizona Study of Aging and Neurodegenerative Disorders and Brain and Body Donation Program (BBDP) at Banner Sun Health Research Institute (Beach et al , ). The BBDP recruits independently‐living normal and neurologically‐impaired elderly subjects predominantly from the surrounding Sun City's retirement communities. These volunteer research subjects are followed prospectively with annual standardized clinical assessments for the rest of their lives. Participants included in this study ranged from cognitively unimpaired to mild cognitive impairment and AD patients, as well as patients with other neurodegenerative diseases. We selected participants with both plasma and neuropathological exams available, including only those with all biomarkers available in the cross‐sectional analyses. Participants in the cross‐sectional analysis were also restricted as to those having blood drawing up to 5 years before death (mean (SD) [range] time: 482 (355) [9–1,760] days). All experiments were conducted in accordance with the Declaration of Helsinki. The operations of the Brain and Body Donation Program are approved by Institutional Review Boards and all participants or their legal representatives gave informed consent. Plasma biomarkers Plasma p‐tau217 and p‐tau181 concentrations were measured in‐house using an immunoassay developed by Lilly Research Laboratories (IN, USA), each of which had performed very well in multiple studies and cohorts (Palmqvist et al , ; Janelidze et al , ; Mielke et al , ; Thijssen et al , ). Plasma p‐tau231 concentration was also measured in‐house using a Simoa approach which was developed at the University of Gothenburg, which can detect Aβ pathology with high accuracy (Ashton et al , ). The remaining plasma biomarkers (Aβ42, Aβ40, GFAP, and NfL) were measured with prototype fully automated Elecsys® plasma immunoassays (updated versions for Aβ42 and Aβ40 (Palmqvist et al , )) plasma immunoassays (not commercially available) cobas e 601 and cobas e 411 analyzers (Roche Diagnostics International Ltd, Rotkreuz, Switzerland), also in‐house (Palmqvist et al , ). Longitudinal samples were only analyzed for p‐tau217 and p‐tau181, and not the Elecsys measurements or p‐tau231 for logistic reasons. Outliers were defined as subjects with values above or below more than 5 interquartile range of the third or the first quartile and were excluded from subsequent analyses. Only one plasma NfL value was considered an outlier. We also excluded another plasma NfL value that strongly affected the models (based on the standardized residuals) due to the very young age of the subject (44 years) and the strong correlation between NfL and age. Pathological measures All neuropathological measures were performed by a single US‐certified neuropathologist (TGB). Core AD pathology Tissue processing methods have been detailed previously (Beach et al , ). Histopathological scoring was performed blinded to clinical and neuropathological diagnosis, as well as levels of the plasma biomarkers. Amyloid plaque and neurofibrillary tangle density were graded at standard sites in frontal, temporal, and parietal cortices, as well as the hippocampus and entorhinal cortex, as previously explained (Beach et al , ). To obtain the total plaque score, each region was first rated as none, sparse, moderate, or frequent, using the published CERAD templates (Mirra et al , ). These descriptive measures were then converted into 0–3 scores in each region that combined and gave the total plaque score with a maximum value of 15. Neurofibrillary tangle abundance was measured similarly using the CERAD templates. Additionally, Braak staging was performed based on the topographical distribution of neurofibrillary tangle change (Braak & Braak, ). Global CERAD scoring and Thal phases (Thal et al , ) were also assessed as a measure of amyloid pathology. Using these three global scales we obtained a global measure of ADNC as described in the NIA‐AA guidelines (Montine et al , ). Although the ADNC score was used as a four‐scale measure in some analyses, we also dichotomized as a significant AD pathology (nonsignificant AD pathology) ADNC when scores were intermediate or high (none or low). Non‐AD pathology CAA was graded on a 0–3 scale based analogously on CERAD templates (Mirra et al , ) and dichotomized as positive if the score was above 1. Immunohistochemical staining in 10 brain regions for p129 alpha‐synuclein, as well as Thioflavin‐S (for substantia nigra) was used as a secondary stain to detect Lewy bodies and were staged based on the Unified Staging System (Beach et al , ). TDP‐43 positivity, location of positivity, and morphology were recorded as explained previously (Arnold et al , ). Significant CWMR was defined as exceeding 25% of the total centrum semiovale area within one or more cerebral lobes using hematoxylin and eosin on large format section protocol (Dugger et al , ). AGD was defined as typical spindle‐shaped structures revealed by the Gallyas silver stain (Josephs et al , ; Sabbagh et al , ). Statistical analyses First, partial Spearman's ρ was used to assess associations between each plasma biomarker (as dependent variable) and both plaques and tangles (as independent variables), independently. These models were adjusted for age, sex, and time between blood draw and death. To assess specific associations with each of these pathological measures, we also used partial Spearman's ρ further adjusting for the other pathology. Differences between two correlation coefficients were tested using a bootstrapping approach ( n = 1,000). Specific contribution of plaque and tangle loads on each biomarker concentration was obtained as the percentage of partial Spearman's ρ of each pathology over the sum of the partial Spearman's ρ of the two pathologies. Diagnostic accuracies of plasma biomarkers were assessed using ROC curve analysis, with age, sex, and time between blood draw and death as covariates. When assessing the diagnostic accuracy of non‐AD pathologies, the ADNC status as a dichotomous variable was also included as a covariate. Differences in the area under the curve (AUC) between two ROC curves were compared with the DeLong test (Robin et al , ). Differences in plasma levels between pathological groups were assessed with Kruskal–Wallis tests, with the pairwise Wilcoxon test as a post hoc comparison among groups (only differences between consecutive groups tested). We used the R package MuMIn to select the most parsimonious models to predict both continuous and dichotomous pathological measures following procedures described earlier (Palmqvist et al , ). Only plasma biomarkers showing a significant association with each pathological measure in the initial univariable models were included as possible predictors and the aforementioned covariates. All covariates were included in the final parsimonious model (even when they were not significant) to allow a fair comparison against univariable models. When multiple p‐tau (at different phosphorylation sites) biomarkers were possible predictors, independent models for each p‐tau marker were performed and then compared based on the corrected Akaike criterion (AICc) to avoid multicollinearity problems. Plasma levels were log‐transformed for this analysis. Linear mixed models (LME) were used to assess longitudinal changes in plasma biomarkers. Three independent models were performed for each biomarker. In each one, the interaction between time and one measure of AD pathology (i. plaques, ii. tangles, or iii. presence of ADNC) was used as a predictor. The LME models were adjusted for age and sex and included random intercepts and fixed slopes due to the limited number of datapoints (median[range]: 2[2–5]). Statistical analyses were done using R version 4.1.0. Significance was set at P < 0.05 (two‐tailed) and corrected for multiple comparisons using a false discovery rate (FDR). All samples were obtained through autopsies of subjects enrolled in the Arizona Study of Aging and Neurodegenerative Disorders and Brain and Body Donation Program (BBDP) at Banner Sun Health Research Institute (Beach et al , ). The BBDP recruits independently‐living normal and neurologically‐impaired elderly subjects predominantly from the surrounding Sun City's retirement communities. These volunteer research subjects are followed prospectively with annual standardized clinical assessments for the rest of their lives. Participants included in this study ranged from cognitively unimpaired to mild cognitive impairment and AD patients, as well as patients with other neurodegenerative diseases. We selected participants with both plasma and neuropathological exams available, including only those with all biomarkers available in the cross‐sectional analyses. Participants in the cross‐sectional analysis were also restricted as to those having blood drawing up to 5 years before death (mean (SD) [range] time: 482 (355) [9–1,760] days). All experiments were conducted in accordance with the Declaration of Helsinki. The operations of the Brain and Body Donation Program are approved by Institutional Review Boards and all participants or their legal representatives gave informed consent. Plasma p‐tau217 and p‐tau181 concentrations were measured in‐house using an immunoassay developed by Lilly Research Laboratories (IN, USA), each of which had performed very well in multiple studies and cohorts (Palmqvist et al , ; Janelidze et al , ; Mielke et al , ; Thijssen et al , ). Plasma p‐tau231 concentration was also measured in‐house using a Simoa approach which was developed at the University of Gothenburg, which can detect Aβ pathology with high accuracy (Ashton et al , ). The remaining plasma biomarkers (Aβ42, Aβ40, GFAP, and NfL) were measured with prototype fully automated Elecsys® plasma immunoassays (updated versions for Aβ42 and Aβ40 (Palmqvist et al , )) plasma immunoassays (not commercially available) cobas e 601 and cobas e 411 analyzers (Roche Diagnostics International Ltd, Rotkreuz, Switzerland), also in‐house (Palmqvist et al , ). Longitudinal samples were only analyzed for p‐tau217 and p‐tau181, and not the Elecsys measurements or p‐tau231 for logistic reasons. Outliers were defined as subjects with values above or below more than 5 interquartile range of the third or the first quartile and were excluded from subsequent analyses. Only one plasma NfL value was considered an outlier. We also excluded another plasma NfL value that strongly affected the models (based on the standardized residuals) due to the very young age of the subject (44 years) and the strong correlation between NfL and age. All neuropathological measures were performed by a single US‐certified neuropathologist (TGB). Core AD pathology Tissue processing methods have been detailed previously (Beach et al , ). Histopathological scoring was performed blinded to clinical and neuropathological diagnosis, as well as levels of the plasma biomarkers. Amyloid plaque and neurofibrillary tangle density were graded at standard sites in frontal, temporal, and parietal cortices, as well as the hippocampus and entorhinal cortex, as previously explained (Beach et al , ). To obtain the total plaque score, each region was first rated as none, sparse, moderate, or frequent, using the published CERAD templates (Mirra et al , ). These descriptive measures were then converted into 0–3 scores in each region that combined and gave the total plaque score with a maximum value of 15. Neurofibrillary tangle abundance was measured similarly using the CERAD templates. Additionally, Braak staging was performed based on the topographical distribution of neurofibrillary tangle change (Braak & Braak, ). Global CERAD scoring and Thal phases (Thal et al , ) were also assessed as a measure of amyloid pathology. Using these three global scales we obtained a global measure of ADNC as described in the NIA‐AA guidelines (Montine et al , ). Although the ADNC score was used as a four‐scale measure in some analyses, we also dichotomized as a significant AD pathology (nonsignificant AD pathology) ADNC when scores were intermediate or high (none or low). Non‐AD pathology CAA was graded on a 0–3 scale based analogously on CERAD templates (Mirra et al , ) and dichotomized as positive if the score was above 1. Immunohistochemical staining in 10 brain regions for p129 alpha‐synuclein, as well as Thioflavin‐S (for substantia nigra) was used as a secondary stain to detect Lewy bodies and were staged based on the Unified Staging System (Beach et al , ). TDP‐43 positivity, location of positivity, and morphology were recorded as explained previously (Arnold et al , ). Significant CWMR was defined as exceeding 25% of the total centrum semiovale area within one or more cerebral lobes using hematoxylin and eosin on large format section protocol (Dugger et al , ). AGD was defined as typical spindle‐shaped structures revealed by the Gallyas silver stain (Josephs et al , ; Sabbagh et al , ). Tissue processing methods have been detailed previously (Beach et al , ). Histopathological scoring was performed blinded to clinical and neuropathological diagnosis, as well as levels of the plasma biomarkers. Amyloid plaque and neurofibrillary tangle density were graded at standard sites in frontal, temporal, and parietal cortices, as well as the hippocampus and entorhinal cortex, as previously explained (Beach et al , ). To obtain the total plaque score, each region was first rated as none, sparse, moderate, or frequent, using the published CERAD templates (Mirra et al , ). These descriptive measures were then converted into 0–3 scores in each region that combined and gave the total plaque score with a maximum value of 15. Neurofibrillary tangle abundance was measured similarly using the CERAD templates. Additionally, Braak staging was performed based on the topographical distribution of neurofibrillary tangle change (Braak & Braak, ). Global CERAD scoring and Thal phases (Thal et al , ) were also assessed as a measure of amyloid pathology. Using these three global scales we obtained a global measure of ADNC as described in the NIA‐AA guidelines (Montine et al , ). Although the ADNC score was used as a four‐scale measure in some analyses, we also dichotomized as a significant AD pathology (nonsignificant AD pathology) ADNC when scores were intermediate or high (none or low). CAA was graded on a 0–3 scale based analogously on CERAD templates (Mirra et al , ) and dichotomized as positive if the score was above 1. Immunohistochemical staining in 10 brain regions for p129 alpha‐synuclein, as well as Thioflavin‐S (for substantia nigra) was used as a secondary stain to detect Lewy bodies and were staged based on the Unified Staging System (Beach et al , ). TDP‐43 positivity, location of positivity, and morphology were recorded as explained previously (Arnold et al , ). Significant CWMR was defined as exceeding 25% of the total centrum semiovale area within one or more cerebral lobes using hematoxylin and eosin on large format section protocol (Dugger et al , ). AGD was defined as typical spindle‐shaped structures revealed by the Gallyas silver stain (Josephs et al , ; Sabbagh et al , ). First, partial Spearman's ρ was used to assess associations between each plasma biomarker (as dependent variable) and both plaques and tangles (as independent variables), independently. These models were adjusted for age, sex, and time between blood draw and death. To assess specific associations with each of these pathological measures, we also used partial Spearman's ρ further adjusting for the other pathology. Differences between two correlation coefficients were tested using a bootstrapping approach ( n = 1,000). Specific contribution of plaque and tangle loads on each biomarker concentration was obtained as the percentage of partial Spearman's ρ of each pathology over the sum of the partial Spearman's ρ of the two pathologies. Diagnostic accuracies of plasma biomarkers were assessed using ROC curve analysis, with age, sex, and time between blood draw and death as covariates. When assessing the diagnostic accuracy of non‐AD pathologies, the ADNC status as a dichotomous variable was also included as a covariate. Differences in the area under the curve (AUC) between two ROC curves were compared with the DeLong test (Robin et al , ). Differences in plasma levels between pathological groups were assessed with Kruskal–Wallis tests, with the pairwise Wilcoxon test as a post hoc comparison among groups (only differences between consecutive groups tested). We used the R package MuMIn to select the most parsimonious models to predict both continuous and dichotomous pathological measures following procedures described earlier (Palmqvist et al , ). Only plasma biomarkers showing a significant association with each pathological measure in the initial univariable models were included as possible predictors and the aforementioned covariates. All covariates were included in the final parsimonious model (even when they were not significant) to allow a fair comparison against univariable models. When multiple p‐tau (at different phosphorylation sites) biomarkers were possible predictors, independent models for each p‐tau marker were performed and then compared based on the corrected Akaike criterion (AICc) to avoid multicollinearity problems. Plasma levels were log‐transformed for this analysis. Linear mixed models (LME) were used to assess longitudinal changes in plasma biomarkers. Three independent models were performed for each biomarker. In each one, the interaction between time and one measure of AD pathology (i. plaques, ii. tangles, or iii. presence of ADNC) was used as a predictor. The LME models were adjusted for age and sex and included random intercepts and fixed slopes due to the limited number of datapoints (median[range]: 2[2–5]). Statistical analyses were done using R version 4.1.0. Significance was set at P < 0.05 (two‐tailed) and corrected for multiple comparisons using a false discovery rate (FDR). Gemma Salvadó: Formal analysis; writing – original draft. Rik Ossenkoppele: Supervision; methodology; writing – review and editing. Nicholas J Ashton: Methodology; writing – review and editing. Thomas G Beach: Conceptualization; resources; funding acquisition; methodology; writing – review and editing. Geidy E Serrano: Resources; methodology; writing – review and editing. Eric M Reiman: Resources; funding acquisition. Henrik Zetterberg: Resources; methodology; writing – review and editing. Niklas Mattsson‐Carlgren: Supervision; writing – review and editing. Shorena Janelidze: Methodology; writing – review and editing. Kaj Blennow: Software; funding acquisition; methodology; writing – review and editing. Oskar Hansson: Conceptualization; resources; supervision; funding acquisition; project administration; writing – review and editing. In addition to the CRediT author contributions listed above, the contributions in detail are: GS, RO, and OH were involved in the study conception and design. GS analyzed the data. GS, RO, NM‐C, and OH interpreted the data. NJA and JS performed the experiments. GS wrote the manuscript. TGB, GES, ER, HZ, and KB participated in the acquisition of data. All authors critically reviewed and approved the final manuscript. HZ has served at scientific advisory boards and/or as a consultant for Abbvie, Alector, ALZPath, Annexon, Apellis, Artery Therapeutics, AZTherapies, CogRx, Denali, Eisai, Nervgen, Novo Nordisk, Passage Bio, Pinteon Therapeutics, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, and Wave, has given lectures in symposia sponsored by Cellectricon, Fujirebio, Alzecure, Biogen, and Roche, and is a co‐founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program (outside submitted work). KB has served as a consultant, at advisory boards, or at data monitoring committees for Abcam, Axon, BioArctic, Biogen, JOMDD/Shimadzu. Julius Clinical, Lilly, MagQu, Novartis, Ono Pharma, Pharmatrophix, Prothena, Roche Diagnostics, and Siemens Healthineers, and is a co‐founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program, outside the work presented in this paper. OH has acquired research support (for the institution) from ADx, AVID Radiopharmaceuticals, Biogen, Eli Lilly, Eisai, Fujirebio, GE Healthcare, Pfizer, and Roche. In the past 2 years, he has received consultancy/speaker fees from AC Immune, Amylyx, Alzpath, BioArctic, Biogen, Cerveau, Fujirebio, Genentech, Novartis, Roche, and Siemens. Appendix Click here for additional data file.
A toolbox of different approaches to analyze and present PRO-CTCAE data in oncology studies	d614515d-9184-4245-9707-a1824b2ac655	10165480	Internal Medicine[mh]	Overview of DREAMM-2 study and PRO-CTCAE data collection In the DREAMM-2 study, patients were randomly assigned to receive belantamab mafodotin 2.5 mg/kg or 3.4 mg/kg; the study was not designed to identify statistically significant differences in outcomes between doses . The DREAMM-2 study evaluated 28 items corresponding to 15 symptomatic toxicities (PRO-CTCAE version 1.0, with a 7-day recall period; , available online). PRO-CTCAE data were collected at screening, on day 1 of the first treatment cycle before drug administration, during each subsequent treatment cycle (starting at week 4 after first dose) or every 3 weeks, and at the end of the treatment visit (which occurred within 45 days after the last dose). The analyses were conducted using all available PRO-CTCAE data from patients who received at least 1 dose of belantamab mafodotin as part of the study in the 2.5- or 3.4-mg/kg arm. In this exploration, DREAMM-2 study PRO-CTCAE data are used to illustrate the analysis methods. Statistical analyses Methods for analyzing and visualizing PRO-CTCAE data were selected to be adaptable to the specific features of longitudinal PRO-CTCAE data for which independent single items with 4-level response scales are frequently collected. The methods selected were not exhaustive or intended as a definitive list of available approaches. Rather, the intention was to demonstrate how a variety of statistical methods could be applied to address a number of relevant research questions associated with the PRO-CTCAE. No statistical comparisons were made between the selected analysis methods. Equally, selection of specific AEs was not intended to compare results from the analyses but to serve only as illustrations of the methods. The administration of up to 28 items produced complex datasets with the possibility of skipped and missing data. Statistical techniques with various underlying assumptions and levels of sophistication were tested, as described in . Conditional branching logic was applied to PRO-CTCAE items in the study to reduce patient burden . For example, if a patient reported a symptomatic AE, they were asked about the severity and the extent to which it interfered with daily activities. If a patient did not report an AE, the follow-up questions were not posed, resulting in higher skipped data rates for some of the severity and interference items. In addition to these data skipped by design, participant discontinuation from treatment (mostly due to disease progression) was a major cause of increasing numbers of missing data over the duration of the study ( , available online). Descriptive and graphical visualizations and analysis based on modeling methods are shown in . As each method implied multiple statistical tests, given the number of PRO-CTCAE items , the Bonferroni adjustment was used to control for multiple comparisons, and a threshold of 0.0018 was used for statistical significance based on P values (.05/28). Missing data were not imputed, except for the area under the operating curve (AUC) analysis in the toxicity over time (ToxT) approach, which required this imputation to be valid (missing data would have created different follow-up time between participants, making the calculation of AUC invalid). Data analysis was performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA). In the DREAMM-2 study, patients were randomly assigned to receive belantamab mafodotin 2.5 mg/kg or 3.4 mg/kg; the study was not designed to identify statistically significant differences in outcomes between doses . The DREAMM-2 study evaluated 28 items corresponding to 15 symptomatic toxicities (PRO-CTCAE version 1.0, with a 7-day recall period; , available online). PRO-CTCAE data were collected at screening, on day 1 of the first treatment cycle before drug administration, during each subsequent treatment cycle (starting at week 4 after first dose) or every 3 weeks, and at the end of the treatment visit (which occurred within 45 days after the last dose). The analyses were conducted using all available PRO-CTCAE data from patients who received at least 1 dose of belantamab mafodotin as part of the study in the 2.5- or 3.4-mg/kg arm. In this exploration, DREAMM-2 study PRO-CTCAE data are used to illustrate the analysis methods. Methods for analyzing and visualizing PRO-CTCAE data were selected to be adaptable to the specific features of longitudinal PRO-CTCAE data for which independent single items with 4-level response scales are frequently collected. The methods selected were not exhaustive or intended as a definitive list of available approaches. Rather, the intention was to demonstrate how a variety of statistical methods could be applied to address a number of relevant research questions associated with the PRO-CTCAE. No statistical comparisons were made between the selected analysis methods. Equally, selection of specific AEs was not intended to compare results from the analyses but to serve only as illustrations of the methods. The administration of up to 28 items produced complex datasets with the possibility of skipped and missing data. Statistical techniques with various underlying assumptions and levels of sophistication were tested, as described in . Conditional branching logic was applied to PRO-CTCAE items in the study to reduce patient burden . For example, if a patient reported a symptomatic AE, they were asked about the severity and the extent to which it interfered with daily activities. If a patient did not report an AE, the follow-up questions were not posed, resulting in higher skipped data rates for some of the severity and interference items. In addition to these data skipped by design, participant discontinuation from treatment (mostly due to disease progression) was a major cause of increasing numbers of missing data over the duration of the study ( , available online). Descriptive and graphical visualizations and analysis based on modeling methods are shown in . As each method implied multiple statistical tests, given the number of PRO-CTCAE items , the Bonferroni adjustment was used to control for multiple comparisons, and a threshold of 0.0018 was used for statistical significance based on P values (.05/28). Missing data were not imputed, except for the area under the operating curve (AUC) analysis in the toxicity over time (ToxT) approach, which required this imputation to be valid (missing data would have created different follow-up time between participants, making the calculation of AUC invalid). Data analysis was performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA). Descriptive analyses of PRO-CTCAE data from DREAMM-2 study In the present analyses, 221 patients were included at baseline and by week 10, 132 patients remained in the analysis. Data for 136 patients were available at the end of treatment. An overview of the advantages and disadvantages of each analysis method can be found in . Visualization of maximum baseline-adjusted symptomatic AEs An overview of symptomatic AEs over the course of the DREAMM-2 study using aggregated maximum baseline-adjusted values for all PRO-CTCAE items is shown in . This visualization provides an overview of the symptomatic AEs experienced by patients during the study, displaying full distribution and variation in AE ratings, related to symptoms at baseline. This figure is a good summary but does not show how symptoms change over time, and handling of missing items requires careful consideration. Visualization of symptomatic AEs over the study follow-up Line graphs showing the mean change from baseline at each visit, per symptom type, provide a conventional approach to show the change of a symptomatic AE over time. The mean change from baseline in severity of nausea over time in the DREAMM-2 study is shown in as an illustrative example. Other approaches to describe PRO-CTCAE ratings over time are butterfly charts and stacked bar charts. Examples of these for severity of nausea in the DREAMM-2 study are presented in , respectively. These charts show the distribution of ratings for a given symptomatic AE at each timepoint per dose arm. Butterfly charts are similar to the visualization method used by the US Food and Drug Administration’s Project Patient Voice, an online platform developed to display patient-reported symptom data in a consistent manner across selected cancer trials . Visualization of cumulative frequencies of participants experiencing symptomatic AEs The number of participants who experience certain symptomatic AEs over the course of the study may be relevant information to present. Cumulative frequencies are useful tools for this purpose; provides example cumulative frequencies for participants with nausea during the DREAMM-2 study. These graphical representations show the proportion of patients experiencing a given level of a symptomatic AE and how quickly it occurs and provide an overall comparison of change in symptomatic AEs over time between treatment groups. Modeling longitudinal PRO-CTCAE data Comparing symptomatic AEs over time (ToxT) Accounting for within-patient variation with repeated-measures analysis of variance (ANOVA) A first analytical component of the ToxT approach is repeated-measures ANOVA. It is based on a linear model for repeated measures and controls for within-individual correlations. shows the adjusted mean PRO-CTCAE ratings for severity of constipation in DREAMM-2 participants, predicted from the model at each visit, for each dose arm along with confidence intervals. More details on model fit for the model assessing severity of constipation can be found in (available online). Repeated-measures ANOVA provides model-based estimates of mean PRO-CTCAE ratings that permit comparison of 2 treatment arms of a trial while considering between-patient variation. Time-to-event analysis using Kaplan–Meier (KM) estimates The second analytical component of the ToxT approach is a KM estimate of occurrence of the symptomatic AE of interest. The KM estimate for the time-to-event analysis of severity of constipation is shown in . This approach compares the time until participants in 2 treatment groups experience the specific symptomatic AE. Comparing symptomatic toxicities using AUC analysis with a single number The third analytical component of the ToxT approach is an AUC analysis. It provides a single number to quantify “cumulative” toxicity within each treatment arm. An example for severity of constipation in the DREAMM-2 study is shown in . AUC analysis is used to perform an overall comparison of the overall toxicity between treatment groups. Comparing symptomatic AEs considering the ordinal nature of PRO-CTCAE ratings with generalized estimating equations (GEE) models Dose arms in a study can be compared using GEE model analysis using the full data available while accounting for within-patient variation and the ordinal nature ratings for PRO-CTCAEs. The GEE model provides a point estimate with an associated P value to compare the PRO-CTCAE ratings over time between the 2 groups. An example GEE model output is shown in for shivering and shaking chills. More details on model fit for the model assessing frequency of shivering and shaking chills can be found in (available online). This approach, which is the most sophisticated tested here, may also be the most appropriate for PRO-CTCAE data given their ordinal nature. The odds ratio for the cumulative logit link function can be derived from model parameter estimates, the interpretation of which might be challenging and counterintuitive. As an example, GEE results are displayed in for frequency of shivering and shaking chills. The odds ratio of 0.58 reported in the table refers to the probability of having a rating in the 2.5-mg/kg arm of the DREAMM-2 study that is lower or equal to a rating in the 3.4-mg/kg arm. This implies that overall higher ratings are expected for this symptom in the 2.5-mg/kg arm (note that the difference was not statistically significant [ P = .007] after the Bonferroni adjustment). Comparing symptomatic AEs adjusting for baseline PRO-CTCAE ratings with ordinal log-linear models (OLLMs) OLLMs are versatile descriptive approaches permitting comparison of symptomatic AEs using aggregated variables or at specific timepoints, adjusting from baseline ratings. OLLMs provide interpretable estimates—odds ratios—comparing the PRO-CTCAE ratings between groups of interest, with an associated P value. OLLM output exemplars are shown in for frequency of heart palpitations, where the odds ratio of 0.72 indicates that the maximum postbaseline rating for the PRO-CTCAE item is likely to be lower in the 2.5-mg/kg arm of the DREAMM-2 study (though this was not statistically significant [ P = .10] after the Bonferroni adjustment). Various models with different specifications of the response scale coding were compared using Akaike information criterion (AIC), with a lower AIC indicating better fit. The model with equal spacing (assuming interval scale) was the one with the lowest AIC, suggesting the possible linearity of PRO-CTCAE ratings in the DREAMM-2 data ( , available online). In the present analyses, 221 patients were included at baseline and by week 10, 132 patients remained in the analysis. Data for 136 patients were available at the end of treatment. An overview of the advantages and disadvantages of each analysis method can be found in . Visualization of maximum baseline-adjusted symptomatic AEs An overview of symptomatic AEs over the course of the DREAMM-2 study using aggregated maximum baseline-adjusted values for all PRO-CTCAE items is shown in . This visualization provides an overview of the symptomatic AEs experienced by patients during the study, displaying full distribution and variation in AE ratings, related to symptoms at baseline. This figure is a good summary but does not show how symptoms change over time, and handling of missing items requires careful consideration. Visualization of symptomatic AEs over the study follow-up Line graphs showing the mean change from baseline at each visit, per symptom type, provide a conventional approach to show the change of a symptomatic AE over time. The mean change from baseline in severity of nausea over time in the DREAMM-2 study is shown in as an illustrative example. Other approaches to describe PRO-CTCAE ratings over time are butterfly charts and stacked bar charts. Examples of these for severity of nausea in the DREAMM-2 study are presented in , respectively. These charts show the distribution of ratings for a given symptomatic AE at each timepoint per dose arm. Butterfly charts are similar to the visualization method used by the US Food and Drug Administration’s Project Patient Voice, an online platform developed to display patient-reported symptom data in a consistent manner across selected cancer trials . Visualization of cumulative frequencies of participants experiencing symptomatic AEs The number of participants who experience certain symptomatic AEs over the course of the study may be relevant information to present. Cumulative frequencies are useful tools for this purpose; provides example cumulative frequencies for participants with nausea during the DREAMM-2 study. These graphical representations show the proportion of patients experiencing a given level of a symptomatic AE and how quickly it occurs and provide an overall comparison of change in symptomatic AEs over time between treatment groups. An overview of symptomatic AEs over the course of the DREAMM-2 study using aggregated maximum baseline-adjusted values for all PRO-CTCAE items is shown in . This visualization provides an overview of the symptomatic AEs experienced by patients during the study, displaying full distribution and variation in AE ratings, related to symptoms at baseline. This figure is a good summary but does not show how symptoms change over time, and handling of missing items requires careful consideration. Line graphs showing the mean change from baseline at each visit, per symptom type, provide a conventional approach to show the change of a symptomatic AE over time. The mean change from baseline in severity of nausea over time in the DREAMM-2 study is shown in as an illustrative example. Other approaches to describe PRO-CTCAE ratings over time are butterfly charts and stacked bar charts. Examples of these for severity of nausea in the DREAMM-2 study are presented in , respectively. These charts show the distribution of ratings for a given symptomatic AE at each timepoint per dose arm. Butterfly charts are similar to the visualization method used by the US Food and Drug Administration’s Project Patient Voice, an online platform developed to display patient-reported symptom data in a consistent manner across selected cancer trials . The number of participants who experience certain symptomatic AEs over the course of the study may be relevant information to present. Cumulative frequencies are useful tools for this purpose; provides example cumulative frequencies for participants with nausea during the DREAMM-2 study. These graphical representations show the proportion of patients experiencing a given level of a symptomatic AE and how quickly it occurs and provide an overall comparison of change in symptomatic AEs over time between treatment groups. Comparing symptomatic AEs over time (ToxT) Accounting for within-patient variation with repeated-measures analysis of variance (ANOVA) A first analytical component of the ToxT approach is repeated-measures ANOVA. It is based on a linear model for repeated measures and controls for within-individual correlations. shows the adjusted mean PRO-CTCAE ratings for severity of constipation in DREAMM-2 participants, predicted from the model at each visit, for each dose arm along with confidence intervals. More details on model fit for the model assessing severity of constipation can be found in (available online). Repeated-measures ANOVA provides model-based estimates of mean PRO-CTCAE ratings that permit comparison of 2 treatment arms of a trial while considering between-patient variation. Time-to-event analysis using Kaplan–Meier (KM) estimates The second analytical component of the ToxT approach is a KM estimate of occurrence of the symptomatic AE of interest. The KM estimate for the time-to-event analysis of severity of constipation is shown in . This approach compares the time until participants in 2 treatment groups experience the specific symptomatic AE. Comparing symptomatic toxicities using AUC analysis with a single number The third analytical component of the ToxT approach is an AUC analysis. It provides a single number to quantify “cumulative” toxicity within each treatment arm. An example for severity of constipation in the DREAMM-2 study is shown in . AUC analysis is used to perform an overall comparison of the overall toxicity between treatment groups. Comparing symptomatic AEs considering the ordinal nature of PRO-CTCAE ratings with generalized estimating equations (GEE) models Dose arms in a study can be compared using GEE model analysis using the full data available while accounting for within-patient variation and the ordinal nature ratings for PRO-CTCAEs. The GEE model provides a point estimate with an associated P value to compare the PRO-CTCAE ratings over time between the 2 groups. An example GEE model output is shown in for shivering and shaking chills. More details on model fit for the model assessing frequency of shivering and shaking chills can be found in (available online). This approach, which is the most sophisticated tested here, may also be the most appropriate for PRO-CTCAE data given their ordinal nature. The odds ratio for the cumulative logit link function can be derived from model parameter estimates, the interpretation of which might be challenging and counterintuitive. As an example, GEE results are displayed in for frequency of shivering and shaking chills. The odds ratio of 0.58 reported in the table refers to the probability of having a rating in the 2.5-mg/kg arm of the DREAMM-2 study that is lower or equal to a rating in the 3.4-mg/kg arm. This implies that overall higher ratings are expected for this symptom in the 2.5-mg/kg arm (note that the difference was not statistically significant [ P = .007] after the Bonferroni adjustment). Comparing symptomatic AEs adjusting for baseline PRO-CTCAE ratings with ordinal log-linear models (OLLMs) OLLMs are versatile descriptive approaches permitting comparison of symptomatic AEs using aggregated variables or at specific timepoints, adjusting from baseline ratings. OLLMs provide interpretable estimates—odds ratios—comparing the PRO-CTCAE ratings between groups of interest, with an associated P value. OLLM output exemplars are shown in for frequency of heart palpitations, where the odds ratio of 0.72 indicates that the maximum postbaseline rating for the PRO-CTCAE item is likely to be lower in the 2.5-mg/kg arm of the DREAMM-2 study (though this was not statistically significant [ P = .10] after the Bonferroni adjustment). Various models with different specifications of the response scale coding were compared using Akaike information criterion (AIC), with a lower AIC indicating better fit. The model with equal spacing (assuming interval scale) was the one with the lowest AIC, suggesting the possible linearity of PRO-CTCAE ratings in the DREAMM-2 data ( , available online). Accounting for within-patient variation with repeated-measures analysis of variance (ANOVA) A first analytical component of the ToxT approach is repeated-measures ANOVA. It is based on a linear model for repeated measures and controls for within-individual correlations. shows the adjusted mean PRO-CTCAE ratings for severity of constipation in DREAMM-2 participants, predicted from the model at each visit, for each dose arm along with confidence intervals. More details on model fit for the model assessing severity of constipation can be found in (available online). Repeated-measures ANOVA provides model-based estimates of mean PRO-CTCAE ratings that permit comparison of 2 treatment arms of a trial while considering between-patient variation. Time-to-event analysis using Kaplan–Meier (KM) estimates The second analytical component of the ToxT approach is a KM estimate of occurrence of the symptomatic AE of interest. The KM estimate for the time-to-event analysis of severity of constipation is shown in . This approach compares the time until participants in 2 treatment groups experience the specific symptomatic AE. Comparing symptomatic toxicities using AUC analysis with a single number The third analytical component of the ToxT approach is an AUC analysis. It provides a single number to quantify “cumulative” toxicity within each treatment arm. An example for severity of constipation in the DREAMM-2 study is shown in . AUC analysis is used to perform an overall comparison of the overall toxicity between treatment groups. A first analytical component of the ToxT approach is repeated-measures ANOVA. It is based on a linear model for repeated measures and controls for within-individual correlations. shows the adjusted mean PRO-CTCAE ratings for severity of constipation in DREAMM-2 participants, predicted from the model at each visit, for each dose arm along with confidence intervals. More details on model fit for the model assessing severity of constipation can be found in (available online). Repeated-measures ANOVA provides model-based estimates of mean PRO-CTCAE ratings that permit comparison of 2 treatment arms of a trial while considering between-patient variation. The second analytical component of the ToxT approach is a KM estimate of occurrence of the symptomatic AE of interest. The KM estimate for the time-to-event analysis of severity of constipation is shown in . This approach compares the time until participants in 2 treatment groups experience the specific symptomatic AE. The third analytical component of the ToxT approach is an AUC analysis. It provides a single number to quantify “cumulative” toxicity within each treatment arm. An example for severity of constipation in the DREAMM-2 study is shown in . AUC analysis is used to perform an overall comparison of the overall toxicity between treatment groups. Dose arms in a study can be compared using GEE model analysis using the full data available while accounting for within-patient variation and the ordinal nature ratings for PRO-CTCAEs. The GEE model provides a point estimate with an associated P value to compare the PRO-CTCAE ratings over time between the 2 groups. An example GEE model output is shown in for shivering and shaking chills. More details on model fit for the model assessing frequency of shivering and shaking chills can be found in (available online). This approach, which is the most sophisticated tested here, may also be the most appropriate for PRO-CTCAE data given their ordinal nature. The odds ratio for the cumulative logit link function can be derived from model parameter estimates, the interpretation of which might be challenging and counterintuitive. As an example, GEE results are displayed in for frequency of shivering and shaking chills. The odds ratio of 0.58 reported in the table refers to the probability of having a rating in the 2.5-mg/kg arm of the DREAMM-2 study that is lower or equal to a rating in the 3.4-mg/kg arm. This implies that overall higher ratings are expected for this symptom in the 2.5-mg/kg arm (note that the difference was not statistically significant [ P = .007] after the Bonferroni adjustment). OLLMs are versatile descriptive approaches permitting comparison of symptomatic AEs using aggregated variables or at specific timepoints, adjusting from baseline ratings. OLLMs provide interpretable estimates—odds ratios—comparing the PRO-CTCAE ratings between groups of interest, with an associated P value. OLLM output exemplars are shown in for frequency of heart palpitations, where the odds ratio of 0.72 indicates that the maximum postbaseline rating for the PRO-CTCAE item is likely to be lower in the 2.5-mg/kg arm of the DREAMM-2 study (though this was not statistically significant [ P = .10] after the Bonferroni adjustment). Various models with different specifications of the response scale coding were compared using Akaike information criterion (AIC), with a lower AIC indicating better fit. The model with equal spacing (assuming interval scale) was the one with the lowest AIC, suggesting the possible linearity of PRO-CTCAE ratings in the DREAMM-2 data ( , available online). The analyses reported here explored various statistical methods and data visualization techniques to improve the analysis and presentation and address a variety of relevant research questions related to PRO-CTCAE data from oncology clinical trials. The overall aim was to explore some of the different options and issues relating to data visualization and analyses of PRO-CTCAE data to promote further research as to how best to move beyond descriptive analyses and include examinations of the longitudinal trajectory of symptomatic AEs with statistical models. In some instances, visualization may precede the hypothesis in an exploratory setting when the objective may be to generate the research question. In other circumstances, the visualization is used to illustrate the analysis that has been carried out to address a predefined hypothesis. The appraisal of the analytical tools that we applied in this posthoc analysis of the DREAMM-2 study data will help in the selection of relevant techniques for future oncology studies. Multiple options were considered for the description and visualization of PRO-CTCAE data, including replication of the visualization of the pilot case of the US Food and Drug Administration’s Project Patient Voice . These techniques highlighted the importance of precise definition of the message, objective, and targeted audience in order to select the best visualization and that a fine balance is needed between simplicity and interpretable display of the complexity of the results. Additionally, we investigated the modeling methods that would be appropriate for statistical comparison of longitudinal PRO-CTCAE data between groups in a clinical trial. The most sophisticated methods, such as GEE models or OLLM, did not provide additional insight beyond the more standard techniques included in the ToxT approach. The ToxT approach, which to our knowledge has been applied to PRO-CTCAE for the first time, may be a good compromise for standard use in this context. The methods we tested, particularly the more sophisticated ones, may be used on a supplemental basis to address specific objectives. For instance, OLLM can be used to study the appropriateness of treating PRO-CTCAE item ratings as ordinal or interval variables by comparing the results with different specifications of the response coding. The OLLM analysis result showed that using a “linear” scoring system for the PRO-CTCAE items worked well for the DREAMM-2 study data. This suggests that using raw PRO-CTCAE ratings in models that assume linearity (such as those used in the ToxT approach) can be acceptable and may inform further future development of methods to analyze PRO-CTCAE data. Although our findings did not identify any statistically significant differences between arms using the more sophisticated modeling methods for longitudinal PRO-CTCAE data, these methods should not be dismissed solely based on their performance in the DREAMM-2 study because the trial compared 2 doses of belantamab mafodotin and was not designed to detect differences between the PRO-CTCAE items. In clinical studies containing cohorts treated with different drugs instead of the same drug at different doses, these statistical modeling methods would be more likely to show greater differences in PRO-CTCAE item ratings, and their utility should be confirmed in such studies. Our work is certainly not a comprehensive list of all of the valid or useful approaches that can be applied for the analysis of PRO-CTCAE data. We identified the application cases corresponding to typical research questions and objectives that could be informed by the analyses of PRO-CTCAE data, and we searched for appropriate analyses in this setting. This led to the series of analyses and visualization techniques applied here, but others could certainly be imagined. This is true for visualization where the graphical representation options are almost infinite [eg, Sankey diagrams could be used ] but also for modeling approaches where other methods also could have been appropriate (eg, cumulative link mixed models) . For example, the toxicity index might be useful because this can detect difference in toxicity profile between treatments and does not rely on maximum grades often used for CTCAE data or average summary measures as used by ToxT . Inferential statistics are typically not relevant for PRO-CTCAE items but may be an option in some cases . Selection of the most appropriate approach may depend on the research question being asked and the target audience. Collection of PRO-CTCAE data generates large and complex datasets that present a number of challenges in terms of analysis. The main challenge we identified was the high dimensionality of the data, where many data points needed handling (eg, individual item scores) and a high number of assessments occurred for each individual over the course of the study. In this context, multiplicity of statistical testing and type I error rate inflation should be considered; several approaches are possible for this purpose (eg, Bonferroni adjustment but also Holm or Hochberg procedures). The choice of the best method to deal with multiplicity will have to be informed by the objective of the analyses and the risk of type I error that the analyst is ready to accept. Given that PRO-CTCAE items are increasingly being collected in clinical studies and the need for both clinicians and patients to clearly interpret results, future work should include optimizing the visual presentation of the data to improve clarity for these different audiences. This could potentially include developing dynamic tools or interactive interfaces for PRO-CTCAE item visualization. It is critical that the message conveyed by any visualization is well received by the target audience. Therefore, further research to gain feedback from the targeted audience, including clinicians and patients, on the various graphical visualizations of PRO-CTCAE would identify the most effective graphics. This could be informed by previous research conducted for more general longitudinal PRO data . For example, researchers have found that clearly indicating the directionality of PRO scoring was important for accurate interpretation, particularly for patient audiences, while normed data were more likely to be misinterpreted than non-normed data . Similarly, studies have found line graphs easy to understand and pie charts easier to interpret than bar charts for data presenting proportions changed. Recent work explored aggregating PRO-CTCAE items through the use of composite scores . In this approach, scores on up to 3 attributes of frequency, severity, and interference with daily activities for each AE are combined into a single metric for each AE, which may be beneficial when visualizing a more succinct version of the data is preferred. These approaches are promising, but they also require further specific research, especially to inform their interpretation, which goes beyond the simple illustration purpose that we had with this research. The second major challenge we identified was handling the large amount of skipped and missing data of different types, including missing attributes, intermittent missing data, and informative missing data (in other words, missing data that do not randomly occur but relate to the quality of life of the patient; eg, if the patient is too ill to complete the questionnaire). Identifying the precise reasons for skipped and missing data in longitudinal studies is important to best match the analysis and imputation approach and may provide additional insight into interpretation of the study. For example, if the missingness is informative, excluding these data could bias the interpretation of the results . Similarly, clarifying whether missing data due to study discontinuation is due to disease progression as opposed to toxicity would be revealing, especially if the research question was to summarize tolerability over the course of the study. Indeed, the importance of correctly handling skipped and missing data is recognized in recommendations on using PRO-CTCAE data being developed by the NCI Cancer Moonshot Standardization Working Group . In the DREAMM-2 study, some data were skipped due to conditional branching logic, where, due to the format of the questioning algorithm, certain follow-up questions were not posed if a patient did not report experiencing a specific AE. PRO-CTCAE data were electronically collected during clinic visits, so there may have been a lower burden of missing data than might be expected from postal survey methods in which patients may not return every survey . There was an increasing burden of missing data over time as a progressive number of patients dropped out of the study over time due to disease progression. This was expected given the natural history of the disease in question and the longitudinal nature of the study. More work on the missing PRO-CTCAE data challenges and solution, including how to calculate percentages, accounting for patient discontinuations due to disease progression during interpretation, and how to handle missing data (best imputation strategy), would be useful for the increasing uptake of PRO-CTCAE collection and analysis. In conclusion, many analytical and visualization techniques are available to process PRO CTCAE data depending on the research question of interest, but also on the targeted audience. This study highlights a toolbox of potential methods that could be optimized to the specific requirements of oncology trials, which may be complemented by other analytical techniques in the future. djad018_Supplementary_Data Click here for additional data file.
Effectiveness of system navigation programs linking primary care with community-based health and social services: a systematic review	8a77e605-d769-43f0-b86d-fffa75f91dcb	10165767	Patient-Centered Care[mh]	Patients and their caregivers often face significant challenges when navigating increasingly complex health and social services. Frequently left to locate and access these siloed services alone , adults living with multifaceted health and social needs have described their care as disjointed, confusing, and uncoordinated . Barriers to accessing available health and social services may include restrictive eligibility criteria and wait lists for services, financial constraints, health literacy and communication challenges, lack of transportation, and poor coordination between primary care providers and health and social service agencies . In an effort to overcome this fragmentation and efficiently access the care they need, patients and caregivers often spend an extraordinary amount of time becoming informal system navigators and de facto care coordinators . This can have significant physical, emotional, social, relational, and financial repercussions [ , , ]. Given the rising prevalence of chronic diseases and multimorbidity worldwide , in addition to urgent calls to address the social and structural inequities that exist in health systems , identifying effective strategies to support individuals in accessing high-quality health and social care is of vital importance. Over the last 30 years, system navigation programs have gained popularity globally as a person-centred approach to support individuals to access health and social care [ – ] . Established initially to overcome health inequities in cancer care , system navigation has since expanded into areas such as chronic disease management , mental health , and to facilitate access to care for marginalized and historically underserved populations (e.g., persons experiencing homelessness, food insecurity, living in low-income countries) . Various terms are used in the literature to describe individuals who provide navigation support, such as patient navigators, community health workers, case managers, and link workers . For this review, system navigation is defined as programs that link the patient’s primary healthcare delivery and community-based health and social services to create integrated, patient-focused care . System navigation can be facilitated by an individual or team of lay and/or healthcare professionals to reduce barriers and facilitate access to continuous, effective, and efficient care for patients, caregivers, and providers . Despite growing interest and calls to expand navigation programs for the general public to enhance integrated care delivery , an understanding of the effectiveness of system navigation overall, and characteristics of effective models is largely unknown. A previous scoping review to identify navigation models and factors influencing the implementation of navigation programs linking primary care with community-based health and social services found the key motivators for implementing such programs included improving the delivery of health and social services to meet patient/population health needs and improve quality of life; however, this review included primarily descriptive, observational, and qualitative studies. In conclusion, Valaitis and colleagues recommended a systematic review of primary care-based system navigation programs as a critical next step to determine program effectiveness and inform practice and policy decision-making related to optimal models and impacts. As the body of literature has grown, this systematic review builds upon the previous scoping review of system navigation programs to identify the effectiveness of system navigation programs linking primary care with community-based health and social services to improve patient, caregiver, and health system outcomes when compared to usual care. This systematic review was registered with PROSPERO (CRD42020205050). The reporting of this review is based on PRISMA guidelines . Search strategy The search strategy was built upon the previous scoping review of navigation programs linking primary care with community-based health and social services . Updating the previously conducted search, the electronic databases PsychInfo, EMBASE, CINAHL, OVID MEDLINE, and Cochrane Clinical Trials Registry were searched from January 1, 2013, to August 10, 2020 (Additional file ). A health science librarian trained in building searches for systematic reviews consulted on the search strategy. In line with the previous scoping review, database searches were limited to studies published in the English language only. Study selection Identified references were uploaded to Covidence (Veritas Health Innovation Ltd., Melbourne, Australia) and duplicates were removed. Titles and abstracts were independently screened by two reviewers for inclusion based on predetermined eligibility criteria. Full texts of potentially relevant studies were retrieved and screened by two independent reviewers. Conflicts were resolved through discussion or with the input of a third reviewer, as needed. Included studies from the previous scoping review were also reviewed independently and in duplicate to determine eligibility, as the previous review included qualitative and observational studies, in addition to intervention studies. Eligibility criteria Types of studies To determine intervention effectiveness, eligible studies were limited to experimental and quasi-experimental designs, including randomized controlled trials (RCTs), non-randomized controlled trials, and single group, pre-test/post-test intervention studies. Mixed methods studies with eligible quantitative designs were also included; however, only quantitative data were extracted. Qualitative, observational, descriptive, and cross-sectional studies were excluded. Participants Eligible studies included adults 18 years of age and older utilizing primary care. In contrast to the previous scoping review, studies that focused on disease-specific populations (e.g., cancer, mental health) were excluded to allow broader transferability and inform effective interventions to support health and social care access among general patient populations. However, studies that included patients with a variety of chronic diseases or chronic disease risk factors were eligible, given that the interventions described were not disease specific. Interventions System navigation programs based in a primary care setting that aimed to link patients to appropriate community-based health and social services were included. Primary care was defined as care delivered at the entry point into the healthcare system, which is typically provided by a physician or nurse practitioner . Social prescription programs, which link users to community social services that may be considered outside of the healthcare system , were eligible. In line with the original scoping review, we initially intended to include system navigation programs linking primary care to other medical specialty care services. However, we later decided to include interventions that went beyond health system navigation alone to focus on integrated, upstream, and community-based approaches. This decision was made in light of mounting evidence that integrated health and social care interventions focused on addressing the social determinants of health can improve health outcomes and reduce the use of costlier health services . Given the distinct role and function of case managers as clinical care providers, which may extend beyond the scope of system navigation , interventions that focused exclusively on case management were excluded. However, interventions that included a case management component in addition to system navigation were eligible. Comparators Studies that compared an intervention to any non-intervention comparison group were eligible, including pre-intervention data or data from a non-exposed control group. Outcomes The primary outcomes of interest were access to care (i.e., timely use of healthcare and/or social services to achieve improved health outcomes) and health and social service utilization. Secondary outcomes included patient-related (e.g., general health and wellbeing, quality of life, self-efficacy) and caregiver outcomes (e.g., caregiver burden, self-efficacy). Upon review of included studies, it became apparent that experience measures (e.g., satisfaction with the quality of care) and cost-related outcomes were also relevant. Thus, these other outcomes were added after the initial PROSPERO registration. Assessment of methodological quality Two independent reviewers critically appraised all eligible studies to assess methodological quality using the Joanna Briggs Institute Critical Appraisal tools for experimental and quasi-experimental studies . Conflicts were resolved through discussion between reviewers and input from a third reviewer when needed. Data extraction Two independent reviewers extracted data using a pre-tested template; discrepancies were resolved through discussion or input from a third reviewer when needed. The data abstraction template included study characteristics (i.e., aim, study design, country), participant characteristics (i.e., number of participants, population description, age, sex, ethnicity, socioeconomic status), description of any comparator groups, limitations, and conclusions as reported by study authors. The Template for Intervention Description and Replication (TIDieR) checklist guided extraction of intervention components . For relevant outcomes, the measure, effect, variation, and statistical significance were extracted. Authors were contacted to obtain missing data. Data collection forms are available upon request. Data synthesis System navigation programs were grouped based on the navigation models identified in the previous scoping review, including lay person-led (i.e., non-healthcare professionals within primary care who perform specific activities related to system navigation), health professional-led (e.g., nurse or social worker who performs specific activities related to system navigation), and team-based (i.e., lay persons and health professionals together, or teams of health professionals) . Results of individual studies were organized into tables by intervention type and outcomes (i.e., type, data collection tool, and measure of effect and significance) to facilitate synthesis and identify possible sources of heterogeneity. A meta-analysis was deemed inappropriate given the wide range of system navigation models and outcomes identified; instead, a narrative approach to synthesis was used , with data presented in corresponding tables. Reporting bias was not explored because most studies did not cite trial registrations or protocols. A comprehensive approach to assess the overall certainty of the evidence for each outcome (e.g., GRADE) was not used due to heterogeneity across interventions and outcomes. Patient and public involvement Key research partners, including four older adult citizens and one community-based social service provider, were included in the review team. The aim of patient and public involvement in this systematic review was to support the interpretation of the results and identify key takeaways to inform the co-design of a community-based intervention to enhance physical activity, nutrition, and system navigation among older adults experiencing health inequities . This was achieved through virtual working group meetings and the collaborative development of knowledge translation products, including a public-facing infographic and research brief. The search strategy was built upon the previous scoping review of navigation programs linking primary care with community-based health and social services . Updating the previously conducted search, the electronic databases PsychInfo, EMBASE, CINAHL, OVID MEDLINE, and Cochrane Clinical Trials Registry were searched from January 1, 2013, to August 10, 2020 (Additional file ). A health science librarian trained in building searches for systematic reviews consulted on the search strategy. In line with the previous scoping review, database searches were limited to studies published in the English language only. Identified references were uploaded to Covidence (Veritas Health Innovation Ltd., Melbourne, Australia) and duplicates were removed. Titles and abstracts were independently screened by two reviewers for inclusion based on predetermined eligibility criteria. Full texts of potentially relevant studies were retrieved and screened by two independent reviewers. Conflicts were resolved through discussion or with the input of a third reviewer, as needed. Included studies from the previous scoping review were also reviewed independently and in duplicate to determine eligibility, as the previous review included qualitative and observational studies, in addition to intervention studies. Types of studies To determine intervention effectiveness, eligible studies were limited to experimental and quasi-experimental designs, including randomized controlled trials (RCTs), non-randomized controlled trials, and single group, pre-test/post-test intervention studies. Mixed methods studies with eligible quantitative designs were also included; however, only quantitative data were extracted. Qualitative, observational, descriptive, and cross-sectional studies were excluded. Participants Eligible studies included adults 18 years of age and older utilizing primary care. In contrast to the previous scoping review, studies that focused on disease-specific populations (e.g., cancer, mental health) were excluded to allow broader transferability and inform effective interventions to support health and social care access among general patient populations. However, studies that included patients with a variety of chronic diseases or chronic disease risk factors were eligible, given that the interventions described were not disease specific. Interventions System navigation programs based in a primary care setting that aimed to link patients to appropriate community-based health and social services were included. Primary care was defined as care delivered at the entry point into the healthcare system, which is typically provided by a physician or nurse practitioner . Social prescription programs, which link users to community social services that may be considered outside of the healthcare system , were eligible. In line with the original scoping review, we initially intended to include system navigation programs linking primary care to other medical specialty care services. However, we later decided to include interventions that went beyond health system navigation alone to focus on integrated, upstream, and community-based approaches. This decision was made in light of mounting evidence that integrated health and social care interventions focused on addressing the social determinants of health can improve health outcomes and reduce the use of costlier health services . Given the distinct role and function of case managers as clinical care providers, which may extend beyond the scope of system navigation , interventions that focused exclusively on case management were excluded. However, interventions that included a case management component in addition to system navigation were eligible. Comparators Studies that compared an intervention to any non-intervention comparison group were eligible, including pre-intervention data or data from a non-exposed control group. Outcomes The primary outcomes of interest were access to care (i.e., timely use of healthcare and/or social services to achieve improved health outcomes) and health and social service utilization. Secondary outcomes included patient-related (e.g., general health and wellbeing, quality of life, self-efficacy) and caregiver outcomes (e.g., caregiver burden, self-efficacy). Upon review of included studies, it became apparent that experience measures (e.g., satisfaction with the quality of care) and cost-related outcomes were also relevant. Thus, these other outcomes were added after the initial PROSPERO registration. To determine intervention effectiveness, eligible studies were limited to experimental and quasi-experimental designs, including randomized controlled trials (RCTs), non-randomized controlled trials, and single group, pre-test/post-test intervention studies. Mixed methods studies with eligible quantitative designs were also included; however, only quantitative data were extracted. Qualitative, observational, descriptive, and cross-sectional studies were excluded. Eligible studies included adults 18 years of age and older utilizing primary care. In contrast to the previous scoping review, studies that focused on disease-specific populations (e.g., cancer, mental health) were excluded to allow broader transferability and inform effective interventions to support health and social care access among general patient populations. However, studies that included patients with a variety of chronic diseases or chronic disease risk factors were eligible, given that the interventions described were not disease specific. System navigation programs based in a primary care setting that aimed to link patients to appropriate community-based health and social services were included. Primary care was defined as care delivered at the entry point into the healthcare system, which is typically provided by a physician or nurse practitioner . Social prescription programs, which link users to community social services that may be considered outside of the healthcare system , were eligible. In line with the original scoping review, we initially intended to include system navigation programs linking primary care to other medical specialty care services. However, we later decided to include interventions that went beyond health system navigation alone to focus on integrated, upstream, and community-based approaches. This decision was made in light of mounting evidence that integrated health and social care interventions focused on addressing the social determinants of health can improve health outcomes and reduce the use of costlier health services . Given the distinct role and function of case managers as clinical care providers, which may extend beyond the scope of system navigation , interventions that focused exclusively on case management were excluded. However, interventions that included a case management component in addition to system navigation were eligible. Studies that compared an intervention to any non-intervention comparison group were eligible, including pre-intervention data or data from a non-exposed control group. The primary outcomes of interest were access to care (i.e., timely use of healthcare and/or social services to achieve improved health outcomes) and health and social service utilization. Secondary outcomes included patient-related (e.g., general health and wellbeing, quality of life, self-efficacy) and caregiver outcomes (e.g., caregiver burden, self-efficacy). Upon review of included studies, it became apparent that experience measures (e.g., satisfaction with the quality of care) and cost-related outcomes were also relevant. Thus, these other outcomes were added after the initial PROSPERO registration. Two independent reviewers critically appraised all eligible studies to assess methodological quality using the Joanna Briggs Institute Critical Appraisal tools for experimental and quasi-experimental studies . Conflicts were resolved through discussion between reviewers and input from a third reviewer when needed. Two independent reviewers extracted data using a pre-tested template; discrepancies were resolved through discussion or input from a third reviewer when needed. The data abstraction template included study characteristics (i.e., aim, study design, country), participant characteristics (i.e., number of participants, population description, age, sex, ethnicity, socioeconomic status), description of any comparator groups, limitations, and conclusions as reported by study authors. The Template for Intervention Description and Replication (TIDieR) checklist guided extraction of intervention components . For relevant outcomes, the measure, effect, variation, and statistical significance were extracted. Authors were contacted to obtain missing data. Data collection forms are available upon request. System navigation programs were grouped based on the navigation models identified in the previous scoping review, including lay person-led (i.e., non-healthcare professionals within primary care who perform specific activities related to system navigation), health professional-led (e.g., nurse or social worker who performs specific activities related to system navigation), and team-based (i.e., lay persons and health professionals together, or teams of health professionals) . Results of individual studies were organized into tables by intervention type and outcomes (i.e., type, data collection tool, and measure of effect and significance) to facilitate synthesis and identify possible sources of heterogeneity. A meta-analysis was deemed inappropriate given the wide range of system navigation models and outcomes identified; instead, a narrative approach to synthesis was used , with data presented in corresponding tables. Reporting bias was not explored because most studies did not cite trial registrations or protocols. A comprehensive approach to assess the overall certainty of the evidence for each outcome (e.g., GRADE) was not used due to heterogeneity across interventions and outcomes. Key research partners, including four older adult citizens and one community-based social service provider, were included in the review team. The aim of patient and public involvement in this systematic review was to support the interpretation of the results and identify key takeaways to inform the co-design of a community-based intervention to enhance physical activity, nutrition, and system navigation among older adults experiencing health inequities . This was achieved through virtual working group meetings and the collaborative development of knowledge translation products, including a public-facing infographic and research brief. Description of included studies The updated search identified 15,226 unique records (Fig. ). Following title and abstract screening, 387 full texts were retrieved and assessed for eligibility. A total of 21 studies published between 2009 and 2020 were included (Table ); 19 of these were newly identified, and 2 were included in the previous scoping review. A list of excluded studies with reasons for exclusion is provided in Additional file . Study designs included RCTs ( n = 8, 38%) [ – ], single group, pre-test/post-test designs ( n = 7, 33%) [ – ], and two group, non-randomized designs ( n = 6, 29%) [ – ]. Studies most often took place in the United States of America ( n = 9, 43%) [ , , , , , , – ] or the United Kingdom ( n = 8, 38%) [ , , , – , ]. A total of 10,743 participants (range 19 to 2,325 across studies) are represented, and, when mean ages were reported, the median mean age across studies was 72 years (range 49 to 82 years). Primary care-based system navigation program models included 1) lay person-led ( n = 10, 48%) [ – , , – , , ], 2) health professional-led ( n = 4, 19%) [ , , , ], and 3) team-based ( n = 6, 29%) [ , – , , ]. A fourth model was also identified, which included self-navigation based on a personalized list of local resources with lay support available ( n = 1, 5%) . In studies that used a primarily lay person-led model, most ( n = 7, 70%) described comprehensive navigator training and employed lay navigators as staff [ , , , , , , ]. This training ranged from 3 h of online training to a 16-week community college health coaching course . In studies that used health professional-led models, system navigation was primarily nurse-led [ , , ] or social worker-led ; however, in one multi-site study, health professionals varied by setting and also included a nurse practitioner or physician assistant in system navigation roles . The team-based navigation models included either lay person(s) and health professional(s) together [ , , , ] or teams of health professionals who provided system navigation support. Intervention duration and frequency of contact were highly variable across the included studies. The median length of system navigation programs was 6 months (range 2 to 30 months). Of the 17 studies that reported intervention frequency, most programs were delivered variably based on individual patient needs ( n = 9, 53%) [ , , – , – , ], while others occurred monthly ( n = 4, 24%) [ , , , ], weekly ( n = 2, 12%) , bi-monthly ( n = 1, 6%) , or one-time-only ( n = 1, 6%) . Theoretical models or frameworks were reported in only 33% ( n = 7) of studies to support the rationale for system navigation programs; these included the Chronic Care Model [ , , , ], the biopsychosocial model , the integral conceptual model of frailty , and a theory of community-based primary care . A full description of intervention characteristics based on the TIDieR framework is presented in Table . Methodological quality Overall, the included studies had generally low to moderate risk of bias. Within the 8 RCTs, the risk of bias was primarily attributed to the absence of blinding among participants and interventionists (Fig. ). The lack of control groups and incomplete follow-up predominantly contributed to the risk of bias among the 13 quasi-experimental studies (Fig. ). Full critical appraisal assessments for each study are reported in Additional files and for RCTs and quasi-experimental studies, respectively. Effectiveness of system navigation programs A summary of findings by system navigation model and outcome category alongside a summary of the risk of bias is provided in Table . Complete data used for analyses for each outcome are provided in Additional files – . Health and social service access and utilization outcomes The 13 studies that reported health service utilization evaluated lay person-led ( n = 6, 46%) [ , , , , , ], health professional-led ( n = 4, 31%) [ , , , ], and team-based ( n = 3, 23%) [ , , ] system navigation models. Health service utilization was primarily captured through administrative, health record, and/or health insurance data related to the number of primary care visits ( n = 10, 77%) [ , , , – , , , , ], hospital admissions and/or readmissions ( n = 9, 69%) [ – , , , , , ], emergency care visits ( n = 7, 54%) [ , , , , , , ], and home care visits ( n = 4, 31%) [ , , , ] (Additional file ). None of the included studies reported healthcare access or social service utilization outcomes. Overall, findings for lay person-led models were mixed. Three studies demonstrated improvements in health service utilization following lay person-led system navigation programs [ , , ]. Compared to baseline, patients at high risk for avoidable hospital admissions due to medical or psychosocial issues who accessed the lay person-led Integrated Care Coordination Service had a statistically significant decrease in emergency department attendance and hospital admissions nine months post-referral (low risk of bias) . Patients living in high-poverty areas who participated in the standardized, 6-month community health worker-led goal setting plus Individualized Management for Patient-Centered Targets (IMPaCT) program (tailored coaching, social support, navigation, advocacy) also had significantly lower odds of repeat admissions, but no difference in overall hospital admissions or length of stay when compared to goal setting plus usual care (low risk of bias) . Compared to usual care, community health worker-led system navigation including patient education, appointment scheduling, and assistance overcoming barriers to healthcare access significantly increased the rate of primary care provider and/or chronic disease nurse visits among patients with chronic health needs who were classified as unengaged with their medical care (i.e., had not seen a primary care physician in last 6 months) (moderate risk of bias) . Further, a higher percentage of these patients visited a primary care provider before seeking other providers for their health needs . However, three studies demonstrated no significant changes following lay person-led system navigation programs when compared to baseline or usual care (moderate risk of bias) [ , , ]. Similarly, the effectiveness of health professional-led system navigation on health service utilization outcomes was unclear. A social worker-led social prescribing program for patients with chronic conditions, polypharmacy, or frequent primary care attendance was associated with a significant decrease in the number of primary care physician visits, but no difference in home visits, telephone visits, or care contacts when compared to usual care in one study (moderate risk of bias) . No significant impacts on health service utilization were observed in three other studies following health professional-led system navigation programs when compared to usual care (low-moderate risk of bias) [ , , ]. In contrast, team-based system navigation models demonstrated some positive impacts on health service utilization across three studies with low risk of bias [ , , ]. In the 6-month Health TAPESTRY program, volunteer-led home visits followed by action planning with the healthcare team and links to community support resulted in a statistically significant increase in primary care visits and reduced rates of hospitalization among older adults, with no significant changes in emergency department visits when compared to usual care . Similarly, social worker and volunteer-led social prescribing to community services resulted in a significantly lower rate of annual general practitioner consultations with no significant impact on emergency department visits among adult patients experiencing social isolation with a history of frequent primary care visits, as compared to matched patients from a neighbouring area . However, it should be noted that this study lacked randomization, and patients assigned to the intervention group had a significantly higher rate of general practitioner consultations at baseline compared to their matched counterparts. Finally, a health coach and link worker-led intervention involving a needs assessment and referral to relevant community services also significantly decreased primary care use over a 3-month time period among patients managing at least one long-term health condition and experiencing social isolation when compared to baseline . Patient-related outcomes In total, 16 studies captured patient-related outcomes [ – , , , – , – ]. These were grouped into four categories: 1) quality of life/health-related quality of life, mental health, and wellbeing, 2) social participation and function, 3) health behaviours, and 4) theoretical constructs related to behaviour change. Quality of life/health-related quality of life, mental health, and wellbeing In total, 13 studies investigated the impact of lay person-led ( n = 5, 39%) [ – , , ], health professional-led ( n = 3, 23%) [ , , ], team-based ( n = 4, 31%) [ , , , ], and self-navigation with lay support as needed ( n = 1, 8%) system navigation models on quality of life/health-related quality of life, mental health, and wellbeing outcomes. These outcomes were most often measured using the 12- or 36-Item Short Form Survey (SF-12, SF-36) ( n = 5, 39%) [ , , , , ], EuroQol-5 Dimension ( n = 5, 39%) [ , , , , ], Hospital Anxiety and Depression Scale ( n = 2, 15%) , or the Warwick-Edinburgh Mental Wellbeing Scale ( n = 2, 15%) . Various other single-item and self-report measures were used (Additional file ). Findings for lay person-led system navigation models were mixed. Social prescribing to local community health and wellbeing resources resulted in reduced anxiety and depression, better self-reported health, as well as a statistically and clinically significant improvement in patient wellbeing when compared to baseline in one study (moderate risk of bias) . However, another social prescribing program found a statistically significant, but not clinically significant difference in wellbeing among patients with multiple chronic conditions experiencing social isolation/loneliness when compared to baseline (moderate risk of bias) . Further, no significant changes in wellbeing, anxiety, depression, or health-related quality of life were found following the Community Links Practitioner intervention when compared to usual care (high risk of bias) . The standardized goal setting plus IMPaCT intervention significantly improved health-related quality of life in the mental domain, but not the physical domain of the SF-12 when compared to goal setting plus usual care in one study (moderate risk of bias) . However, no significant changes were observed in physical or mental health-related quality of life in another study evaluating the goal setting plus IMPaCT intervention when compared to usual care (low risk of bias) . Findings for health professional-led system navigation models were also mixed. The Urban Health Centres Europe approach including health assessment, shared decision making, and referral to appropriate health and social service care pathways (led by either a social worker, nurse, nurse practitioner, or physician assistant based on the setting) significantly improved health-related quality of life compared to usual care (low risk of bias) . However, two studies using nurse-led system navigation models did not result in significant improvements in health-related quality of life compared to usual care (low-moderate risk of bias) . None of the team-based or self-navigation with lay support system navigation models significantly improved quality of life/health-related quality of life, mental health, or wellbeing outcomes compared to baseline or usual care (low-moderate risk of bias) [ , , , , ]. Social participation and function Social participation and function was evaluated in eight studies including lay person-led ( n = 2, 25%) , health professional-led ( n = 2, 25%) , and team-based ( n = 4, 50%) [ , , , ] system navigation models. Various measures were used, including heterogeneous assessments of loneliness [ , , ], social networks , participation in social roles [ , , ], and social group memberships (Additional file ). Overall, the findings were mixed. Of the lay person-led models, social prescribing by wellbeing coordinators significantly increased social networks compared to baseline in one study (moderate risk of bias) . However, no changes in social participation were found following the Community Links Practitioner intervention compared to usual care in another study (high risk of bias) . Neither of the studies that used a health professional-led model found significant differences in social participation and function outcomes (low risk of bias) . Of the team-based models, the health coach and link worker-led intervention for adults managing long-term health conditions and experiencing social isolation, loneliness, or anxiety significantly improved the number of social group memberships from baseline, but did not impact community belonging or loneliness (low risk of bias) . Three additional studies evaluating team-based system navigation models found no significant differences in social participation and function outcomes (low-moderate risk of bias) [ , , ]. Health behaviours Health behaviours were assessed in seven studies evaluating lay person-led ( n = 4, 57%) [ – , ], health professional-led ( n = 1, 14%) , and team-based ( n = 2, 29%) system navigation models. Outcomes included heterogeneous measurements of physical activity/exercise [ , , , , ], cigarette smoking [ , , ], alcohol intake , and diet (Additional file ). Overall, the findings were mixed. Lay person-led social prescribing significantly increased physical activity compared to baseline in one study (moderate risk of bias) . However, three additional studies evaluating lay person-led models found no significant differences in health behaviour outcomes, including cigarette smoking or exercise level (low-moderate risk of bias) [ – ]. The study that evaluated a health professional-led model compared to usual care did not find significant differences in healthy lifestyle behaviours (low risk of bias) . Of the team-based system navigation models, an integrated health management intervention with referral to community programs led by community health centre staff and a multidisciplinary care team led to significant improvements in health behaviours including physical activity, alcohol intake, diet, and smoking habits when compared to bimonthly health education (high risk of bias) . However, another team-based model did not significantly impact physical activity levels compared to usual care (low risk of bias) . Patient activation, self-efficacy, and empowerment Patient activation, self-efficacy, and empowerment were evaluated in five studies including lay person-led ( n = 3, 60%) [ , , ], team-based ( n = 1, 20%) , and self-navigation with lay support as needed ( n = 1, 20%) system navigation models. Heterogeneous measurements of self-efficacy [ , , ], patient activation , and empowerment were used. Overall, the findings were mixed. Of the lay person-led models, the Cities for Live Program including linkage to community programs following an assessment of needs, barriers, and stage of change significantly improved self-efficacy compared to baseline (moderate risk of bias) . However, the standardized lay person-led goal setting plus IMPaCT intervention did not change patient activation in two studies (low-moderate risk of bias) . No significant changes in goal attainment, self-efficacy, or patient empowerment were observed following team-based system navigation in one study (low risk of bias) . Although limited to evidence from one study evaluating a self-navigation with lay support system navigation model, patients who participated in the “HealtheRx” intervention involving an electronic-medical record generated personalized list of local community resources with access to a community health information specialist as needed were more likely to report higher confidence in finding resources in their community to help manage their health compared to usual care (low risk of bias) . Patient experience outcomes Patient experience outcomes were reported in five studies, including lay person-led ( n = 2, 40%) , health professional-led ( n = 2, 40%) , and team-based ( n = 1, 20%) system navigation models. Patient experiences with care quality were measured using the Consumer Assessment of Healthcare Providers and Systems-Patient Centered Medical Home survey , Patient Assessment of Chronic Illness Care tool , and Canadian Institute for Health Information common indicators (Additional file ). Both lay person-led and health professional-led system navigation models consistently improved patient experiences with quality of care. The community health worker-led goal setting plus IMPaCT intervention significantly improved care comprehensiveness and self-management supportiveness when compared to goal setting plus usual care in two RCTs (low-moderate risk of bias) . Compared to usual care, the nurse-led Guided Care and Community Connections Program also significantly improved overall patient experiences with the quality of their care (low-moderate risk of bias). Only one study evaluated the impact of team-based system navigation on patient experiences; the Health TAPESTRY program did not significantly improve patient experiences (i.e., level of difficulty accessing healthcare resources, care comprehensiveness, patient-centeredness, satisfaction) when compared to usual care (low risk of bias) . Caregiver outcomes Caregiver experience and health outcomes were reported in two studies that investigated health professional-led system navigation models . Overall, the findings were unclear. Compared to usual care, caregiver experiences (i.e., perception of patient care quality) improved after the nurse-led Guided Care intervention (moderate risk of bias) but not after the nurse-led Community Connections Program (low risk of bias) . Evidence from only one study demonstrated no impact of the nurse-led Guided Care intervention on caregiver strain and depression (moderate risk of bias) (Additional file ). Cost-related outcomes Only two studies reported on cost-related outcomes; both evaluated a lay person-led system navigation model . The cost of emergency department/hospital visits and emergency care per patient were compared to costs in a matched control group in one study (moderate risk of bias) and projected annual cost savings based on mathematical modelling in another (low risk of bias) . Although both studies reported differences between groups, no formal statistical tests were reported (Additional file ). The updated search identified 15,226 unique records (Fig. ). Following title and abstract screening, 387 full texts were retrieved and assessed for eligibility. A total of 21 studies published between 2009 and 2020 were included (Table ); 19 of these were newly identified, and 2 were included in the previous scoping review. A list of excluded studies with reasons for exclusion is provided in Additional file . Study designs included RCTs ( n = 8, 38%) [ – ], single group, pre-test/post-test designs ( n = 7, 33%) [ – ], and two group, non-randomized designs ( n = 6, 29%) [ – ]. Studies most often took place in the United States of America ( n = 9, 43%) [ , , , , , , – ] or the United Kingdom ( n = 8, 38%) [ , , , – , ]. A total of 10,743 participants (range 19 to 2,325 across studies) are represented, and, when mean ages were reported, the median mean age across studies was 72 years (range 49 to 82 years). Primary care-based system navigation program models included 1) lay person-led ( n = 10, 48%) [ – , , – , , ], 2) health professional-led ( n = 4, 19%) [ , , , ], and 3) team-based ( n = 6, 29%) [ , – , , ]. A fourth model was also identified, which included self-navigation based on a personalized list of local resources with lay support available ( n = 1, 5%) . In studies that used a primarily lay person-led model, most ( n = 7, 70%) described comprehensive navigator training and employed lay navigators as staff [ , , , , , , ]. This training ranged from 3 h of online training to a 16-week community college health coaching course . In studies that used health professional-led models, system navigation was primarily nurse-led [ , , ] or social worker-led ; however, in one multi-site study, health professionals varied by setting and also included a nurse practitioner or physician assistant in system navigation roles . The team-based navigation models included either lay person(s) and health professional(s) together [ , , , ] or teams of health professionals who provided system navigation support. Intervention duration and frequency of contact were highly variable across the included studies. The median length of system navigation programs was 6 months (range 2 to 30 months). Of the 17 studies that reported intervention frequency, most programs were delivered variably based on individual patient needs ( n = 9, 53%) [ , , – , – , ], while others occurred monthly ( n = 4, 24%) [ , , , ], weekly ( n = 2, 12%) , bi-monthly ( n = 1, 6%) , or one-time-only ( n = 1, 6%) . Theoretical models or frameworks were reported in only 33% ( n = 7) of studies to support the rationale for system navigation programs; these included the Chronic Care Model [ , , , ], the biopsychosocial model , the integral conceptual model of frailty , and a theory of community-based primary care . A full description of intervention characteristics based on the TIDieR framework is presented in Table . Overall, the included studies had generally low to moderate risk of bias. Within the 8 RCTs, the risk of bias was primarily attributed to the absence of blinding among participants and interventionists (Fig. ). The lack of control groups and incomplete follow-up predominantly contributed to the risk of bias among the 13 quasi-experimental studies (Fig. ). Full critical appraisal assessments for each study are reported in Additional files and for RCTs and quasi-experimental studies, respectively. A summary of findings by system navigation model and outcome category alongside a summary of the risk of bias is provided in Table . Complete data used for analyses for each outcome are provided in Additional files – . Health and social service access and utilization outcomes The 13 studies that reported health service utilization evaluated lay person-led ( n = 6, 46%) [ , , , , , ], health professional-led ( n = 4, 31%) [ , , , ], and team-based ( n = 3, 23%) [ , , ] system navigation models. Health service utilization was primarily captured through administrative, health record, and/or health insurance data related to the number of primary care visits ( n = 10, 77%) [ , , , – , , , , ], hospital admissions and/or readmissions ( n = 9, 69%) [ – , , , , , ], emergency care visits ( n = 7, 54%) [ , , , , , , ], and home care visits ( n = 4, 31%) [ , , , ] (Additional file ). None of the included studies reported healthcare access or social service utilization outcomes. Overall, findings for lay person-led models were mixed. Three studies demonstrated improvements in health service utilization following lay person-led system navigation programs [ , , ]. Compared to baseline, patients at high risk for avoidable hospital admissions due to medical or psychosocial issues who accessed the lay person-led Integrated Care Coordination Service had a statistically significant decrease in emergency department attendance and hospital admissions nine months post-referral (low risk of bias) . Patients living in high-poverty areas who participated in the standardized, 6-month community health worker-led goal setting plus Individualized Management for Patient-Centered Targets (IMPaCT) program (tailored coaching, social support, navigation, advocacy) also had significantly lower odds of repeat admissions, but no difference in overall hospital admissions or length of stay when compared to goal setting plus usual care (low risk of bias) . Compared to usual care, community health worker-led system navigation including patient education, appointment scheduling, and assistance overcoming barriers to healthcare access significantly increased the rate of primary care provider and/or chronic disease nurse visits among patients with chronic health needs who were classified as unengaged with their medical care (i.e., had not seen a primary care physician in last 6 months) (moderate risk of bias) . Further, a higher percentage of these patients visited a primary care provider before seeking other providers for their health needs . However, three studies demonstrated no significant changes following lay person-led system navigation programs when compared to baseline or usual care (moderate risk of bias) [ , , ]. Similarly, the effectiveness of health professional-led system navigation on health service utilization outcomes was unclear. A social worker-led social prescribing program for patients with chronic conditions, polypharmacy, or frequent primary care attendance was associated with a significant decrease in the number of primary care physician visits, but no difference in home visits, telephone visits, or care contacts when compared to usual care in one study (moderate risk of bias) . No significant impacts on health service utilization were observed in three other studies following health professional-led system navigation programs when compared to usual care (low-moderate risk of bias) [ , , ]. In contrast, team-based system navigation models demonstrated some positive impacts on health service utilization across three studies with low risk of bias [ , , ]. In the 6-month Health TAPESTRY program, volunteer-led home visits followed by action planning with the healthcare team and links to community support resulted in a statistically significant increase in primary care visits and reduced rates of hospitalization among older adults, with no significant changes in emergency department visits when compared to usual care . Similarly, social worker and volunteer-led social prescribing to community services resulted in a significantly lower rate of annual general practitioner consultations with no significant impact on emergency department visits among adult patients experiencing social isolation with a history of frequent primary care visits, as compared to matched patients from a neighbouring area . However, it should be noted that this study lacked randomization, and patients assigned to the intervention group had a significantly higher rate of general practitioner consultations at baseline compared to their matched counterparts. Finally, a health coach and link worker-led intervention involving a needs assessment and referral to relevant community services also significantly decreased primary care use over a 3-month time period among patients managing at least one long-term health condition and experiencing social isolation when compared to baseline . Patient-related outcomes In total, 16 studies captured patient-related outcomes [ – , , , – , – ]. These were grouped into four categories: 1) quality of life/health-related quality of life, mental health, and wellbeing, 2) social participation and function, 3) health behaviours, and 4) theoretical constructs related to behaviour change. Quality of life/health-related quality of life, mental health, and wellbeing In total, 13 studies investigated the impact of lay person-led ( n = 5, 39%) [ – , , ], health professional-led ( n = 3, 23%) [ , , ], team-based ( n = 4, 31%) [ , , , ], and self-navigation with lay support as needed ( n = 1, 8%) system navigation models on quality of life/health-related quality of life, mental health, and wellbeing outcomes. These outcomes were most often measured using the 12- or 36-Item Short Form Survey (SF-12, SF-36) ( n = 5, 39%) [ , , , , ], EuroQol-5 Dimension ( n = 5, 39%) [ , , , , ], Hospital Anxiety and Depression Scale ( n = 2, 15%) , or the Warwick-Edinburgh Mental Wellbeing Scale ( n = 2, 15%) . Various other single-item and self-report measures were used (Additional file ). Findings for lay person-led system navigation models were mixed. Social prescribing to local community health and wellbeing resources resulted in reduced anxiety and depression, better self-reported health, as well as a statistically and clinically significant improvement in patient wellbeing when compared to baseline in one study (moderate risk of bias) . However, another social prescribing program found a statistically significant, but not clinically significant difference in wellbeing among patients with multiple chronic conditions experiencing social isolation/loneliness when compared to baseline (moderate risk of bias) . Further, no significant changes in wellbeing, anxiety, depression, or health-related quality of life were found following the Community Links Practitioner intervention when compared to usual care (high risk of bias) . The standardized goal setting plus IMPaCT intervention significantly improved health-related quality of life in the mental domain, but not the physical domain of the SF-12 when compared to goal setting plus usual care in one study (moderate risk of bias) . However, no significant changes were observed in physical or mental health-related quality of life in another study evaluating the goal setting plus IMPaCT intervention when compared to usual care (low risk of bias) . Findings for health professional-led system navigation models were also mixed. The Urban Health Centres Europe approach including health assessment, shared decision making, and referral to appropriate health and social service care pathways (led by either a social worker, nurse, nurse practitioner, or physician assistant based on the setting) significantly improved health-related quality of life compared to usual care (low risk of bias) . However, two studies using nurse-led system navigation models did not result in significant improvements in health-related quality of life compared to usual care (low-moderate risk of bias) . None of the team-based or self-navigation with lay support system navigation models significantly improved quality of life/health-related quality of life, mental health, or wellbeing outcomes compared to baseline or usual care (low-moderate risk of bias) [ , , , , ]. Social participation and function Social participation and function was evaluated in eight studies including lay person-led ( n = 2, 25%) , health professional-led ( n = 2, 25%) , and team-based ( n = 4, 50%) [ , , , ] system navigation models. Various measures were used, including heterogeneous assessments of loneliness [ , , ], social networks , participation in social roles [ , , ], and social group memberships (Additional file ). Overall, the findings were mixed. Of the lay person-led models, social prescribing by wellbeing coordinators significantly increased social networks compared to baseline in one study (moderate risk of bias) . However, no changes in social participation were found following the Community Links Practitioner intervention compared to usual care in another study (high risk of bias) . Neither of the studies that used a health professional-led model found significant differences in social participation and function outcomes (low risk of bias) . Of the team-based models, the health coach and link worker-led intervention for adults managing long-term health conditions and experiencing social isolation, loneliness, or anxiety significantly improved the number of social group memberships from baseline, but did not impact community belonging or loneliness (low risk of bias) . Three additional studies evaluating team-based system navigation models found no significant differences in social participation and function outcomes (low-moderate risk of bias) [ , , ]. Health behaviours Health behaviours were assessed in seven studies evaluating lay person-led ( n = 4, 57%) [ – , ], health professional-led ( n = 1, 14%) , and team-based ( n = 2, 29%) system navigation models. Outcomes included heterogeneous measurements of physical activity/exercise [ , , , , ], cigarette smoking [ , , ], alcohol intake , and diet (Additional file ). Overall, the findings were mixed. Lay person-led social prescribing significantly increased physical activity compared to baseline in one study (moderate risk of bias) . However, three additional studies evaluating lay person-led models found no significant differences in health behaviour outcomes, including cigarette smoking or exercise level (low-moderate risk of bias) [ – ]. The study that evaluated a health professional-led model compared to usual care did not find significant differences in healthy lifestyle behaviours (low risk of bias) . Of the team-based system navigation models, an integrated health management intervention with referral to community programs led by community health centre staff and a multidisciplinary care team led to significant improvements in health behaviours including physical activity, alcohol intake, diet, and smoking habits when compared to bimonthly health education (high risk of bias) . However, another team-based model did not significantly impact physical activity levels compared to usual care (low risk of bias) . Patient activation, self-efficacy, and empowerment Patient activation, self-efficacy, and empowerment were evaluated in five studies including lay person-led ( n = 3, 60%) [ , , ], team-based ( n = 1, 20%) , and self-navigation with lay support as needed ( n = 1, 20%) system navigation models. Heterogeneous measurements of self-efficacy [ , , ], patient activation , and empowerment were used. Overall, the findings were mixed. Of the lay person-led models, the Cities for Live Program including linkage to community programs following an assessment of needs, barriers, and stage of change significantly improved self-efficacy compared to baseline (moderate risk of bias) . However, the standardized lay person-led goal setting plus IMPaCT intervention did not change patient activation in two studies (low-moderate risk of bias) . No significant changes in goal attainment, self-efficacy, or patient empowerment were observed following team-based system navigation in one study (low risk of bias) . Although limited to evidence from one study evaluating a self-navigation with lay support system navigation model, patients who participated in the “HealtheRx” intervention involving an electronic-medical record generated personalized list of local community resources with access to a community health information specialist as needed were more likely to report higher confidence in finding resources in their community to help manage their health compared to usual care (low risk of bias) . Patient experience outcomes Patient experience outcomes were reported in five studies, including lay person-led ( n = 2, 40%) , health professional-led ( n = 2, 40%) , and team-based ( n = 1, 20%) system navigation models. Patient experiences with care quality were measured using the Consumer Assessment of Healthcare Providers and Systems-Patient Centered Medical Home survey , Patient Assessment of Chronic Illness Care tool , and Canadian Institute for Health Information common indicators (Additional file ). Both lay person-led and health professional-led system navigation models consistently improved patient experiences with quality of care. The community health worker-led goal setting plus IMPaCT intervention significantly improved care comprehensiveness and self-management supportiveness when compared to goal setting plus usual care in two RCTs (low-moderate risk of bias) . Compared to usual care, the nurse-led Guided Care and Community Connections Program also significantly improved overall patient experiences with the quality of their care (low-moderate risk of bias). Only one study evaluated the impact of team-based system navigation on patient experiences; the Health TAPESTRY program did not significantly improve patient experiences (i.e., level of difficulty accessing healthcare resources, care comprehensiveness, patient-centeredness, satisfaction) when compared to usual care (low risk of bias) . Caregiver outcomes Caregiver experience and health outcomes were reported in two studies that investigated health professional-led system navigation models . Overall, the findings were unclear. Compared to usual care, caregiver experiences (i.e., perception of patient care quality) improved after the nurse-led Guided Care intervention (moderate risk of bias) but not after the nurse-led Community Connections Program (low risk of bias) . Evidence from only one study demonstrated no impact of the nurse-led Guided Care intervention on caregiver strain and depression (moderate risk of bias) (Additional file ). Cost-related outcomes Only two studies reported on cost-related outcomes; both evaluated a lay person-led system navigation model . The cost of emergency department/hospital visits and emergency care per patient were compared to costs in a matched control group in one study (moderate risk of bias) and projected annual cost savings based on mathematical modelling in another (low risk of bias) . Although both studies reported differences between groups, no formal statistical tests were reported (Additional file ). The 13 studies that reported health service utilization evaluated lay person-led ( n = 6, 46%) [ , , , , , ], health professional-led ( n = 4, 31%) [ , , , ], and team-based ( n = 3, 23%) [ , , ] system navigation models. Health service utilization was primarily captured through administrative, health record, and/or health insurance data related to the number of primary care visits ( n = 10, 77%) [ , , , – , , , , ], hospital admissions and/or readmissions ( n = 9, 69%) [ – , , , , , ], emergency care visits ( n = 7, 54%) [ , , , , , , ], and home care visits ( n = 4, 31%) [ , , , ] (Additional file ). None of the included studies reported healthcare access or social service utilization outcomes. Overall, findings for lay person-led models were mixed. Three studies demonstrated improvements in health service utilization following lay person-led system navigation programs [ , , ]. Compared to baseline, patients at high risk for avoidable hospital admissions due to medical or psychosocial issues who accessed the lay person-led Integrated Care Coordination Service had a statistically significant decrease in emergency department attendance and hospital admissions nine months post-referral (low risk of bias) . Patients living in high-poverty areas who participated in the standardized, 6-month community health worker-led goal setting plus Individualized Management for Patient-Centered Targets (IMPaCT) program (tailored coaching, social support, navigation, advocacy) also had significantly lower odds of repeat admissions, but no difference in overall hospital admissions or length of stay when compared to goal setting plus usual care (low risk of bias) . Compared to usual care, community health worker-led system navigation including patient education, appointment scheduling, and assistance overcoming barriers to healthcare access significantly increased the rate of primary care provider and/or chronic disease nurse visits among patients with chronic health needs who were classified as unengaged with their medical care (i.e., had not seen a primary care physician in last 6 months) (moderate risk of bias) . Further, a higher percentage of these patients visited a primary care provider before seeking other providers for their health needs . However, three studies demonstrated no significant changes following lay person-led system navigation programs when compared to baseline or usual care (moderate risk of bias) [ , , ]. Similarly, the effectiveness of health professional-led system navigation on health service utilization outcomes was unclear. A social worker-led social prescribing program for patients with chronic conditions, polypharmacy, or frequent primary care attendance was associated with a significant decrease in the number of primary care physician visits, but no difference in home visits, telephone visits, or care contacts when compared to usual care in one study (moderate risk of bias) . No significant impacts on health service utilization were observed in three other studies following health professional-led system navigation programs when compared to usual care (low-moderate risk of bias) [ , , ]. In contrast, team-based system navigation models demonstrated some positive impacts on health service utilization across three studies with low risk of bias [ , , ]. In the 6-month Health TAPESTRY program, volunteer-led home visits followed by action planning with the healthcare team and links to community support resulted in a statistically significant increase in primary care visits and reduced rates of hospitalization among older adults, with no significant changes in emergency department visits when compared to usual care . Similarly, social worker and volunteer-led social prescribing to community services resulted in a significantly lower rate of annual general practitioner consultations with no significant impact on emergency department visits among adult patients experiencing social isolation with a history of frequent primary care visits, as compared to matched patients from a neighbouring area . However, it should be noted that this study lacked randomization, and patients assigned to the intervention group had a significantly higher rate of general practitioner consultations at baseline compared to their matched counterparts. Finally, a health coach and link worker-led intervention involving a needs assessment and referral to relevant community services also significantly decreased primary care use over a 3-month time period among patients managing at least one long-term health condition and experiencing social isolation when compared to baseline . In total, 16 studies captured patient-related outcomes [ – , , , – , – ]. These were grouped into four categories: 1) quality of life/health-related quality of life, mental health, and wellbeing, 2) social participation and function, 3) health behaviours, and 4) theoretical constructs related to behaviour change. Quality of life/health-related quality of life, mental health, and wellbeing In total, 13 studies investigated the impact of lay person-led ( n = 5, 39%) [ – , , ], health professional-led ( n = 3, 23%) [ , , ], team-based ( n = 4, 31%) [ , , , ], and self-navigation with lay support as needed ( n = 1, 8%) system navigation models on quality of life/health-related quality of life, mental health, and wellbeing outcomes. These outcomes were most often measured using the 12- or 36-Item Short Form Survey (SF-12, SF-36) ( n = 5, 39%) [ , , , , ], EuroQol-5 Dimension ( n = 5, 39%) [ , , , , ], Hospital Anxiety and Depression Scale ( n = 2, 15%) , or the Warwick-Edinburgh Mental Wellbeing Scale ( n = 2, 15%) . Various other single-item and self-report measures were used (Additional file ). Findings for lay person-led system navigation models were mixed. Social prescribing to local community health and wellbeing resources resulted in reduced anxiety and depression, better self-reported health, as well as a statistically and clinically significant improvement in patient wellbeing when compared to baseline in one study (moderate risk of bias) . However, another social prescribing program found a statistically significant, but not clinically significant difference in wellbeing among patients with multiple chronic conditions experiencing social isolation/loneliness when compared to baseline (moderate risk of bias) . Further, no significant changes in wellbeing, anxiety, depression, or health-related quality of life were found following the Community Links Practitioner intervention when compared to usual care (high risk of bias) . The standardized goal setting plus IMPaCT intervention significantly improved health-related quality of life in the mental domain, but not the physical domain of the SF-12 when compared to goal setting plus usual care in one study (moderate risk of bias) . However, no significant changes were observed in physical or mental health-related quality of life in another study evaluating the goal setting plus IMPaCT intervention when compared to usual care (low risk of bias) . Findings for health professional-led system navigation models were also mixed. The Urban Health Centres Europe approach including health assessment, shared decision making, and referral to appropriate health and social service care pathways (led by either a social worker, nurse, nurse practitioner, or physician assistant based on the setting) significantly improved health-related quality of life compared to usual care (low risk of bias) . However, two studies using nurse-led system navigation models did not result in significant improvements in health-related quality of life compared to usual care (low-moderate risk of bias) . None of the team-based or self-navigation with lay support system navigation models significantly improved quality of life/health-related quality of life, mental health, or wellbeing outcomes compared to baseline or usual care (low-moderate risk of bias) [ , , , , ]. Social participation and function Social participation and function was evaluated in eight studies including lay person-led ( n = 2, 25%) , health professional-led ( n = 2, 25%) , and team-based ( n = 4, 50%) [ , , , ] system navigation models. Various measures were used, including heterogeneous assessments of loneliness [ , , ], social networks , participation in social roles [ , , ], and social group memberships (Additional file ). Overall, the findings were mixed. Of the lay person-led models, social prescribing by wellbeing coordinators significantly increased social networks compared to baseline in one study (moderate risk of bias) . However, no changes in social participation were found following the Community Links Practitioner intervention compared to usual care in another study (high risk of bias) . Neither of the studies that used a health professional-led model found significant differences in social participation and function outcomes (low risk of bias) . Of the team-based models, the health coach and link worker-led intervention for adults managing long-term health conditions and experiencing social isolation, loneliness, or anxiety significantly improved the number of social group memberships from baseline, but did not impact community belonging or loneliness (low risk of bias) . Three additional studies evaluating team-based system navigation models found no significant differences in social participation and function outcomes (low-moderate risk of bias) [ , , ]. Health behaviours Health behaviours were assessed in seven studies evaluating lay person-led ( n = 4, 57%) [ – , ], health professional-led ( n = 1, 14%) , and team-based ( n = 2, 29%) system navigation models. Outcomes included heterogeneous measurements of physical activity/exercise [ , , , , ], cigarette smoking [ , , ], alcohol intake , and diet (Additional file ). Overall, the findings were mixed. Lay person-led social prescribing significantly increased physical activity compared to baseline in one study (moderate risk of bias) . However, three additional studies evaluating lay person-led models found no significant differences in health behaviour outcomes, including cigarette smoking or exercise level (low-moderate risk of bias) [ – ]. The study that evaluated a health professional-led model compared to usual care did not find significant differences in healthy lifestyle behaviours (low risk of bias) . Of the team-based system navigation models, an integrated health management intervention with referral to community programs led by community health centre staff and a multidisciplinary care team led to significant improvements in health behaviours including physical activity, alcohol intake, diet, and smoking habits when compared to bimonthly health education (high risk of bias) . However, another team-based model did not significantly impact physical activity levels compared to usual care (low risk of bias) . Patient activation, self-efficacy, and empowerment Patient activation, self-efficacy, and empowerment were evaluated in five studies including lay person-led ( n = 3, 60%) [ , , ], team-based ( n = 1, 20%) , and self-navigation with lay support as needed ( n = 1, 20%) system navigation models. Heterogeneous measurements of self-efficacy [ , , ], patient activation , and empowerment were used. Overall, the findings were mixed. Of the lay person-led models, the Cities for Live Program including linkage to community programs following an assessment of needs, barriers, and stage of change significantly improved self-efficacy compared to baseline (moderate risk of bias) . However, the standardized lay person-led goal setting plus IMPaCT intervention did not change patient activation in two studies (low-moderate risk of bias) . No significant changes in goal attainment, self-efficacy, or patient empowerment were observed following team-based system navigation in one study (low risk of bias) . Although limited to evidence from one study evaluating a self-navigation with lay support system navigation model, patients who participated in the “HealtheRx” intervention involving an electronic-medical record generated personalized list of local community resources with access to a community health information specialist as needed were more likely to report higher confidence in finding resources in their community to help manage their health compared to usual care (low risk of bias) . In total, 13 studies investigated the impact of lay person-led ( n = 5, 39%) [ – , , ], health professional-led ( n = 3, 23%) [ , , ], team-based ( n = 4, 31%) [ , , , ], and self-navigation with lay support as needed ( n = 1, 8%) system navigation models on quality of life/health-related quality of life, mental health, and wellbeing outcomes. These outcomes were most often measured using the 12- or 36-Item Short Form Survey (SF-12, SF-36) ( n = 5, 39%) [ , , , , ], EuroQol-5 Dimension ( n = 5, 39%) [ , , , , ], Hospital Anxiety and Depression Scale ( n = 2, 15%) , or the Warwick-Edinburgh Mental Wellbeing Scale ( n = 2, 15%) . Various other single-item and self-report measures were used (Additional file ). Findings for lay person-led system navigation models were mixed. Social prescribing to local community health and wellbeing resources resulted in reduced anxiety and depression, better self-reported health, as well as a statistically and clinically significant improvement in patient wellbeing when compared to baseline in one study (moderate risk of bias) . However, another social prescribing program found a statistically significant, but not clinically significant difference in wellbeing among patients with multiple chronic conditions experiencing social isolation/loneliness when compared to baseline (moderate risk of bias) . Further, no significant changes in wellbeing, anxiety, depression, or health-related quality of life were found following the Community Links Practitioner intervention when compared to usual care (high risk of bias) . The standardized goal setting plus IMPaCT intervention significantly improved health-related quality of life in the mental domain, but not the physical domain of the SF-12 when compared to goal setting plus usual care in one study (moderate risk of bias) . However, no significant changes were observed in physical or mental health-related quality of life in another study evaluating the goal setting plus IMPaCT intervention when compared to usual care (low risk of bias) . Findings for health professional-led system navigation models were also mixed. The Urban Health Centres Europe approach including health assessment, shared decision making, and referral to appropriate health and social service care pathways (led by either a social worker, nurse, nurse practitioner, or physician assistant based on the setting) significantly improved health-related quality of life compared to usual care (low risk of bias) . However, two studies using nurse-led system navigation models did not result in significant improvements in health-related quality of life compared to usual care (low-moderate risk of bias) . None of the team-based or self-navigation with lay support system navigation models significantly improved quality of life/health-related quality of life, mental health, or wellbeing outcomes compared to baseline or usual care (low-moderate risk of bias) [ , , , , ]. Social participation and function was evaluated in eight studies including lay person-led ( n = 2, 25%) , health professional-led ( n = 2, 25%) , and team-based ( n = 4, 50%) [ , , , ] system navigation models. Various measures were used, including heterogeneous assessments of loneliness [ , , ], social networks , participation in social roles [ , , ], and social group memberships (Additional file ). Overall, the findings were mixed. Of the lay person-led models, social prescribing by wellbeing coordinators significantly increased social networks compared to baseline in one study (moderate risk of bias) . However, no changes in social participation were found following the Community Links Practitioner intervention compared to usual care in another study (high risk of bias) . Neither of the studies that used a health professional-led model found significant differences in social participation and function outcomes (low risk of bias) . Of the team-based models, the health coach and link worker-led intervention for adults managing long-term health conditions and experiencing social isolation, loneliness, or anxiety significantly improved the number of social group memberships from baseline, but did not impact community belonging or loneliness (low risk of bias) . Three additional studies evaluating team-based system navigation models found no significant differences in social participation and function outcomes (low-moderate risk of bias) [ , , ]. Health behaviours were assessed in seven studies evaluating lay person-led ( n = 4, 57%) [ – , ], health professional-led ( n = 1, 14%) , and team-based ( n = 2, 29%) system navigation models. Outcomes included heterogeneous measurements of physical activity/exercise [ , , , , ], cigarette smoking [ , , ], alcohol intake , and diet (Additional file ). Overall, the findings were mixed. Lay person-led social prescribing significantly increased physical activity compared to baseline in one study (moderate risk of bias) . However, three additional studies evaluating lay person-led models found no significant differences in health behaviour outcomes, including cigarette smoking or exercise level (low-moderate risk of bias) [ – ]. The study that evaluated a health professional-led model compared to usual care did not find significant differences in healthy lifestyle behaviours (low risk of bias) . Of the team-based system navigation models, an integrated health management intervention with referral to community programs led by community health centre staff and a multidisciplinary care team led to significant improvements in health behaviours including physical activity, alcohol intake, diet, and smoking habits when compared to bimonthly health education (high risk of bias) . However, another team-based model did not significantly impact physical activity levels compared to usual care (low risk of bias) . Patient activation, self-efficacy, and empowerment were evaluated in five studies including lay person-led ( n = 3, 60%) [ , , ], team-based ( n = 1, 20%) , and self-navigation with lay support as needed ( n = 1, 20%) system navigation models. Heterogeneous measurements of self-efficacy [ , , ], patient activation , and empowerment were used. Overall, the findings were mixed. Of the lay person-led models, the Cities for Live Program including linkage to community programs following an assessment of needs, barriers, and stage of change significantly improved self-efficacy compared to baseline (moderate risk of bias) . However, the standardized lay person-led goal setting plus IMPaCT intervention did not change patient activation in two studies (low-moderate risk of bias) . No significant changes in goal attainment, self-efficacy, or patient empowerment were observed following team-based system navigation in one study (low risk of bias) . Although limited to evidence from one study evaluating a self-navigation with lay support system navigation model, patients who participated in the “HealtheRx” intervention involving an electronic-medical record generated personalized list of local community resources with access to a community health information specialist as needed were more likely to report higher confidence in finding resources in their community to help manage their health compared to usual care (low risk of bias) . Patient experience outcomes were reported in five studies, including lay person-led ( n = 2, 40%) , health professional-led ( n = 2, 40%) , and team-based ( n = 1, 20%) system navigation models. Patient experiences with care quality were measured using the Consumer Assessment of Healthcare Providers and Systems-Patient Centered Medical Home survey , Patient Assessment of Chronic Illness Care tool , and Canadian Institute for Health Information common indicators (Additional file ). Both lay person-led and health professional-led system navigation models consistently improved patient experiences with quality of care. The community health worker-led goal setting plus IMPaCT intervention significantly improved care comprehensiveness and self-management supportiveness when compared to goal setting plus usual care in two RCTs (low-moderate risk of bias) . Compared to usual care, the nurse-led Guided Care and Community Connections Program also significantly improved overall patient experiences with the quality of their care (low-moderate risk of bias). Only one study evaluated the impact of team-based system navigation on patient experiences; the Health TAPESTRY program did not significantly improve patient experiences (i.e., level of difficulty accessing healthcare resources, care comprehensiveness, patient-centeredness, satisfaction) when compared to usual care (low risk of bias) . Caregiver experience and health outcomes were reported in two studies that investigated health professional-led system navigation models . Overall, the findings were unclear. Compared to usual care, caregiver experiences (i.e., perception of patient care quality) improved after the nurse-led Guided Care intervention (moderate risk of bias) but not after the nurse-led Community Connections Program (low risk of bias) . Evidence from only one study demonstrated no impact of the nurse-led Guided Care intervention on caregiver strain and depression (moderate risk of bias) (Additional file ). Only two studies reported on cost-related outcomes; both evaluated a lay person-led system navigation model . The cost of emergency department/hospital visits and emergency care per patient were compared to costs in a matched control group in one study (moderate risk of bias) and projected annual cost savings based on mathematical modelling in another (low risk of bias) . Although both studies reported differences between groups, no formal statistical tests were reported (Additional file ). Building upon a previous scoping review, this systematic review synthesizes a growing body of evidence regarding the effectiveness of system navigation programs linking primary care with community-based health and social services. Whereas 1,248 records were screened in the original review, our search identified 15,226 new studies published since 2013, suggesting a substantial increase in interest in this field. Overall, there was variation in impacts across models of system navigation programs linking primary care with community-based health and social services on patient, caregiver, and health system outcomes. Evidence from three studies with low risk of bias [ , , ] suggests a team-based system navigation approach may result in slightly more appropriate health service utilization (e.g., increases in primary care use versus use of costlier health services) compared to baseline or usual care. These results may indicate a shift from reactive to more preventative care and self-management support, with health and social needs being better managed at the most appropriate level of care. Evidence from four studies [ , , , ] with moderate risk of bias suggests either lay person-led or health professional-led system navigation models may improve patient experiences with the quality of care when compared to usual care. This is consistent with patient descriptions of such programs as empowering, generally meeting their identified needs, and allowing patients to form positive relationships with their healthcare providers . It is unclear whether system navigation may improve patient-related outcomes (e.g., health-related quality of life, mental health and wellbeing, health behaviours). The evidence is very uncertain about the effect of system navigation programs on caregiver and cost-related outcomes as these were evaluated in a small number of studies. Although promising trends were observed, the potential impacts of lay person-led system navigation models on cost-related outcomes are unclear due to limited data, heterogeneous outcome measurements, and a lack of reporting concerning statistical significance. Our findings are consistent with those of another systematic review that demonstrated inconsistent effects of social prescribing programs in the United Kingdom on healthcare usage outcomes, generally consistent improvements in patient experiences, and limited evidence on costs . Also consistent with our findings, a recent mixed methods systematic review identified variable effectiveness of social prescribing services on health, wellbeing, health-related behaviours, self-confidence, social isolation/loneliness, and daily functioning . Although qualitative findings demonstrated that social prescribing service users generally experienced positive improvements in health/wellbeing and health behaviours, this was not consistently demonstrated by quantitative measures , in line with the patient-related findings in our review. Heterogeneous measurements across patient-related outcomes may explain some of the variation in findings within this category. Further, the presence of wide confidence intervals for many effect measures suggests that small sample sizes may have contributed to the lack of significant findings observed. While it is possible that quantitative measurements alone are insufficient to capture the holistic impact of system navigation, it is also conceivable that interventions focused primarily on linking patients to existing community-based health and social services may be insufficient to influence significant changes in patient-related health and health behaviour outcomes. For example, evidence from a recent systematic review demonstrates that chronic disease/case management and disease prevention initiatives led by registered nurses in primary care settings are effective for improving health outcomes and health-related behaviours such as weight loss, smoking cessation, diet and physical activity, self-efficacy, and social activity . Thus, while team-based system navigation may be effective for improving health service utilization by supporting patients to access the most appropriate services to meet their needs, the lack of clinical care provision within system navigation programs, when compared to primary care-based chronic disease and/or case management interventions , may limit the possible impact of system navigation alone on health-related outcomes. Several studies in this systematic review focused on populations who may face structural barriers to accessing care and found generally positive results. This included patients experiencing social isolation and/or chronic conditions with high use of primary care [ , , , ], individuals managing a chronic condition with previously limited engagement with their primary care team , patients with multiple chronic conditions living in high-poverty areas , and those deemed to be at high risk for avoidable and costly health services use due to medical or psychosocial conditions . These findings suggest that the greatest impacts of system navigation programs may be observed among populations who stand to benefit the most from improved connections to community-based health and social services. This hypothesis is supported by existing evidence that patients with chronic conditions, unmanaged behavioural health needs, and those experiencing health inequities (e.g., poverty, limited social support) tend to be the highest drivers of potentially avoidable and costly health services use . Further research is needed to identify which populations may benefit the most from system navigation. Several limitations should be considered when interpreting the results of this review. Although the individual studies within the review were appraised as having a generally low to moderate risk of bias, it is important to note that most were quasi-experimental, therefore lacking randomized controlled groups to facilitate strong comparisons. Further, most studies took place in the United States of America or the United Kingdom, which may limit generalizability to other health and social care contexts. Challenges with outcome measurements in the included studies also limited our conclusions. Although the primary outcomes of interest were access to care and health and social service utilization, none of the included studies objectively measured access to care or social service use outcomes, making it difficult to determine intervention effectiveness. For example, while changes in health services utilization were observed in several studies, we cannot definitively say that this was a direct result of increased connections to community-based social services because outcomes were typically only measured in the primary and/or acute care sectors. Another recent systematic review of social prescribing interventions identified similar limitations when analyzing the available evidence, suggesting that it is important to assess community-level changes (e.g., social service use, belonging, social support) and their associated impacts on health services use . Finally, given the generally small number of studies per outcome and high heterogeneity in results, our certainty regarding the effectiveness of system navigation programs on user and health system outcomes is low. The number of intervention studies has notably increased since the original scoping review, in which most studies were descriptive in nature. As more high-quality data becomes available regarding system navigation programs linking primary care with community-based health and social services, more robust and definitive conclusions may be observed. Implications for research Our synthesis of the effectiveness of system navigation programs, alongside existing synthesized evidence regarding social prescribing services , suggests that the potential impacts of these types of interventions may not be adequately captured through quantitative measurement tools alone. Although the decision to limit included studies to experimental and quasi-experimental designs was justified based on the objective of this systematic review to determine intervention effectiveness, future review authors may want to consider alternate research questions and types of evidence syntheses (e.g., integrative review, realist review) that would allow for the inclusion of both qualitative and quantitative data. This may also help determine the acceptability and feasibility of system navigation programs, given the generally high loss to follow up observed across studies (Table ) and the lack of reporting concerning intervention adherence and fidelity (Table ). Although we did not review qualitative data when studies used mixed methods, which may be a limitation, less than one quarter ( n = 5) [ , , , , ] of included studies conducted mixed methods evaluations. While only one study evaluated a self-navigation model by providing individuals with a personalized list of local services with lay support available , further research is warranted to evaluate similar novel approaches to system navigation. Researchers should ensure appropriate facilitation and support are available when designing self-navigation interventions, as this is known to be key for overcoming fluctuating health status concerns in persons managing chronic conditions or challenges with health literacy . Our review also highlights a need for more research related to the impact of system navigation programs on caregiver and cost-related outcomes. Although this review focused on patients’ and caregivers’ perspectives, it would be salient for future research to also consider the health professional perspective, given the rising levels of burnout and strain reported among primary care providers . Implications for practice Assisting patients and families to navigate and access programs and services is a current mandate for primary care providers . Integration of system navigation within primary care settings is proposed as a potential approach to alleviate some of the current and projected demands on the primary care sector . Providers should consider prioritizing individuals at greater risk for potentially avoidable and costly health services use when implementing system navigation programs. Findings from this review suggest that persons managing chronic conditions, experiencing social isolation, and/or living with health inequities (e.g., low income) may stand to benefit the most from navigation support, although further research is warranted. While this review included adults aged 18 + , the median age of 72 years across included studies also suggests that older adults are key targets for system navigation support, consistent with the complex, multimorbid health and social conditions older adults often face . Implications for policy Given the current orientation of health systems toward delivering integrated and coordinated health and community services , this systematic review is also highly relevant to policy makers. We identified system navigation models that may support outcomes relevant to the Quintuple Aim framework for healthcare improvement , which is top of mind for decision makers to advance health equity and improve patient and provider experiences, health system utilization, and cost-effectiveness. Our findings highlight the potential benefit of team-based system navigation as a strategy to improve use of primary healthcare services versus costlier healthcare (e.g., emergency department visits, hospitalizations) and enhance patient experiences with care. Our synthesis of the effectiveness of system navigation programs, alongside existing synthesized evidence regarding social prescribing services , suggests that the potential impacts of these types of interventions may not be adequately captured through quantitative measurement tools alone. Although the decision to limit included studies to experimental and quasi-experimental designs was justified based on the objective of this systematic review to determine intervention effectiveness, future review authors may want to consider alternate research questions and types of evidence syntheses (e.g., integrative review, realist review) that would allow for the inclusion of both qualitative and quantitative data. This may also help determine the acceptability and feasibility of system navigation programs, given the generally high loss to follow up observed across studies (Table ) and the lack of reporting concerning intervention adherence and fidelity (Table ). Although we did not review qualitative data when studies used mixed methods, which may be a limitation, less than one quarter ( n = 5) [ , , , , ] of included studies conducted mixed methods evaluations. While only one study evaluated a self-navigation model by providing individuals with a personalized list of local services with lay support available , further research is warranted to evaluate similar novel approaches to system navigation. Researchers should ensure appropriate facilitation and support are available when designing self-navigation interventions, as this is known to be key for overcoming fluctuating health status concerns in persons managing chronic conditions or challenges with health literacy . Our review also highlights a need for more research related to the impact of system navigation programs on caregiver and cost-related outcomes. Although this review focused on patients’ and caregivers’ perspectives, it would be salient for future research to also consider the health professional perspective, given the rising levels of burnout and strain reported among primary care providers . Assisting patients and families to navigate and access programs and services is a current mandate for primary care providers . Integration of system navigation within primary care settings is proposed as a potential approach to alleviate some of the current and projected demands on the primary care sector . Providers should consider prioritizing individuals at greater risk for potentially avoidable and costly health services use when implementing system navigation programs. Findings from this review suggest that persons managing chronic conditions, experiencing social isolation, and/or living with health inequities (e.g., low income) may stand to benefit the most from navigation support, although further research is warranted. While this review included adults aged 18 + , the median age of 72 years across included studies also suggests that older adults are key targets for system navigation support, consistent with the complex, multimorbid health and social conditions older adults often face . Given the current orientation of health systems toward delivering integrated and coordinated health and community services , this systematic review is also highly relevant to policy makers. We identified system navigation models that may support outcomes relevant to the Quintuple Aim framework for healthcare improvement , which is top of mind for decision makers to advance health equity and improve patient and provider experiences, health system utilization, and cost-effectiveness. Our findings highlight the potential benefit of team-based system navigation as a strategy to improve use of primary healthcare services versus costlier healthcare (e.g., emergency department visits, hospitalizations) and enhance patient experiences with care. System navigation programs linking primary care with community-based health and social services demonstrated mixed results. The ideal model of system navigation for improving patient, caregiver, and health system outcomes remains unclear. Nevertheless, a multidisciplinary team of healthcare providers and lay persons performing system navigation activities within primary care settings may result in slightly more appropriate health service utilization. Lay person-led or health professional-led system navigation may improve patient experiences with quality of care. Further research is warranted, specifically to understand the impact of system navigation on caregiver and cost-related outcomes, and to identify which populations may benefit the most from integrated health and social service care delivery programs. Additional file 1. Search Strategies. Additional file 2. List of Excluded Studies. Additional file 3. JBI Critical Appraisal Checklist for Randomized Controlled Trials. Additional file 4. JBI Critical Appraisal Checklist for Quasi-Experimental Studies. Additional file 5. Health Service Utilization Outcomes. Additional file 6. Patient-Related Outcomes. Additional file 7. Patient Experience Outcomes. Additional file 8. Caregiver Outcomes. Additional file 9. Cost-Related Outcomes.
Ocular manifestations in Iranian patients referred to rheumatology clinics from 2018 to 2020	a2e88158-870c-496c-9e5f-5b0a00b68ecc	10165950	Internal Medicine[mh]	INTRODUCTION Autoimmune diseases (ADs) affect approximately 7.6%–9.4% of the general population worldwide, based on recent studies. The prevalence of various rheumatologic diseases among Iranian patients are as follows: Rheumatoid arthritis (RA) 0.37%, Seronegative spondyloarthritis 0.24%, Ankylosing spondylitis (AS) 0.12%, Systemic lupus erythematosus (SLE) 0.06%, Behçet's disease (BD) 0.08%. ADs can show various signs and symptoms. The eyes are frequently involved in these diseases, and ocular manifestations can be the early presentation of rheumatologic diseases. Ocular complications can have inflammatory, vascular, infectious, or iatrogenic causes, with symptoms ranging from eye dryness to blindness. , The most common ocular involvement in AD patients consists of keratoconjunctivitis sicca, episcleritis, scleritis, uveitis, vitritis, retinal vasculitis, and panophthalmitis. Uveitis is one of the major causes of visual loss and blindness, and based on the primarily affected site; it is categorized into anterior, intermediate, posterior, and panuveitis. , Uveitides associated with juvenile idiopathic arthritis (JIA) are severe conditions with different visual outcomes; uveitis is also one of the most common eye complications in seronegative spondyloarthropathies. , , Acute anterior uveitis is the most common extra‐articular presentation of AS and occurs in almost 40% of patients; it can also be observed as an early sign before diagnosing AS. Ocular complications, including uveitis, can occur without rheumatologic implications too. Previous studies have shown that 30% to 60% of uveitis cases are idiopathic. The eyes are the most commonly affected organ in BD. The most common forms of involvement are retinal vasculitis and panuveitis; the most horrifying ocular complication of BD is double‐sided scarring panuveitis since it might rapidly lead to permanent blindness. , Other rheumatologic diseases, namely sarcoidosis, vasculitides, and other spondyloarthropathies, can also demonstrate some forms of eye involvement well before diagnosis of the background disease. To estimate the prevalence of different ophthalmological complications in patients diagnosed with rheumatologic diseases and suspected patients with ocular complications, we separated them into two groups (rheumatologic group and control group); we compared these two types of patients to have a better understanding of ocular complications in immune‐mediated rheumatological diseases patients; and suspected AD patients. MATERIALS AND METHODS 2.1 Study population The current study is a retrospective cross‐sectional study of patients referred to rheumatology clinics by an ophthalmologist in Kermanshah, Iran, from 2018 to 2020. We collected data retrospectively through the medical files; 106 patients were comprised, after an initial physical examination and requesting appropriate imaging and laboratory tests by a rheumatologist for each patient, seven different immune‐mediated rheumatological diseases based on the College of Rheumatology (ACR) classification criteria of rheumatologic diseases were diagnosed in 54 patients (Tables and ), and 52 patients did not satisfy any ACR classification criteria for immune‐mediated rheumatological diseases. We divided patients into two groups, patients who met the ACR classification criteria of rheumatologic diseases were designated as the first group (rheumatologic group), and patients with no diagnosed rheumatologic diseases according to the ACR classification criteria were considered as the second group (control group). 2.2 Data collection By using a dedicated checklist following information, including symptoms, different organ involvements, ocular disease diagnosed by an ophthalmologist, rheumatologic disease diagnosed by a rheumatologist, lab tests, and disease progression was recorded, then a database in Microsoft Excel Microsoft (Microsoft Corp.) was created. 2.3 Statistical analysis The database was then transformed to STATA 17 (StataCorp. 2021. Stata Statistical Software: Release 17: StataCorp LLC.) statistical software, and variables, namely: sex, age, type of rheumatologic disease, type of ocular involvement, ophthalmic symptoms, the affected eye, comorbidity were analyzed. The results were collected as means, standard deviation, the percentage for continuous variables, and frequency distribution for categorical variables. To prove an association between variables, we used the Chi‐square independence test was used, and then we drew charts and graphs. 2.4 Ethics approval This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Kermanshah University of Medical Sciences (IR.KUMS.REC.1400.430). Study population The current study is a retrospective cross‐sectional study of patients referred to rheumatology clinics by an ophthalmologist in Kermanshah, Iran, from 2018 to 2020. We collected data retrospectively through the medical files; 106 patients were comprised, after an initial physical examination and requesting appropriate imaging and laboratory tests by a rheumatologist for each patient, seven different immune‐mediated rheumatological diseases based on the College of Rheumatology (ACR) classification criteria of rheumatologic diseases were diagnosed in 54 patients (Tables and ), and 52 patients did not satisfy any ACR classification criteria for immune‐mediated rheumatological diseases. We divided patients into two groups, patients who met the ACR classification criteria of rheumatologic diseases were designated as the first group (rheumatologic group), and patients with no diagnosed rheumatologic diseases according to the ACR classification criteria were considered as the second group (control group). Data collection By using a dedicated checklist following information, including symptoms, different organ involvements, ocular disease diagnosed by an ophthalmologist, rheumatologic disease diagnosed by a rheumatologist, lab tests, and disease progression was recorded, then a database in Microsoft Excel Microsoft (Microsoft Corp.) was created. Statistical analysis The database was then transformed to STATA 17 (StataCorp. 2021. Stata Statistical Software: Release 17: StataCorp LLC.) statistical software, and variables, namely: sex, age, type of rheumatologic disease, type of ocular involvement, ophthalmic symptoms, the affected eye, comorbidity were analyzed. The results were collected as means, standard deviation, the percentage for continuous variables, and frequency distribution for categorical variables. To prove an association between variables, we used the Chi‐square independence test was used, and then we drew charts and graphs. Ethics approval This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Kermanshah University of Medical Sciences (IR.KUMS.REC.1400.430). RESULTS One hundred and six patients were enlisted in the study, 71 of whom were females, and 35 were males. The minimum age was two, and the maximum was 83 years, with a mean age of 40.43 ± 13.26 years. Table shows age and sex distribution. Ocular symptoms affecting both eyes simultaneously were more common (43.40%), and one‐sided involvement of the left or right eye was almost the same (left eye 30.19% and right eye 26.42%). Anterior uveitis was the most prevalent ophthalmic disease in both groups, and BD was the most frequently diagnosed rheumatologic disease. Approximately half of the patients suffered from blurred vision; ophthalmic disease diagnoses and ophthalmic symptoms in the rheumatologic and control groups are listed in Tables and , respectively. Comorbidities were assessed as well, and 20 comorbidities were found among 16 patients (four patients had two comorbidities), and 90 patients had no significant comorbidity. Hypertension and hypothyroidism were the most common ones, with seven cases each, followed by three diabetes mellitus type 2 (DM2), two chronic kidney disease (CKD), and one hepatitis (silent hepatitis B). While we assessed skin and mucous involvement, we found that five patients had two separate skin and mucous involvements, and two patients had three different involvements; 76 patients did not report any skin and mucous involvement; aphthous ulcer was the most common involvement, followed by nonsexually acquired genital ulcers (NSGU) and pseudo‐folliculitis barbae (PB). One hundred (94.33%) patients had no pulmonary complications, and only six (5.66%) had lung involvement. While approximately 66% of the patients had no joint involvement, among those with joint involvement, inflammatory back pain was the most common complication affecting 23 (21.7%) patients, followed by arthralgia affecting 12 (11.3%) patients and arthritis affecting 5 (4.7%) patients, two patients suffered from inflammatory back pain and arthritis simultaneously, and one patient was identified having inflammatory back pain, arthritis, and arthralgia at the same time. Patients with regard to their ophthalmic symptoms were also followed up, and three different outcomes were classified: completely controlled, partially controlled, and not controlled; if the ophthalmic symptoms disappeared and the physical examination showed no abnormality, the patient's ophthalmic symptoms were considered being completely controlled; if the symptoms were relieved but the physical examination was abnormal, the patient's ophthalmic symptoms were considered being partially controlled, and if neither the symptoms were relieved nor the physical examination was normal, the patient's ophthalmic symptoms were considered not being controlled. Symptoms of 35 (33%) patients were completely controlled, 9 (8.5%) were not controlled at all, and 62 (58.5%) were controlled partially. The findings mentioned earlier are shown visually in Figure . Patients in the rheumatologic disease group had more positive or elevated tests; the number of positive or elevated results for each test in this group were as follows: HLAB5 = 15, HLAB27 = 15, HLAB51 = 6, ACE = 5, ANCA = 2, ANA = 1, PPD = 1, ANTIdsDANA = 1, in addition, the number of positive or elevated results for each test in the control group were as follows: HLAB5 = 5, ANA = 4, ACE = 4, RF = 2, PPD = 2, HLAB27 = 1, AntiRO = 1, ANTIdsDNA = 1; Figure visualizes laboratory test results. Ophthalmic diseases were statistically associated with rheumatologic background ( p ‐value = .036), indicating that patients with rheumatic diseases more often had ophthalmic diseases than the control group; pulmonary ( p ‐value = .013), skin and mucous ( p ‐value < .001), and joint involvement ( p ‐value < .001) were also statistically different between two groups, suggesting patients with rheumatologic diseases (rheumatologic group) more frequently had pulmonary, skin, mucous, and joint involvement. Chi‐squared test results are listed in Table . DISCUSSION Based on the best of our knowledge, we disclose the first study, which presents ocular manifestations in patients with immune‐mediated rheumatologic diseases, and suspected AD patients in the Middle East and North Africa region. In our study, we included 106 patients with ophthalmic complications who were referred to rheumatologic clinics for further investigation. Rheumatologic diseases have broad extra‐articular manifestations; skin, bones, eyes, mouth, and lungs can be affected frequently. On the one hand, a sophisticated ocular immune system and the blood‐aqueous barrier protect the eyes; not having a lymphatic outflow adds up to this protection. Still, on the other hand, they are commonly affected in rheumatologic diseases; not only early diagnosis of ocular involvement in rheumatologic diseases can prevent severe ophthalmic complications like vision loss and blindness, but also it can be a sensitive indicator for the severity of such diseases, having said that, familiarity with ophthalmic complications in rheumatologic diseases is essential for health care workers and especially for rheumatologists. Based on other studies, autoimmune diseases are far more common among females; similarly, in our study, most of the patients were also females (67%), and gender was not associated with immune‐mediated rheumatologic diseases in our study ( p ‐value = .085); this event might be because of our small sample size. , , We found that the mean age was 40.43 ± 13.26 years in our participants; Adelowo et al. and Pathanapitoon et al. reported the age of the participants 44.3 ± 13.7 and 48.9 ± 19.3 years, respectively, which were similar to our study; still, Uribe‐Reina et al. have reported the mean age 54.6 ± 15.6 years, which was higher than our results. , , Although approximately half of our participants were older than 40, only 16 patients reported comorbidities, and the comorbidities were not statistically associated with rheumatologic disease history ( p ‐value = .198). Involvement of a singular eye was more common (56.6%) than involvement of both eyes (43.40%), but the involvement of both eyes was more common than the isolated left or right eye involvement. The distribution of the affected eye was not significantly different in the control and rheumatologic group ( p ‐value = .185). Blurred vision and eye redness were the most prevalent, and metamorphopsia and photophobia were the least frequent ophthalmic symptoms among patients. Ophthalmic symptoms were not associated with having rheumatologic diseases ( p ‐value = .592). In the Uribe‐Reina et al., study, 35% of patients had at least one ocular symptom; the most frequent ones were dry eyes (30.8%), and ocular pain (2.7%), which is in contrast with our findings, yet the overall percentage of ocular pain in their study is similar to ours (1.8%); similarly, Ausayakhun et al. reported dry eyes (19.9%) as their most frequent ophthalmic symptom. Isolated anterior uveitis was the most observed ophthalmic disease, with 35.8% of total cases, which adds up to 41.2% of total cases if we consider anterior uveitis with or without other ophthalmic complications. Following anterior uveitis, posterior uveitis was the second utmost ophthalmic complication with an incidence of 20.7%, and panuveitis (17.9%) was the third most observed ophthalmic disease. In a study by Jiménez‐Balderas et al. on 57 patients diagnosed with uveitis, anterior uveitis was the most frequent type of uveitis, and vision loss was associated with uveitis recurrence. In a study conducted by Cakan et al., the most frequent type of uveitis was anterior uveitis, followed by bilateral intermediate uveitis and bilateral pan uveitis. A significant association exists between having a rheumatologic disease and having ophthalmic diseases ( p ‐value = .036). We diagnosed immune‐mediated rheumatologic diseases by order of frequency as follows: Behçet's Disease (21%), ankylosing spondylitis (17.9%), sarcoidosis (4.7%), other spondylopathies (2.8%), Vogt–Koyanagi–Harada disease (1.9%), juvenile rheumatoid arthritis, and isolated retinal vasculitis (0.9%). Other studies have reported different rheumatologic diseases' occurrence frequencies; in Tseng et al. study, they reported the frequency of rheumatologic diseases occurrence from more to less as follows: ankylosing spondylitis, Behçet's Disease, sarcoidosis, psoriasis, and juvenile rheumatoid arthritis; Uribe‐Reina et al. reported prevalence of rheumatologic diseases in their study population as follows: rheumatoid arthritis (33.3%), fibromyalgia (22.7%), Sjögren's syndrome (19.7%), and lupus (9.9%); and in Jiménez‐Balderas et al. study, ankylosing spondylitis was the most common rheumatologic disease. We found that lung involvement was associated with rheumatologic background ( p ‐value = .013), and five patients out of six having lung involvement were diagnosed with sarcoidosis. Patients with rheumatologic diseases commonly had more positive and elevated laboratory tests, and laboratory tests were associated with rheumatologic background ( p ‐value = .004). Results on the outcome of ophthalmic symptoms were not associated with rheumatologic disease history, meaning patients with rheumatologic diseases necessarily did not have the worse ophthalmic symptoms outcome ( p ‐value = .920). Investigating the frequency and outcome of ocular diseases among patients with autoimmune diseases is barely studied; to find more concerning ophthalmic symptoms with regard to diagnosing rheumatological diseases and to determine the most common ophthalmic symptoms, diseases, and other characteristics, including age, sex, the affected eye, comorbidities, laboratory tests, and other tissues involvement among patients with autoimmune diseases and suspected patients, this study was conducted. To better understand the key characteristics of patients with ophthalmic complications who develop rheumatologic diseases, we recommend further studies with larger sample sizes be conducted. CONCLUSION Since eye complications can be the first indicator of an autoimmune disease, understanding the prevalence of ocular symptoms and diseases among patients with autoimmune diseases is essential for proper referral to rheumatology clinics. Alternatively, early diagnosis of autoimmune diseases can prevent disastrous ophthalmic complications and improve patients' quality of life. LIMITATION The access to patients was limited due to the low prevalence of the disease, and the coronavirus disease restrictions further bounded our access to patients. To address this issue, we used medical records in rheumatology clinics. Meharan Pournazari : Conceptualization; data curation; methodology; project administration; resources; supervision. Tara Hashemi : Conceptualization; data curation; resources; validation. Mahsa Zarpoosh : writing—original draft; editing; Conceptualization; resources. Parsa Amirian : Formal analysis; project administration; visualization; writing—original draft; writing—review & editing. The authors declare no conflicts of interest. Supporting information. Click here for additional data file.
Geriatric Radiation Oncology: What We Know and What Can We Do Better?	24c001a3-68e9-4231-ba2c-96b3650ea539	10166100	Internal Medicine[mh]	Due to demographic changes, radiation oncologists are facing an increasing number of elderly patients in their daily practice at least in developed countries. There is no clear preclinical, radiobiological or clinical evidence for a generally reduced radiation tolerance with increasing age. Nevertheless, an emerging body of evidence shows that older patients are still less likely to receive radiation therapy (RT) in similar clinical scenarios compared to younger patients, especially with regard to adjuvant treatment settings. A variety of factors may play a role in the clinical decision-making process: Elderly patients are clearly underrepresented or even excluded from most randomized trials creating the evidence for certain treatment decisions. Thus, for many physicians, it seems at least questionable if results commonly used for patient guidance can be simply transferred to the elderly population. Elderly patients are also more likely to have either distinct serious comorbidities or at least an accumulation of several minor limitations of organ functions. This may lead to a (anticipated) decreased treatment tolerance. Moreover, socioeconomic factors such as limited mobility or inadequate resources for general care and/or management of common side effects at home may further impair their (anticipated) ability for outpatient RT treatments. This seems especially true if long travel distances to the next RT facility are present. Different views on treatment aims focusing on the preservation on quality of life (Qol) and personal independence rather than pure survival may further prompt elderly patients more likely to refuse (anticipated) intense treatment approaches or long-lasting inpatient therapies. From a scientific point of view, addressing the question of an optimal RT approach in elderly patients is even more difficult. First of all, there is no generally agreed definition of the term “elderly”, which seems to be even disease- and/or treatment specific. While most physicians would agree to designate a 70-year-old person with glioblastoma as “elderly”, this would probably not been the case for prostate cancer. Moreover, there is no clear demarcation to patients with comorbidities and/or limited performance status. Many studies reporting outcomes of “elderly” patients include also “frail” patients without reporting subgroup analyses. Severe comorbidities of organs within the radiation volumes obviously can limit RT tolerance, but this does not represent a specific difference compared to younger patients. While limitations in organ functions outside the treated area usually do not compromise the ability to tolerate RT, this might be true for combination approaches. Radiation techniques as well as supportive care have constantly evolved over time, resulting in less severe toxicities and/or improved capabilities to reduce side effects. The widespread adoption of intensity-modulated or stereotactic RT constantly reduced doses to adjacent organs at risk. Combined with the use of daily image-guidance, those advances resulted in improved tolerability of RT. Therefore, reports with outdated radiation techniques often describing increased toxicity in elderly patients should not be used anymore to draw conclusions on treatment tolerability. In contrast, non-randomized analyses reporting equal toxicity inherit always some risk of selection bias. Randomized comparisons specifically addressing elderly patients are desirable, however they likely face difficulties regarding the definitions of inclusion criteria: Because elderly patients show a greater variety in performance status and comorbidities, wide inclusion criteria would result in inhomogeneous cohorts. In contrast, strict inclusion criteria and/or stratification rules would compromise recruitment. However, evolvement of radiation techniques offers also a chance to improve the adherence to RT in elderly and/or frail patients. Hypofractionated regimes using less fractions with higher single doses resulting in shorter overall treatment time have been widely adopted. This will likely reduce the impact of socioeconomic factors on treatment decisions on both sides (physicians and patients). Nevertheless, more research is warranted, specifically addressing the value and tolerability of RT and/or combination approaches in elderly patients. This should include investigations about the optimal study design for the evaluation of more personalized treatment approaches. While clear evidence and consequently specific recommendations or guidelines are still rare, our review aims at summarizing the current evidence for RT in elderly patients to support radiation oncologists and other treating physicians in their decision-making process. We hereby focused on the most relevant disease types presented in daily practice of radiation oncology. Approximately 310.000 new cases of central nervous system (CNS) tumours are diagnosed globally and 67.114 new cases in Europe each year corresponding to a crude incidence rate of 4.0/100.000 and 9.0/100.000. Almost half of them are diagnosed in patients aged 65 and older. Around 50–70% of those tumours are diagnosed as glioblastoma with incidence rates peaking at ages 60–80 years. Despite this fact, elderly patients are an underrepresented age group in clinical trials. Outcome in patients diagnosed with glioblastoma is bleak: 5-year overall survival (OS) still ranges between 3% (patients >65 years) to 27% (patients between 20 and 39 years). Combined chemoradiation with temozolomide following resection has been established as treatment standard in 2005. Stupp et al could show impressive OS improvement by combining simultaneous radiochemotherapy with temozolomide plus adjuvant temozolomide. They observed an increase median survival from 12.1 months (radiotherapy only) to 14.6 months (chemoradiation). In their subgroup analysis of patients ≥60 years, chemoradiation still improved OS (11.8 months vs 10.9 months), but the observed effect was not as pronounced as in younger patients. The benefit also seemed to decrease with increasing RPA class corresponding to decreasing overall performance status. Patients with MGMT promotor methylation were shown to benefit from combined treatment significantly more than patients without methylation. However, age itself was not a significant prognostic factor in this post-hoc analysis. Treatment outcomes can be further improved by tumour-treating fields. Their addition after completion of a regular Stupp regimen resulted in improved OS without detrimental impact on Qol. , Intensification of chemotherapy by the addition of CCNU to temozolomide in patients with MGMT promotor methylation in the CeTeG trial also increased median OS (48.1 vs 31.8 months) in patients ≤65 years. However, fewer elderly patients seem to exhibit favourable prognostic markers such as MGMT promotor methylation. Iwamoto et al analyzed patterns of care in 4137 glioblastoma patients aged >65 years and could show that age and comorbidities significantly influenced treatment choice (resection, chemotherapy, radiotherapy) in this cohort. Defining adequate treatment paths in elderly patients is therefore an important issue. The considerable time burden of a standard course of radiotherapy may play an important role specifically in the elderly. Nevertheless, the French group showed that patients aged 70-85 years benefited significantly from 50.4Gy in 28 fractions vs best supportive care (BSC) in terms of OS and symptom control. In order to shorten overall treatment time, Roa et al established hypofractionated radiotherapy with 15×2.67Gy in elderly patients. Their regimen was equally effective compared to standard radiotherapy with 30x2Gy in a cohort with a median age of 71–72 years. No difference in either OS or performance score was noted. However, median survival was only 5.1 months (60Gy) and 5.6 months (40.05Gy). The EORTC-trial therefore investigated the use of combining hypofractionated radiotherapy with temozolomide in a trial randomizing 562 patients with a median age of 73 years. Hypofractionated chemoradiation improved OS from 7.6 months (RT alone) to 9.3 months (chemoradiation). Patients with MGMT-promotor methylation again benefited more from the addition of chemotherapy (13.5 months vs 7.7 months) than patients with unmethylated MGMT-promotor (10.0 months vs 7.9 months). Qol was not affected by treatment arm. With the aim to further reduce in-hospital treatment time, Wick et al investigated temozolomide only as an alternative to standard radiotherapy in patients with a median age of 71/72 years within the NOA-08 trial. Overall and event-free survival were higher in the radiotherapy group. In their subgroup analysis, MGMT-promotor methylated patients receiving temozolomide showed superior overall and event-free survival than MGMTpromotor methylated patients receiving radiotherapy. Hence, temozolomide only may be an option for this subset of patients. The NORDIC-trial was designed to clarify this question further: Malmstrom et al randomized patients aged 60–83 years to standard radiotherapy (60Gy), 10×3.4Gy or temozolomide. Protocol completion in the 60Gy and temozolomide groups were only 72% and 34% vs 95% in the hypofractionated group. OS in the temozolomide group was 8.3 months vs 7.5 months (hypofractionated) vs 6.0 months (standard RT). As described above, patient performance score, comorbidities and age often influence treatment choice in clinical routine. Mirimanoff et al could show the influence of those aspects through analysis of the Stupp trial according to RPA classes: patients with lower RPA classes showed significantly improved outcomes. Gerstein et al and Combs et al retrospectively showed significant differences in outcome based on overall performance state or RPA class in elderly glioblastoma patients. Unfortunately, comorbidities are not reported in either of those trials specifically including elderly patients. Median age as well as minimum age also varies substantially between the trials. While two trials report Karnofsky performance and mini-mental state exam scores, , Malmstrom et al does not include this information. Neither of those trials include any form of specific geriatric or frailty assessment, hence interpretation of those trials has some limits regarding those aspects. As opposed to many other malignancies though, treatment of glioblastoma in elderly patients can be guided by prospective evidence albeit more detailed information is still somewhat scarce. Elderly patients well enough to undergo treatment should at least receive hypofractionated radiotherapy. Elderly patients with MGMT-promotor methylation and good performance status may receive either combined chemoradiation or temozolomide alone, whereas patients with poor performance status may receive hypofractionated radiotherapy, temozolomide or BSC. Squamous cell carcinoma represents the most common malignancy in the head and neck (HNSCC). Approximately 745.000 new cases were diagnosed in 2020 worldwide corresponding to a crude incidence rate of 9.6/100.000. Combined chemoradiation has been well established as a potentially curative treatment approach resulting in improved organ preservation, locoregional control and OS of around 60-70% in locally advanced HNSCC. A recent SEER database analysis could demonstrate that modern regimens have improved survival for HNSCC patients over the past decades for all age groups except for patients aged ≥75 years. Current cancer statistics in the UK show peak incidence rates of HNSCC in males at 70–74 years and in females at 85–89 years. Around 33% (male) to 36% (female) of all new HNSCC patients are ≥70 years, so a significant number of patients appear not to share the benefit from recent advances. Elderly patients (≥70–75 years) are often excluded from participation in clinical trials which represents a relevant knowledge gap in the treatment of patients in this age group. We will therefore address the following issues in the treatment of elderly head and neck patients based on available – albeit mostly retrospective – data: Do elderly HNSCC patients routinely receive standard treatment? Is standard treatment feasible for elderly patients? Which factors adversely affect prognosis in elderly patients? Are there any objective criteria to guide treatment selection in elderly patients? Do Elderly HNSCC Patients Routinely Receive Standard Treatment and is It Feasible for Elderly Patients? A recent pattern-of-care survey in Germany, Austria and Switzerland reported that the majority of respondents apparently treat elderly patients within standard regimens. Several groups also report outcomes of elderly patients undergoing chemoradiation. Maggiore et al share their experience in 89 patients ≥70 years treated with a 5-FU/hydroxyurea-based regimen between 1997 and 2009. At a median follow-up of approx. 40 months, OS was 32%. Most patients (86.5%) completed their treatment; toxicity, however, was considerable: grade 3–4 neutropenia occurred in 34%, 44% of patients experienced aspiration, 62% required feeding tubes. The treatment was still deemed feasible in selected patients. Treatment toxicity in 246 elderly patients (65 years) mainly treated by chemoradiation was found to be high also in the Freiburg analyses: incidence of at least one grade 3/4 toxicity was 56%. While treatment adherence was high (87%), OS only reached 57% at 2 years. Brown et al reported the outcome of patients ≥80 years treated with intensity-modulated techniques from 2003 to 2015. Only 7% of those patients received chemoradiation. Twenty-six percent required in-patient treatment during radiotherapy. However, overall grade 3/4 late toxicity at a median follow-up of 25 months was only 2%. The authors concluded that treatment tolerance was high even in this age group. Considering low overall patient numbers and also low rates of patients treated with concomitant chemoradiation, it is doubtful whether elderly patients routinely receive standard treatment. As reported, standard treatment may be feasible for some patients, though selection is still largely subjective. Potentially, an evaluation of larger databases and registries regarding treatment specifics – which to our knowledge is currently lacking – may help to answer these questions. Which Factors Adversely Affect Prognosis and are There Any Objective Criteria to Guide Treatment Selection in Elderly Patients? As mentioned, chemoradiation is associated with considerable treatment-related toxicity. In cumulative analysis of RTOG studies, approximately 43% of patients developed late toxicities grade 3+. Patient age was one of the most important predictors for development of severe side effects. Age was also found to predict for higher rates of treatment modifications, interruptions or discontinuation in two retrospective cohorts of 272 and 40 elderly patients receiving chemoradiation. , Moreover, age adversely affected outcomes in elderly patients receiving chemoradiation according to a SEER analysis. Comorbidity has also been shown to adversely impact survival , and toxicity, , while prevalence of comorbidity in HNSCC patients increases with age. , Paleri et al analyzed 180 patients with larynx carcinomas: 64% of patients had comorbidities, 26% more than one comorbid condition. OS in patients without comorbidities was significantly higher than in patients with one or more comorbidities. Patients with independent neurological disorders as part of their comorbidities showed the highest mortality rate of 70% in this analysis. No reliable statement can be made regarding the frequency of treatment modifications in elderly patients (see above). However, Derks et al could show in their analyses that both age and comorbidities independently influenced treatment choice in HNSCC patients. Frailty is a relatively new term summarizing age-related physiological decline and functioning leading to increased vulnerability. De Vries and collaborators analyzed 160 patients undergoing radiotherapy for head and neck cancer from 2014 to 2016. Type of treatment and neither frailty or geriatric assessment were predictive of treatment toxicity in their cohort. In contrast, Morse et al found that patients showing sarcopenia as a surrogate for frailty were more prone to developing treatment toxicities during chemoradiation. This finding is supported by Chou et al and Karavolia et al, who showed associations between increased vulnerability and OS and toxicity. The most concordant parameter to predict for adverse outcome in the elderly seems to be overall performance state. Despite several attempts to identify other predictors, performance state was found to be an independent predictor for OS in several analyses. , However, Nicolay et al recently proposed a novel prognostic score. Based on their initial cohort of 284 patients, they found performance state, comorbidity index, CRP level and tumour and nodal stage as independent prognostic factors. Their nomogram based on CRP, Charlson comorbidity index and Karnofsky performance state could predict for outcomes in their validation cohort (217 elderly patients) reasonably well and may be used to select treatment in the elderly population. Further prospective data are, however, pending. In summary, data regarding standard treatment in elderly HNSCC patients are scarce. In addition, consensus regarding absolute age-groups termed “elderly” seems to be lacking. Retrospective analyses frequently report on comparatively low patient numbers in view of the considerable incidence of HNSCC in the elderly, hence one may safely assume that treatment is frequently modified and adjusted. Several factors, including age, comorbidities, frailty and performance score impact treatment as well as treatment outcome in HNSCC. Prospective data to guide objective scores and hence treatment recommendations are still lacking. A recent pattern-of-care survey in Germany, Austria and Switzerland reported that the majority of respondents apparently treat elderly patients within standard regimens. Several groups also report outcomes of elderly patients undergoing chemoradiation. Maggiore et al share their experience in 89 patients ≥70 years treated with a 5-FU/hydroxyurea-based regimen between 1997 and 2009. At a median follow-up of approx. 40 months, OS was 32%. Most patients (86.5%) completed their treatment; toxicity, however, was considerable: grade 3–4 neutropenia occurred in 34%, 44% of patients experienced aspiration, 62% required feeding tubes. The treatment was still deemed feasible in selected patients. Treatment toxicity in 246 elderly patients (65 years) mainly treated by chemoradiation was found to be high also in the Freiburg analyses: incidence of at least one grade 3/4 toxicity was 56%. While treatment adherence was high (87%), OS only reached 57% at 2 years. Brown et al reported the outcome of patients ≥80 years treated with intensity-modulated techniques from 2003 to 2015. Only 7% of those patients received chemoradiation. Twenty-six percent required in-patient treatment during radiotherapy. However, overall grade 3/4 late toxicity at a median follow-up of 25 months was only 2%. The authors concluded that treatment tolerance was high even in this age group. Considering low overall patient numbers and also low rates of patients treated with concomitant chemoradiation, it is doubtful whether elderly patients routinely receive standard treatment. As reported, standard treatment may be feasible for some patients, though selection is still largely subjective. Potentially, an evaluation of larger databases and registries regarding treatment specifics – which to our knowledge is currently lacking – may help to answer these questions. As mentioned, chemoradiation is associated with considerable treatment-related toxicity. In cumulative analysis of RTOG studies, approximately 43% of patients developed late toxicities grade 3+. Patient age was one of the most important predictors for development of severe side effects. Age was also found to predict for higher rates of treatment modifications, interruptions or discontinuation in two retrospective cohorts of 272 and 40 elderly patients receiving chemoradiation. , Moreover, age adversely affected outcomes in elderly patients receiving chemoradiation according to a SEER analysis. Comorbidity has also been shown to adversely impact survival , and toxicity, , while prevalence of comorbidity in HNSCC patients increases with age. , Paleri et al analyzed 180 patients with larynx carcinomas: 64% of patients had comorbidities, 26% more than one comorbid condition. OS in patients without comorbidities was significantly higher than in patients with one or more comorbidities. Patients with independent neurological disorders as part of their comorbidities showed the highest mortality rate of 70% in this analysis. No reliable statement can be made regarding the frequency of treatment modifications in elderly patients (see above). However, Derks et al could show in their analyses that both age and comorbidities independently influenced treatment choice in HNSCC patients. Frailty is a relatively new term summarizing age-related physiological decline and functioning leading to increased vulnerability. De Vries and collaborators analyzed 160 patients undergoing radiotherapy for head and neck cancer from 2014 to 2016. Type of treatment and neither frailty or geriatric assessment were predictive of treatment toxicity in their cohort. In contrast, Morse et al found that patients showing sarcopenia as a surrogate for frailty were more prone to developing treatment toxicities during chemoradiation. This finding is supported by Chou et al and Karavolia et al, who showed associations between increased vulnerability and OS and toxicity. The most concordant parameter to predict for adverse outcome in the elderly seems to be overall performance state. Despite several attempts to identify other predictors, performance state was found to be an independent predictor for OS in several analyses. , However, Nicolay et al recently proposed a novel prognostic score. Based on their initial cohort of 284 patients, they found performance state, comorbidity index, CRP level and tumour and nodal stage as independent prognostic factors. Their nomogram based on CRP, Charlson comorbidity index and Karnofsky performance state could predict for outcomes in their validation cohort (217 elderly patients) reasonably well and may be used to select treatment in the elderly population. Further prospective data are, however, pending. In summary, data regarding standard treatment in elderly HNSCC patients are scarce. In addition, consensus regarding absolute age-groups termed “elderly” seems to be lacking. Retrospective analyses frequently report on comparatively low patient numbers in view of the considerable incidence of HNSCC in the elderly, hence one may safely assume that treatment is frequently modified and adjusted. Several factors, including age, comorbidities, frailty and performance score impact treatment as well as treatment outcome in HNSCC. Prospective data to guide objective scores and hence treatment recommendations are still lacking. RT is an important component of lung cancer therapy alone or in combination with systemic treatments. With a median age of about 70 years at diagnosis, lung cancer is clearly a disease of the elderly, often associated with comorbidities and limited performance status. Many patients cannot receive standard treatment and were excluded from randomized trials, leading to limited evidence in the treatment of this patient group. Therefore, the decision-making process should aim at maximizing benefits and minimizing harms depending on patients' individual values and preferences. Lung cancer can be generally divided into small-cell (SCLC) and non-small-cell lung cancer (NSCLC). In SCLC, systemic therapy is the dominant option for the vast majority of patients, while the role of RT is restricted to early stages or palliative treatments. Within these situations, the used RT regimens (and their general limitations) for SCLC are similar to those for locally advanced or metastatic NSCLC. Therefore, it will not be specifically addressed in the following: NSCLC Stage I/II Standard of care in patients with Stage I/II is resection, alternatively stereotactic ablative body radiotherapy (SABR) is well established. A pooled analysis of the prospective randomized STARS and ROSEL trials showed even improved survival with SABR compared to surgery for resectable patients, although not focusing on elderly patients. In 2021, the updated propensity score matched results of the STARS trial confirmed that SABR is a comparable curative treatment with minimal toxicity, noninvasive character and possibility for an outpatient setting with a short treatment period (generally 1 week), which is often important for frail or elderly patients. Brooks et al retrospectively analysed 772 patients in Stage I/II NSCLC (n = 442 < 75 years, n = 330 >75 years) treated with SABR and found no difference in major endpoints (time to progression, lung-cancer-specific survival, 2yr-OS and toxicity) according to age. OS in the group >75 years deteriorated after 5 years, due to natural shorter life expectancy. Toxicity did also not differ, no grade 4/5 toxicity was observed in the elderly group. A large retrospective database confirmed these results. In conclusion, SABR should be offered preferably to elderly and frail patients. NSCLC Stage IIIA/B (Resectable) In potentially resectable Stage IIIA, multimodal treatment with neoadjuvant chemotherapy/chemoradiation followed by surgery is recommended. However, this seems feasible only in highly selected elderly patients without relevant comorbidities and high performance score. Although beneficial in other endpoints, prospective data showed no difference in OS comparing chemoradiation alone vs chemoradiation followed by surgery , in well selected Stage IIIA/B tumors. The number of patients ≥70 years was not reported in the study of Eberhardt et al and very small (16%) in the study of Albain et al. No relevant prospective trials are available comparing trimodal therapy vs definitive chemoradiation in the elderly group. Both treatments are generally valid therapeutic options, while chemoradiation alone is less invasive and toxic and therefore likely the preferable option in frail or elderly patients. Furthermore, no prospective data are available addressing the role of surgery specifically in the elderly. Retrospective reports also included only very small numbers of patients ≥75 years (17%) evaluated for surgery due to the strict pulmonary/cardiac functional requirements for resections. , One small prospective trial specifically assessed the role of RT in patients ≥75 years and reported a median survival of 19 months with very limited toxicity. Pignon et al found no significant differences regarding OS and toxicity after curative intent RT according to age. Although comorbidities and poor functional status may influence the tolerability of radiotherapy, modern radiation techniques have clearly increased the therapeutic index. Especially if conventional chemoradiation seems not tolerable, hypofractionated RT should be evaluated. NSCLC Stage IIIB/C (Unresectable) Concurrent chemoradiation was found to be superior to RT alone already during the 1990s, but came along with increased toxicity. Consequently, subsequent trials included only small numbers of elderly patients. First, Atagi et al investigated the role of chemoradiation in patients ≥70 years. OS benefit for chemoradiation was confirmed, although the applied regimen (carboplatin only) was less intensive than usually used doublets. Similar results were shown in a meta-analysis for combined treatments despite increased toxicity. Esophageal and pulmonary toxicities are usually the clinically most relevant side effects with regard to treatment compliance and Qol and are clearly increased within combination approaches. , Therefore, sequential chemoradiation or (hypofractionated) RT are still reasonable treatment options in elderly patients deemed unfit for concurrent chemoradiation, showing limited toxicity and promising results. , , A clinical trial comparing concurrent and sequential chemoradiation in patients ≥75 years regarding quality-adjusted survival is ongoing. Patients are stratified into frail, vulnerable and fit groups based on geriatric assessment. Immunotherapy has gained attraction also in non-metastatic NSCLC patients since the PACIFIC trial showed an OS benefit for adding consolidation therapy with Durvalumab after chemoradiation compared to chemoradiation alone. , Immunotherapy generally inherits a different safety profile, which is not affected by age , at least if used as a sole treatment. Therefore, checkpoint inhibitors seem worth to be evaluated either as a substitute for concomitant chemotherapy during chemoradiation or as an adjunct after less intensive chemoradiation specifically in elderly patients. The first approach is currently evaluated in the randomized Phase II TRADE-hypo trial (NCT04351256). It includes patients >70 years, ECOG 1/2 with locally advanced, unresectable NSCLC, who are unfit for chemotherapy because of age and/or frailty and compares concomitant durvalumab with either conventionally (60Gy in 30 fractions) or hypofractionated (55Gy in 20 fractions) RT. Moreover, a Japanese prospective trial evaluating conventionally fractionated RT with low-dose carboplatin and Durvalumab followed by Durvalumab consolidation in frail/elderly stage III NSCLC patients is ongoing. Efficacy and toxicity results, especially regarding pneumonitis rates with simultaneous application, are awaited. NSCLC Stage IV (Metastatic) If RT is used for palliation or prevention of symptoms (pain, bleeding, obstruction), hypofractionated regimens are clearly preferable. In case of oligometastatic disease, the individual situation has to be evaluated. Palma et al showed in his prospective trial that SABR in oligometastatic patients may improve OS with minimal toxicity compared to chemotherapy alone. Furthermore, SABR may help to avoid or delay the need for systemic therapies, which can improve Qol especially in frail/elderly patients. Adding RT to Pembrolizumab is also a promising option with improved outcome in metastatic lung cancer. Standard of care in patients with Stage I/II is resection, alternatively stereotactic ablative body radiotherapy (SABR) is well established. A pooled analysis of the prospective randomized STARS and ROSEL trials showed even improved survival with SABR compared to surgery for resectable patients, although not focusing on elderly patients. In 2021, the updated propensity score matched results of the STARS trial confirmed that SABR is a comparable curative treatment with minimal toxicity, noninvasive character and possibility for an outpatient setting with a short treatment period (generally 1 week), which is often important for frail or elderly patients. Brooks et al retrospectively analysed 772 patients in Stage I/II NSCLC (n = 442 < 75 years, n = 330 >75 years) treated with SABR and found no difference in major endpoints (time to progression, lung-cancer-specific survival, 2yr-OS and toxicity) according to age. OS in the group >75 years deteriorated after 5 years, due to natural shorter life expectancy. Toxicity did also not differ, no grade 4/5 toxicity was observed in the elderly group. A large retrospective database confirmed these results. In conclusion, SABR should be offered preferably to elderly and frail patients. In potentially resectable Stage IIIA, multimodal treatment with neoadjuvant chemotherapy/chemoradiation followed by surgery is recommended. However, this seems feasible only in highly selected elderly patients without relevant comorbidities and high performance score. Although beneficial in other endpoints, prospective data showed no difference in OS comparing chemoradiation alone vs chemoradiation followed by surgery , in well selected Stage IIIA/B tumors. The number of patients ≥70 years was not reported in the study of Eberhardt et al and very small (16%) in the study of Albain et al. No relevant prospective trials are available comparing trimodal therapy vs definitive chemoradiation in the elderly group. Both treatments are generally valid therapeutic options, while chemoradiation alone is less invasive and toxic and therefore likely the preferable option in frail or elderly patients. Furthermore, no prospective data are available addressing the role of surgery specifically in the elderly. Retrospective reports also included only very small numbers of patients ≥75 years (17%) evaluated for surgery due to the strict pulmonary/cardiac functional requirements for resections. , One small prospective trial specifically assessed the role of RT in patients ≥75 years and reported a median survival of 19 months with very limited toxicity. Pignon et al found no significant differences regarding OS and toxicity after curative intent RT according to age. Although comorbidities and poor functional status may influence the tolerability of radiotherapy, modern radiation techniques have clearly increased the therapeutic index. Especially if conventional chemoradiation seems not tolerable, hypofractionated RT should be evaluated. Concurrent chemoradiation was found to be superior to RT alone already during the 1990s, but came along with increased toxicity. Consequently, subsequent trials included only small numbers of elderly patients. First, Atagi et al investigated the role of chemoradiation in patients ≥70 years. OS benefit for chemoradiation was confirmed, although the applied regimen (carboplatin only) was less intensive than usually used doublets. Similar results were shown in a meta-analysis for combined treatments despite increased toxicity. Esophageal and pulmonary toxicities are usually the clinically most relevant side effects with regard to treatment compliance and Qol and are clearly increased within combination approaches. , Therefore, sequential chemoradiation or (hypofractionated) RT are still reasonable treatment options in elderly patients deemed unfit for concurrent chemoradiation, showing limited toxicity and promising results. , , A clinical trial comparing concurrent and sequential chemoradiation in patients ≥75 years regarding quality-adjusted survival is ongoing. Patients are stratified into frail, vulnerable and fit groups based on geriatric assessment. Immunotherapy has gained attraction also in non-metastatic NSCLC patients since the PACIFIC trial showed an OS benefit for adding consolidation therapy with Durvalumab after chemoradiation compared to chemoradiation alone. , Immunotherapy generally inherits a different safety profile, which is not affected by age , at least if used as a sole treatment. Therefore, checkpoint inhibitors seem worth to be evaluated either as a substitute for concomitant chemotherapy during chemoradiation or as an adjunct after less intensive chemoradiation specifically in elderly patients. The first approach is currently evaluated in the randomized Phase II TRADE-hypo trial (NCT04351256). It includes patients >70 years, ECOG 1/2 with locally advanced, unresectable NSCLC, who are unfit for chemotherapy because of age and/or frailty and compares concomitant durvalumab with either conventionally (60Gy in 30 fractions) or hypofractionated (55Gy in 20 fractions) RT. Moreover, a Japanese prospective trial evaluating conventionally fractionated RT with low-dose carboplatin and Durvalumab followed by Durvalumab consolidation in frail/elderly stage III NSCLC patients is ongoing. Efficacy and toxicity results, especially regarding pneumonitis rates with simultaneous application, are awaited. If RT is used for palliation or prevention of symptoms (pain, bleeding, obstruction), hypofractionated regimens are clearly preferable. In case of oligometastatic disease, the individual situation has to be evaluated. Palma et al showed in his prospective trial that SABR in oligometastatic patients may improve OS with minimal toxicity compared to chemotherapy alone. Furthermore, SABR may help to avoid or delay the need for systemic therapies, which can improve Qol especially in frail/elderly patients. Adding RT to Pembrolizumab is also a promising option with improved outcome in metastatic lung cancer. Breast cancer (BC) remains the most common cancer diagnosis in women. Of those diagnosed with BC, about 30% are ≥70 years. In early stage BC, breast conserving surgery (BCS) is the first and one of the main pillars in oncological treatment. Even though RT is seen as a standard treatment after BCS, the treatment of elderly BC patients differs from the treatment of younger ones due to geriatric frailty or comorbidities – especially in low-risk constellations. Nevertheless, the range of general conditions in elderly patients is wide – from highly morbid patients in their early 70s to athletic patients in their late 80s – leaving the establishment of a standardized procedure hardly achievable. Especially in radiation oncology, where the side effects of a breast treatment decreased continually over the last decades, critical selection of patients is crucial, yet not always easy. If patients are not eligible for postoperative RT, mastectomy was historically considered as the alternative surgical treatment. This dogma, that all patients who cannot receive postoperative radiotherapy have to undergo mastectomy, falters in low-risk constellations (pT1, pN0, ER/PR+, Her2-). Several analyses showed that in those situations, RT can most likely be dispensed – even after BCS. Hughes et al reported on the results of the CALGB 9343 trial, which showed an advantage of tamoxifen and postoperative RT compared to tamoxifen alone after 10 years (local control 98% vs 90%, sig.). Nevertheless, this did not translate in a significant difference in time to mastectomy, time to distant metastasis, cancer-specific survival or OS. A more recent analysis by Stueber et al came to similar conclusions with a chance of relapse within 5 years of <3% in patients aged 70 years and older with low-risk features. Furthermore, the authors recommend to spare elderly low-risk patients from mastectomy with the intention of avoiding postoperative RT. In elderly high-risk patients, RT following BCS is recommended. The recent NCCN-guidelines also feature the recommendation to consider omitting breast irradiation in patients aged 70 years and older with ER-positive, cN0, pT1 tumors who receive adjuvant endocrine therapy. The same is true for the national German guideline. If the omission of radiotherapy is no option, several hypofractionated options are eligible. The most elegant option is an intraoperative single-dose radiotherapy in which the “adjuvant” treatment is already completed intraoperatively. A major disadvantage may be that the full pathological report is not available at the time of the radiotherapy and that this option is not available in all centers. Nevertheless, the randomized TARGIT-IORT trial showed a comparable long-term control to the classic RT. Accelerated breast irradiation over the course of one week was also described as non-inferior to the standard treatment course of three weeks, though long-term follow-up concerning cosmesis is still pending. Partial breast irradiation after BCS is also often discussed as an additional RT technique. However, one meta-analysis reported on significantly higher rates of in breast recurrences. In cases with higher risk constellations (T3/T4, N+, ER/PR-, Her2+, TNBC) radiotherapy according to guidelines is clearly recommended with a significant effect on recurrence-free survival being proven and confirmed by a recent analysis. The variety of treatment options (especially in low-risk constellations) is wide and needs to be discussed in detail with the patient, desirably before surgery. The patient's clinical condition and the patient’s preferences must be taken into account to reach an informed consent on the best possible treatment for the individual case. Curative intent RT for gastrointestinal cancers is usually not used as a sole modality but often within a bimodal (combined with simultaneous chemotherapy) or even trimodal (followed by surgery) approach. This complicates the evaluation of its role in elderly patients. A relevant part of toxicities and treatment compliance will be related to the concurrently applied chemotherapy rather than to RT alone. Detailed discussion of all combination approaches within gastrointestinal cancer would exceed the scope of this review. Therefore, we focused on three major gastrointestinal entities with (chemo)radiation as a major part of curative intent therapy strategies. Esophageal Cancer Neoadjuvant or definitive chemoradiation represents the current standard of care in locally advanced esophageal cancer (EC) based on resectability, while RT alone has to be considered a palliative treatment. Although the median age at diagnosis for EC is 68–70 years and roughly 30–40% of patients will be ≥75 years, , elderly patients are still underrepresented in randomized trials. This seems to be partly related to the inclusion of surgery into the treatment approach. For example, the median age within the CROSS trial (setting the standard of neoadjuvant chemoradiation followed by surgery in resectable locally advanced EC) was only 60 years. However, modern trials focusing on definitive chemoradiation analysed cohorts comparable to the general age distribution. In the randomized SCOPE-1 and ARTDECO trials, , median age was 67 and 71 years, with the latter including patients up to 90 years. Nevertheless, two SEER analyses focusing on patients ≥65 years showed an underutilization of treatment in older patients per se: , 34% of patients ≥65 years did not receive any treatment, which was significantly correlated with decreased OS. Interestingly, increasing age was associated with the non-receipt of surgery or chemotherapy but not RT. Trimodal Therapy Several studies showed that trimodal therapy (chemoradiation followed by surgery) results in improved OS also in selected elderly patients compared to surgery alone. Guttman et al analysed 1364 patients >70 years and found lower R+ resection and increased OS rates with the addition of neoadjuvant chemoradiation. Postoperative morbidity and mortality were similar. However, most studies analyzing age-dependent outcomes after trimodal therapy or resection alone showed an increase in postoperative cardiopulmonary toxicity , , in elderly patients. This finding translated into consecutively increased postoperative mortality at least in some of the trials. , Cardiopulmonary toxicity after trimodal therapy increased roughly linear with age in a pooled US analysis (+61% per decade). Similarly, postop. mortality after resection increased with age (65–69 yrs: 9%, 70–79 yrs: 13%, >80 yrs: 20%) in a large population-based study. However, this might be rather related to comorbidity than age, as a significant association between an increased Charlson score (CCI >2) with postoperative mortality was described in the latter analysis. Definitive Chemoradiation Data on definitive chemoradiation focusing on elderly patients are limited and mainly retrospective. Some reports found decreased survival rates and increased major toxicities. , However, most data suggest that chemoradiation is equally effective compared to younger patients without a major increase in adverse events. Clinical complete remissions (cCR) are achieved in 50–65% of the patients with median OS times of 12–26 months and 2-year OS-rates of 30–40%. Moreover, in the prospective SCOPE-1 trial, age (< vs ≥ 65 years) had no statistical impact on PFS or OS according to multivariate analyses. Only one prospective trial specifically addressed chemoradiation feasibility in (selected) elderly patients: Servagi-Vernat et al included 22 patients (mean 79 years) if they had a CCI ≤4, a baseline weight loss <15% and ECOG ≤2. They were treated with 50Gy and concurrent cisplatin. The treatment compliance was 100%, 64% achieved cCR at 6 weeks and 1-year OS was 62% with no acute grade 4 toxicities. The authors concluded that chemoradiation is well tolerated using these inclusion criteria, however some data suggest an increase in pulmonary complications in patients ≥80 years. Tolerability of chemoradiation may depend also on several treatment factors. Carboplatin/Paclitaxel has shown similar OS and DFS with lower toxicity rates compared to the long-time standard of Cisplatin/5-FU also in the definitive setting. The introduction of modern radiation techniques (namely image-guided intensity-modulated RT) has resulted in similar or improved outcomes with clearly reduced side effects in patients of all age groups. , The issue of the necessity of elective nodal irradiation is not finally solved in the absence of randomized trials. Indeed, increasing evidence suggests the use of smaller treatment volumes confined to the areas of gross disease because of increased tolerability. , A recent randomized trial further suggests that dose escalation beyond 50Gy to the primary tumor does not result in significantly improved outcome, but may result in increased toxicity. Therefore, it seems reasonable to restrict the total dose in elderly/frail patients to 50Gy. Nutritional status (based on nutritional risk index) or weight loss at baseline have been identified as major prognostic factors for DFS and OS in large retrospective series. , There is no clear evidence supporting the prognostic value of further weight loss or deterioration of nutritional status nor its therapeutic correction during chemoradiation. Nevertheless, care should be taken to ensure adequate nutritional support especially in elderly patients. In summary, elderly patients in good shape (especially with no or limited cardiopulmonary comorbidity) might be selected for trimodal therapy. For most patients, definitive chemoradiation (with selective reevaluation for surgery) should be preferred because of better treatment compliance and less treatment-related mortality. Chemoradiation should be performed preferably with limited treatment volumes and total doses not exceeding 50Gy. Modern imaging for treatment planning and modern RT techniques should be used. Care should be taken especially for adequate nutritional support. Future trials specifically designed for elderly cohorts are warranted. They should include the evaluation of geriatric assessment tools for the prediction of chemoradiation outcomes as for example in the ongoing OSAGE trial. Rectal Cancer Neoadjuvant (short course) RT or chemoradiation therapy represents the current standard of care in locally advanced rectal cancer (LARC) to improve locoregional control and/or resectability. Recently, neoadjuvant (chemo)radiation has been incorporated into so-called total neoadjuvant therapy (TNT) concepts combined with induction or consolidation chemotherapy prior to surgery. Several randomized trials have shown increased overall response, pathologic complete remissions (pCR) and distant control rates compared to the standard approach. , In parallel, so-called NOM approaches omitting surgery in cases of cCR after neoadjuvant chemoradiation or TNT have gained attraction, especially if sphincter-sparing surgery seems not possible. Both concepts have not been specifically studied in elderly patients. As TNT concepts usually include doublet or triplet consolidation or induction chemotherapy regimens, they seem hardly suitable for the majority of elderly patients. Nevertheless, alternative (non-surgical) treatment concepts might offer new options in elderly patients hardly suitable for extended surgery. Median age at diagnosis of rectal cancer is roughly 70 years with a peak incidence at 80–85 years, which also represents the peak prevalence age for comorbidities. , In comparison, median age in major landmark trials was roughly 10 years less. , Moreover, several population-based or retrospective studies showed an underutilization of surgery, neoadjuvant radiotherapy and palliative systemic treatment with less adherence to guidelines compared to younger patients (<70–75 years). In contrast, elderly patients received more often palliative and/or hypofractionated RT. , While treatment outcome in terms of OS has clearly improved in the last decades in younger patients (5y-OS increased from 60% to 70%), this was not the case in patients aged >75 years (5-year OS remained stable at around 40%). As the vast majority of LARC patients is treated within multimodal concepts, suitability for subsequent treatments has to be included into the evaluation of radiation approaches. Most available data indicate no distinct differences in treatment compliance or tolerance within neoadjuvant (chemo)radiation concepts comparing younger and older patients (cutoff typically 70–75 years). Some trials showed higher rates of acute G3+ toxicity with limited clinical consequences. In contrast, a large Dutch study found a clear and steady increase in 1- and 6-month mortality after surgery with increasing age starting at 75 years. , While postoperative complication rates per se showed no significant difference, the onset of a complication resulted in clearly worse outcome in elderly patients. , For example, anastomotic leakage occurred at a rate of roughly 10% but resulted in a mortality rate of 8% in younger vs 57% in older patients. Interestingly, the same studies did not find a significant association of preoperative RT with postoperative complication rates. , However, they showed an improved outcome with the addition of RT to surgery in elderly, which is also supported by a SEER analysis. The authors concluded that RT had little or no impact on postoperative complication rates or mortality, while the surgical trauma itself remains most important. In contrast, comorbidity seems to be clearly linked with postoperative complications and 30-day-mortality. A Dutch study restricted to patients >75 years showed only a moderate difference in postoperative complication rates and none in 30- day-mortality with or without neoadjuvant short-course RT in the entire cohort. However, they described a roughly 5-fold increase in complication rates and a more than 10-fold increase in 30- day-mortality, if severe comorbidity like COPD, diabetes or cerebrovascular disease was present. Therefore, the most important question to answer prior to RT is suitability for major pelvic surgery. If a patient is deemed suitable, indication for neoadjuvant (chemo)radiation can usually follow standard recommendations. Several retrospective studies showed high compliance rates and similar results in patients aged >70 and deemed fit for surgery compared to younger ones. , There is no clear evidence for a distinctly different outcome comparing neoadjuvant short-course RT with long-course chemoradiation specifically in elderly patients, thus the indication may follow the institutional standards and general recommendations. However, short-course RT might be preferred with regard to patient’s convenience. In elderly patients deemed less suitable for major pelvic surgery, several algorithms have been proposed. , Paradoxically, some include treatment intensification of chemoradiation because of a higher likeliness of response. Moderate treatment intensification by localized dose escalation either via external beam RT or addition of brachytherapy can usually be achieved without a major increase in side effects. A consequently pronounced response then might enable local excision (LE) or even omission of surgery without compromising the overall results. Several trials suggested similar local control rates at least in node-negative patients with chemoradiation followed by LE in responding patients compared to more extended surgery, although not specifically for elderly patients. Short-course RT seems less suitable for this approach because several reports suggest increased complication rates with LE after short-course RT compared to chemoradiation. In frail (medically inoperable) patients, chemoradiation alone , or EBRT with or without brachytherapy are reasonable options. , If RT alone is used, hypofractionation should be strongly considered, because it achieved similar results to conventional fractionation. Short-course RT is an effective regimen for palliation with >80% complete or partial symptom relief at four weeks and reasonable rates of colostomy-free and overall survival. The addition of brachytherapy may further enhance the results as shown by a recent Phase I trial: increasing weekly doses of brachytherapy were added to a moderately hypofractionated EBRT concept and resulted in good response rates (cCR 61%) and acceptable OS. However, within the MTD cohort of this trial, rectal grade 3+ toxicity was roughly 30% prompting the authors to recommend some form of optimization. Several structured reviews have proposed similar treatment algorithms for elderly patients with rectal cancer. , Based on the available evidence, elderly patients in good shape (suitable for major pelvic surgery) should be preferably treated according to standard recommendations. Patients with intermediate features (some comorbidity, less suitable for major surgery) might be treated with chemoradiation (with or without dose escalation) and LE or omission or surgery in case of partial or complete response. In patients with severe comorbidity (unlikely to undergo surgery at all), hypofractionated radiotherapy achieves good palliation and seems feasible in most cases. Addition of brachytherapy can improve results but has to be weighed against increased complications risks. Anal Cancer In contrast to many other gastrointestinal malignancies, most patients with anal cancer are managed without surgery. The cornerstone of treatment is definitive chemoradiation with curative intent. Chemoradiation results in significantly improved outcomes compared to RT alone based on randomized trials. The median age at diagnosis is 60–65 years and roughly onethird of the patients are aged ≥70 years. , Standard treatment usually includes a doublet chemotherapy regimen (Mitomycin C or Cisplatin combined with 5-FU or capecitabine) simultaneously applied to RT. This may result in considerable rates of acute gastrointestinal, hematological and skin toxicities. Therefore, its suitability to elderly patients is often questioned by clinicians. Data specifically addressing elderly patients are rare, mainly retrospective and therefore susceptible for selection biases. Moreover, the vast majority of published series used outdated staging modalities and radiation techniques (2D- or 3D-conformal RT). In contrast, modern techniques like intensity-modulated RT (IMRT) have shown clearly reduced rates of side effects in younger populations. Nevertheless, most published data suggest that age per se is not limiting the capability to tolerate standard therapy. In a large population-based analysis of roughly 12,000 patients treated with curative intent, age was not an independent factor for receiving chemoradiation in multivariate analysis. However, patients with two or more comorbidities were more likely to receive RT alone. Several authors analyzed elderly cohorts with cutoffs of 70–80 years treated by chemoradiation or RT alone. They showed high treatment compliance for RT but 25–50% of the patients needed dose reductions of chemotherapy, especially if comorbidity was present. , Addition of chemotherapy resulted in significantly increased toxicity but also in improved outcome in most reports, including colostomy-free survival. Some investigators analyzed patients treated with either RT or chemoradiation according to age. , They observed less CHT use in elderly patients, which showed also worse performance scores. If only patients with chemoradiation were compared, overall toxicity was not clearly increased in elderly patients, but they tended to have less skin but more hematological side effects. Outcome parameters were reported only for the overall cohorts. While cCR, colostomy and LC rates were similar, MFS, DFS and OS were worse in older patients, probably related to less chemotherapy use and increased mortality by other causes. , Recent studies analyzed larger cohorts receiving chemoradiation by age groups. , They consistently found reduced treatment compliance (mainly to chemotherapy) with increasing age but no clear difference in overall toxicity. However, none of the measured outcome parameters was associated with age, while comorbidity was associated with more toxicity, more dose reductions and worse LC. Thus, it seems reasonable to offer standard regimens also to elderly patients at least if performance score is good and no severe comorbidity is present. In patients with comorbidities or limited performance status, several adjustments are possible with still curative intent. Chemoradiation can be performed using only one sensitizing agent (either 5-FU, Mitomycin or Cisplatin). Although chemoradiation with simultaneous doublet therapy (MMC/5-FU) was superior to 5-FU mono regarding LC, MFS and DFS in RTOG 87–04, no significant survival difference was reported. However, patients with cardiopulmonary comorbidities might be less suitable for 5-FU or capecitabine therapy. In those patients, MMC or Cisplatin mono may represent a reasonable alternative, although not supported by sufficient data. In patients deemed unsuitable for chemotherapy at all, conventionally fractionated RT alone may still result in good outcome with less toxicity. In the RT only arms of both randomized trials showing superiority of chemoradiation, 5-year OS was still >50%. RT alone can be tolerated by most patients of advanced age even in the presence of comorbidities, , especially if modern radiation techniques are used. RTOG 0529 evaluated IMRT for anal cancer in a single-arm phase II design and demonstrated significant reductions in acute toxicity compared to historic data from RTOG 98–11. Moreover, they showed that moderate de-escalation of dose did not result in decreased outcome. Therefore, the use of modern RT techniques should be mandatory in elderly patients. In frail patients, further adjustments can be made regarding target volume and dose. Charnley et al reported the results of limited treatment (30Gy in 10 fractions to the primary tumor excluding elective nodal irradiation combined with low-dose 5-FU) in frail patients and found 100% treatment compliance with low toxicity and still tolerable outcomes. In summary, elderly patients in good shape should be preferably treated according to standard recommendations. In patients with comorbidities or limited performance status, stepwise adjustments can be made, ranging from chemoradiation with a single agent to standard RT alone. While toxicity will be clearly reduced with every step, long-term survival is still likely in the majority of patients. In patients with severe comorbidity, hypofractionated radiotherapy with limited volumes still achieves good palliation. The use of modern RT techniques like IMRT is strongly recommended to reduce acute and late side effects. Neoadjuvant or definitive chemoradiation represents the current standard of care in locally advanced esophageal cancer (EC) based on resectability, while RT alone has to be considered a palliative treatment. Although the median age at diagnosis for EC is 68–70 years and roughly 30–40% of patients will be ≥75 years, , elderly patients are still underrepresented in randomized trials. This seems to be partly related to the inclusion of surgery into the treatment approach. For example, the median age within the CROSS trial (setting the standard of neoadjuvant chemoradiation followed by surgery in resectable locally advanced EC) was only 60 years. However, modern trials focusing on definitive chemoradiation analysed cohorts comparable to the general age distribution. In the randomized SCOPE-1 and ARTDECO trials, , median age was 67 and 71 years, with the latter including patients up to 90 years. Nevertheless, two SEER analyses focusing on patients ≥65 years showed an underutilization of treatment in older patients per se: , 34% of patients ≥65 years did not receive any treatment, which was significantly correlated with decreased OS. Interestingly, increasing age was associated with the non-receipt of surgery or chemotherapy but not RT. Several studies showed that trimodal therapy (chemoradiation followed by surgery) results in improved OS also in selected elderly patients compared to surgery alone. Guttman et al analysed 1364 patients >70 years and found lower R+ resection and increased OS rates with the addition of neoadjuvant chemoradiation. Postoperative morbidity and mortality were similar. However, most studies analyzing age-dependent outcomes after trimodal therapy or resection alone showed an increase in postoperative cardiopulmonary toxicity , , in elderly patients. This finding translated into consecutively increased postoperative mortality at least in some of the trials. , Cardiopulmonary toxicity after trimodal therapy increased roughly linear with age in a pooled US analysis (+61% per decade). Similarly, postop. mortality after resection increased with age (65–69 yrs: 9%, 70–79 yrs: 13%, >80 yrs: 20%) in a large population-based study. However, this might be rather related to comorbidity than age, as a significant association between an increased Charlson score (CCI >2) with postoperative mortality was described in the latter analysis. Data on definitive chemoradiation focusing on elderly patients are limited and mainly retrospective. Some reports found decreased survival rates and increased major toxicities. , However, most data suggest that chemoradiation is equally effective compared to younger patients without a major increase in adverse events. Clinical complete remissions (cCR) are achieved in 50–65% of the patients with median OS times of 12–26 months and 2-year OS-rates of 30–40%. Moreover, in the prospective SCOPE-1 trial, age (< vs ≥ 65 years) had no statistical impact on PFS or OS according to multivariate analyses. Only one prospective trial specifically addressed chemoradiation feasibility in (selected) elderly patients: Servagi-Vernat et al included 22 patients (mean 79 years) if they had a CCI ≤4, a baseline weight loss <15% and ECOG ≤2. They were treated with 50Gy and concurrent cisplatin. The treatment compliance was 100%, 64% achieved cCR at 6 weeks and 1-year OS was 62% with no acute grade 4 toxicities. The authors concluded that chemoradiation is well tolerated using these inclusion criteria, however some data suggest an increase in pulmonary complications in patients ≥80 years. Tolerability of chemoradiation may depend also on several treatment factors. Carboplatin/Paclitaxel has shown similar OS and DFS with lower toxicity rates compared to the long-time standard of Cisplatin/5-FU also in the definitive setting. The introduction of modern radiation techniques (namely image-guided intensity-modulated RT) has resulted in similar or improved outcomes with clearly reduced side effects in patients of all age groups. , The issue of the necessity of elective nodal irradiation is not finally solved in the absence of randomized trials. Indeed, increasing evidence suggests the use of smaller treatment volumes confined to the areas of gross disease because of increased tolerability. , A recent randomized trial further suggests that dose escalation beyond 50Gy to the primary tumor does not result in significantly improved outcome, but may result in increased toxicity. Therefore, it seems reasonable to restrict the total dose in elderly/frail patients to 50Gy. Nutritional status (based on nutritional risk index) or weight loss at baseline have been identified as major prognostic factors for DFS and OS in large retrospective series. , There is no clear evidence supporting the prognostic value of further weight loss or deterioration of nutritional status nor its therapeutic correction during chemoradiation. Nevertheless, care should be taken to ensure adequate nutritional support especially in elderly patients. In summary, elderly patients in good shape (especially with no or limited cardiopulmonary comorbidity) might be selected for trimodal therapy. For most patients, definitive chemoradiation (with selective reevaluation for surgery) should be preferred because of better treatment compliance and less treatment-related mortality. Chemoradiation should be performed preferably with limited treatment volumes and total doses not exceeding 50Gy. Modern imaging for treatment planning and modern RT techniques should be used. Care should be taken especially for adequate nutritional support. Future trials specifically designed for elderly cohorts are warranted. They should include the evaluation of geriatric assessment tools for the prediction of chemoradiation outcomes as for example in the ongoing OSAGE trial. Neoadjuvant (short course) RT or chemoradiation therapy represents the current standard of care in locally advanced rectal cancer (LARC) to improve locoregional control and/or resectability. Recently, neoadjuvant (chemo)radiation has been incorporated into so-called total neoadjuvant therapy (TNT) concepts combined with induction or consolidation chemotherapy prior to surgery. Several randomized trials have shown increased overall response, pathologic complete remissions (pCR) and distant control rates compared to the standard approach. , In parallel, so-called NOM approaches omitting surgery in cases of cCR after neoadjuvant chemoradiation or TNT have gained attraction, especially if sphincter-sparing surgery seems not possible. Both concepts have not been specifically studied in elderly patients. As TNT concepts usually include doublet or triplet consolidation or induction chemotherapy regimens, they seem hardly suitable for the majority of elderly patients. Nevertheless, alternative (non-surgical) treatment concepts might offer new options in elderly patients hardly suitable for extended surgery. Median age at diagnosis of rectal cancer is roughly 70 years with a peak incidence at 80–85 years, which also represents the peak prevalence age for comorbidities. , In comparison, median age in major landmark trials was roughly 10 years less. , Moreover, several population-based or retrospective studies showed an underutilization of surgery, neoadjuvant radiotherapy and palliative systemic treatment with less adherence to guidelines compared to younger patients (<70–75 years). In contrast, elderly patients received more often palliative and/or hypofractionated RT. , While treatment outcome in terms of OS has clearly improved in the last decades in younger patients (5y-OS increased from 60% to 70%), this was not the case in patients aged >75 years (5-year OS remained stable at around 40%). As the vast majority of LARC patients is treated within multimodal concepts, suitability for subsequent treatments has to be included into the evaluation of radiation approaches. Most available data indicate no distinct differences in treatment compliance or tolerance within neoadjuvant (chemo)radiation concepts comparing younger and older patients (cutoff typically 70–75 years). Some trials showed higher rates of acute G3+ toxicity with limited clinical consequences. In contrast, a large Dutch study found a clear and steady increase in 1- and 6-month mortality after surgery with increasing age starting at 75 years. , While postoperative complication rates per se showed no significant difference, the onset of a complication resulted in clearly worse outcome in elderly patients. , For example, anastomotic leakage occurred at a rate of roughly 10% but resulted in a mortality rate of 8% in younger vs 57% in older patients. Interestingly, the same studies did not find a significant association of preoperative RT with postoperative complication rates. , However, they showed an improved outcome with the addition of RT to surgery in elderly, which is also supported by a SEER analysis. The authors concluded that RT had little or no impact on postoperative complication rates or mortality, while the surgical trauma itself remains most important. In contrast, comorbidity seems to be clearly linked with postoperative complications and 30-day-mortality. A Dutch study restricted to patients >75 years showed only a moderate difference in postoperative complication rates and none in 30- day-mortality with or without neoadjuvant short-course RT in the entire cohort. However, they described a roughly 5-fold increase in complication rates and a more than 10-fold increase in 30- day-mortality, if severe comorbidity like COPD, diabetes or cerebrovascular disease was present. Therefore, the most important question to answer prior to RT is suitability for major pelvic surgery. If a patient is deemed suitable, indication for neoadjuvant (chemo)radiation can usually follow standard recommendations. Several retrospective studies showed high compliance rates and similar results in patients aged >70 and deemed fit for surgery compared to younger ones. , There is no clear evidence for a distinctly different outcome comparing neoadjuvant short-course RT with long-course chemoradiation specifically in elderly patients, thus the indication may follow the institutional standards and general recommendations. However, short-course RT might be preferred with regard to patient’s convenience. In elderly patients deemed less suitable for major pelvic surgery, several algorithms have been proposed. , Paradoxically, some include treatment intensification of chemoradiation because of a higher likeliness of response. Moderate treatment intensification by localized dose escalation either via external beam RT or addition of brachytherapy can usually be achieved without a major increase in side effects. A consequently pronounced response then might enable local excision (LE) or even omission of surgery without compromising the overall results. Several trials suggested similar local control rates at least in node-negative patients with chemoradiation followed by LE in responding patients compared to more extended surgery, although not specifically for elderly patients. Short-course RT seems less suitable for this approach because several reports suggest increased complication rates with LE after short-course RT compared to chemoradiation. In frail (medically inoperable) patients, chemoradiation alone , or EBRT with or without brachytherapy are reasonable options. , If RT alone is used, hypofractionation should be strongly considered, because it achieved similar results to conventional fractionation. Short-course RT is an effective regimen for palliation with >80% complete or partial symptom relief at four weeks and reasonable rates of colostomy-free and overall survival. The addition of brachytherapy may further enhance the results as shown by a recent Phase I trial: increasing weekly doses of brachytherapy were added to a moderately hypofractionated EBRT concept and resulted in good response rates (cCR 61%) and acceptable OS. However, within the MTD cohort of this trial, rectal grade 3+ toxicity was roughly 30% prompting the authors to recommend some form of optimization. Several structured reviews have proposed similar treatment algorithms for elderly patients with rectal cancer. , Based on the available evidence, elderly patients in good shape (suitable for major pelvic surgery) should be preferably treated according to standard recommendations. Patients with intermediate features (some comorbidity, less suitable for major surgery) might be treated with chemoradiation (with or without dose escalation) and LE or omission or surgery in case of partial or complete response. In patients with severe comorbidity (unlikely to undergo surgery at all), hypofractionated radiotherapy achieves good palliation and seems feasible in most cases. Addition of brachytherapy can improve results but has to be weighed against increased complications risks. In contrast to many other gastrointestinal malignancies, most patients with anal cancer are managed without surgery. The cornerstone of treatment is definitive chemoradiation with curative intent. Chemoradiation results in significantly improved outcomes compared to RT alone based on randomized trials. The median age at diagnosis is 60–65 years and roughly onethird of the patients are aged ≥70 years. , Standard treatment usually includes a doublet chemotherapy regimen (Mitomycin C or Cisplatin combined with 5-FU or capecitabine) simultaneously applied to RT. This may result in considerable rates of acute gastrointestinal, hematological and skin toxicities. Therefore, its suitability to elderly patients is often questioned by clinicians. Data specifically addressing elderly patients are rare, mainly retrospective and therefore susceptible for selection biases. Moreover, the vast majority of published series used outdated staging modalities and radiation techniques (2D- or 3D-conformal RT). In contrast, modern techniques like intensity-modulated RT (IMRT) have shown clearly reduced rates of side effects in younger populations. Nevertheless, most published data suggest that age per se is not limiting the capability to tolerate standard therapy. In a large population-based analysis of roughly 12,000 patients treated with curative intent, age was not an independent factor for receiving chemoradiation in multivariate analysis. However, patients with two or more comorbidities were more likely to receive RT alone. Several authors analyzed elderly cohorts with cutoffs of 70–80 years treated by chemoradiation or RT alone. They showed high treatment compliance for RT but 25–50% of the patients needed dose reductions of chemotherapy, especially if comorbidity was present. , Addition of chemotherapy resulted in significantly increased toxicity but also in improved outcome in most reports, including colostomy-free survival. Some investigators analyzed patients treated with either RT or chemoradiation according to age. , They observed less CHT use in elderly patients, which showed also worse performance scores. If only patients with chemoradiation were compared, overall toxicity was not clearly increased in elderly patients, but they tended to have less skin but more hematological side effects. Outcome parameters were reported only for the overall cohorts. While cCR, colostomy and LC rates were similar, MFS, DFS and OS were worse in older patients, probably related to less chemotherapy use and increased mortality by other causes. , Recent studies analyzed larger cohorts receiving chemoradiation by age groups. , They consistently found reduced treatment compliance (mainly to chemotherapy) with increasing age but no clear difference in overall toxicity. However, none of the measured outcome parameters was associated with age, while comorbidity was associated with more toxicity, more dose reductions and worse LC. Thus, it seems reasonable to offer standard regimens also to elderly patients at least if performance score is good and no severe comorbidity is present. In patients with comorbidities or limited performance status, several adjustments are possible with still curative intent. Chemoradiation can be performed using only one sensitizing agent (either 5-FU, Mitomycin or Cisplatin). Although chemoradiation with simultaneous doublet therapy (MMC/5-FU) was superior to 5-FU mono regarding LC, MFS and DFS in RTOG 87–04, no significant survival difference was reported. However, patients with cardiopulmonary comorbidities might be less suitable for 5-FU or capecitabine therapy. In those patients, MMC or Cisplatin mono may represent a reasonable alternative, although not supported by sufficient data. In patients deemed unsuitable for chemotherapy at all, conventionally fractionated RT alone may still result in good outcome with less toxicity. In the RT only arms of both randomized trials showing superiority of chemoradiation, 5-year OS was still >50%. RT alone can be tolerated by most patients of advanced age even in the presence of comorbidities, , especially if modern radiation techniques are used. RTOG 0529 evaluated IMRT for anal cancer in a single-arm phase II design and demonstrated significant reductions in acute toxicity compared to historic data from RTOG 98–11. Moreover, they showed that moderate de-escalation of dose did not result in decreased outcome. Therefore, the use of modern RT techniques should be mandatory in elderly patients. In frail patients, further adjustments can be made regarding target volume and dose. Charnley et al reported the results of limited treatment (30Gy in 10 fractions to the primary tumor excluding elective nodal irradiation combined with low-dose 5-FU) in frail patients and found 100% treatment compliance with low toxicity and still tolerable outcomes. In summary, elderly patients in good shape should be preferably treated according to standard recommendations. In patients with comorbidities or limited performance status, stepwise adjustments can be made, ranging from chemoradiation with a single agent to standard RT alone. While toxicity will be clearly reduced with every step, long-term survival is still likely in the majority of patients. In patients with severe comorbidity, hypofractionated radiotherapy with limited volumes still achieves good palliation. The use of modern RT techniques like IMRT is strongly recommended to reduce acute and late side effects. RT is an integral part of the treatment in many gynecological cancers, however often used as an adjunct to surgery. We therefore focused on cervical cancer as the main entity using nonsurgical radiation-containing approaches. Recently, the American Cancer Society reported that more than 15% of cervical cancer cases were found in women aged over 65 years. In large clinical studies, only a few elderly patients were included. Venkatesulu et al conducted a systematic review about patterns of care of cervical cancer in elderly and found only 24 out of 17,338 publications addressing the outcome in elderly cohorts. In these publications, 11,279 out of 14,479 patients aged ≥60 years (78%) received EBRT with or without concurrent chemotherapy and/or brachytherapy. With regard to the latter, low dose rate (LDR) was the most common modality, followed by high dose rate (HDR). However, in some studies with scheduled brachytherapy, up to 30% of the patients did not receive it due technical reasons (48.7%), comorbidities (69.4%) or patient refusal (38.3%). Five-year OS was generally inferior (27–69%) for elderly patients compared to younger populations (58–75%). Suboptimal radiation dose resulted in clearly reduced 5-year OS (11%) compared to patients treated with chemoradiation followed by brachytherapy (74%). However, it remains unclear if elderly patients with cervical cancer have a worse prognosis per se, due to comorbidity or limited performance scores or due to limited treatment application or tolerability. Limited historical data showed inconsistent results regarding generally worse , or similar , survival outcomes in elderly compared to younger patients. The performance status has been reported as a significant prognostic factor for OS and PFS. However, a retrospective registry-based study showed that patients ≥60 years were less likely to receive standard therapy compared to younger ones, mainly because treatment was not recommended. In a Japanese study, some elderly patients experienced severe toxicity, although radiotherapy was generally effective for them. These conditions might further contribute to the adverse effect on the prognosis of elderly patients. Interestingly, Hou et al reported comparable treatment outcomes with regard to CSS and loco-regional control in elderly patients with cervical cancer despite receiving less comprehensive treatment (including less concurrent chemoradiation, less brachytherapy, lower total RT dose and limited EBRT volumes). The authors concluded that, for patients ≥70 years a conservative treatment strategy with RT alone could be appropriate, especially in those with a favorable stage or histopathology. Similarly, Shimamoto et al found equivalent DFS but worse OS in patients aged ≥65 years compared to younger ones. Modern EBRT techniques with a reduction of treatment-related side effects might be helpful for elderly patients to improve outcomes and RT completion rates. Several studies have confirmed clearly reduced side effects with IMRT techniques compared to conventional 2D- or 3D-conformal treatments. , Brachytherapy was also often refused in elderly patients due to fear of toxicities or other reasons in elderly patients, , , but seems applicable with acceptable toxicity using modern image-guided approaches. In a prospective cohort study, Seppenwoolde et al analyzed morbidity and dose–volume effects in definitive radiochemotherapy for locally advanced cervical cancer. This cohort was treated with modern treatment techniques according to the EMBRACE protocol (94% received IMRT techniques). Most common gastrointestinal toxicities were low-grade diarrhea, stool urgency, rectal incontinence, rectal bleeding and weight loss. Most common genitourinary toxicities were dysuria and bladder incontinence. Only stool urgency, rectal or bladder incontinence and weight loss showed significantly increasing rates over the entire dose range. Correlations of toxicities with certain dose-volume factors (V40Gy) were observed, linking dysuria to bladder dose and stool urgency and rectal incontinence to bowel and rectum doses. Based on these data, the following treatment planning objectives were recommended to minimize stool urgency, rectal and urinary incontinence: bowel V40Gy≤250cm 3 , rectum V40Gy≤80% and bladder V40Gy≤80–90%, respectively. Using modern EBRT techniques considering those assumptions during treatment planning will likely increase treatment tolerability and ensure treatment completion in a larger proportion of elderly patients in the future. In summary, age influences treatment strategies for cervical cancer world-wide, although based on scarce evidence. Only very few studies addressed treatment outcome in elderly patients with cervical cancer. However, most elderly patients may be treated with curative intent using modern radiation techniques with adequate supportive care. Recent studies have provided more insights into dose constraints for organs at risk to minimize acute and late toxicity by sophisticated treatment planning. The use of concurrent chemotherapy should be carefully evaluated based on patients´ performance status and comorbidities balancing possible benefit and harms. Brachytherapy should be an integral part of the treatment regimens in elderly patients as it can be safely applied by modern image-guided approaches with acceptable toxicities. Prostate Cancer Prostate cancer is a disease of the elderly and age remains to be the most important risk factor for its development. Due to its high prevalence and since the population will continue to grow older, prostate cancer and its management is a major health and socioeconomic factor. The population-wide PSA screening combined with the good prognosis of prostate cancer carries the inherent risk of overtreatment, which is aggravated in elderly patients with a limited life span due to comorbidities or frailty. This has to be taken into account in screening strategies and treatment decisions. EAU, S3 as well as international NCCN guidelines have abandoned general population-based PSA screening in favor of offering early detection programs to select patients after discussion of pros and cons. For men ≥70 years, EAU guidelines for example give a grade D recommendation (rather than C for men aged 55–69) to inform about the possibility of PSA screening. A definition of ‘select patients’ is not provided, but data from the ERSPC and PIVOT trial suggest that men with a life expectancy of ≤15 years are unlikely to benefit from screening. , However, after decades of routine PSA screening of men aged ≥45, the current recommendations do not seem to be fully adopted in clinical practice yet. The same is true for the primary treatment of elderly patients with very low risk and low risk prostate cancer for whom active surveillance is the preferred approach when the life expectancy is below 20 and 10 years, respectively. The ProtecT trial has shown that active surveillance had similar very high 10-year-OS compared to surgery or primary RT. For patients with a life expectancy below 5 years, observation (not active surveillance!) is suggested in NCCN guidelines. The dependency on life expectancy underlines the importance of geriatric assessment of prostate cancer patients as a basis of treatment decisions. The task force of the International Society of Geriatric Oncology has issued recommendations for the management of prostate cancer in elderly patients. They emphasize that patients should be managed according to their individual health status rather than age. Therefore, initial evaluation of health status is mandatory. A two-step process is suggested using the validated G8 screening tool which identifies patients who should undergo full comprehensive geriatric assessment (CGA). For elderly patients having aggressive (Gleason grade 8–10) or locally advanced disease, immediate treatment is usually warranted. Local complications such as bleeding, urinary retention and problems arising from distant metastases may occur within a short timeframe. In these patients, EBRT is the treatment of choice since it is less invasive and can be better tailored to the individual situation in terms of total dose and treatment time. For such patients, ultrahypofractionated radiation schedules, using 4–7 high-dose fractions daily or every other day, may be the preferred treatment. Such regimens reduce strain on patients in terms of required hospital visits (and transportation) and have been shown to be non-inferior to moderate or normofractionated schedules in randomized Phase III trials , and metaanalyses. Androgen deprivation therapy (ADT) should be added according to current guidelines. However, cardiovascular risk and long-term complications as a result of reduced bone density need to be considered in the elderly. Notably, Qol (hot flashes, depression, fatigue) is typically less affected by long-term ADT in elderly patients than in younger patients featuring higher testosterone levels. In the oligometastatic setting, data have shown OS benefits for both, RT of the primary tumor as well as metastases directed therapy for patients with low-volume disease. These treatments should not be held back thoughtlessly in elderly patients. They inherit a very low risk of severe toxicity, especially when delivered via SBRT. Thus, elderly patients will likely benefit in a similar range than younger ones. If life expectancy is severely impaired by comorbidities such as secondary cancers, palliative androgen deprivation monotherapy may be considered. Prostate cancer is a disease of the elderly and age remains to be the most important risk factor for its development. Due to its high prevalence and since the population will continue to grow older, prostate cancer and its management is a major health and socioeconomic factor. The population-wide PSA screening combined with the good prognosis of prostate cancer carries the inherent risk of overtreatment, which is aggravated in elderly patients with a limited life span due to comorbidities or frailty. This has to be taken into account in screening strategies and treatment decisions. EAU, S3 as well as international NCCN guidelines have abandoned general population-based PSA screening in favor of offering early detection programs to select patients after discussion of pros and cons. For men ≥70 years, EAU guidelines for example give a grade D recommendation (rather than C for men aged 55–69) to inform about the possibility of PSA screening. A definition of ‘select patients’ is not provided, but data from the ERSPC and PIVOT trial suggest that men with a life expectancy of ≤15 years are unlikely to benefit from screening. , However, after decades of routine PSA screening of men aged ≥45, the current recommendations do not seem to be fully adopted in clinical practice yet. The same is true for the primary treatment of elderly patients with very low risk and low risk prostate cancer for whom active surveillance is the preferred approach when the life expectancy is below 20 and 10 years, respectively. The ProtecT trial has shown that active surveillance had similar very high 10-year-OS compared to surgery or primary RT. For patients with a life expectancy below 5 years, observation (not active surveillance!) is suggested in NCCN guidelines. The dependency on life expectancy underlines the importance of geriatric assessment of prostate cancer patients as a basis of treatment decisions. The task force of the International Society of Geriatric Oncology has issued recommendations for the management of prostate cancer in elderly patients. They emphasize that patients should be managed according to their individual health status rather than age. Therefore, initial evaluation of health status is mandatory. A two-step process is suggested using the validated G8 screening tool which identifies patients who should undergo full comprehensive geriatric assessment (CGA). For elderly patients having aggressive (Gleason grade 8–10) or locally advanced disease, immediate treatment is usually warranted. Local complications such as bleeding, urinary retention and problems arising from distant metastases may occur within a short timeframe. In these patients, EBRT is the treatment of choice since it is less invasive and can be better tailored to the individual situation in terms of total dose and treatment time. For such patients, ultrahypofractionated radiation schedules, using 4–7 high-dose fractions daily or every other day, may be the preferred treatment. Such regimens reduce strain on patients in terms of required hospital visits (and transportation) and have been shown to be non-inferior to moderate or normofractionated schedules in randomized Phase III trials , and metaanalyses. Androgen deprivation therapy (ADT) should be added according to current guidelines. However, cardiovascular risk and long-term complications as a result of reduced bone density need to be considered in the elderly. Notably, Qol (hot flashes, depression, fatigue) is typically less affected by long-term ADT in elderly patients than in younger patients featuring higher testosterone levels. In the oligometastatic setting, data have shown OS benefits for both, RT of the primary tumor as well as metastases directed therapy for patients with low-volume disease. These treatments should not be held back thoughtlessly in elderly patients. They inherit a very low risk of severe toxicity, especially when delivered via SBRT. Thus, elderly patients will likely benefit in a similar range than younger ones. If life expectancy is severely impaired by comorbidities such as secondary cancers, palliative androgen deprivation monotherapy may be considered. Generally, RT is an important tool in the treatment of cancer patients. It has been estimated that more than 50% of all cancer patients benefit from some form of RT. With the latest improvements in radiation technique, cure rates have further increased, paralleled by a reduction in side effects. Thus, RT is a particularly attractive local treatment in elderly patients because of its non-invasive nature with limited systemic toxicities. Recently, hypofractionated regimens using higher single doses but fewer fractions have been established in many cancer entities. They achieved similar rates of tumor control and toxicity compared to conventionally fractionated approaches. This seems specifically relevant for older and/or frail patients with lack of social, financial or practical support for transportation and may further improve patient’s acceptance and tolerance of RT. However, many curative intent radiation therapies still involve simultaneous applications of chemotherapy (Chemoradiation), which impedes strong hypofractionation approaches. Nevertheless, many older patients can tolerate conventionally fractionated chemoradiation regimens without increased side effects as shown by our summarized data. This is especially true, if older patients present in good performance score without significant comorbidities. Therefore, age alone should not be a decisive factor for treatment assignment. However, some of the mentioned studies also showed decreased treatment tolerance or the need for treatment adaptions, especially within combination approaches in relevant parts of elderly cohorts. Therefore, tools for an adequate selection of elderly patients for standard regimens as well as for adapted regimens or even omission of RT are urgently needed. Several authors have used different assessment tools to predict treatment tolerance or side effects in elderly cancer patients with regard to surgery, chemotherapy and RT. The most sophisticated tool is a comprehensive geriatric assessment (CGA). This includes a multidisciplinary diagnostic process that evaluates medical, psychological, social and functional capacity. However, CGA is time consuming (it takes roughly two hours) and therefore unlikely to be applied in a relevant proportion of elderly patients in daily routine practice prior to RT. Therefore, screening tools have been developed either as surrogate markers to guide treatment decisions directly or at least to reduce the proportion of elderly patients amended to a full CGA. For example, Hurria et al developed an 11-item scale that correctly predicted the risk of severe toxicities in the majority of patients treated with chemotherapy in an external validation cohort. However, most of these screening tools still lack adequate discriminative power for the prediction of tolerability according to a recent review. Moreover, most of the studies did not involve patients with combined chemoradiation approaches. In studies specifically addressing elderly patients with planned RT, the use of either screening tools or CGA was even less promising. Szumacher et al performed a systematic review and found no significant association with RT-related toxicity in the majority of the included studies. One possible reason lies within the different spectrum of acute side effects caused by RT. In contrast to chemotherapy, most severe side effects are locoregionally limited, site-specific and depend on irradiated volume, total dose and fractionation. For example, lung cancer treatment by a small-volume SABR in 3 fractions with total doses equivalent to ≥80Gy in conventional fractionation will hardly result in any severe acute toxicity. Therefore, it can be used even in very old, frail patients with severe pulmonal comorbidity. In contrast, large field chemoradiation for lung cancer (even with much lower equivalent doses) will likely result in unacceptable side effects in similar patients. Moreover, RT inherits the risk of causing late toxicities possibly limiting long-term Qol. Again, those are site-specific, depend on volume, dose and fractionation and additionally on preexisting mainly organ-specific comorbidities. While an older or frail patient with limited pulmonary reserve will be at higher risk for all major surgeries, this is not true for all high-dose radiation treatments. Moreover, the timely onset of late radiation toxicities is highly variable and their detection needs long and careful follow-up strategies. Taken together, it is unlikely that common CGA or screening tests will be able to accurately predict tolerability or toxicity caused by RT, because they do not include site-specific and treatment-specific variables. Therefore, one key priority of future radiation research must be the development of modified evaluation tools taken the specific conditions of RT into account. Ideally, they would use at least some information already ascertained during routine RT planning. Prediction tools using such information could be more conveniently incorporated into daily routine compared to long lasting assessments. For example, radiomics and deep-learning algorithms might be used to gain more site-specific information already inherited in routinely performed treatment planning CTs. The latter methods also allow the incorporation of multiple information into new models. This may include site-specific clinical information, molecular data, imaging and (RT)-treatment-related factors resulting in more complex, nevertheless easily usable models. Aside from patient selection for standard treatments, underrepresentation of elderly and/or comorbid patients in clinical trials is a major issue. Some progress has been made since major study groups recommended to skip an upper age limit within the eligibility criteria of randomized trials. This has already resulted in an increased rate of protocols allowing the inclusion of older patients. However, real inclusion of those patients is far from being adequate. , Several patient-related barriers for study participation of elderly patients have been suggested: poor performance status, number of comorbidities, need for emergent therapy, lack of social support, reluctance to travel or difficulties due to travel distance (especially regarding fractionated RT), unwillingness to participate in trials per se, unability to manage the intensified follow-up procedures. While some of these factors may truly prevent patients from entering trials or from being eligible, there is no reliable evidence of a distinctly different attitude of elderly patients towards study participation. In fact, physicians recommendations and bias may play a more important role. Restrictive inclusion criteria regarding organ function, comorbidity and performance status and/or complex trial designs with multiple endpoints and/or additional preclinical research may further reduce trial participation of elderly patients. Moreover, most trials rarely address endpoints of particular interest for older adults such as preservation of physical or cognitive function. , While there is clearly a need to generate more high-level evidence regarding the optimal treatment for elderly patients, the way to do so is challenging and debatable. One option would be to establish general rules for trials to ensure enrolling of a particular percentage of older patients. However, this may lengthen the trial period or increase the number of prematurely stopped trials if accrual in the elderly age group is slower than in younger patients. Even if achieved, subgroup analyses according to age will often lack statistical power for definitive conclusions. Another option is the design of trials specifically addressing elderly patients. This approach would allow adaptions of the study design: inclusion of geriatric assessments and recording of parameters related to frailty, but less complex trial procedures and more adequate endpoints to increase the willingness to participate. Several studies suggest that age-specific trials will result in the inclusion of much older patients, many of whom otherwise will not have entered a clinical trial. , Moreover, tailored treatment regimens based on combinations of age, performance status, comorbidity and results of geriatric assessments could be evaluated. However, caution is mandatory in exploring tailored treatment approaches: age-specific trials should aim at defining an optimal treatment rather than simply testing less aggressive interventions. Otherwise, undertreatment may result in suboptimal outcomes. Consensus recommendations for the design of therapeutic trials in older and frail patients have been already published. In summary, RT plays an important role in the treatment of elderly patients because of its noninvasive nature. In several situations, RT can replace surgical interventions with lower risks for severe complications but similar outcome. Most treatments using RT as the sole modality are well tolerated by elderly patients. If RT is embedded into multimodal treatment approaches (for example combined with simultaneous chemotherapy or as neoadjuvant/adjuvant treatment), tolerability of the overall treatment is mainly dependent on the systemic or surgical therapy. Nevertheless, the (limited) available data suggest that standard treatments including RT are feasible and well tolerated in a large proportion of elderly patient with similar outcomes to younger ones. Therefore, age alone should not be a decisive factor with regard to treatment selection. Geriatric assessments or screening might be used for patient selection. However, the value of the available tests is limited with regard to predictability of specific RT side effects. Care should be taken to use modern radiation techniques, which clearly have reduced toxicity rates. Hypofractionated regimens might be preferred due to a reduction in socioeconomic burden for the patients. Future research should focus on the development of specific assessment tools for RT and improved representation of elderly patients in randomized trials. Moreover, specific trials addressing the optimal treatment strategy for elderly patients including endpoints of particular interest for elderly should be developed and conducted.
Missed Opportunities in Geriatric Oncology Research	bcb7f70b-f395-4ba3-a0f3-46d14c29b1f9	10166148	Internal Medicine[mh]
Supporting Biomarker-Driven Therapies in Oncology: A Genomic Testing Cost Calculator	b42f355e-1366-404a-88b3-60b8e121e968	10166172	Internal Medicine[mh]	Rapid progress in identifying oncogenic driver mutations, along with advances in molecular diagnostics, has paved the way for precision oncology, contributing to growing opportunities to develop new therapies targeted against “clinically actionable” genomic alterations (eg, trastuzumab for HER2-positive breast cancer, EGFR inhibitors for EGFR mutation-positive non-small cell lung cancer [NSCLC], and Philadelphia chromosome [BCR-ABL fusion] in chronic myelogenous leukemia). The introduction of tumor-agnostic therapies that target genomic driver alterations independent of histology (eg, NTRK gene fusions, microsatellite instability-high [MSI-H]/deficient mismatch repair [dMMR], tumor mutational burden-high [TMB-H]) has expanded the number of vulnerable tumors with genomic targets. Improving patient outcomes has compelled widespread adoption of precision oncology, including some real-world studies that suggest improvement in survival with genomically matched vs. unmatched therapies. The growing compendium of genomic biomarkers has led to guideline recommendations regarding biomarker-guided diagnostics and treatment by the European Society for Medical Oncology (ESMO) and the American Society of Clinical Oncology, including recommendations on sequencing for approved biomarkers in advanced/metastatic cancers. , Next-generation sequencing (NGS) is a high-throughput DNA sequencing technology that offers the advantage of simultaneous analysis of multiple targets from a single-tissue sample, providing comprehensive genomic profiles. It is the only method for identifying multigene molecular signatures (eg, TMB, homologous recombination deficiency ). With the expected advances in genomic science, targeted panel NGS is poised to become the preferred approach for optimizing time to correct diagnosis and treatment. In 2020, ESMO’s Precision Medicine Working Group recommended the use of NGS for lung adenocarcinomas and prostate cancers and its consideration for colorectal carcinoma (CRC), cholangiocarcinoma, and ovarian cancers, but not squamous cell lung, breast, gastric, pancreatic, or liver cancers. Implementation of routine NGS testing has been hindered by the lack of harmonization of clinical infrastructure and insufficient guidance and clinical standardization, while entangled with challenges to equitable reimbursement and the lack of value assessment processes. For many countries, the lack of investment in infrastructure and inadequate reimbursement have hampered its access. Although NGS has demonstrated better cost-effectiveness than single-gene testing (SGT) in NSCLC and CRC, , data from other tumors are lacking. For analyses to be meaningful, they must account for factors that affect both costs and probability of a correct diagnosis in a given laboratory, healthcare system, or region. Analyses should be specific to: (1) the tumor, (2) prices and test performance characteristics of selected tests, and (3) a specific set of genomic alterations whose prevalence may vary across regions or treatment settings. To ensure that economic data are appropriately tailored, we developed a novel metric—cost per correctly identified patient (CCIP)—and tested it in a newly developed genomic testing cost calculator that enables stakeholders (eg, clinicians, pathologists, and pathology advisory groups) to compare the cost of targeted panel NGS with the standard practice of sequential SGT in achieving an accurate diagnosis of a patient’s genomic alterations. To evaluate the clinical utility of this approach—defined as the net benefit to patients and health systems with regard to clinical outcomes, patient access, and shared decision-making , —we evaluated the applicability of the calculator based on ESMO–issued NGS recommendations in 2020 for approved targeted therapies. Targeted Literature Review The scope of the calculator was determined based on ESMO guidelines for selecting tumor types (advanced/metastatic non-squamous NSCLC, squamous NSCLC, breast cancer [mBC], metastatic colorectal carcinoma [mCRC], prostate, gastric [mGC], pancreatic ductal [PDAC], hepatocellular cancers, and cholangiocarcinoma) and genomic alterations of the ESMO Scale for Clinical Actionability of molecular Targets (ESCAT) Tier 1 (ie, genomic alterations for which there is an approved targeted therapy). Sensitivity and specificity values were identified using a targeted PubMed search and supplemented by Google searches of gray scientific literature (through December 2021). Relevant citations were reviewed for outcomes relating to sensitivity/specificity of relevant diagnostic tests (fluorescence or chromogenic in situ hybridization [FISH/CISH], PCR, Sanger, small/large-targeted panel NGS, and quantitative PCR [qPCR]) for 34 genetic targets (ESCAT I & II genomic alterations). Raw numbers for each true positive (TP), false positive (FP), true negative (TN), and false negative (FN) were extracted from the literature search and summed to calculate a single sensitivity and specificity measure for each SGT: IHC, PCR/qPCR, FISH, and NGS. Sensitivity was defined as probability of true positive, and specificity was defined as probability of a true negative, with probability conditioned on being truly positive or truly negative, respectively. Prevalence data were collected from Mosele et al. except for that of NTRK fusions and MSI-H. For NTRK , prevalence data were collected from Forsythe et al. a meta–analysis that synthesized all prevalence data on NTRK derived from a systematic literature review. It also included uncertainty estimates against each prevalence estimate, thus, is believed to provide a more robust estimate. Bonneville et al. was used for prevalence rates of MSI-H. Normanno et al. was used for price estimates for SGTs. Prevalence data are shown in . Cost calculations were made for a “base case” scenario using the published prices for each test, with “cost” defined as direct cost of NGS or sequential SGT, and “price” referring to published price estimates for each test. Given that prices can vary between different countries and health systems, calculations were also made using a range of prices that spanned the base case price for each test. Calculation of Sensitivity and Specificity Data for sensitivity and specificity, such as TP, FP, TN, and FN, were extracted from eligible papers. Summed TP, FP, TN, and FN values were used to calculate an aggregate sensitivity and specificity parameter for each SGT and NGS. In cases where the first genetic test was a screening test, followed by a confirmatory test, we used a serial testing approach proposed by Parikh et al using the following formulas: S e n s i t i v i t y [ S G T 1 ] ∗ S e n s i t i v i t y [ S G T 2 ] = S e n s i t i v i t y S p e c i f i c i t y [ S G T 1 ] + S p e c i f i c i t y [ 1 − S G T 1 ] ∗ S p e c i f i c i t y [ S G T 2 ] = S p e c i f i c i t y shows the sequential SGTs used in this study by tumor type. Definitions and Equations Assuming that the objective is to maximize the overall predictive accuracy of a test, it was important to identify a metric that accounts for both positive predictive value (PPV) and negative predictive value (NPV) in a target population. When comparing tests, those associated with higher false-positive and false-negative rates would be “penalized,” while those associated with lower false-positive and false-negative rates would be considered as having better value. These considerations led to use of a metric based on number needed to predict (NNP: number of patients that need to be examined within the patient population in order to correctly predict the diagnosis of 1 person ), which was monetarized to create the metric of CCIP. The input data required to calculate NNP include the following: test sensitivity, specificity, and prevalence/yield in a patient population. Prices can be customized or adapted to a particular lab, healthcare system, or country. Predictive summary index (PSI) is a metric for measuring test performance for persons who test positive or negative, defined as a total net gain in certainty from a diagnostic test which may be of interest to clinicians, patients, and policy makers/economists. Comparisons of SGT and NGS were conducted using CCIP such that lower values reflect lower total costs to achieve one correctly identified patient. The following equations were used: C C I P = N N P c o s t p e r p a t i e n t C o s t p e r p a t i e n t = N N P ∗ c u m u l a t i v e t e s t c o s t a c r o s s g e n o m i c a l t e r a t i o n b y t u m o r t y p e D i a g n o s t i c y i e l d = S u m o f F P + T P a c r o s s g e n o m i c a l t e r a t i o n / t o t a l t e s t ( N ) N N P = 1 P S I P S I = P P V + N P V 1 P P V = t r u e - p o s i t i v e / ( t r u e - p o s i t i v e + f a l s e - p o s i t i v e ) N P V = t r u e - n e g a t i v e / ( t r u e - n e g a t i v e + f a l s e - n e g a t i v e ) Calculations of NNP, CCIP, and PSI We ran a simulation of 1000 lab tests ( N ). Starting with the most prevalent gene alteration, we generated an algorithm, using the prevalence, recommended test, and its associated sensitivity and specificity. To align with standard laboratory workflow, if the first test was a screening test (eg, IHC), a second (confirmatory) test (eg, PCR/qPCR) was performed on individuals with a positive test result. The matrices were readjusted using the sensitivity and specificity of the serial testing approach. While evaluating each gene alteration sequentially in order of highest to lowest prevalence, all positives (TP + FP) were subtracted from N (total lab tests), and the above procedure was repeated for individuals with a negative test result until the list of ESCAT 1 category genomic alterations was exhausted. A tumor–specific algorithm was then generated to calculate PPV, NPV, PSI, NNP, and consequently, CCIP. CCIP estimates are deterministic; smaller or larger lab cohort sizes will only impact estimates of uncertainty (95% CIs) and can be easily tailored within the model. The diagnostic yield for each tumor was calculated as the sum of positives (TP + FP) over the total number of lab tests ( N ). The same process was repeated for NGS, but without repeating procedures used for sequential SGT. The prevalence across the ESCAT 1 category was summed, and using the relevant sensitivity and specificity, an algorithm was similarly created to calculate the same parameters as SGT. Parameter uncertainty was assessed by conducting a probabilistic sensitivity analysis (PSA), where probability distributions were used to reflect individual parameter uncertainty and analyzed using 1000 Monte Carlo simulations. For model inputs that varied in PSA, data for sensitivity and specificity were varied using a random draw from a binomial distribution. With each iteration, sensitivity and specificity were calculated. Test price was not varied in the PSA because it is expected to be a lab–specific, fixed parameter with a user-modifiable input. For NTRK prevalence, we calculated a standard error using the reported CI in Forsythe et al. and a random draw from a beta distribution was used in the PSA. For other prevalence proportions, we assumed 20% of the mean as a measure of standard error and repeated the above procedure. PSA results were used to generate 95% uncertainty intervals by calculating the 2.5th and 97.5th percentiles across the PSA iterations. The scope of the calculator was determined based on ESMO guidelines for selecting tumor types (advanced/metastatic non-squamous NSCLC, squamous NSCLC, breast cancer [mBC], metastatic colorectal carcinoma [mCRC], prostate, gastric [mGC], pancreatic ductal [PDAC], hepatocellular cancers, and cholangiocarcinoma) and genomic alterations of the ESMO Scale for Clinical Actionability of molecular Targets (ESCAT) Tier 1 (ie, genomic alterations for which there is an approved targeted therapy). Sensitivity and specificity values were identified using a targeted PubMed search and supplemented by Google searches of gray scientific literature (through December 2021). Relevant citations were reviewed for outcomes relating to sensitivity/specificity of relevant diagnostic tests (fluorescence or chromogenic in situ hybridization [FISH/CISH], PCR, Sanger, small/large-targeted panel NGS, and quantitative PCR [qPCR]) for 34 genetic targets (ESCAT I & II genomic alterations). Raw numbers for each true positive (TP), false positive (FP), true negative (TN), and false negative (FN) were extracted from the literature search and summed to calculate a single sensitivity and specificity measure for each SGT: IHC, PCR/qPCR, FISH, and NGS. Sensitivity was defined as probability of true positive, and specificity was defined as probability of a true negative, with probability conditioned on being truly positive or truly negative, respectively. Prevalence data were collected from Mosele et al. except for that of NTRK fusions and MSI-H. For NTRK , prevalence data were collected from Forsythe et al. a meta–analysis that synthesized all prevalence data on NTRK derived from a systematic literature review. It also included uncertainty estimates against each prevalence estimate, thus, is believed to provide a more robust estimate. Bonneville et al. was used for prevalence rates of MSI-H. Normanno et al. was used for price estimates for SGTs. Prevalence data are shown in . Cost calculations were made for a “base case” scenario using the published prices for each test, with “cost” defined as direct cost of NGS or sequential SGT, and “price” referring to published price estimates for each test. Given that prices can vary between different countries and health systems, calculations were also made using a range of prices that spanned the base case price for each test. Data for sensitivity and specificity, such as TP, FP, TN, and FN, were extracted from eligible papers. Summed TP, FP, TN, and FN values were used to calculate an aggregate sensitivity and specificity parameter for each SGT and NGS. In cases where the first genetic test was a screening test, followed by a confirmatory test, we used a serial testing approach proposed by Parikh et al using the following formulas: S e n s i t i v i t y [ S G T 1 ] ∗ S e n s i t i v i t y [ S G T 2 ] = S e n s i t i v i t y S p e c i f i c i t y [ S G T 1 ] + S p e c i f i c i t y [ 1 − S G T 1 ] ∗ S p e c i f i c i t y [ S G T 2 ] = S p e c i f i c i t y shows the sequential SGTs used in this study by tumor type. Assuming that the objective is to maximize the overall predictive accuracy of a test, it was important to identify a metric that accounts for both positive predictive value (PPV) and negative predictive value (NPV) in a target population. When comparing tests, those associated with higher false-positive and false-negative rates would be “penalized,” while those associated with lower false-positive and false-negative rates would be considered as having better value. These considerations led to use of a metric based on number needed to predict (NNP: number of patients that need to be examined within the patient population in order to correctly predict the diagnosis of 1 person ), which was monetarized to create the metric of CCIP. The input data required to calculate NNP include the following: test sensitivity, specificity, and prevalence/yield in a patient population. Prices can be customized or adapted to a particular lab, healthcare system, or country. Predictive summary index (PSI) is a metric for measuring test performance for persons who test positive or negative, defined as a total net gain in certainty from a diagnostic test which may be of interest to clinicians, patients, and policy makers/economists. Comparisons of SGT and NGS were conducted using CCIP such that lower values reflect lower total costs to achieve one correctly identified patient. The following equations were used: C C I P = N N P c o s t p e r p a t i e n t C o s t p e r p a t i e n t = N N P ∗ c u m u l a t i v e t e s t c o s t a c r o s s g e n o m i c a l t e r a t i o n b y t u m o r t y p e D i a g n o s t i c y i e l d = S u m o f F P + T P a c r o s s g e n o m i c a l t e r a t i o n / t o t a l t e s t ( N ) N N P = 1 P S I P S I = P P V + N P V 1 P P V = t r u e - p o s i t i v e / ( t r u e - p o s i t i v e + f a l s e - p o s i t i v e ) N P V = t r u e - n e g a t i v e / ( t r u e - n e g a t i v e + f a l s e - n e g a t i v e ) We ran a simulation of 1000 lab tests ( N ). Starting with the most prevalent gene alteration, we generated an algorithm, using the prevalence, recommended test, and its associated sensitivity and specificity. To align with standard laboratory workflow, if the first test was a screening test (eg, IHC), a second (confirmatory) test (eg, PCR/qPCR) was performed on individuals with a positive test result. The matrices were readjusted using the sensitivity and specificity of the serial testing approach. While evaluating each gene alteration sequentially in order of highest to lowest prevalence, all positives (TP + FP) were subtracted from N (total lab tests), and the above procedure was repeated for individuals with a negative test result until the list of ESCAT 1 category genomic alterations was exhausted. A tumor–specific algorithm was then generated to calculate PPV, NPV, PSI, NNP, and consequently, CCIP. CCIP estimates are deterministic; smaller or larger lab cohort sizes will only impact estimates of uncertainty (95% CIs) and can be easily tailored within the model. The diagnostic yield for each tumor was calculated as the sum of positives (TP + FP) over the total number of lab tests ( N ). The same process was repeated for NGS, but without repeating procedures used for sequential SGT. The prevalence across the ESCAT 1 category was summed, and using the relevant sensitivity and specificity, an algorithm was similarly created to calculate the same parameters as SGT. Parameter uncertainty was assessed by conducting a probabilistic sensitivity analysis (PSA), where probability distributions were used to reflect individual parameter uncertainty and analyzed using 1000 Monte Carlo simulations. For model inputs that varied in PSA, data for sensitivity and specificity were varied using a random draw from a binomial distribution. With each iteration, sensitivity and specificity were calculated. Test price was not varied in the PSA because it is expected to be a lab–specific, fixed parameter with a user-modifiable input. For NTRK prevalence, we calculated a standard error using the reported CI in Forsythe et al. and a random draw from a beta distribution was used in the PSA. For other prevalence proportions, we assumed 20% of the mean as a measure of standard error and repeated the above procedure. PSA results were used to generate 95% uncertainty intervals by calculating the 2.5th and 97.5th percentiles across the PSA iterations. Test Costs, Sensitivity, and Specificity The objective for the genomic testing cost calculator was to estimate the CCIP using NGS vs. sequential SGT. A targeted literature search was conducted to identify publications on sensitivity and specificity of IHC, PCR/qPCR, FISH, and NGS . Normanno et al was used for estimating the price of each test. For IHC, sensitivity was estimated at 92.54%, and specificity at 86.45%. The price of IHC screening was estimated at €242 per test. Alternate scenarios were calculated at €200, €300, and €350. When IHC was followed by PCR, test sensitivity was 86.26% and specificity was 99.49%. When IHC was followed by FISH, test sensitivity was 82.90% and specificity was 99.98%. For FISH, sensitivity was estimated at 89.58% and specificity at 97.67%. The price of FISH was estimated at €664 per test. Alternate scenarios were calculated at €600, €700, and €750. For PCR/qPCR, sensitivity was estimated at 93.41% and specificity at 94.79%. The price of PCR/qPCR was estimated at €218 per test. Alternate scenarios were calculated at €200, €250, and €300. For NGS, sensitivity was estimated at 84.98% and specificity at 98.50%. Estimates from studies suggest much higher sensitivity rates for NGS. The effect of assuming higher sensitivity for NGS tests would improve CCIP relative to SGT. The price of NGS testing was estimated at €593 for an up to 50-gene panel. Alternate scenarios were calculated at €500, €800, and €1000. Impact of Test Specificity and Prevalence on Number Needed to Predict a Correctly Identified Patient shows the impact of test specificity, sensitivity, and prevalence of a given genomic alteration on the NNP. At the highest sensitivity and specificity levels for a given test, the NNP remains low, independent of mutational prevalence. However, as specificity of a test decreases, the NNP becomes inversely proportional to the mutational prevalence, such that NNP is higher for low prevalence alterations and vice versa. For tests with low specificity, NNP is also inversely proportional to the test sensitivity, such that at a given mutational prevalence, NNP increases as test sensitivity decreases. At low mutational prevalence (eg, 1%), NNP is high, while at high prevalence (eg, 20%), NNP remains low, regardless of sensitivity or specificity of the test. CCIP Using Sequential SGT vs. NGS Using the base case estimates, a more favorable CCIP was observed using NGS vs. SGT, for advanced non-squamous NSCLC (€1983 for sequential SGT vs. €658 for NGS), mBC (€1202 vs. €695), mCRC (€1226 vs. €659), mGC (€1202 vs. €695), and cholangiocarcinoma (€1661 vs. €667) . Cost differences between sequential SGT and NGS were greatest in non–squamous NSCLC, which had the highest number of clinically actionable mutations. Lower costs were also observed at base case for advanced squamous NSCLC (€35 259 vs. €21 637), hepatocellular carcinoma (€4596 vs. €1825), and PDAC (€8190 vs. €3153), but with overlap in CIs . The high relative costs for squamous NSCLC for either sequential SGT or NGS can be attributed to the low prevalence (0.17%, ) of NTRK gene fusions found in this cancer type, which drives the NNP higher . CCIP at base case was lower for sequential SGT than NGS in advanced prostate cancer (€540 vs. €2340), for which MSI-H was the only actionable marker. The cost difference for this cancer type would become negligible if the diagnostic yield was increased, for example, if both ESCAT 1 and 2 were to be included. When genomic alterations from the ESCAT 2 category are included, the CCIP with NGS further decreases, such that it becomes lower than sequential SGT for advanced prostate cancer (data not shown). Alternate scenarios considering a range of prices showed the same general trends across all scenarios, with incremental changes in the magnitude of difference between NGS and SGT related to the incremental change in the price estimate of each test . For example, when NGS price was increased to €1000, CCIP for squamous NSCLC became higher with NGS than sequential SGT (€36 487 vs. €35 259, respectively) and differences between sequential SGT and NGS became negligible (~€31) for metastatic breast cancer. illustrates hypothetical examples of cost differentials at various price estimates for each test, compared with the base case scenario. The objective for the genomic testing cost calculator was to estimate the CCIP using NGS vs. sequential SGT. A targeted literature search was conducted to identify publications on sensitivity and specificity of IHC, PCR/qPCR, FISH, and NGS . Normanno et al was used for estimating the price of each test. For IHC, sensitivity was estimated at 92.54%, and specificity at 86.45%. The price of IHC screening was estimated at €242 per test. Alternate scenarios were calculated at €200, €300, and €350. When IHC was followed by PCR, test sensitivity was 86.26% and specificity was 99.49%. When IHC was followed by FISH, test sensitivity was 82.90% and specificity was 99.98%. For FISH, sensitivity was estimated at 89.58% and specificity at 97.67%. The price of FISH was estimated at €664 per test. Alternate scenarios were calculated at €600, €700, and €750. For PCR/qPCR, sensitivity was estimated at 93.41% and specificity at 94.79%. The price of PCR/qPCR was estimated at €218 per test. Alternate scenarios were calculated at €200, €250, and €300. For NGS, sensitivity was estimated at 84.98% and specificity at 98.50%. Estimates from studies suggest much higher sensitivity rates for NGS. The effect of assuming higher sensitivity for NGS tests would improve CCIP relative to SGT. The price of NGS testing was estimated at €593 for an up to 50-gene panel. Alternate scenarios were calculated at €500, €800, and €1000. shows the impact of test specificity, sensitivity, and prevalence of a given genomic alteration on the NNP. At the highest sensitivity and specificity levels for a given test, the NNP remains low, independent of mutational prevalence. However, as specificity of a test decreases, the NNP becomes inversely proportional to the mutational prevalence, such that NNP is higher for low prevalence alterations and vice versa. For tests with low specificity, NNP is also inversely proportional to the test sensitivity, such that at a given mutational prevalence, NNP increases as test sensitivity decreases. At low mutational prevalence (eg, 1%), NNP is high, while at high prevalence (eg, 20%), NNP remains low, regardless of sensitivity or specificity of the test. Using the base case estimates, a more favorable CCIP was observed using NGS vs. SGT, for advanced non-squamous NSCLC (€1983 for sequential SGT vs. €658 for NGS), mBC (€1202 vs. €695), mCRC (€1226 vs. €659), mGC (€1202 vs. €695), and cholangiocarcinoma (€1661 vs. €667) . Cost differences between sequential SGT and NGS were greatest in non–squamous NSCLC, which had the highest number of clinically actionable mutations. Lower costs were also observed at base case for advanced squamous NSCLC (€35 259 vs. €21 637), hepatocellular carcinoma (€4596 vs. €1825), and PDAC (€8190 vs. €3153), but with overlap in CIs . The high relative costs for squamous NSCLC for either sequential SGT or NGS can be attributed to the low prevalence (0.17%, ) of NTRK gene fusions found in this cancer type, which drives the NNP higher . CCIP at base case was lower for sequential SGT than NGS in advanced prostate cancer (€540 vs. €2340), for which MSI-H was the only actionable marker. The cost difference for this cancer type would become negligible if the diagnostic yield was increased, for example, if both ESCAT 1 and 2 were to be included. When genomic alterations from the ESCAT 2 category are included, the CCIP with NGS further decreases, such that it becomes lower than sequential SGT for advanced prostate cancer (data not shown). Alternate scenarios considering a range of prices showed the same general trends across all scenarios, with incremental changes in the magnitude of difference between NGS and SGT related to the incremental change in the price estimate of each test . For example, when NGS price was increased to €1000, CCIP for squamous NSCLC became higher with NGS than sequential SGT (€36 487 vs. €35 259, respectively) and differences between sequential SGT and NGS became negligible (~€31) for metastatic breast cancer. illustrates hypothetical examples of cost differentials at various price estimates for each test, compared with the base case scenario. The need for NGS or other multiplex diagnostics in routine cancer care is widely recognized, but impeded by concerns about cost. We propose the use of a simple metric and analytic framework to inform national policies regarding genomic testing strategies, illustrating the methodology in a comparison of a sequential SGT algorithm vs. NGS. To demonstrate use of the calculator, we applied NGS recommendations from ESCAT 1 to compare sequential SGT vs. NGS and showed that CCIP favored NGS in advanced/metastatic non-squamous NSCLC, mBC, mCRC, mGC, and cholangiocarcinoma. CCIP also favored NGS for advanced squamous NSCLC, PDAC, and hepatocellular carcinoma, but with overlapping CIs. For prostate cancer, CCIP favored sequential SGT over NGS. Development of the genomic testing cost calculator was based on previously reported prevalence and cost. , Mosele et al. was selected to provide a EU perspective on NGS, based on ESMO recommendations. Forsythe et al. was a systemic literature review and meta–analysis that provided the most robust estimate of NTRK prevalence to date. For price estimates, Normanno et al. was chosen for providing representative European estimates that would facilitate comparisons for this study, but estimates for comparisons may be adapted by the user to reflect individual cases or support lab practices. We developed the calculator to allow full cost comparisons involving initial setup and maintenance costs for a given diagnostic test (eg, purchasing of equipment, laboratory and IT infrastructure, personnel, validation, etc), as well as the ongoing costs (eg, sequencing kits and flow cells) of validated, genomic tests. Of note, labor costs, initial infrastructure, and ongoing genomic testing costs vary significantly across countries, and the calculator allows individual labs and hospitals to use comprehensive local costs as input source. As such, these results compare the costs of validated diagnostic tests with an approach consistent with prior cost-effectiveness analyses of genomic tests. To illustrate this point, CCIPs were calculated using a range of published prices for each test. While some variability was observed at the different SGT cost estimates, the same general trends were seen as base case. At the higher cost estimates for NGS, the cost differential from SGTs became smaller, and vice versa. As testing strategies have shifted in favor of NGS, its perceived high cost has raised concerns about its value and sustainability, particularly its budget impact compared to SGT. Our results are largely consistent with previous cost-effectiveness studies on NGS and SGT. In a study investigating costs of different diagnostic approaches in 3 Italian hospitals, Pruneri et al. applied several different scenarios (eg, current testing pathway, minimum set per local guidelines, and anticipated future mutational load) for patients with advanced NSCLC or mCRC and found that the NGS-based strategy was cost-saving in all scenarios except one, where the additional cost for NGS was modest. Savings per patient were higher in scenarios where NGS encompassed a more comprehensive set of mutations, attributed both to the volume of detectable alterations via NGS and to the reduction in personnel time needed. In another analysis from the perspective of the US Centers for Medicare & Medicaid Services (CMS) or US commercial payers, an economic impact model showed that, compared with SGT, NGS for metastatic NSCLC was associated with cost savings for both CMS and commercial payers, while also providing shorter time-to-test results by 2-3 weeks. In contrast, despite the importance in advancing national policies regarding optimal use of NGS in routine cancer care, some cost-effectiveness analyses for genomic testing approaches have been limited and/or lacked gravitas to decision makers because of the nature of rapidly changing key parameters, such as test prices, test performance characteristics (eg, sensitivity and specificity), and the number and type of clinically actionable biomarkers. For example, a 2020 analysis in Brazil of cost-effectiveness of SGT vs. NGS for EGFR , ALK , and ROS1 in NSCLC concluded that NGS-facilitated identification was not cost-effective due to an incremental $3479 per correct case detected; however, the limited number of biomarkers included in that study was insufficient to accurately estimate the cost differential between diagnostic methods. On the other hand, a study in Singapore found that use of a targeted NGS panel for DNA alterations (29 selected genes including BRAF , EGFR , ERBB2 , and TP53 ) and an RNA fusion panel ( ALK , ROS1 , and RET ) resulted in identification of an additional 1% of patients with actionable alterations, without significant added costs. The development of our genomic testing cost calculator and findings from its initial application have important implications for the oncology community, not only in terms of economic value but also for informing policy and how physicians approach diagnostics, both in clinical practice and in clinical trial design. Within the current clinical landscape, where there is a drive to develop companion diagnostics in parallel with clinical trials, offering a model that can estimate cost differentials in trials will facilitate adoption and access to both drug and diagnostic approaches following marketing authorization. As illustrated in the cancer types selected for our analysis, NGS provided favorable CCIP for some but not all cancers, with the differences related to the number of biomarkers tested, their prevalence within a given cancer type, and the sensitivity and selectivity of each test. In the US, NGS testing has increased to 48% for advanced NSCLC, but remains <20% for mCRC, mBC, and advanced melanoma. For those who remain skeptical about value of NGS, the genomic testing cost calculator makes comparative cost calculations accessible to clinicians, guideline developers, and other decision makers who otherwise may not have specialized health economics training. Pruneri et al. has suggested that the increased adoption of NGS over SGT can lead to cost reduction, particularly at a given threshold of patient numbers or molecular alterations. Furthermore, the anticipated growing number of molecular alterations will also increase the potential savings generated by NGS. Indeed, prevalence data collected from Mosele et al. in this study are conservative and perhaps outdated, given the growing availability of targeted therapies and potential for increased yield via NGS. Recent years have seen a growing number of clinically actionable biomarkers: 58% of 62 cancer drugs approved by FDA and 59% of 46 cancer drugs authorized by EMA in the last 5 years have been granted pharmacogenomic labels. This rapid expansion of biomarker-guided therapies is apparent even in the short timeframe from when the ESCAT rankings were published in August 2020. For example, at time of the ESMO NGS publication, there were 5 genomic alterations ( EGFR , ALK , MET , BRAF V600E , and NTRK ) for which targeted therapies had been approved for NSCLC by FDA and/or EMA. Over the last 2 years, at least 11 new therapies have been approved that depend on testing of genomic/molecular alterations, 7 of which target ESCAT 1 markers, 3 which target non-ESCAT 1 markers, and 1 indicated for NSCLC without certain genomic alterations. , Given the pace and volume of emerging new therapies that target genomic alterations, we believe the CCIP differences between NGS and single SGTs presented in our analysis are conservative and will increase in favor of NGS over time. Along with the expansion in biomarker-specific labels, the treatment landscape has shifted towards use of precision oncology agents that target specific actionable genomic alterations operating in many cancer types (“tumor agnostic”; eg, tumor–agnostic therapies for NTRK gene fusions, MSI-H/dMMR, TMB-H, BRAF V600E ) rather than the classic one cancer type—one alteration—one drug approach. Furthermore, the number of late stage, multi–indication trials has increased, along with the potential for increased use of pan–tumor therapies. The interest in genomic–based diagnostics is also evident in new initiatives, such as Europe’s Beating Cancer Plan—which recently invested €4 billion in the Knowledge Centre on Cancer, Genomics for Public Health, and Partnership on Personalized Medicine, among other groups. Indeed, as the number of clinically actionable biomarkers continues to grow, evolving guidelines and new initiatives have kept pace to accelerate genomics for research, prevention, diagnostics, and treatment. In a consensus report published in 2020 by a panel of international experts from Europe, US, and Asia, NTRK fusion testing was suggested for all patients with advanced solid tumors without other known actionable and driver gene mutations, with testing to occur both before and during standard treatment. MSI/MMR testing was recommended for patients with advanced solid tumors with high incidence of MSI-H/dMMR, and weighing the economic considerations of testing with potential clinical benefit, it was suggested that advanced tumors with low incidence of MSI/dMMR should also be considered to inform treatment decisions. Additional guidance is anticipated following the recent FDA approval of dabrafenib plus trametinib for unresectable or metastatic solid tumors with BRAF V600E mutation. NGS testing has been described as having potential to become the standard-of-care for determining eligibility for treatment with PD-(L)1 inhibitors and for assessing tumor responses. , A limitation of the calculator is the assumption that it treats false-positives and false-negatives equally, consequences of each may be different and may include suboptimal or incorrectly assigned treatments to patients, resulting in different outcomes and costs over time. By combining values for TP/FP and TN/FN to calculate a single value for sensitivity and specificity parameters for each SGT, neither differences by tumor type/gene alteration nor uncertainty in PSA are taken into account. While the NNP metric takes these rates into consideration, subsequent treatment decisions and their impact on patients were considered out of scope for the calculation and not taken into account. Further research on such impact is warranted. Evolving treatment guidelines reflect an unprecedented expansion in precision oncology. Recent approvals of molecularly targeted therapies and expanded use of basket trials are uncovering genomic signatures that can inform treatment decisions and improve prediction of outcomes. This rapid pace of change points toward a new era where NGS will enable more efficient oncology testing with demonstrated value than the multiplicity of tests required for multiple single-gene alterations. Indeed, while the genomic testing cost calculator described herein provides comparative benchmarks from different diagnostic methodologies, it can be expanded or tailored to further substantiate the need for adopting NGS or other multiplex diagnostic advances to optimize individual benefits of biomarker-driven, tumor-agnostic precision oncology. Investing in a transition to NGS offers the opportunity to optimize personalized patient care via early identification of efficacious matched therapies and achieve improvements in patient outcomes and quality of life, and cost offsets through minimizing sequential SGTs and ineffective therapeutic regimens and other treatments. oyad005_suppl_Supplementary_Material Click here for additional data file.
NOHA: A Promising Biomarker for Determining Estrogen Receptor Status Among Patients With Breast Cancer in Resource-Constrained Settings	ce1fcf05-f33a-45b3-b02e-80d6005ab7f2	10166386	Anatomy[mh]	Breast cancer is the leading cause of cancer morbidity and mortality in women globally with nearly 2.3 million cases and 700k deaths yearly. Mortality rates are substantially higher among women in low-and middle-income countries (LMICs), , largely due to advanced stages at diagnosis, reflecting diagnostic delays because of limited care access. Mortality-to-incidence rates in sub-Saharan Africa (SSA) reach 0.55 versus 0.16 in North America. , Breast cancer incidence rates are projected to increase in SSA because of shifting risk factors as countries transition economically. In East Africa, annual breast cancer cases are expected to climb from 45.7k in 2020 to 125k in 2040. Advancements in breast cancer control require longitudinal assessment of a nation's disease burden, reflecting accurate diagnoses and collection of prognostic information, including that from a dependable pathology report. CONTEXT Key Objective The overall goal of this work was to pilot test an innovative, US-validated and patented blood-based assay, N w -hydroxy-L-Arginine, to determine estrogen receptor (ER) status among a cohort of patients with newly diagnosed breast cancer from a single Tanzanian cancer center. Knowledge Generated Offers insight into a promising, sensitive, blood-based technology that differentiates ER-negative versus ER-positive breast cancers and uses inexpensive, easy-to-maintain equipment and reagents, suitable for point-of-care use by laboratory personnel in low-and middle-income countries. Relevance N w -hydroxy-L-Arginine holds promise as an attractive and scalable replacement for costly immunohistochemistry-based ER testing, with potential clinical applications extending beyond cancer diagnostics to surveillance, determination of prognosis, and disease monitoring in resource-constrained settings. Despite a significant burden of late-stage breast cancer in SSA, individuals can benefit from surgery, chemotherapy, or endocrine therapy, depending on tumor biology, advancing both quality of life and survival. Included among the 2019 WHO essential medicines list are two oral, easy to administer, and accessible breast cancer hormonal agents: tamoxifen and anastrozole. The offer of surgery or systemic therapy, including hormonal therapy, requires pathological confirmation of disease with optimal treatment dependent on stage and other markers, the most critical of which is estrogen receptor (ER) alpha expression. Although prognostic, progesterone receptor (PR) status is not significantly predictive of response to hormonal agents, and costly human epidermal growth factor receptor 2 (HER2)–directed therapies are not widely available in LMICs. Thus, PR analysis is of limited value when added to ER, and HER2 testing is not warranted when related therapy is cost-prohibitive , in resource-constrained settings such as Tanzania. Fundamentally, ER determination classifies a breast cancer as ER-positive (ER+) versus ER-negative (ER−) and improves the potential of disease control through identification of candidates for cost-efficient and easy-to-access and administer endocrine therapies, promising to improve survival and quality of life. However, there are marked shortages in professional and technical pathology services in LMICs, with many of the lowest pathologist-to-population ratios and associated technical services, such as hormone receptor immunohistochemical (IHC) testing in Africa, including Tanzania. , We have identified N w -hydroxy-L-Arginine (NOHA) as a blood-based marker that is 100% sensitive and specific, respectively, (95% CI, 94.5% to 100% each) in determining ER status in five ethnically and racially diverse US groups (US Utility Patent 10073099). - We also find NOHA to be effective in differentiating ER– tumors by grade and molecular phenotype. NOHA can be measured at a fraction of the cost of traditional IHC and can be run on fresh or dried plasma extracted from a needle prick amount of blood. , In addition, NOHA remains stable in dried plasma for 14 days, at ≤ 42°C, allowing for ease of sample shipment. NOHA promises to be a cost-effective and accessible tool for disease analysis in LMICs. The overall goal of this work was to pilot test and validate the use of this new, US-validated, blood-based assay to determine ER status among a cohort of patients with newly diagnosed breast cancer from a single Tanzanian cancer center. Key Objective The overall goal of this work was to pilot test an innovative, US-validated and patented blood-based assay, N w -hydroxy-L-Arginine, to determine estrogen receptor (ER) status among a cohort of patients with newly diagnosed breast cancer from a single Tanzanian cancer center. Knowledge Generated Offers insight into a promising, sensitive, blood-based technology that differentiates ER-negative versus ER-positive breast cancers and uses inexpensive, easy-to-maintain equipment and reagents, suitable for point-of-care use by laboratory personnel in low-and middle-income countries. Relevance N w -hydroxy-L-Arginine holds promise as an attractive and scalable replacement for costly immunohistochemistry-based ER testing, with potential clinical applications extending beyond cancer diagnostics to surveillance, determination of prognosis, and disease monitoring in resource-constrained settings. The Kilimanjaro Christian Medical Centre (KCMC) and the National Institute of Medical Research institutional review boards approved this work (ie, KCMC #2445 and NIMR lHQ/R.8a/Vol. IX/3249). Written informed consent was required from all participants and was collected at enrollment by the KCMC study coordinator. There were no incentives offered for enrollment. Settings This study was done in partnership with KCMC staff in Moshi, serving the Northern Zone of Tanzania. All NOHA analysis was performed in the Mohan laboratory at the University of New England (UNE; Portland, ME). Study-based histologic and hormone receptor, ER, and PR IHC testing was performed at Maine Medical Center (MMC; Portland, ME). Participants A total of 70 individuals were recruited for this study. Participant enrollment was based on a suspected or proven breast cancer diagnosis, before surgical or medical management. Basic clinical and sociodemographic data from participants were collected by KCMC clinical staff and stored in a secure study database, without identifiers. Sample Collection, Handling, and Processing Following informed consent and before any treatment, a finger prick amount of blood (approximately 25 μL) was collected from all participants onto a Shimatzu Noviplex plasma prep card (West Lafayette, IN). After a 3-minute incubation period, the top layer of the Noviplex card was peeled off; the resultant plasma containing disc was air dried for 15 minutes, individually vacuum sealed, and stored at −80°C before shipment directly from KCMC via DHL international overnight service every 3 months. The plasma card shipment protocol was changed after the loss of 34 samples because of the commercial shipper's package mishandling, resulting in sample denaturation. Subsequently, all plasma cards were secured in individually sealed vacuum bags, wrapped in aluminum foil to protect the sample from atmospheric moisture and photon radiation. Secured samples were hand carried by research or clinical volunteers returning to the United States for ultimate stateside shipment to the Mohan laboratory at 2-3 month intervals. This resulted in a total of 36 Noviplex plasma prep card samples available for stateside NOHA analysis. Surgical specimens from all participants were submitted for routine pathology at KCMC, where five unstained slides were prepared and shipped from KCMC via DHL international overnight service at 3-month intervals for study-based histologic and IHC analysis at MMC. When available, clinical hormone receptor IHC testing was also performed on site at KCMC with results recorded in the secure study database. Study-based hormone receptor IHC testing was performed on shipped unstained tissues slides from study participants, as described by Allison et al, by a single pathologist (coauthor, M.J.) at MMC, following CAP guidelines wherein a tumor was classified as ER+ if ≥ 1% of tumor cells demonstrated nuclear staining. NOHA and IHC assay operators were blinded to each other's results and to patient clinical status. During the study period, among those cases undergoing clinical IHC analysis at KCMC, a tumor was classified as ER+ if at least 10% of cells revealed nuclear staining. Sample Preparation for NOHA Enzyme-Linked Immunosorbent Assay Analysis The dried plasma disk from each enrolled participant was soaked in 100 μL of deionized water for 15 minutes at room temperature (25°C), without stirring. The soaked discs were then placed in 200 μL of extraction buffer (ie, 9:1, acetonitrile-water, v/v), vortexed for 1 minutes, sonicated for 5 minutes, incubated in extraction buffer for 10 minutes, disc discarded, and centrifuged at 14,000 g for 5 minutes. The resulting supernatant was dried in a lyophilizer and reconstituted in 50 μL of water and stored at −80°C, until NOHA measurement by enzyme-linked immunosorbent assay (ELISA) assay. NOHA ELISA Assay We previously validated a competitive ELISA assay for NOHA quantification with a proprietary monoclonal antibody, as a simple yet sensitive alternative method to analytical NOHA detection by LC-MS. The competitive NOHA ELISA assay was used here. All reagents and chemicals were purchased from either Invitrogen (Carlsbad, CA) or Sigma-Aldrich (St Louis, MO), and all supplies were obtained from VWR (Bridgeport, NJ). In brief, BSA-NOHA binding strips were washed at least 3 times with 200 μL of 1× phosphate-buffered saline (PBS). 20 µL of reconstituted dried plasma sample from each research participant (as duplicates) was mixed with 5 μL of NOHA-monoclonal antibody at 5 ng/mL and 75 μL 1× PBS. The resultant mixture was added to each well of the bovine serum albumin–NOHA binding strip and incubated for 1 hour at 25°C. After incubation, well contents were discarded and washed 8 times with 200 μL of 1× PBS, before adding 100 μL of polyclonal horseradish peroxidase conjugate from Abcam (Cambridge, MA), at 1:20,000 dilution (in 1× PBS wash buffer). The wells were incubated for 1 hour, at 25°C, before decanting and washing them 8 times with 200 μL of 1× PBS. 50 μL of tetramethyl-benzidine substrate from Mossbio (Pasadena, MD) was added to each well and incubated for 10 minutes in the dark, before stopping the horseradish peroxidase-tetramethyl benzidine interaction with 50 μL of 0.1 N HCl. The binding strip wells were read for absorbance at 450 nm using a VersaMax Spectrophotometer (Molecular Devices, NH). Absorbance values were plotted on a polynomial second order trendline, and R -squared values were added to the standard curve to assess its confidence in NOHA measurement. Average NOHA value from duplicate assessment of each patient sample was used for IHC comparison. NOHA versus IHC Clinical Utility Assessment As further assessment of the clinical utility of NOHA as a promising replacement for IHC, we examined sample requirements, assay cost, turnaround time, and facility and personnel needs, comparing these with reference standard ER IHC testing in Tanzania. Statistical Analysis Our pilot study assumed that 50% of patients would have ER+ disease on the basis of preliminary published evidence. ‐ Our sample size of a total of 70 patients (35 ER+ and 35 ER−) was predicted to provide 80% power to detect a one standard deviation difference in means between the two groups with an alpha level of .05. Statistical comparisons of participant NOHA levels versus US-determined ER IHC results were performed using Student's t test (GraphPad Prism, Version 8.0, La Jolla, CA). Statistical significance was determined at P < .05. This study was done in partnership with KCMC staff in Moshi, serving the Northern Zone of Tanzania. All NOHA analysis was performed in the Mohan laboratory at the University of New England (UNE; Portland, ME). Study-based histologic and hormone receptor, ER, and PR IHC testing was performed at Maine Medical Center (MMC; Portland, ME). A total of 70 individuals were recruited for this study. Participant enrollment was based on a suspected or proven breast cancer diagnosis, before surgical or medical management. Basic clinical and sociodemographic data from participants were collected by KCMC clinical staff and stored in a secure study database, without identifiers. Following informed consent and before any treatment, a finger prick amount of blood (approximately 25 μL) was collected from all participants onto a Shimatzu Noviplex plasma prep card (West Lafayette, IN). After a 3-minute incubation period, the top layer of the Noviplex card was peeled off; the resultant plasma containing disc was air dried for 15 minutes, individually vacuum sealed, and stored at −80°C before shipment directly from KCMC via DHL international overnight service every 3 months. The plasma card shipment protocol was changed after the loss of 34 samples because of the commercial shipper's package mishandling, resulting in sample denaturation. Subsequently, all plasma cards were secured in individually sealed vacuum bags, wrapped in aluminum foil to protect the sample from atmospheric moisture and photon radiation. Secured samples were hand carried by research or clinical volunteers returning to the United States for ultimate stateside shipment to the Mohan laboratory at 2-3 month intervals. This resulted in a total of 36 Noviplex plasma prep card samples available for stateside NOHA analysis. Surgical specimens from all participants were submitted for routine pathology at KCMC, where five unstained slides were prepared and shipped from KCMC via DHL international overnight service at 3-month intervals for study-based histologic and IHC analysis at MMC. When available, clinical hormone receptor IHC testing was also performed on site at KCMC with results recorded in the secure study database. Study-based hormone receptor IHC testing was performed on shipped unstained tissues slides from study participants, as described by Allison et al, by a single pathologist (coauthor, M.J.) at MMC, following CAP guidelines wherein a tumor was classified as ER+ if ≥ 1% of tumor cells demonstrated nuclear staining. NOHA and IHC assay operators were blinded to each other's results and to patient clinical status. During the study period, among those cases undergoing clinical IHC analysis at KCMC, a tumor was classified as ER+ if at least 10% of cells revealed nuclear staining. The dried plasma disk from each enrolled participant was soaked in 100 μL of deionized water for 15 minutes at room temperature (25°C), without stirring. The soaked discs were then placed in 200 μL of extraction buffer (ie, 9:1, acetonitrile-water, v/v), vortexed for 1 minutes, sonicated for 5 minutes, incubated in extraction buffer for 10 minutes, disc discarded, and centrifuged at 14,000 g for 5 minutes. The resulting supernatant was dried in a lyophilizer and reconstituted in 50 μL of water and stored at −80°C, until NOHA measurement by enzyme-linked immunosorbent assay (ELISA) assay. We previously validated a competitive ELISA assay for NOHA quantification with a proprietary monoclonal antibody, as a simple yet sensitive alternative method to analytical NOHA detection by LC-MS. The competitive NOHA ELISA assay was used here. All reagents and chemicals were purchased from either Invitrogen (Carlsbad, CA) or Sigma-Aldrich (St Louis, MO), and all supplies were obtained from VWR (Bridgeport, NJ). In brief, BSA-NOHA binding strips were washed at least 3 times with 200 μL of 1× phosphate-buffered saline (PBS). 20 µL of reconstituted dried plasma sample from each research participant (as duplicates) was mixed with 5 μL of NOHA-monoclonal antibody at 5 ng/mL and 75 μL 1× PBS. The resultant mixture was added to each well of the bovine serum albumin–NOHA binding strip and incubated for 1 hour at 25°C. After incubation, well contents were discarded and washed 8 times with 200 μL of 1× PBS, before adding 100 μL of polyclonal horseradish peroxidase conjugate from Abcam (Cambridge, MA), at 1:20,000 dilution (in 1× PBS wash buffer). The wells were incubated for 1 hour, at 25°C, before decanting and washing them 8 times with 200 μL of 1× PBS. 50 μL of tetramethyl-benzidine substrate from Mossbio (Pasadena, MD) was added to each well and incubated for 10 minutes in the dark, before stopping the horseradish peroxidase-tetramethyl benzidine interaction with 50 μL of 0.1 N HCl. The binding strip wells were read for absorbance at 450 nm using a VersaMax Spectrophotometer (Molecular Devices, NH). Absorbance values were plotted on a polynomial second order trendline, and R -squared values were added to the standard curve to assess its confidence in NOHA measurement. Average NOHA value from duplicate assessment of each patient sample was used for IHC comparison. As further assessment of the clinical utility of NOHA as a promising replacement for IHC, we examined sample requirements, assay cost, turnaround time, and facility and personnel needs, comparing these with reference standard ER IHC testing in Tanzania. Our pilot study assumed that 50% of patients would have ER+ disease on the basis of preliminary published evidence. ‐ Our sample size of a total of 70 patients (35 ER+ and 35 ER−) was predicted to provide 80% power to detect a one standard deviation difference in means between the two groups with an alpha level of .05. Statistical comparisons of participant NOHA levels versus US-determined ER IHC results were performed using Student's t test (GraphPad Prism, Version 8.0, La Jolla, CA). Statistical significance was determined at P < .05. Participant Pathology, Staging, and Sociodemographic Data Because of technical and staffing constraints, complete pathology, clinical staging, and sociodemographic data were available for 53 of 70 enrolled patients (Fig ). Of these, 33 had associated NOHA and IHC results (Fig , Table ). Sociodemographic data among these 33 participants revealed that all were women, with almost three-fourths being between the self-reported age of 20-60 years and a quarter being above a self-reported age of 60 years. A majority were from the Chagga tribe, reflecting geographic setting of the study in Northern Tanzania. The sociodemographic data collected from the 33 participants with a complete data set correlated reasonably well with the overall population. Clinical staging at enrollment revealed that most participants had T4 and/or nodal involvement, with 18% presenting with evidence of metastatic disease. Slides from 43 participants were available for histologic and ER and PR IHC testing in the United States. Of these, four (9%) were not shown to have cancer. Of the remaining 39 patients with IHC results, 33 had matching NOHA values. Most of the 33 had invasive ductal carcinoma (64%). IHC results in these 33 patients revealed that 45% of tumors were ER+ while 55% were ER−, with 10 of the 15 ER+ cases shown to be PR+ (67%). Three of the 33 tumors were ER− and PR+ (9%). A total of 20 participants had hormone receptor IHC testing performed on-site at KCMC. Of these, two had inconclusive IHC results at MMC. Excluding these two cases, IHC concordance was 78% (14 of 18) and 89% (16 of 18) for ER and PR, respectively, comparing on-site with US-run results, recognizing that the two sites varied in their classification of ER+ disease, that is, positivity cutoff values of at least 1% at MMC versus at least 10% at KCMC. NOHA to IHC Comparison Coupled with the 34 cards destroyed during commercial shipment at study outset, unstained slides were not available from three participants for US-based IHC testing, and three of the 36 dried plasma discs from the Noviplex prep cards were blood-contaminated and could not be analyzed for NOHA. This resulted in 33 of 70 participants with matching NOHA and IHC data for comparative ER assessment. Duplicate NOHA values deviated ≤ 0.2 nM from their average in this study population. As shown in Fig , study comparison of participant average NOHA level with IHC ER results revealed a distinct NOHA threshold that correlated 100% to the tumor's IHC categorization as ER−, ER+, and no tumor (Fig ). A NOHA level < 4.0 nM served as a reliable indicator of ER– breast cancer status. Participants with a NOHA level of 4.0-8.0 nM corresponded to ER+ tumor status, and three cases with a plasma NOHA level above 8.0 nM showed no tumor on study-based analysis of the corresponding pathology specimens. Additional NOHA-IHC comparisons relevant to the biomarker's clinical utility in the low-resource setting are summarized in Table . Because of technical and staffing constraints, complete pathology, clinical staging, and sociodemographic data were available for 53 of 70 enrolled patients (Fig ). Of these, 33 had associated NOHA and IHC results (Fig , Table ). Sociodemographic data among these 33 participants revealed that all were women, with almost three-fourths being between the self-reported age of 20-60 years and a quarter being above a self-reported age of 60 years. A majority were from the Chagga tribe, reflecting geographic setting of the study in Northern Tanzania. The sociodemographic data collected from the 33 participants with a complete data set correlated reasonably well with the overall population. Clinical staging at enrollment revealed that most participants had T4 and/or nodal involvement, with 18% presenting with evidence of metastatic disease. Slides from 43 participants were available for histologic and ER and PR IHC testing in the United States. Of these, four (9%) were not shown to have cancer. Of the remaining 39 patients with IHC results, 33 had matching NOHA values. Most of the 33 had invasive ductal carcinoma (64%). IHC results in these 33 patients revealed that 45% of tumors were ER+ while 55% were ER−, with 10 of the 15 ER+ cases shown to be PR+ (67%). Three of the 33 tumors were ER− and PR+ (9%). A total of 20 participants had hormone receptor IHC testing performed on-site at KCMC. Of these, two had inconclusive IHC results at MMC. Excluding these two cases, IHC concordance was 78% (14 of 18) and 89% (16 of 18) for ER and PR, respectively, comparing on-site with US-run results, recognizing that the two sites varied in their classification of ER+ disease, that is, positivity cutoff values of at least 1% at MMC versus at least 10% at KCMC. Coupled with the 34 cards destroyed during commercial shipment at study outset, unstained slides were not available from three participants for US-based IHC testing, and three of the 36 dried plasma discs from the Noviplex prep cards were blood-contaminated and could not be analyzed for NOHA. This resulted in 33 of 70 participants with matching NOHA and IHC data for comparative ER assessment. Duplicate NOHA values deviated ≤ 0.2 nM from their average in this study population. As shown in Fig , study comparison of participant average NOHA level with IHC ER results revealed a distinct NOHA threshold that correlated 100% to the tumor's IHC categorization as ER−, ER+, and no tumor (Fig ). A NOHA level < 4.0 nM served as a reliable indicator of ER– breast cancer status. Participants with a NOHA level of 4.0-8.0 nM corresponded to ER+ tumor status, and three cases with a plasma NOHA level above 8.0 nM showed no tumor on study-based analysis of the corresponding pathology specimens. Additional NOHA-IHC comparisons relevant to the biomarker's clinical utility in the low-resource setting are summarized in Table . In contrast to costly cytotoxic agents and HER2-directed therapies that carry significant risk and require close monitoring by an oncologist and IV administration in a dedicated infusion center, cost-efficient oral hormonal agents are more widely available and easier to administer in low-resource and community settings, including nontraditional settings. Despite the promise of broader availability of these effective therapies, their use is greatly hampered by little to no access to costly and resource-demanding IHC-based ER testing in SSA. Although approximately 80% of breast tumors are ER+ in US women, the proportion of ER+ breast tumors in SSA ranges from 20% to 70%. This variation likely reflects differential risk factor distributions (including reproductive and lifestyle) born from socioeconomic development efforts, histopathologic methods, and genetic heterogeneity across the continent. , Consistent with the results reported here, a single-institution study in urban Tanzania, using tightly controlled IHC testing methodology, revealed that approximately 50% of the cases presenting for care were ER− with these results subsequently confirmed in a national urban referral hospital population. , These data reinforce the need for reliable ER determination, supporting effective treatment decision making in Tanzania. Accurate ER testing requires high-quality histology and IHC facilities as part of a pathological review. A quality control issue in SSA is inappropriate handling of biopsy or excision specimens obtained in the community setting, compounded by sparse to no access to tissue processing facilities resulting in diagnostic delays, high IHC equipment and reagent costs, frequent equipment failure requiring hard-to-access technical support, and need for a skilled pathologist to accurately interpret and distribute results. Access to guidelines-based basic resources and care is unavailable among many with breast cancer in SSA because of these diagnostic barriers. Reflected in the data shown here, lack of access to pathology services in LMIC settings results in both late stage at disease presentation and limited treatment options among those ultimately presenting for care, with a critical component of treatment decision making, both curative and palliative intent, dependent on reliable ER status assessment. Here, we tested the generalizability of this US-validated, , blood-based assay to determine breast cancer ER status among a cohort of patients with breast cancer recruited from a single rural Tanzanian cancer center. The results presented provide preliminary evidence that the NOHA biomarker is valid within the Tanzanian context. We confirm that NOHA can be reliably measured via our existing ELISA assay using dried blood samples from TZ. Consistent with our previous US-based work, among the population studied here, distinct NOHA threshold levels correlate 100% to both tumor ER IHC and disease categorization where a level below 4 nM, from 4 to 8 nM, and above 8 nM signified ER−, ER+ and no cancer, respectively. , These pilot data suggest that NOHA is equivalent to IHC in breast cancer ER status determination and that NOHA analysis may offer utility in the identification of breast cancer in LMICs where tissue-based diagnosis is limited. Essential to the low-resource setting, NOHA stability, using plasma cards, was retained during up to 3 months of −80°C storage at KCMC, followed by ambient temperature, that is, at 25-30°C, hand carriage by air, followed by ground shipment to the US-based Mohan laboratory for analysis. This correlates with our prior laboratory-based storage testing outcomes, where NOHA maintained its stability at least 14 days in dried plasma at ambient or higher temperatures of ≤ 42°C. , These data are critical to real-world LMIC settings where sample storage and shipment conditions are often affected by unreliable cold storage, significant temperature extremes, and unreliable interinstitutional handling. Our own initial international sample shipping experience revealed this where the commercial shipper mishandled 34 plasma sample cards, rendering them unusable. Although this was a major barrier to the work described herein, relative to the scalability of NOHA in Tanzania and elsewhere, it reflected our pilot study design where all NOHA testing was performed in the United States. As the ELISA testing technology relies on easy to acquire and maintain equipment and reagents, it promises to be adaptable for use by point-of-care laboratory personnel receiving local training at the regional or community level, overcoming the sample shipment barriers experienced herein. As shown in Table , the cost to perform the NOHA ELISA in the Mohan (US) laboratory is $11.36/sample, including supplies, sample processing, data analysis, and technician time. It is expected that these costs will be less in Tanzania because of lower personnel expenses. Technical and analytic development efforts are underway for a portable NOHA assay that could further impact point-of-care access to this biomarker. With a turnaround time of ≤ 2.5 h/sample, NOHA would be an attractive and scalable replacement for costly IHC-based ER testing, with promising clinical applications extending beyond cancer diagnostics to surveillance, determination of prognosis, and disease monitoring. - This work is highly innovative as, to our knowledge, NOHA is the first blood-based technology that differentiates ER− versus ER+ breast cancers, offering rapid results through use of inexpensive, easy-to-maintain equipment and reagents, suitable for use by laboratory personnel at LMIC point of care. Furthermore, data reported here hold promise relative to NOHA's utility in supporting the diagnosis of breast cancer in clinical settings with little to no access to pathology services. If validated among broader breast cancer populations throughout Tanzania and SSA, this assay could be scaled globally. Taking bold and creative steps to address burgeoning breast cancer morbidity and mortality rates globally holds promise in addressing cancer care disparities. Limitations to this study include the small sample size and enrollment of patients from a single Tanzanian institution. On the basis of the results presented here, broader assessment of NOHA's clinical utility in the low-resource setting is warranted.
International Collaboration on Palliative Care Development Between ASCO and the Land of Hornbills	71215cf1-2c96-4ddb-986a-0538888fd64a	10166424	Internal Medicine[mh]	Sarawak, nicknamed the Land of Hornbills, located on northwest Borneo Island is the largest and fourth most populated state of Malaysia. It spans across 120,000 square kilometers hosting 2.9 million population. As of 2010, 54% of the population is urban while 46% are residing in rural areas. Given its widespread geographical area, Sarawak has the most dispersed rural communities in Malaysia. Despite the progressive development of infrastructure in Sarawak, inadequate connectivity in the rural area with consequential accessibility issues especially for health care service remains a major challenge in the state. CONTEXT Key Objective This report highlights the impact of the international collaborative work between ASCO and Sarawak. Knowledge Generated The collaborative effort meets the palliative care educational needs which resulted in positive changes in practice of the oncology department of Sarawak General Hospital. Trainer skills were taught in the trainer program, building a reservoir of local health care champions to ensure sustainability of palliative care service in Sarawak. The collaborative effort also led to translational work of the education materials, ensuring language is not a barrier in learning palliative care. Relevance All these efforts are integral to the development in Sarawak to be a center of excellence for palliative care. Challenges in providing palliative care in Sarawak are many and include lack of trained personnel, lack of awareness and buy-ins on palliative care service, financial constraints, geographical challenge, and limited access to health care service in rural areas. To our knowledge, to date, there is only one palliative care physician in the state of Sarawak since mid-2021. Therefore, palliative care services are often provided by doctors and nurses who are passionate in palliative care, many with limited formal training in palliative care. Sarawak General Hospital (SGH) is a tertiary public hospital and the only oncology center in the state which also serves as a teaching hospital for University Malaysia Sarawak (UNIMAS). Palliative care service in SGH has been uniquely initiated and developed under the leadership of the oncology unit. The services provided include a nine-bedded inpatient ward, intrahospital and interhospital consult services, outpatient palliative care clinic, and combined multidisciplinary motor neuron disease outpatient clinic. In addition, there are several independent, nonprofit organizations, namely National Cancer Society (Sarawak Branch), Two Tree Lodge Hospice and Kuching Life Care Society providing support and palliative care services in the community. The close-knitted collaboration between hospital-hospices enables patients who live within a 20-km radius to have seamless transition into the community, back to the comfort of their home and vice versa when needed. ASCO International Cancer Corps (ICC) was first introduced to the oncology SGH team by Malaysian Oncological Society in 2019. ICC is designed to improve the quality of cancer care of medical institutions in low- and middle-income countries, addressing three primary areas of need namely, multidisciplinary management of common cancers, integration of palliative care into cancer care, and improvement of quality of care using evidence-based quality measures. In 2020, collaborative work between SGH, UNIMAS, and ASCO began with the signing of Memorandum of Understanding. Unforeseen emergence of COVID-19 pandemic has resulted in setbacks such as scheduling, in-person training, and onsite visits of the collaboration. Nonetheless, this did not deter the team from successfully initiating the first online palliative care educational curriculum in 2021. Key Objective This report highlights the impact of the international collaborative work between ASCO and Sarawak. Knowledge Generated The collaborative effort meets the palliative care educational needs which resulted in positive changes in practice of the oncology department of Sarawak General Hospital. Trainer skills were taught in the trainer program, building a reservoir of local health care champions to ensure sustainability of palliative care service in Sarawak. The collaborative effort also led to translational work of the education materials, ensuring language is not a barrier in learning palliative care. Relevance All these efforts are integral to the development in Sarawak to be a center of excellence for palliative care. Objectives of Palliative Care Collaboration To advocate and promote palliative care awareness, competency and service development in Sarawak To provide holistic palliative care through unified efforts by all health care providers and independent, nonprofit organizations, with optimal usage of available resources to provide a support system for patients and family members. To achieve optimal pain and symptom control in patients with advanced cancer and ensure safe and appropriate opioid administration and titration. To educate and conduct workshops on cancer pain and symptom management to empower health care providers with palliative care knowledge and skill sets via efficient and cost-effective training programs. To develop trainer skills among local health care champions to expand training and educational programs and ensure sustainable service provision. Report of Palliative Care Collaborations ASCO Palliative Care e-Course 2021. The collaborative initiative between Sarawak, UNIMAS, and ASCO kicked off with the rollout of the first ASCO Palliative Care e-Course (APCeC) titled “A Taste of Palliative Medicine” in March to June 2021 attended by 32 participants (Fig ) across Sarawak, with an average of 6.2 years' experience in palliative care. This was a 12-session comprehensive and personalized online course conducted weekly with participants having access to precourse self-study material, namely PowerPoints notes, Palliative Care Interdisciplinary Curriculum (PCIC) videos, and related journals/articles. The course featured case-based presentations and breakout sessions on different aspects of palliative care. The main objectives of the course are Understanding the concepts and principles of palliative and end-of-life care. Effective communication with patients and their families. Optimal pain and symptom control including safe and effective opioids usage and titration Networking of palliative care providers across Sarawak A postcourse impact assessment was done 6 months later, with 22 of 32 participants answering the survey voluntarily, representing 69% of attendees. All respondents (n = 22) reported that they were able to provide and use the skills learned in their daily practice. The most reported changes were related to communication, pain, and symptoms assessment/management (Fig ). Seventy-one percent of respondents stated that they made these changes within 4 weeks of attending the course. Overall, of those who responded to the follow-up impact assessment, the results suggested that the course was successful. Train the Trainer Program 2022. After the successful first APCeC course, a weekly virtual Train the Trainer (TTT) course was conducted over a span of 3 weeks from February 2022 to March 2022. There were a total of 27 participants, primarily palliative care physicians, and palliative care nurses across Sarawak and other parts of Malaysia (Fig ). The participants have an average of 6.9 years of experience in their current profession, with respondents reported giving training to up to 100 participants annually and conducting up to 12 trainings per year. The objectives of the course were to equip participants to effectively Present information. Facilitate discussions. Provide feedback. A postcourse evaluation survey obtained from 23 respondents, representing 85% of all participants showed that 87% (n = 20) reported an increase in their ability to present information effectively, 91% (n = 21) reported an increase in ability to facilitate discussion effectively, and 83% (n = 19) reported an increase in ability in providing feedback effectively. These results suggested that the course has further enhanced the ability of respondents to be trainers and to present information, provide feedback, and facilitate discussion effectively. International Development and Education Award—Palliative Care. International Development and Education Award—Palliative Care award from the Conquer Cancer Foundation encompasses 3 years of complimentary membership to ASCO, attendance to ASCO's 2022 Annual Meeting in Chicago, TTT Workshop, mentorship program, and extended tour award to University of Chicago. It also provides a platform for networking and future collaborations among award recipients. Dr Wong Yin Yee, palliative care unit registrar of SGH and one of the local palliative care champions is the first recipient from Malaysia of this globally competitive award. In July 2022, Dr Wong Yin Yee, along with other awardees of International Development and Education Award—Palliative Care, were able to attend an onsite TTT workshop by Dr Frank Ferris and team in Chicago. Essential trainer skills including providing effective feedback and elevator pitches were imparted, and there was plenty of opportunity to practice these skills during the course. Furthermore, the mentorship program creates opportunities for mentor and mentee to work on projects to improve palliative care outcomes. The cultural exchange provided a platform for sharing of ideas, bench markings, support, and opportunity for future collaboration. Translation of ASCO PCIC Resources. ASCO PCIC has an extensive and comprehensive educational curriculum available in 10 languages to date. During the first APCeC virtual course, some participants reported that they had trouble following the precourse videos because of language limitations as the primary language in Malaysia is Malay, while English is the secondary language. Therefore, major efforts are underway to translate these resources to Malay language, to ensure that language is not a barrier in learning palliative care. The translations are done in partnership with developers of the curricular materials. They were guided by Malaysian palliative care providers from various states who volunteered their time and expertise to ensure translational accuracy and availability of resources. To advocate and promote palliative care awareness, competency and service development in Sarawak To provide holistic palliative care through unified efforts by all health care providers and independent, nonprofit organizations, with optimal usage of available resources to provide a support system for patients and family members. To achieve optimal pain and symptom control in patients with advanced cancer and ensure safe and appropriate opioid administration and titration. To educate and conduct workshops on cancer pain and symptom management to empower health care providers with palliative care knowledge and skill sets via efficient and cost-effective training programs. To develop trainer skills among local health care champions to expand training and educational programs and ensure sustainable service provision. ASCO Palliative Care e-Course 2021. The collaborative initiative between Sarawak, UNIMAS, and ASCO kicked off with the rollout of the first ASCO Palliative Care e-Course (APCeC) titled “A Taste of Palliative Medicine” in March to June 2021 attended by 32 participants (Fig ) across Sarawak, with an average of 6.2 years' experience in palliative care. This was a 12-session comprehensive and personalized online course conducted weekly with participants having access to precourse self-study material, namely PowerPoints notes, Palliative Care Interdisciplinary Curriculum (PCIC) videos, and related journals/articles. The course featured case-based presentations and breakout sessions on different aspects of palliative care. The main objectives of the course are Understanding the concepts and principles of palliative and end-of-life care. Effective communication with patients and their families. Optimal pain and symptom control including safe and effective opioids usage and titration Networking of palliative care providers across Sarawak A postcourse impact assessment was done 6 months later, with 22 of 32 participants answering the survey voluntarily, representing 69% of attendees. All respondents (n = 22) reported that they were able to provide and use the skills learned in their daily practice. The most reported changes were related to communication, pain, and symptoms assessment/management (Fig ). Seventy-one percent of respondents stated that they made these changes within 4 weeks of attending the course. Overall, of those who responded to the follow-up impact assessment, the results suggested that the course was successful. Train the Trainer Program 2022. After the successful first APCeC course, a weekly virtual Train the Trainer (TTT) course was conducted over a span of 3 weeks from February 2022 to March 2022. There were a total of 27 participants, primarily palliative care physicians, and palliative care nurses across Sarawak and other parts of Malaysia (Fig ). The participants have an average of 6.9 years of experience in their current profession, with respondents reported giving training to up to 100 participants annually and conducting up to 12 trainings per year. The objectives of the course were to equip participants to effectively Present information. Facilitate discussions. Provide feedback. A postcourse evaluation survey obtained from 23 respondents, representing 85% of all participants showed that 87% (n = 20) reported an increase in their ability to present information effectively, 91% (n = 21) reported an increase in ability to facilitate discussion effectively, and 83% (n = 19) reported an increase in ability in providing feedback effectively. These results suggested that the course has further enhanced the ability of respondents to be trainers and to present information, provide feedback, and facilitate discussion effectively. International Development and Education Award—Palliative Care. International Development and Education Award—Palliative Care award from the Conquer Cancer Foundation encompasses 3 years of complimentary membership to ASCO, attendance to ASCO's 2022 Annual Meeting in Chicago, TTT Workshop, mentorship program, and extended tour award to University of Chicago. It also provides a platform for networking and future collaborations among award recipients. Dr Wong Yin Yee, palliative care unit registrar of SGH and one of the local palliative care champions is the first recipient from Malaysia of this globally competitive award. In July 2022, Dr Wong Yin Yee, along with other awardees of International Development and Education Award—Palliative Care, were able to attend an onsite TTT workshop by Dr Frank Ferris and team in Chicago. Essential trainer skills including providing effective feedback and elevator pitches were imparted, and there was plenty of opportunity to practice these skills during the course. Furthermore, the mentorship program creates opportunities for mentor and mentee to work on projects to improve palliative care outcomes. The cultural exchange provided a platform for sharing of ideas, bench markings, support, and opportunity for future collaboration. Translation of ASCO PCIC Resources. ASCO PCIC has an extensive and comprehensive educational curriculum available in 10 languages to date. During the first APCeC virtual course, some participants reported that they had trouble following the precourse videos because of language limitations as the primary language in Malaysia is Malay, while English is the secondary language. Therefore, major efforts are underway to translate these resources to Malay language, to ensure that language is not a barrier in learning palliative care. The translations are done in partnership with developers of the curricular materials. They were guided by Malaysian palliative care providers from various states who volunteered their time and expertise to ensure translational accuracy and availability of resources. The collaborative initiative between Sarawak, UNIMAS, and ASCO kicked off with the rollout of the first ASCO Palliative Care e-Course (APCeC) titled “A Taste of Palliative Medicine” in March to June 2021 attended by 32 participants (Fig ) across Sarawak, with an average of 6.2 years' experience in palliative care. This was a 12-session comprehensive and personalized online course conducted weekly with participants having access to precourse self-study material, namely PowerPoints notes, Palliative Care Interdisciplinary Curriculum (PCIC) videos, and related journals/articles. The course featured case-based presentations and breakout sessions on different aspects of palliative care. The main objectives of the course are Understanding the concepts and principles of palliative and end-of-life care. Effective communication with patients and their families. Optimal pain and symptom control including safe and effective opioids usage and titration Networking of palliative care providers across Sarawak A postcourse impact assessment was done 6 months later, with 22 of 32 participants answering the survey voluntarily, representing 69% of attendees. All respondents (n = 22) reported that they were able to provide and use the skills learned in their daily practice. The most reported changes were related to communication, pain, and symptoms assessment/management (Fig ). Seventy-one percent of respondents stated that they made these changes within 4 weeks of attending the course. Overall, of those who responded to the follow-up impact assessment, the results suggested that the course was successful. After the successful first APCeC course, a weekly virtual Train the Trainer (TTT) course was conducted over a span of 3 weeks from February 2022 to March 2022. There were a total of 27 participants, primarily palliative care physicians, and palliative care nurses across Sarawak and other parts of Malaysia (Fig ). The participants have an average of 6.9 years of experience in their current profession, with respondents reported giving training to up to 100 participants annually and conducting up to 12 trainings per year. The objectives of the course were to equip participants to effectively Present information. Facilitate discussions. Provide feedback. A postcourse evaluation survey obtained from 23 respondents, representing 85% of all participants showed that 87% (n = 20) reported an increase in their ability to present information effectively, 91% (n = 21) reported an increase in ability to facilitate discussion effectively, and 83% (n = 19) reported an increase in ability in providing feedback effectively. These results suggested that the course has further enhanced the ability of respondents to be trainers and to present information, provide feedback, and facilitate discussion effectively. International Development and Education Award—Palliative Care award from the Conquer Cancer Foundation encompasses 3 years of complimentary membership to ASCO, attendance to ASCO's 2022 Annual Meeting in Chicago, TTT Workshop, mentorship program, and extended tour award to University of Chicago. It also provides a platform for networking and future collaborations among award recipients. Dr Wong Yin Yee, palliative care unit registrar of SGH and one of the local palliative care champions is the first recipient from Malaysia of this globally competitive award. In July 2022, Dr Wong Yin Yee, along with other awardees of International Development and Education Award—Palliative Care, were able to attend an onsite TTT workshop by Dr Frank Ferris and team in Chicago. Essential trainer skills including providing effective feedback and elevator pitches were imparted, and there was plenty of opportunity to practice these skills during the course. Furthermore, the mentorship program creates opportunities for mentor and mentee to work on projects to improve palliative care outcomes. The cultural exchange provided a platform for sharing of ideas, bench markings, support, and opportunity for future collaboration. ASCO PCIC has an extensive and comprehensive educational curriculum available in 10 languages to date. During the first APCeC virtual course, some participants reported that they had trouble following the precourse videos because of language limitations as the primary language in Malaysia is Malay, while English is the secondary language. Therefore, major efforts are underway to translate these resources to Malay language, to ensure that language is not a barrier in learning palliative care. The translations are done in partnership with developers of the curricular materials. They were guided by Malaysian palliative care providers from various states who volunteered their time and expertise to ensure translational accuracy and availability of resources. Because of limitations in resources and manpower, the standard practice for referrals to the palliative care team in SGH is largely for patients on best supportive care. APCeC helps to provide a fundamental framework for palliative care education that is invaluable in equipping oncologists and oncology trainees with the necessary knowledge and skill sets to better identify and meet palliative care needs among their patients. It ensures a more competent and timely palliative care provision at a general level by the oncology team of SGH and enables the team to incorporate basic palliative care management early in the course of illness alongside active oncological treatment. With the availability of the first palliative care physician at SGH since mid-2021, the palliative care team has opened for earlier referrals for oncology patients still undergoing active oncological treatment with more complex palliative care needs as observed in Figure . These early referrals are known as early introduction. The ASCO collaboration enhances teamwork and helps the oncology team to recognize their limitations while providing general palliative care, thereby encouraging more timely palliative care referrals when appropriate to ensure that patients with more complex physical, psychosocial, and spiritual needs have the necessary input and support from the palliative care team throughout the course of patients' illnesses. The palliative care international collaboration with ASCO ICC was set in motion with virtual meetings to understand local challenges and the palliative care educational needs of Sarawakians, followed by elaborate discussions on training objectives and intended outcomes. These had led to the development of a highly personalized palliative care education program which is one of the highlights of this collaboration. The first APCeC program allowed participants from various districts across Sarawak to gain knowledge and skills effectively in the comfort of their home. This eliminated financial costs for travel, accommodation, and time away from work. The 12-session e-course spanned out across a few months, allowing participants to read up materials and learn and digest information at a reasonable pace. The successful rollout of the first APCeC program had sparked interest among palliative care providers from the Malaysia Ministry of Health. Subsequently, it has led to meaningful developments such as the TTT program and second APCeC program. The TTT program was developed to address the need to equip and nurture trainer skills among local palliative care providers so as to ensure high quality of palliative care education and sustainability in training. The program benefitted not only Sarawak palliative care providers but also included palliative care providers across Malaysia. The TTT program encouraged participants to effectively practice the skills taught in small breakout groups, namely presentation, facilitating, and feedback skills. Meanwhile, the second APCeC program is a collaborative effort not only between ASCO, UNIMAS, and SGH but also involves Malaysia Ministry of Health. Facilitators from the ASCO team were paired with Malaysian cofacilitators, namely palliative care providers who had completed the TTT program. There was an overwhelming response with 100 applications for a limited 60 slots. The second APCeC was similarly a 12-session comprehensive virtual course that concluded in August 2022. There were 30 participants from Sarawak and 30 participants from other states in Malaysia. Some of the challenges experienced working on international collaboration includes scheduling, cultural differences, language, and technology failure. Scheduling for discussion, preparation, and delivery of course calls for dedication and commitment beyond working hours by both international and local teams. Although cultural differences are present, robust discussion in smaller breakout groups and conscientiously obtaining feedback at the end of every session provided opportunity to discuss on adaptation of skills learnt to local setting. The official and national language of Malaysia is Malay while English is the second language. Communication sessions during break out groups may remain a challenge for participants whose first or second language is not English. Technology failures experienced by some participants are mainly band width issues causing inability to turn on videos throughout the entire course or video lag. Participants are still able to listen to lectures and participate through chat boxes and voice communication. Despite the challenges, the international ASCO ICC collaboration, with aligned goals and objectives, has brought about tremendous positive impacts. Efforts made throughout the collaboration by both parties are not only worthwhile but also enriching. The reported outcomes for the first APCeC program suggest that it was a success. It is worth noting that the majority of participants have basic background palliative care knowledge and are actively involved with palliative care work. Hence, they may be more receptive to the contents delivered and more able to put knowledge and skills learnt into practice in a supportive work environment. Therefore, further evaluation on the outcome of the second APCeC program is crucial as the participants were from a more diverse nonpalliative background with close to one third of the participants reported not having any prior exposure to palliative care. The evaluation would provide a better insight on the impact and value of the e-courses. This will enable the team to evaluate and explore different learning needs and adapt the e-course contents to cater to different groups of participants accordingly. Moving forward, the SGH palliative care team is anticipating the in-person training by Dr Frank Ferris and his team of experts from ASCO in November 2022 to further enhance the skill set taught during the APCeC program. The esteemed ASCO palliative care experts will also conduct an in-person second TTT workshop focusing on advocacy, leadership and mentoring during their anticipated visit. In conclusion, the reproducibility of the comprehensively crafted APCeC curriculum along with the abundant resources from ASCO will serve as an invaluable and pivotal platform for the expansion of palliative care service in various departments of SGH and across districts in Sarawak. In addition, the TTT program will equip and enable local leaders in the fraternity to contribute to palliative care education and development in Sarawak and Malaysia. These international collaborations are integral to the development of palliative care in Sarawak, with the vision to build Sarawak to be the center of excellence for impeccable palliative care service, education and training in Malaysia and beyond.
The PRINCIPAL Network: A Model to Optimize Infection Care and Prevention in Pediatric Oncology in the Latin American Region	a85a95a1-4d46-47ef-bd7d-bc6031f237c9	10166443	Pediatrics[mh]	The survival of children with cancer in low- and middle-income countries (LMICs) remains poor, whereas in high-income countries, it has surpassed 85%. This unequal outcome reflects the more advanced stage of disease at diagnosis, the higher rate of treatment abandonment, and the high treatment-related mortality, mainly because of infections, in LMICs. Outcomes of care for children in LMICs can be improved by implementing measures to prevent and/or control infections. Such measures to improve the quality of care rely on the presence of a sufficient number of health care providers with the necessary expertise. CONTEXT Key Objective The Prevencionistas e Infectólogos para Cáncer Pediátrico en América Latina (PRINCIPAL) network is a community of health care professionals focused on decreasing the risk for poor outcomes of infections in children in Latin America. Knowledge Generated The network stimulates knowledge exchange, cross-mentoring, and collaboration. Both intraregional collaboration and inter-regional collaboration allow for a unified approach to address mutual concerns and for advocacy to global agencies. Relevance PRINCIPAL is an imperative to improving infection-related care and has progressively evolved to its current structure. Although St Jude Global has supported its inception and early performance, PRINCIPAL is envisioned to become a self-sustaining structure with active participation of other institutions, organizations, and agencies to strengthen its work. A lasting effect of the network will be the improvement of local infectious disease expertise and management; and the sharing of relevant data that will lead to better patient outcomes. In many LMICs, a shortage of skilled health care professionals hinders the provision of high-quality care, and postgraduate education and training are often unavailable. In such an environment, knowledge sharing, fluid communication, and collaboration between health care professionals can maximize the use of their limited time and expertise. Regional and global professional networks in infection care and prevention (IC&P) with focused goals and objectives can be resources for capacity building, education and studies in clinical research, quality improvement, and implementation science. - In 2017, we initiated an educational program for health care providers caring for children with cancer and infections. The alumni of the first training cohort established a network that remained active after the completion of the course (Fig , Table ). Here, we describe how this network was built, organized, and sustained. Additionally, we share how this type of professional network can be used to stimulate medical education and collaboration in IC&P and thereby serve as a model to improve the quality-of-care delivery for children with cancer. Key Objective The Prevencionistas e Infectólogos para Cáncer Pediátrico en América Latina (PRINCIPAL) network is a community of health care professionals focused on decreasing the risk for poor outcomes of infections in children in Latin America. Knowledge Generated The network stimulates knowledge exchange, cross-mentoring, and collaboration. Both intraregional collaboration and inter-regional collaboration allow for a unified approach to address mutual concerns and for advocacy to global agencies. Relevance PRINCIPAL is an imperative to improving infection-related care and has progressively evolved to its current structure. Although St Jude Global has supported its inception and early performance, PRINCIPAL is envisioned to become a self-sustaining structure with active participation of other institutions, organizations, and agencies to strengthen its work. A lasting effect of the network will be the improvement of local infectious disease expertise and management; and the sharing of relevant data that will lead to better patient outcomes. Resources The St Jude Global infectious disease program. The goal of the Global infectious disease (Global ID) program is to help decrease the rate of and risks for infection among children with cancer, mainly—but not exclusively—at institutions that are members of the St Jude Children's Research Hospital (St Jude) Global Alliance. The administrative and programmatic financial support for Global ID program is provided by St Jude. Three coordinators dedicate up to 40% time to managing the networks as part of their responsibilities in the Global ID program. This program works to develop and promote standards for IC&P by establishing, participating in, and facilitating capacity-building initiatives, education and training opportunities, and research studies on infectious disease (ID)–related topics at collaborative sites around the world. To satisfy the learning needs of collaborators, the Global ID Program established two main training courses: the ID Training Seminars for leaders and the Intensive Infection Control Course for preventionists. The Global ID Program stimulates communication with and among participants during and after completion of the courses by inviting them to participate in program activities, such as contributing to a monthly bulletin or communicating via social media venues, for example, WhatsApp. As these engagements were not sufficient by themselves, and as the number of course graduates grew over time, the need for a more sustainable and more organized structure for engaging with Global ID and with each other was increasingly apparent. Accordingly, 2017 saw the birth of the concept of a network to address the needs of participants with respect to IC&P capacity building, education, and research. The St Jude Global ID training seminars for leaders. This is a 10-week blended course in IC&P consisting of 8 weeks of distance learning followed by 2 weeks of in-person learning. Since its launch in 2017, 189 participants have attended this course. Overall, participants showed improved knowledge in topics taught and positive behavioral changes in their institutional engagement. The members of each cohort of participants bond throughout their training and continue to communicate and cooperate with each other and the Global ID Program after graduation, including joining and participating in our networks. The St Jude Infection Prevention and Control course for preventionists. This is a 10-week blended course in infection prevention and control (IPC) for infection preventionists. It consists of 9 weeks of distance learning followed by 1 week of in-person learning. The original version of the course, consisting of 4 weeks of in-person learning, was launched in 2005 but subsequently underwent a series of transformations to accommodate the needs of the audience. The course is offered in Spanish and English and, to date, has trained 581 participants. Participant outcomes include increased expertise in IPC and plans for improving or implementing IPC infrastructure at their institutions. IPC course attendees also communicate with each other as part of course assignments, and these virtual relationships continue after completion of the course, including through the Global ID networks. Global health sessions of the annual St Jude/PIDS Pediatric Infectious Diseases Research Conference (St Jude/PIDS). The global health session (GHS) is a program within the St Jude/Pediatric Infectious Diseases Society (PIDS) annual research conference conceived to satisfy a growing need for a dedicated forum in which participants could present and discuss IC&P developments that concern pediatric populations globally and to create opportunities for domestic and global participants to meet, learn from each other, and collaborate. The GHS, created by the Global ID program in 2017 and comanaged with the St Jude/PIDS conference, has the goal of exposing attendees to pediatric global health issues, stimulating participation in efforts to better understand health determinants that affect children around the globe and decrease IC&P health inequities. The GHS provides a setting for network members to present and discuss the results of their studies on infectious pathogens, especially those affecting children with depressed immunity, as poster or oral presentations. The St Jude Global Alliance online community infectious disease portal. The ID portal is a platform through which health care providers, including members of the Global ID networks, can connect, communicate, and collaborate. Through the ID portal, network members interact, ask questions, seek advice, and find information on topics of interest. The ID portal provides access to a collection of tools and educational resources, along with opportunities for project-specific interactions. Currently, this portal is password-protected, limiting access to members of the community. Network Design and Development Network design. Graduates of the first ID course wished to continue communicating and cooperating with each other after the course ended in 2017. With that desire in mind, a professional network was thought to be a way to continue engagement and a network name was selected: PRINCIPAL, a Spanish acronym for Prevencionistas e Infectologos de Cancer Pediátrico en America Latina, meaning Preventionists and Infectologists for Pediatric Cancer in Latin America. Moreover, basic concepts, such as the vision, mission, and values, were established in discussion with course participants. After several iterations and corrections, the current vision, mission, and values were finalized and are provided in Table . Network development. Once the guiding concepts were formulated, we addressed issues such as membership eligibility and benefits, network activities, and communication venues, including internet messaging (WhatsApp). One early benefit of membership was the ability to obtain the support of fellow members as speakers at IC&P conferences and workshops to impart knowledge in IC&P in pediatric oncology. Requirements for membership included the following: employment or privileges at a health care facility that provides care to children with cancer or other catastrophic illnesses; involvement in IC&P activities; agreement to collaborate with Global ID and network members on strategic goals; and permission to engage in network activities from a direct work supervisor. The benefits of being a member now include access to additional education and training resources through Global ID and St Jude Global, opportunities to obtain support for travel to regional and international scientific meetings, opportunities to collaborate on multicenter research and collaborative studies, and the availability of support for study design and data management. To engage network members and increase their knowledge of and expertise in IC&P in the areas of clinical care and research, we created the following resources: Case-based learning: We established a weekly event, Kiosko de Casos , to improve IC&P professional skills through the discussion of difficult and complex patient cases. These cases mainly involve patients with cancer or recipients of hematopoietic cell transplants. Research in human subject certification: We stimulated network members to become certified in human subject research by providing access to the Collaborative Institutional Training Initiative program. Collaborative Institutional Training Initiative training is mandatory for participating in network-associated research activities but is voluntary for other interested members. Quality improvement skills: We partnered with a Lean 6-Sigma instructor through the St Jude Department of Quality and Patient Safety to collaborate on and deliver a 6-month distance-learning training program through which participants earned Green Belt certification. Cross-collaboration: The PRINCIPAL network has become a focal point for expertise in IC&P in immunocompromised children. As a result, network members have participated in multiple institutional, national, and regional IC&P meetings and conferences, collaborating with institutions and organizers of scientific meetings, short courses, and seminars as speakers on identified gaps in IC&P topics in immunocompromised patients (Table ). Presentation of professional work: We have encouraged members to share local experiences, the results of clinical research, and the outcomes of quality improvement efforts in IC&P at scientific conferences, particularly at the St Jude/PIDS GHS (Table ). Participating in research related to IC&P: We have also encouraged network members to establish and/or collaborate in clinical or basic research and qualitative studies. - Network annual meeting: Since the network was formally established in 2017, an annual meeting has been scheduled every year thereafter. The meeting was held in person before the pandemic (2018-2019) and was virtual in 2020-2022. In these meetings, we review the state of the network and provide reports on achievements and future directions. Monthly bulletin: We produce a monthly bulletin in which we announce upcoming events and summarize the results of ongoing activities. Every other month, we invite network members to contribute a review of a recently published IC&P paper, to share their experiences in IC&P, and/or to communicate some important facts regarding IC&P. This bulletin is distributed to all graduates of our courses and other interested individuals. After publication, the issues are archived and are accessible through our ID portal. PRINCIPAL network structure and leadership. The PRINCIPAL network activities are coordinated by the Global ID program with the input of members. This includes enrolling new members, coordinating meetings, finding speakers, and assisting working groups (WGs) in executing their projects. To integrate the PRINCIPAL network into a global structure, representatives of this network serve on the recently formed Global ID network steering committee (SC). This committee will provide a leadership structure that will in turn lead to a more sustainable and participatory organization. Committee members are currently working to outline the duties of the committee, as well as the roles of the WGs. Leadership consists of network sponsors, SC members, and WG leaders. The WGs will execute projects as they emerge, on the basis of the interest and resources available. Support of these WGs will be essential to ensure productivity across the domains of education, capacity building, and research. Members. Prospective members of the network must be qualified by their training, be involved in pediatric oncology care, and have the support of their workplace. Network membership is free of charge but requires completing an application and periodic renewal. The benefits of membership include invitations to all network-related activities and initiatives, as well as leadership participation. Members also have the opportunity to apply for travel awards for attending scientific conferences and courses, and they receive assistance with finding support for project execution and publication. Sponsors. The sponsors speak about and advocate for the network. They promote and discuss its function and accomplishments, and they assist with matters relating to representation, the procurement of resources, and funding. Currently, the Global ID program is the main sponsor of the network. Funding. Department and programmatic resources support network operations and maintenance. Educational programs and conference awards are funded through the Global ID budget, but members are encouraged to apply for conference-sponsored awards as well as other funding opportunities circulated by the network. Global ID network SC. The SC deliberates, makes decisions, advises, provides strategic oversight, and serves as the primary advocate for initiatives carried out by WGs. Their roles are to build, review, and amend (if needed) long- and short-term goals, objectives, and strategies; to guide network activities on the basis of these goals, objectives, and strategies; to monitor network activities; and to develop ideas to advance network objectives by conducting periodic evaluations of the network. The members of the inaugural SC were selected by the sponsors, and subsequent members will be selected by the departing SC members and the sponsors. SC will facilitate the transition of network activity ownership to its members. This will include, but will not be limited to, leading and managing network project initiatives and organizing coordination and administrative support. Working groups. The performance of the network will be reflected in the participation, productivity, and diverse themes of the WGs, categorized into three groups: education, capacity building, and research. These WGs, which are highly collaborative, will be led by interested members and may be composed of fellow network members or nonmembers. Before activation, the WG proposal is reviewed and approved by the SC. To date, there are three operational WGs in the global network and one WG exclusively for the PRINCIPAL network region (the use of a clinical care pathway in the initial management of a pediatric patient with fever). The St Jude Global infectious disease program. The goal of the Global infectious disease (Global ID) program is to help decrease the rate of and risks for infection among children with cancer, mainly—but not exclusively—at institutions that are members of the St Jude Children's Research Hospital (St Jude) Global Alliance. The administrative and programmatic financial support for Global ID program is provided by St Jude. Three coordinators dedicate up to 40% time to managing the networks as part of their responsibilities in the Global ID program. This program works to develop and promote standards for IC&P by establishing, participating in, and facilitating capacity-building initiatives, education and training opportunities, and research studies on infectious disease (ID)–related topics at collaborative sites around the world. To satisfy the learning needs of collaborators, the Global ID Program established two main training courses: the ID Training Seminars for leaders and the Intensive Infection Control Course for preventionists. The Global ID Program stimulates communication with and among participants during and after completion of the courses by inviting them to participate in program activities, such as contributing to a monthly bulletin or communicating via social media venues, for example, WhatsApp. As these engagements were not sufficient by themselves, and as the number of course graduates grew over time, the need for a more sustainable and more organized structure for engaging with Global ID and with each other was increasingly apparent. Accordingly, 2017 saw the birth of the concept of a network to address the needs of participants with respect to IC&P capacity building, education, and research. The St Jude Global ID training seminars for leaders. This is a 10-week blended course in IC&P consisting of 8 weeks of distance learning followed by 2 weeks of in-person learning. Since its launch in 2017, 189 participants have attended this course. Overall, participants showed improved knowledge in topics taught and positive behavioral changes in their institutional engagement. The members of each cohort of participants bond throughout their training and continue to communicate and cooperate with each other and the Global ID Program after graduation, including joining and participating in our networks. The St Jude Infection Prevention and Control course for preventionists. This is a 10-week blended course in infection prevention and control (IPC) for infection preventionists. It consists of 9 weeks of distance learning followed by 1 week of in-person learning. The original version of the course, consisting of 4 weeks of in-person learning, was launched in 2005 but subsequently underwent a series of transformations to accommodate the needs of the audience. The course is offered in Spanish and English and, to date, has trained 581 participants. Participant outcomes include increased expertise in IPC and plans for improving or implementing IPC infrastructure at their institutions. IPC course attendees also communicate with each other as part of course assignments, and these virtual relationships continue after completion of the course, including through the Global ID networks. Global health sessions of the annual St Jude/PIDS Pediatric Infectious Diseases Research Conference (St Jude/PIDS). The global health session (GHS) is a program within the St Jude/Pediatric Infectious Diseases Society (PIDS) annual research conference conceived to satisfy a growing need for a dedicated forum in which participants could present and discuss IC&P developments that concern pediatric populations globally and to create opportunities for domestic and global participants to meet, learn from each other, and collaborate. The GHS, created by the Global ID program in 2017 and comanaged with the St Jude/PIDS conference, has the goal of exposing attendees to pediatric global health issues, stimulating participation in efforts to better understand health determinants that affect children around the globe and decrease IC&P health inequities. The GHS provides a setting for network members to present and discuss the results of their studies on infectious pathogens, especially those affecting children with depressed immunity, as poster or oral presentations. The St Jude Global Alliance online community infectious disease portal. The ID portal is a platform through which health care providers, including members of the Global ID networks, can connect, communicate, and collaborate. Through the ID portal, network members interact, ask questions, seek advice, and find information on topics of interest. The ID portal provides access to a collection of tools and educational resources, along with opportunities for project-specific interactions. Currently, this portal is password-protected, limiting access to members of the community. The goal of the Global infectious disease (Global ID) program is to help decrease the rate of and risks for infection among children with cancer, mainly—but not exclusively—at institutions that are members of the St Jude Children's Research Hospital (St Jude) Global Alliance. The administrative and programmatic financial support for Global ID program is provided by St Jude. Three coordinators dedicate up to 40% time to managing the networks as part of their responsibilities in the Global ID program. This program works to develop and promote standards for IC&P by establishing, participating in, and facilitating capacity-building initiatives, education and training opportunities, and research studies on infectious disease (ID)–related topics at collaborative sites around the world. To satisfy the learning needs of collaborators, the Global ID Program established two main training courses: the ID Training Seminars for leaders and the Intensive Infection Control Course for preventionists. The Global ID Program stimulates communication with and among participants during and after completion of the courses by inviting them to participate in program activities, such as contributing to a monthly bulletin or communicating via social media venues, for example, WhatsApp. As these engagements were not sufficient by themselves, and as the number of course graduates grew over time, the need for a more sustainable and more organized structure for engaging with Global ID and with each other was increasingly apparent. Accordingly, 2017 saw the birth of the concept of a network to address the needs of participants with respect to IC&P capacity building, education, and research. This is a 10-week blended course in IC&P consisting of 8 weeks of distance learning followed by 2 weeks of in-person learning. Since its launch in 2017, 189 participants have attended this course. Overall, participants showed improved knowledge in topics taught and positive behavioral changes in their institutional engagement. The members of each cohort of participants bond throughout their training and continue to communicate and cooperate with each other and the Global ID Program after graduation, including joining and participating in our networks. This is a 10-week blended course in infection prevention and control (IPC) for infection preventionists. It consists of 9 weeks of distance learning followed by 1 week of in-person learning. The original version of the course, consisting of 4 weeks of in-person learning, was launched in 2005 but subsequently underwent a series of transformations to accommodate the needs of the audience. The course is offered in Spanish and English and, to date, has trained 581 participants. Participant outcomes include increased expertise in IPC and plans for improving or implementing IPC infrastructure at their institutions. IPC course attendees also communicate with each other as part of course assignments, and these virtual relationships continue after completion of the course, including through the Global ID networks. The global health session (GHS) is a program within the St Jude/Pediatric Infectious Diseases Society (PIDS) annual research conference conceived to satisfy a growing need for a dedicated forum in which participants could present and discuss IC&P developments that concern pediatric populations globally and to create opportunities for domestic and global participants to meet, learn from each other, and collaborate. The GHS, created by the Global ID program in 2017 and comanaged with the St Jude/PIDS conference, has the goal of exposing attendees to pediatric global health issues, stimulating participation in efforts to better understand health determinants that affect children around the globe and decrease IC&P health inequities. The GHS provides a setting for network members to present and discuss the results of their studies on infectious pathogens, especially those affecting children with depressed immunity, as poster or oral presentations. The ID portal is a platform through which health care providers, including members of the Global ID networks, can connect, communicate, and collaborate. Through the ID portal, network members interact, ask questions, seek advice, and find information on topics of interest. The ID portal provides access to a collection of tools and educational resources, along with opportunities for project-specific interactions. Currently, this portal is password-protected, limiting access to members of the community. Network design. Graduates of the first ID course wished to continue communicating and cooperating with each other after the course ended in 2017. With that desire in mind, a professional network was thought to be a way to continue engagement and a network name was selected: PRINCIPAL, a Spanish acronym for Prevencionistas e Infectologos de Cancer Pediátrico en America Latina, meaning Preventionists and Infectologists for Pediatric Cancer in Latin America. Moreover, basic concepts, such as the vision, mission, and values, were established in discussion with course participants. After several iterations and corrections, the current vision, mission, and values were finalized and are provided in Table . Network development. Once the guiding concepts were formulated, we addressed issues such as membership eligibility and benefits, network activities, and communication venues, including internet messaging (WhatsApp). One early benefit of membership was the ability to obtain the support of fellow members as speakers at IC&P conferences and workshops to impart knowledge in IC&P in pediatric oncology. Requirements for membership included the following: employment or privileges at a health care facility that provides care to children with cancer or other catastrophic illnesses; involvement in IC&P activities; agreement to collaborate with Global ID and network members on strategic goals; and permission to engage in network activities from a direct work supervisor. The benefits of being a member now include access to additional education and training resources through Global ID and St Jude Global, opportunities to obtain support for travel to regional and international scientific meetings, opportunities to collaborate on multicenter research and collaborative studies, and the availability of support for study design and data management. To engage network members and increase their knowledge of and expertise in IC&P in the areas of clinical care and research, we created the following resources: Case-based learning: We established a weekly event, Kiosko de Casos , to improve IC&P professional skills through the discussion of difficult and complex patient cases. These cases mainly involve patients with cancer or recipients of hematopoietic cell transplants. Research in human subject certification: We stimulated network members to become certified in human subject research by providing access to the Collaborative Institutional Training Initiative program. Collaborative Institutional Training Initiative training is mandatory for participating in network-associated research activities but is voluntary for other interested members. Quality improvement skills: We partnered with a Lean 6-Sigma instructor through the St Jude Department of Quality and Patient Safety to collaborate on and deliver a 6-month distance-learning training program through which participants earned Green Belt certification. Cross-collaboration: The PRINCIPAL network has become a focal point for expertise in IC&P in immunocompromised children. As a result, network members have participated in multiple institutional, national, and regional IC&P meetings and conferences, collaborating with institutions and organizers of scientific meetings, short courses, and seminars as speakers on identified gaps in IC&P topics in immunocompromised patients (Table ). Presentation of professional work: We have encouraged members to share local experiences, the results of clinical research, and the outcomes of quality improvement efforts in IC&P at scientific conferences, particularly at the St Jude/PIDS GHS (Table ). Participating in research related to IC&P: We have also encouraged network members to establish and/or collaborate in clinical or basic research and qualitative studies. - Network annual meeting: Since the network was formally established in 2017, an annual meeting has been scheduled every year thereafter. The meeting was held in person before the pandemic (2018-2019) and was virtual in 2020-2022. In these meetings, we review the state of the network and provide reports on achievements and future directions. Monthly bulletin: We produce a monthly bulletin in which we announce upcoming events and summarize the results of ongoing activities. Every other month, we invite network members to contribute a review of a recently published IC&P paper, to share their experiences in IC&P, and/or to communicate some important facts regarding IC&P. This bulletin is distributed to all graduates of our courses and other interested individuals. After publication, the issues are archived and are accessible through our ID portal. PRINCIPAL network structure and leadership. The PRINCIPAL network activities are coordinated by the Global ID program with the input of members. This includes enrolling new members, coordinating meetings, finding speakers, and assisting working groups (WGs) in executing their projects. To integrate the PRINCIPAL network into a global structure, representatives of this network serve on the recently formed Global ID network steering committee (SC). This committee will provide a leadership structure that will in turn lead to a more sustainable and participatory organization. Committee members are currently working to outline the duties of the committee, as well as the roles of the WGs. Leadership consists of network sponsors, SC members, and WG leaders. The WGs will execute projects as they emerge, on the basis of the interest and resources available. Support of these WGs will be essential to ensure productivity across the domains of education, capacity building, and research. Members. Prospective members of the network must be qualified by their training, be involved in pediatric oncology care, and have the support of their workplace. Network membership is free of charge but requires completing an application and periodic renewal. The benefits of membership include invitations to all network-related activities and initiatives, as well as leadership participation. Members also have the opportunity to apply for travel awards for attending scientific conferences and courses, and they receive assistance with finding support for project execution and publication. Sponsors. The sponsors speak about and advocate for the network. They promote and discuss its function and accomplishments, and they assist with matters relating to representation, the procurement of resources, and funding. Currently, the Global ID program is the main sponsor of the network. Funding. Department and programmatic resources support network operations and maintenance. Educational programs and conference awards are funded through the Global ID budget, but members are encouraged to apply for conference-sponsored awards as well as other funding opportunities circulated by the network. Global ID network SC. The SC deliberates, makes decisions, advises, provides strategic oversight, and serves as the primary advocate for initiatives carried out by WGs. Their roles are to build, review, and amend (if needed) long- and short-term goals, objectives, and strategies; to guide network activities on the basis of these goals, objectives, and strategies; to monitor network activities; and to develop ideas to advance network objectives by conducting periodic evaluations of the network. The members of the inaugural SC were selected by the sponsors, and subsequent members will be selected by the departing SC members and the sponsors. SC will facilitate the transition of network activity ownership to its members. This will include, but will not be limited to, leading and managing network project initiatives and organizing coordination and administrative support. Working groups. The performance of the network will be reflected in the participation, productivity, and diverse themes of the WGs, categorized into three groups: education, capacity building, and research. These WGs, which are highly collaborative, will be led by interested members and may be composed of fellow network members or nonmembers. Before activation, the WG proposal is reviewed and approved by the SC. To date, there are three operational WGs in the global network and one WG exclusively for the PRINCIPAL network region (the use of a clinical care pathway in the initial management of a pediatric patient with fever). Graduates of the first ID course wished to continue communicating and cooperating with each other after the course ended in 2017. With that desire in mind, a professional network was thought to be a way to continue engagement and a network name was selected: PRINCIPAL, a Spanish acronym for Prevencionistas e Infectologos de Cancer Pediátrico en America Latina, meaning Preventionists and Infectologists for Pediatric Cancer in Latin America. Moreover, basic concepts, such as the vision, mission, and values, were established in discussion with course participants. After several iterations and corrections, the current vision, mission, and values were finalized and are provided in Table . Once the guiding concepts were formulated, we addressed issues such as membership eligibility and benefits, network activities, and communication venues, including internet messaging (WhatsApp). One early benefit of membership was the ability to obtain the support of fellow members as speakers at IC&P conferences and workshops to impart knowledge in IC&P in pediatric oncology. Requirements for membership included the following: employment or privileges at a health care facility that provides care to children with cancer or other catastrophic illnesses; involvement in IC&P activities; agreement to collaborate with Global ID and network members on strategic goals; and permission to engage in network activities from a direct work supervisor. The benefits of being a member now include access to additional education and training resources through Global ID and St Jude Global, opportunities to obtain support for travel to regional and international scientific meetings, opportunities to collaborate on multicenter research and collaborative studies, and the availability of support for study design and data management. To engage network members and increase their knowledge of and expertise in IC&P in the areas of clinical care and research, we created the following resources: Case-based learning: We established a weekly event, Kiosko de Casos , to improve IC&P professional skills through the discussion of difficult and complex patient cases. These cases mainly involve patients with cancer or recipients of hematopoietic cell transplants. Research in human subject certification: We stimulated network members to become certified in human subject research by providing access to the Collaborative Institutional Training Initiative program. Collaborative Institutional Training Initiative training is mandatory for participating in network-associated research activities but is voluntary for other interested members. Quality improvement skills: We partnered with a Lean 6-Sigma instructor through the St Jude Department of Quality and Patient Safety to collaborate on and deliver a 6-month distance-learning training program through which participants earned Green Belt certification. Cross-collaboration: The PRINCIPAL network has become a focal point for expertise in IC&P in immunocompromised children. As a result, network members have participated in multiple institutional, national, and regional IC&P meetings and conferences, collaborating with institutions and organizers of scientific meetings, short courses, and seminars as speakers on identified gaps in IC&P topics in immunocompromised patients (Table ). Presentation of professional work: We have encouraged members to share local experiences, the results of clinical research, and the outcomes of quality improvement efforts in IC&P at scientific conferences, particularly at the St Jude/PIDS GHS (Table ). Participating in research related to IC&P: We have also encouraged network members to establish and/or collaborate in clinical or basic research and qualitative studies. - Network annual meeting: Since the network was formally established in 2017, an annual meeting has been scheduled every year thereafter. The meeting was held in person before the pandemic (2018-2019) and was virtual in 2020-2022. In these meetings, we review the state of the network and provide reports on achievements and future directions. Monthly bulletin: We produce a monthly bulletin in which we announce upcoming events and summarize the results of ongoing activities. Every other month, we invite network members to contribute a review of a recently published IC&P paper, to share their experiences in IC&P, and/or to communicate some important facts regarding IC&P. This bulletin is distributed to all graduates of our courses and other interested individuals. After publication, the issues are archived and are accessible through our ID portal. The PRINCIPAL network activities are coordinated by the Global ID program with the input of members. This includes enrolling new members, coordinating meetings, finding speakers, and assisting working groups (WGs) in executing their projects. To integrate the PRINCIPAL network into a global structure, representatives of this network serve on the recently formed Global ID network steering committee (SC). This committee will provide a leadership structure that will in turn lead to a more sustainable and participatory organization. Committee members are currently working to outline the duties of the committee, as well as the roles of the WGs. Leadership consists of network sponsors, SC members, and WG leaders. The WGs will execute projects as they emerge, on the basis of the interest and resources available. Support of these WGs will be essential to ensure productivity across the domains of education, capacity building, and research. Members. Prospective members of the network must be qualified by their training, be involved in pediatric oncology care, and have the support of their workplace. Network membership is free of charge but requires completing an application and periodic renewal. The benefits of membership include invitations to all network-related activities and initiatives, as well as leadership participation. Members also have the opportunity to apply for travel awards for attending scientific conferences and courses, and they receive assistance with finding support for project execution and publication. Sponsors. The sponsors speak about and advocate for the network. They promote and discuss its function and accomplishments, and they assist with matters relating to representation, the procurement of resources, and funding. Currently, the Global ID program is the main sponsor of the network. Funding. Department and programmatic resources support network operations and maintenance. Educational programs and conference awards are funded through the Global ID budget, but members are encouraged to apply for conference-sponsored awards as well as other funding opportunities circulated by the network. Global ID network SC. The SC deliberates, makes decisions, advises, provides strategic oversight, and serves as the primary advocate for initiatives carried out by WGs. Their roles are to build, review, and amend (if needed) long- and short-term goals, objectives, and strategies; to guide network activities on the basis of these goals, objectives, and strategies; to monitor network activities; and to develop ideas to advance network objectives by conducting periodic evaluations of the network. The members of the inaugural SC were selected by the sponsors, and subsequent members will be selected by the departing SC members and the sponsors. SC will facilitate the transition of network activity ownership to its members. This will include, but will not be limited to, leading and managing network project initiatives and organizing coordination and administrative support. Working groups. The performance of the network will be reflected in the participation, productivity, and diverse themes of the WGs, categorized into three groups: education, capacity building, and research. These WGs, which are highly collaborative, will be led by interested members and may be composed of fellow network members or nonmembers. Before activation, the WG proposal is reviewed and approved by the SC. To date, there are three operational WGs in the global network and one WG exclusively for the PRINCIPAL network region (the use of a clinical care pathway in the initial management of a pediatric patient with fever). Prospective members of the network must be qualified by their training, be involved in pediatric oncology care, and have the support of their workplace. Network membership is free of charge but requires completing an application and periodic renewal. The benefits of membership include invitations to all network-related activities and initiatives, as well as leadership participation. Members also have the opportunity to apply for travel awards for attending scientific conferences and courses, and they receive assistance with finding support for project execution and publication. The sponsors speak about and advocate for the network. They promote and discuss its function and accomplishments, and they assist with matters relating to representation, the procurement of resources, and funding. Currently, the Global ID program is the main sponsor of the network. Department and programmatic resources support network operations and maintenance. Educational programs and conference awards are funded through the Global ID budget, but members are encouraged to apply for conference-sponsored awards as well as other funding opportunities circulated by the network. The SC deliberates, makes decisions, advises, provides strategic oversight, and serves as the primary advocate for initiatives carried out by WGs. Their roles are to build, review, and amend (if needed) long- and short-term goals, objectives, and strategies; to guide network activities on the basis of these goals, objectives, and strategies; to monitor network activities; and to develop ideas to advance network objectives by conducting periodic evaluations of the network. The members of the inaugural SC were selected by the sponsors, and subsequent members will be selected by the departing SC members and the sponsors. SC will facilitate the transition of network activity ownership to its members. This will include, but will not be limited to, leading and managing network project initiatives and organizing coordination and administrative support. The performance of the network will be reflected in the participation, productivity, and diverse themes of the WGs, categorized into three groups: education, capacity building, and research. These WGs, which are highly collaborative, will be led by interested members and may be composed of fellow network members or nonmembers. Before activation, the WG proposal is reviewed and approved by the SC. To date, there are three operational WGs in the global network and one WG exclusively for the PRINCIPAL network region (the use of a clinical care pathway in the initial management of a pediatric patient with fever). Network outcomes. Since inception, network productivity has been reflected in the number of members participating in various activities in the areas of education, collaboration, advocacy, and research. Importantly, network members have shared their work at meetings through abstracts and poster/oral presentations (Tables and ) and in selected publications (Table ). We anticipate that the lasting effect of the network will be an improvement in the use of evidence-based practice and data standardization. Report standardization begins by teaching disease definitions and their use in our courses, , which can be followed by mentoring IC&P teams at local sites. Recently, we concluded a 3-year mentoring project of IC&P teams, members of the PRINCIPAL network, in three hospitals on Hispaniola Island, two in the Dominican Republic, and one in Haiti. - A PRINCIPAL network member also concluded the experience in standardizing central line–associated bloodstream infections in another institution in Latin America. The ID portal provides resources and training on disease definitions as well. Building these resources for data management and interpretation is ongoing. Network challenges and barriers. The emergence and resolution of the multiple challenges and barriers during the life of the network have been dynamic. These were finding a model for building a network such as PRINCIPAL; developing suitable and attractive activities for members; developing performance evaluation indicators (process and outcomes affecting patient care); building and maintaining a framework to respond to membership demands; defining leadership structure and roles and responsibilities of members and administrative personnel; and developing a plan to sustain the network. These shortcomings were addressed by consulting with experts, using helpful ideas from published literature, collaborators, and importantly from members of the network. Future Directions and Sustainability We expect the PRINCIPAL network to be a venue for building and bringing expertise to pediatric oncology centers throughout Latin America. The PRINCIPAL network must continue to grow and support local and regional expertise in IC&P in pediatric cancer. The members are a critical resource for the clinical workforce, educators, and investigators in IC&P in pediatric cancer. Importantly, through the network, members can unify their voices, message, and efforts at the institutional, country, and regional level and bring their collective thoughts as a network to the global forum. To capitalize on the enthusiasm of new members, robust support for education, capacity building, and research must be in place. There is also a clear need for the network to provide mentoring to WGs. To meet this need, the SC and WG members are identifying partnerships with other networks, professional organizations, and higher education institutions that will result in mutual benefits. PRINCIPAL network members have begun this type of engagement with the International Society for Pediatric Oncology (Société Internationale d'Oncologie Pédiatrique) Supportive Care and the Société Internationale d'Oncologie Pédiatrique Global Health networks in IC&P in pediatric oncology. These networks are excellent sources of expertise in patient care and collaborators in IC&P. The PRINCIPAL network is evolving as a model for establishing collaboration, data sharing, resource procurement, and IC&P proficiency globally. Since inception, network productivity has been reflected in the number of members participating in various activities in the areas of education, collaboration, advocacy, and research. Importantly, network members have shared their work at meetings through abstracts and poster/oral presentations (Tables and ) and in selected publications (Table ). We anticipate that the lasting effect of the network will be an improvement in the use of evidence-based practice and data standardization. Report standardization begins by teaching disease definitions and their use in our courses, , which can be followed by mentoring IC&P teams at local sites. Recently, we concluded a 3-year mentoring project of IC&P teams, members of the PRINCIPAL network, in three hospitals on Hispaniola Island, two in the Dominican Republic, and one in Haiti. - A PRINCIPAL network member also concluded the experience in standardizing central line–associated bloodstream infections in another institution in Latin America. The ID portal provides resources and training on disease definitions as well. Building these resources for data management and interpretation is ongoing. The emergence and resolution of the multiple challenges and barriers during the life of the network have been dynamic. These were finding a model for building a network such as PRINCIPAL; developing suitable and attractive activities for members; developing performance evaluation indicators (process and outcomes affecting patient care); building and maintaining a framework to respond to membership demands; defining leadership structure and roles and responsibilities of members and administrative personnel; and developing a plan to sustain the network. These shortcomings were addressed by consulting with experts, using helpful ideas from published literature, collaborators, and importantly from members of the network. We expect the PRINCIPAL network to be a venue for building and bringing expertise to pediatric oncology centers throughout Latin America. The PRINCIPAL network must continue to grow and support local and regional expertise in IC&P in pediatric cancer. The members are a critical resource for the clinical workforce, educators, and investigators in IC&P in pediatric cancer. Importantly, through the network, members can unify their voices, message, and efforts at the institutional, country, and regional level and bring their collective thoughts as a network to the global forum. To capitalize on the enthusiasm of new members, robust support for education, capacity building, and research must be in place. There is also a clear need for the network to provide mentoring to WGs. To meet this need, the SC and WG members are identifying partnerships with other networks, professional organizations, and higher education institutions that will result in mutual benefits. PRINCIPAL network members have begun this type of engagement with the International Society for Pediatric Oncology (Société Internationale d'Oncologie Pédiatrique) Supportive Care and the Société Internationale d'Oncologie Pédiatrique Global Health networks in IC&P in pediatric oncology. These networks are excellent sources of expertise in patient care and collaborators in IC&P. The PRINCIPAL network is evolving as a model for establishing collaboration, data sharing, resource procurement, and IC&P proficiency globally. In a collaborative undertaking, we designed and operationalized a regional network for professionals involved in IC&P in Latin America. This network, initially in response to the desire of initial members to continue engaging with each other after finishing a training course, is becoming a model for the global engagement of professionals in sharing ideas and opportunities for collaboration in IC&P and serves as a model for creating other regional networks. Born of necessity, assembling the network incorporated the essentials, namely the mission, vision, and values, and the rules for engaging stakeholders. Well-functioning networks can speed communication, grow collaboration, and rapidly introduce changes, providing benefits unavailable to individuals working alone. These qualities are needed to confront the ever-growing threats to global health, including the ongoing COVID-19 pandemic. Other professional networks have been created in response to a natural disaster or to study a disease common in children. , As with our network, the structure of these networks became more defined over time. The participation of members in identifying the optimal structure and continuously striving to ensure the network satisfies the growing needs of its members can guide not only its inauguration but also measures to sustain it. The PRINCIPAL network and its members aim to improve the outcomes of infectious threats in the pediatric oncology population at engaged institutions in Latin America. In REKAMLATINA (Red de Enfermedad de Kawasaki en América Latina), members identified challenges in dealing with Kawasaki disease in Latin America and suggested actions to improve outcomes. , Health care professions such as nursing see structured gatherings as important ways to stimulate conversations, learn about emerging ideas, and find support for the needs of members. The Federation of Gynecology and Obstetrics, a global advocate for the health of women and children, originated as a structure to address the pressing need for measures to prevent maternal and newborn deaths in Africa and Southeast Asia. However, as with our network, the engagement of members continued and expanded, and the Federation of Gynecology and Obstetrics is currently the only organization that brings together professional societies of obstetricians and gynecologists on a global basis. Guided by its foundational mission, vision, and values, and through its inclusive leadership structure, the PRINCIPAL network will engage its members and sponsors, focusing specifically, but not exclusively, on pediatric cancer regionally and globally. Building expertise by participating in IC&P for patients with a rare disease such as cancer can take a long time. Furthermore, the variety and complexity of pediatric cancers and the rapid evolution of new therapies and emerging complications necessitate continuous learning. By using its resources, our network will enable its members to share their experience and knowledge of IC&P in pediatric cancer and will speed the learning process, thereby developing local expertise necessary for better IC&P outcomes in a given region. Implementing the PRINCIPAL network included challenges, but sustaining it imposes additional factors to overcome. A survey of the specialty of Pediatric Infectious Diseases in Latin America found that scarcity of support personnel, time availability, outside-the-specialty medical duties at multiple institutions, high patient volume, and lack of protective time for research and administrative duties were limiting factors for pediatric ID physicians' participation in academic endeavors including those promoted by the PRINCIPAL network. Furthermore, the infrastructure for research and data collection is often suboptimal at many participating institutions, and ethics committees may not be familiar with the types of research conducted by network members. Therefore, participation in WGs often requires effort characterized by excellent time management, temperate character, and high motivation. The benefits of participating in the PRINCIPAL network are quickly emerging. Outcomes amenable to measurement include the number of members participating in conferences, research collaborations, and publications. The intangible benefits for members are far more abundant and difficult to measure, but they are no less important. They include developing new skills and competencies, sharing expertise and knowledge, linking up with colleagues, and contributing to regional proficiency and resources. According to Mata et al, the values of belonging to a professional organization are numerous. These benefits are especially useful for young professionals, for whom network activities can provide a place for interaction, sharing, and finding collaborators and mentors, especially in areas of common interest. Another important benefit of the network is the ability to gain access to members from all areas of a vast region such as the Americas for collaborative studies, especially ones aimed at learning the regional status of health or health care. It is frequently recommended that individuals initiating academic work, especially in health sciences, network for collaborations. Despite the intricacies of some collaborative arrangements, learning how to navigate them is crucial for healthy growth of the network. , Recognizing the needs of network members and their role in education, capacity building, and research as part of building the network, the sponsors offered to provide additional assistance with respect to improving clinical skills, thematic training sessions, quality improvement, and leadership. Building a professional network has stimulated knowledge exchange, cross-mentoring, and collaboration among our course graduates. Strengthening the network structure and operations, recruiting course graduates as members, and retaining them in the network have emerged as essential to the long-term objectives of our course. Following the course, we assessed members' confidence with clinical care, collaborations, and academic productivity in IC&P, and continue to monitor these outcomes. To optimize the use of limited resources and to ensure the success of health care interventions, health care providers in pediatric cancer centers in LMICs, being themselves limited in number, need to be master clinicians and need to know about the best practices in IC&P. Therefore, improving the knowledge of IC&P among health care providers is the first step in building an effective institutional workforce and sustaining the quality of health care for pediatric patients with cancer. We expect that better expertise will promote better patient care and better quality of educational content and that it will stimulate studies about local issues and areas of interest. Considering that, in any country, the number of professionals with capabilities in IC&P for pediatric cancer are concentrated in a few tertiary care hospitals, the effect of improving the quality of IC&P professionals can be considerable. A similar effect can occur with training to promote research leadership skills. Other focused specialties, such as pediatric critical care, stress the importance of partnerships and networking to promote the global specialty agenda. The PRINCIPAL network provides members with a suitable environment in which to communicate their research and, importantly, to identify best practices and opportunities for improvement. As network members, they can collaborate widely with members from other regions, uniting their voices to present their concerns to global agencies to advocate for greater commitment for improving care. The resources provided by network member contributions include building and teaching in trainings, speaking at conferences, providing expertise for difficult IC&P cases, and collaborating research initiatives. Our training courses and the resulting community of graduates operating in networks can build and sustain IC&P knowledge and expertise globally. Participating in our IC&P networks will augment attendees' professional contributions locally, stimulate their leadership potential, and encourage their collaboration in professional networks. Ongoing efforts continue to target IC&P training, thereby strengthening local human capacity and augmenting local and collective expertise, which results in improved care and prevention of infection in children with cancer. In conclusion, the St Jude Global ID program built a regional network, PRINCIPAL, to engage members, build their expertise, disseminate new IC&P evidence-based information on best practices, and collaborate in clinical, quality improvement, and implementation research. A professional network for health care professionals who are interested in and dedicated to improving IC&P will rapidly enhance the impact of educational, capacity building, and research interventions in a region. The PRINCIPAL network is a model for building a similar network in any geographic region that aims to rapidly deploy expertise building and collaboration.
Using Artificial Intelligence for Optimization of the Processes and Resource Utilization in Radiotherapy	63cf4f72-8507-4e72-b7d9-6a2fd5d9d87d	10166445	Internal Medicine[mh]	The anticipated increase in cancer burden over the next few years could potentially overwhelm the oncology care system, especially in resource-constrained low- and middle-income countries (LMICs). , Although radiotherapy (RT) is an indispensable component of cancer care, access to it worldwide is very inequitable, with the current density of RT machines per million population ranging from 0 to 11.6, depending on the economic situation of a country. The RT process starts with a series of visits to radiation oncology (RO) clinic, culminating in the final diagnosis, staging, and prognostication after which a radiation treatment protocol is assigned. Once a protocol is assigned, the subsequent RT treatment process can be categorized into imaging, target and organs-at-risk (OARs) segmentation, treatment plan generation, onboard imaging, treatment delivery, and quality assurance (QA) checks. These steps are labor-intensive and time-consuming, requiring multiple levels of human-machine interaction and a high degree of precision. The patient continues to visit the clinic on conclusion of therapy for toxicity management and follow-up. This workflow is summarized in Figure . CONTEXT Key Objective We review the role of artificial intelligence (AI) and machine learning (ML) in improving the efficiency of various radiotherapy processes and the challenges in their clinical integration. Knowledge Generated AI and ML can improve the accuracy, robustness, and speed of radiotherapy processes by reducing or eliminating human interference, aiding in decision making, and efficiently executing time-consuming, repetitive tasks. As its clinical utility remains yet to be proven, multi-institutional collaborative effort between various stakeholders is urgently needed, before the revolutionary impact of AI and ML bears fruition. Relevance The anticipated increase in cancer burden over the next few years coupled with cancer care becoming more personalized and tailored could potentially pressurize the oncology care system in the years to come, especially in low- and middle-income countries. The RT workflow requires meticulous coordination between trained medical professionals with diverse expertise, i.e., radiation oncologists, medical physicists, dosimetrists, and radiation therapists. Understaffing and workforce burnout is, unfortunately, a common problem plaguing RO in LMICs heightened by the ongoing COVID-19 pandemic. Training the highly specialized RO workforce requires a high cost and time commitment. The role of artificial intelligence (AI) and machine learning (ML) in optimizing RT processes to achieve the best human, technological, and financial resource utilization is worth exploring. Key Objective We review the role of artificial intelligence (AI) and machine learning (ML) in improving the efficiency of various radiotherapy processes and the challenges in their clinical integration. Knowledge Generated AI and ML can improve the accuracy, robustness, and speed of radiotherapy processes by reducing or eliminating human interference, aiding in decision making, and efficiently executing time-consuming, repetitive tasks. As its clinical utility remains yet to be proven, multi-institutional collaborative effort between various stakeholders is urgently needed, before the revolutionary impact of AI and ML bears fruition. Relevance The anticipated increase in cancer burden over the next few years coupled with cancer care becoming more personalized and tailored could potentially pressurize the oncology care system in the years to come, especially in low- and middle-income countries. Russell and Norvig have defined AI as “the designing and building of intelligent agents that receive precepts from the environment and take actions that affect that environment.” A more perceptive definition given by Goel is “the science of building artificial minds by understanding how natural minds work and understanding how natural minds work by building artificial minds.” ML is a branch of AI that allows computer systems to progressively learn, train, and improve on the knowledge gained from a variety of input data without being overtly programed. AI and ML can improve the accuracy, robustness, and speed of RT processes by reducing or eliminating human interference, aiding in decision making, and efficiently executing lengthy, repetitive tasks. Using ML can free up time for more rewarding tasks such as education, research, patient counseling, and quality checks. The following review focuses on the potential use of ML and AI to transform the existing RT workflow and create a sustainable model that can be adopted in LMICs to supplement human efforts in labor-intensive tasks: segmentation, planning, and QA. Manual segmentation (or contouring) of the target and OARs is a time-consuming and highly subjective task that lies at the core of RT planning. Historical solutions offered by technology to this conundrum include edge- and region-based methods and atlas-based methods of autosegmentation. Deep learning (DL), a subset of ML, is essentially a neural network with three or more layers. These can comprise simple feed-forward models such as artificial neural network or complex models such as convolutional neural networks (CNN) and recurrent neural networks. CNN has been increasingly used to learn complex nonlinear relationships within the imaging data to speed up and improve OARs delineation in mediastinum, pelvis, thorax, brain, and head/neck. Image segmentation on the basis of DL uses either patches or regions of an image or the entire image as input to estimate the likelihood that a given image sample belongs to the object being segmented. The likelihood map can be further enhanced by methods combining DL and deformable models. Multiple papers published using ML for autosegmentation of OARs have demonstrated no clinically meaningful difference between segmentation by model and clinicians or radiographers with very high values of dice similarity coefficient, while recording a significant reduction in time needed for the segmentation. - For example, the average segmentation time for abdominal OARs liver, stomach, duodenum, and kidneys was 7.1 minutes with automation versus 22.6 minutes when done manually. Lesion segmentation is more complex than OARs segmentation because of the heterogeneity in shape, size, and location of the target and, therefore, is still in nascent stages. Computer-aided diagnosis methods, including conventional radiomics and CNN-based algorithms that enhance lesion detection in diagnostic radiology, can potentially be used in lesion segmentation during RT planning. The advent of intensity-modulated RT and volumetric modulated arc therapy that offer exceptionally conformal RT delivery has increased manifold the intricacy and complexity of RT planning. High-precision treatment procedures such as stereotactic body ablative RT often consume hours or even days of human effort for planning. Knowledge-based planning (KBP), which uses data from previous good cases to inform current patient planning parameters, has emerged as a powerful tool to accelerate the process of RT planning. Efforts are ongoing to establish indigenous KBP models for cancers common in LMICs, such as cervical cancers, and validate them in various geoethnic populations to test efficacy in patients with different anatomies on the basis of geographical locations. , Supervised DL algorithms have been used for beam direction optimization, where the possible subsequent beam distribution is predicted on the basis of patient anatomy. Use of DL in the prediction of spatial dose distribution has been extensively explored, with different architectures of CNN being used to predict the geometric and planning parameters of historical patients. A significant gain in time has been reported with ML over non-ML methods such as column generation to select beam orientations, calculate the dose influence matrices, and finally solve the fluence map optimization with comparable dosimetry. ML algorithms have also been used to enhance KBP further to generate treatment plans. Recent studies have even attempted to emulate the decision-making strategy of human planners when solving a specific dosimetric trade-off problem, thereby potentially reducing the element of subjectivity. Another unique approach is the use of Pareto surface–based techniques for multicriteria optimization, where a database of plans is created for a single patient and the plan that achieves the best balance between different treatment planning goals is chosen by the planner and the physician. , The Erasmus i-cycle (created in an academic university) is a vendor-neutral algorithm using multicriteria optimization that is in clinical use for external beam therapy and is being developed for CyberKnife, proton therapy, and brachytherapy (BiCyle). - ML techniques, including DL approaches, have dealt with intra- and interfraction patient and organ motion during RT treatment delivery to aid tumor gating and motion tracking. Frameworks have been built using neural networks trained on collected patient breathing data to predict the breathing pattern while delivering RT. ML has been used to aid motion tracking by assisting in the detection of the tumor (marker-less tracking) or surrogate markers. ML has been used to help avert setup errors and patient safety hazards by tracking the treatment room components and the patient's body in real time using 3D cameras to fine tune a CNN for object recognition. A group of scientists have developed a computer vision–based pneumatic soft robot actuator to better estimate a patient's head pitch motion and to manipulate the patient head position on the basis of sensed head pitch motion, thereby potentially eliminating the need for immobilization with a thermoplastic mask. The role of ML in online adaptive RT planning has been extensively explored, mainly in deformable registration and dose warping, facilitating high registration accuracy and efficient execution even if graphical processing units are unavailable. A proof-of-concept study investigates online multileaf collimator tracking to generate appropriate safety margins for online adaptation of the treatment plan on the basis of the patient's motion and the ability of the machine to follow these excursions. Algorithms can assist physicians in supervising variations during treatment course by evaluating daily setup variations and anatomic changes, for early identification of adaptive replanning requirement. Implementing regular and meticulous QA in RT is expected to lead to more accurate treatment delivery and better clinical outcomes. ML has excellent potential to enhance the efficacy and efficiency of RT QA processes as they are often repetitive and time-consuming. ML techniques have been used to predict gamma passing rates and the probability of the plan failing patient-specific intensity-modulated RT QA by analyzing plan complexity; multiple components of the delivery system such as multileaf collimator, imaging system, and mechanical and dosimetric parameters, and plan delivery log files over time. - Another approach has identified RT treatment delivery errors using radiomics-based feature extraction from patient-specific gamma images. A study has applied artificial neural network–based time series prediction modeling to predict the performance of beam symmetry of linear accelerators over time. Although these in silico approaches of various ML tools have augmented RT QA procedures, it is pertinent to establish its real utility in the clinical context before implementation. Data Annotation, Radiomics, and Response Prediction Radiomics is a method that extracts a large number of features from medical images using data characterization algorithms. Many institutions and health networks, including from India, are working to create repositories of annotated medical data and medical images including outcomes of treatment for furthering radiomic research in large image data sets. , Distributed learning approaches with AI support have been used to conduct population-based studies on routine data and build decision support models. , Image banking combined with predictive/prescriptive AI is a cost-effective and efficient alternative to identify signatures for response, toxicity, and outcome prediction after cancer treatment. - Natural Language Processing Natural language processing (NLP) is a branch of AI that enables computers to interpret human language. NLP has already found application in the medical world to facilitate data extraction from free text in electronic medical records. The specific utility of NLP being explored in RO is standardization of contours and plans nomenclature to enable efficient data extraction. Radiomics is a method that extracts a large number of features from medical images using data characterization algorithms. Many institutions and health networks, including from India, are working to create repositories of annotated medical data and medical images including outcomes of treatment for furthering radiomic research in large image data sets. , Distributed learning approaches with AI support have been used to conduct population-based studies on routine data and build decision support models. , Image banking combined with predictive/prescriptive AI is a cost-effective and efficient alternative to identify signatures for response, toxicity, and outcome prediction after cancer treatment. - Natural language processing (NLP) is a branch of AI that enables computers to interpret human language. NLP has already found application in the medical world to facilitate data extraction from free text in electronic medical records. The specific utility of NLP being explored in RO is standardization of contours and plans nomenclature to enable efficient data extraction. We searched the PubMed database using the search term (Artificial Intelligence[tw] OR Machine Learning[tw] OR Deep Learning[tw] OR Automation[tw] automated[tw] OR knowledge-based planning[tw] AND Radiotherapy[tw]) AND (list of Low- and Middle-Income Countries as defined by the World Bank at the time of writing this review). The search yielded a total of 90 results, of which results with first authors from the LMICs were chosen. The reference lists of retrieved articles were also reviewed to search for more studies. No language restrictions were imposed. A total of 20 research items with unique study objectives conducted with the aim of enhancing RT processes were studied in detail and are presented in Table . The majority of studies have focused on the utilization of CNN and other networks for autosegmentation. - The striking reduction in time burden seen with incorporating these algorithms while maintaining the accuracy of contours can prove to be pivotal in resource allocation in LMICs. The studies on autoplanning have mainly used KBP. , , , Developing indigenous models with local data rather than adopting western models to fit into the local workflow seems to be the standard approach, which is undoubtedly remarkable. Studies from LMICs using AI/ML to assist in online adaptive planning, treatment delivery, and QA are few and far between, and more work in this area must be encouraged. , ‐ Although AI has already become pervasive in our day-to-day activities and has the potential to influence how medicine is practiced, many challenges remain before its complete integration into RT processes, as listed below. Clinical utility yet to be determined: Most ML-based solutions are still in the stage of technological incubation, with the onus on the RO team to establish their clinical value. Risk analysis of automation and AI: QA studies for automated treatment planning tools have been conducted, which stress the need for comprehensive manual review of the plans by physicians and physicists before implementation. Black-box nature of AI algorithms: In the case of failure of AI-based solutions, there is no straightforward framework to fix the outcome or predict errors. This lack of transparency and difficulty in understanding the outputs and predicting failures may make physicians hesitant and distrustful to rely on AI in patient-related decisions, further delaying the adoption of AI into clinical practice. It is essential to train the RO staff to correctly use the ML model and accurately interpret the intended utility, scope, and limitations. Interpretation: It is necessary to remember that even if some algorithms can perform at near-human ability, the way they perceive and interpret the inputs differs from the human mind. Training data: A machine learning algorithm's accuracy and generalizability are influenced heavily by the quality and quantity of the training data more than the mathematical parameters. Since individual institutional data sets are bound to be minor, data sharing across multiple institutions can make the ML/DL algorithms more robust. Distributed learning is an emerging approach to securely transferring data sets between institutions. Patient privacy and anonymity: The potential of distributed learning to provide evidence-based personalized care in LMICs is immense. However, care should be taken to uphold the rules of ethics, standardization, and stringent privacy regulations. In conclusion, integrating AI and ML in RT processes may allow radiation oncologists to spend more time on patient consultation, teaching, and research in resource-constrained setups with a heavy workload. Given the transformative impact that AI-based technology can bring to clinical processes and workflow, it is essential to integrate these concepts early in medical education and RO residency to facilitate a better understanding of the methods and encourage innovation. We have in front of us a means to revolutionize the practice of RT as we know it. It is our responsibility toward the future generation to understand, plan, prioritize, conduct meaningful research, integrate and constantly improve AI and ML in RT without bias or prejudice, and deliver it to settings where the impact would be maximum.
ASCO's Leadership Development Program: Focusing on the Next Generation of Leaders in Asia Pacific	ea21b70e-b5d9-41b6-b002-2d26c0807348	10166472	Internal Medicine[mh]	Developing new leaders is critical for any organization. The complexity of health care organizations and the pressing challenges of modern health care call for great leadership from within, thus providing the motivation to develop physician leaders. - Traditionally, the criteria set to advance physicians to leadership positions are based on academic and clinical accomplishments. Good clinicians and researchers do not necessarily make good and effective leaders. Distinct competencies for leadership in medicine are different from the competencies and skills needed in the practice of the profession. - There should be a balance between managerial skills focused on efficiency, financial stringency, and human resource optimization and skills focused on patient care. Consequently, the clinical and scientific skills must form part of the managerial focus and are able to be provided if there is an identified management role for physicians. , It is unknown to what extent previous medical training and experiences of doctors affect the performance of physician executives. Previous studies support the importance of including doctors in hospital governing boards. - Given the complexity encountered within modern health care organizations, it is difficult to demonstrate the impact of leadership on outcomes when assessing medical and nonmedical leadership in the same setting. It remains unclear in the literature whether nonfaculty career administrators or clinical staff are better suited. , Studies found no difference in the performance of medical and nonmedical managers. - CONTEXT Key Objective How important is it to build and develop the next generation of oncology leaders in the Asia Pacific region? Knowledge Generated ASCO through its Asia Pacific Regional Council has initiated the Leadership Development Program to create a pool of oncology professionals from the region. Building effective Asian leaders with diverse backgrounds, culture and specializations will bridge the leadership gap and build leadership capacity in Asia, the region with the heaviest cancer burden. Relevance The complexity of modern health care and the demands in the field of oncology call for empowered, competent, and motivated leaders to help improve the quality of cancer care in Asia Pacific. Identifying the fundamental principles of physician leadership is critical. The principles are the core components that characterize leadership competency, and these abilities are teachable skills. - Physician leadership has been recognized to correlate with organizational success. Implementing a physician leadership program cultivates and strengthens the skills of emerging clinical leaders and develops a pipeline of physician leaders. , Oncology is facing enormous challenges and changes. Despite much progress in cancer management, the increasing number of cases and mortality globally is a continuous challenge. This calls for strong leadership from cancer care providers and adequate succession planning to ensure continuity of purpose. ASCO is committed to promoting lifelong learning and professional development of oncology professionals. In 2009, ASCO launched its Leadership Development Program (LDP) where participants learn strategies and critical skillsets, gain exposure to the roles and mission of ASCO, and network with and receive mentorship from ASCO leaders. In 2013, eligibility for the prestigious LDP was extended to international members. Three participants from Asia had been admitted to this program. Although there are different challenges for leaders in countries outside the United States, the LDP provided the participants a strong foundation of leadership skills with impactful and long-lasting personal and professional benefits. Key Objective How important is it to build and develop the next generation of oncology leaders in the Asia Pacific region? Knowledge Generated ASCO through its Asia Pacific Regional Council has initiated the Leadership Development Program to create a pool of oncology professionals from the region. Building effective Asian leaders with diverse backgrounds, culture and specializations will bridge the leadership gap and build leadership capacity in Asia, the region with the heaviest cancer burden. Relevance The complexity of modern health care and the demands in the field of oncology call for empowered, competent, and motivated leaders to help improve the quality of cancer care in Asia Pacific. The Asia Pacific (APAC) region is composed of a diverse group of high-, middle-, and low-income countries. Each has distinctive cultural and socioeconomic backgrounds, which pose unique challenges in delivering high-quality, equitable patient care. - Asia has the heaviest cancer burden because of its high population density comprising 60% of the global population. The region has 50% of cancer cases and is estimated to account for 58% of cancer deaths worldwide. In an effort to learn more about these challenges and understand the landscape of oncology in APAC, ASCO launched the Asia Pacific Regional Council in May 2019. The Council, representing the eastern APAC countries, advises ASCO on the needs of members in the region and facilitates and encourages member involvement in the society's global activities. The Council recognized the need to develop leaders from the region and to further provide members with a new avenue to contribute to ASCO's work. The demand for a high-quality and comprehensive cancer care highlights the challenges and opportunities in developing oncology leaders in APAC. In a study performed by the Center for Creative Leadership, in the top 200 organizations (by revenue), leaders of Asian ethnicity and/or nationality represent only about 4% of the executive teams in US-headquartered companies and 3% in Europe-headquartered firms. It is unknown if similar data would suggest very limited representation of ethnic Asian talent in oncology leadership roles. One explanation on the under-representation of Asians in leadership positions in the United States is an issue of cultural fit. There is a mismatch between East Asian's low assertiveness in communication and the American norms of how leaders should communicate. The Council has been tasked with adapting content and lessons learned from the LDP to offer a new leadership program tailored to the needs of an APAC audience. The first-ever ASCO APAC LDP was thus launched in Nov 2021. The program was funded by the Conquer Cancer, the ASCO Foundation Mission Endowment. With more than half of the world's population, the region's role in the global socioeconomic landscape and in global health is becoming increasingly important. Workplaces are becoming more multicultural. Organizations, companies, and institutions are expanding globally. Asian leaders should be ready to navigate these challenges ahead. There are common demands that leaders face across the world. However, leadership styles and behaviors are different and are shaped within the context of cultural and philosophical views, beliefs, and approaches. Traditionally, Asians have a strong respect for age and seniority. They are conditioned to revere seniors, follow established customs and accepted norms, and avoid opposing viewpoints. Filial piety, an attitude of respect for parents, elders, and ancestors in societies, is influenced by Confucian thought. It has shaped family care giving and other aspects of individual, social, political, and legal relations in Asian countries. Filial attitudes emphasize obedience. It has been related to resistance to cognitive change, which tends to adopt a passive, uncreative, and uncritical orientation toward learning. , Therefore, these attitudes tend to result in loyalty and groupthink and may hamper creative thinking and innovation. The characters of different Asian leadership styles are also reflected not only from the influences of Confucianism but also from Daoism, Mohism, and other philosophy. - Leadership is built on a foundation of moral character and exercised through virtuous examples. A leader or a ruler is a moral manager and a person of benevolence, wisdom, and courage. , A Daoist leader needs to be trustworthy and unbiased and use action over words. Such leadership should promote harmony and balance. The Mohism philosophy and major ethical tenets promote universal, unbiased respect and concern for all regardless of affiliations or relations. Some Asian countries were under colonization and gained freedom from foreign political rule. Most earned their independence through protests and resistance or after going through the ravages of war. Asian leaders had to put much work on changing institutional structures, policies, and laws, which are rooted in the colonial past. There are individuals who may not be used to having the freedom to find and develop their own leadership strengths and style. Some are not accustomed to developing themselves to their full potential. These are historical roots born from decades and even centuries of colonial oppression. Asian countries have inherently humble cultures with strong family and societal bonds. A paternalistic leadership style is built on the premise that dad knows best. This is claimed to be the one dominant leadership style in Asia. Several research studies have looked into the elements of paternalistic leadership. - It entails authoritarianism, benevolence, and moral character. The moral character dimension of paternalistic leadership expects the leader to behave on high moral standards, and the subordinate sees them as a role model, believes in their moral integrity and benevolence, and follows their authoritarian guidance. Authoritarian leadership is characterized by the hierarchical dynamics between the leader's authority, control, and power. It is also marked by the compliance, obedience, and respect of the subordinates. Benevolent leaders provide individualized and holistic concern to the subordinate. A recent meta-analysis found that two dimensions of paternalistic leadership, benevolent leadership and moral leadership, were positively associated with employee innovation. By contrast, the dimension of authoritarian leadership was negatively associated with innovation. There are strategies that stakeholders must adopt to maximize leadership talent in the region. Focusing on strength-based leadership culture will develop leaders with authentic and unique strengths. A strong culture of coaching and mentoring can give some flexibility to the rigid and hierarchical Asian culture. Giving recognition for significant and unique contributions is a powerful motivator for emerging leaders. Building cultural intelligence is particularly important as individuals exposed to working in different environments and cultural contexts adapt to a diverse range of scenarios and situations. The objectives of the LDP are thus to create a pool of oncology professionals from APAC who are empowered to identify and fill leadership roles in oncology, to learn the skills for strategic planning tools and time management that will prepare them for potential future advocacy roles, and to promote interdisciplinary collaboration across the region. The pan-Asian interdisciplinary group of young leaders are expected to gain knowledge and skillsets including Understanding the different types of leaders and how to apply individual strengths to be an effective leader. Knowledge about advocacy, importance of advocacy efforts, and ability to participate in advocacy initiatives. Ability to collaborate with others on regional efforts to improve cancer care. Presenting information to others and delivering media interviews. Effectively managing conflict and ability to have difficult conversations. After the yearlong program, key performance indicators on the success and the impact of the program will look at Collaborative projects fostered between participants (joint regional educational initiatives; regional cancer research, health services research, etc). Appointments to various leadership positions within ASCO, national committees, and regional and international societies and committees. After graduating from the program, participants are able to stay connected to ASCO through opportunities to serve as volunteers in advisory groups, committees, and working groups. They could be in roles where they excel and where their talents are truly leveraged. As they develop senior leadership roles within their own countries and the region, they can also serve as ambassadors between their institutions or local societies and ASCO, organizing or participating in networking events and collaborative initiatives that would keep the engagement and make it a strong and well-connected community. Since the program represents diverse countries, resource settings, gender, and specialties, participants will learn shared views on leadership models from the different Asian countries. The participants will gain knowledge on the key differences that are unique to specific Asian countries. This knowledge will stand them in good stead in formulating culture- and resource-appropriate strategies and guidelines for cancer control activities. The program would also highlight crucial factors for a leader in Asia to succeed, which may not be familiar to faculties from the West. Profiling the APAC Leader in Oncology The selection criteria had to be modified given the diverse backgrounds of the candidates who would come from various countries. However, some of the overarching key criteria for assessment of leadership qualities were retained from the original program, which included leadership roles in their departments, civic engagement, and clinical or research groups; strong track records of new project initiatives in their institutions; active involvement in national society committees; and a strong letter of support from their current mentors and heads of departments. Collectively, the Council intended to ensuring diversity at all levels, be it country, sex, and seniority with the candidates' selection. English language competency was a necessity, given that the program will be conducted in English. Another thing considered during the profiling of the candidates was their disparate environments, which correlated with their country's income status (as defined by the World Bank classification). It was apparent that leaders from institutions of developed countries had more resources to kickstart programs. In the same vein, the leaders from different countries faced divergent sets of challenges; for example, leaders from low- and middle-income countries were restrained by a chronic lack of resources and manpower, - whereas leaders from more developed ecosystems were more vocal about institutional bureaucracy. Nonetheless, ASCO and the Council were open to exploring how the leaders from such diverse backgrounds and environments would interact overall and in their small project groups. Selecting the Pioneer Batch of APAC Young Oncology Leaders As this was the pioneer cohort, there were very limited insights into the background and caliber of the applicants. To encourage applications, publicity was done via several streams—word of mouth by the council members, through the national societies, through electronic dissemination by e-mails to the ASCO international membership, or through social media platforms. In total, there were 45 applicants from 12 different countries, and impressively, the applicants were well-matched in terms of their clinical and leadership track records. This was a testament to the appeal and enthusiasm toward an ASCO-led LDP in APAC. Selection was led primarily by the Council members, who were unanimous in ensuring diversity for the pioneer class. The members prioritized equal opportunities during the selection process, so that candidates were not penalized because of limitations imposed on them by the country that they represent. A challenge that was, however, encountered was the difficulty in organizing interviews for the large number of applicants by the Council. This inevitably resulted in a process that was heavily dependent on the sentiments of the Council to the candidates' curricula vitae and personal statements, leading to some degree of uncertainty surrounding the suitability of the chosen candidates. Going forward, the Council has opined on the need to review this process and to consider the inclusion of a face-to-face interview either in-person or online. Conducting the LDP in a Virtual World Amidst a COVID-19 Pandemic Apart from the candidate selection process, the other big unknown for ASCO and the Council were the feasibility of conducting an online leadership course and whether such a platform is conducive for coaching and mentoring and learning for the leaders. Unlike the US LDP, which consisted of four quarterly sessions of 2-3 days of intensive boot camps, the APAC LDP was split into 12 monthly plenary sessions conducted exclusively online over Zoom. Each plenary session was participated by all the 12 leaders (Table ), along with three mentors, three coaches, two course directors, and ASCO education staff who oversaw the curriculum and conduct of the LDP. At the initial phase of the program, it was evident that an online forum was not ideal to foster interaction between the leaders themselves and with the faculty members. This lack of participation is compounded by the discrepancy in their abilities to communicate in English, which speaks to another blind spot of the candidate interview and screening process—something to be reviewed with the subsequent iterations of the APAC LDP. Nonetheless, participation levels gradually improved over time when the leaders, coaches, and mentors were given the opportunity to share about themselves before every plenary session, which helped to reduce the virtual social barriers. Another inevitable challenge encountered by the leaders, coaches, and mentors is related to the stress and burden of having to undergo an intensive LDP course, while managing their routine clinical and administrative duties during the peaks of the COVID-19 pandemic. At times, it was obvious that some of the leaders were struggling to cope, and this affected the progress of respective small group projects (Table ). Nonetheless, ASCO staff, coaches/mentors, and fellow team-mates were ready to aid and cover for each other to ensure that the projects were able to continue. All things considered, the first version of ASCO APAC LDP was a successful effort to provide training of the future leaders from this part of the world. The selection criteria had to be modified given the diverse backgrounds of the candidates who would come from various countries. However, some of the overarching key criteria for assessment of leadership qualities were retained from the original program, which included leadership roles in their departments, civic engagement, and clinical or research groups; strong track records of new project initiatives in their institutions; active involvement in national society committees; and a strong letter of support from their current mentors and heads of departments. Collectively, the Council intended to ensuring diversity at all levels, be it country, sex, and seniority with the candidates' selection. English language competency was a necessity, given that the program will be conducted in English. Another thing considered during the profiling of the candidates was their disparate environments, which correlated with their country's income status (as defined by the World Bank classification). It was apparent that leaders from institutions of developed countries had more resources to kickstart programs. In the same vein, the leaders from different countries faced divergent sets of challenges; for example, leaders from low- and middle-income countries were restrained by a chronic lack of resources and manpower, - whereas leaders from more developed ecosystems were more vocal about institutional bureaucracy. Nonetheless, ASCO and the Council were open to exploring how the leaders from such diverse backgrounds and environments would interact overall and in their small project groups. As this was the pioneer cohort, there were very limited insights into the background and caliber of the applicants. To encourage applications, publicity was done via several streams—word of mouth by the council members, through the national societies, through electronic dissemination by e-mails to the ASCO international membership, or through social media platforms. In total, there were 45 applicants from 12 different countries, and impressively, the applicants were well-matched in terms of their clinical and leadership track records. This was a testament to the appeal and enthusiasm toward an ASCO-led LDP in APAC. Selection was led primarily by the Council members, who were unanimous in ensuring diversity for the pioneer class. The members prioritized equal opportunities during the selection process, so that candidates were not penalized because of limitations imposed on them by the country that they represent. A challenge that was, however, encountered was the difficulty in organizing interviews for the large number of applicants by the Council. This inevitably resulted in a process that was heavily dependent on the sentiments of the Council to the candidates' curricula vitae and personal statements, leading to some degree of uncertainty surrounding the suitability of the chosen candidates. Going forward, the Council has opined on the need to review this process and to consider the inclusion of a face-to-face interview either in-person or online. Apart from the candidate selection process, the other big unknown for ASCO and the Council were the feasibility of conducting an online leadership course and whether such a platform is conducive for coaching and mentoring and learning for the leaders. Unlike the US LDP, which consisted of four quarterly sessions of 2-3 days of intensive boot camps, the APAC LDP was split into 12 monthly plenary sessions conducted exclusively online over Zoom. Each plenary session was participated by all the 12 leaders (Table ), along with three mentors, three coaches, two course directors, and ASCO education staff who oversaw the curriculum and conduct of the LDP. At the initial phase of the program, it was evident that an online forum was not ideal to foster interaction between the leaders themselves and with the faculty members. This lack of participation is compounded by the discrepancy in their abilities to communicate in English, which speaks to another blind spot of the candidate interview and screening process—something to be reviewed with the subsequent iterations of the APAC LDP. Nonetheless, participation levels gradually improved over time when the leaders, coaches, and mentors were given the opportunity to share about themselves before every plenary session, which helped to reduce the virtual social barriers. Another inevitable challenge encountered by the leaders, coaches, and mentors is related to the stress and burden of having to undergo an intensive LDP course, while managing their routine clinical and administrative duties during the peaks of the COVID-19 pandemic. At times, it was obvious that some of the leaders were struggling to cope, and this affected the progress of respective small group projects (Table ). Nonetheless, ASCO staff, coaches/mentors, and fellow team-mates were ready to aid and cover for each other to ensure that the projects were able to continue. All things considered, the first version of ASCO APAC LDP was a successful effort to provide training of the future leaders from this part of the world. A key component of any leadership course is developing teamwork and cooperation. A pathway to achieve this is through the commissioning of a project for a team to develop and present a finalized proposal. Each project team was selected to ensure diversity of countries, specialties, and sex. The projects (Table ) were put forth by various members of the Council and felt to be relevant to the regions' needs and the background of the selected candidates. Eventually, three projects were finalized on the basis of the highest number of votes by the Council. Fundamental principles in project selection included a clear recognition that the region was diverse in socioeconomic and cultural aspects, that projects had the potential to make an impact on all countries by addressing problems common to all patients with cancer irrespective of where they lived, and that project outcomes incorporated flexibility to suit local needs and had clear tangible outcomes. Each team consisted of one coach, one mentor, and four LDP participants who rotated serving as a team leader. The role of the coach was to provide information that team members requested, educate the team about the project topic, and provide networking leads and connections in support of the team's work on the project. The mentor provided guidance to focus and organize their work; monitored team progress against their plan with regular formal and informal assessments; and facilitated team communication, planning, decision making, conflict resolution, and lessons learned. Teams were encouraged to engage with other LDP teams and mentors. Project-based learning was designed to encourage development of many skills including open communication, respect others’ views, encourage and build on all ideas, identify problems and barriers and explore solutions, and manage effective team meetings. Working on team projects allowed the participants to take ownership and responsibility, by developing models of teamwork and accountability. Team leadership was intended to nurture soft skills related to team management, such as facilitating the team members to define and implement their mission, goals, priorities, roles and accountabilities, operating protocols, and development of collaborative and mutually supportive relationships. Additional skills to be developed included ensuring that the team operates efficiently and that members participate actively in discussions and decision making; facilitating debates but also mitigating conflicts, clarifying issues to be addressed and keeping members focused; stakeholder engagement; and monitoring of members' performance against their assigned tasks, ensuring that they are meeting targets, promoting visibility and recognition of team member contributions and performance. The projects are outcomes-driven, they represent goals that the Council wishes to promote to ensure better care of our patients and better engagement of health services throughout the region in key aspects of cancer treatment, prevention, and control. Equally, the projects are a vehicle to instill leadership qualities through experiential learning and mentoring. Leadership has different styles, perspectives, and philosophical approaches in different parts of the world. Organizational structures of leadership also have some differences. - Western organizations usually have flatter structures and are less prescriptive. Eastern organizations have hierarchical structures and therefore are more directive. Western leaders build open relationships. Superiors and subordinates perceive each other as inherently equal in hierarchy at work. This may mean that work organization, roles, and positions can be changed easily. On the other hand, Asian leaders maintain a distance with the subordinates affecting work organization, structure, and relations between them. Subordinates can have inhibitions in approaching their managers and superiors. Individualistic culture emphasizes the unique personal characteristics of an individual. The needs and motives are the focal point of understanding the individual's actions. Western approach to leadership tends to focus on individual achievement. Collectivistic culture places more emphasis on the identification of an individual within a group, such as roles and duties associated with being a group member. Eastern leadership approach focuses on the collective activities of followers. The ties between members of the group are very strong, and loyalty to the group is one of the basic values. Asian leaders focus on collective activities of followers and collective achievement. The traditional Western style of leadership may not be well suited to an Asian context, and given the high cancer burden, a fit-for-purpose model is essential. Notably, as cancer care globalizes, modern approaches to leadership in the East and the West continue to evolve. Leaders should adopt principles from each other and effectively apply them in any culture, country, or region. APAC is the largest and most populous region with more than 60% of the world's population. There are more than 50 countries with diverse ethnic groups, histories, languages, cultures, environments, economics, and sociopolitical systems. Unfortunately, it has long been under-represented, despite the highest cancer burden, and suffers from a huge unmet need for optimal oncologic care. Effective leadership is particularly important in difficult times. The best way to prepare future leaders in oncology is to get as diverse a set of experiences as possible under their belt. To prepare them to meet today's demands, stakeholders should support their development as a journey of change and growth. Leadership development is not only a yearlong program but also a career-long process. It requires continuous learning with multiple touch points, follow-ups, and accountabilities. The sense of purpose and mission is important to be reinforced. Young leaders are greatly energized and motivated when they are connected to a broader and higher purpose. Institutions and organizations will benefit from a deeper and more sustained focus on leadership development. It is important to focus on the changes that are needed from the perspectives of leaders, rather than investing in showcase programs that may not address the immediate needs of the respective health care systems. Hence, successfully addressing the challenges on leadership development is important as we need a pipeline of future leaders. Organizations must ensure that their talented staff are constantly upskilling, getting ready for the challenges of tomorrow. They must remain committed to developing the skillsets and leadership qualities of the next-generation Asian leaders. These approaches lay the foundation for superior transformation and performance. In conclusion, the demand for high-quality and more comprehensive oncology care will continue to rise across APAC in the years to come. This growth will intensify the need for capable leaders who are able to navigate the complex dynamics of health care services in the countries. The need for an Asian-focused leadership program was exemplified by the wide geographic spread of the participants. Critical capabilities such as collaboration and resilience will be must-haves to prepare the young leaders for whatever future Asia throws at them. The evolving endeavor of ASCO to reach out globally will not only help discover the untapped talent of the region but through them improve the quality of cancer care in APAC. In addition to developing future leaders of the region, it will also provide the potential opportunity for rising APAC stars to serve in leadership positions for ASCO in due course. The Council is very proud to be involved in this ambitious program and confident that this will represent a great milestone for the whole APAC oncology society to move forward for quality care and global leadership of oncology.
Physical characteristics of soil-biodegradable and nonbiodegradable plastic mulches impact conidial splash dispersal of	a3b2fe7e-b3c8-4fed-a64e-b75a3479c162	10166481	Microbiology[mh]	Fungi have evolved different mechanisms to aid dispersal of their propagules . The main dispersal mechanisms of fungal plant pathogens are wind and rain splash [ – ]. Understanding dispersal mechanisms is important for limiting fungal propagule movement and reducing the spread of global diseases that threaten food security and plant survival . Fungal plant pathogens belonging to phylum Ascomycota are capable of producing asexual spores known as conidia and these conidia are regarded as the major dispersal unit of many important plant pathogens . Conidia can be dispersed over short distances (e.g., a few centimeters) by rain-generated splash droplets or over long distances (e.g., several kilometers) by wind . Infected plant tissues are a source of conidial inoculum and conidia incorporated into water droplets can be transported to neighboring plants by rain or irrigation splash events . Rain splash dispersal can become critical in the development of fungal disease epidemics when a large number of conidia are transported to neighboring plants . Previous research has demonstrated there is variability in conidial dispersal numbers through rain splash [ , , ] and this variability is attributed to multiple factors including the surfaces that splashed droplets are interacting with [ , , ]. However, the overall role of surface characteristics on splash dispersal of plant pathogens is understudied and in need of further investigation to inform disease risk and associated management. Gray mold is an important fungal disease caused by Botrytis cinerea that severely impacts yields in commercial strawberry ( Fragaria × ananassa ) production systems [ , , ]. The pathogen is found worldwide , produces abundant inoculum in the form of conidia , and is considered a primary pathogen responsible for significant yield losses of harvested strawberry across the world . Yield losses due to this pathogen can exceed 80% in the absence of fungicides and under favorable environmental conditions . The pathogen is currently considered hemibiotrophic and can infect a variety of tissues including flowers and fruits ( ). Ripe and green strawberry fruits are both susceptible to this pathogen in the field . Botrytis is also regarded as a post-harvest pathogen because the disease can develop during transportation and storage . Rain splash dispersal may cause severe secondary gray mold infections through the dispersal of conidia from infected flowers, buds, or fruits to susceptible tissues [ , , ], especially in high humidity (>80%) environments or when rainfall occurs preharvest . Commercial strawberry production is often done under plasticulture using non-degradable polyethylene (PE) mulch ( ) . PE mulch utilization promotes plant growth and increases yield primarily through weed suppression and optimization of soil temperatures [ – ]. Weedmat (a semi-permeable, woven, PE- or polypropylene-based geotextile) is an alternative to PE that is preferred by some farmers due to its durability and potential to be re-used over multiple seasons. Soil-biodegradable plastic mulch (BDM) is perceived as a more sustainable alternative to PE and weedmat because it is designed to biodegrade in soils and consequently reduce plastic waste generation . Ground covers including mulches can vary in surface characteristics and impact splash dispersal of plant pathogens. For example, splash dispersal patterns of Colletotrichum acutatum (cause of anthracnose) and Phytophthora cactorum (cause of leather rot on strawberry fruits and Phytophthora crown and root rots) were highly related to ground cover physical characteristics [ , , ]. Random roughness was determined to be one of the key physical features of ground cover surface microtopography that influence splash dispersal with decreasing roughness increasing the magnitude of splash distance . For example, smooth PE mulch generated more conidial splash dispersal of C . acutatum compared to a straw mulch . However, the impact of groundcover random roughness is not consistent as a buckwheat ( Fagopyrum esculentum ) husk mulch increased gray mold incidence in strawberry compared to other mulches including black plastic, wood chips, and straw . Other ground cover characteristics such as permeability and surface adhesion of water may impact the dynamics of pathogen splash dispersal. However, there are no recent studies on splash dispersal from mulch surfaces and a systematic understanding of B . cinerea conidial dispersal via rain splash in strawberry productions systems under plasticulture is lacking. The objective of this study was to investigate splash dispersal dynamics of B . cinerea conidia with three plastic mulch surfaces: PE mulch, weedmat, and BDM. Specific sub-objectives were to: 1) Evaluate mulch surface physical characteristics that could impact splash dispersal dynamics and 2) Determine whether different plastic mulches affect conidial splash dispersal patterns of B . cinerea . Mulch surface physical characteristics New, unweathered PE, weedmat, and embossed BDM mulches included in this study ( ) were the same products that had been used in a previous mulch experiment . Surface microtopographic features of mulch samples were assessed using a digital microscope (Keyence VHX-7000) in the Composite Materials and Engineering Center at Washington State University (WSU) in Pullman, Washington, USA. Mulch permeability was characterized to assess how material properties affect infiltration of liquid involved in splash dispersal and subsequent conidial dispersal outcomes. For each mulch treatment, a large plastic Petri dish (14 cm in diameter) was filled with greenhouse growing medium (Steuber Promix BX General Purpose, Premier Tech Horticulture, Quakertown, PA, USA) and covered with one of the mulch treatments. A 20 cm-long square piece of mulch was secured tightly around the dish by taping the mulch to the bottom of the Petri dish and adjusting the mulch surface tension to 0.74 N at the center using a tension meter (Chatillon 516 Series linear push/pull scale, Ametek, Berwyn, PA, USA), which is comparable to mulch tension measurements collected in the field ( ). A silicone ring (6 cm diameter and 1 cm high ) was placed on the mulch surface in the center of the Petri dish and sealed to the mulch using a white, 100% silicone sealant (Gorilla Glue Company; Berlin, Germany). The sealant was allowed to dry completely (~30 min) before 10 ml of water was gently poured into the area within the silicone ring. The volume of remaining water within the silicone ring was measured after 2, 4, 8, 16, and 32 minutes at 25°C using a 10 ml-graduated cylinder. Permeability measurements were repeated three times as technical replicates for each mulch treatment using new mulch for each repetition. Characterization of splash dynamics To better understand splash dynamics of water on PE mulch, weedmat, and BDM surfaces, a preliminary non-replicated experiment using a single droplet was conducted. A water nozzle, syringe pump (NE-1000, Pump Systems LLC., Dickinson, ND, USA), and needle attached to the nozzle were used to generate individual water droplets falling onto either PE mulch, weedmat, or BDM surfaces. A high-speed camera (NOVA S6, Photron USA, Inc., San Diego, CA, USA) with a backlit LED light source was used to record the splash dynamics initiated by a second drop falling onto the former drop placed on mulch surfaces. The number of resulting satellite droplets as well as qualitative data describing splash dynamics were recorded. Conidial splash dispersal of B . cinerea This experiment was conducted in a fully enclosed screen house (15 m long, 6 m wide, 5.5 m tall with 270 × 770 μm mesh size; Gable Series 7500, U.S. Global Resources, Seattle, WA, USA) at the WSU Mount Vernon Northwestern Washington Research and Extension Center (WSU NWREC) in Mount Vernon, WA to minimize potential contamination. The ground of the screenhouse was covered with woven, polyethylene landscape fabric. Within the screenhouse an Eurmax Premium instant canopy tent with enclosed sidewalls (3 m long, 2.9 m wide, 2.3 m tall; Eurmax Canopy, EI Monte, CA, USA) was assembled to further minimize contamination potential and wind effects ( ). A rain simulator system was built within the canopy tent consisting of three metal pulsating sprinklers (Hunter MP Rotator® MP800, Hunter Industries, San Marcos, CA, USA) mounted on tripods (Melnor 65066-AMZ; Melnor Inc., Frederick County, VA, USA; ). There was one sprinkler head per tripod and individual sprinkler heads were arranged in an equilateral triangle with each side measuring 1.8 m long between sprinkler heads ( ). Each sprinkler head consisted of a high-efficiency nozzle paired with a pop-up body (Hunter Pro-Spray PRS30, Hunter Industries, San Marcos, CA, USA) for pressure regulation to 207 KPa. A schedule 80 PVC nipple (Grainger Industrial Supply, Lake Forest, IL, USA) was used to connect the sprinkler head and irrigation pipes. Water was discharged from nozzles at a height of 1.5 m. Uniformity tests were conducted prior to conidial dispersal experiments to position the nozzles, maximize head-to-head coverage, and optimize uniformity of water discharged from the rain simulator system ( ). Each uniformity test was performed for five minutes and repeated three times as technical replicates for each configuration until maximum uniformity was achieved. Water was collected into five empty Petri plates (10 cm diameter) placed in the central area where water was discharged by the rain simulator system. A total of 40 plates were used for each uniformity test with plates arranged in eight cardinal and ordinal directions (with 5 plates in each direction) radiating from the center (North, South, West, East, Northwest, Northeast, Southwest, and Southeast). Accumulated water was collected from each plate and measured at the end of each uniformity test. All three nozzles were adjusted to a 210˚ arc and a low flow rate (68 liter/hour). The central position of the sprinkler system was arranged so it was 1 m equidistant from the inoculum source ( ). The configuration of the rain simulator system used for subsequent conidial dispersal experiments achieved an 85–90% uniformity within 1.8 m ( ), which included the area where the dispersal experiment was conducted. Additionally, multiple drops of water emitted from the rain simulator system were collected on water-sensitive paper. From this, the average size (i.e., diameter) of the drops was measured. Then 10, 20, 30, 40, 50, and 60ul of water were dropped individually on a separate piece of water-sensitive paper using a pipette. The size of the drops from the rain simulator system was compared to the drops containing a known volume of water to estimate average droplet volume from the rain simulator system. This approach allowed us to determine the average volume of a drop from the rain simulator, which was 50ul. Precipitation rate was also measured during uniformity tests and the pressure was adjusted to achieve a rate of 21 mm/hour. Following establishment of the rain simulator system, a conidial suspension was prepared using a pathogenic, SDHI (succinate dehydrogenase inhibitor; FRAC-7) fungicide-resistant isolate of B . cinerea to distinguish it from other airborne Botrytis spores during this experiment ( ). This isolate was characterized as having the P225F mutation which confers resistance to all SDHI active ingredients (Kozhar, personal communication). To prepare the conidial suspension, the isolate was cultured on full-strength potato dextrose agar amended with 100 μg/ml salicylhydroxamic acid (SHAM), 32 μg/ml isofetamid, and 35 mg/ml chloramphenicol. Then B . cinerea culture plates were incubated at 22°C in an incubator (Percival Intellus Ultra temperature controller, Percival Scientific, Inc., Perry, IA, USA) with 18 hours of white light (light intensity 320 lux) and 6 hours dark for 10 days. Sterile deionized (DI) water (20 mL) was added onto 10-day-old B . cinerea culture plates and a lab spatula was used to dislodge conidia into the water solution. The solution was then filtered through one layer of cheesecloth and transferred into 50 ml centrifuge tubes. The conidial suspension was vortexed for 5 minutes to distribute clumped conidia and then the conidial concentration was determined using a hemocytometer (Bright-line Model 1492, Hausser Scientific, Horsham, PA, USA). The final conidial concentration was adjusted to 2 ±0.2 ×10 6 conidia/ml. The conidial germination rate prior to running the splash dispersal experiment was nearly 100% ( ). Small Petri dishes (60 mm diameter) containing Botrytis spore trap media (BSTM) (modified from ) was used to collect the conidia splashed from the inoculum source. BSTM consisted of 2 g glucose, 20 g bacto TM agar, 0.1 g NaNO 3 , 0.1 g K 2 HPO 4 , 0.2 g MgSO 4 ⋅7H 2 O and 0.1 g KCl in one liter of DI water. The solution was autoclaved and cooled to 56°C before adding 2.5 g/l tannic acid and adjusting the pH to 4.5 with 1 mol/l NaOH. Next, 0.2 g/l chloramphenicol, 0.02 g/l pentachloronitrobenzene (PCNB), 0.02 g/l dithane, and 0.1 ml/l 12% fenarimol were added and the media was poured into Petri dishes. Before running the rain simulator, a large piece of Pig ® Absorbent Mat Pad (30 cm × 50 cm) was placed on the ground to limit secondary splash from outside of the primary inoculum source and each replicate was placed on the mat at a marked position within the center of the rain simulator system ( ). The absorbent mat was immediately replaced after each replicate. For each replicate, five of the BSTM plates were placed immediately adjacent to the Petri dish containing the inoculum and one of the three mulch treatments and arranged linearly in a southward direction ( ). This direction was based on uniformity results and maintained throughout the experiment for consistency. Each BSTM plate was immediately adjacent to the other, so the BSTM plate distance between each center of an adjacent BSTM plate was standardized at 6 cm. The center of the first BSTM plate was placed 10 cm away from the center of inoculum source and mulch treatment. Succeeding plates were placed at 16, 22, 28, and 34 cm from the center of the first plate. These distances from the inoculum source were assigned as dispersal distances (denoted as i, i = 10, 16, 22, 28, 34). The experiment was repeated five times as technical replicates per mulch treatment with the construction of the mulch treatment done using the procedures outlined above for mulch permeability characterization. Additionally, after the silicone sealant dried, the annulus of the mulch surface was covered by a water absorbent mat (Pig ® Absorbent Mat Pad, model MAT 203; New Pig Corporation, Tipton, PA) to reduce production of secondary splash droplets ( ). New mulch and absorbent mats were used for each replicate. The mulch surface and silicone band were then disinfected with 70% ethanol immediately before exposure to the rain simulator system. The B . cinerea conidial suspension (10 ml per replicate) was pipetted into the well created by the silicone ring and formed a pool on the treatment mulch. Immediately after the placement of BSTM plates and the addition of conidial suspension, the rain simulator system was initiated for two minutes with three people turning the irrigation on and off at the same time. BSTM plates were immediately collected and incubated at 22 ˚C for 18 hours with lids open in the same incubator listed above to dry the media surface and provide time for conidia to germinate. Prior to counting conidial germination, lids were placed back on the BSTM plates, and maintained at 1°C for a maximum of 4 hours while conidial counts were being performed to minimize changes in spore germination during the counting process. Presence of a germ tube (length was greater than half the width of the conidia) growing out of a conidium was used to assign germination status. All conidia from BSTM plates were counted within 3 hours after removal from the incubator. Total dispersed conidia (TC i ) and the number of dispersed conidia that germinated after 18h (GC i ) were enumerated for each dispersal distance (i) from each BSTM plate using a compound microscope (Nikon Eclipse 50i, Nikon Inc., Melville, NY, USA) at 40X magnification. Data analysis All quantitative data were analyzed using R Studio software (Version 1.4.1106, Rstudio PBC, Boston, MA, USA). Mulch permeability was graphed using ggplot2 package , while surface characteristics and splash dynamics were visually assessed. Findings from the non-replicated splash dynamics study were treated as qualitative observations and not analyzed due to the lack of replication. Conidial splash dispersal TC i data were first fit to an exponential model = N = N 0 e x p ( − d / τ ) (1) with N 0 as the prefactor of the exponential model, which represents the number of conidia at the impact location ( d = 0). τ is the decay length and N represents the number of conidia at distance d . Data are presented in their original units. Then two additional variables were calculated and added into further data analysis as follows: The percentage (%) of TC i of total dispersed conidia at all distances [abbreviated as T i byT] = T C i ∑ i T C i × 100 , i ∈ { 10 , 16 , 22 , 28 , 34 } (2) The percentage (%) of GC i of germinated conidia at all distances [abbreviated as G i byG] = G C i ∑ i G C i × 100 , i ∈ { 10 , 16 , 22 , 28 , 34 } (3) Technical replicate was regarded as a random factor, while mulch treatment and dispersal distance were treated as fixed factors. Four variables (TC i , GC i , T i byT, and G i byG) were checked to ensure they met distribution assumptions and fitted into a generalized linear mixed-effects model. Normal distributions of those four variables were checked before depositing data into analysis using the his() function on residuals. According to different distribution patterns, TC i and GC i data were subjected to a Poisson distribution and log transformed, while T i byT and G i byG data were subjected to a normal distribution. An interaction between mulch treatment and dispersal distance was considered for these four variables. Multivariate Analysis of variance (MANOVA) was conducted using estimated marginal means (least-square means) with a post hoc Tukey–Kramer test for detection of treatment effects at a significance of α = 0.05. Data were presented in their original units. Additionally, a Pearson product-moment correlation coefficient analysis was conducted between each of TC i , GC i , T i byT, G i byG and dispersal distance, between TC i and GC i , and T i byT and G i byG. All data are available in Excel file S5. New, unweathered PE, weedmat, and embossed BDM mulches included in this study ( ) were the same products that had been used in a previous mulch experiment . Surface microtopographic features of mulch samples were assessed using a digital microscope (Keyence VHX-7000) in the Composite Materials and Engineering Center at Washington State University (WSU) in Pullman, Washington, USA. Mulch permeability was characterized to assess how material properties affect infiltration of liquid involved in splash dispersal and subsequent conidial dispersal outcomes. For each mulch treatment, a large plastic Petri dish (14 cm in diameter) was filled with greenhouse growing medium (Steuber Promix BX General Purpose, Premier Tech Horticulture, Quakertown, PA, USA) and covered with one of the mulch treatments. A 20 cm-long square piece of mulch was secured tightly around the dish by taping the mulch to the bottom of the Petri dish and adjusting the mulch surface tension to 0.74 N at the center using a tension meter (Chatillon 516 Series linear push/pull scale, Ametek, Berwyn, PA, USA), which is comparable to mulch tension measurements collected in the field ( ). A silicone ring (6 cm diameter and 1 cm high ) was placed on the mulch surface in the center of the Petri dish and sealed to the mulch using a white, 100% silicone sealant (Gorilla Glue Company; Berlin, Germany). The sealant was allowed to dry completely (~30 min) before 10 ml of water was gently poured into the area within the silicone ring. The volume of remaining water within the silicone ring was measured after 2, 4, 8, 16, and 32 minutes at 25°C using a 10 ml-graduated cylinder. Permeability measurements were repeated three times as technical replicates for each mulch treatment using new mulch for each repetition. To better understand splash dynamics of water on PE mulch, weedmat, and BDM surfaces, a preliminary non-replicated experiment using a single droplet was conducted. A water nozzle, syringe pump (NE-1000, Pump Systems LLC., Dickinson, ND, USA), and needle attached to the nozzle were used to generate individual water droplets falling onto either PE mulch, weedmat, or BDM surfaces. A high-speed camera (NOVA S6, Photron USA, Inc., San Diego, CA, USA) with a backlit LED light source was used to record the splash dynamics initiated by a second drop falling onto the former drop placed on mulch surfaces. The number of resulting satellite droplets as well as qualitative data describing splash dynamics were recorded. B . cinerea This experiment was conducted in a fully enclosed screen house (15 m long, 6 m wide, 5.5 m tall with 270 × 770 μm mesh size; Gable Series 7500, U.S. Global Resources, Seattle, WA, USA) at the WSU Mount Vernon Northwestern Washington Research and Extension Center (WSU NWREC) in Mount Vernon, WA to minimize potential contamination. The ground of the screenhouse was covered with woven, polyethylene landscape fabric. Within the screenhouse an Eurmax Premium instant canopy tent with enclosed sidewalls (3 m long, 2.9 m wide, 2.3 m tall; Eurmax Canopy, EI Monte, CA, USA) was assembled to further minimize contamination potential and wind effects ( ). A rain simulator system was built within the canopy tent consisting of three metal pulsating sprinklers (Hunter MP Rotator® MP800, Hunter Industries, San Marcos, CA, USA) mounted on tripods (Melnor 65066-AMZ; Melnor Inc., Frederick County, VA, USA; ). There was one sprinkler head per tripod and individual sprinkler heads were arranged in an equilateral triangle with each side measuring 1.8 m long between sprinkler heads ( ). Each sprinkler head consisted of a high-efficiency nozzle paired with a pop-up body (Hunter Pro-Spray PRS30, Hunter Industries, San Marcos, CA, USA) for pressure regulation to 207 KPa. A schedule 80 PVC nipple (Grainger Industrial Supply, Lake Forest, IL, USA) was used to connect the sprinkler head and irrigation pipes. Water was discharged from nozzles at a height of 1.5 m. Uniformity tests were conducted prior to conidial dispersal experiments to position the nozzles, maximize head-to-head coverage, and optimize uniformity of water discharged from the rain simulator system ( ). Each uniformity test was performed for five minutes and repeated three times as technical replicates for each configuration until maximum uniformity was achieved. Water was collected into five empty Petri plates (10 cm diameter) placed in the central area where water was discharged by the rain simulator system. A total of 40 plates were used for each uniformity test with plates arranged in eight cardinal and ordinal directions (with 5 plates in each direction) radiating from the center (North, South, West, East, Northwest, Northeast, Southwest, and Southeast). Accumulated water was collected from each plate and measured at the end of each uniformity test. All three nozzles were adjusted to a 210˚ arc and a low flow rate (68 liter/hour). The central position of the sprinkler system was arranged so it was 1 m equidistant from the inoculum source ( ). The configuration of the rain simulator system used for subsequent conidial dispersal experiments achieved an 85–90% uniformity within 1.8 m ( ), which included the area where the dispersal experiment was conducted. Additionally, multiple drops of water emitted from the rain simulator system were collected on water-sensitive paper. From this, the average size (i.e., diameter) of the drops was measured. Then 10, 20, 30, 40, 50, and 60ul of water were dropped individually on a separate piece of water-sensitive paper using a pipette. The size of the drops from the rain simulator system was compared to the drops containing a known volume of water to estimate average droplet volume from the rain simulator system. This approach allowed us to determine the average volume of a drop from the rain simulator, which was 50ul. Precipitation rate was also measured during uniformity tests and the pressure was adjusted to achieve a rate of 21 mm/hour. Following establishment of the rain simulator system, a conidial suspension was prepared using a pathogenic, SDHI (succinate dehydrogenase inhibitor; FRAC-7) fungicide-resistant isolate of B . cinerea to distinguish it from other airborne Botrytis spores during this experiment ( ). This isolate was characterized as having the P225F mutation which confers resistance to all SDHI active ingredients (Kozhar, personal communication). To prepare the conidial suspension, the isolate was cultured on full-strength potato dextrose agar amended with 100 μg/ml salicylhydroxamic acid (SHAM), 32 μg/ml isofetamid, and 35 mg/ml chloramphenicol. Then B . cinerea culture plates were incubated at 22°C in an incubator (Percival Intellus Ultra temperature controller, Percival Scientific, Inc., Perry, IA, USA) with 18 hours of white light (light intensity 320 lux) and 6 hours dark for 10 days. Sterile deionized (DI) water (20 mL) was added onto 10-day-old B . cinerea culture plates and a lab spatula was used to dislodge conidia into the water solution. The solution was then filtered through one layer of cheesecloth and transferred into 50 ml centrifuge tubes. The conidial suspension was vortexed for 5 minutes to distribute clumped conidia and then the conidial concentration was determined using a hemocytometer (Bright-line Model 1492, Hausser Scientific, Horsham, PA, USA). The final conidial concentration was adjusted to 2 ±0.2 ×10 6 conidia/ml. The conidial germination rate prior to running the splash dispersal experiment was nearly 100% ( ). Small Petri dishes (60 mm diameter) containing Botrytis spore trap media (BSTM) (modified from ) was used to collect the conidia splashed from the inoculum source. BSTM consisted of 2 g glucose, 20 g bacto TM agar, 0.1 g NaNO 3 , 0.1 g K 2 HPO 4 , 0.2 g MgSO 4 ⋅7H 2 O and 0.1 g KCl in one liter of DI water. The solution was autoclaved and cooled to 56°C before adding 2.5 g/l tannic acid and adjusting the pH to 4.5 with 1 mol/l NaOH. Next, 0.2 g/l chloramphenicol, 0.02 g/l pentachloronitrobenzene (PCNB), 0.02 g/l dithane, and 0.1 ml/l 12% fenarimol were added and the media was poured into Petri dishes. Before running the rain simulator, a large piece of Pig ® Absorbent Mat Pad (30 cm × 50 cm) was placed on the ground to limit secondary splash from outside of the primary inoculum source and each replicate was placed on the mat at a marked position within the center of the rain simulator system ( ). The absorbent mat was immediately replaced after each replicate. For each replicate, five of the BSTM plates were placed immediately adjacent to the Petri dish containing the inoculum and one of the three mulch treatments and arranged linearly in a southward direction ( ). This direction was based on uniformity results and maintained throughout the experiment for consistency. Each BSTM plate was immediately adjacent to the other, so the BSTM plate distance between each center of an adjacent BSTM plate was standardized at 6 cm. The center of the first BSTM plate was placed 10 cm away from the center of inoculum source and mulch treatment. Succeeding plates were placed at 16, 22, 28, and 34 cm from the center of the first plate. These distances from the inoculum source were assigned as dispersal distances (denoted as i, i = 10, 16, 22, 28, 34). The experiment was repeated five times as technical replicates per mulch treatment with the construction of the mulch treatment done using the procedures outlined above for mulch permeability characterization. Additionally, after the silicone sealant dried, the annulus of the mulch surface was covered by a water absorbent mat (Pig ® Absorbent Mat Pad, model MAT 203; New Pig Corporation, Tipton, PA) to reduce production of secondary splash droplets ( ). New mulch and absorbent mats were used for each replicate. The mulch surface and silicone band were then disinfected with 70% ethanol immediately before exposure to the rain simulator system. The B . cinerea conidial suspension (10 ml per replicate) was pipetted into the well created by the silicone ring and formed a pool on the treatment mulch. Immediately after the placement of BSTM plates and the addition of conidial suspension, the rain simulator system was initiated for two minutes with three people turning the irrigation on and off at the same time. BSTM plates were immediately collected and incubated at 22 ˚C for 18 hours with lids open in the same incubator listed above to dry the media surface and provide time for conidia to germinate. Prior to counting conidial germination, lids were placed back on the BSTM plates, and maintained at 1°C for a maximum of 4 hours while conidial counts were being performed to minimize changes in spore germination during the counting process. Presence of a germ tube (length was greater than half the width of the conidia) growing out of a conidium was used to assign germination status. All conidia from BSTM plates were counted within 3 hours after removal from the incubator. Total dispersed conidia (TC i ) and the number of dispersed conidia that germinated after 18h (GC i ) were enumerated for each dispersal distance (i) from each BSTM plate using a compound microscope (Nikon Eclipse 50i, Nikon Inc., Melville, NY, USA) at 40X magnification. All quantitative data were analyzed using R Studio software (Version 1.4.1106, Rstudio PBC, Boston, MA, USA). Mulch permeability was graphed using ggplot2 package , while surface characteristics and splash dynamics were visually assessed. Findings from the non-replicated splash dynamics study were treated as qualitative observations and not analyzed due to the lack of replication. Conidial splash dispersal TC i data were first fit to an exponential model = N = N 0 e x p ( − d / τ ) (1) with N 0 as the prefactor of the exponential model, which represents the number of conidia at the impact location ( d = 0). τ is the decay length and N represents the number of conidia at distance d . Data are presented in their original units. Then two additional variables were calculated and added into further data analysis as follows: The percentage (%) of TC i of total dispersed conidia at all distances [abbreviated as T i byT] = T C i ∑ i T C i × 100 , i ∈ { 10 , 16 , 22 , 28 , 34 } (2) The percentage (%) of GC i of germinated conidia at all distances [abbreviated as G i byG] = G C i ∑ i G C i × 100 , i ∈ { 10 , 16 , 22 , 28 , 34 } (3) Technical replicate was regarded as a random factor, while mulch treatment and dispersal distance were treated as fixed factors. Four variables (TC i , GC i , T i byT, and G i byG) were checked to ensure they met distribution assumptions and fitted into a generalized linear mixed-effects model. Normal distributions of those four variables were checked before depositing data into analysis using the his() function on residuals. According to different distribution patterns, TC i and GC i data were subjected to a Poisson distribution and log transformed, while T i byT and G i byG data were subjected to a normal distribution. An interaction between mulch treatment and dispersal distance was considered for these four variables. Multivariate Analysis of variance (MANOVA) was conducted using estimated marginal means (least-square means) with a post hoc Tukey–Kramer test for detection of treatment effects at a significance of α = 0.05. Data were presented in their original units. Additionally, a Pearson product-moment correlation coefficient analysis was conducted between each of TC i , GC i , T i byT, G i byG and dispersal distance, between TC i and GC i , and T i byT and G i byG. All data are available in Excel file S5. Splash dynamics and dispersal of B . cinerea conidia differed due to mulch treatment and distance with treatment effects attributed to differences in the physical characteristics of the mulch surfaces. PE mulch, weedmat, and embossed BDM showed different surface topographic characteristics ( ). PE mulch had a smooth, flat surface compared to the embossed BDM. Weedmat mulch had a rigid surface due to the woven fibers of PE. Furthermore, PE mulch and embossed BDM were impermeable to water and weedmat was semi-permeable based on laboratory observations ( ). There was no water loss through infiltration for PE mulch and embossed BDM, whereas >94% of the water infiltrated through the weedmat after 32 mins. Mulch permeability in the screenhouse was not measured given there was additional water delivered via the rain simulator system and water displacement occurring from rain droplets impacting the inoculum pool. Although the woven construction of weedmat seemed to produce visibly similar surface characteristics as embossed BDM, the semi-permeability of weedmat led to infiltration of water containing the inoculum through the mulch and effectively reduced the volume of the conidial suspension. Results from the rain simulator system showed the means of all TC i across the five distances for PE mulch, weedmat, and embossed BDM were 26.8, 31, and 47 conidia, respectively. All mulch treatments showed an inverse relationship in that as the horizontal distance from the inoculum source increased, the number of TC i and GC i decreased ( ). Inverse relationships were also observed between T i byT, G i byG and horizontal distance from the inoculum source ( ). Interestingly, the TC i and GC i differed due to mulch treatment ( P <0.001 for TC i and GC i ,) and dispersal distance from the inoculum source ( P <0.001) ( ). The T i byT and G i byG differed due to dispersal distance ( P <0.001) but not due to mulch treatment ( P = 0.97–0.98). There was an interaction between mulch treatment and dispersal distance ( P = 0.006) for TC i , but not for GC i ( P = 0.20), T i byT ( P = 0.64), and G i byG ( P = 0.88). The TC i and GC i splashed from embossed BDM were the highest, followed by PE and weedmat mulches. PE and weedmat had similar TC i and GC i ( P = 0.43 and P = 0.23 respectively). The dispersal distance at 10 cm and 34 cm had the highest and lowest TC i and GC i , respectively, across all mulch treatments. GC 28 equaled to zero for PE mulch and GC 34 equaled to zero for PE and weedmat mulches. T 10 byT and G 10 byG was > 60% and T 10 byT and T 16 byT were approximately 80% across all mulch sources. Differences in conidia dispersal distance were likely due to mulch surface topographic characteristics as the number of conidia over distance followed a simple exponential decay model ( ). The measured N 0 [number of conidia at impact location ( d = 0)] was 50.50 for PE mulch, 187.58 for embossed BDM, and 320.53 for weedmat. The decay lengths ( τ ) were 8.33 cm for PE mulch, 5.26 cm for embossed BDM, and 3.70 cm for weedmat. This indicates that as mulch roughness increases, more conidia spread with shorter horizontal distance ( N 0 increases and τ decreases). Physically, this may be observed as the rough surface of mulches create splashes more upward compared to a smooth surface. Random roughness of groundcover surface microtopography has been determined to be a key physical feature that influences splash dispersal . Preliminary observations from the non-replicated splash dynamics experiment showed that the second single drop falling onto an existing single drop on the weedmat and embossed BDM surfaces produced similar numbers of splashed satellite droplets (approximately 20–22), whereas PE produced fewer satellite droplets (approximately 15). Both PE and embossed BDM showed a well-connected splash crown upon secondary drop impact ( ). Furthermore, the rim of the crown ejected what appeared to be visually smaller splashed satellite droplets when interacting on the PE mulch surface than embossed BDM. Weedmat exhibited an irregular pattern of splashing ( ), and this was attributed to the large-scale roughness of the interwoven weedmat fibers. We also observed PE and embossed BDM surfaces produced a similar spherical adhesion to a single drop of water whereas the weedmat surface produced a flat shape ( ), which may be due variations in mulch permeability ( ) . Although this was not a replicated trail, observations suggest that an irregular splashing pattern may produce a variable but sometimes large number of splashed satellite droplets. Additionally, the embossed surface of BDM and interwoven fibers of weedmat lead to relatively upward splash patterns and a greater number of droplets than the PE mulch and this may increase the number of conidia dispersed in satellite droplets but quickly decay horizontally as described in the exponential model fit. Regardless of mulch treatment, most B . cinerea conidia were dispersed within 10 cm from the inoculum source in this study. Conidia captured on media plates decreased greatly from 10 to 22 cm and were nearly absent at 34 cm from the inoculum source. A similar trend was observed for GC i . It is also to be noted that there was a highly significant linear correlation between TCi and GCi, as well as between T i byT and G i byG ( ). Results from this study align well with previous research on plant pathogen spread by splash dispersal in rain-simulated systems. Previous work has demonstrated that the majority of splashed droplets are dispersed over short distances and larger droplets containing dozens to hundreds of conidia fall within 30 cm from inoculum sources [ , , – ]. Perryman et al. also found larger splashed droplets containing an average of 308 conidia/droplet of Pyllosticta citricarpa (cause of Citrus black spot) were dispersed within 30 cm from the inoculum source while smaller splash droplets can be dispersed further and as far as 70 cm. Findings from our study are also in agreement with the general understanding that rain splash dispersal of fungal plant pathogens occurs at smaller scales (e.g., centimeters) compared to wind dispersal, which can disperse propagules several meters or even kilometers from an inoculum source [ – ]. The conidial density of B . cinerea necessary for infection and lesion development on strawberry flowers and fruits as well as the splash dispersal distance of water droplets with conidial concentrations needed for infection are currently unknown. Previous inoculation studies on grape ( Vitis vinifera ) berry surfaces showed an individual airborne conidium could infect berry tissue and lead to lesion development, but overall infection was governed by host resistance that varied between fresh and cold-stored fruits . For strawberry, a B . cinerea concentration of 3×10 3 conidia/m 3 in air caused only 0.56% Botrytis fruit rot . Studies demonstrating the density of B . cinerea needed in a liquid suspension to cause disease are limited. In lentil ( Lens culinaris ) seedlings, a Botrytis spp. suspension at 10 4 spores/ml concentration caused a 20% chance of infection whereas a 10 2 spores/ml concentration did not result in any symptom development . The concentration of our suspension was 2 ±0.2 ×10 6 conidia/ml, which may be sufficient to cause disease. Hence, although our splash dispersal results showed a mulch treatment effect, whether the greater conidial numbers generated from embossed BDM cause more severe disease needs to be demonstrated using field-relevant conidial concentrations in suspension and is an area for future research. Decreasing random roughness (defined as the standard deviation of elevations over an area of interest from a base plane) of ground covers, including mulches, has been previously reported to increase the magnitude of pathogen splash distance [ , , , ]. In this study, mulch random roughness was not measured, but visual assessment of mulch surface microtopography indicated PE to have lower mulch roughness than embossed BDM and weedmat due to their respective embossed and ridged surface features. PE can also be made with an embossed surface, so it is still an open question if similar responses will be observed between PE and BDM depending on whether or not they are embossed. Despite this variation in surface microtopography, all mulch treatments in the current study were associated with a decrease in the number of dispersed conidia with increasing horizontal dispersal distance. Furthermore, other factors such as the degree, uniformity, and scale of roughness may contribute to the magnitude of splash dispersal of pathogens. It is to be noted that the assayed dispersal distances under our experimental conditions was designed on a small scale to reflect the range of typical plant spacings (25 to 30 cm) observed in commercial strawberry systems . Our results suggest that wider within-row strawberry plant spacing could reduce the spread of B . cinerea conidia transported through splash dispersal. Wider within-row plant spacing has been shown to reduce disease incidence of Botrytis in field-grown strawberry in Florida with 40% less disease at a wider spacing of 45.6 cm compared to 22.9 cm . Other splash dispersal studies similarly found plant density can affect splash dispersal of C . acutatum in strawberry . Although wider spacing can be beneficial for reducing plant pathogens spread by rain splash and reducing the period of free moisture on plants, there is a trade-off as this effectively reduces plant density and yield potential by having fewer plants per unit area. Rain characteristics, such as rain density, duration, and droplet size are also known to impact pathogen dispersal distance [ , , , , ]. Pathogens could furthermore have different splash dispersal distances due to different density and/or sizes of pathogen spores incorporated within a droplet [ , , , ]. However, studies on splash dispersal epidemiology of fungal plant pathogens are few and there are many unexplored aspects that remain understudied. It is also unclear if the feedstock ingredients of mulches contribute to differences in surface characteristics and how they interact with water, which could impact splash dispersal. BDMs are comprised of 75–95% biodegradable feedstock ingredients with the remainder being additives such as plasticizers, fillers, colorants, pigments, UV-stabilizers, antioxidants, nucleating agents, and antibacterial additives . The PE and weedmat mulches used in this study were both made of PE feedstock whereas the embossed BDM feedstock comprised of PLA (polylactic acid) and PBAT (polybutylene adipate-co-terephthalate). If ingredients differ in hydrophobicity or hydrophilicity, this could in turn influence the interaction between water droplets and resultant splash dispersal. In addition, different plastic mulches have varying surface characteristics due to their manufacturing (i.e., film vs. woven material vs. embossing), which could impact splash dispersal of pathogen spores and disease risk of fungal pathogens such as B . cinerea . In conclusion, physical characteristics including permeability, surface physical features, and splashing dynamics of PE mulch, weedmat, and embossed BDM can impact splash dispersal dynamics of B . cinerea . Differences in the number of conidia dispersed among PE mulch, weedmat, and embossed BDM was observed. The embossed BDM facilitated greater dispersal of splashed conidia compared to PE mulch and weedmat. Based on these results, mulch physical features created by embossing may contribute to enhanced B . cinerea inoculum availability in strawberry production under plasticulture. However, the conidial concentrations observed among treatments were overall low and differences in splashed conidia may not be great enough to elicit differential responses in disease incidence at field scale. Future studies should focus on translating these differences into potential real-world risks within commercial strawberry fields for B . cinerea as well as other pathogens dispersed through rain splash. S1 Table Mulch tension measurements in a mulched ‘Albion’ strawberry field in Mount Vernon, WA. (DOCX) Click here for additional data file. S2 Table Uniformity tests for standardizing rain simulator system before running splash dispersal experiments. (DOCX) Click here for additional data file. S3 Table Assessment of conidial germination rates prior to running the splash dispersal Experiment. (DOCX) Click here for additional data file. S4 Table Germination rate of all dispersed Botrytis cinerea conidia recovered from the rain splash dispersal experiment. (DOCX) Click here for additional data file. S1 Data Data used in the experiment. (XLSX) Click here for additional data file.
Addressing the Gap in Research Methodologies Education in Pediatric Oncology in the Eastern Mediterranean Region	83e33f42-2847-4130-9320-f581ab0082c0	10166560	Internal Medicine[mh]	In limited-resource settings, there is little or no focus on formal clinical research methodology training. Most fellowship and specialty training programs rely on trainees to run clinical services, with limited time for research education. Additionally, opportunities for clinical research for nonphysician health care professionals including nurses are limited. Barriers to research capacity building in limited-resource settings include absence of protected time for trainees, unavailability of local or regional educational funding opportunities, and a lack of mentorship time. Over the past few years, driven by the WHO Global Initiative for Childhood Cancer (GICC), there has been a growing impetus to improve the outcome of childhood cancer across the world. This initiative has identified investment in cancer research infrastructure and participation in collaborative research networks as a priority action. While building local capacity to improve direct patient care is an essential component for the success of the GICC initiative, local clinical research is also crucial to identify areas for improvement and barriers to implementation so that improved outcomes are achieved and can inform health policies for future sustainable actions. CONTEXT Key Objective Is it feasible and useful to conduct clinical research workshops focusing on pediatric oncology in limited-resource settings? Knowledge Generated A virtual workshop spanning 2 half-days per week for 2 weeks was successfully conducted through collaborative planning and delivery, using regional and international expertise within the Pediatric Oncology East & Mediterranean/St Jude Alliance collaborative group. Participants were of variable specialty backgrounds, included physicians, nurses, and other health care specialists. Evaluation of workshop impact and achievement of learning objectives was positive. Limitations of time and internet connectivity were identified. Participant feedback prioritized specific focus areas for future workshops. Relevance Short clinical research training workshops targeting specific specialties such as pediatric oncology using existing international, regional, and local alliances and platforms can be effective in enhancing the infrastructure and regional investigator knowledge in sound research methodologies and clinical research planning. The Pediatric Oncology East & Mediterranean Group (POEM) is a regional network of pediatric oncology health care professionals in the Middle East, North Africa, and West Asia regions. It aims to improve pediatric oncology care through collaborative initiatives, working closely with the St Jude Global (SJG) Alliance network. Since 2013, the POEM group has been conducting regional scientific and educational meetings for pediatric oncology professionals, as well as business meetings to prioritize focus areas of potential impact. Over the course of several clinical workshops, and through feedback from attendees, POEM identified a requirement for formal training in clinical research within the local context. The POEM Group collaborated with SJG on developing a framework for training in research methodologies for pediatric oncology practitioners from different backgrounds and varying levels of research experience. Key Objective Is it feasible and useful to conduct clinical research workshops focusing on pediatric oncology in limited-resource settings? Knowledge Generated A virtual workshop spanning 2 half-days per week for 2 weeks was successfully conducted through collaborative planning and delivery, using regional and international expertise within the Pediatric Oncology East & Mediterranean/St Jude Alliance collaborative group. Participants were of variable specialty backgrounds, included physicians, nurses, and other health care specialists. Evaluation of workshop impact and achievement of learning objectives was positive. Limitations of time and internet connectivity were identified. Participant feedback prioritized specific focus areas for future workshops. Relevance Short clinical research training workshops targeting specific specialties such as pediatric oncology using existing international, regional, and local alliances and platforms can be effective in enhancing the infrastructure and regional investigator knowledge in sound research methodologies and clinical research planning. A research workshop plan was prepared and implemented, as detailed in Figure . Research Workshop Committee A planning committee was formed, encompassing regional clinical researchers from within the POEM group, as well as from St Jude Children's Research Hospital. Committee members were chosen based on their interest and experience in teaching, mentorship, and clinical research. In total, the committee was composed of five researchers from the POEM region (all pediatric oncologists, with two from India and one each from Jordan, Lebanon, and Pakistan), and five researchers from St Jude Children's Research Hospital in the United States (three pediatric oncologists, one biostatistician, and one clinical trials manager). Online Planning Meetings Nine online meetings were conducted to plan the workshop agenda and curriculum. It was expected that the workshop participants would have varying backgrounds in research experience and knowledge base. The intent was to evaluate interest in the topic and identify the interested audience and their personal goals, in addition to imparting basic knowledge regarding formulating, planning, executing, and publishing a research project. Continuing medical education accreditation by the American Medical Association was provided. Although initial plans were for a 2-day onsite in-person meeting, this was modified to a virtual format because of the COVID-19 pandemic travel restrictions in 2020. To enable attendance within the practitioners' busy schedules and to best accommodate participants across various time zones, the workshop was set up as a half-day (3 hours per day), for 2 days in a row over 2 consecutive weeks. The timing of 15:00-18:00 GMT allowed participants from all time zones to join with minimal disruption. Finalization of Curriculum To best utilize the virtual setup, yet maintain a low-cost platform, electronic whiteboards were used for discussion, set up by the organizing team using PowerPoint and Word files. The workshop format included a mix of five didactic lectures, five small group hands-on sessions, and three discussion sessions delivered and mentored by committee members. Participants were required to submit an abstract for a proposed study before the workshop. The plan was to divide the attendees into groups (breakout sessions) of 6-8, with each group choosing one of their abstracts to work on and further develop over the course of the workshop. The final output would be presented by each group at the conclusion of the workshop. Workshop Delivery The workshop was delivered as planned, in 4 half-days over 2 weeks in November 2020. Didactic presentations covered clinical studies and methodologies, cancer data registries, introduction to health care statistics, research ethics and governance, and guidance for scientific writing and publishing. Breakout sessions covered discussions of abstract evaluation, and formulation of the objectives, output and deliverables, methodology, timeline, and resources, respectively. Workshop Follow-Up At the conclusion of the workshop, an evaluation survey (Data Supplement) was distributed to all participants to rate the quality of the workshop and the knowledge gained (competencies). A few weeks after the workshop, a second survey (Data Supplement) was distributed by email to all workshop applicants to identify reasons for choosing to apply for the workshop, prior experience with other research training, and rating of knowledge gained. A planning committee was formed, encompassing regional clinical researchers from within the POEM group, as well as from St Jude Children's Research Hospital. Committee members were chosen based on their interest and experience in teaching, mentorship, and clinical research. In total, the committee was composed of five researchers from the POEM region (all pediatric oncologists, with two from India and one each from Jordan, Lebanon, and Pakistan), and five researchers from St Jude Children's Research Hospital in the United States (three pediatric oncologists, one biostatistician, and one clinical trials manager). Nine online meetings were conducted to plan the workshop agenda and curriculum. It was expected that the workshop participants would have varying backgrounds in research experience and knowledge base. The intent was to evaluate interest in the topic and identify the interested audience and their personal goals, in addition to imparting basic knowledge regarding formulating, planning, executing, and publishing a research project. Continuing medical education accreditation by the American Medical Association was provided. Although initial plans were for a 2-day onsite in-person meeting, this was modified to a virtual format because of the COVID-19 pandemic travel restrictions in 2020. To enable attendance within the practitioners' busy schedules and to best accommodate participants across various time zones, the workshop was set up as a half-day (3 hours per day), for 2 days in a row over 2 consecutive weeks. The timing of 15:00-18:00 GMT allowed participants from all time zones to join with minimal disruption. To best utilize the virtual setup, yet maintain a low-cost platform, electronic whiteboards were used for discussion, set up by the organizing team using PowerPoint and Word files. The workshop format included a mix of five didactic lectures, five small group hands-on sessions, and three discussion sessions delivered and mentored by committee members. Participants were required to submit an abstract for a proposed study before the workshop. The plan was to divide the attendees into groups (breakout sessions) of 6-8, with each group choosing one of their abstracts to work on and further develop over the course of the workshop. The final output would be presented by each group at the conclusion of the workshop. The workshop was delivered as planned, in 4 half-days over 2 weeks in November 2020. Didactic presentations covered clinical studies and methodologies, cancer data registries, introduction to health care statistics, research ethics and governance, and guidance for scientific writing and publishing. Breakout sessions covered discussions of abstract evaluation, and formulation of the objectives, output and deliverables, methodology, timeline, and resources, respectively. At the conclusion of the workshop, an evaluation survey (Data Supplement) was distributed to all participants to rate the quality of the workshop and the knowledge gained (competencies). A few weeks after the workshop, a second survey (Data Supplement) was distributed by email to all workshop applicants to identify reasons for choosing to apply for the workshop, prior experience with other research training, and rating of knowledge gained. Registration for the workshop was opened for a period of 30 days. A total of 38 registrations were received on a first-come basis, after which the application platform was closed because of preplanned capping of participants. Registrants were distributed across 12 countries and six disciplines (23 medical doctors, six research fellows/associates, five nurses, two nutritionists, one pharmacist, and one data manager). Eight registrants (21%) were not previously POEM group members and had the announcement forwarded to them by a colleague. All but one were interested in American Medical Association continuing medical education credits. Of the 38 initial registrants, 29 (76%) attended the workshop. Of those participants, 19 (66%) attended more than 70% of the workshop, eight (27%) attended more than 50% but < 70% of the content, and two (7%) attended < 50% of the content time. The submitted abstracts included retrospective chart review studies, prospective observational studies, prospective interventional studies, quality improvement projects, cross-sectional survey studies, longitudinal survey studies, and registry studies. The breakout groups were accordingly divided by abstract themes, specifically cohort studies (two groups), clinical trials (two groups), and registries (one group). On the first day, each group of 6-8 participants reviewed all the abstracts submitted by their group and assigned scores on the areas of significance, feasibility, generalizability, and value added, while listing strengths and weaknesses in each of these categories (Data Supplement). This was facilitated by the committee members. One abstract was then selected in each group to further develop over the remainder of the workshop. Didactic sessions covered the following topics: Fundamentals of clinical research and formulating a research question, research designs and types of biomedical research, introduction to statistical analytic methods, implementation and use of cancer data registries, and how to write up results and publish a manuscript. In the breakout sessions, each of the five groups used preformatted worksheet templates to guide them through developing their proposal. The worksheet templates (Data Supplement) included exercises on how to review and assess an abstract and in writing out elements for a proposed study. Each group used these templates to develop the selected abstract. On the last day, each group compiled their work using a common summary template (Data Supplement) and presented their proposed study to all participants using a common PowerPoint template, detailing the study's background, research question, objectives, deliverables, methodology and statistical plan, timelines and projected deliverable dates, resources needed, expected impact, and next steps. The final output from the breakout groups included a retrospective study, a prospective observational study, a prospective interventional study, and a registry-planning proposal—on the basis of the abstracts selected for development during the first day of the workshop. The postworkshop appraisal by attendees showed a highly positive overall experience, with more than 90% of participants scoring all content aspects of the course as excellent, very good, or good (Data Supplement). As expected, free text feedback regarding barriers revealed internet connectivity issues and an interest in blended format or on-site in-person future workshops for face-to-face interactions. A few comments advocated for more detailed sessions and a longer course, to be spread out over more days, with more sessions on statistical analysis planning, and on learning to publish. When asked about the most effective aspects of the workshop, 45% pointed out the breakout sessions and the chance for discussion within small groups. All attendees stated that what they learned will impact their practice moving forward. When asked to suggest topics for future workshops on research methodologies, the three topics that were scored highest were: statistical analysis, designing quality improvement projects, and scientific writing and publishing. After the workshop, participants were advised to further develop their proposals on the basis of what they learned and were offered a private session to review and discuss their modified abstract with the workshop facilitators. A deadline of two months was given to submit revised proposals. Of the initial 38 abstracts, five were submitted (13%), of which three (8%) were reviewed and discussed, as the remaining two did not follow-up further. Of the three reviewed, one was moved forward as a quality improvement study at the local institution, one was developed further within the POEM group for a multi-institutional prospective observational study, and one was used as part of the participant's graduate thesis at their institution. All applicants reported that the discussions and feedback greatly improved their proposals. An anonymous survey was conducted after the workshop to better understand the reasons for attendance, availability of other training opportunities, barriers to attending educational workshops, and general feedback. Of the 38 who were e-mailed the survey, 14 responded (37%). Responses to the survey are shown in Table . Of all respondents, 50% had previous research experience, and the majority were more than 5 years within profession. Eight (57%) had previously attended research workshops. Eleven (78%) felt that the current workshop met their learning goals, and five (36%) reported barriers related to internet connectivity and time management. Regarding length of the workshop, three (21%) preferred if it were shorter or spread out over more weeks, one (7%) preferred it to be longer and more in depth, and 10 (71%) did not comment on any needed changes in length. All expressed improved knowledge after the workshop. Capacity building for clinical and health research in limited-resource settings has long been identified as a need. , However, progress has been hampered because of multiple factors such as lack of protected time, lack of funding for research training, and limited national research funding support, among others. , Clinicians and trainees in limited resource countries usually are focused on clinical training with very limited education in research methodologies because of the source of funding and competing priorities. With the recent progress in pediatric oncology collaborative efforts across the world, bolstered further by the recent WHO-endorsed GICC, there has been a renewed interest in improving local research infrastructure, to enable effective collaborative efforts both in assessing the status of cancer care delivery and implementing changes to improve pediatric cancer outcomes. Thus, the timing of such interventional efforts is optimal and is likely to synergize with ongoing global, regional, and local initiatives. There have been multiple efforts to enhance training in health and medical research in limited-resource settings with different formats including formal postgraduate degrees, or short courses and online learning, with most initiatives deployed at institutional collaboration levels, - and a few targeting specific fields of research such as mental health, , cardiovascular diseases, or hematology. However, there has been very limited training in clinical research focused on pediatric oncology specifically, and none to our knowledge targeting the Middle East, North Africa, and West Asia region, which is the geographical area comprising the POEM consortium. Our approach was built on an existing long-term strategic collaboration between the POEM group and St Jude. We aimed for this first research methodology workshop to target a wide range of interested clinicians of different backgrounds and to broadly improve the understanding of basic concepts of how to approach a research project from hypothesis formulation to planning its execution and publication. This included not only pediatric oncologists at different levels of their professional careers but also nurses interested in conducting research studies, where traditionally little engagement in limited-resource settings has been possible. Several opportunities were identified through this initiative. There was a high level of interest and engagement from a range of not only early- and mid-career but also established practitioners. In addition, there was notable engagement by participants from nursing and other health care specialty backgrounds, in addition to medical doctors. The feedback assessment identified specific topics of interest and relevance, which will be prioritized for the upcoming workshops. In addition, this exercise was very informative in how to use virtual platforms for small-group discussions and hands-on workshops within the POEM group and has allowed us to further develop the tools used. We also identified a need for regional mentorship in research and have since developed a Research Committee within POEM to work toward addressing these needs. Through such research workshops, and in collaboration with the POEM Research Committee, we hope to assist in bolstering and further development of the regional research infrastructure and enhance the interest and understanding of robust research strategies in pediatric oncology in constrained-resource settings. Furthermore, as this initiative expands into next phases, we will address the optimization and scalability of our efforts. Limitations were also identified through this project. The virtual nature of the workshop, relatively new at the time, and internet connectivity issues in limited-resource settings, likely contributed to the noted difficulties with on-line attendance. Indeed, the feedback survey from the participants listed internet connectivity issues and time management for virtual meeting attendance as the primary problems hampering effective participation. As for the length of the workshop, the feedback was mixed. We had initially decided on a short workshop to mitigate the time commitment that busy practitioners are able to spare and to limit screen time on any particular day. The feedback received showed this to be agreeable to most of the survey respondents (71%), although three (21%) felt it was too time-intensive and one (7%) preferred a longer format with more in-depth topic discussion. Moving forward, we plan to focus the topics of future workshops, which would allow in-depth discussion of a specific area, while maintaining a limited time commitment and short screen time per day. As such, a higher number of focused workshops would likely confer the most benefit while mitigating time barriers to participation. This initiative highlights the importance of systematically identifying strengths and gaps in the local and regional health care delivery systems that could identify opportunities for improvement through carefully designed clinical research via quality and improvement science. Since this workshop was conducted in November 2020, the POEM-SJG partnership has conducted a second research workshop in November 2021, this time with a focused objective on understanding quality improvement methodologies and assessing published research manuscripts. A third workshop is in planning phase for March 2023, for mentored writing of a research study, and will include prework self-paced study sessions. Future workshops will incorporate online courses and recordings as supplements to try to circumvent live internet connectivity problems. The current workshop, as well as the future ones in planning, will be compiled into a Research Methodologies Curriculum, which can then be adapted and used for participants with specific levels of research skills, focusing on contexts relevant to limited-resource settings.
Dilemme en néonatalogie moderne: prise en charge de l’extrême prématurité	78c1eae5-9d59-49a9-982e-22e7315e4508	10166619	Pediatrics[mh]	Quand le travail a commencé, à 21 semaines et 5 jours de gestation, on nous a dit que j’allais (Shakina) faire une fausse-couche et qu’il n’y avait rien à faire pour sauver les bébés parce qu’ils n’étaient « pas viables ». Tout ce que l’hôpital pouvait offrir était des soins de confort, ce qui signifiait placer les bébés sur nous après leur naissance et attendre leur décès. Aucune autre intervention ne serait tentée pour les garder en vie. Une infirmière m’a remis une cuvette pour que je puisse « attraper » les bébés dans la salle de bain s’ils étaient « expulsés ». Rien ne peut exprimer l’ampleur du traumatisme émotionnel, mental et physique que j’ai subi. Je (Kevin) me rappelle qu’il était 2 h 30 du matin, je ne dormais pas, je pleurais et je priais Dieu désespérément pour un signe d’espoir. Peu après, une amie proche, Jennifer, m’a envoyé le contact du compte Instagram de TwentyTwo Matters, qui milite pour la défense des fœtus de 22 semaines et fournit une liste d’hôpitaux qui les prennent en charge. L’Hôpital Mount Sinai était au sommet de cette liste et le lendemain matin, notre demande de transfert a été acceptée. À l’Hôpital Mount Sinai, on nous a informés qu’aucune mesure de réanimation n’allait être appliquée si les bébés arrivaient avant 22 semaines de gestation. On nous a aussi informés que même s’ils survivaient, les bébés souffriraient de plusieurs handicaps pouvant nuire à leur qualité de vie. Nous avons reconnu ces risques et maintenu notre position: les bébés méritaient une chance de vivre. Leur qualité de vie allait être déterminée non pas par d’éventuels handicaps, mais bien par l’amour, les soins et le soutien qu’ils recevraient. Par miracle, les bébés sont nés tout juste après minuit, à 22 semaines et 0 jour de gestation et ils ont été réanimés avec succès. Les quelques premières semaines ont été les plus éprouvantes de nos vies. Ç’a été extrêmement pénible d’assister à toutes les interventions effractives (piqûres, tests et traitements) auxquelles Adiah et Adrial ont été soumis. Nos bébés ont failli mourir sous nos yeux à plusieurs reprises et les médecins nous demandaient de réfléchir au moment où nous souhaiterions arrêter les soins. Même atterrés, nous avons réaffirmé notre désir de continuer à nous battre pour eux. Leurs vies étaient importantes et valaient la peine d’être sauvées. Les médecins s’inquiétaient des défis, des risques et des complications à venir. Nous avons choisi de défendre nos bébés coûte que coûte et de nous concentrer sur leurs progrès au jour le jour. Nous avons toujours été soutenus par le personnel en soins infirmiers et en travail social et par les autres parents de l’USIN qui appuyaient nos décisions; aux jours les plus sombres, ils nous ont aidés à demeurer positifs et à garder espoir. Malgré les limites de la médecine, nous avons cru que d’autres facteurs pouvaient améliorer les chances de nos bébés. Nous sommes de confession chrétienne et extrêmement croyants, alors nous avons demandé à nos amis et à nos proches de lancer un appel; et de tous les coins du monde, une prière s’est élevée pour nos bébés. Nous avons décidé que peu importe notre inquiétude, nous allions toujours sourire, chanter et célébrer chaque étape de leur vie avec nos bébés. Le modèle de soins familiaux intégrés nous a permis de participer pleinement à leurs soins quotidiens. Ils pouvaient sentir notre présence et se savoir aimés, et nous croyons que cela a changé la donne pour eux. L’immense aide émotionnelle, psychologique, financière, spirituelle et matérielle que nous avons reçue de nos amis, de nos familles et de nos communautés nous a également permis de consacrer tout notre temps et toute notre énergie à nos bébés hospitalisés à l’USIN pendant près de 6 mois. Ramener nos jumeaux à la maison a été une sorte de miracle qui nous a donné la force de nous engager pour la défense d’autres nourrissons prématurés, comme Adiah et Adrial, qui ne seraient pas en vie aujourd’hui si les limites de la viabilité n’avaient pas été remises en question par les équipes soignantes. — Shakina Rajendram et Kevin Nadarajah Les prestataires de soins de santé qui travaillent à l’USIN sont souvent perçus par le public comme des magiciens et des précurseurs en sciences médicales, mais on entend peu parler de leurs dilemmes d’ordre éthique et de leurs combats intérieurs. Pendant une période de ma carrière en soins infirmiers à l’USIN, je me suis demandé si je faisais partie de la solution ou du problème. Les progrès de la médecine ont permis de revoir l’âge de la viabilité. Mais malgré de grands succès, les soins intensifs néonataux comportent des défis et des risques particuliers. Nous pouvons certes maintenir les bébés en vie, mais nous savons que toute complication peut entraîner des séquelles permanentes, voire la mort. J’ai souvent déploré de ne pas connaître à l’avance les possibles répercussions négatives de nos soins, de ne pas savoir si le maintien des fonctions vitales ne faisait que retarder l’inéluctable ou permettre une survie sans qualité de vie. Ce questionnement est encore plus fort quand des parents semblent se désintéresser des soins qu’on prodigue à leur enfant, tout en refusant de cesser les soins face à un pronostic sombre; il semble alors que tous nos efforts et nos meilleures intentions ont des répercussions négatives sur la famille et peutêtre même sur la société entière. La nature est peut-être déterministe, mais les soins attentifs fournis par chaque personne qui se sent concernée maximisent les chances de réaliser pleinement le potentiel des nourrissons prématurés. C’était mon point de vue lorsque Kevin et Shakina m’ont demandé d’être l’une des infirmières attitrées aux soins de leurs jumeaux. Ma première rencontre avec la famille a eu lieu durant la première semaine de vie des bébés. Kevin et Shakina ne manifestaient aucun signe de désintérêt ou d’accablement, mais plutôt une volonté de s’impliquer. Étant donné la petite taille et la fragilité extrêmes des bébés, il était encore impossible de les prendre dans nos bras comme s’ils étaient des bébés à terme. J’ai donc encouragé les parents à les entourer de leurs mains dans l’incubateur, comme façon de les enlacer et j’ai pu immortaliser en photo et en vidéo les premiers moments de ce lien qui se créait. Lorsque les jumeaux ont atteint l’âge de 4 semaines, j’ai remarqué que la photo de cette étreinte était affichée devant la chambre. À partir de ce moment, j’ai su que ces parents allaient saisir chaque infime occasion de consolider leur lien avec les bébés et j’ai décidé d’être l’infirmière attitrée à cette famille, un rôle créé par notre unité pour les nouveau-nés susceptibles d’y séjourner longtemps. Cela signifiait que j’allais être affectée aux soins d’Adiah et d’Adrial à chacun de mes quarts de travail, jusqu’à leur congé afin d’assurer la cohésion et la régularité des soins. J’ai été mise au fait des réunions familiales précédentes et de la détermination des parents à toujours pousser les soins. Je savais que des soins intensifs leur avaient été administrés et que leur avenir était incertain. La route s’annonçait parsemée d’embûches pour la famille. Les 2 parents comprenaient bien la situation; ils n’étaient pas dans le déni. La plupart des familles ne comprennent pas toujours bien ce qu’impliquent les soins infirmiers 24 heures sur 24 à domicile pour l’enfant et la maisonnée; elles restent dans le déni. C’est leur façon de composer avec un pronostic dévastateur. Ce ne sont pas toutes les familles qui arrivent à franchir les 5 étapes de ce type de deuil. Or, cette famille se donnait la peine de réfléchir aux défis qui la guettaient. Ses membres se sont montrés résilients et proactifs et ils ont continué d’apprendre et de se battre pour les enfants. C’est la détermination de Kevin et de Shakina à surmonter tous ces défis qui nous ont motivées, moimême et toute l’équipe de l’USIN, à laisser de côté nos doutes à nous donner à fond. Pendant les 5 mois qu’a duré notre relation infirmière–famille, nous avons eu l’occasion de parler sérieusement. Au cours d’une de ces conversations, Shakina nous a dit: « Nous savons qu’Adiah et Adrial vont franchir les étapes de leur développement à un rythme différent. Nous avons convenu d’enseigner à nos enfants à ne pas célébrer leurs propres étapes, mais bien celles de l’autre jumeau. De cette façon, nous pouvons nous rappeler, et rappeler à nos enfants de ne pas se comparer l’un à l’autre, mais d’apprécier chaque petite ou grande victoire ». À l’USIN, nous n’avons pas toutes les réponses, mais je crois que nous avons eu un effet positif en conseillant ces parents, en les soulageant d’une part de leur fardeau et en les préparant aux défis à venir. — Luzia Leong S’occuper de jumeaux nés à 22 semaines de gestation comportait une bonne part d’inconnu pour notre équipe. Lors de ma première rencontre avec les parents, j’ai voulu connaître leur point de vue. Ils nous ont résumé ce qu’on leur avait dit au sujet de la réanimation à cet âge gestationnel. En sachant que leurs enfants avaient commencé leur vie en recevant de justesse des soins actifs (à 1 heure près), Shakina et Kevin ont choisi d’y voir une décision des bébés eux-mêmes et ils ont voulu lutter à leur côté pour leur survie. J’ai résumé les grandes lignes du modèle des soins familiaux intégrés que nous pratiquons ici et j’ai mentionné l’importance d’être présents pour les bébés, de leur parler, de les toucher et, éventuellement, de participer à leurs soins, y compris par contact peau à peau. Kevin m’a fait remarquer que j’étais la première personne à leur parler de la vie et de l’implication, plutôt que d’une éventuelle issue négative. Je me suis alors demandé si je n’avais fait que présenter mon introduction de routine ou si j’avais été utopiste. Aurais-je dû être plus réservée? Nous sommes revenus là-dessus quelques semaines plus tard et les parents m’ont assuré que cet échange avait été positif: ils étaient heureux que je ne me sois pas censurée, car cette conversation avait nourri leur confiance en la survie de leurs bébés. Shakina et Kevin étaient là tous les jours. Ils participaient aux soins des bébés de toutes les façons possibles. Certains membres de l’équipe s’inquiétaient que les parents « ne comprennent pas » et qu’ils se donnent autant, au lieu de baisser les bras et de pleurer. C’était frustrant pour moi parce que ma perception était toute autre et je poussais en ce sens: je faisais face à des parents qui célébraient la vie, qui apportaient une énergie positive, les pieds bien ancrés dans la réalité malgré leur peine face à la situation. Les parents m’ont dit qu’ils ne nourrissaient pas de rêves d’avenir, qu’ils essayaient simplement d’aider leurs bébés, et à l’évidence, cette tâche bien terre à terre comptait plus que tout pour eux. Au fil des discussions et des tournées, avec le temps, l’équipe a graduellement adopté la même position que moi; nous avons développé une compréhension commune de la situation de cette famille. C’est dans cet esprit, qui consistait à se réjouir de chaque étape, qu’au bout de 7 jours, j’ai remis aux parents une carte de souhaits pour souligner la « 1 re semaine de vie » des bébés. Ils l’ont appréciée. Ils ont dit à quel point ils se sentaient validés du fait qu’un membre de l’équipe célèbre avec eux. Bien sûr, l’avenir n’était pas rose, mais ces parents ne lâchaient pas, ils soulignaient chaque moment et ma place était à leur côté. Nous avons eu plusieurs réunions familiales au sujet de l’effet des traitements et de leur orientation. Ils ont continué de demander le maximum pour offrir les meilleures chances possibles aux bébés et je me suis demandé comment ils pouvaient à la fois comprendre la situation et ne pas « rêver en couleur ». Une telle prématurité est un monde inconnu pour n’importe quelle famille. Nous avons discuté de façons d’en parler avec l’équipe; ils ont ainsi acquis un nouveau sentiment d’autonomie. Ç’a été formidable de voir grandir leur confiance à l’endroit des soins et leur détermination à défendre leurs bébés; ils ont appris de nous, mais nous avons appris d’eux à faire confiance et mesurer les répercussions d’une intention positive. — Sara Gambino J’ai reçu un appel d’un jeune collègue au sujet de soins de réanimation pour des nourrissons à 21 semaines et 6 jours de gestation, et Shakina nous a été transférée à l’Hôpital Mount Sinai. La conduite a été de discuter avec les parents pour leur expliquer les conséquences à court et à long terme d’une si grande prématurité et les possibilités de la technologie, comprendre leurs attentes et convenir ensemble d’un plan. Au Canada, la plupart des centres n’offrent la réanimation néonatale et les soins intensifs actifs aux nouveau-nés qu’à partir de 23 semaines de gestation, après consultation avec les parents, en raison du taux élevé de mortalité et de morbidité neurosensorielle et développementale substantielles chez les survivants qui reçoivent ces types de soins avant la 23 e semaine de gestation. Le débat sur la réanimation selon l’âge de viabilité n’est pas nouveau en néonatalogie et l’âge de la viabilité est constamment remis en question. La position qui consiste à croire la réanimation active superflue en raison du pronostic sombre et à ne faire état d’aucune survie soulève une foule de questions d’ordre médical, éthique et moral. Les opinions et les croyances personnelles, et le besoin de les justifier ont fait en sorte que beaucoup de prestataires de soins de santé sont sur la défensive ou prudents. Quand j’ai fait la connaissance de Shakina et Kevin, 2 semaines après la naissance d’Adiah et d’Adrial, j’ai compris à quel point ils se souciaient de l’état de santé et des progrès de leurs enfants. Ils ont fermement réitéré leur refus qu’on suspende ou qu’on cesse les soins de maintien des fonctions vitales. Leur problème était qu’on ait sans cesse remis en question leur choix de maintenir des soins actifs complets face aux plus graves complications de leurs jumeaux. Après cette conversation, l’orientation des soins a changé pour assurer leur survie et l’équipe a pris en compte les objectifs globaux de cette famille. Cette semaine-là, Adrial a été très malade en raison d’une perforation intestinale et d’une infection, et Adiah a perdu beaucoup de poids. Ces complications auraient pu entraîner la mort ou de graves séquelles, mais les 2 parents ont gardé espoir et se sont accrochés à chaque amélioration de l’état de l’un ou l’autre des bébés et n’ont cessé de s’appuyer mutuellement. J’ai choisi la néonatologie principalement parce que cette spécialité peut constamment évoluer. Aucun d’entre nous ne pouvait prédire quelles seraient les conséquences à long terme des problèmes de santé des bébés; mais une chose restait sûre, les parents, eux, n’ont jamais perdu espoir. Plusieurs parmi mes collègues et dans l’équipe se demandaient: « Est-ce qu’on fait la bonne chose? », « Qu’essaie-t-on de prouver ici? » et surtout « Les parents comprennent-ils les difficultés qui les guettent? ». J’aurais bien aimé connaître la réponse à ces questions. J’ai écouté les doutes des membres de l’équipe et j’ai fait de mon mieux pour que tous comprennent le point de vue des parents durant les réunions. Quand j’ai rencontré les parents, avec leur infirmière et leur travailleuse sociale, je me suis rallié à leur optimisme, tout en essayant de ne pas les induire en erreur. Ce dont j’étais sûr, c’est que ces parents avaient tout ce qu’il faut pour accomplir leur tâche. Je voyais dans leur regard un espoir, une joie et une fierté immenses. Le débat autour de la réanimation chez les bébés à la limite de la viabilité n’est pas près de s’éteindre; mais, les leçons d’humilité de toutes les familles touchées me rappellent à chaque fois à quel point il est important qu’ils participent aux soins, à l’établissement des objectifs et aux prises de décisions. C’est un domaine où il faut guider les parents et les familles en restant à l’écoute de leurs croyances, de leurs valeurs et de leurs attentes. Je les appuie autant que possible, à l’intérieur des limites de ce que peut véritablement offrir l’équipe aux enfants dans les circonstances. — Prakesh S. Shah Les articles Consultation-360 ° mettent en lumière certains aspects interpersonnels et systémiques des soins de santé rarement abordés dans les autres articles de cette section du JAMC . Chaque article comporte un résumé des antécédents médicaux et les réflexions personnelles de 2–4 personnes impliquées dans une même consultation clinique. Un des auteurs est toujours un patient, un membre de la famille ou un proche aidant. Les réflexions des autres auteurs (p. ex., médecins, personnel infirmier, travailleurs sociaux, diététiciennes, etc.) mettent en évidence divers points de vue représentés au cours de la consultation. Pour plus d’information, veuillez consulter https://www.cmaj.ca/submission-guidelines ou communiquer avec [email protected] .
The landscape of neurophysiological outcome measures in ALS interventional trials: A systematic review	41a87059-c401-4a59-9028-40994730bce1	10166714	Physiology[mh]	Introduction Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease affecting both upper and lower motor neurons , characterised by loss of glial cells and degeneration of motor neurons in the brain and spinal cord . ALS is relentlessly progressive with a median survival from symptom onset of 3–5 years . There is currently no cure for ALS; the only approved drug in Europe is riluzole , while edaravone has also been licensed in Japan and USA. Despite many clinical trials in ALS, most have yielded negative results. A contributing factor may be the relative insensitivity of trial outcome measures, emphasising the importance of a carefully chosen set of measures relevant to the intervention’s proposed mechanism of action . Current outcome measures employed include the revised ALS functional rating scale (ALSFRS-R), forced/slow vital capacity and neurophysiological assessments. Neurophysiological biomarkers assess motor unit dysfunction, death and/or compensatory adaptations, thereby aiding diagnosis and tracking disease progression in ALS. Their diagnostic importance has been highlighted in both the revised El Escorial criteria and the Gold Coast criteria . Not only can they provide detailed insight into disease mechanisms, but they can also be used as non-invasive outcome measures in therapeutic clinical trials . Moreover, since neurophysiological techniques can detect subclinical disease progression , they have the potential to increase the statistical power of future clinical trials . In this systematic review, we provide a comprehensive overview of all ALS clinical interventional trials that have employed neurophysiological outcome measures, assessing their advantages and limitations alongside future recommendations. Methods The reporting of this systematic review is in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines . The protocol for this review has not been published. 2.1 Study selection Studies were included in the review according to the following inclusion criteria: (1) an interventional clinical trial; (2) conducted on patients with ALS; (3) the primary and/or secondary outcome measure(s) included at least one neurophysiological measure; and (4) the study was written in English. 2.2 Search strategy A systematic computerised search was completed by one author (NA) in February 2021 in eight databases: MEDLINE (1946–), PubMed (1972–), Web of Science (1955–), Embase (1947–), PsychInfo, OpenGrey, Cumulative Index to Nursing and Allied Health Literature (CINAHL) and Education Resources Information Centre (ERIC). Searches were tailored to the databases and included key text-word terms, phrases, and medical subject headings (MESH) terms with “and/or” for “amyotrophic lateral sclerosis”, “motor neuron disease”, “clinical trial”, “interventional trial”, “neurophysiological outcome” and related. A full search strategy is presented as a table in Supplementary Table 1. Duplicates amongst the identified studies were removed. The remaining studies were screened by one author (NA), initially based on title and abstract. Of those remaining, the full-text article was assessed, and any queries were resolved by discussion with the senior author (JB). The final list of selected studies was agreed upon by two authors (NA/JB). A repeat search was conducted on 14th October 2021 to include any updates. 2.3 Data extraction and quality assessment Relevant data were extracted from the selected studies in a table with headings as follows: study details (location, sponsors, and type of trial), study design (number of patients and the proposed intervention including mechanism of action), outcome measures, number and duration of follow-ups, results collected, and any strengths/limitations of the neurophysiological measurements used. Additionally, the outcome with respect to the neurophysiological measure has been recorded as positive (where p < 0.05) or negative. To assess the risk of bias in the selected trials, we used a modified form of the Newcastle-Ottawa Scale and the Cochrane Risk of Bias Assessment Tool . Both scales assess the risk of bias in three main domains: (1) Selection of the study groups; (2) Comparability between the two groups; and (3) Ascertainment of the outcome and/or the exposure. With the maximum score being 8, studies with a score of ≥7 were considered to have low risk of bias, those with a score of 4–6 were considered to have moderate bias, and scores of <4 were considered to have high risk of bias. We determined adequate follow-up length to be a period of at least 6 months. Study selection Studies were included in the review according to the following inclusion criteria: (1) an interventional clinical trial; (2) conducted on patients with ALS; (3) the primary and/or secondary outcome measure(s) included at least one neurophysiological measure; and (4) the study was written in English. Search strategy A systematic computerised search was completed by one author (NA) in February 2021 in eight databases: MEDLINE (1946–), PubMed (1972–), Web of Science (1955–), Embase (1947–), PsychInfo, OpenGrey, Cumulative Index to Nursing and Allied Health Literature (CINAHL) and Education Resources Information Centre (ERIC). Searches were tailored to the databases and included key text-word terms, phrases, and medical subject headings (MESH) terms with “and/or” for “amyotrophic lateral sclerosis”, “motor neuron disease”, “clinical trial”, “interventional trial”, “neurophysiological outcome” and related. A full search strategy is presented as a table in Supplementary Table 1. Duplicates amongst the identified studies were removed. The remaining studies were screened by one author (NA), initially based on title and abstract. Of those remaining, the full-text article was assessed, and any queries were resolved by discussion with the senior author (JB). The final list of selected studies was agreed upon by two authors (NA/JB). A repeat search was conducted on 14th October 2021 to include any updates. Data extraction and quality assessment Relevant data were extracted from the selected studies in a table with headings as follows: study details (location, sponsors, and type of trial), study design (number of patients and the proposed intervention including mechanism of action), outcome measures, number and duration of follow-ups, results collected, and any strengths/limitations of the neurophysiological measurements used. Additionally, the outcome with respect to the neurophysiological measure has been recorded as positive (where p < 0.05) or negative. To assess the risk of bias in the selected trials, we used a modified form of the Newcastle-Ottawa Scale and the Cochrane Risk of Bias Assessment Tool . Both scales assess the risk of bias in three main domains: (1) Selection of the study groups; (2) Comparability between the two groups; and (3) Ascertainment of the outcome and/or the exposure. With the maximum score being 8, studies with a score of ≥7 were considered to have low risk of bias, those with a score of 4–6 were considered to have moderate bias, and scores of <4 were considered to have high risk of bias. We determined adequate follow-up length to be a period of at least 6 months. Results 3.1 Study selection A total of 703 studies were obtained from reference and database searches after the removal of duplicates . The titles and abstracts were screened, and 667 studies were excluded because they did not meet the inclusion criteria. Of the 36 studies remaining, the full text was assessed for eligibility and four were excluded: one did not provide sufficient trial details; one trial had been withdrawn; and the remaining two did not involve neurophysiological outcome measures. Thus, a total of 32 studies were included for qualitative synthesis. 3.2 Study and patient characteristics Eleven of the 32 trials were conducted in Europe , nine in North America , six in Asia , two in Australia , and four were intercontinental . Under half of the trials (41%) were randomised, placebo-controlled and blinded, while the majority of the remainder were reported as either open-label or pilot trials. Sixteen trials (50%) were labelled as either phase 1 (two trials; 6.2%), phase 1/2 (three trials; 9.4%), phase 2 (ten trials; 31.3%), or phase 2/3 (one trial; 3.1%). No trials were labelled as phase 3. Fifteen trials (47%) were conducted over multiple sites. All patients had a diagnosis of definite or probable ALS. The total number of participants from the completed, published trials was 1128, with a median of 30 participants per trial (range: 3–300). Of the completed studies, the mean age varied from 20 to 62.4 years, with 51% of patients being male. The number of estimated patients participating in ongoing studies was 445. 3.3 Neurophysiological outcome measures The neurophysiological assessments used were needle electromyography (NEMG; n = 12), compound muscle action potential (CMAP; n = 8), neurophysiological index (NI; n = 7), motor unit number estimation (MUNE; n = 7), motor unit number index (MUNIX; n = 5), peripheral motor nerve excitability testing (PMNET; n = 5), short-interval intracortical inhibition (SICI; n = 4), resting motor threshold (RMT; n = 3), surface electromyography (SEMG; n = 3), and fasciculation frequency detected by ultrasound (n = 2; , ) . A neurophysiological measure was the primary outcome measure in seven trials (22%), and at least one neurophysiological measure was included as a secondary outcome measure in 28 trials (88%). Fifteen trials (47%) employed a combination of neurophysiological parameters. Final follow-up assessments varied with a median of 4 follow-up assessments per study. The median follow-up time was 1 month with a range of 6 hours – 2 years. Of the completed trials, 1 out of 7 employing NEMG described an improvement in the outcome measure with the intervention. For CMAP and NI, the positivity rates were 2/8 and 2/6, respectively. Out of the transcranial magnetic stimulation measures, the use of RMT and SICI were associated with positive outcomes in 2/3 and 2/4 trials, respectively. Of the 7 studies using MUNE and/or MUNIX, only 1 reported a positive finding (a MUNIX study). With regards to MUNE, three trials used statistical MUNE, two used the multipoint incremental method, and one used spike-triggering MUNE. Positive effects were seen in 3 out of 4 completed trials reporting PMNET as an outcome measure. contains a summary of all trials. 3.4 Study interventions Almost a third (28%) of the included trials tested an intervention that targeted a neuronal ion channel, mainly sodium-channel blockade, NMDA blockade or potassium-channel agonism . The other major groups of interventions targeted either the immune system (50%) or neuronal metabolic pathways (22%). Most interventions were oral treatments (52%), while a minority were administered either intravenously (6%), subcutaneously (3%), intrathecally (13%), or using a mixture of methods (26%). Only one trial had a non-pharmacological intervention in the form of transcranial direct current stimulation. The majority of trials (69%) took place during the last ten years. Under half of the completed trials (39%) were positive with respect to at least one neurophysiological outcome measure. A quarter of reported studies remain incomplete. 3.5 Study quality The full risk of bias assessment table is presented in Supplementary Table 2. The study quality ranged from high risk of bias (n = 2) to a very low risk of bias (n = 2) . All included studies accurately assessed the outcome using the neurophysiological measurements detailed above. As defined above, adequate follow up length was determined to be at least 6 months and most studies did not have an adequate follow up, thus reducing the overall quality of the studies. All completed studies highlighted the attrition rate, apart from three studies , which did not describe the loss to follow up. Overall, the studies included in this systematic review had a low risk of bias with 15 of the 24 studies assessed for risk scoring 7 and above. . Study selection A total of 703 studies were obtained from reference and database searches after the removal of duplicates . The titles and abstracts were screened, and 667 studies were excluded because they did not meet the inclusion criteria. Of the 36 studies remaining, the full text was assessed for eligibility and four were excluded: one did not provide sufficient trial details; one trial had been withdrawn; and the remaining two did not involve neurophysiological outcome measures. Thus, a total of 32 studies were included for qualitative synthesis. Study and patient characteristics Eleven of the 32 trials were conducted in Europe , nine in North America , six in Asia , two in Australia , and four were intercontinental . Under half of the trials (41%) were randomised, placebo-controlled and blinded, while the majority of the remainder were reported as either open-label or pilot trials. Sixteen trials (50%) were labelled as either phase 1 (two trials; 6.2%), phase 1/2 (three trials; 9.4%), phase 2 (ten trials; 31.3%), or phase 2/3 (one trial; 3.1%). No trials were labelled as phase 3. Fifteen trials (47%) were conducted over multiple sites. All patients had a diagnosis of definite or probable ALS. The total number of participants from the completed, published trials was 1128, with a median of 30 participants per trial (range: 3–300). Of the completed studies, the mean age varied from 20 to 62.4 years, with 51% of patients being male. The number of estimated patients participating in ongoing studies was 445. Neurophysiological outcome measures The neurophysiological assessments used were needle electromyography (NEMG; n = 12), compound muscle action potential (CMAP; n = 8), neurophysiological index (NI; n = 7), motor unit number estimation (MUNE; n = 7), motor unit number index (MUNIX; n = 5), peripheral motor nerve excitability testing (PMNET; n = 5), short-interval intracortical inhibition (SICI; n = 4), resting motor threshold (RMT; n = 3), surface electromyography (SEMG; n = 3), and fasciculation frequency detected by ultrasound (n = 2; , ) . A neurophysiological measure was the primary outcome measure in seven trials (22%), and at least one neurophysiological measure was included as a secondary outcome measure in 28 trials (88%). Fifteen trials (47%) employed a combination of neurophysiological parameters. Final follow-up assessments varied with a median of 4 follow-up assessments per study. The median follow-up time was 1 month with a range of 6 hours – 2 years. Of the completed trials, 1 out of 7 employing NEMG described an improvement in the outcome measure with the intervention. For CMAP and NI, the positivity rates were 2/8 and 2/6, respectively. Out of the transcranial magnetic stimulation measures, the use of RMT and SICI were associated with positive outcomes in 2/3 and 2/4 trials, respectively. Of the 7 studies using MUNE and/or MUNIX, only 1 reported a positive finding (a MUNIX study). With regards to MUNE, three trials used statistical MUNE, two used the multipoint incremental method, and one used spike-triggering MUNE. Positive effects were seen in 3 out of 4 completed trials reporting PMNET as an outcome measure. contains a summary of all trials. Study interventions Almost a third (28%) of the included trials tested an intervention that targeted a neuronal ion channel, mainly sodium-channel blockade, NMDA blockade or potassium-channel agonism . The other major groups of interventions targeted either the immune system (50%) or neuronal metabolic pathways (22%). Most interventions were oral treatments (52%), while a minority were administered either intravenously (6%), subcutaneously (3%), intrathecally (13%), or using a mixture of methods (26%). Only one trial had a non-pharmacological intervention in the form of transcranial direct current stimulation. The majority of trials (69%) took place during the last ten years. Under half of the completed trials (39%) were positive with respect to at least one neurophysiological outcome measure. A quarter of reported studies remain incomplete. Study quality The full risk of bias assessment table is presented in Supplementary Table 2. The study quality ranged from high risk of bias (n = 2) to a very low risk of bias (n = 2) . All included studies accurately assessed the outcome using the neurophysiological measurements detailed above. As defined above, adequate follow up length was determined to be at least 6 months and most studies did not have an adequate follow up, thus reducing the overall quality of the studies. All completed studies highlighted the attrition rate, apart from three studies , which did not describe the loss to follow up. Overall, the studies included in this systematic review had a low risk of bias with 15 of the 24 studies assessed for risk scoring 7 and above. . Discussion Neurophysiological measures have gained increased attention as biomarkers of the neurodegenerative process in ALS . As the natural history of each measure becomes better characterised over time, their implementation as outcome measures in interventional clinical trials has become increasingly robust. In the context of ALS, neurophysiological measures can be categorised into three major groups: (1) Those that pertain to dysfunctional neuronal excitability (e.g., SICI or peripheral excitability testing) ; (2) Those that relate directly to neuronal death (e.g., NI or MUNIX) ; and (3) Those that detect a compensatory protective adaptation as a result of neuronal loss (e.g., EMG detection of motor unit reinnervation) . By definition, alterations in neuronal excitability must precede neuronal death in individual neurons, while neuronal death must have already begun within the motor neuron pool to induce compensatory changes in surviving neurons. However, when considering the entire motor neuron pool during a cross-sectional evaluation, all three categories can be captured simultaneously. Certainly, the detection of changes in the living neuron, either in excitability or compensatory adaptations, should be the focus when aiming to retain maximal neuronal function at the earliest stages of disease. By the time significant neuronal death has occurred, it is likely to be too late to halt disease progression and prevent significant disability and death. We found that interventional clinical trials in ALS have utilised a broad variety of neurophysiological measures. We included US-FF in this review as an imaging modality that detects the mechanical sequela (muscular contraction) of an electrophysiological event (fasciculation potential). The use of other imaging modalities (e.g. MRI) that focus on structural abnormalities were beyond the scope of this review. Although electrical impedance myography has shown potential value as an outcome measure in ALS , this systematic review did not identify any interventional trials that have employed this technique. While standardisation of the included measures was evident for the 47% of trials conducted over multiple sites, we found limited evidence for standardisation between trials. This will undoubtedly have a detrimental effect on the robustness and objectivity of these outcome measures. Key aspects related to standardisation include the choice of muscle(s), sensor type, recording equipment, technical parameters for electrical stimulation (where required), software, analytical processes, and reporting methods . Ideally, these choices should maximise the cost-effectiveness and global availability of the technique in question, so that attempts by other research groups to replicate a set of results are readily achievable . The optimal sampling frequency and duration of trials should be decided based on recommendations from previous trials, considering the extra burden placed on patients for more intensive regimes . The relative ability of each outcome measure to detect disease progression in a quantitative and sensitive way should be appreciated, recognising key influences such as test–retest replicability and inter-rater variation . In this regard, MUNIX and MUNE are amongst the most robust techniques, having been shown to decline by 2.4–9.0% per month in natural history studies . Appropriate training and assessor validation tests should be clarified and made available by those with sufficient experience of the technique . Consensus guidelines detailing the preferred methodology, training requirements, analysis and reporting for each technique would improve the current variations across clinical trials. Such a repository would guide future study design, ensuring that the advantages (e.g., low cost or non-invasiveness) and limitations (e.g., floor effect, reduced patient tolerability) are adequately considered when selecting the best measure for a new trial. It is anticipated that this would eliminate the apparent ‘scatter-gun’ approach revealed in the results of this systematic review. It was clear from the studies in this systematic review that neurophysiological measures should not be used in isolation. Their combination with clinical, biofluid and imaging biomarkers is vital to ensure a multi-modal output. Direct comparisons between measures from these different categories were limited (9% of trials performed this type of analysis), but it helps to assess how representative the recruited cohort is compared to the wider ALS population (e.g., ALSFRS-R average decline can be compared with other large cohorts) . It was unsurprising to find that the most commonly reported phase amongst the trials was phase 2 (almost a third), whereby neurophysiological measures (principally as secondary/exploratory outcomes) contributed to decisions over which interventions might benefit from clinical efficacy testing in phase 3. This aligns with recommendations set out in the 2019 revised Airlie House consensus guidelines for the design and implementation of ALS clinical trials, which emphasises that phase 3 outcomes should instead be the remit of well-established survival and functional measures . It should be highlighted that outcome measures are not the only way neurophysiological techniques can be employed in clinical trials. They can also help to make the recruited cohort more homogenous, reducing the variability and improving the study power . This should be chosen based on the presumed mechanism of action of the intervention, ensuring target engagement with the neurophysiological measure used. In some trials, where this approach has been adopted, there was a very clear rationale for the choice of neurophysiological measure. The best example is for the K v 7 channel agonist, retigabine/ezogabine, which demanded an outcome measure that could detect changes in excitability, more specifically a reduction in pathological hyperexcitability . The proposed mechanism of action for some interventions is indirect, for example the COX inhibitor celecoxib, which reduces prostaglandin release, in turn leading to lower glutamate levels . Interestingly, for those interventions targeting the immune system, measures of motor neuronal loss (e.g., MUNE, NI) were preferentially chosen, as opposed to measures of neuronal excitability. Another important consideration when matching the best outcome measure with the intervention stems from the neuroanatomical distinction between upper and lower motor neurons. Some neurophysiological measures, such as those obtained by transcranial magnetic stimulation (e.g., SICI, RMT), predominantly detect upper motor neuron disease, while the remainder (e.g., MUNIX, NI) highlight lower motor neuron degeneration. Conclusion It is widely agreed that neurophysiological outcome measures can play a significant role in the assessment of novel drug effectiveness in ALS. In this review, we have highlighted 32 clinical trials that have put a wide variety of these tools to the test, involving interventions with disparate mechanisms of action. While NEMG demonstrated relatively poor utility in this context, newer techniques such as MUNIX and SICI showed the greatest promise. What is evident from this analysis is that greater standardisation between trials is required, especially as trials begin to include greater numbers of patients, spanning multiple international sites. This would be particularly important for emerging neurophysiological techniques to ensure there is a solid knowledge and experience base for the future. In selecting the most appropriate outcome measure(s), the relative strengths and pitfalls of each should be considered, whilst ensuring target engagement with the intervention’s mechanism of action. In this regard, we hope this analysis can serve as a guide for future trial design in this field. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Understanding hope at diagnosis: A study among Guatemalan parents of children with cancer	7a1ab07f-57ab-4ba2-9979-eab5b907a7d2	10166945	Internal Medicine[mh]	INTRODUCTION For parents of children with cancer, hope has been demonstrated to alleviate distress and support a positive outlook. Hope‐centered conversations help clinicians foster trust and build the clinician–family relationship. , However, the majority of hope research has been conducted in high‐income countries (HICs) and in the setting of poor prognosis. In these contexts, parents emphasize the importance of hope, describing it as a guiding force that gives them the strength to navigate the uncertainties of cancer treatment. , Additionally, parental hope fluctuates as their child's condition changes and parents' ability to focus on the positive is challenged by factors like uncertainty and fear. How parents of children with diverse prognoses balance hope and concern from the time of diagnosis remains poorly understood, particularly in low‐ and middle‐income countries (LMICs), where 90% of children with cancer live. Hope theory suggests a goal‐pathway‐agency model. According to this model, to have hope an individual must be able to identify what they want to happen (a goal), believe a path to their desired outcome exists (a pathway), and feel capable of following that pathway (agency). , Hopelessness arises when an individual feels unable to change their circumstances in lieu of their goals. A recent extension of hope theory suggests that hope has a culture‐dependent “locus” dimension. In individualist cultures, hope relies on cognitions about one's own ability to influence outcomes (an internal locus), whereas in collectivist cultures people anchor hope externally, deriving hope from beliefs about how family, friends, and supernatural beings can influence outcomes (an external locus). These findings highlight the need to extend research on hope in pediatric oncology beyond HICs to settings with different customs and beliefs. The purpose of this analysis was to explore hopes and concerns of parents of children with cancer, including a range of prognoses, during diagnostic conversations in Guatemala, a collectivist, upper‐middle income country. METHODS This is a secondary analysis of a study for which complete methodology has been previously reported. Details pertaining to this analysis are described below. 2.1 Settings and participants This study was conducted at Unidad Nacional de Oncología Pediátrica (UNOP) in Guatemala, which is funded through a public‐private partnership and is Guatemala's only national pediatric cancer center. Guatemala is a small diverse country with 24 principal ethnic groups and a collectivist culture in which identity is centered in the family or community group rather than in individual traits. UNOP uses a multidisciplinary approach known as “medicina integral” which provides psychosocial services and resources to help cover expenses for families with children in treatment. The childhood cancer survival rate in Guatemala is ~65%. Spanish‐speaking parents of children under age 18 with a new or probable cancer diagnosis were eligible to participate in this study. Families were approached sequentially with additional purposive sampling to ensure a range of ages, diagnoses, and socioeconomic statuses. Data was collected between April and August of 2019; 32 families were approached, 20 chose to participate. Most families who declined participation cited hesitancy about being audio‐recorded; one family was unable to participate due to language. The children of families who declined participation had similar diagnoses, ages, and genders to those of participating families. Participants gave written informed consent prior to participation. Study design complied with international regulations for the protection of human subjects and was approved by the UNOP ethics committee and St. Jude IRB. 2.2 Study design and data collection At UNOP, parents participate in a two‐part diagnostic process. Initially, parents meet with a psychologist who conducts a psychosocial intake and provides a general explanation of cancer. Within approximately a week, an oncologist provides the official diagnosis and treatment plan with the psychologist present for support. In this study, both conversations were audio‐recorded, and a recorded semi‐structured interview of one parent was conducted within 14 days of diagnosis. In interviews, parents expanded on experiences with the diagnostic process and reflected on hopes and concerns. All study materials including the protocol and interview guides were translated and reviewed by bilingual team members. The interview guide was iteratively revised in Spanish and back translated into English. Audio‐recordings of conversations and interviews were professionally transcribed and translated into English; bilingual research team members checked translated texts against Spanish recordings to verify the transcripts accurately reflected the content of original audio recordings. 2.3 Data analysis Analysis was conducted using a priori and novel codes. Two authors coded each transcript independently and disagreements were resolved by consensus with a third‐party adjudicator. All data was coded for whether a parent, psychologist, or oncologist was speaking. This secondary analysis focused on data coded under a priori code “Expression of concern” defined as, “Utterances in which the caregiver expresses worry, anxiety, fear, anger, frustration, and other forms of negative affect or emotion; includes references to things being hard or difficult” and novel code “Expression of hope”, defined as “Utterances in which the caregiver expresses hope, desire for the future, or other form of positive emotion”. Additional novel codes informed the analysis including, “Uncertainty”, “God/Faith/Fatalism”, “Family Factors”, and “Child's Best Interest”. Thematic content analysis using constant comparative methodology , explored parental perceptions of hope and its perceived impacts during the diagnostic process. MAXQDA software facilitated data management. The Consolidated Criteria for Reporting Qualitative Studies guidelines were used to ensure rigor. Settings and participants This study was conducted at Unidad Nacional de Oncología Pediátrica (UNOP) in Guatemala, which is funded through a public‐private partnership and is Guatemala's only national pediatric cancer center. Guatemala is a small diverse country with 24 principal ethnic groups and a collectivist culture in which identity is centered in the family or community group rather than in individual traits. UNOP uses a multidisciplinary approach known as “medicina integral” which provides psychosocial services and resources to help cover expenses for families with children in treatment. The childhood cancer survival rate in Guatemala is ~65%. Spanish‐speaking parents of children under age 18 with a new or probable cancer diagnosis were eligible to participate in this study. Families were approached sequentially with additional purposive sampling to ensure a range of ages, diagnoses, and socioeconomic statuses. Data was collected between April and August of 2019; 32 families were approached, 20 chose to participate. Most families who declined participation cited hesitancy about being audio‐recorded; one family was unable to participate due to language. The children of families who declined participation had similar diagnoses, ages, and genders to those of participating families. Participants gave written informed consent prior to participation. Study design complied with international regulations for the protection of human subjects and was approved by the UNOP ethics committee and St. Jude IRB. Study design and data collection At UNOP, parents participate in a two‐part diagnostic process. Initially, parents meet with a psychologist who conducts a psychosocial intake and provides a general explanation of cancer. Within approximately a week, an oncologist provides the official diagnosis and treatment plan with the psychologist present for support. In this study, both conversations were audio‐recorded, and a recorded semi‐structured interview of one parent was conducted within 14 days of diagnosis. In interviews, parents expanded on experiences with the diagnostic process and reflected on hopes and concerns. All study materials including the protocol and interview guides were translated and reviewed by bilingual team members. The interview guide was iteratively revised in Spanish and back translated into English. Audio‐recordings of conversations and interviews were professionally transcribed and translated into English; bilingual research team members checked translated texts against Spanish recordings to verify the transcripts accurately reflected the content of original audio recordings. Data analysis Analysis was conducted using a priori and novel codes. Two authors coded each transcript independently and disagreements were resolved by consensus with a third‐party adjudicator. All data was coded for whether a parent, psychologist, or oncologist was speaking. This secondary analysis focused on data coded under a priori code “Expression of concern” defined as, “Utterances in which the caregiver expresses worry, anxiety, fear, anger, frustration, and other forms of negative affect or emotion; includes references to things being hard or difficult” and novel code “Expression of hope”, defined as “Utterances in which the caregiver expresses hope, desire for the future, or other form of positive emotion”. Additional novel codes informed the analysis including, “Uncertainty”, “God/Faith/Fatalism”, “Family Factors”, and “Child's Best Interest”. Thematic content analysis using constant comparative methodology , explored parental perceptions of hope and its perceived impacts during the diagnostic process. MAXQDA software facilitated data management. The Consolidated Criteria for Reporting Qualitative Studies guidelines were used to ensure rigor. RESULTS 3.1 Demographics Most families were Christian; 65% identified as Evangelical and 20% identified as Catholic. Although 80% of patients had a hematological malignancy, a range of other diagnoses were represented. Participating clinicians included three psychologists and seven pediatric hematologist/oncologists. Complete family and clinician demographic information is included in Table . 3.2 Hopes and concerns across the cancer continuum At diagnosis, parents expressed a breadth of hopes and concerns that pertained to the entire cancer continuum. Parents reported a “ pre‐diagnosis ” process prior to arrival at UNOP, citing expensive tests, unsuccessful treatments, and worsening symptoms. During this time, lack of knowledge about why their child was sick or what to do created helplessness and fear. One parent said, “Because we didn't know what to do with her back there… right now I'm with the anxiety to know what's wrong with her, the nurse said we don't know, and I asked why, why” (mother of a child with a hematological malignancy, psychosocial intake). Concerns dominated parents' reflections on the pre‐diagnosis period, although some expressed hope that UNOP would be able to treat their child. During the transition to UNOP, hopes and concerns about “ diagnosis ” arose. Parents worried about the cause of cancer, prognosis, and extent of disease. One parent explained, “My family and I were really worried because we thought cancer didn't have a cure” (father of a teenager with a hematological malignancy, interview). Some parents expressed hopes for a favorable diagnosis, good prognosis, and treatment options: “If the diagnosis is positive, I trust the doctors here, that they will give him the right treatment” (mother of a child with a hematological malignancy, psychosocial intake). Once parents met with clinicians at UNOP and heard a cure might be possible, they articulated hopes and concerns related to “ treatment .” Guatemalan parents worried about disease progression, interventions, and acute toxicities. A mother relayed her concern, “he is not eating enough, because the doctor said that if he doesn't eat, he can be malnourished, and he can even require a tube” (mother of a child with a hematological malignancy, interview). Parents hoped for positive disease response and minimal treatment side effects. Many participants hoped to return home. One parent shared her hope “that my girl can leave before the two years of treatment,” (mother of a child with a hematological malignancy, interview). Parents also considered life “ after completion of therapy ,” seeking normality for their child, including graduation, marriage, and long‐term health, from the time of diagnosis. One Guatemalan parent hoped that her child, “… could grow, play, and have a happy life” (mother of a child with a hematological malignancy, interview). Thoughts about life after treatment were predominantly hope‐focused; however, participants also expressed concerns about bullying, long‐term side effects, and relapse. Table provides additional quotes describing hopes and concerns related to each phase of the cancer continuum. 3.3 Supporting hope The clinical team at UNOP facilitated hope in four discrete ways: creating a supportive environment , providing information , affirming religious beliefs , and empowering parents . Staff at UNOP created a supportive environment that encouraged hopes and mitigated concerns. Parents' concerns were amplified by separation from support systems. UNOP staff reassured parents that they were not alone. One participant voiced the importance of this explaining, “Sometimes the support we need is not monetary but morally, they encourage us to keep fighting” (mother of a teenager with a hematological malignancy, interview). Clinicians established shared priorities to build a supportive environment, “We always ask this because whatever is important to you, it's important to us” (psychologist to parents of a child with a hematological malignancy). Parents found these interactions helpful for maintaining a positive outlook. One explained, “The staff, the workers, maintenance, the nurses, the doctors sometimes come inside and chat a little and say don't worry you will be home soon. It's a hope” (father of a child with a hematological malignancy, interview). Clinicians also provided information which supported hope. During diagnostic conversations, clinicians focused on cure and addressed fatalistic beliefs. One oncologist explained, “there is a possibility and that's what we want to offer you today, treatment” (oncologist to parents of a teenager with a hematological malignancy). Parents acknowledged the impact of information. One parent said, “Now that you talked to us, we have hope and now we understand more about the illness,” (father of a child with a hematological malignancy, diagnostic conversation). Clinicians affirmed religious beliefs during diagnostic conversations by establishing shared values, encouraging coping strategies like prayer, and highlighting God's role. One psychologist explained, “Just like you, we are people of faith here at the hospital” (psychologist to a father of a teenager with a hematological malignancy). Clinicians emphasized the possibility of a cure using religious language, “Sometimes we ask God for a miracle, and he uses humans to do it. And he left knowledge to doctors” (psychologist to parents of a child with a hematological malignancy) to reinforce faith in treatment and establish a connection between the clinicians and God. Finally, the UNOP team empowered parents . Participants expressed distress related to feeling unable to help their child; one told a psychologist, “I'd like to do something, but I cannot do anything” (father of a child with a hematological malignancy, psychosocial intake). In response, clinicians emphasized needs parents could fulfill: “give her love, letting her know she's not alone, that's what you can control” (psychologist to a father of a teenager with a CNS tumor). Clinicians also articulated ways parents could facilitate treatment. They encouraged parents to find blood donors and emphasized the importance of bringing the child to appointments. One psychologist explained, “Bring him to his appointments and as [the] hospital, we will give him his treatment, and God will do his part as well” (psychologist to a father of a child with a hematological malignancy). As clinicians created a supportive environment, provided information, affirmed religious beliefs, and empowered parents, they strengthened trust between the healthcare team and families, establishing a pathway for parents to embrace hope. One parent described this impact saying, “Well, yes, my vision changed because it was told to me by a doctor, a person with experience, he gave me more hope” (father of a teenager with a hematological malignancy, interview). In addition to the environment of trust and support cultivated by clinicians, some parents described how their religious beliefs and community or family support facilitated a growing sense of hope: “Even though it was hard for me to accept it, my husband told me to have faith that we will be in those 7 kids. There is the hope” (mother of a child with a hematological malignancy, interview). Table includes quotes from parents that further highlight the impact of these four clinical actions on hope. Clinicians at UNOP used these strategies to help parents shift from uncertainty and concern to possibility, control, and ultimately hope for life beyond cancer; Figure depicts this transition. 3.4 Hope and parental adjustment Parents referenced how their hopes were supported by clinical teams as they discussed acceptance, strength, happiness, and peace after diagnosis. Table highlights parental perceptions of emotional benefits. Initially, many Guatemalan parents grappled with their child's diagnosis wondering what caused their child to become sick. One participant questioned, “I've given her good food, I've taken good care of her, then why is this happening?” (mother of a child with a hematological malignancy, psychosocial intake). As parents discussed finding acceptance, they often mentioned hope, “But like she said, nothing is impossible. With treatment they try to cure the children. Whatever happens we must accept the situation” (mother of a child with a hematological malignancy, interview). Throughout diagnostic conversations and interviews, parents discussed how support, religion, and hope contributed to new‐found strength. One mother explained, “When you come here, you arrive petrified but here you recover your strength. When you started to talk to her and she started to analyze, she told me mom this place is nice and they are very kind. I said yes, they are lovely and you will be fine” (mother of a teenager with a hematological malignancy, interview). Another mother reflected on the strength she derived from religion, “In the middle of all this I try to give up and say God I know you know the things happening…I know he'll be okay; I want to think that way” (mother of a child with a hematological malignancy, diagnostic conversation). Many parents described happiness. One father said, “The important thing is…that you are explaining that there's treatment, a cure. That makes me feel a little bit happy because there's still hope” (father of a child with a hematological malignancy, diagnostic conversation). Participants also associated peace with hope. One commented, “yes, we have a lot of concerns, but we close our eyes and pray, and we feel relieved. I teach him to do it. There is always hope” (mother of a child with a hematological malignancy, interview). Finally, parents explicitly discussed how UNOPs supportive environment, knowledge, and empowerment helped alleviate concerns and allowed them to care for their own needs. One parent described time before UNOP saying, “…during those 4 days I couldn't sleep any because I was too scared and uncomfortable,” later commenting, “I feel I can rest here, although my distress is always there, I can rest here,” (mother of a child with a solid tumor, interview). Another participant responded to a psychologist's comment about hope explaining, “We were feeling so weak, and yesterday we were able to go, have lunch, and we ate in peace” (mother of a child with a hematological malignancy, diagnostic conversation), highlighting the impact of hope on self‐care. Demographics Most families were Christian; 65% identified as Evangelical and 20% identified as Catholic. Although 80% of patients had a hematological malignancy, a range of other diagnoses were represented. Participating clinicians included three psychologists and seven pediatric hematologist/oncologists. Complete family and clinician demographic information is included in Table . Hopes and concerns across the cancer continuum At diagnosis, parents expressed a breadth of hopes and concerns that pertained to the entire cancer continuum. Parents reported a “ pre‐diagnosis ” process prior to arrival at UNOP, citing expensive tests, unsuccessful treatments, and worsening symptoms. During this time, lack of knowledge about why their child was sick or what to do created helplessness and fear. One parent said, “Because we didn't know what to do with her back there… right now I'm with the anxiety to know what's wrong with her, the nurse said we don't know, and I asked why, why” (mother of a child with a hematological malignancy, psychosocial intake). Concerns dominated parents' reflections on the pre‐diagnosis period, although some expressed hope that UNOP would be able to treat their child. During the transition to UNOP, hopes and concerns about “ diagnosis ” arose. Parents worried about the cause of cancer, prognosis, and extent of disease. One parent explained, “My family and I were really worried because we thought cancer didn't have a cure” (father of a teenager with a hematological malignancy, interview). Some parents expressed hopes for a favorable diagnosis, good prognosis, and treatment options: “If the diagnosis is positive, I trust the doctors here, that they will give him the right treatment” (mother of a child with a hematological malignancy, psychosocial intake). Once parents met with clinicians at UNOP and heard a cure might be possible, they articulated hopes and concerns related to “ treatment .” Guatemalan parents worried about disease progression, interventions, and acute toxicities. A mother relayed her concern, “he is not eating enough, because the doctor said that if he doesn't eat, he can be malnourished, and he can even require a tube” (mother of a child with a hematological malignancy, interview). Parents hoped for positive disease response and minimal treatment side effects. Many participants hoped to return home. One parent shared her hope “that my girl can leave before the two years of treatment,” (mother of a child with a hematological malignancy, interview). Parents also considered life “ after completion of therapy ,” seeking normality for their child, including graduation, marriage, and long‐term health, from the time of diagnosis. One Guatemalan parent hoped that her child, “… could grow, play, and have a happy life” (mother of a child with a hematological malignancy, interview). Thoughts about life after treatment were predominantly hope‐focused; however, participants also expressed concerns about bullying, long‐term side effects, and relapse. Table provides additional quotes describing hopes and concerns related to each phase of the cancer continuum. Supporting hope The clinical team at UNOP facilitated hope in four discrete ways: creating a supportive environment , providing information , affirming religious beliefs , and empowering parents . Staff at UNOP created a supportive environment that encouraged hopes and mitigated concerns. Parents' concerns were amplified by separation from support systems. UNOP staff reassured parents that they were not alone. One participant voiced the importance of this explaining, “Sometimes the support we need is not monetary but morally, they encourage us to keep fighting” (mother of a teenager with a hematological malignancy, interview). Clinicians established shared priorities to build a supportive environment, “We always ask this because whatever is important to you, it's important to us” (psychologist to parents of a child with a hematological malignancy). Parents found these interactions helpful for maintaining a positive outlook. One explained, “The staff, the workers, maintenance, the nurses, the doctors sometimes come inside and chat a little and say don't worry you will be home soon. It's a hope” (father of a child with a hematological malignancy, interview). Clinicians also provided information which supported hope. During diagnostic conversations, clinicians focused on cure and addressed fatalistic beliefs. One oncologist explained, “there is a possibility and that's what we want to offer you today, treatment” (oncologist to parents of a teenager with a hematological malignancy). Parents acknowledged the impact of information. One parent said, “Now that you talked to us, we have hope and now we understand more about the illness,” (father of a child with a hematological malignancy, diagnostic conversation). Clinicians affirmed religious beliefs during diagnostic conversations by establishing shared values, encouraging coping strategies like prayer, and highlighting God's role. One psychologist explained, “Just like you, we are people of faith here at the hospital” (psychologist to a father of a teenager with a hematological malignancy). Clinicians emphasized the possibility of a cure using religious language, “Sometimes we ask God for a miracle, and he uses humans to do it. And he left knowledge to doctors” (psychologist to parents of a child with a hematological malignancy) to reinforce faith in treatment and establish a connection between the clinicians and God. Finally, the UNOP team empowered parents . Participants expressed distress related to feeling unable to help their child; one told a psychologist, “I'd like to do something, but I cannot do anything” (father of a child with a hematological malignancy, psychosocial intake). In response, clinicians emphasized needs parents could fulfill: “give her love, letting her know she's not alone, that's what you can control” (psychologist to a father of a teenager with a CNS tumor). Clinicians also articulated ways parents could facilitate treatment. They encouraged parents to find blood donors and emphasized the importance of bringing the child to appointments. One psychologist explained, “Bring him to his appointments and as [the] hospital, we will give him his treatment, and God will do his part as well” (psychologist to a father of a child with a hematological malignancy). As clinicians created a supportive environment, provided information, affirmed religious beliefs, and empowered parents, they strengthened trust between the healthcare team and families, establishing a pathway for parents to embrace hope. One parent described this impact saying, “Well, yes, my vision changed because it was told to me by a doctor, a person with experience, he gave me more hope” (father of a teenager with a hematological malignancy, interview). In addition to the environment of trust and support cultivated by clinicians, some parents described how their religious beliefs and community or family support facilitated a growing sense of hope: “Even though it was hard for me to accept it, my husband told me to have faith that we will be in those 7 kids. There is the hope” (mother of a child with a hematological malignancy, interview). Table includes quotes from parents that further highlight the impact of these four clinical actions on hope. Clinicians at UNOP used these strategies to help parents shift from uncertainty and concern to possibility, control, and ultimately hope for life beyond cancer; Figure depicts this transition. Hope and parental adjustment Parents referenced how their hopes were supported by clinical teams as they discussed acceptance, strength, happiness, and peace after diagnosis. Table highlights parental perceptions of emotional benefits. Initially, many Guatemalan parents grappled with their child's diagnosis wondering what caused their child to become sick. One participant questioned, “I've given her good food, I've taken good care of her, then why is this happening?” (mother of a child with a hematological malignancy, psychosocial intake). As parents discussed finding acceptance, they often mentioned hope, “But like she said, nothing is impossible. With treatment they try to cure the children. Whatever happens we must accept the situation” (mother of a child with a hematological malignancy, interview). Throughout diagnostic conversations and interviews, parents discussed how support, religion, and hope contributed to new‐found strength. One mother explained, “When you come here, you arrive petrified but here you recover your strength. When you started to talk to her and she started to analyze, she told me mom this place is nice and they are very kind. I said yes, they are lovely and you will be fine” (mother of a teenager with a hematological malignancy, interview). Another mother reflected on the strength she derived from religion, “In the middle of all this I try to give up and say God I know you know the things happening…I know he'll be okay; I want to think that way” (mother of a child with a hematological malignancy, diagnostic conversation). Many parents described happiness. One father said, “The important thing is…that you are explaining that there's treatment, a cure. That makes me feel a little bit happy because there's still hope” (father of a child with a hematological malignancy, diagnostic conversation). Participants also associated peace with hope. One commented, “yes, we have a lot of concerns, but we close our eyes and pray, and we feel relieved. I teach him to do it. There is always hope” (mother of a child with a hematological malignancy, interview). Finally, parents explicitly discussed how UNOPs supportive environment, knowledge, and empowerment helped alleviate concerns and allowed them to care for their own needs. One parent described time before UNOP saying, “…during those 4 days I couldn't sleep any because I was too scared and uncomfortable,” later commenting, “I feel I can rest here, although my distress is always there, I can rest here,” (mother of a child with a solid tumor, interview). Another participant responded to a psychologist's comment about hope explaining, “We were feeling so weak, and yesterday we were able to go, have lunch, and we ate in peace” (mother of a child with a hematological malignancy, diagnostic conversation), highlighting the impact of hope on self‐care. DISCUSSION From the time of diagnosis, Guatemalan parents expressed hopes and concerns pertaining to the entire cancer continuum and described how support from the UNOP clinical team helped them shift their focus from concern toward hope. Ultimately, parents associated hope with acceptance, strength, happiness, peace, and coping. These findings are aligned with studies examining hope among parents of pediatric oncology patients in HICs. , Our results further establish the relevance of hope in LMICs at the time of diagnosis for all populations, including those treated with curative intent. Supporting hope has been previously defined as one of eight communication functions essential to pediatric oncology. While this function was initially established through work with populations within the United States, our results confirm its significance within Guatemala. Furthermore, our results identify four discrete actions that parents cited during discussions of hope, creating a supportive environment , providing information , affirming religious beliefs , and empowering parents , which clinicians can use to operationalize this function regardless of the cultural context. Table highlights cultural factors including literacy, religious beliefs, and cultural norms that may impact the role of the clinician in supporting hope and suggests ways to adapt clinical actions accordingly. The experiences of Guatemalan parents reflect the goal‐pathway‐agency model of hope theory. During the pre‐diagnosis period, participants' concerns about the unknown reflected a perceived lack of pathways and agency. Throughout the diagnostic process, clinicians provided information to help parents realize a pathway to survival and empowered them to recover agency, which facilitated feelings of hope. Furthermore, our findings reflect the conclusion that the locus of hope is dependent on cultural setting. Parents in Guatemala frequently mentioned religion and cited the power of supernatural beings as integral to hope, suggesting an external locus. By affirming religious beliefs, Guatemalan clinicians reinforced parents' external anchors of hope and utilized them to build trust, strengthen faith in the medical team, and emphasize possibility. The supportive environment at UNOP allows parents to expand their anchors of hope to include the medical team, thereby supporting the process by which parents find hope. Finally, our results demonstrate how supporting hope may influence outcomes. This study's findings suggest that hope may have a role in facilitating parental acceptance and coping following diagnosis. As clinicians supported hope, they empowered parents, emphasized dignity, and demonstrated respect, interactions which may be associated with longer‐term outcomes including increased quality of care and lower rates of treatment abandonment. Perhaps the most telling evidence of the importance of culturally sensitive hope communication on outcomes is the resounding success of UNOPs implementation of integrated psychosocial services through medicina integral . This multidisciplinary approach included tailored services and adapted clinical conversations that reflected the cultural needs of the Guatemalan people and facilitated clinicians' ability to support hope and build a therapeutic alliance. Implementation of the medicina integral team contributed to treatment abandonment rates falling from 27% to 7%. This study had several limitations. Understanding hope was not the primary objective of the original study and thus was not central to the design of the interview guide. Data was exclusively collected during the diagnostic period; future work is needed to assess how hope evolves throughout treatment. In addition, this analysis focused on hope at UNOP, a single public‐private hospital in Guatemala, a predominantly Christian, upper‐middle income country. Further work should explore parental hope in other contexts to understand the applicability of identified themes in regions with different cultures and resources. While bilingual study team members checked translations, English‐Spanish translation during study tool development, data collection, and analysis may have impacted results. The use of Spanish during data collection may have limited the ability of parents whose primary language was a Mayan dialect to fully express themselves. Additionally, future work is needed to explore pediatric patients' perspectives on hope as this study included only parent participants. CONCLUSION Our results highlight the relevance of parental hope to pediatric oncology regardless of cultural setting. While the importance of hope appears to be universal, culture affects how it manifests, impacting parents' hope‐related needs and influencing the role of the clinician in supporting hope. These results indicate that clinicians, regardless of resource setting, should tailor communication based on families' cultural contexts. While previous studies identified processes that underly the ways clinicians support hope, this study delineated four discrete actions that facilitate these processes providing insight into how hope can be integrated into clinical communication. Further work should explore similar themes in other contexts to expand on the role of culture in hope and focus on interventions that promote culturally adaptive hope communication. Anneliese H. Williams: Formal analysis (lead); visualization (equal); writing – original draft (lead); writing – review and editing (equal). Silvia Rivas: Methodology (supporting); project administration (equal); resources (equal); writing – review and editing (equal). Lucia Fuentes: Investigation (lead); methodology (supporting); resources (equal); writing – review and editing (equal). Ana Caceres‐Serrano: Investigation (lead); methodology (supporting); resources (equal); writing – review and editing (equal). Gia Ferrara: Formal analysis (supporting); writing – review and editing (equal). Tegan Reeves: Formal analysis (supporting); writing – review and editing (equal). Federico Antillon‐Klussmann: Methodology (supporting); project administration (equal); resources (equal); writing – review and editing (equal). Carlos Rodriguez‐Galindo: Methodology (supporting); writing – review and editing (equal). Jennifer W. Mack: Methodology (supporting); writing – review and editing (equal). Dylan E. Graetz: Conceptualization (lead); formal analysis (supporting); investigation (supporting); methodology (lead); project administration (equal); supervision (equal); visualization (equal); writing – review and editing (equal). This work was funded by American Lebanese Syrian Associated Charities of St. Jude Children's Research Hospital and a Conquer Cancer Young Investigator Award (award number 17290). Any opinions, findings, and conclusions expressed in this material are those of the authors and do not necessarily reflect those of the American Society of Clinical Oncology or Conquer Cancer. The authors have no conflicts of interest relevant to this article to disclose.
Optimizing induction chemotherapy regimens for radiotherapy in patients with locoregionally advanced nasopharyngeal carcinoma	06d3f753-143e-4063-9306-16e761fe612c	10166969	Internal Medicine[mh]	INTRODUCTION Nasopharyngeal carcinoma (NPC), known to be endemic in Southern China and Southeast Asia, is prone to recurrence and metastasis. The anatomic location of NPC does not lend surgery as a viable management option. Taking advantage of the high sensitivity of nasopharyngeal carcinoma cells to radiation, radiotherapy (RT) has become an effective and dominant definitive therapy for NPC, with over 90% of early stage NPC experiencing overall survival (OS) at 5 years. However, for LANPC, the 5‐year OS was only between 41% and 63%. , In Mainland China, about 75% of newly diagnosed NPC are LANPC. Therefore, there remains a compelling need to improve the outcomes of LANPC. With the development of radiotherapy technology, the local control of LANPC has been greatly enhanced, with distant metastasis has become the major mode of treatment failure for NPC. According to meta‐analyses of randomized trials, the combination of RT and chemotherapy reduces mortality by 18% and increases the 5‐year OS by 4%–6%. Among regimens, concurrent chemoradiotherapy (CCRT) has shown OS benefits and has become the standard of care for LANPC. , Regarding induction chemotherapy (IC), a meta‐analysis revealed that the addition of IC to CCRT reduces distant failure in LANPC compared with CCRT alone. Additionally, some studies also showed that the benefit of IC did not translate into an increase in OS. , , However, a publication with the largest sample size and the longest follow‐up period suggested that IC is a better treatment model for NPC patients with clinically staged T 1‐2 /N 2‐3 and T 3‐4 /N 1‐3 diseases, especially in combination with CCRT. Because of the above clinical evidence, IC has been widely used in the treatment of patients with LANPC. However, the optimal number of cycles of IC has not been studied and hence remains controversial. In previous studies from different institutions, , , , , , , LANPC patients received 2 or 3, or even 4 cycles of IC. Despite the benefits, IC also has its own potential drawbacks, such as causing a delay in radiotherapy, leading to accelerated reproliferation of tumor cells, and yielding reduced radiotherapy tolerance due to adverse reactions of chemotherapy. The delayed start of radiotherapy may reduce the survival rate. , , Therefore, optimal numbers of IC cycles for different stages of NPC and the best interventional time for radiotherapy are important problems to investigate for prolonging the survival of patients with LANPC. In this work, we analyzed the optimal number of IC cycles from the time pending for radical radiotherapy. All patients with LANPC underwent multiple computed tomography (CT) scans, obtained the imaging data before and during the IC period, we analyzed the pattern of tumor reduction, dynamically monitored the effects of IC, and objectively evaluated the optimal number of IC cycles. This study provides a basis for the optimization of radiotherapy and chemotherapy combined therapy for LANPC. MATERIALS AND METHODS 2.1 Patient characteristics From December 2017 to September 2018, 54 patients with pathologically confirmed nonkeratinizing NPC‐undifferentiated type were retrospectively analyzed for this study. The inclusion criteria were the completion of a pretreatment evaluation including physical examination, chest CT, contrast‐enhanced nasopharyngeal/neck magnetic resonance imaging (MRI), abdominal ultrasonography, and a bone scan; T 3‐4 and/or N 2‐3 , M 0 stage; III/IV stage based on 7th edition UICC (International Union against Cancer/American Joint Committee on Cancer); age between 18 and 70 years old; and Eastern Cooperative Oncology Group (ECOG) score of 0 to 1. 2.2 Induction chemotherapy All patients received CCRT, prior to which they all were given a total of three cycles of platinum‐based IC, once every 3 weeks. Available IC regimens included DP (docetaxel 75 mg/m 2 /d on day 1, cisplatin 25 mg/m 2 /d on days 1–3) or PF (5‐fluorouracil 800–1000 mg/m 2 , continuous intravenous infusion for 96 h, cisplatin 25 mg/m 2 /d on days 1–3). Three cycles of induction chemotherapy with PF (dose is the same as above) were administered in 18 patients, and the other 36 patients underwent DP. Concurrent chemotherapy consisted of cisplatin (30–45 mg/m 2 , intravenous infusion on day 1) given weekly during radiotherapy. 2.3 Imaging acquisition and registration Immobilization was achieved using individualized custom thermoplastic masks encompassing the head, neck, and shoulders in the supine position. Before IC, a CT scan was acquired on a Brilliance Big Bore CT simulator (Philips Inc., Cleveland, OH, USA) with intravenous contrast. Each patient then underwent CT scans on the same CT simulator 14 days after each cycle of chemotherapy following the same protocol except without intravenous contrast. All scans were acquired with a 0.5 × 0.5 mm axial pixel size and a 2.5 mm slice thickness from the vertex to 2 cm caudal to the sternal manubrium. In this article, the three post‐IC CT image sets are denoted as CT n ( n = 1, 2, 3) for the three cycles of IC, respectively, and the pre‐IC CT image set is denoted as CT 0 . All scanned images were transferred to the MIM system (MIM Software Inc.) for image registration and target delineation. 2.4 Definition of target volume The four CT image sets of the same patient together with their respective diagnostic CT and MRI images were automatically registered in the MIM software and then adjusted manually when necessary. In order to reduce the interobserver variability, the GTVs in each CT image set were contoured by a single trained head and neck radiation oncologist using the MIM software system. In case of doubt, the agreement of another senior radiation oncologist was obtained. Target volumes were delineated manually according to the NPC intensity‐modulated radiotherapy (IMRT) protocol of our institution. The GTV of the primary tumor (GTV_T n ), involved cervical lymph nodes (GTV_N n ), and retropharyngeal lymph node (GTV_RP n , n = 0, 1, 2, 3) was delineated. 2.5 Calculations of tumor response The study recorded and calculated volume changes during IC. The GTV volumes of the primary tumor and involved lymphatics were evaluated using the contours on the corresponding CT image sets and recorded. The percent change after varying cycles of IC was calculated as follows: (V CTn −V CT0 )/V CT0 × 100% ( n = 1,2,3), where V CTn represents the GTV volume on the CT image, and V CT0 is the target volume in the primary CT taken before the IC. The coordinates of the center of mass (COM) in three orthogonal directions of the GTV, left–right (LR), cranial‐caudal (CC), and anterior–posterior (AP) directions were automatically calculated. In each direction, three displacements could be obtained for each GTV, from the pre‐IC image to each of the post‐IC image. The three‐dimensional (3D) vector of displacement, which combines errors recorded along all three axes, was defined as the square root of the DLR 2 , DCC 2 , and DAP 2 mean errors (DLR, DCC, and DAP were the displacements in the LR, CC, and AP directions, respectively). 2.6 Statistical analysis The changes of each GTV volume during IC were analyzed using the Wilcoxon signed‐rank test, with p < 0.05 meaning statistically significant changes. The GTV volume reduction was also analyzed against bodyweight change during the IC period using linear regression analysis. The Statistical Package for the Social Sciences (SPSS version 26; IBM Corporation) and Microsoft Office Excel (Microsoft Corporation, Redmond) were used. Patient characteristics From December 2017 to September 2018, 54 patients with pathologically confirmed nonkeratinizing NPC‐undifferentiated type were retrospectively analyzed for this study. The inclusion criteria were the completion of a pretreatment evaluation including physical examination, chest CT, contrast‐enhanced nasopharyngeal/neck magnetic resonance imaging (MRI), abdominal ultrasonography, and a bone scan; T 3‐4 and/or N 2‐3 , M 0 stage; III/IV stage based on 7th edition UICC (International Union against Cancer/American Joint Committee on Cancer); age between 18 and 70 years old; and Eastern Cooperative Oncology Group (ECOG) score of 0 to 1. Induction chemotherapy All patients received CCRT, prior to which they all were given a total of three cycles of platinum‐based IC, once every 3 weeks. Available IC regimens included DP (docetaxel 75 mg/m 2 /d on day 1, cisplatin 25 mg/m 2 /d on days 1–3) or PF (5‐fluorouracil 800–1000 mg/m 2 , continuous intravenous infusion for 96 h, cisplatin 25 mg/m 2 /d on days 1–3). Three cycles of induction chemotherapy with PF (dose is the same as above) were administered in 18 patients, and the other 36 patients underwent DP. Concurrent chemotherapy consisted of cisplatin (30–45 mg/m 2 , intravenous infusion on day 1) given weekly during radiotherapy. Imaging acquisition and registration Immobilization was achieved using individualized custom thermoplastic masks encompassing the head, neck, and shoulders in the supine position. Before IC, a CT scan was acquired on a Brilliance Big Bore CT simulator (Philips Inc., Cleveland, OH, USA) with intravenous contrast. Each patient then underwent CT scans on the same CT simulator 14 days after each cycle of chemotherapy following the same protocol except without intravenous contrast. All scans were acquired with a 0.5 × 0.5 mm axial pixel size and a 2.5 mm slice thickness from the vertex to 2 cm caudal to the sternal manubrium. In this article, the three post‐IC CT image sets are denoted as CT n ( n = 1, 2, 3) for the three cycles of IC, respectively, and the pre‐IC CT image set is denoted as CT 0 . All scanned images were transferred to the MIM system (MIM Software Inc.) for image registration and target delineation. Definition of target volume The four CT image sets of the same patient together with their respective diagnostic CT and MRI images were automatically registered in the MIM software and then adjusted manually when necessary. In order to reduce the interobserver variability, the GTVs in each CT image set were contoured by a single trained head and neck radiation oncologist using the MIM software system. In case of doubt, the agreement of another senior radiation oncologist was obtained. Target volumes were delineated manually according to the NPC intensity‐modulated radiotherapy (IMRT) protocol of our institution. The GTV of the primary tumor (GTV_T n ), involved cervical lymph nodes (GTV_N n ), and retropharyngeal lymph node (GTV_RP n , n = 0, 1, 2, 3) was delineated. Calculations of tumor response The study recorded and calculated volume changes during IC. The GTV volumes of the primary tumor and involved lymphatics were evaluated using the contours on the corresponding CT image sets and recorded. The percent change after varying cycles of IC was calculated as follows: (V CTn −V CT0 )/V CT0 × 100% ( n = 1,2,3), where V CTn represents the GTV volume on the CT image, and V CT0 is the target volume in the primary CT taken before the IC. The coordinates of the center of mass (COM) in three orthogonal directions of the GTV, left–right (LR), cranial‐caudal (CC), and anterior–posterior (AP) directions were automatically calculated. In each direction, three displacements could be obtained for each GTV, from the pre‐IC image to each of the post‐IC image. The three‐dimensional (3D) vector of displacement, which combines errors recorded along all three axes, was defined as the square root of the DLR 2 , DCC 2 , and DAP 2 mean errors (DLR, DCC, and DAP were the displacements in the LR, CC, and AP directions, respectively). Statistical analysis The changes of each GTV volume during IC were analyzed using the Wilcoxon signed‐rank test, with p < 0.05 meaning statistically significant changes. The GTV volume reduction was also analyzed against bodyweight change during the IC period using linear regression analysis. The Statistical Package for the Social Sciences (SPSS version 26; IBM Corporation) and Microsoft Office Excel (Microsoft Corporation, Redmond) were used. RESULTS 3.1 Clinical characteristics Fifty‐four LANPC patients were analyzed in this study, all managed from December 2017 to September 2018, with a median age of 47 years (range = 21–66). The detailed patient characteristics are shown in Table . 3.2 Volume reduction following IC The GTV volume reduction following IC was found to have distinct magnitudes for GTV_T, GTV_RP, and GTV_N, and across different patients, as detailed below. No significant differences were apparent in the mean volume reduction of the gross tumor or nodes between those who received 3 cycles of PF ( n = 18) versus DP ( n = 36). 3.2.1 During IC , the gross tumor volume of the primary lesion gradually shrinks (Figures and ) but tapers off after 2 cycles Compared with the initial volume on pre‐IC CT (GTV_T 0 ), the total volume reduction of GTV_T n ( n = 1, 2, 3) after each IC cycle was −3.1 ± 0.7 cc (12.0%), 7.0 ± 1.9 cc (22.5%), and 6.9 ± 2.0 cc (20.1%), respectively. GTV_T significantly decreased its volume following the first two IC cycles ( p < 0.05), but not after the third cycle (Table ). 3.2.2 During IC , the size of cervical lymph node metastases continues to shrink following each cycle, at reduction rates higher than those of primary nasopharyngeal lesions (Figures and ) Using the initial volume on pre‐IC CT (GTV_N 0 ) as a comparison, the total volume reduction of GTV_N n ( n = 1, 2, 3) after each IC cycle was 7.3 ± 1.7 cc (25.3%), 13.5 ± 2.9 cc (43.2%), and 17.2 ± 3.4 cc (54.7%), respectively (Table ). For GTV_N, the volume reduction was significant ( p < 0.05) following all three IC cycles, with a single‐cycle GTV_N relative volume reduction rate of 25.3%, 26.9%, and 24.4% respectively after each cycle of IC. 3.2.3 The volume change trend of retropharyngeal lymph nodes is similar to primary nasopharyngeal lesions (Figures and ) Compared with the initial volume on pre‐IC CT (GTV_RP 0 ), the volume changes of GTV_RP n ( n = 1, 2, 3) after IC was 1.4 ± 0.4 cc (26.0%), 2.5 ± 0.5 cc (44.1%), and 2.4 ± 0.5 cc (42.2%), respectively. Significant volume reduction occurred only in the first two cycles ( p < 0.05), and after the third cycle, the volume even slightly increased (Table ). 3.3 Displacement of COM The displacements of all GTVs are shown in Table . The average COM displacements of each of the GTV_T, GTV_N, and GTV_RP were all less than 1.5 mm in individual orthogonal directions, and their mean 3D displacements were 2.6, 4.0, and 1.7 mm, respectively (Table ). We also noticed that the largest magnitude of displacements of lymph node GTVs was in the CC direction, with most of them shifted toward the cranial direction. 3.4 Volume reduction and bodyweight change During IC, the body weight change of the patients ranged from −10.2% to 11.6%. The average body weight increased 3.1%, 3.1%, and 1.6% respectively after each cycle of chemotherapy. The volume reductions of the GTVs do not have a strong correlation with weight change, which makes it hard to predict GTV volume reduction from the bodyweight change. 3.5 Adverse reactions during IC The most common side effects during IC were myelosuppression, mainly leukopenia and neutropenia, and no neutropenia fever. Most of the patients experienced grade I‐II myelosuppression. Only one patient developed grade III myelosuppression after three cycles of IC which improved after being treated with recombinant human granulocyte colony‐stimulating factor. Another common reaction was gastrointestinal reaction, mainly nausea and vomiting, including 48 cases of grade I gastrointestinal reaction and 6 cases of grade II delayed vomiting. After strengthening symptomatic support treatment, all patients completed three cycles of IC. Table describes the toxicity details. 3.6 Impact on subsequent CCRT All patients in this study completed 3 cycles of induction chemotherapy prior to concurrent chemoradiotherapy. All patients received 5–7 cycles of weekly cisplatin for current chemotherapy during radiation treatment. The completion rate was 100%. Clinical characteristics Fifty‐four LANPC patients were analyzed in this study, all managed from December 2017 to September 2018, with a median age of 47 years (range = 21–66). The detailed patient characteristics are shown in Table . Volume reduction following IC The GTV volume reduction following IC was found to have distinct magnitudes for GTV_T, GTV_RP, and GTV_N, and across different patients, as detailed below. No significant differences were apparent in the mean volume reduction of the gross tumor or nodes between those who received 3 cycles of PF ( n = 18) versus DP ( n = 36). 3.2.1 During IC , the gross tumor volume of the primary lesion gradually shrinks (Figures and ) but tapers off after 2 cycles Compared with the initial volume on pre‐IC CT (GTV_T 0 ), the total volume reduction of GTV_T n ( n = 1, 2, 3) after each IC cycle was −3.1 ± 0.7 cc (12.0%), 7.0 ± 1.9 cc (22.5%), and 6.9 ± 2.0 cc (20.1%), respectively. GTV_T significantly decreased its volume following the first two IC cycles ( p < 0.05), but not after the third cycle (Table ). 3.2.2 During IC , the size of cervical lymph node metastases continues to shrink following each cycle, at reduction rates higher than those of primary nasopharyngeal lesions (Figures and ) Using the initial volume on pre‐IC CT (GTV_N 0 ) as a comparison, the total volume reduction of GTV_N n ( n = 1, 2, 3) after each IC cycle was 7.3 ± 1.7 cc (25.3%), 13.5 ± 2.9 cc (43.2%), and 17.2 ± 3.4 cc (54.7%), respectively (Table ). For GTV_N, the volume reduction was significant ( p < 0.05) following all three IC cycles, with a single‐cycle GTV_N relative volume reduction rate of 25.3%, 26.9%, and 24.4% respectively after each cycle of IC. 3.2.3 The volume change trend of retropharyngeal lymph nodes is similar to primary nasopharyngeal lesions (Figures and ) Compared with the initial volume on pre‐IC CT (GTV_RP 0 ), the volume changes of GTV_RP n ( n = 1, 2, 3) after IC was 1.4 ± 0.4 cc (26.0%), 2.5 ± 0.5 cc (44.1%), and 2.4 ± 0.5 cc (42.2%), respectively. Significant volume reduction occurred only in the first two cycles ( p < 0.05), and after the third cycle, the volume even slightly increased (Table ). During IC , the gross tumor volume of the primary lesion gradually shrinks (Figures and ) but tapers off after 2 cycles Compared with the initial volume on pre‐IC CT (GTV_T 0 ), the total volume reduction of GTV_T n ( n = 1, 2, 3) after each IC cycle was −3.1 ± 0.7 cc (12.0%), 7.0 ± 1.9 cc (22.5%), and 6.9 ± 2.0 cc (20.1%), respectively. GTV_T significantly decreased its volume following the first two IC cycles ( p < 0.05), but not after the third cycle (Table ). During IC , the size of cervical lymph node metastases continues to shrink following each cycle, at reduction rates higher than those of primary nasopharyngeal lesions (Figures and ) Using the initial volume on pre‐IC CT (GTV_N 0 ) as a comparison, the total volume reduction of GTV_N n ( n = 1, 2, 3) after each IC cycle was 7.3 ± 1.7 cc (25.3%), 13.5 ± 2.9 cc (43.2%), and 17.2 ± 3.4 cc (54.7%), respectively (Table ). For GTV_N, the volume reduction was significant ( p < 0.05) following all three IC cycles, with a single‐cycle GTV_N relative volume reduction rate of 25.3%, 26.9%, and 24.4% respectively after each cycle of IC. The volume change trend of retropharyngeal lymph nodes is similar to primary nasopharyngeal lesions (Figures and ) Compared with the initial volume on pre‐IC CT (GTV_RP 0 ), the volume changes of GTV_RP n ( n = 1, 2, 3) after IC was 1.4 ± 0.4 cc (26.0%), 2.5 ± 0.5 cc (44.1%), and 2.4 ± 0.5 cc (42.2%), respectively. Significant volume reduction occurred only in the first two cycles ( p < 0.05), and after the third cycle, the volume even slightly increased (Table ). Displacement of COM The displacements of all GTVs are shown in Table . The average COM displacements of each of the GTV_T, GTV_N, and GTV_RP were all less than 1.5 mm in individual orthogonal directions, and their mean 3D displacements were 2.6, 4.0, and 1.7 mm, respectively (Table ). We also noticed that the largest magnitude of displacements of lymph node GTVs was in the CC direction, with most of them shifted toward the cranial direction. Volume reduction and bodyweight change During IC, the body weight change of the patients ranged from −10.2% to 11.6%. The average body weight increased 3.1%, 3.1%, and 1.6% respectively after each cycle of chemotherapy. The volume reductions of the GTVs do not have a strong correlation with weight change, which makes it hard to predict GTV volume reduction from the bodyweight change. Adverse reactions during IC The most common side effects during IC were myelosuppression, mainly leukopenia and neutropenia, and no neutropenia fever. Most of the patients experienced grade I‐II myelosuppression. Only one patient developed grade III myelosuppression after three cycles of IC which improved after being treated with recombinant human granulocyte colony‐stimulating factor. Another common reaction was gastrointestinal reaction, mainly nausea and vomiting, including 48 cases of grade I gastrointestinal reaction and 6 cases of grade II delayed vomiting. After strengthening symptomatic support treatment, all patients completed three cycles of IC. Table describes the toxicity details. Impact on subsequent CCRT All patients in this study completed 3 cycles of induction chemotherapy prior to concurrent chemoradiotherapy. All patients received 5–7 cycles of weekly cisplatin for current chemotherapy during radiation treatment. The completion rate was 100%. DISCUSSION The patients enrolled in this study were clinically classified as T 3‐4 or N 2 . According to Du and Lan, these patients are all in the high‐risk group for local recurrence and distant metastasis and can benefit from IC. With the imaging data in this study, the optimal number of IC cycles was investigated and identified for the first time. Our study showed that the GTVs of LANPC changed relatively substantially throughout IC. The effective rate was 89% with only six cases failing to show a tumor reduction after IC. The results of our study suggest that GTV_T continues to shrink after the first two cycles of IC. However, further tumor reduction was not significant after the third cycle and some patients even showed slight volume increases, especially those patients with T4 stage. Our data analysis showed that 27 (50%) of the 54 cases experienced a GTV_T increase after the third cycle of IC. The volume change of retropharyngeal lymph nodes shows the same trend as GTV_T. After the third cycle of IC, the volume of retropharyngeal lymph nodes remained unchanged or even slightly increased. Volume reductions of the GTV during IC have been previously noted. Li et al. showed a relative volume reduction of 12.7% in GTV_T after IC in nonmetastatic NPC patients. Zheng et al. reported a mean volume regression of 23.1% in GTV_T and 33.1% in GTV_N among 44 NPC patients who received two cycles of IC. Lee et al. reported that in 20 cases of advanced nasopharyngeal carcinoma, 35% of patients had achieved significant downstaging. In their study, significant GTV_T volume reductions were observed, with the mean reduction being 61.4% (range = 36.7–83.6%), with 70% of patients having experienced a >50% reduction in GTV_T following IC. In addition, He et al. compared survival outcomes from patients treated with two, three, and four IC cycles and found similar survival between two and three cycles IC groups, while four cycles IC was associated with worse overall survival and higher incidence of treatment‐related toxicities. The findings of our study showed tumor reduction in magnitudes to a certain extent consistent with the results of the above clinical studies. However, in clinical practice, further shrinkage in the gross tumor and lymph nodes may not meaningfully correspond with outcomes; rather, it is likely that the degree of benefit is limited to a subset of patients. Therefore, judging from our results, we recommend that for patients with stage T 3‐4 or large retropharyngeal lymph nodes, two cycles of IC are optimal, and radiotherapy should be initiated after two IC cycles. The delay of radiation therapy may undermine the efficacy of IC, and further increases in the number of IC cycles may not benefit the patients. Our study also shows that the volume of GTV_N steadily decreased after all three IC cycles and exhibited a greater reduction in contrast to GTV_T and GTV_RP. The single‐cycle volume reduction of GTV_N after each IC cycle was statistically significant. After three cycles of induction, the volume of GTV_N decreased by an average of 55%. This is consistent with the results of several other publications. With NPC patients included, these head and neck patients showed average GTV_N volume reduction in the range of 33%–68%. , A study also reported that the number of IC appeared to be an independent predictor and associated with an improvement in survival. For N2‐3 NPC, survival data of the 4 cycles of IC were better than those of 2 cycles. In our study, all but one patient showed a GTV_N volume reduction after three cycles of IC. The exception patient even showed a GTV_N volume increase. The clinical stage of this patient was T 3 N 2 M 0 stage III. For this patient, the right cervical lymph node continued to enlarge after IC, and the GTV_T also increased after the second cycle of IC. The most likely explanation was the primary drug resistance in this patient. In a sense, the COM displacement was used to gauge the tumor response, since the degree of response impacts the location of the remaining disease. There are little data published about the position variations of LANPC of IC. In our study, the average COM displacement of all GTVs was less than 1.5 mm in all three orthogonal directions, and the average 3D vector displacements ranged from 1.7–4.0 mm. The GTV_T displacements were larger in the LR and AP directions relative to the CC direction. This could be because the primary tumor receded from the adjacent air cavity as well as a shift in the posterior portion of the disease abutting the relatively stable vertebrae and paravertebral muscles (as shown in Figure ). However, the displacements of the cervical lymph nodes were more considerable in the CC direction. We also found that the cervical lymph nodes (except for one patient whose GTV_T increased during IC) moved in the lateral–medial direction. The left and right sides of GTV_N shifted medially during IC by an average of 0.8 mm and 0.4 mm, respectively. While the GTV displacement results obtained in this study is not directly useful during IC, they may be important for CCRT if similar trends also follow chemotherapy concurrent to radiation therapy. To our knowledge, this is the first time that GTV volume reduction was explored through imaging in effort to optimize IC delivery for LANPC. However, there are three major limitations. First, repeat CTs were done without contrast, which may not be of sufficient quality in order to allow for accurate target definition. Although it is logical to use contrast on each scan, frequent contrast enhancement may lead to a risk of renal impairment and will also increase the financial burden for patients. Second, because surgery is not standard for NPC, it is hard to obtain the pathological remission rate, this study focused on the trend of tumor volume change to optimize the number of IC cycles, and only investigated the completion rate of the subsequent CCRT. Long‐term follow‐up may be studied in future studies. Finally, even though our cohort had only 54 patients, our findings were able to show the changes in tumor shrinkage, and then provide evidence and new ideas for clinical practice. Because the sample size was only 54 patients, our conclusion remains to be validated in large‐scale studies. Further large‐sample with long‐term follow‐up and a prospective clinical trial is warranted. CONCLUSION Three cycles of platinum‐based doublet IC were well tolerated and effective for LANPC patients. The nasopharyngeal carcinoma patients with the same stage (III and IV) but different T and N stages may benefit from different cycles of IC. Volume reductions in the primary tumor and retropharyngeal lymph nodes are similar, based on which two cycles of IC are recommended for patients with large volumes of these targets. However, cervical lymph nodes showed continuous and significant volume reductions following all three cycles of IC. Therefore, a three‐cycle IC course are recommended for patients with large cervical lymph node volumes. Ying Li: Data curation (lead); funding acquisition (equal); methodology (equal); writing – original draft (lead). Jianping Bi: Data curation (equal); methodology (equal); writing – original draft (equal). Guoliang Pi: Investigation (equal). Hanping He: Data curation (equal); formal analysis (equal). Yanping Li: Data curation (equal). Dandan Zheng: Writing – review and editing (equal). Zecheng Wei: Investigation (equal). Guang Han: Conceptualization (lead); data curation (equal); funding acquisition (equal); methodology (lead); writing – review and editing (lead). This work was supported by Hubei Provincial Health Commission (grant no. WJ2017M145 and ZY2021M008). The authors declare that they have no conflict of interest to disclose. The Medical Ethics Committee of Hubei Cancer Hospital, Tongji Medical College, Huazhong University of Science and Technology approved this study. Informed consent was obtained from all patients. Not applicable.
A novel “7 sutures and 8 knots” surgical technique in reverse shoulder arthroplasty for proximal humeral fractures: tuberosity healing improves short-term clinical results	05ad9c75-fba7-461a-b842-daae9125ec36	10167075	Suturing[mh]	Proximal humeral fractures (PHFs) are the seventh most commonly observed fractures in adults and account for 4–10% of all fracture types. A bimodal distribution has been described: PHFs occur in elderly patients with decreased bone strength after low-energy traumas, while most high-energy injuries involve patients under the age of 55 . PHF incidence is rising in the elderly, especially in women, and it now constitutes the third most common osteoporotic fracture [ – ]. The choice of the most effective treatment option for PHFs should take into account the fracture morphology, patient co-morbidities and functional expectations, and it should aim to achieve a pain-free functional shoulder . Also, since PHFs in the elderly are fragility fractures, regardless of the treatment option, a multidisciplinary approach such as a fracture liaison service is fundamental in order to reduce the risk of further fractures . A variety of surgical options can be employed, including closed reduction and percutaneous fixation, closed or open reduction and internal fixation , and arthroplasty . Non-operative treatment is generally accepted for undisplaced or minimally displaced PHFs, or for displaced fractures in the elderly with low functional demands or who are not cleared for surgery [ , , ]. The most appropriate treatment for complex PHFs (cPHFs) in the elderly is still a topic of debate, as concomitant osteoporosis and significant comminution prevent the achievement of stable fixation, so they may benefit from arthroplasty rather than osteosynthesis [ , , ]. Historically, hemiarthroplasty (HA) was considered the preferred choice for operative treatment of cPHFs ; nevertheless, its outcomes are heterogeneous, so reverse shoulder arthroplasty (RSA) has emerged as an alternative treatment option [ – ]. The main theoretical advantage of RSA is that tuberosity healing and cuff rotator integrity are not prerequisites for a satisfactory outcome since RSA primarily depends on the deltoid muscle to restore shoulder function [ , , , , – ]. Nevertheless, it has been shown that tuberosity healing leads to better functional results and active motion, even in RSA [ – ]. This is due to the influence of the volume of the greater tuberosity in restoring the lateral offset, improving the deltoid wrapping over the RSA, and maintaining the function of the subscapularis. As a result, recent efforts to enhance the tuberosity healing rate have been made [ , – ], but a gold standard technique has not been identified. In the present paper, we present the results of a retrospective observational study conducted on patients older than 65 years of age who underwent RSA for cPHFs with the application of a novel “7 sutures and 8 knots” tuberosity fixation technique to achieve better tuberosity healing. Study design A retrospective and observational study was performed. Inclusion criteria were as follows: (1) a cPHF categorized as a Neer three- or four-part fracture, a head-splitting fracture, or with more than 40% of the joint surface head involved; (2) a cPHF occurring in a patient over 65 years of age; (3) a cPHF treated with RSA, a fracture-specific stem, and a standardized novel technique of tuberosity fixation including bone grafting between the metaphyseal part of the stem and the tuberosities performed by a single surgeon; and (4) a minimum clinical and radiological follow-up of 12 months. Patients with previous failed open reduction and internal fixation for PHFs, patients undergoing revision surgery, and patients whose tuberosity comminution did not allow fixation were excluded. At our Institution, no ethical committee nor institutional review board approval is necessary for retrospective studies, and all patients gave their informed consent to data collection and their anonymous use for scientific and teaching purposes. Surgical procedure All surgeries were performed by the same senior surgeon with great experience in the RSA procedure performed both for trauma and chronic pathologies. The same prosthesis was implanted in all cases (Equinoxe Reverse-Fracture System Prosthesis; Exactech Inc., Gainesville, FL, USA). All fractures were evaluated by plain radiographs and then further assessed via computed tomography scans with the multiplanar reconstruction technique. A deltopectoral approach was used in all cases. After identifying the fracture planes, the greater and lesser tuberosities were detected and tagged within the context of the tendons (the infraspinatus and teres minor and the subscapularis, respectively) with #2 nonabsorbable sutures (Fig. , green threads). The tenotomy of the long head of the biceps brachii tendon was performed. In cases where the supraspinatus tendon was still attached to the greater tuberosity, it was removed, leaving the posterior portion of the rotator cuff intact to facilitate greater tuberosity reduction to the humeral stem during repair. If the bicipital groove was still intact, the tuberosities were separated from each other using a chisel. The glenoid was prepared first after careful retraction of the tuberosities. Reaming of the glenoid surface was performed with a cannulated reamer inserted over a guidewire, and a hole for the central peg was drilled. A standard glenoid baseplate was implanted and secured with the required number of screws, followed by the glenosphere. Whenever possible, pre-operative planning and intraoperative navigation were employed, as previously described . Next, the humeral canal was prepared, and the appropriate fracture stem size was chosen and cemented in 25° of retroversion, being careful to limit the cement to the meta-diaphyseal level. Placing the humeral stem in such retroversion causes the major fin of the stem to be placed at the bicipital groove so that the tuberosity reconstruction can be as anatomical as possible. Before placing the stem, a #2 high-strength suture was passed through the medial fenestration of the prosthesis and around the stem (Fig. A, blue thread). Two drill holes were made in the humeral diaphysis before the hardening of the cement, and then two needles were inserted and left in place during cement polymerization to prevent their obstruction (Fig. ). This expedient is used to avoid possible fragmentation of the cement mantle with subsequent drilling. A #2 high-strength suture was passed into each drill hole and through the superior part of the subscapularis tendon and the external rotator tendon, respectively (Fig. A, pink threads). Then, a #2 nonabsorbable suture was passed horizontally through the external rotator tendon, the two cranial holes of the major fin, and again through the tendon (Fig. A, orange thread). A similar technique was employed for the subscapularis tendon, engaging the two distal holes of the prosthesis’s major fin (Fig. A, yellow thread). Figure B shows an intraoperative image of the sutures. At this point, a cancellous bone graft harvested from the humeral head was placed underneath and next to the major fin of the prosthesis and the tuberosities were secured with eight knots according to the technique illustrated in Fig. A, with the seven sutures placed previously. First, the greater and the lesser tuberosities were stabilized on the tuberosity bed (two knots in total, a knot for each tuberosity, yellow and orange threads) and then further tightened together with the medial thread of each suture (one knot). At this point, the two vertical sutures were secured, one for each tuberosity, and then knotted together (three knots in total, pink threads). Thereafter, the sutures used to detect and tag the tuberosities were tied together (a single knot with all four ends of the sutures, green threads). Lastly, the horizontal suture that had been passed through the medial fenestration of the prosthesis and behind the stem was knotted (one knot, blue thread), to further compress the tuberosities onto the stem and onto the humeral shaft (Fig. B). To control bleeding, in the absence of contraindications, tranexamic acid was administered both intravenously and locally, as previously described . One suction drain was left in place for 24 h. Post-operative care The same standardized post-operative protocol was used in all patients to minimize possible differences in the functional outcome due to differences in rehabilitation. The arm was rested in a neutral rotation sling in 45° abduction for 4 weeks to minimize tension on the tuberosities and enhance their union. Active and passive range of motion (ROM) of the elbow and the wrist was allowed. The sling was removed at 4 weeks and rehabilitation of the shoulder with a physiotherapist began. Passive ROM exercises in forward elevation and abduction were encouraged at 4 weeks, while active exercises were allowed at 5 weeks. External and internal rotation and strengthening exercises were not allowed until 6 weeks from surgery. No heavy lifting was allowed until 9 weeks post-operatively, and a return to all activities was permitted at 3 months post-operatively. Clinical and radiological assessment All patients were evaluated clinically and radiographically at 1, 3, 6 and 12 months after surgery and then annually. Shoulder function was assessed using the Constant scoring systems , and active ROM was recorded in forward elevation, abduction, and external and internal rotation. Overall subjective patient satisfaction was evaluated through a four-grade rating scale (very disappointed, disappointed, satisfied and very satisfied) and the Disability of the Arm, Shoulder and Hand (DASH) scoring system . Plain radiographs in the true antero-posterior view (Grashey projection) of the shoulder with a standardized “shoulder protocol” (65–70 kV, 16 mAs) were obtained at each visit. The greater tuberosity was considered healed when it was visible on the X-rays (Grashey projection in neutral rotation) and fused to the humeral shaft. Glenoid notching was evaluated according to the Nerot–Sirveaux classification . A retrospective and observational study was performed. Inclusion criteria were as follows: (1) a cPHF categorized as a Neer three- or four-part fracture, a head-splitting fracture, or with more than 40% of the joint surface head involved; (2) a cPHF occurring in a patient over 65 years of age; (3) a cPHF treated with RSA, a fracture-specific stem, and a standardized novel technique of tuberosity fixation including bone grafting between the metaphyseal part of the stem and the tuberosities performed by a single surgeon; and (4) a minimum clinical and radiological follow-up of 12 months. Patients with previous failed open reduction and internal fixation for PHFs, patients undergoing revision surgery, and patients whose tuberosity comminution did not allow fixation were excluded. At our Institution, no ethical committee nor institutional review board approval is necessary for retrospective studies, and all patients gave their informed consent to data collection and their anonymous use for scientific and teaching purposes. All surgeries were performed by the same senior surgeon with great experience in the RSA procedure performed both for trauma and chronic pathologies. The same prosthesis was implanted in all cases (Equinoxe Reverse-Fracture System Prosthesis; Exactech Inc., Gainesville, FL, USA). All fractures were evaluated by plain radiographs and then further assessed via computed tomography scans with the multiplanar reconstruction technique. A deltopectoral approach was used in all cases. After identifying the fracture planes, the greater and lesser tuberosities were detected and tagged within the context of the tendons (the infraspinatus and teres minor and the subscapularis, respectively) with #2 nonabsorbable sutures (Fig. , green threads). The tenotomy of the long head of the biceps brachii tendon was performed. In cases where the supraspinatus tendon was still attached to the greater tuberosity, it was removed, leaving the posterior portion of the rotator cuff intact to facilitate greater tuberosity reduction to the humeral stem during repair. If the bicipital groove was still intact, the tuberosities were separated from each other using a chisel. The glenoid was prepared first after careful retraction of the tuberosities. Reaming of the glenoid surface was performed with a cannulated reamer inserted over a guidewire, and a hole for the central peg was drilled. A standard glenoid baseplate was implanted and secured with the required number of screws, followed by the glenosphere. Whenever possible, pre-operative planning and intraoperative navigation were employed, as previously described . Next, the humeral canal was prepared, and the appropriate fracture stem size was chosen and cemented in 25° of retroversion, being careful to limit the cement to the meta-diaphyseal level. Placing the humeral stem in such retroversion causes the major fin of the stem to be placed at the bicipital groove so that the tuberosity reconstruction can be as anatomical as possible. Before placing the stem, a #2 high-strength suture was passed through the medial fenestration of the prosthesis and around the stem (Fig. A, blue thread). Two drill holes were made in the humeral diaphysis before the hardening of the cement, and then two needles were inserted and left in place during cement polymerization to prevent their obstruction (Fig. ). This expedient is used to avoid possible fragmentation of the cement mantle with subsequent drilling. A #2 high-strength suture was passed into each drill hole and through the superior part of the subscapularis tendon and the external rotator tendon, respectively (Fig. A, pink threads). Then, a #2 nonabsorbable suture was passed horizontally through the external rotator tendon, the two cranial holes of the major fin, and again through the tendon (Fig. A, orange thread). A similar technique was employed for the subscapularis tendon, engaging the two distal holes of the prosthesis’s major fin (Fig. A, yellow thread). Figure B shows an intraoperative image of the sutures. At this point, a cancellous bone graft harvested from the humeral head was placed underneath and next to the major fin of the prosthesis and the tuberosities were secured with eight knots according to the technique illustrated in Fig. A, with the seven sutures placed previously. First, the greater and the lesser tuberosities were stabilized on the tuberosity bed (two knots in total, a knot for each tuberosity, yellow and orange threads) and then further tightened together with the medial thread of each suture (one knot). At this point, the two vertical sutures were secured, one for each tuberosity, and then knotted together (three knots in total, pink threads). Thereafter, the sutures used to detect and tag the tuberosities were tied together (a single knot with all four ends of the sutures, green threads). Lastly, the horizontal suture that had been passed through the medial fenestration of the prosthesis and behind the stem was knotted (one knot, blue thread), to further compress the tuberosities onto the stem and onto the humeral shaft (Fig. B). To control bleeding, in the absence of contraindications, tranexamic acid was administered both intravenously and locally, as previously described . One suction drain was left in place for 24 h. The same standardized post-operative protocol was used in all patients to minimize possible differences in the functional outcome due to differences in rehabilitation. The arm was rested in a neutral rotation sling in 45° abduction for 4 weeks to minimize tension on the tuberosities and enhance their union. Active and passive range of motion (ROM) of the elbow and the wrist was allowed. The sling was removed at 4 weeks and rehabilitation of the shoulder with a physiotherapist began. Passive ROM exercises in forward elevation and abduction were encouraged at 4 weeks, while active exercises were allowed at 5 weeks. External and internal rotation and strengthening exercises were not allowed until 6 weeks from surgery. No heavy lifting was allowed until 9 weeks post-operatively, and a return to all activities was permitted at 3 months post-operatively. All patients were evaluated clinically and radiographically at 1, 3, 6 and 12 months after surgery and then annually. Shoulder function was assessed using the Constant scoring systems , and active ROM was recorded in forward elevation, abduction, and external and internal rotation. Overall subjective patient satisfaction was evaluated through a four-grade rating scale (very disappointed, disappointed, satisfied and very satisfied) and the Disability of the Arm, Shoulder and Hand (DASH) scoring system . Plain radiographs in the true antero-posterior view (Grashey projection) of the shoulder with a standardized “shoulder protocol” (65–70 kV, 16 mAs) were obtained at each visit. The greater tuberosity was considered healed when it was visible on the X-rays (Grashey projection in neutral rotation) and fused to the humeral shaft. Glenoid notching was evaluated according to the Nerot–Sirveaux classification . A series of 32 consecutive patients (33 shoulders) met the inclusion criteria: there were 6 males (18.2%) and 26 females (27 shoulders) (81.8%) with a mean (± standard deviation, SD) age of 77.1 ± 7.3 (range 65–92) years who were evaluated at a mean follow-up of 35.9 ± 16.2 (range 12–64) months after surgery. Intra-operatively, a standard glenoid baseplate was implanted in each case and secured with a mean of 2.4 ± 0.8 (range 2–5) screws with an average length of 31.4 ± 4.4 (range 22–42) mm. The diameter of the glenosphere was chosen to optimally fit the patient’s anatomy: a 38-mm-diameter glenosphere was used in most of the cases (24 shoulders), a 36-mm-diameter glenosphere was used in smaller subjects (5 shoulders, all females), and a 42-mm-diameter glenosphere was used for larger shoulders (4 males). Post-operative anaemia requiring blood transfusion occurred in 15 patients (46.9%), and a patient suffered from a Clostridium difficile infection. Inferior glenoid notching (grade 1) was observed in only one patient. Patients were hospitalized for a mean of 8.0 ± 4.3 (range 3–26) days. The mean ± SD active forward elevation was 129° ± 31° (range 60–180°), the mean abduction was 118° ± 27° (range 70–160°), the mean external rotation was 37° ± 8° (range 23–55°), and the mean internal rotation was 6 ± 3 (range 2–10) points on a converted scale which corresponded to reaching the L1–L3 vertebral level. Only one patient achieved less than 90° of forward elevation. The mean Constant score was 66.7 ± 20.5 (range 29–100) points, and the mean DASH score was 33.4 ± 22.6 (range 2–85) points. At the last follow-up, most patients were satisfied or very satisfied with the results of the surgical procedure. Only three patients (9.1%) were disappointed, and none was very disappointed. Despite the advanced ages of the patients, the use of the “7 sutures and 8 knots” technique plus an autologous bone graft added to a specific reverse shoulder fracture stem resulted in a high tuberosity healing rate and good functional outcomes. Twenty-nine out of 33 shoulders (87.9%) had complete tuberosity healing (Fig. ), while four patients (12.1%) presented tuberosity resorption. Although not statistically significant, the healed tuberosities group showed better clinical and radiographical results with respect to the non-healed tuberosities group. Overall demographics and functional outcomes for the 33 shoulders are summarized in Table , while differences in functional results between the two groups are summarized in Table . Anatomical tuberosity healing and rotator cuff integrity have been shown to be essential for good functional recovery after HA for PHFs [ – ]. The unreliable results achieved with HA in the elderly suffering from cPHFs led to attempts to treat these patients with RSA, since the functional outcomes are less dependent on tuberosity healing and cuff integrity [ – , , , , ]. However, recent studies have demonstrated that although tuberosity healing is not a prerequisite for a satisfactory outcome after RSA for cPHFs, it still leads to better clinical results [ – , , ]. It has been proven that tuberosity osteotomy or excision is associated with worse functional results, with particular reference to a loss of external rotation and to a higher risk of RSA instability . The main advantage is better deltoid wrapping, which helps to improve both the function and the stability of the prosthesis . In order to improve the tuberosity healing rate, many surgical techniques have been investigated, but a gold standard reinsertion technique has not been identified . Despite the lack of consensus, it appears from a recent metanalysis that the main fixation method relies on the combination of vertical and horizontal fixation with or without cerclage. Other than the suture techniques and construct, the use of a fracture-specific humeral stem with a large ingrowth surface for tuberosity healing, space for a bone graft and partial cementation techniques could also enhance tuberosity healing . The most pertinent findings of the present study are that tuberosity healing in RSA for cPHFs can be obtained, even in the elderly, by employing a standardized surgical technique, and that tuberosity healing leads to an improved functional outcome and increased patient satisfaction, even if these are not statistically significantly enhanced according to the present data. The pivotal points of this novel surgical technique are the use of a standardized and reproducible tuberosity suture technique, the use of a fracture-specific prosthetic stem associated with an autologous bone graft (harvested from the fractured head), and the application of a partial cementation technique. In our opinion, of paramount importance for obtaining a high tuberosity healing rate is “friendly” tuberosity management, with an optimal balance between the tension and elasticity of the construct. The present “7 sutures and 8 knots” technique employs three high-strength sutures combined with four nonabsorbable sutures and aims to achieve this correct balance, which can favour a high healing rate. In addition, as the geometry of the stem affects the bone integration around it [ , , ], a fracture-specific prosthesis was implanted in all cases. By reducing the proximal metal surface, a larger ingrowth surface is obtained, which can allow better reduction of the tuberosities to the stem and to each other and the use of an autologous bone graft, which may further enhance bone healing [ , , , ]. Given the poor bone quality in the elderly, our preference is to cement the stem in all patients. According to previous studies reported in the literature [ , , , ], the advantages of cementation include a low rate of iatrogenic fracture, the ability to provide optimal initial stability of the implant and fixation independent from osteogenesis, and the anti-infection ability of the antibiotic-loaded bone cement. However, direct thermal reactions and disturbance of the local blood flow might inhibit tuberosity healing. Accordingly, we limited the cementation to the meta-diaphyseal humeral portion, as suggested by Singh et al. . Despite the advanced age of our study group, fixation of the tuberosities associated with an autologous bone graft and the use of a fracture-specific stem and the partial cementation technique resulted in a high rate of tuberosity healing (> 85%). Our results confirm those of Grubhofer et al. , Boileau et al. , and Levy and Badman , who all observed a similar tuberosity healing rate (> 84%). In the present study, neither shoulder instability nor loosening were reported within the study period, which is consistent with previous studies that also underlined the importance of achieving tuberosity healing to prevent such complications . The present series demonstrates that the restoration of a better active ROM and better subjective results can be expected after tuberosity reconstruction and healing. Among the four patients in whom the tuberosities did not heal, two were disappointed with the results of the surgical procedure and complained about difficulties with activities of daily living (ADLs) which required active forward elevation or rotations. On the other hand, only one patient with healed tuberosities complained about difficulties with ADLs, probably because the dominant shoulder was involved by the fracture and the contralateral one was affected by a severe cuff arthropathy. This study has limitations. First, it is a retrospective study with a relatively small sample size: it may be difficult to correctly generalize the obtained data. Moreover, the relatively short follow-up could underestimate additional functional improvement or complications beyond 1 year post-operatively. However, no dislocations were observed, which is the primary early complication after RSA for cPHFs (it typically occurs within the first 3 months post-operatively) . Second, this novel technique has been shown to be reproducible, but there may be occurrences where it is not feasible. Extreme tuberosity comminution and therefore a complete lack of bone could prevent the use of such a suturing technique, even though, in our experience, we were not able to employ the “7 sutures and 8 knots” technique in only a single case after we standardized it. On that occasion, we tried to reconstruct the tuberosities with bone cement to guarantee implant stability. RSA has been demonstrated to be a feasible surgical option to treat cPHFs in the elderly and, although its function relies mainly on the deltoid muscle, reattaching the tuberosities leads to better functional and clinical outcomes. However, there is no consensus regarding the best surgical technique to obtain the highest rate of tuberosity consolidation. The present study shows that the “7 sutures and 8 knots” technique is a relatively straightforward and reproducible method, and, given the results (a tuberosity consolidation rate of > 85%), it is possible to affirm that—despite the above-mentioned limitations—it can provide an excellent success rate, considering both the mean age and the poor bone quality of the study group and the previous results reported in the literature.
Understanding the patient and family experience of nutrition and dietetic support during childhood cancer treatment	96f781b7-5e21-4bac-8331-6fc62f6274ea	10167176	Internal Medicine[mh]	A cancer diagnosis and treatment can significantly impact a child’s ability to maintain adequate dietary intake and nutrition . Appetite suppression, vomiting, nausea, mucositis, changes in taste and smell, and pain are side effects of anti-cancer therapies that can negatively affect a child’s nutrition status . Maintaining good nutrition status throughout cancer treatment is crucial for children with cancer , as poor dietary intake impacts growth and development, may result in excessive weight gain, decreased tolerance to chemotherapy, and increased risk of infection . Therefore, it is unsurprising that nutrition status is an independent predictor of survival . With many decisions about their child’s treatment being out of their control, feeding their child during treatment is one of the few areas of care parents and families feel they can influence . However, this can significantly strain family relationships and child eating practices. For example, treatment-related weight changes can prompt negative feeding practices such as increasing pressure to eat or compromising diet quality to maintain intake . As parents are highly motivated to seek information about nutrition and health , providing reliable and trustworthy nutrition support during treatment is necessary. Patient and family experiences with nutrition during treatment have been documented , with most studies investigating patient and parent perceptions of food intake or strategies used by parents to cope with changes in intake . Two studies in Australia and America have explored patient and family perceptions of nutrition support, such as enteral nutrition (EN) and parenteral nutrition (PN) . Given that optimising nutrition status has far-reaching benefits on treatment tolerance, adherence to treatment schedules, and improved quality of life , understanding the experience of patients and their families with nutrition as supportive care during treatment is warranted. This study aimed to understand the experience of families caring for a child with cancer in NZ, who have received dietetic support during cancer treatment and their preferences for the delivery, format, and timing of nutrition information. Setting This study was conducted at a specialist paediatric oncology centre in Auckland, NZ. Treatment for childhood cancer is coordinated with one other specialist centre, and fourteen shared care centres around NZ. Childhood cancer patients aged 0 to 15 years receiving treatment (which could include chemotherapy, radiation, immunotherapy, or surgery alone or in combination with other modalities) and their families were invited to participate in this study. Study design The study used a mixed methods design with eligible participants identified through discussions with the ward charge nurse and the patient’s assigned nurse. Patients who were classed as palliative, had been recently diagnosed and were not yet ready to receive additional information (at the direction of medical staff), or were deemed too unwell to participate were excluded. The inclusion criteria were otherwise kept broad for patients and their families due to the heterogeneity of this population. Semi-structured interviews and questionnaire completion was facilitated by one researcher (E.C.) between January and March 2022. The interviewer (E.C) was not involved in the clinical care of participants. Parents or guardians provided written, informed consent before completing the questionnaire and interview. If the patient was older than 6 years, assent was also obtained. The study received ethical approval from the Auckland Health Research Ethics Committee (AHREC) on 20/12/2021 (AH23378). Institutional approval was granted by the ADHB Research Office (Project number: A + 9288). Data collection Participants completed a Health and Nutrition Questionnaire adapted from the literature and a semi-structured interview. The questionnaire collected data on the patient’s demographics and diagnosis , eating behaviours , nutrition education and support received , symptom assessment , and requirements for nutrition support. The questionnaire was completed via direct entry on an iPad or computer using REDCap electronic data capture hosted by The University of Auckland . The Behavioural Paediatric Feeding Assessment Scale (BPFAS) is a parent-reported questionnaire that gathers information on mealtime behaviours . Only questions related to child eating behaviours were included, which assessed frequencies of behavioural feeding problems and whether parents considered these a problem . Five items from the food responsiveness sub-scale of the Child Eating Behaviour Questionnaire (CEBQ) were also included . These additional questions aimed to capture behaviours of overeating. The interviews followed a semi-structured interview guide (Supplementary Table ) and were audio-recorded. Interviews were conducted until thematic saturation on current nutrition support, effectiveness, and requirements for support was achieved. Thematic saturation was determined by repeated themes and an absence of new themes in subsequent interviews. Interview recordings were transcribed verbatim. Analysis Quantitative data were evaluated using SPSS (SPSS Statistics for Windows Version 26, IBM Corp., Armonk, NY, USA), and descriptive statistics reported. Continuous data were presented as mean (SD) and categorical data as frequencies (percentage). Qualitative transcripts were coded line-by-line and analysed using NVivo, 2022, v.12 (QSR International Pty Ltd., Victoria, Australia) using Braun and Clarke’s thematic analysis framework. A multilevel consensus coding methodology was completed to ensure accurate coding and thematic analysis. Seventy-one percent of all interviews ( n = 15) were coded independently by two investigators (E.C and G.P), who then reviewed their analysis and discussed any discrepancies. This study was conducted at a specialist paediatric oncology centre in Auckland, NZ. Treatment for childhood cancer is coordinated with one other specialist centre, and fourteen shared care centres around NZ. Childhood cancer patients aged 0 to 15 years receiving treatment (which could include chemotherapy, radiation, immunotherapy, or surgery alone or in combination with other modalities) and their families were invited to participate in this study. The study used a mixed methods design with eligible participants identified through discussions with the ward charge nurse and the patient’s assigned nurse. Patients who were classed as palliative, had been recently diagnosed and were not yet ready to receive additional information (at the direction of medical staff), or were deemed too unwell to participate were excluded. The inclusion criteria were otherwise kept broad for patients and their families due to the heterogeneity of this population. Semi-structured interviews and questionnaire completion was facilitated by one researcher (E.C.) between January and March 2022. The interviewer (E.C) was not involved in the clinical care of participants. Parents or guardians provided written, informed consent before completing the questionnaire and interview. If the patient was older than 6 years, assent was also obtained. The study received ethical approval from the Auckland Health Research Ethics Committee (AHREC) on 20/12/2021 (AH23378). Institutional approval was granted by the ADHB Research Office (Project number: A + 9288). Participants completed a Health and Nutrition Questionnaire adapted from the literature and a semi-structured interview. The questionnaire collected data on the patient’s demographics and diagnosis , eating behaviours , nutrition education and support received , symptom assessment , and requirements for nutrition support. The questionnaire was completed via direct entry on an iPad or computer using REDCap electronic data capture hosted by The University of Auckland . The Behavioural Paediatric Feeding Assessment Scale (BPFAS) is a parent-reported questionnaire that gathers information on mealtime behaviours . Only questions related to child eating behaviours were included, which assessed frequencies of behavioural feeding problems and whether parents considered these a problem . Five items from the food responsiveness sub-scale of the Child Eating Behaviour Questionnaire (CEBQ) were also included . These additional questions aimed to capture behaviours of overeating. The interviews followed a semi-structured interview guide (Supplementary Table ) and were audio-recorded. Interviews were conducted until thematic saturation on current nutrition support, effectiveness, and requirements for support was achieved. Thematic saturation was determined by repeated themes and an absence of new themes in subsequent interviews. Interview recordings were transcribed verbatim. Quantitative data were evaluated using SPSS (SPSS Statistics for Windows Version 26, IBM Corp., Armonk, NY, USA), and descriptive statistics reported. Continuous data were presented as mean (SD) and categorical data as frequencies (percentage). Qualitative transcripts were coded line-by-line and analysed using NVivo, 2022, v.12 (QSR International Pty Ltd., Victoria, Australia) using Braun and Clarke’s thematic analysis framework. A multilevel consensus coding methodology was completed to ensure accurate coding and thematic analysis. Seventy-one percent of all interviews ( n = 15) were coded independently by two investigators (E.C and G.P), who then reviewed their analysis and discussed any discrepancies. Participants Baseline characteristics are displayed in Table . Twenty-one participants completed the Health and Nutrition Questionnaire and a semi-structured interview. Most (48%) were a family member of a child with cancer between the ages of 2 and 4 years. Most (48%) children were of NZ European ethnicity, and one in ten children (9%) were of self-defined Māori ethnicity. Diagnoses were recorded according to ICCC-3 classification , with acute lymphoblastic leukaemia (ALL) and Wilms tumour being the most common diagnoses (29% each). Ninety-five percent of children ( n = 20) were receiving chemotherapy at the time of study participation. Health and nutrition questionnaire Eighteen (86%) participants reported being concerned about their child’s diet or nutrition during treatment (Table ), with the most common concern being anorexia, or loss of appetite (62%), followed by vomiting (29%), and weight loss (24%). Fourteen (67%) participants rated nutrition as extremely important during childhood cancer; however, only 8 (38%) participants reported nutrition being addressed at every clinic or hospital visit. Over half (52%) of the children had experienced weight loss between 0 and 5 kg since diagnosis, of which 48% ( n = 10) of parents found worrying. Most (67%) participants reported changes in their child’s diet following diagnosis (Supplementary Table ). Common dietary changes included a focus on food safety (29%), deterioration in intake (21%), and an impact on the types and amounts of foods accepted (21%). All participants (100%) had received nutrition care or advice from the hospital dietitian, and seven (33%) reported receiving nutrition information from their nurse specialist. The types of advice provided by their health care team focused on food safety (62%), general healthy eating for cancer (62%), and how to gain weight (24%). Most participants (85%) rated the nutrition advice from the dietitian as helpful (ranging from somewhat, very, or extremely helpful) and that the dietitians were responsive to their queries. Reasons for rating advice from the dietitian as ‘somewhat helpful’ included a resulting lack of knowledge on appropriate low-risk foods, that the focus appeared to be on weight maintenance rather than diet quality, and that they were only contacted when nasogastric (NG) feeding was recommended. One-third ( n = 7) of participants reported that they would have liked to have received more support from their dietitian when their child was on treatment. Suggestions for further support included online weekend support, meal plans/family meal ideas when at home, a list of ‘safe’ foods, early support to prevent oral aversions and poor eating behaviours, and general lifestyle advice. The Memorial Symptom Assessment Scale (MSAS) was mostly parent-completed (Table ). The mean (SD) number of symptoms experienced was 6.3 (4.2). Symptom prevalence in the last week experienced by most children included as follows: lack of energy (71%), pain (62%), lack of appetite (62%), diarrhoea (57%), nausea (57%), vomiting (48%), irritability (48%). Of these, symptoms such as lack of appetite, nausea, and vomiting directly impact nutrition. Other symptoms reported as less frequent within the last week but can indirectly impact nutrition include anxiety, depression, or pain impacting desire to eat or low energy/stamina to complete meals. Over half (55%) of the symptoms in the last week were rated moderate to very severe in all children who experienced them. Weight loss and constipation caused high levels of distress and severity, whereas nausea, vomiting, and lack of appetite were more likely to be rated as severe than distressing. Semi-structured interviews Twenty-one ( n = 21) semi-structured interviews were conducted with childhood cancer patients and members of their families in person ( n = 19) or via telephone ( n = 2). These included three patients and twenty-three family members. The emerging themes were organised into four categories: (1) patients experience significant and distressing nutrition challenges; (2) patients and families have mixed perceptions of EN; (3) there are gaps in the current nutrition support system for inpatients; and (4) a desire for more accessible nutrition support. Sub-categories for these themes and representative quotes are outlined in Table . Theme 1: Patients experience significant and distressing nutrition challenges Most patients experienced challenges related to limited dietary intake or fussy eating, which caused frustration and distress. Managing taste changes, difficulties communicating with young children, and accepting that weight loss was inevitable were all significant challenges during treatment. Yeah, I’d kind of eat one food, throw up and I would just – I couldn’t even think about the food, let alone like look at it. So, for a little bit I went off potatoes . — Interview 9 (CCS: male, osteosarcoma, 12–15 years) Fussy eating encompassed food aversions, periods of inadequate intake, poor diet quality, and missed feeding milestones. While some families described being able to cope with these challenges and viewed it as a phase, other families described how it significantly impacted their child’s healthy eating practices. Usually [healthy eating is] quite important to me. But, right now it’s like, he’s alive. And we’re just trying to keep him from losing weight so the fruits and veggies can, you know, be put on hold, unless he feels like it. — Interview 8 (Mother: male, Wilms tumour, 2–4 years) Theme 2: Patients and families have mixed perceptions of enteral nutrition Placement of NG tubes for feeding was common, with families having a mixed view of their benefits. Some parents were positive about the role of NG feeds for maintaining weight and providing optimal nutrition. However, initiating NG feeding was often communicated to parents as a suggestion rather than an instruction or part of the treatment protocol, leading to confusion about their necessity. I think [the NGT] shouldn’t be suggested; I think it should be informed that you have to have it […] yeah, no, it’s more reassuring to know that he’s getting intake . — Interview 3 (Father: male, ALL, 2–4 years) Reasons for hesitancy included fear of complications, developing a reliance on EN, and disruption to oral intake. Two families delayed NG insertion as they felt they had not received enough explanation of the mechanism of feeding, the benefits of EN, and the proposed duration of use. I knew we were probably gonna reach [inserting an NG tube] at a point, but I was trying to stall it for as long as I could. — Interview 19 (Mother: female, osteosarcoma, 12–15 years) We need to be guided by, yes, okay, cool, we’re gonna be on the tube. She’ll need that for probably a week. And then she’ll probably go back to feeding…… And then, this is how we introduce solids into the mix as well,” is sort of, you know, how I would imagine that would sort of get – get there . — Interview 10 (Mother: female, CNS tumour, < 2 years) Theme 3: There are gaps in the current nutrition support system Some patients and families reported limited contact with the dietitian. Dietitians on the wards often need to prioritise patients with significant weight loss or intensive nutrition support, e.g. PN. Families expressed difficulties accessing dietitians for other concerns and often sought alternative sources of information. If they could see a dietitian, participants reported significant barriers to implementing nutrition advice, such as treatment side effects impacting intake. When you’re only coming in once a week to clinic and you’re hoping, I guess, you’re gonna see a dietitian then. I don’t know whether you’re guaranteed that you will. — Interview 21 (Mother: female, ALL, 2–4 years) There was a consensus that there was information overload at diagnosis, with limited capacity to absorb information beyond diagnosis and treatment protocols. Parents get sometimes get really overwhelmed in ‘diagnosis days.’ They cannot process information . — Interview 4 (Father: male, neuroblastoma, 9–12 years) Parent desire to delay addressing nutrition until after treatment had finished was apparent, as was the limited food availability in/near the hospital. He’s got three months left of chemo and then he should be finished completely. So, I’m hoping that once this finishes and then he might come back to his old self. — Interview 17 (Mother; male, Lymphoma, 5–8 years) Parents reported hesitation when asked about alternative or homoeopathic nutrition approaches. Most parents did not report using any natural remedies, but when they did, there was always a certain level of care and caution taken to ensure that it did not interact with any of the therapies they were receiving and have unintended consequences. Instead of using like the paraffin creams all the time, we asked if maybe we could use some manuka honey. […] But again, the doctors – with his immunity dropping so low, any bacteria sometimes can be really harmful. — Interview 13 (Mother: male, AML, 2–4 years) My husband found that [manioc] has like B19 or something so it’s like an antioxidant, yeah. And it’s, like you know, it’s a traditional breakfast we eat it so […] those things which—which are like sort of highlighted, we would incorporate it more. — Interview 5 (Mother: male, neuroblastoma, 2–4 years) Other parents were more confident and would openly disclose natural remedies they had used; When it comes to her vomiting, […] I’ll try give her like gingernuts or something like that or make sure there’s like ginger like in our food. — Interview 16 (Mother: female, neuroblastoma, 2–4 years) Theme 4: Desire for more accessible nutrition support Finally, while some families felt they were receiving adequate support, many wanted more accessible nutrition support. Easily accessible, general nutrition support (e.g. healthy eating and nutritious meal ideas) that could be accessed when patients or caregivers had the time and headspace were requested. I found that pamphlets were quite good. ‘Cause the thing is – oh, even online as well ‘cause we’re here at the hospital. There’s a lot of downtime here. — Interview 15 (Mother: female, Burkitt’s lymphoma, 2–4 years) Others requested increased access to dietitians for more personalised support. For example, guidance on making up for missed feeding milestones, or support for healthy eating while their child was immunocompromised. In terms of timing, preferences varied and included support at diagnosis, and on treatment completion. Some families wanted support to be available pre-emptive of nutrition challenges. I mean the problem with written or online resources […] is, if you’ve got a question about your particular scenario, you’ve gotta have someone else to ask ‘cause it might not be covered in that information. — Interview 21 (Mother: female, ALL, 2–4 years) Relationship between emergent themes A concept map is displayed in Fig. to illustrate the relationship between the emergent themes. The current gaps in nutrition support that may contribute to the challenges faced by childhood cancer patients and their families were identified. Stretched dietetic resource and existing barriers to implementing dietary advice create an environment where nutrition support is often deescalated during treatment. This may create feelings of isolation when faced with challenges such as fussy eating/food aversions, feelings of frustration and distress, and mixed perceptions about the benefits of EN. Some families were very happy with the current dietary support provided. However, requests for more readily available additional support throughout their cancer treatment journey were made. Increased access to the hospital dietitian is required for more personalised, patient-centred support. Additionally, pamphlets and online resources providing general nutrition support were useful for patients and families to access in their own time. Baseline characteristics are displayed in Table . Twenty-one participants completed the Health and Nutrition Questionnaire and a semi-structured interview. Most (48%) were a family member of a child with cancer between the ages of 2 and 4 years. Most (48%) children were of NZ European ethnicity, and one in ten children (9%) were of self-defined Māori ethnicity. Diagnoses were recorded according to ICCC-3 classification , with acute lymphoblastic leukaemia (ALL) and Wilms tumour being the most common diagnoses (29% each). Ninety-five percent of children ( n = 20) were receiving chemotherapy at the time of study participation. Eighteen (86%) participants reported being concerned about their child’s diet or nutrition during treatment (Table ), with the most common concern being anorexia, or loss of appetite (62%), followed by vomiting (29%), and weight loss (24%). Fourteen (67%) participants rated nutrition as extremely important during childhood cancer; however, only 8 (38%) participants reported nutrition being addressed at every clinic or hospital visit. Over half (52%) of the children had experienced weight loss between 0 and 5 kg since diagnosis, of which 48% ( n = 10) of parents found worrying. Most (67%) participants reported changes in their child’s diet following diagnosis (Supplementary Table ). Common dietary changes included a focus on food safety (29%), deterioration in intake (21%), and an impact on the types and amounts of foods accepted (21%). All participants (100%) had received nutrition care or advice from the hospital dietitian, and seven (33%) reported receiving nutrition information from their nurse specialist. The types of advice provided by their health care team focused on food safety (62%), general healthy eating for cancer (62%), and how to gain weight (24%). Most participants (85%) rated the nutrition advice from the dietitian as helpful (ranging from somewhat, very, or extremely helpful) and that the dietitians were responsive to their queries. Reasons for rating advice from the dietitian as ‘somewhat helpful’ included a resulting lack of knowledge on appropriate low-risk foods, that the focus appeared to be on weight maintenance rather than diet quality, and that they were only contacted when nasogastric (NG) feeding was recommended. One-third ( n = 7) of participants reported that they would have liked to have received more support from their dietitian when their child was on treatment. Suggestions for further support included online weekend support, meal plans/family meal ideas when at home, a list of ‘safe’ foods, early support to prevent oral aversions and poor eating behaviours, and general lifestyle advice. The Memorial Symptom Assessment Scale (MSAS) was mostly parent-completed (Table ). The mean (SD) number of symptoms experienced was 6.3 (4.2). Symptom prevalence in the last week experienced by most children included as follows: lack of energy (71%), pain (62%), lack of appetite (62%), diarrhoea (57%), nausea (57%), vomiting (48%), irritability (48%). Of these, symptoms such as lack of appetite, nausea, and vomiting directly impact nutrition. Other symptoms reported as less frequent within the last week but can indirectly impact nutrition include anxiety, depression, or pain impacting desire to eat or low energy/stamina to complete meals. Over half (55%) of the symptoms in the last week were rated moderate to very severe in all children who experienced them. Weight loss and constipation caused high levels of distress and severity, whereas nausea, vomiting, and lack of appetite were more likely to be rated as severe than distressing. Twenty-one ( n = 21) semi-structured interviews were conducted with childhood cancer patients and members of their families in person ( n = 19) or via telephone ( n = 2). These included three patients and twenty-three family members. The emerging themes were organised into four categories: (1) patients experience significant and distressing nutrition challenges; (2) patients and families have mixed perceptions of EN; (3) there are gaps in the current nutrition support system for inpatients; and (4) a desire for more accessible nutrition support. Sub-categories for these themes and representative quotes are outlined in Table . Theme 1: Patients experience significant and distressing nutrition challenges Most patients experienced challenges related to limited dietary intake or fussy eating, which caused frustration and distress. Managing taste changes, difficulties communicating with young children, and accepting that weight loss was inevitable were all significant challenges during treatment. Yeah, I’d kind of eat one food, throw up and I would just – I couldn’t even think about the food, let alone like look at it. So, for a little bit I went off potatoes . — Interview 9 (CCS: male, osteosarcoma, 12–15 years) Fussy eating encompassed food aversions, periods of inadequate intake, poor diet quality, and missed feeding milestones. While some families described being able to cope with these challenges and viewed it as a phase, other families described how it significantly impacted their child’s healthy eating practices. Usually [healthy eating is] quite important to me. But, right now it’s like, he’s alive. And we’re just trying to keep him from losing weight so the fruits and veggies can, you know, be put on hold, unless he feels like it. — Interview 8 (Mother: male, Wilms tumour, 2–4 years) Theme 2: Patients and families have mixed perceptions of enteral nutrition Placement of NG tubes for feeding was common, with families having a mixed view of their benefits. Some parents were positive about the role of NG feeds for maintaining weight and providing optimal nutrition. However, initiating NG feeding was often communicated to parents as a suggestion rather than an instruction or part of the treatment protocol, leading to confusion about their necessity. I think [the NGT] shouldn’t be suggested; I think it should be informed that you have to have it […] yeah, no, it’s more reassuring to know that he’s getting intake . — Interview 3 (Father: male, ALL, 2–4 years) Reasons for hesitancy included fear of complications, developing a reliance on EN, and disruption to oral intake. Two families delayed NG insertion as they felt they had not received enough explanation of the mechanism of feeding, the benefits of EN, and the proposed duration of use. I knew we were probably gonna reach [inserting an NG tube] at a point, but I was trying to stall it for as long as I could. — Interview 19 (Mother: female, osteosarcoma, 12–15 years) We need to be guided by, yes, okay, cool, we’re gonna be on the tube. She’ll need that for probably a week. And then she’ll probably go back to feeding…… And then, this is how we introduce solids into the mix as well,” is sort of, you know, how I would imagine that would sort of get – get there . — Interview 10 (Mother: female, CNS tumour, < 2 years) Theme 3: There are gaps in the current nutrition support system Some patients and families reported limited contact with the dietitian. Dietitians on the wards often need to prioritise patients with significant weight loss or intensive nutrition support, e.g. PN. Families expressed difficulties accessing dietitians for other concerns and often sought alternative sources of information. If they could see a dietitian, participants reported significant barriers to implementing nutrition advice, such as treatment side effects impacting intake. When you’re only coming in once a week to clinic and you’re hoping, I guess, you’re gonna see a dietitian then. I don’t know whether you’re guaranteed that you will. — Interview 21 (Mother: female, ALL, 2–4 years) There was a consensus that there was information overload at diagnosis, with limited capacity to absorb information beyond diagnosis and treatment protocols. Parents get sometimes get really overwhelmed in ‘diagnosis days.’ They cannot process information . — Interview 4 (Father: male, neuroblastoma, 9–12 years) Parent desire to delay addressing nutrition until after treatment had finished was apparent, as was the limited food availability in/near the hospital. He’s got three months left of chemo and then he should be finished completely. So, I’m hoping that once this finishes and then he might come back to his old self. — Interview 17 (Mother; male, Lymphoma, 5–8 years) Parents reported hesitation when asked about alternative or homoeopathic nutrition approaches. Most parents did not report using any natural remedies, but when they did, there was always a certain level of care and caution taken to ensure that it did not interact with any of the therapies they were receiving and have unintended consequences. Instead of using like the paraffin creams all the time, we asked if maybe we could use some manuka honey. […] But again, the doctors – with his immunity dropping so low, any bacteria sometimes can be really harmful. — Interview 13 (Mother: male, AML, 2–4 years) My husband found that [manioc] has like B19 or something so it’s like an antioxidant, yeah. And it’s, like you know, it’s a traditional breakfast we eat it so […] those things which—which are like sort of highlighted, we would incorporate it more. — Interview 5 (Mother: male, neuroblastoma, 2–4 years) Other parents were more confident and would openly disclose natural remedies they had used; When it comes to her vomiting, […] I’ll try give her like gingernuts or something like that or make sure there’s like ginger like in our food. — Interview 16 (Mother: female, neuroblastoma, 2–4 years) Theme 4: Desire for more accessible nutrition support Finally, while some families felt they were receiving adequate support, many wanted more accessible nutrition support. Easily accessible, general nutrition support (e.g. healthy eating and nutritious meal ideas) that could be accessed when patients or caregivers had the time and headspace were requested. I found that pamphlets were quite good. ‘Cause the thing is – oh, even online as well ‘cause we’re here at the hospital. There’s a lot of downtime here. — Interview 15 (Mother: female, Burkitt’s lymphoma, 2–4 years) Others requested increased access to dietitians for more personalised support. For example, guidance on making up for missed feeding milestones, or support for healthy eating while their child was immunocompromised. In terms of timing, preferences varied and included support at diagnosis, and on treatment completion. Some families wanted support to be available pre-emptive of nutrition challenges. I mean the problem with written or online resources […] is, if you’ve got a question about your particular scenario, you’ve gotta have someone else to ask ‘cause it might not be covered in that information. — Interview 21 (Mother: female, ALL, 2–4 years) Most patients experienced challenges related to limited dietary intake or fussy eating, which caused frustration and distress. Managing taste changes, difficulties communicating with young children, and accepting that weight loss was inevitable were all significant challenges during treatment. Yeah, I’d kind of eat one food, throw up and I would just – I couldn’t even think about the food, let alone like look at it. So, for a little bit I went off potatoes . — Interview 9 (CCS: male, osteosarcoma, 12–15 years) Fussy eating encompassed food aversions, periods of inadequate intake, poor diet quality, and missed feeding milestones. While some families described being able to cope with these challenges and viewed it as a phase, other families described how it significantly impacted their child’s healthy eating practices. Usually [healthy eating is] quite important to me. But, right now it’s like, he’s alive. And we’re just trying to keep him from losing weight so the fruits and veggies can, you know, be put on hold, unless he feels like it. — Interview 8 (Mother: male, Wilms tumour, 2–4 years) Placement of NG tubes for feeding was common, with families having a mixed view of their benefits. Some parents were positive about the role of NG feeds for maintaining weight and providing optimal nutrition. However, initiating NG feeding was often communicated to parents as a suggestion rather than an instruction or part of the treatment protocol, leading to confusion about their necessity. I think [the NGT] shouldn’t be suggested; I think it should be informed that you have to have it […] yeah, no, it’s more reassuring to know that he’s getting intake . — Interview 3 (Father: male, ALL, 2–4 years) Reasons for hesitancy included fear of complications, developing a reliance on EN, and disruption to oral intake. Two families delayed NG insertion as they felt they had not received enough explanation of the mechanism of feeding, the benefits of EN, and the proposed duration of use. I knew we were probably gonna reach [inserting an NG tube] at a point, but I was trying to stall it for as long as I could. — Interview 19 (Mother: female, osteosarcoma, 12–15 years) We need to be guided by, yes, okay, cool, we’re gonna be on the tube. She’ll need that for probably a week. And then she’ll probably go back to feeding…… And then, this is how we introduce solids into the mix as well,” is sort of, you know, how I would imagine that would sort of get – get there . — Interview 10 (Mother: female, CNS tumour, < 2 years) Some patients and families reported limited contact with the dietitian. Dietitians on the wards often need to prioritise patients with significant weight loss or intensive nutrition support, e.g. PN. Families expressed difficulties accessing dietitians for other concerns and often sought alternative sources of information. If they could see a dietitian, participants reported significant barriers to implementing nutrition advice, such as treatment side effects impacting intake. When you’re only coming in once a week to clinic and you’re hoping, I guess, you’re gonna see a dietitian then. I don’t know whether you’re guaranteed that you will. — Interview 21 (Mother: female, ALL, 2–4 years) There was a consensus that there was information overload at diagnosis, with limited capacity to absorb information beyond diagnosis and treatment protocols. Parents get sometimes get really overwhelmed in ‘diagnosis days.’ They cannot process information . — Interview 4 (Father: male, neuroblastoma, 9–12 years) Parent desire to delay addressing nutrition until after treatment had finished was apparent, as was the limited food availability in/near the hospital. He’s got three months left of chemo and then he should be finished completely. So, I’m hoping that once this finishes and then he might come back to his old self. — Interview 17 (Mother; male, Lymphoma, 5–8 years) Parents reported hesitation when asked about alternative or homoeopathic nutrition approaches. Most parents did not report using any natural remedies, but when they did, there was always a certain level of care and caution taken to ensure that it did not interact with any of the therapies they were receiving and have unintended consequences. Instead of using like the paraffin creams all the time, we asked if maybe we could use some manuka honey. […] But again, the doctors – with his immunity dropping so low, any bacteria sometimes can be really harmful. — Interview 13 (Mother: male, AML, 2–4 years) My husband found that [manioc] has like B19 or something so it’s like an antioxidant, yeah. And it’s, like you know, it’s a traditional breakfast we eat it so […] those things which—which are like sort of highlighted, we would incorporate it more. — Interview 5 (Mother: male, neuroblastoma, 2–4 years) Other parents were more confident and would openly disclose natural remedies they had used; When it comes to her vomiting, […] I’ll try give her like gingernuts or something like that or make sure there’s like ginger like in our food. — Interview 16 (Mother: female, neuroblastoma, 2–4 years) Finally, while some families felt they were receiving adequate support, many wanted more accessible nutrition support. Easily accessible, general nutrition support (e.g. healthy eating and nutritious meal ideas) that could be accessed when patients or caregivers had the time and headspace were requested. I found that pamphlets were quite good. ‘Cause the thing is – oh, even online as well ‘cause we’re here at the hospital. There’s a lot of downtime here. — Interview 15 (Mother: female, Burkitt’s lymphoma, 2–4 years) Others requested increased access to dietitians for more personalised support. For example, guidance on making up for missed feeding milestones, or support for healthy eating while their child was immunocompromised. In terms of timing, preferences varied and included support at diagnosis, and on treatment completion. Some families wanted support to be available pre-emptive of nutrition challenges. I mean the problem with written or online resources […] is, if you’ve got a question about your particular scenario, you’ve gotta have someone else to ask ‘cause it might not be covered in that information. — Interview 21 (Mother: female, ALL, 2–4 years) A concept map is displayed in Fig. to illustrate the relationship between the emergent themes. The current gaps in nutrition support that may contribute to the challenges faced by childhood cancer patients and their families were identified. Stretched dietetic resource and existing barriers to implementing dietary advice create an environment where nutrition support is often deescalated during treatment. This may create feelings of isolation when faced with challenges such as fussy eating/food aversions, feelings of frustration and distress, and mixed perceptions about the benefits of EN. Some families were very happy with the current dietary support provided. However, requests for more readily available additional support throughout their cancer treatment journey were made. Increased access to the hospital dietitian is required for more personalised, patient-centred support. Additionally, pamphlets and online resources providing general nutrition support were useful for patients and families to access in their own time. This study aimed to understand the experience of families caring for a child with cancer in NZ, who received nutrition and dietetic support during cancer treatment and determine preferences for the delivery, format, and timing of nutrition information/support. Nutrition-related challenges during treatment included fussy eating, weight changes, and mixed perceptions of EN, which families found distressing. These challenges were often a result of the intensive cancer therapies; however, perceived limitations to accessing ward dietitians and existing barriers to implementing nutrition advice may exacerbate these experiences. Two prevalent nutrition-related challenges resulting in feelings of frustration and distress were changes in weight and fussiness. Over half of the children had experienced weight loss since diagnosis, and almost a quarter of parents reported an impact on the types and amounts of foods accepted by their child. Food aversions, inadequate intake, poor diet quality, and missing feeding milestones were all reported by families and were categorised as ‘fussy eating challenges’. These challenges have previously been reported in the literature alongside malnutrition and are often intertwined . Lower BMI percentile at diagnosis and during treatment are related to greater parent-reported resistance to eating and aversions to mealtimes . The contribution of treatment side effects to overall intake exacerbated parent stress, with an average of 6.3 symptoms known to disrupt oral intake reported . These sentiments are echoed in cohorts around the world, with Swedish parents reporting the responsibility of getting their child to eat as distressing, often resulting in coercive feeding practices to increase intake . Negative feeding practices, such as coercion, were also commonly discussed in this study. Families used strategies to prioritise total food intake over diet quality, pressuring their child to eat or allowing their child to dictate their food choices. Similar findings were reported in Australia, where parents reported feeling stressed and engaging in conflict with their child at mealtimes, often turning to the use of pressure, food bribes, or threats of an NG tube to coerce their child into eating . Perceptions of EN, such as NG feeding, were mixed. Some families had faced mechanical challenges (e.g. tube expulsion or blocking), which contributed to negative perceptions. Other reasons for resistance to EN included concerns about becoming reliant on this form of nutrition or that EN would disrupt oral intake and the body’s ‘natural’ hunger and fullness sensations. Often it was felt that health professionals did not adequately address these questions and concerns. The use of EN during childhood cancer treatment is common, particularly in the prevention of malnutrition . However, inconsistent recommendations from health care professionals may contribute to families’ uncertainty around its benefit . Cohorts of parents in Australia and America have demonstrated similar conclusions, with 100% of parents interviewed in an American cohort preferring the use of PN if their child was unable to eat, despite EN being the safer method . Both studies concluded that standardised, evidence-based information from health professionals was essential for patients and families to make informed decisions . This was also apparent in the present study, where one caregiver reported delaying NG tube insertion because they believed they would have to prepare and blend up meals to use as the feed. Standardised information and nutrition pathways may minimise delays to appropriate nutrition support and reduce malnutrition risk. Current guidelines for best practice reinforce the importance of early and sustained dietitian involvement in patient care . Families perceived limited access to the ward dietitian as a potential risk factor for increased nutrition challenges. All participants had contact with a dietitian, and the quality of the consults was reported to be strong. However, there was a need for increased frequency of support outside of weight loss or EN. Modifying practice to a more proactive approach could reduce the risk of significant malnutrition in this patient group . Feeling stressed or overwhelmed while receiving large amounts of new information has been shown to impact families’ recall . Poor retention of information has previously been identified, with > 50% of parents reporting little to no recall of the first two consults with their child’s oncologist . Despite the importance of adequate nutrition in a patient’s treatment plan, decision-making around initiating nutrition support is difficult for parents and healthcare professionals. Shared decision-making (SDM) is the best practice in patient-centred care, where treatment decisions are made in collaboration with the patient (their family) and health professional . Decision aids (DA) are one way to implement SDM in clinical practice by providing patients with evidence-based options within the context of their preferences and values . Using a DA for nutrition support following a childhood cancer diagnosis would guide patients through a deliberative process of actively weighing up the requirements for nutrition with the pros and cons of engaging in nutrition support during treatment. Nutrition DA’s have been piloted in Australian paediatric populations and warrant implementation in NZ. This study validates findings from international studies and provides an NZ context to the priorities and needs of patients and families. Hearing perspectives from Māori, Pacific, and Asian families reflects the diverse population in NZ. However, the participation of Māori (9%) and Pacific (5%) families was low compared to national averages . Sampling bias must be considered, as participation was voluntary and during a defined period; therefore, families with particularly distressing experiences with nutrition may have been more likely to participate. Excluding patients who were too unwell to participate (at the discretion of the medical team) may have led to the exclusion of patients and their families most at risk of malnutrition. The distribution of diagnoses was not reflective of the annual distribution, where the most common cancer diagnosis in New Zealand in 2021 were CNS tumours (24%), then leukaemias (23%), and lymphomas (17%) . The limited data collection timeframe may have influenced this, as would the diagnoses admitted to the ward, which were majority ALL and solid tumours such as Wilm’s tumours or neuroblastomas. Additionally, this study may have limited generalisability due to its focus on NZ experiences. Childhood cancer patients and families experience significant and distressing nutrition challenges during treatment, including fussy eating, weight fluctuations, and mixed attitudes towards EN. Despite the importance of nutrition support in a patient’s treatment plan, decision-making around its initiation is difficult for parents and healthcare professionals. Standardising information given to patients and their families through a DA may optimise nutrition support for paediatric oncology patients and reduce the discordance between families and health professionals. Future implementation of a nutrition DA in this population is warranted. Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 21 KB)
What Works Where and How for Uptake and Impact of Artificial Intelligence in Pathology: Review of Theories for a Realist Evaluation	9af9871c-2e5b-4d2f-b1aa-ef5c9fb83353	10167589	Pathology[mh]	Pathologists diagnose cancer and other diseases by using a microscope to examine glass slides containing thin sections of human tissue. They perform a variety of tasks including recognizing patterns at high and low power, counting and measuring particular features, and using this information to classify, grade, and stage tumors. Until recently, pathology had remained largely unchanged since the specialty first emerged over 100 years ago . While there had been huge advances in how tissue is stained and visualized, with the discovery of new chemicals and the development of labeled antibodies, the fundamental process of morphological assessment of tissue had not altered, with the microscope remaining as essential as it ever was . However, technological advances over the past 20 years mean that it is now possible to scan slides quickly and at high resolution so that they can be viewed on a computer display through the use of whole slide imaging (WSI) . This echoes the digitization of radiology but with one considerable difference; WSI necessitates an analog (glass slide) to digital (WSI image) conversion, whereas radiology captures images in digital format . These digital images can be subjected to computer-based image analysis, which can aid pathologists in a range of areas such as biomarker quantification, object measurement, and object counting (cells or nuclei) . There has been widespread interest in developing artificial intelligence (AI) image analysis tools in pathology . The combination of WSI and AI means there is not only the possibility of replacing the microscope but also radically altering the role of pathologists. However, while there are a number of commentaries about the challenges of introducing AI into health care , there are few empirical studies of the implementation of AI into practice . A recent scoping review of studies of perceptions of AI among clinicians, patients, and the public, which identified 26 studies, found moderate acceptability, but a number of concerns were identified, including a lack of trust in patient safety and technology maturity . Only 11 of the 26 studies included health care staff, and none explored the perceptions of pathologists. This review explores the current literature in order to understand stakeholders’ perspectives on factors that may support or constrain the implementation and uptake of AI in pathology. Given the lack of literature on pathologists’ perceptions of AI, we have drawn on the approach of realist evaluation, allowing us to make use of a wider range of literature, including relevant theory. Study Design The use of AI in health care can be characterized as a complex intervention comprising a number of different elements that act both independently and interdependently. These include technological (eg, functionality and user interface), organizational (eg, implementation process, including training and support), and social components (eg, staff attitudes). Studying complex interventions requires a strong theoretical foundation . Realist evaluation is a theory-driven approach to understanding for whom and in what circumstances complex interventions work and has been used for studying a number of complex interventions, including health information technology . Realist approaches can be used both to evaluate complex interventions and inform intervention design . For these reasons, a realist approach is highly appropriate for the conduct of this review. Technology depends on human agency to work; technology in and of itself does not cause change, it is how people choose to make use of (or not) the resources that a technology offers to them that lead to what we typically consider to be the impacts of technology. Such choices are highly dependent on context. So, while a technology may provide the desired impact in one context, it is unlikely to produce the same impact across all settings. Realist evaluation is a methodology that explicitly recognizes this. It involves constructing, testing, and refining stakeholders’ ideas or theories about how and in what contexts a technology is supposed to work. These theories detail how particular contexts shape users’ responses to components of the technology (intervention mechanisms) to generate outcomes. They are presented as context-mechanism-outcome (CMO) configurations, where context (C) + mechanism (M) = outcome (O). In this way, and in comparison to more general qualitative approaches, realist evaluation moves beyond listing barriers and facilitators, offering specificity in understanding the relationship between contexts, mechanisms, and outcomes. The elicitation of stakeholders’ theories can be done in a number of ways, such as interviewing stakeholders, reviewing the existing literature on the topic, identifying relevant theories from the sociological or other works of literature, or some combination of these approaches. To ensure we are building on existing work, we have chosen to elicit stakeholders’ theories through a review of the literature related to the use of AI in pathology. In contrast to a full realist review, where published evidence is used to test and refine stakeholders’ theories , theories elicited in this review will be refined through interviews with pathologists in later stages of the research. While this is not a full realist review, where relevant, we have reported the items included in the Realist and Meta-narrative Evidence Synthesis: Evolving Standards (RAMESES) reporting guidelines for the reporting of realist reviews (see for completed RAMESES checklist). Search Strategy The overriding question for the review was what works, for whom, in what circumstances, and how to encourage uptake and impact of AI in pathology. To address this question, several searches were undertaken. Search 1 sought studies, reports, and policy documents from the following databases: arXiv.org (Cornell University) repository, Ovid MEDLINE(R), and HMIC Health Management Information Consortium (Ovid). Searches were developed for the concepts of artificial intelligence and histopathology. Subject headings and free text words were identified for use in the search concepts by an information specialist (JW) and project team members. Further terms were identified and tested from known relevant papers. Search results were limited to English-language publications. We also limited search results to publications published since 2000, given the recent development of AI in pathology. The searches were peer-reviewed by a second information specialist using the Peer Review of Electronic Search Strategies (PRESS) checklist (see for full search strategies) . Search 2 sought reports discussing AI, authored by Eric Topol in Ovid MEDLINE(R), Sciences Citation Index (Clarivate Analytics Web of Science), and Emerging Sources Citation Index (Clarivate Analytics Web of Science; see ). While searching for a particular author is not typically used in traditional systematic reviews, named author searches are recognized as a method for increasing the number of outputs included in a review . Such an approach is recommended for realist reviews and more generally for understanding the theories that underpin complex interventions and the contexts in which they are implemented . Named author searches are used in the theory elicitation stage of realist reviews as a means of gathering opinion pieces, remembering that we are looking for theories rather than empirical evidence . Eric Topol was chosen as an “opinion leader,” given his popular science book on the topic of AI in health care and his input into England’s strategy on digital health training and education, including around AI . Searches were also undertaken of the following websites: the Food and Drugs Administration, the College of American Pathologists, the Royal College of Pathologists, and the Digital Pathology Association along with a number of Google searches (see ). Additional papers were identified through personal recommendation and “snowballing” (pursuing references of references) . Search results were collated and deduplicated in EndNote (Clarivate). Selection and Appraisal of Documents As the aim of this review is to identify and characterize stakeholders’ perspectives regarding contextual factors that will enhance or restrict the uptake of AI in pathology rather than to assess the validity of these perspectives, the identified papers were screened not based on rigor but on relevance to the review question. Titles and abstracts returned from the search results were screened in EndNote, asking the following questions: is the paper about AI in pathology or health care? And does the paper contain ideas about how, for whom, and in what circumstance AI can work (in the clinical setting)? Full-text copies of potentially relevant documents were then obtained and read by the reviewers to identify if they contained ideas about the introduction of AI into pathology or other relevant health care settings. While these searches were primarily concerned with pathology, papers from outside this field (particularly radiology) were also returned. This is due to the medical use of “pathology” to describe the features typical of the way a disease presents, in addition to it being the branch of medicine that deals with the analysis of body tissue for diagnostic or forensic purposes. These tangential papers were retained as they potentially contained concepts that were transferable or relevant to the field of pathology. Similarly, a number of papers concerned with image analysis without the use of AI were retained on the basis that some of the potential supports and constraints to implementation they described would have relevance to the use of AI-based image analysis. Data Extraction, Analysis, and Synthesis Documents were entered into NVivo 12 (QSR International Pty Ltd) software for qualitative analysis. Sections of text were indexed in an iterative process using a series of codes that evolved to represent topics relevant to the review question. Following the realist strategy, these codes sought to capture different contexts, mechanisms, and outcomes that could affect the introduction of AI in pathology. The coded data were used to produce narrative summaries of each of the identified contexts, mechanisms, and outcomes. The reviewers and other authors then discussed these narratives and translated them into CMO configurations. While the documents provided data about outcomes associated with the use of AI and some data could be drawn out about contexts, there was little about the mechanisms through which these were achieved. Therefore, to guide our thinking, we also drew on substantive theories concerning the implementation of technology and complex interventions more generally, identifying potentially relevant theories through team discussions of the emerging themes. Use of substantive theory is in line with the realist approach, which argues that the design of interventions tends to be based on a limited number of theories regarding human behavior and therefore, rather than starting from scratch when evaluating a new intervention, researchers should also make use of existing theory . The use of AI in health care can be characterized as a complex intervention comprising a number of different elements that act both independently and interdependently. These include technological (eg, functionality and user interface), organizational (eg, implementation process, including training and support), and social components (eg, staff attitudes). Studying complex interventions requires a strong theoretical foundation . Realist evaluation is a theory-driven approach to understanding for whom and in what circumstances complex interventions work and has been used for studying a number of complex interventions, including health information technology . Realist approaches can be used both to evaluate complex interventions and inform intervention design . For these reasons, a realist approach is highly appropriate for the conduct of this review. Technology depends on human agency to work; technology in and of itself does not cause change, it is how people choose to make use of (or not) the resources that a technology offers to them that lead to what we typically consider to be the impacts of technology. Such choices are highly dependent on context. So, while a technology may provide the desired impact in one context, it is unlikely to produce the same impact across all settings. Realist evaluation is a methodology that explicitly recognizes this. It involves constructing, testing, and refining stakeholders’ ideas or theories about how and in what contexts a technology is supposed to work. These theories detail how particular contexts shape users’ responses to components of the technology (intervention mechanisms) to generate outcomes. They are presented as context-mechanism-outcome (CMO) configurations, where context (C) + mechanism (M) = outcome (O). In this way, and in comparison to more general qualitative approaches, realist evaluation moves beyond listing barriers and facilitators, offering specificity in understanding the relationship between contexts, mechanisms, and outcomes. The elicitation of stakeholders’ theories can be done in a number of ways, such as interviewing stakeholders, reviewing the existing literature on the topic, identifying relevant theories from the sociological or other works of literature, or some combination of these approaches. To ensure we are building on existing work, we have chosen to elicit stakeholders’ theories through a review of the literature related to the use of AI in pathology. In contrast to a full realist review, where published evidence is used to test and refine stakeholders’ theories , theories elicited in this review will be refined through interviews with pathologists in later stages of the research. While this is not a full realist review, where relevant, we have reported the items included in the Realist and Meta-narrative Evidence Synthesis: Evolving Standards (RAMESES) reporting guidelines for the reporting of realist reviews (see for completed RAMESES checklist). The overriding question for the review was what works, for whom, in what circumstances, and how to encourage uptake and impact of AI in pathology. To address this question, several searches were undertaken. Search 1 sought studies, reports, and policy documents from the following databases: arXiv.org (Cornell University) repository, Ovid MEDLINE(R), and HMIC Health Management Information Consortium (Ovid). Searches were developed for the concepts of artificial intelligence and histopathology. Subject headings and free text words were identified for use in the search concepts by an information specialist (JW) and project team members. Further terms were identified and tested from known relevant papers. Search results were limited to English-language publications. We also limited search results to publications published since 2000, given the recent development of AI in pathology. The searches were peer-reviewed by a second information specialist using the Peer Review of Electronic Search Strategies (PRESS) checklist (see for full search strategies) . Search 2 sought reports discussing AI, authored by Eric Topol in Ovid MEDLINE(R), Sciences Citation Index (Clarivate Analytics Web of Science), and Emerging Sources Citation Index (Clarivate Analytics Web of Science; see ). While searching for a particular author is not typically used in traditional systematic reviews, named author searches are recognized as a method for increasing the number of outputs included in a review . Such an approach is recommended for realist reviews and more generally for understanding the theories that underpin complex interventions and the contexts in which they are implemented . Named author searches are used in the theory elicitation stage of realist reviews as a means of gathering opinion pieces, remembering that we are looking for theories rather than empirical evidence . Eric Topol was chosen as an “opinion leader,” given his popular science book on the topic of AI in health care and his input into England’s strategy on digital health training and education, including around AI . Searches were also undertaken of the following websites: the Food and Drugs Administration, the College of American Pathologists, the Royal College of Pathologists, and the Digital Pathology Association along with a number of Google searches (see ). Additional papers were identified through personal recommendation and “snowballing” (pursuing references of references) . Search results were collated and deduplicated in EndNote (Clarivate). As the aim of this review is to identify and characterize stakeholders’ perspectives regarding contextual factors that will enhance or restrict the uptake of AI in pathology rather than to assess the validity of these perspectives, the identified papers were screened not based on rigor but on relevance to the review question. Titles and abstracts returned from the search results were screened in EndNote, asking the following questions: is the paper about AI in pathology or health care? And does the paper contain ideas about how, for whom, and in what circumstance AI can work (in the clinical setting)? Full-text copies of potentially relevant documents were then obtained and read by the reviewers to identify if they contained ideas about the introduction of AI into pathology or other relevant health care settings. While these searches were primarily concerned with pathology, papers from outside this field (particularly radiology) were also returned. This is due to the medical use of “pathology” to describe the features typical of the way a disease presents, in addition to it being the branch of medicine that deals with the analysis of body tissue for diagnostic or forensic purposes. These tangential papers were retained as they potentially contained concepts that were transferable or relevant to the field of pathology. Similarly, a number of papers concerned with image analysis without the use of AI were retained on the basis that some of the potential supports and constraints to implementation they described would have relevance to the use of AI-based image analysis. Documents were entered into NVivo 12 (QSR International Pty Ltd) software for qualitative analysis. Sections of text were indexed in an iterative process using a series of codes that evolved to represent topics relevant to the review question. Following the realist strategy, these codes sought to capture different contexts, mechanisms, and outcomes that could affect the introduction of AI in pathology. The coded data were used to produce narrative summaries of each of the identified contexts, mechanisms, and outcomes. The reviewers and other authors then discussed these narratives and translated them into CMO configurations. While the documents provided data about outcomes associated with the use of AI and some data could be drawn out about contexts, there was little about the mechanisms through which these were achieved. Therefore, to guide our thinking, we also drew on substantive theories concerning the implementation of technology and complex interventions more generally, identifying potentially relevant theories through team discussions of the emerging themes. Use of substantive theory is in line with the realist approach, which argues that the design of interventions tends to be based on a limited number of theories regarding human behavior and therefore, rather than starting from scratch when evaluating a new intervention, researchers should also make use of existing theory . Overview The search identified 1420 papers, 4 web pages, and 9 government or institution or foundation documents, providing a total of 1433 unique records (see for PRISMA [Preferred Reporting Items for Systematic Reviews and Meta-Analyses] diagram). After title and abstract screening, 1294 documents were determined to not be relevant, leaving 139 potentially relevant documents. After screening of full texts of these, 101 documents were identified as relevant. All 101 documents were coded, although there was much repetition of the themes contained within them. Below we summarize these themes, organizing them according to the realist concepts of context, mechanism, and outcome. As realist evaluation typically starts with looking at outcome patterns before identifying the contexts and mechanisms that lead to the outcome pattern, the anticipated impacts (outcomes), both positive and negative, of AI in pathology were first considered. The contexts that may impact uptake and impacts of AI were then considered. The mechanisms that may be triggered in particular contexts for the outcomes to be achieved were finally considered. The analysis also suggested practical challenges of introducing AI, such as infrastructure and the need for adequate training data, but we do not discuss these in what follows as our focus is on contextual factors that are likely to shape pathologists’ responses to AI. Impacts of AI in Pathology Accuracy The most commonly mentioned benefit of the use of AI, whether anticipated or determined experimentally, was increased accuracy . Some authors have argued that currently pathology is a subjective specialty relying on manual observation subject to human skill, procedural errors, or inefﬁciencies in processes and bias . From this perspective, increased accuracy results from the standardization of diagnosis through quantitative evaluation of samples . Increased accuracy was also seen as a consequence of using AI to remove variance in results attributed to decrease in pathologist performance that occurs when working under time constraints . It has also been argued that, if AI is used to augment the decision-making of pathologists, it may increase the knowledge of pathologists, and in that way increase accuracy both when AI is used and when the pathologist works alone . Speed Another predicted benefit of AI in pathology was the ability of computers to make a diagnosis quickly. This was expressed as increasing the speed of the diagnosis , improving workflow , keeping pace with increasing demand , and increasing efficiency . One way in which authors theorized this increased efficiency could be achieved was by reducing the number of slides that the pathologist has to look at . For example, in a screening, triage, or prioritization scenario, the pathologist would be presented with all positive slides for rapid review, while in a diagnostic or fully automated scenario, the pathologist would not need to review all benign slides, as the AI tool would review them instead. This could remove a significant percentage of slides from the workload of a pathologist. Another way in which it was theorized this could be achieved was through identifying regions of interest in a slide . An anticipated knock-on benefit of increasing efficiency was addressing the shortage of pathologists . Combination of Data Sources The ability to analyze multiple disparate data sources was highlighted as another possible benefit of AI. This would include combining images with patient records , combining pathology with other imaging techniques (magnetic resonance imaging, computerized tomography, and x-ray) , and integrating pathology with genomics, metabolomics, and other diagnostic techniques (eg, Raman spectroscopy) . While currently hypothetical, authors argue that this has the possibility to better characterize disease , which may lead to new or enhanced predictive models . Role of the Pathologist Another potential impact is on the role of the pathologist. Most authors opine that pathologists are unlikely to be replaced by AI but that their role is likely to change substantially , echoing what has regularly happened when disruptive technology has been introduced . In the immediate term, it is thought that the role of the pathologist is unlikely to change, as currently, the field of AI in image analysis is in its infancy. Some have argued that, if anything, the pathologist will be more indispensable than ever since their knowledge will be essential for algorithm design and the generation of annotated data for AI training . Furthermore, there will always be a role in assessing whether enough tissue of sufficient quality (eg, lacking artifacts and representative of the targeted lesion) is present at the tissue processing stage and directing an AI algorithm to assess specific areas of tissue identified as of interest by a pathologist. Jha and Topol and others theorize an alternative view of the future where, as AI for image analysis improves, pathologists, along with radiologists, will become information specialists, managing information extracted by AI in the clinical context of the patient rather than extracting information from images themselves. Contexts for AI Use Collaborative Working In the literature, we identified 2 key contextual factors that authors argue have the potential to affect whether or not the anticipated benefits of AI are achieved. The first of these is the size of and expertise within a department and to what extent the pathologists within that department work collaboratively. Studies looking at the accuracy of AI have typically compared AI with the decision-making of a single pathologist, and many papers present a scenario of the pathologist working alone (with or without the support of AI). However, Campanella et al note the collaborative nature of pathology, arguing that, with access to additional information provided by immunohistochemistry, it could be assumed “that a team of pathologists at a comprehensive cancer centre will operate with 100% sensitivity and specificity,” and therefore, AI in such a context should not seek to achieve the impossible goal of surpassing the performance of pathologists. Campanella et al go on to argue that, in such a context, the focus should be on the AI achieving 100% sensitivity with an acceptable false positive rate, so that pathologists can focus on those cases and slides where the AI has identified a tumor, thereby increasing efficiency. While some may question this claim of 100% sensitivity and specificity, the underlying theory seems to be that, when pathologists work collaboratively to generate consensus diagnoses, sensitivity and specificity are likely to be greater than when a pathologist is working alone. Conversely, a theory stemming from this is that AI could provide greater benefit in terms of increasing accuracy in smaller, nonspecialist departments. Thus, while AI is often touted as a replacement for a variable or inconsistent human opinion, in the real world, pathologists can get opinions from others and use ancillary testing to confirm diagnoses. So, depending on the clinical context, accuracy at a single point in the diagnostic process may be less important than the overall output of the process. Regulation The second contextual factor is that of regulation. A key issue raised is the tension between rapid technological advancement and safety . Allen describes the difficulty of balancing the need to ensure patient safety without stifling development. A particular issue is the lack of transparency about how AI algorithms work; it is not possible to see inside the “black box” of their decision-making and know what features the algorithm is using to make its decision . Some have argued that one of the benefits of AI is the potential to reduce medical malpractice liability by improving diagnosis and treatment, reducing medical error, and preventing ineffective and unnecessary care . However, concerns regarding how the use of AI will impact pathologists’ liability are more dominant in the literature, suggesting that a failure to resolve these issues could constrain the uptake of AI. Some have argued that the rise of AI in health care challenges the traditional liability structures used if AI moves beyond augmenting the work of pathologists and begins to, for example, automate certain tasks . A survey of pathologists found that almost 50% thought that the platform vendor should bear some of the liability . Mechanisms In realist evaluation, mechanisms are understood as a combination of the resources that a technology provides and the users’ responses to those resources. The literature described a broad range of actual and potential technologies that could be considered as resources. These different technologies can be conceptualized at a more abstract level, using a typology that we identified in the literature. This typology describes 3 different models of the relationship between the pathologist and AI . In the first model, the pathologist is “in the loop,” or what some refer to as the “augmented pathologist” , where AI is a tool that pathologists use to aid their diagnosis . It has been proposed that this is the model that we are most likely to see in the short term. Some have suggested that AI should be seen as a colleague who can provide second opinions on difficult cases, but without individual challenges, such as tiredness, and collaborative challenges, such as those that result from hierarchies . The AI-human combination is believed to be more accurate because the errors made by AI are not strongly correlated with the errors made by humans . In the second model, the pathologist will become “on the loop,” and some authors suggest we will see this in the medium term . In this scenario, AI is capable of making independent decisions but pathologists are still involved. This could initially be a system whereby benign or normal tissue is screened out at an early stage leaving the pathologist to concentrate on diseased tissue. As algorithms further advance, it is possible that AI diagnoses more and more conditions, with the pathologist only needed for highly unusual or ambiguous cases. In this situation, the pathologist plays a role in quality control and oversight, checking that the decisions being made are appropriate rather than making the decisions. In the third model, the pathologist is “out of the loop,” a scenario that some authors predict we may reach in the long term , at least for some decisions. It has been suggested AI could automate routine tasks and more time-consuming tasks . It is theorized that this will enable pathologists to spend more time on high-level decision-making tasks, particularly those related to disease presentations with more confounding features . Tasks that have been identified in the literature for automation include those “tedious routine diagnostic tasks that require great accuracy” such as finding metastases in lymph node sections, with the potential for a significant reduction in the workload of pathologists . In this situation, AI has become autonomous, and decision-making has shifted away from human control. This may be unlikely to happen in the foreseeable future, and the general consensus in the literature is that there will always be human involvement in the diagnostic process . How pathologists might reason about and respond to these different models of working alongside AI was less clear from the literature. However, broader literature on the use of AI in health care provides some useful insights about which of these models is most likely to be responded to positively. For uptake of AI, the literature suggests that there is the need for advice in a way that recognizes the expertise of the user, making it clear that it is designed to inform and assist but not replace the clinician , implying that the scenario of the pathologist being “in the loop” will be most acceptable to pathologists. In thinking about this scenario, we drew on substantive theories regarding the implementation of technology and complex interventions more generally. Here, normalization process theory (NPT) provided valuable understanding. Successful introduction of technology involves interactions between individual clinicians and their work environment until the technology becomes embedded (routinely incorporated into everyday work) and integrated (sustained over time) into routine practice, a process known as “normalization” . NPT suggests that, for normalization to occur, 4 key constructs need to be considered: coherence: sense-making—where individuals make sense of the new technology and how it differs from existing practice; cognitive participation: the process of engaging individuals with the introduction of the technology; collective action: how the work processes are adapted and altered to make the intervention happen; and reflexive monitoring: the formal and informal appraisal of the benefits and costs of the intervention . This suggests that if pathologists have been able to “make sense” of AI, have been engaged in the adoption process, have been able to adapt their work processes, and are able to identify potential benefits to its introduction, it is more likely to become embedded into practice. We also looked at theories relating to the adoption of a clinical decision support system (CDSS), with AI being a form of CDSS . Relevant theories are the user acceptance and system adaptation design model and the input-process-output-engage (IPOE) model . The user acceptance and system adaptation design model suggests that, for users to accept a CDSS, an iterative design process with early end-user involvement is needed, along with rigorous usability testing in both laboratory and natural settings, to ensure that the system works within the cognitive and environmental constraints of the intended user. The IPOE model suggests that acceptance of CDSS requires the CDSS to provide users with the rules that the machine followed to generate the output, so the user can make informed decisions when deciding whether to follow the recommendation. A challenge, highlighted by NPT and the IPOE model, may be making sense of the black box of AI. In the pathology literature, it is theorized that trust is needed for uptake of AI . Drawing on studies of AI implementation in other health care settings, we can also theorize that pathologists’ trust will be eroded when the AI recommendations conflict with their own observations and experience . Consequently, it is theorized that the “resource” of explainable AI is necessary for building pathologists’ trust and will increase acceptance of the scenario of pathologists being “on the loop” . The search identified 1420 papers, 4 web pages, and 9 government or institution or foundation documents, providing a total of 1433 unique records (see for PRISMA [Preferred Reporting Items for Systematic Reviews and Meta-Analyses] diagram). After title and abstract screening, 1294 documents were determined to not be relevant, leaving 139 potentially relevant documents. After screening of full texts of these, 101 documents were identified as relevant. All 101 documents were coded, although there was much repetition of the themes contained within them. Below we summarize these themes, organizing them according to the realist concepts of context, mechanism, and outcome. As realist evaluation typically starts with looking at outcome patterns before identifying the contexts and mechanisms that lead to the outcome pattern, the anticipated impacts (outcomes), both positive and negative, of AI in pathology were first considered. The contexts that may impact uptake and impacts of AI were then considered. The mechanisms that may be triggered in particular contexts for the outcomes to be achieved were finally considered. The analysis also suggested practical challenges of introducing AI, such as infrastructure and the need for adequate training data, but we do not discuss these in what follows as our focus is on contextual factors that are likely to shape pathologists’ responses to AI. Accuracy The most commonly mentioned benefit of the use of AI, whether anticipated or determined experimentally, was increased accuracy . Some authors have argued that currently pathology is a subjective specialty relying on manual observation subject to human skill, procedural errors, or inefﬁciencies in processes and bias . From this perspective, increased accuracy results from the standardization of diagnosis through quantitative evaluation of samples . Increased accuracy was also seen as a consequence of using AI to remove variance in results attributed to decrease in pathologist performance that occurs when working under time constraints . It has also been argued that, if AI is used to augment the decision-making of pathologists, it may increase the knowledge of pathologists, and in that way increase accuracy both when AI is used and when the pathologist works alone . Speed Another predicted benefit of AI in pathology was the ability of computers to make a diagnosis quickly. This was expressed as increasing the speed of the diagnosis , improving workflow , keeping pace with increasing demand , and increasing efficiency . One way in which authors theorized this increased efficiency could be achieved was by reducing the number of slides that the pathologist has to look at . For example, in a screening, triage, or prioritization scenario, the pathologist would be presented with all positive slides for rapid review, while in a diagnostic or fully automated scenario, the pathologist would not need to review all benign slides, as the AI tool would review them instead. This could remove a significant percentage of slides from the workload of a pathologist. Another way in which it was theorized this could be achieved was through identifying regions of interest in a slide . An anticipated knock-on benefit of increasing efficiency was addressing the shortage of pathologists . Combination of Data Sources The ability to analyze multiple disparate data sources was highlighted as another possible benefit of AI. This would include combining images with patient records , combining pathology with other imaging techniques (magnetic resonance imaging, computerized tomography, and x-ray) , and integrating pathology with genomics, metabolomics, and other diagnostic techniques (eg, Raman spectroscopy) . While currently hypothetical, authors argue that this has the possibility to better characterize disease , which may lead to new or enhanced predictive models . Role of the Pathologist Another potential impact is on the role of the pathologist. Most authors opine that pathologists are unlikely to be replaced by AI but that their role is likely to change substantially , echoing what has regularly happened when disruptive technology has been introduced . In the immediate term, it is thought that the role of the pathologist is unlikely to change, as currently, the field of AI in image analysis is in its infancy. Some have argued that, if anything, the pathologist will be more indispensable than ever since their knowledge will be essential for algorithm design and the generation of annotated data for AI training . Furthermore, there will always be a role in assessing whether enough tissue of sufficient quality (eg, lacking artifacts and representative of the targeted lesion) is present at the tissue processing stage and directing an AI algorithm to assess specific areas of tissue identified as of interest by a pathologist. Jha and Topol and others theorize an alternative view of the future where, as AI for image analysis improves, pathologists, along with radiologists, will become information specialists, managing information extracted by AI in the clinical context of the patient rather than extracting information from images themselves. The most commonly mentioned benefit of the use of AI, whether anticipated or determined experimentally, was increased accuracy . Some authors have argued that currently pathology is a subjective specialty relying on manual observation subject to human skill, procedural errors, or inefﬁciencies in processes and bias . From this perspective, increased accuracy results from the standardization of diagnosis through quantitative evaluation of samples . Increased accuracy was also seen as a consequence of using AI to remove variance in results attributed to decrease in pathologist performance that occurs when working under time constraints . It has also been argued that, if AI is used to augment the decision-making of pathologists, it may increase the knowledge of pathologists, and in that way increase accuracy both when AI is used and when the pathologist works alone . Another predicted benefit of AI in pathology was the ability of computers to make a diagnosis quickly. This was expressed as increasing the speed of the diagnosis , improving workflow , keeping pace with increasing demand , and increasing efficiency . One way in which authors theorized this increased efficiency could be achieved was by reducing the number of slides that the pathologist has to look at . For example, in a screening, triage, or prioritization scenario, the pathologist would be presented with all positive slides for rapid review, while in a diagnostic or fully automated scenario, the pathologist would not need to review all benign slides, as the AI tool would review them instead. This could remove a significant percentage of slides from the workload of a pathologist. Another way in which it was theorized this could be achieved was through identifying regions of interest in a slide . An anticipated knock-on benefit of increasing efficiency was addressing the shortage of pathologists . The ability to analyze multiple disparate data sources was highlighted as another possible benefit of AI. This would include combining images with patient records , combining pathology with other imaging techniques (magnetic resonance imaging, computerized tomography, and x-ray) , and integrating pathology with genomics, metabolomics, and other diagnostic techniques (eg, Raman spectroscopy) . While currently hypothetical, authors argue that this has the possibility to better characterize disease , which may lead to new or enhanced predictive models . Another potential impact is on the role of the pathologist. Most authors opine that pathologists are unlikely to be replaced by AI but that their role is likely to change substantially , echoing what has regularly happened when disruptive technology has been introduced . In the immediate term, it is thought that the role of the pathologist is unlikely to change, as currently, the field of AI in image analysis is in its infancy. Some have argued that, if anything, the pathologist will be more indispensable than ever since their knowledge will be essential for algorithm design and the generation of annotated data for AI training . Furthermore, there will always be a role in assessing whether enough tissue of sufficient quality (eg, lacking artifacts and representative of the targeted lesion) is present at the tissue processing stage and directing an AI algorithm to assess specific areas of tissue identified as of interest by a pathologist. Jha and Topol and others theorize an alternative view of the future where, as AI for image analysis improves, pathologists, along with radiologists, will become information specialists, managing information extracted by AI in the clinical context of the patient rather than extracting information from images themselves. Collaborative Working In the literature, we identified 2 key contextual factors that authors argue have the potential to affect whether or not the anticipated benefits of AI are achieved. The first of these is the size of and expertise within a department and to what extent the pathologists within that department work collaboratively. Studies looking at the accuracy of AI have typically compared AI with the decision-making of a single pathologist, and many papers present a scenario of the pathologist working alone (with or without the support of AI). However, Campanella et al note the collaborative nature of pathology, arguing that, with access to additional information provided by immunohistochemistry, it could be assumed “that a team of pathologists at a comprehensive cancer centre will operate with 100% sensitivity and specificity,” and therefore, AI in such a context should not seek to achieve the impossible goal of surpassing the performance of pathologists. Campanella et al go on to argue that, in such a context, the focus should be on the AI achieving 100% sensitivity with an acceptable false positive rate, so that pathologists can focus on those cases and slides where the AI has identified a tumor, thereby increasing efficiency. While some may question this claim of 100% sensitivity and specificity, the underlying theory seems to be that, when pathologists work collaboratively to generate consensus diagnoses, sensitivity and specificity are likely to be greater than when a pathologist is working alone. Conversely, a theory stemming from this is that AI could provide greater benefit in terms of increasing accuracy in smaller, nonspecialist departments. Thus, while AI is often touted as a replacement for a variable or inconsistent human opinion, in the real world, pathologists can get opinions from others and use ancillary testing to confirm diagnoses. So, depending on the clinical context, accuracy at a single point in the diagnostic process may be less important than the overall output of the process. Regulation The second contextual factor is that of regulation. A key issue raised is the tension between rapid technological advancement and safety . Allen describes the difficulty of balancing the need to ensure patient safety without stifling development. A particular issue is the lack of transparency about how AI algorithms work; it is not possible to see inside the “black box” of their decision-making and know what features the algorithm is using to make its decision . Some have argued that one of the benefits of AI is the potential to reduce medical malpractice liability by improving diagnosis and treatment, reducing medical error, and preventing ineffective and unnecessary care . However, concerns regarding how the use of AI will impact pathologists’ liability are more dominant in the literature, suggesting that a failure to resolve these issues could constrain the uptake of AI. Some have argued that the rise of AI in health care challenges the traditional liability structures used if AI moves beyond augmenting the work of pathologists and begins to, for example, automate certain tasks . A survey of pathologists found that almost 50% thought that the platform vendor should bear some of the liability . In the literature, we identified 2 key contextual factors that authors argue have the potential to affect whether or not the anticipated benefits of AI are achieved. The first of these is the size of and expertise within a department and to what extent the pathologists within that department work collaboratively. Studies looking at the accuracy of AI have typically compared AI with the decision-making of a single pathologist, and many papers present a scenario of the pathologist working alone (with or without the support of AI). However, Campanella et al note the collaborative nature of pathology, arguing that, with access to additional information provided by immunohistochemistry, it could be assumed “that a team of pathologists at a comprehensive cancer centre will operate with 100% sensitivity and specificity,” and therefore, AI in such a context should not seek to achieve the impossible goal of surpassing the performance of pathologists. Campanella et al go on to argue that, in such a context, the focus should be on the AI achieving 100% sensitivity with an acceptable false positive rate, so that pathologists can focus on those cases and slides where the AI has identified a tumor, thereby increasing efficiency. While some may question this claim of 100% sensitivity and specificity, the underlying theory seems to be that, when pathologists work collaboratively to generate consensus diagnoses, sensitivity and specificity are likely to be greater than when a pathologist is working alone. Conversely, a theory stemming from this is that AI could provide greater benefit in terms of increasing accuracy in smaller, nonspecialist departments. Thus, while AI is often touted as a replacement for a variable or inconsistent human opinion, in the real world, pathologists can get opinions from others and use ancillary testing to confirm diagnoses. So, depending on the clinical context, accuracy at a single point in the diagnostic process may be less important than the overall output of the process. The second contextual factor is that of regulation. A key issue raised is the tension between rapid technological advancement and safety . Allen describes the difficulty of balancing the need to ensure patient safety without stifling development. A particular issue is the lack of transparency about how AI algorithms work; it is not possible to see inside the “black box” of their decision-making and know what features the algorithm is using to make its decision . Some have argued that one of the benefits of AI is the potential to reduce medical malpractice liability by improving diagnosis and treatment, reducing medical error, and preventing ineffective and unnecessary care . However, concerns regarding how the use of AI will impact pathologists’ liability are more dominant in the literature, suggesting that a failure to resolve these issues could constrain the uptake of AI. Some have argued that the rise of AI in health care challenges the traditional liability structures used if AI moves beyond augmenting the work of pathologists and begins to, for example, automate certain tasks . A survey of pathologists found that almost 50% thought that the platform vendor should bear some of the liability . In realist evaluation, mechanisms are understood as a combination of the resources that a technology provides and the users’ responses to those resources. The literature described a broad range of actual and potential technologies that could be considered as resources. These different technologies can be conceptualized at a more abstract level, using a typology that we identified in the literature. This typology describes 3 different models of the relationship between the pathologist and AI . In the first model, the pathologist is “in the loop,” or what some refer to as the “augmented pathologist” , where AI is a tool that pathologists use to aid their diagnosis . It has been proposed that this is the model that we are most likely to see in the short term. Some have suggested that AI should be seen as a colleague who can provide second opinions on difficult cases, but without individual challenges, such as tiredness, and collaborative challenges, such as those that result from hierarchies . The AI-human combination is believed to be more accurate because the errors made by AI are not strongly correlated with the errors made by humans . In the second model, the pathologist will become “on the loop,” and some authors suggest we will see this in the medium term . In this scenario, AI is capable of making independent decisions but pathologists are still involved. This could initially be a system whereby benign or normal tissue is screened out at an early stage leaving the pathologist to concentrate on diseased tissue. As algorithms further advance, it is possible that AI diagnoses more and more conditions, with the pathologist only needed for highly unusual or ambiguous cases. In this situation, the pathologist plays a role in quality control and oversight, checking that the decisions being made are appropriate rather than making the decisions. In the third model, the pathologist is “out of the loop,” a scenario that some authors predict we may reach in the long term , at least for some decisions. It has been suggested AI could automate routine tasks and more time-consuming tasks . It is theorized that this will enable pathologists to spend more time on high-level decision-making tasks, particularly those related to disease presentations with more confounding features . Tasks that have been identified in the literature for automation include those “tedious routine diagnostic tasks that require great accuracy” such as finding metastases in lymph node sections, with the potential for a significant reduction in the workload of pathologists . In this situation, AI has become autonomous, and decision-making has shifted away from human control. This may be unlikely to happen in the foreseeable future, and the general consensus in the literature is that there will always be human involvement in the diagnostic process . How pathologists might reason about and respond to these different models of working alongside AI was less clear from the literature. However, broader literature on the use of AI in health care provides some useful insights about which of these models is most likely to be responded to positively. For uptake of AI, the literature suggests that there is the need for advice in a way that recognizes the expertise of the user, making it clear that it is designed to inform and assist but not replace the clinician , implying that the scenario of the pathologist being “in the loop” will be most acceptable to pathologists. In thinking about this scenario, we drew on substantive theories regarding the implementation of technology and complex interventions more generally. Here, normalization process theory (NPT) provided valuable understanding. Successful introduction of technology involves interactions between individual clinicians and their work environment until the technology becomes embedded (routinely incorporated into everyday work) and integrated (sustained over time) into routine practice, a process known as “normalization” . NPT suggests that, for normalization to occur, 4 key constructs need to be considered: coherence: sense-making—where individuals make sense of the new technology and how it differs from existing practice; cognitive participation: the process of engaging individuals with the introduction of the technology; collective action: how the work processes are adapted and altered to make the intervention happen; and reflexive monitoring: the formal and informal appraisal of the benefits and costs of the intervention . This suggests that if pathologists have been able to “make sense” of AI, have been engaged in the adoption process, have been able to adapt their work processes, and are able to identify potential benefits to its introduction, it is more likely to become embedded into practice. We also looked at theories relating to the adoption of a clinical decision support system (CDSS), with AI being a form of CDSS . Relevant theories are the user acceptance and system adaptation design model and the input-process-output-engage (IPOE) model . The user acceptance and system adaptation design model suggests that, for users to accept a CDSS, an iterative design process with early end-user involvement is needed, along with rigorous usability testing in both laboratory and natural settings, to ensure that the system works within the cognitive and environmental constraints of the intended user. The IPOE model suggests that acceptance of CDSS requires the CDSS to provide users with the rules that the machine followed to generate the output, so the user can make informed decisions when deciding whether to follow the recommendation. A challenge, highlighted by NPT and the IPOE model, may be making sense of the black box of AI. In the pathology literature, it is theorized that trust is needed for uptake of AI . Drawing on studies of AI implementation in other health care settings, we can also theorize that pathologists’ trust will be eroded when the AI recommendations conflict with their own observations and experience . Consequently, it is theorized that the “resource” of explainable AI is necessary for building pathologists’ trust and will increase acceptance of the scenario of pathologists being “on the loop” . Principal Findings This review has described the anticipated benefits of AI in digital pathology. While these benefits are not surprising, reflecting broader claims made for AI in the health care literature—namely, increased accuracy and efficiency—the value of this review comes from the use of a realist approach, which has allowed us to go beyond a simple listing of benefits to theorizing the contexts in which, and the mechanisms through which, such benefits are likely to be achieved. These are summarized as CMO configurations in . It has also enabled us to draw on a wider range of literature than in more traditional review approaches, an important feature given the absence of studies exploring pathologists’ attitudes toward AI . We began this review with the intention of identifying stakeholders’ theories about the contextual factors that may support or constrain the adoption of AI in pathology. However, the review revealed a gap in the literature regarding the discussion of this topic, likely a reflection of the current state of progress in the development of AI in pathology. As we begin to introduce AI into pathology, there is a need for empirical research to address this gap. However, this points to an additional benefit of a realist approach, which allowed us to integrate existing theory concerning the implementation of technology and complex interventions more generally. This enabled us to develop some tentative theories regarding the mechanisms that may lead pathologists to choose to integrate AI into their work practice, providing a strong theoretical basis for future research in this area. The review has several implications for the design and reporting of studies as we begin to move from experimental studies to real-world evaluation and the use of AI in pathology departments. Firstly, the findings highlight tasks where AI is likely to provide the greatest benefit and which benefits are most desirable in a given setting. For example, in large specialist centers, the emphasis should be on reducing workload rather than increasing accuracy. This responds to calls for developers to carefully consider the tasks that are best performed by AI and those best performed by clinicians . Alternatively, the review findings can inform the selection of sites for evaluation. For example, if the ambition is to increase the accuracy of diagnosis, then smaller, nonspecialist departments are a more appropriate choice than large specialist centers. Designers also need to give careful thought to usability and how AI is integrated into the pathologist’s workflow. The findings and resulting CMO configurations also suggest that, except for simple quantitative tasks, explainable AI will be needed for pathologists to trust the recommendations provided. However, this may be based on a misunderstanding of the current nature of explainable AI; broad descriptions of how the AI system works in a general sense can be produced but they are rarely informative with respect to individual decisions . For example, in radiology, the trustworthiness of saliency maps, a widely used method to provide explainable AI in medical imaging, has been questioned . Instead, rigorous evaluations should be used to provide evidence of the trustworthiness of AI, as is the case with other black-box systems in health care, such as medicines where the mechanisms of action are only partially understood . Following on from this, the review findings highlight information that should be reported in evaluation studies of AI, regarding not only the size and nature of the department but also how pathologists worked in the different departments, working alone or in teams. The issue of reporting AI studies in health care has been highlighted by other authors, with recommendations for describing the study setting, the target user, the digitized workflow, and the extent of use . Our analysis suggests that for real-world studies, a finer-grained level of reporting would be beneficial; by capturing and reporting such details, it will enable identification of the specific contexts in which AI is likely to provide the greatest benefit and therefore, where its implementation is most justified. To capture this information regarding the extent of collaborative work, as well as to capture information about the contextual factors that support or constrain the use of AI in different departments, real-world trials need to be complemented by mixed method or qualitative process evaluations. The findings also have implications for the introduction of AI into pathology. Pathologists should be involved in the decision to introduce AI and have the opportunity to feed into evaluations of the costs and benefits of the system following its introduction. Limitations The CMO configurations presented earlier are tentative theories, as they have not been tested with empirical data. In the next stage of this research, these will be explored and refined through interviews with pathologists. This will provide the basis for a future realist evaluation of AI in pathology, gathering empirical data to test the theories. Conclusions This paper has presented a review of stakeholders’ theories of how and in what contexts AI will be adopted and provide benefit within pathology. The results suggest that for uptake of AI in pathology, for all but the simplest quantitative tasks, measures will be required that either increase confidence in the system or provide users with an understanding of the performance of the system. For specialist centers, efforts should focus on reducing workload, rather than increasing accuracy. Designers also need to give careful thought to usability and how AI is integrated into pathologists’ workflow. This review has described the anticipated benefits of AI in digital pathology. While these benefits are not surprising, reflecting broader claims made for AI in the health care literature—namely, increased accuracy and efficiency—the value of this review comes from the use of a realist approach, which has allowed us to go beyond a simple listing of benefits to theorizing the contexts in which, and the mechanisms through which, such benefits are likely to be achieved. These are summarized as CMO configurations in . It has also enabled us to draw on a wider range of literature than in more traditional review approaches, an important feature given the absence of studies exploring pathologists’ attitudes toward AI . We began this review with the intention of identifying stakeholders’ theories about the contextual factors that may support or constrain the adoption of AI in pathology. However, the review revealed a gap in the literature regarding the discussion of this topic, likely a reflection of the current state of progress in the development of AI in pathology. As we begin to introduce AI into pathology, there is a need for empirical research to address this gap. However, this points to an additional benefit of a realist approach, which allowed us to integrate existing theory concerning the implementation of technology and complex interventions more generally. This enabled us to develop some tentative theories regarding the mechanisms that may lead pathologists to choose to integrate AI into their work practice, providing a strong theoretical basis for future research in this area. The review has several implications for the design and reporting of studies as we begin to move from experimental studies to real-world evaluation and the use of AI in pathology departments. Firstly, the findings highlight tasks where AI is likely to provide the greatest benefit and which benefits are most desirable in a given setting. For example, in large specialist centers, the emphasis should be on reducing workload rather than increasing accuracy. This responds to calls for developers to carefully consider the tasks that are best performed by AI and those best performed by clinicians . Alternatively, the review findings can inform the selection of sites for evaluation. For example, if the ambition is to increase the accuracy of diagnosis, then smaller, nonspecialist departments are a more appropriate choice than large specialist centers. Designers also need to give careful thought to usability and how AI is integrated into the pathologist’s workflow. The findings and resulting CMO configurations also suggest that, except for simple quantitative tasks, explainable AI will be needed for pathologists to trust the recommendations provided. However, this may be based on a misunderstanding of the current nature of explainable AI; broad descriptions of how the AI system works in a general sense can be produced but they are rarely informative with respect to individual decisions . For example, in radiology, the trustworthiness of saliency maps, a widely used method to provide explainable AI in medical imaging, has been questioned . Instead, rigorous evaluations should be used to provide evidence of the trustworthiness of AI, as is the case with other black-box systems in health care, such as medicines where the mechanisms of action are only partially understood . Following on from this, the review findings highlight information that should be reported in evaluation studies of AI, regarding not only the size and nature of the department but also how pathologists worked in the different departments, working alone or in teams. The issue of reporting AI studies in health care has been highlighted by other authors, with recommendations for describing the study setting, the target user, the digitized workflow, and the extent of use . Our analysis suggests that for real-world studies, a finer-grained level of reporting would be beneficial; by capturing and reporting such details, it will enable identification of the specific contexts in which AI is likely to provide the greatest benefit and therefore, where its implementation is most justified. To capture this information regarding the extent of collaborative work, as well as to capture information about the contextual factors that support or constrain the use of AI in different departments, real-world trials need to be complemented by mixed method or qualitative process evaluations. The findings also have implications for the introduction of AI into pathology. Pathologists should be involved in the decision to introduce AI and have the opportunity to feed into evaluations of the costs and benefits of the system following its introduction. The CMO configurations presented earlier are tentative theories, as they have not been tested with empirical data. In the next stage of this research, these will be explored and refined through interviews with pathologists. This will provide the basis for a future realist evaluation of AI in pathology, gathering empirical data to test the theories. This paper has presented a review of stakeholders’ theories of how and in what contexts AI will be adopted and provide benefit within pathology. The results suggest that for uptake of AI in pathology, for all but the simplest quantitative tasks, measures will be required that either increase confidence in the system or provide users with an understanding of the performance of the system. For specialist centers, efforts should focus on reducing workload, rather than increasing accuracy. Designers also need to give careful thought to usability and how AI is integrated into pathologists’ workflow.
OncoCardioDB: a public and curated database of molecular information in onco-cardiology/cardio-oncology	f2a2ef13-eefe-4da5-aafa-1bf8698a3bbd	10167979	Internal Medicine[mh]	Onco-cardiology/cardio-oncology studies the relationship between cardiovascular diseases (CVDs) and cancer . It is well known that these families of diseases have a high incidence in the world’s population. Indeed, among non-communicable diseases, CVDs are the first cause of death , whereas cancer is second . CVDs and cancer historically have been studied separately, but the growing clinical evidence in favour of the relationship between them gave rise to onco-cardiology field in 2010 . These relationships can be approached in two ways: risk factors shared by both diseases and cardiotoxicity derived from cancer treatments. Common risk factors include smoking, physical inactivity, genetic predisposition and advanced age, among others . Cardiotoxicity derived from cancer treatment includes all types of cancer and therapy , and it is recognized as one of the main causes of mortality among cancer survivors , especially with regard to breast cancer . Cardiotoxicity impacts the prognosis of treated patients in both the short and long terms; indeed, it is possible to develop CVDs derived from cancer treatment years after having overcome the disease . Emerging evidence suggests that these diseases share molecular pathways such as chronic inflammation, oxidative stress, aberrant apoptosis and angiogenesis . Other studies have identified common molecular targets for both diseases, including the widely used cardiac biomarkers troponins and natriuretic peptides . Furthermore, evidence (publications and citations) related to cardio-oncology research increased almost exponentially from 2010 to date . From the former consideration, it seems important to store and organize all the molecular information generated so far in the emerging field of onco-cardiology. This would allow the assimilation of all the information related to molecular relationships between CVDs and cancer in a timely manner. It would also be useful in research and clinical practice, since in the long term, it will contribute to identifying new and safer preventive and therapeutic approaches for both diseases. This implies certain challenges, one of them being the differences in the way of expressing onco-cardiology information. Such differences suggest that some effort should be made to represent and store this information in a suitable, common way. This way must be simultaneously well standardized, well organized around the key concepts and fully accessible in order to extract valuable knowledge from it. Several databases exist that store, organize and make available the information, on the one hand, concerning cancer and, on the other hand, related to CVDs and molecular factors associated with pathologies in general. Some reviews have been published regarding the cancer databases , e.g. The Cancer Genome Atlas (TCGA). The TCGA is a research network that stores and analyses a large number of human tumours to discover molecular aberrations at the Deoxyribonucleic acid (DNA), Ribonucleic acid (RNA), protein and epigenetic levels, with the aim of improving our ability to diagnose, treat and prevent cancer. The data are accessible through the Genomic Data Commons (GDC) Data Portal (accessible at https://portal.gdc.cancer.gov/ ). The GDC Data Portal is a robust platform that allows you to search and download omics and clinical cancer data for analysis and includes 70 projects on 67 primary cancer sites with >85 0000 cases. With respect to CVDs, the CVD and CardioGenBase databases have been created to store and organize information related to multi-omic studies. Regarding the molecular bases of the disease, in general, we can mention the Online Mendelian Inheritance in Man (OMIM). The OMIM (available at https://www.omim.org ) contains comprehensive, authoritative and timely information concerning human genes and genetic disorders, obtained from biomedical literature, focusing on the relationship between the phenotype and genotype. It aims to support human genetics research and education and the practice of clinical genetics . Regarding efforts to store information related to onco-cardiology, the Virtual Cardio-Oncology Research Initiative aims to create a large-scale resource, with a long follow-up period, to provide quality data for research and identify new scientific avenues to further knowledge of cardio-oncology . The importance of big data in the area of cardio-oncology has also been highlighted . However, to the best of our knowledge, there is no database for molecular features specifically involved in onco-cardiology/cardio-oncology. Due to all of this, the proposal and main goal of this paper is to create a database whose entities are the key concepts used in studies concerning onco-cardiology and whose relationship reflects the real molecular interactions between them. Apart from the database itself, mechanisms should be provided to allow easy interaction with it by biomedical practitioners. This motivates the need to implement a user-friendly interface and to provide complete accessibility through the Internet at all times. Overview The aim is to create a database that organizes and makes available concepts and knowledge related to molecular information on onco-cardiology revealed by human studies which have been reported in original articles published to date . The first step to accomplish this objective was the design of the database itself; namely, which concepts appear in all studies, what their data types and allowed values are and how they are related to each other. This information was used to choose the entities and relationships of the database. The second step involves human intervention in the form of a knowledgeable reader who can identify in each medical paper the chosen concepts (see part 1 of to access the referenced papers) and take note of them as fields in a key-value table. The said table will be used to feed the database. Possible errors in curation, especially those related to typing mistakes or the use of nonstandard terms and/or invalid values, should be, and were, automatically reported and corrected before the acceptance of the data using text analysis programs. The third step was to identify which queries are interesting and to write them in a formal language, structured query language (SQL) in this case. For users not familiar with SQL, a number of ‘predefined queries’, were provided. Also, a graphical representation of the diagram of tables is provided to allow expert users to write ‘advanced queries’ (any arbitrary query) in SQL. It is important to point out that few biological databases allow such unrestricted access. In any case, the users can contact us if their question is not covered and we will include a predefined query that answers it. The last step was to build a simple user-friendly graphical interface for both the predefined and advanced queries, which would allow the user to access through the Internet, make his/her queries and obtain the results. The results must be provided in a form suitable for two purposes: automatic processing by machines and visual human-understandable representation. The formats chosen for that were CSV (comma-separated values to be read by a spreadsheet) and HTML (Hypertext Markup Language to be shown in a normal web browser), respectively. Data collection The data correspond to all the molecular target information that exists to date on the relationship between cancer and CVDs that has been validated in patients over 18 years of age. The strategy used to obtain the primary information was a systematic search of the literature, according to the PRISMA statement, using four search strategies in Medline, Web of Science, Scopus and Embase, which are detailed in the table at the end of part 1 of , ‘References of the papers used to populate the OncoCardio database with their accessible URLs and search strategies used.’ The inclusion criteria of the articles were original publications, scientific research and clinical studies focused on the molecular relationship between cancer and CVDs in adult patients (>18 years of age). Those studies that do not include molecular features (protein, DNA, RNA, methylation, etc.) or those in which the molecular features were evaluated in animal models, cell lines, and xenograft, among others, not validated directly in human samples, were excluded. Non-original research, grey literature, clinical trial protocols, reviews, consensus documents, short communications, opinions papers, letters, posters and conference abstracts were also excluded. A subselection of this information was published in a scoping review (see . In that previous work, exploratory research was carried out on the available evidence on novel molecular biomarkers associated with cardiotoxicity in the adult population undergoing cancer therapy. A total of 42 studies were included, and the data were collected in a spreadsheet and summarized in the publication. In the present work, the relationship between cancer and CVDs was studied in a broader sense (not only under cardiotoxicity), and the information was stored in a database. Therefore, the information contained in the selected articles was curated, putting in consensus the names of the genes, including their genomics position, an official symbol for human, synonyms, official identifiers, drug codes and pathology codes. This data curation was carried out by expert judgement, and if the information could not be normalized, the paper was discarded. Of the 91 selected papers, 7 were discarded for this reason, while articles that studied metabolites, lipids and other molecules that are not possible to assign easily/directly to an area of the genome were discarded in the full-text reading phase. Database organization The organization of the database was designed using variation as a key concept; variation in this context means ‘a change in amount or level’ or ‘a change of organization or arrangement’. This is because these are the molecular units on which there can be a molecular relationship between cancer and CVDs. This ‘Variation’ object can be a finite number of types (‘VariationType’), which refers to its role in biological processes. Currently, 10 types are admitted, represented in the database under the same name: SNV (single-nucleotide variant), SNP (single-nucleotide polymorphism), InDel (insertion and deletion), CNV (copy number variation), abundance proteins, differential expression, DNA methylation, RNA methylation, histone modifications and post-translational modification. Others can be added if needed. This variation, in turn, may or may not belong to a gene (object ‘Gene’); when it does not, it is intergenic and maps to the closest gene. The gene is identified by its identifiers (namely, Entrez and Ensembl), synonyms, and transcripts. On the other hand, a variation has a specific position in the genome (object ‘GenomeFullPosition’). Also, it may be used as a biomarker in a panel (object ‘PanelKitName’). Furthermore, this variation must have been reported as related to cancer and CVDs in at least one reported study (object ‘Study’). This study has specific fields that define the clinical context in which the variation is associated with both pathologies, including statistics values, abundance (e.g. fold change) and phenotype characteristics (see ‘Tables corresponding to entities’ and ). The relationship between these diseases can be approached from the risk factors that cancer and CVDs have in common, from the possibility that one can be a risk factor for the other one and from the cardiotoxicity that cancer therapies produced in the patients. This is addressed in the database using a combination of fields such as ‘CarditoxicityAppears’ (which can be null when corresponding to another type of association), ‘IdPathologyCD10’ (which can refer to cancer or CVDs) or IdCardiovascularSymptom (to describe the symptoms). See part 2 of for a full description. Knowledge extraction and curation To fill the database with the information extracted from the medical literature, the biomedical people read the papers and fill in a form for each one (in our case simply as cells on a spreadsheet) which is exported in text format and read by programs that process it and automatically generate the SQL sentences. The said programs were written in Perl (Practical Extraction and Report Language). The values of the numerous fields are checked first to verify that they are of the expected type (i.e. strings of characters, integer numbers and floating point numbers in a certain range) and also to be sure that they take a legal value, i.e. if they belong to a finite set, the value is one of its elements. This is done, e.g., to check for the validity of the code of a drug or of a pathology and so on. Furthermore, very large sets of values must be incorporated into the tables. For instance, the table ‘Gene’ contains the Entrez identifier of every gene in the human genome. The same thing occurs with every possible drug classified by the Anatomical Therapeutic Chemical Classification (ATC) , and every disease included in the International Classification of Diseases, ver. 10 (ICD10) . It can be argued that only a small fraction of all the genes, drugs or diseases will appear in the field of onco-cardiology, but we have decided to incorporate the whole list. This has not been a problem since this incorporation was done automatically with Perl macros and the size of the complete database turned out to be well below the capacity manageable by current computers. Moreover, in this way the filled tables can be useful for databases in other biomedical areas. Curation of manually extracted data by automatic or semi-automatic means is relatively common in these kinds of applications; other systems that mention this or similar approaches are Datanator or MarkerDB . Notice that special care has been taken to use reliable sources of information and standardized lists, such as the ATC or ICD10 (see section 1 of part 2 of ). Software architecture of the system The whole system was composed at the software level by an http microserver written in JavaScript and running with node.js . Users can provide an email address that is not permanently stored and is used exclusively to send the results of the queries to the user in CSV and/or HTML. Nevertheless, providing an email address is not compulsory: a user can log in anonymously simply to test, interact with the database and visualize and download the results. The database server runs in MariaDB. Database internal users have been created which have read-only access to the database tables, so even with access to an SQL console, they cannot alter its integrity. A program written in Java functions as the graphical user interface (GUI) to the database and it is accessed through a Guacamole server that opens a session for each user. The user can also access the web of the ATC and the ICD10 to consult drug or pathology codes. A daemon written in Perl orchestrates the former pieces, calling them to be run whenever needed. The system is currently fully functional and can be found at https://biodb.uv.es/oncocardio . The aim is to create a database that organizes and makes available concepts and knowledge related to molecular information on onco-cardiology revealed by human studies which have been reported in original articles published to date . The first step to accomplish this objective was the design of the database itself; namely, which concepts appear in all studies, what their data types and allowed values are and how they are related to each other. This information was used to choose the entities and relationships of the database. The second step involves human intervention in the form of a knowledgeable reader who can identify in each medical paper the chosen concepts (see part 1 of to access the referenced papers) and take note of them as fields in a key-value table. The said table will be used to feed the database. Possible errors in curation, especially those related to typing mistakes or the use of nonstandard terms and/or invalid values, should be, and were, automatically reported and corrected before the acceptance of the data using text analysis programs. The third step was to identify which queries are interesting and to write them in a formal language, structured query language (SQL) in this case. For users not familiar with SQL, a number of ‘predefined queries’, were provided. Also, a graphical representation of the diagram of tables is provided to allow expert users to write ‘advanced queries’ (any arbitrary query) in SQL. It is important to point out that few biological databases allow such unrestricted access. In any case, the users can contact us if their question is not covered and we will include a predefined query that answers it. The last step was to build a simple user-friendly graphical interface for both the predefined and advanced queries, which would allow the user to access through the Internet, make his/her queries and obtain the results. The results must be provided in a form suitable for two purposes: automatic processing by machines and visual human-understandable representation. The formats chosen for that were CSV (comma-separated values to be read by a spreadsheet) and HTML (Hypertext Markup Language to be shown in a normal web browser), respectively. The data correspond to all the molecular target information that exists to date on the relationship between cancer and CVDs that has been validated in patients over 18 years of age. The strategy used to obtain the primary information was a systematic search of the literature, according to the PRISMA statement, using four search strategies in Medline, Web of Science, Scopus and Embase, which are detailed in the table at the end of part 1 of , ‘References of the papers used to populate the OncoCardio database with their accessible URLs and search strategies used.’ The inclusion criteria of the articles were original publications, scientific research and clinical studies focused on the molecular relationship between cancer and CVDs in adult patients (>18 years of age). Those studies that do not include molecular features (protein, DNA, RNA, methylation, etc.) or those in which the molecular features were evaluated in animal models, cell lines, and xenograft, among others, not validated directly in human samples, were excluded. Non-original research, grey literature, clinical trial protocols, reviews, consensus documents, short communications, opinions papers, letters, posters and conference abstracts were also excluded. A subselection of this information was published in a scoping review (see . In that previous work, exploratory research was carried out on the available evidence on novel molecular biomarkers associated with cardiotoxicity in the adult population undergoing cancer therapy. A total of 42 studies were included, and the data were collected in a spreadsheet and summarized in the publication. In the present work, the relationship between cancer and CVDs was studied in a broader sense (not only under cardiotoxicity), and the information was stored in a database. Therefore, the information contained in the selected articles was curated, putting in consensus the names of the genes, including their genomics position, an official symbol for human, synonyms, official identifiers, drug codes and pathology codes. This data curation was carried out by expert judgement, and if the information could not be normalized, the paper was discarded. Of the 91 selected papers, 7 were discarded for this reason, while articles that studied metabolites, lipids and other molecules that are not possible to assign easily/directly to an area of the genome were discarded in the full-text reading phase. The organization of the database was designed using variation as a key concept; variation in this context means ‘a change in amount or level’ or ‘a change of organization or arrangement’. This is because these are the molecular units on which there can be a molecular relationship between cancer and CVDs. This ‘Variation’ object can be a finite number of types (‘VariationType’), which refers to its role in biological processes. Currently, 10 types are admitted, represented in the database under the same name: SNV (single-nucleotide variant), SNP (single-nucleotide polymorphism), InDel (insertion and deletion), CNV (copy number variation), abundance proteins, differential expression, DNA methylation, RNA methylation, histone modifications and post-translational modification. Others can be added if needed. This variation, in turn, may or may not belong to a gene (object ‘Gene’); when it does not, it is intergenic and maps to the closest gene. The gene is identified by its identifiers (namely, Entrez and Ensembl), synonyms, and transcripts. On the other hand, a variation has a specific position in the genome (object ‘GenomeFullPosition’). Also, it may be used as a biomarker in a panel (object ‘PanelKitName’). Furthermore, this variation must have been reported as related to cancer and CVDs in at least one reported study (object ‘Study’). This study has specific fields that define the clinical context in which the variation is associated with both pathologies, including statistics values, abundance (e.g. fold change) and phenotype characteristics (see ‘Tables corresponding to entities’ and ). The relationship between these diseases can be approached from the risk factors that cancer and CVDs have in common, from the possibility that one can be a risk factor for the other one and from the cardiotoxicity that cancer therapies produced in the patients. This is addressed in the database using a combination of fields such as ‘CarditoxicityAppears’ (which can be null when corresponding to another type of association), ‘IdPathologyCD10’ (which can refer to cancer or CVDs) or IdCardiovascularSymptom (to describe the symptoms). See part 2 of for a full description. To fill the database with the information extracted from the medical literature, the biomedical people read the papers and fill in a form for each one (in our case simply as cells on a spreadsheet) which is exported in text format and read by programs that process it and automatically generate the SQL sentences. The said programs were written in Perl (Practical Extraction and Report Language). The values of the numerous fields are checked first to verify that they are of the expected type (i.e. strings of characters, integer numbers and floating point numbers in a certain range) and also to be sure that they take a legal value, i.e. if they belong to a finite set, the value is one of its elements. This is done, e.g., to check for the validity of the code of a drug or of a pathology and so on. Furthermore, very large sets of values must be incorporated into the tables. For instance, the table ‘Gene’ contains the Entrez identifier of every gene in the human genome. The same thing occurs with every possible drug classified by the Anatomical Therapeutic Chemical Classification (ATC) , and every disease included in the International Classification of Diseases, ver. 10 (ICD10) . It can be argued that only a small fraction of all the genes, drugs or diseases will appear in the field of onco-cardiology, but we have decided to incorporate the whole list. This has not been a problem since this incorporation was done automatically with Perl macros and the size of the complete database turned out to be well below the capacity manageable by current computers. Moreover, in this way the filled tables can be useful for databases in other biomedical areas. Curation of manually extracted data by automatic or semi-automatic means is relatively common in these kinds of applications; other systems that mention this or similar approaches are Datanator or MarkerDB . Notice that special care has been taken to use reliable sources of information and standardized lists, such as the ATC or ICD10 (see section 1 of part 2 of ). The whole system was composed at the software level by an http microserver written in JavaScript and running with node.js . Users can provide an email address that is not permanently stored and is used exclusively to send the results of the queries to the user in CSV and/or HTML. Nevertheless, providing an email address is not compulsory: a user can log in anonymously simply to test, interact with the database and visualize and download the results. The database server runs in MariaDB. Database internal users have been created which have read-only access to the database tables, so even with access to an SQL console, they cannot alter its integrity. A program written in Java functions as the graphical user interface (GUI) to the database and it is accessed through a Guacamole server that opens a session for each user. The user can also access the web of the ATC and the ICD10 to consult drug or pathology codes. A daemon written in Perl orchestrates the former pieces, calling them to be run whenever needed. The system is currently fully functional and can be found at https://biodb.uv.es/oncocardio . Graphical interface and remote access design The tasks the user can execute are the following: Perform a predefined query whose input is introduced through a form and whose result is shown on the screen and also temporarily stored to be sent by an email or downloaded at the end of the session. Perform an advanced query, which is freely written in SQL into a text input area. The results are shown and stored, in the same way as it was done for predefined queries. Write his/her query in plain English that will be sent to us, as explained in the section ‘Description and examples of predefined queries’. Report the bibliographic reference of a new publication in this field whose results their authors would like to see incorporated into OncoCardio. Leave the session. At that point, the user is requested to choose the format or formats (CSV, HTML or both) for the results to be either sent by an email or downloaded in the case of anonymous log-in. Additionally and at any time, the user can get help through a menu at the upper side of each window in the form of a text and graphical diagram of the tables and fields in the database. Also, he/she can connect to the websites of ATC and ICD10 to find or verify the drug and disease codes, respectively. A complete demonstration of the use of the system is shown in the video included as Database statistics The statistics information has been directly obtained from the data base (DB) itself through appropriate SQL queries (see part 3 of for full statistics). A total of 93 unique Entrez identifiers were reported from all 61 548 possible identifiers. The most frequent were Entrez 4879 ( NPPB ), 7139 ( TNNT2 ) and 7137 ( TNNI3 ), but most identifiers appeared just once. With respect to variations, 107 different variations appear in at least one study. The most frequent ones were variations in the abundance of proteins encoded by the genes already mentioned . Regarding the types of variations, all those reported to date fall into the categories of SNVs, differential expression and variation in protein abundance, being studied mainly in blood samples. The total number of diseases stored with their corresponding ICD10 codes is 46 654. Out of them, 23 are reported in at least one study. Their ordering by the number of occurrences shows that the most frequent pathology is a malignant neoplasm of breast (ICD10 code: C50), which appears 122 times . Six different basic therapies (chemotherapy, hormonal therapy, immunotherapy, radiotherapy, surgery and targeted therapy) have been considered since no others are mentioned in any study, and also any combination of two or more of them is in principle possible, which generates 64 possibilities. From these 64, only 7 different possibilities appear in the studies. The most frequent is chemotherapy alone, which was used 144 times, followed by radiotherapy with 39 occurrences. The most common combined therapy is chemotherapy+radiotherapy in the fourth place with 15 occurrences. The total number of drugs stored in the DB with their corresponding ATC code is 6460. Out of them, 259 were used in any study. This number is greater than the number of studies since the prescribed treatment in many studies uses two or more drugs. In fact, only 26 of them are different. The most frequently used drug is doxorubicin (ATC code: L01DB01) with 58 occurrences, followed by cyclophosphamide (ATC code: L01AA01) with 35 occurrences. Finally, there are eight drugs mentioned in only one study each. The total number of different cardiovascular symptoms reported in the studies is 42. Ordering by the frequency of appearance shows that the decline in left ventricular ejection fraction (LVEF) is the most common with 125 occurrences, followed by heart failure with 20 occurrences. On the other hand, 15 symptoms are mentioned only once each. Description and examples of predefined queries The database contains a collection of (currently) seven predefined searches (Queries 1–7) that answer common questions. Among them, only one field is left free and the user is requested to provide a value. The predefined queries are as follows (values in typewriter font correspond to their denomination in the tables (see and part 3 of ). Input: a pathology code ( IdPathologyCD10 ). Output: all variations (as IdVariation ) associated to the input pathology and, for each of these variations, all its associated pathologies. Input: a variation name ( IdVariation ). Output: all pathologies (as IdPathologyCD10 ) associated with that variation and, for each of them, all studies (as DOIReference ) and, for each study, all symptoms (if any) associated with it (as IdCardiovascularSymptom ). Input: a variation ( IdVariation ). Output: all panel kits (as IdPanelKitName ) that can test the expression of the variation. Input: a variation ( IdVariation ). Output: all drugs (as IdDrugATC ) which have been used with patients that exhibit the variation. Input: a pathology name ( IdPathologyCD10 ). Output: all variations (as IdVariation ) detected in any study associated with that pathology and, for each of them, the gene (as IdGeneSymbol ) in which that variation occurs and the number of studies that refer to the found association. Input: a gene ( IdGeneSymbol ). Output: all variations (as IdVariation ) that occur in that gene and, for each of them, all the drugs (as IdDrugATC ) that are supposed to modify the variation. Input: a gene ( IdGeneSymbol ). Output: all variations (as IdVariation ) that occur in that gene and, for each of them, all the panel kits (as IdPanelKitName ) that can test the expression of that variation. For example, the user might want to know what onco-cardiology genes have been identified in lung cancer and how many studies have reported said association? This can be answered using Query 5, giving, as input the ICD10-code of the pathology. In this example, we use C34 (malignant neoplasm of bronchus and lung). The result is , which shows all the reported genes and the number of studies. Compound questions are also included, such as which pathology is associated with a molecular feature, which cardiovascular symptoms are associated with it and which papers report said association? This can be answered in Query 2. The input in this case is a variation; for example, let us use myeloperoxidase (MPO). MPO was always reported in breast cancer but associated with three cardiovascular symptoms in two different studies, as shown in . The list of predefined queries is obviously not exhaustive and can be augmented in the future if the authors receive feedback from the users concerning what they would like to get from the database. To do so, the program provides a specific item where the users can write their queries, in plain English, which are sent to us; we will do our best to translate them into SQL and even incorporate them as predefined questions if they prove to be of general interest. Examples of advanced queries Often a researcher’s question cannot be answered using any of the seven predefined queries. Nevertheless, to provide, from the beginning, a way to make any query, knowledgeable users can access a SQL console where they can write any unrestricted query. Since all the contents of the database come from public information, and no personal or sensitive data are stored, this does not represent a legal or privacy problem. Also, since users are allowed to access the database in read-only mode, integrity is not a concern, either. Notice that very few biomedical databases allow this unrestricted access. Thus, new questions can be asked using advanced queries. For instance, how many types of cancer can be found in the database and how often do they appear? To do this, the input in the advanced query would be: SELECT patients_affected_by.IdPathologyCD10, Pathology.Name, COUNT(patients_affected_by.IdPathologyCD10) FROM patients_affected_by,Pathology WHERE patients_affected_by.IdPathologyCD10= Pathology.IdPathologyCD10 GROUP BY patients_affected_by.IdPathologyCD10 ORDER BY COUNT(patients_affected_by.IdPathologyCD10) DESC The output corresponds to the 23 different types of cancer with their ICD10 codes, names and frequencies with which they are reported. The five most frequent types are shown in . Another example of an advanced query is: what is the average percentage of cardiotoxicity reported in all the studies which have measured it? The answer is 14.4 % and to obtain this the following advanced query needs to be introduced: SELECT SUM(ObservedCardiotoxicityNumSubjects)/ (SUM(NumSubjects)) FROM Study WHERE ObservedCardiotoxicity IS NOT NULL; Notice that this result has been calculated weighting the cardiotoxicity percentages reported in each study proportionally to the number of subjects in it with respect to the total number of subjects. Another interesting question is: which are the top three variations in the database? To answer this, the following advanced query needs to be used: SELECT IdVariation, COUNT(IdVariation) FROM variation_studied_by GROUP BY IdVariation ORDER BY COUNT(IdVariation) DESC LIMIT 3; the most frequent being natriuretic peptide B (frequency = 20), troponin T (frequency = 15) and troponin I (frequency = 14). The tasks the user can execute are the following: Perform a predefined query whose input is introduced through a form and whose result is shown on the screen and also temporarily stored to be sent by an email or downloaded at the end of the session. Perform an advanced query, which is freely written in SQL into a text input area. The results are shown and stored, in the same way as it was done for predefined queries. Write his/her query in plain English that will be sent to us, as explained in the section ‘Description and examples of predefined queries’. Report the bibliographic reference of a new publication in this field whose results their authors would like to see incorporated into OncoCardio. Leave the session. At that point, the user is requested to choose the format or formats (CSV, HTML or both) for the results to be either sent by an email or downloaded in the case of anonymous log-in. Additionally and at any time, the user can get help through a menu at the upper side of each window in the form of a text and graphical diagram of the tables and fields in the database. Also, he/she can connect to the websites of ATC and ICD10 to find or verify the drug and disease codes, respectively. A complete demonstration of the use of the system is shown in the video included as The statistics information has been directly obtained from the data base (DB) itself through appropriate SQL queries (see part 3 of for full statistics). A total of 93 unique Entrez identifiers were reported from all 61 548 possible identifiers. The most frequent were Entrez 4879 ( NPPB ), 7139 ( TNNT2 ) and 7137 ( TNNI3 ), but most identifiers appeared just once. With respect to variations, 107 different variations appear in at least one study. The most frequent ones were variations in the abundance of proteins encoded by the genes already mentioned . Regarding the types of variations, all those reported to date fall into the categories of SNVs, differential expression and variation in protein abundance, being studied mainly in blood samples. The total number of diseases stored with their corresponding ICD10 codes is 46 654. Out of them, 23 are reported in at least one study. Their ordering by the number of occurrences shows that the most frequent pathology is a malignant neoplasm of breast (ICD10 code: C50), which appears 122 times . Six different basic therapies (chemotherapy, hormonal therapy, immunotherapy, radiotherapy, surgery and targeted therapy) have been considered since no others are mentioned in any study, and also any combination of two or more of them is in principle possible, which generates 64 possibilities. From these 64, only 7 different possibilities appear in the studies. The most frequent is chemotherapy alone, which was used 144 times, followed by radiotherapy with 39 occurrences. The most common combined therapy is chemotherapy+radiotherapy in the fourth place with 15 occurrences. The total number of drugs stored in the DB with their corresponding ATC code is 6460. Out of them, 259 were used in any study. This number is greater than the number of studies since the prescribed treatment in many studies uses two or more drugs. In fact, only 26 of them are different. The most frequently used drug is doxorubicin (ATC code: L01DB01) with 58 occurrences, followed by cyclophosphamide (ATC code: L01AA01) with 35 occurrences. Finally, there are eight drugs mentioned in only one study each. The total number of different cardiovascular symptoms reported in the studies is 42. Ordering by the frequency of appearance shows that the decline in left ventricular ejection fraction (LVEF) is the most common with 125 occurrences, followed by heart failure with 20 occurrences. On the other hand, 15 symptoms are mentioned only once each. The database contains a collection of (currently) seven predefined searches (Queries 1–7) that answer common questions. Among them, only one field is left free and the user is requested to provide a value. The predefined queries are as follows (values in typewriter font correspond to their denomination in the tables (see and part 3 of ). Input: a pathology code ( IdPathologyCD10 ). Output: all variations (as IdVariation ) associated to the input pathology and, for each of these variations, all its associated pathologies. Input: a variation name ( IdVariation ). Output: all pathologies (as IdPathologyCD10 ) associated with that variation and, for each of them, all studies (as DOIReference ) and, for each study, all symptoms (if any) associated with it (as IdCardiovascularSymptom ). Input: a variation ( IdVariation ). Output: all panel kits (as IdPanelKitName ) that can test the expression of the variation. Input: a variation ( IdVariation ). Output: all drugs (as IdDrugATC ) which have been used with patients that exhibit the variation. Input: a pathology name ( IdPathologyCD10 ). Output: all variations (as IdVariation ) detected in any study associated with that pathology and, for each of them, the gene (as IdGeneSymbol ) in which that variation occurs and the number of studies that refer to the found association. Input: a gene ( IdGeneSymbol ). Output: all variations (as IdVariation ) that occur in that gene and, for each of them, all the drugs (as IdDrugATC ) that are supposed to modify the variation. Input: a gene ( IdGeneSymbol ). Output: all variations (as IdVariation ) that occur in that gene and, for each of them, all the panel kits (as IdPanelKitName ) that can test the expression of that variation. For example, the user might want to know what onco-cardiology genes have been identified in lung cancer and how many studies have reported said association? This can be answered using Query 5, giving, as input the ICD10-code of the pathology. In this example, we use C34 (malignant neoplasm of bronchus and lung). The result is , which shows all the reported genes and the number of studies. Compound questions are also included, such as which pathology is associated with a molecular feature, which cardiovascular symptoms are associated with it and which papers report said association? This can be answered in Query 2. The input in this case is a variation; for example, let us use myeloperoxidase (MPO). MPO was always reported in breast cancer but associated with three cardiovascular symptoms in two different studies, as shown in . The list of predefined queries is obviously not exhaustive and can be augmented in the future if the authors receive feedback from the users concerning what they would like to get from the database. To do so, the program provides a specific item where the users can write their queries, in plain English, which are sent to us; we will do our best to translate them into SQL and even incorporate them as predefined questions if they prove to be of general interest. Often a researcher’s question cannot be answered using any of the seven predefined queries. Nevertheless, to provide, from the beginning, a way to make any query, knowledgeable users can access a SQL console where they can write any unrestricted query. Since all the contents of the database come from public information, and no personal or sensitive data are stored, this does not represent a legal or privacy problem. Also, since users are allowed to access the database in read-only mode, integrity is not a concern, either. Notice that very few biomedical databases allow this unrestricted access. Thus, new questions can be asked using advanced queries. For instance, how many types of cancer can be found in the database and how often do they appear? To do this, the input in the advanced query would be: SELECT patients_affected_by.IdPathologyCD10, Pathology.Name, COUNT(patients_affected_by.IdPathologyCD10) FROM patients_affected_by,Pathology WHERE patients_affected_by.IdPathologyCD10= Pathology.IdPathologyCD10 GROUP BY patients_affected_by.IdPathologyCD10 ORDER BY COUNT(patients_affected_by.IdPathologyCD10) DESC The output corresponds to the 23 different types of cancer with their ICD10 codes, names and frequencies with which they are reported. The five most frequent types are shown in . Another example of an advanced query is: what is the average percentage of cardiotoxicity reported in all the studies which have measured it? The answer is 14.4 % and to obtain this the following advanced query needs to be introduced: SELECT SUM(ObservedCardiotoxicityNumSubjects)/ (SUM(NumSubjects)) FROM Study WHERE ObservedCardiotoxicity IS NOT NULL; Notice that this result has been calculated weighting the cardiotoxicity percentages reported in each study proportionally to the number of subjects in it with respect to the total number of subjects. Another interesting question is: which are the top three variations in the database? To answer this, the following advanced query needs to be used: SELECT IdVariation, COUNT(IdVariation) FROM variation_studied_by GROUP BY IdVariation ORDER BY COUNT(IdVariation) DESC LIMIT 3; the most frequent being natriuretic peptide B (frequency = 20), troponin T (frequency = 15) and troponin I (frequency = 14). The area of onco-cardiology is in its early years of study regarding the molecular relationship between cancer and CVDs. This makes possible the collection of all the information in this regard early. The OncoCardioDB designed in this work allows storing all the molecular information obtained from patients in the area of onco-cardiology, incorporating new relevant studies on a regular basis. The global burden of cancer and CVDs continues to increase, and therefore the growing number of patients who survive cancer have an increasing risk of developing CVDs. This is why we think that the OncoCardio database will be an important contribution to identify molecular targets related to both diseases. Concrete applications that can be mentioned are in precision medicine, e.g. when designing new targeted drugs with fewer cardiotoxic effects, or as a repository of possible molecular biomarkers in onco-cardiology. The above comment is based on the fact that the OncoCardio database contains information on all the molecular targets (variations) identified to date associated with cancer that cause a cardiovascular effect, whether due to cardiotoxicity or other reasons. Another important aspect to be discussed is the usability of the system. It will be really useful only as long as the biomedical community finds it valuable and accessible. Although it is not especially aesthetic, the authors have made an effort to provide simple access and allow only one course of action at any time, so the interface is robust and can be extended or adapted very easily according to the user’s requirements. An important point is the addition of predefined queries on demand; due to the modular organization, this will require the alteration of only a restricted part of the GUI, which is a simple task. The generation of the SQL sentences to fill the DB in the case of the generic tables (those whose information comes from predefined lists or standards), especially gene-related ones, is relatively slow (it takes ~10 min on a normal modern computer) and its introduction into the database (the call to the generated SQL macro) is even slower (~20 min), but this only has to be done once and is valid forever. On the other hand, the generation of the sentences for the specific tables (namely, the studies and their related values such as observed symptoms and pathologies) and their loading into the database is almost instantaneous, so the addition of further studies to update the information is not a problem. Apart from this, probably the main drawback for complete user-friendliness is the need to introduce the entities such as genes, pathologies and drugs using their normalized designations exclusively (Entrez code, ATC and ICD10 codes), instead of common or normally accepted names. The use of fully standardized vocabularies is an almost universally acknowledged need since automatic information processing was introduced in medicine. Nevertheless, this is something that is not always accepted by the medical community enthusiastically. It is our opinion that an effort should be carried out by both parties (doctors and information processing experts) to reduce this gap. To that aim, a proposal for future work is provided in the ‘Future work’ section. Extendibility and reusability are important aspects to be mentioned, too. The GUI is a Java application in which the model interacts with a MariaDB server to make any query and return any result. This makes it possible to use it in any database application with few changes; e.g., more predefined queries can be added without much effort. On the other hand, the controller and the views have been devised for this application and even though they are a good basis for similar ones, a substantial part of their code would have to be changed. Biological knowledge is completely contained inside the database and to some extent can be reused. Tables containing general entities (genes, drugs and pathologies) can be exported separately and used as such for other medical databases. Reorganization of the database by adding fields, or even new tables, is possible, too, but may require rewriting of one or more predefined queries. Finally, the authentication and user-management parts are the most reusable components. They are completely independent of the GUI and the database; indeed, they can work with any other program where access is to be offered through the network with just a browser, even if it is not a user interface or a database. Future work The system is functional and awaits the interaction of interested users. Their suggestions and comments will be important to introduce improvements, but some of them are already planned, namely: In the case that the user does not know the ICD10 code of a pathology but only its name, a search through the ICD10 database by similar words or expressions will be provided. The similarity between character strings, including metrics, to show possible matches ordered by closeness is a habitual task in natural language processing, and several libraries aimed at this objective are available. The interface to look for pathology codes can be organized, too, based on a rational taxonomy of diseases, but this is already provided by the ICD10 web (see https://www.icd10data.com/ICD10CM/Codes ) which has been made accessible as part of the GUI of the system. A similar strategy can be applied to the ATC codes of drugs. In this case, the rational taxonomy is available at https://www.atccode.com/ and is also available in our system. A program option is currently implemented which allows authors of papers on onco-cardiology to inform us about their works so that the content can be incorporated into the database. It will also be possible to provide a form where the authors themselves can fill in the values of each of the relevant fields mentioned in their work. Such information will be curated in its form by the syntactic checks mentioned in the section ‘Knowledge extraction and curation’ and in its content by the database maintainers. With respect to the system’s organization and the experience acquired during its construction, it is likely that can be used to build similar systems in other biomedical areas. The system is sufficiently modular, and in particular, the organization of the user interface and the orchestration of all the parts under Guacamole are fully reusable. In a similar way, some of the tables of the database can be used without changes in applications that require these data (drug names/codes or information on genes and variations). The system is functional and awaits the interaction of interested users. Their suggestions and comments will be important to introduce improvements, but some of them are already planned, namely: In the case that the user does not know the ICD10 code of a pathology but only its name, a search through the ICD10 database by similar words or expressions will be provided. The similarity between character strings, including metrics, to show possible matches ordered by closeness is a habitual task in natural language processing, and several libraries aimed at this objective are available. The interface to look for pathology codes can be organized, too, based on a rational taxonomy of diseases, but this is already provided by the ICD10 web (see https://www.icd10data.com/ICD10CM/Codes ) which has been made accessible as part of the GUI of the system. A similar strategy can be applied to the ATC codes of drugs. In this case, the rational taxonomy is available at https://www.atccode.com/ and is also available in our system. A program option is currently implemented which allows authors of papers on onco-cardiology to inform us about their works so that the content can be incorporated into the database. It will also be possible to provide a form where the authors themselves can fill in the values of each of the relevant fields mentioned in their work. Such information will be curated in its form by the syntactic checks mentioned in the section ‘Knowledge extraction and curation’ and in its content by the database maintainers. With respect to the system’s organization and the experience acquired during its construction, it is likely that can be used to build similar systems in other biomedical areas. The system is sufficiently modular, and in particular, the organization of the user interface and the orchestration of all the parts under Guacamole are fully reusable. In a similar way, some of the tables of the database can be used without changes in applications that require these data (drug names/codes or information on genes and variations). baad029_Supp Click here for additional data file.
Psychosocial supports within pediatric nephrology practices: A pediatric nephrology research consortium survey	ae0efedc-37b6-4fd9-9b50-ee216db64feb	10168552	Internal Medicine[mh]	Pediatric nephrologists are responsible for providing complex and multifaceted medical care across diverse kidney disease groups in children. This is inclusive of acute kidney injury, chronic kidney disease (CKD), chronic kidney failure, congenital or genetic disorders with kidney involvement (e.g., polycystic kidney disease, congenital nephrotic syndrome, and hypo/dysplastic kidneys), glomerular disease, hypertension, kidney stones and urinary or urologic abnormalities. In addition to the breadth of diagnoses, the intricacies of disease management are multifaceted, including dietary and/or fluid modifications along with complex and strict medication regimens that when not followed closely may result in increased morbidity and mortality. Multisite research efforts, including the Chronic Kidney Disease in Children study (CKiD) have sought to better understand the course of mild to moderate impaired kidney function in CKD, including impacts on psychosocial and cognitive functioning . Additionally, groups such as the Pediatric Nephrology Research Consortium (PNRC) have been established to improve and promote high quality care across pediatric nephrology centers, centered on the shared goal of increasing understanding and treatment of pediatric kidney disease. Given the heterogeneity of kidney disease and the chronicity of these conditions, some with a lifelong course, the medical and psychosocial burden of care remains a challenge due to the impact of these disorders on psychosocial development and growth in children, as well as highly variable disease presentation, course, and severity. Psychosocial concerns for kidney patients and families Effects on psychological well-being have been reported across disease groups in pediatric nephrology. In addition to CKD, there are a range of psychosocial and cognitive consequences observed across other specific conditions with kidney involvement, including hypertension , prune belly syndrome , tuberous sclerosis complex , and nephrotic syndrome . The psychosocial and cognitive outcomes associated with CKD are most widely studied among the diverse kidney conditions in pediatric patients. Indeed, a disproportionately high number of children (ages 9–18) with CKD meet diagnostic criteria for depression . Patients with end stage kidney disease undergoing dialysis treatment report even greater rates of depression and anxiety, relative to both their peers without kidney disease and their peers who have pre-dialysis CKD . Over time, pediatric dialysis patients report worse emotional health that often persists into adulthood . Although medical advancements have led to improved long-term survival , the consequences of kidney disease sequelae and treatment in children clearly influence health-related quality-of-life (HRQOL). In a systematic review of pediatric dialysis patients, the impact on HRQOL was best understood across five themes . These included loss of control (reliance on others, machine dependence) restricted lifestyle (limited socialization), coping strategies (hopefulness towards transplant), managing treatment (adherence to dietary and fluid restrictions), and feeling different (the burden of perceived differences in physical appearance). Accordingly, both disease progression and medical treatments (e.g., dialysis) may affect children’s self-esteem, personal identity, independence, and perceived control or agency. Due to elevated concerns for mental health problems and decreased quality of life in children with end stage CKD, the Centers for Medicaid and Medicare Services (CMS) requirements now mandate regularly screening these patients for depression and quality of life. In addition to the effects on mental health, many patients with kidney diseases are also at increased risk of cognitive and academic dysfunction, which may be directly related to the disease process. For instance, the pathophysiologic effects of advanced uremia include impacts on brain metabolism and thereby cognitive functioning . However, kidney replacement treatments may also affect cognitive functions . Attending hemodialysis has a notable impact on limiting a child’s ability to be physically present in the school environment, which not only has implications for academic achievement , but also reduces opportunities for expected peer socialization and skill development. While some neurocognitive deficits may improve following kidney transplant, overall findings of children with CKD suggest lower intellectual functioning compared to those without CKD . Patterns of difficulties include challenges with executive functions including holding information and shifting attention, reduced visual and verbal memory, and lower metacognitive skills . With increasing awareness of the direct and indirect impacts of the medical and emotional burden that kidney disease can have on patient and family functioning, psychosocial professionals have emerged as important providers in kidney care. Increasingly, the importance of specialized training in nephrology for psychosocial professionals have been identified as imperative for maximizing support for patients and families. For example, the National Kidney Foundation Council of Nephrology Social Workers developed a Nephrology Social Worker Certification in 2009 , and the Society for Pediatric Psychology, a division of the American Psychological Association, began a special interest group with a focus in nephrology in 2018 . Specialty training opportunities as well as professional interest groups reflect the need for systematic and coordinated care of patients with high levels of disease complexity and psychosocial needs. The COVID-19 pandemic may amplify the psychosocial issues inherent in kidney disorders, as children with chronic medical conditions such as pediatric nephrology patients are most at risk for experiencing a mental health crisis today . The US Surgeon General recently documented children have increasing rates of depression, suicidal ideation and attempts, and more challenges in relation to social determinants of health (mental health, social services, food, housing, and caregiver health); the vulnerabilities within the pediatric nephrology population have similarly also increased . Impact on the pediatric nephrology workforce The perceived high complexity of patient needs, along with insufficient support for psychosocial services, may serve to dissuade medical residents from pursuing a career in pediatric nephrology . Unfortunately, concerns about the pediatric nephrology workforce have only escalated . Given the high complexity of care needs, it is possible that expansion of the psychosocial services available to meet holistic needs of patients and their families will reduce clinical burdens on nephrologists and lead to improvements in workforce growth. Indeed, given the complex needs of patients in the pediatric nephrology subspecialty, comprehensive disease management is best achieved through multidisciplinary care. Salerno, Weinstein, and Hanevold supported this notion, outlining distinct personnel needs of pediatric nephrology practices in addition to the nephrologist, including nurse specialists, dieticians, social workers, clinical administrators, psychologists, child life specialists, dialysis nurses, and renal transplant coordinators (for those with transplant programs). Through overlapping interests and unique contributions of each discipline, patients may experience a high level of integrated care that may also allay stressors experienced by individual providers. Indeed, adequate psychosocial support is a key recommendation in treatment for infants with kidney failure as well as in treatment of children with hypertension . Multidisciplinary care allows for management of comorbidities, focused on the patient’s unique needs, and has implications for improved clinical outcomes . Review of the nephrologist’s perception of psychosocial care in hemodialysis units revealed that the overwhelming majority (94%) shared the belief that patient outcomes improve with focus on psychosocial care . Additionally, while most (78%) nephrologists identified as participating in psychosocial initiatives, less than half (40%) lead these initiatives and very few (9%) received training in aspects of psychosocial issues, suggesting that while nephrologists report perceived benefit from interdisciplinary collaboration, the nephrologist and each psychosocial provider make unique contributions to patient care. However, the solution is not as simple as having increased access to psychosocial professionals, but also involves careful consideration of division infrastructure and provider support. For example, in many kidney care settings with additional support, social workers are the primary designated psychosocial provider. Hansen and colleagues documented the broad psychosocial support that social workers provide, ranging from psychoeducation, counseling for parents and families, and significant care coordination, such as linkage to community and educational resources. While these supports have been highlighted in the literature to be key factors to help address high rates of depression, lower quality of life, and family stressors associated with higher acuity kidney needs, the burden of who is responsible for addressing these psychosocial needs carries a cost . There is a notable body of research documenting high levels of burnout among social workers and nurses in dialysis centers . Burnout may be hastened by the complex patient needs, high patient volume, and the vast responsibilities largely shouldered by social workers in pediatric nephrology centers. Identified barriers to providing highly valued psychosocial care to pediatric nephrology patients include lack of access to psychosocial healthcare providers, high administrative demands, lack of training in psychosocial care delivery, and empathy fatigue . Therefore, methodical and deliberate integration and inclusion of psychosocial providers, given value of integrated care to patients as well as reduction of burden on other providers that may help decrease burnout and address ongoing workforce shortages. While a multidisciplinary model has been detailed as the standard of care within pediatric nephrology , little is known about the implementation of these guidelines into pediatric nephrology practices in the United States. In the United Kingdom, review of their renal psychosocial workforce showed great variability in service provision, as well as differences in staffing models including disparate availability of psychosocial team members (e.g., social worker, psychologist, counselor) between centers . To address this gap, we surveyed pediatric nephrology centers belonging to the PNRC in the United States, with the goal of gaining a better understanding of the present availability of psychosocial services, as well as elucidate present disparities in access to psychosocial care. Effects on psychological well-being have been reported across disease groups in pediatric nephrology. In addition to CKD, there are a range of psychosocial and cognitive consequences observed across other specific conditions with kidney involvement, including hypertension , prune belly syndrome , tuberous sclerosis complex , and nephrotic syndrome . The psychosocial and cognitive outcomes associated with CKD are most widely studied among the diverse kidney conditions in pediatric patients. Indeed, a disproportionately high number of children (ages 9–18) with CKD meet diagnostic criteria for depression . Patients with end stage kidney disease undergoing dialysis treatment report even greater rates of depression and anxiety, relative to both their peers without kidney disease and their peers who have pre-dialysis CKD . Over time, pediatric dialysis patients report worse emotional health that often persists into adulthood . Although medical advancements have led to improved long-term survival , the consequences of kidney disease sequelae and treatment in children clearly influence health-related quality-of-life (HRQOL). In a systematic review of pediatric dialysis patients, the impact on HRQOL was best understood across five themes . These included loss of control (reliance on others, machine dependence) restricted lifestyle (limited socialization), coping strategies (hopefulness towards transplant), managing treatment (adherence to dietary and fluid restrictions), and feeling different (the burden of perceived differences in physical appearance). Accordingly, both disease progression and medical treatments (e.g., dialysis) may affect children’s self-esteem, personal identity, independence, and perceived control or agency. Due to elevated concerns for mental health problems and decreased quality of life in children with end stage CKD, the Centers for Medicaid and Medicare Services (CMS) requirements now mandate regularly screening these patients for depression and quality of life. In addition to the effects on mental health, many patients with kidney diseases are also at increased risk of cognitive and academic dysfunction, which may be directly related to the disease process. For instance, the pathophysiologic effects of advanced uremia include impacts on brain metabolism and thereby cognitive functioning . However, kidney replacement treatments may also affect cognitive functions . Attending hemodialysis has a notable impact on limiting a child’s ability to be physically present in the school environment, which not only has implications for academic achievement , but also reduces opportunities for expected peer socialization and skill development. While some neurocognitive deficits may improve following kidney transplant, overall findings of children with CKD suggest lower intellectual functioning compared to those without CKD . Patterns of difficulties include challenges with executive functions including holding information and shifting attention, reduced visual and verbal memory, and lower metacognitive skills . With increasing awareness of the direct and indirect impacts of the medical and emotional burden that kidney disease can have on patient and family functioning, psychosocial professionals have emerged as important providers in kidney care. Increasingly, the importance of specialized training in nephrology for psychosocial professionals have been identified as imperative for maximizing support for patients and families. For example, the National Kidney Foundation Council of Nephrology Social Workers developed a Nephrology Social Worker Certification in 2009 , and the Society for Pediatric Psychology, a division of the American Psychological Association, began a special interest group with a focus in nephrology in 2018 . Specialty training opportunities as well as professional interest groups reflect the need for systematic and coordinated care of patients with high levels of disease complexity and psychosocial needs. The COVID-19 pandemic may amplify the psychosocial issues inherent in kidney disorders, as children with chronic medical conditions such as pediatric nephrology patients are most at risk for experiencing a mental health crisis today . The US Surgeon General recently documented children have increasing rates of depression, suicidal ideation and attempts, and more challenges in relation to social determinants of health (mental health, social services, food, housing, and caregiver health); the vulnerabilities within the pediatric nephrology population have similarly also increased . The perceived high complexity of patient needs, along with insufficient support for psychosocial services, may serve to dissuade medical residents from pursuing a career in pediatric nephrology . Unfortunately, concerns about the pediatric nephrology workforce have only escalated . Given the high complexity of care needs, it is possible that expansion of the psychosocial services available to meet holistic needs of patients and their families will reduce clinical burdens on nephrologists and lead to improvements in workforce growth. Indeed, given the complex needs of patients in the pediatric nephrology subspecialty, comprehensive disease management is best achieved through multidisciplinary care. Salerno, Weinstein, and Hanevold supported this notion, outlining distinct personnel needs of pediatric nephrology practices in addition to the nephrologist, including nurse specialists, dieticians, social workers, clinical administrators, psychologists, child life specialists, dialysis nurses, and renal transplant coordinators (for those with transplant programs). Through overlapping interests and unique contributions of each discipline, patients may experience a high level of integrated care that may also allay stressors experienced by individual providers. Indeed, adequate psychosocial support is a key recommendation in treatment for infants with kidney failure as well as in treatment of children with hypertension . Multidisciplinary care allows for management of comorbidities, focused on the patient’s unique needs, and has implications for improved clinical outcomes . Review of the nephrologist’s perception of psychosocial care in hemodialysis units revealed that the overwhelming majority (94%) shared the belief that patient outcomes improve with focus on psychosocial care . Additionally, while most (78%) nephrologists identified as participating in psychosocial initiatives, less than half (40%) lead these initiatives and very few (9%) received training in aspects of psychosocial issues, suggesting that while nephrologists report perceived benefit from interdisciplinary collaboration, the nephrologist and each psychosocial provider make unique contributions to patient care. However, the solution is not as simple as having increased access to psychosocial professionals, but also involves careful consideration of division infrastructure and provider support. For example, in many kidney care settings with additional support, social workers are the primary designated psychosocial provider. Hansen and colleagues documented the broad psychosocial support that social workers provide, ranging from psychoeducation, counseling for parents and families, and significant care coordination, such as linkage to community and educational resources. While these supports have been highlighted in the literature to be key factors to help address high rates of depression, lower quality of life, and family stressors associated with higher acuity kidney needs, the burden of who is responsible for addressing these psychosocial needs carries a cost . There is a notable body of research documenting high levels of burnout among social workers and nurses in dialysis centers . Burnout may be hastened by the complex patient needs, high patient volume, and the vast responsibilities largely shouldered by social workers in pediatric nephrology centers. Identified barriers to providing highly valued psychosocial care to pediatric nephrology patients include lack of access to psychosocial healthcare providers, high administrative demands, lack of training in psychosocial care delivery, and empathy fatigue . Therefore, methodical and deliberate integration and inclusion of psychosocial providers, given value of integrated care to patients as well as reduction of burden on other providers that may help decrease burnout and address ongoing workforce shortages. While a multidisciplinary model has been detailed as the standard of care within pediatric nephrology , little is known about the implementation of these guidelines into pediatric nephrology practices in the United States. In the United Kingdom, review of their renal psychosocial workforce showed great variability in service provision, as well as differences in staffing models including disparate availability of psychosocial team members (e.g., social worker, psychologist, counselor) between centers . To address this gap, we surveyed pediatric nephrology centers belonging to the PNRC in the United States, with the goal of gaining a better understanding of the present availability of psychosocial services, as well as elucidate present disparities in access to psychosocial care. Procedures This study was deemed to be exempt by sponsoring institutions’ Human Research Protection Programs. Participants were supplied with information about the study, so their choice to continue was implied consent. The PNRC is a voluntary consortium of academic pediatric nephrology practices (N = 90) representing more than 60% of pediatric nephrology centers in the US, including the majority of large US centers. Each registered member of the PNRC was contacted via email requesting completion of an electronic survey through a REDCap link . The email recommended that participants complete the survey at a division or section meeting to ensure only one survey was submitted per PNRC affiliated site. Participants received up to three reminders to complete the survey. Of the responses, five PNRC centers completed the survey twice; the first response submitted was deleted from dataset. Authors also collected data from respective program websites for all PNRC centers that did not respond to determine representative nature of responses collected. Data were analyzed using IBM SPSS Version 26. Due to the exploratory nature of the study, statistical analyses were largely descriptive, with appropriate test statistics used to determine significant differences (i.e., Chi-Square/T-Tests). Participants Respondents represented 49 unique PNRC affiliated pediatric nephrology practices across the United States of America, encompassing 56% of PNRC membership in North America (see for geographic distribution). When comparing responding versus non-responding PNRC centers, there were no significant differences in size of centers, number of physicians (M.D. or D.O. providers), or midlevel professionals (N.P. or P.A. providers). Non-responding centers were less likely to have a renal dialysis unit (RDU; 70.2% vs. 92.1%; X 2 (1) = 6.294, p = .012). There were no significant differences in percent of centers that conducted renal transplants between responding and non-responding PNRC centers. Of the top 40 pediatric nephrology centers ranked by the U.S. News and World Report’s Best Children’s Hospitals in 2021 , 34 centers participate in the PNRC; 26 of those centers provided survey responses in the current study. Measures The survey was developed by the authors and revised based on recommendations made by the PNRC study approval committee. The survey included three parts; the first part collected institution characteristics (i.e., PNRC institution, number of providers, dialysis offered, number of chronic dialysis patients treated, type of kidney conditions treated, and whether transplants are performed). The second part assessed availability of psychosocial services (e.g., assigned to division, available for consultation, assigned to division, or not available), use of behavioral screeners, and the allied health professional/psychosocial services that are available to patients receiving chronic dialysis. The third part related to satisfaction, importance, and effectiveness of psychosocial services to meet the center’s patient population needs. This study was deemed to be exempt by sponsoring institutions’ Human Research Protection Programs. Participants were supplied with information about the study, so their choice to continue was implied consent. The PNRC is a voluntary consortium of academic pediatric nephrology practices (N = 90) representing more than 60% of pediatric nephrology centers in the US, including the majority of large US centers. Each registered member of the PNRC was contacted via email requesting completion of an electronic survey through a REDCap link . The email recommended that participants complete the survey at a division or section meeting to ensure only one survey was submitted per PNRC affiliated site. Participants received up to three reminders to complete the survey. Of the responses, five PNRC centers completed the survey twice; the first response submitted was deleted from dataset. Authors also collected data from respective program websites for all PNRC centers that did not respond to determine representative nature of responses collected. Data were analyzed using IBM SPSS Version 26. Due to the exploratory nature of the study, statistical analyses were largely descriptive, with appropriate test statistics used to determine significant differences (i.e., Chi-Square/T-Tests). Respondents represented 49 unique PNRC affiliated pediatric nephrology practices across the United States of America, encompassing 56% of PNRC membership in North America (see for geographic distribution). When comparing responding versus non-responding PNRC centers, there were no significant differences in size of centers, number of physicians (M.D. or D.O. providers), or midlevel professionals (N.P. or P.A. providers). Non-responding centers were less likely to have a renal dialysis unit (RDU; 70.2% vs. 92.1%; X 2 (1) = 6.294, p = .012). There were no significant differences in percent of centers that conducted renal transplants between responding and non-responding PNRC centers. Of the top 40 pediatric nephrology centers ranked by the U.S. News and World Report’s Best Children’s Hospitals in 2021 , 34 centers participate in the PNRC; 26 of those centers provided survey responses in the current study. The survey was developed by the authors and revised based on recommendations made by the PNRC study approval committee. The survey included three parts; the first part collected institution characteristics (i.e., PNRC institution, number of providers, dialysis offered, number of chronic dialysis patients treated, type of kidney conditions treated, and whether transplants are performed). The second part assessed availability of psychosocial services (e.g., assigned to division, available for consultation, assigned to division, or not available), use of behavioral screeners, and the allied health professional/psychosocial services that are available to patients receiving chronic dialysis. The third part related to satisfaction, importance, and effectiveness of psychosocial services to meet the center’s patient population needs. Institution characteristics The number of nephrology providers per responding center was highly variable (physicians ranged from 1 to 21, M = 6.12; midlevel professionals ranged from 0 to 10, M = 1.60). The number of attending nephrologists per institution was used to categorize centers by size for purposes of statistical analysis. Corresponding to 25 th , 50 th , and 75 th percentile values, small centers are classified as those having ≤ 3 nephrologists, medium centers 4–7 nephrologists, and large centers ≥ 8 nephrologists. For additional characteristics of centers by size, see . All centers indicated treating a wide variety of kidney conditions; however, there was pronounced variability in treatment of metabolic bone disease, with five centers (2 small, 1 medium, and 2 large) not treating these conditions within the nephrology division. Most centers performed kidney transplants, with the likelihood of this increasing with the size of the center X 2 (2) = 7.964, p = .019. The total number of providers at a center (physicians plus midlevel professionals) was strongly associated with the number of chronic dialysis patients cared for at the center, r = .738, p < .001. Psychosocial services Availability of psychosocial providers varied based on nephrology division size, such that as nephrology center size increased, access to various psychosocial providers increased. See for availability of psychosocial services to entire nephrology divisions, as well as the variety of psychosocial services to renal dialysis units specifically. Further analysis indicated that the larger the center, the more likely it was to have a dedicated social worker assigned to the division [ X 2 (2) = 11.688, p = .003] and the same was true for pediatric psychology [ X 2 (2) = 12.434, p = .002]. Likewise, responding centers that were ranked in top 40 of U.S. News and World Reports were more likely to have a pediatric psychologist specifically assigned to their nephrology division than those that were not [ X 2 (1) = 5.841, p = .016], but were not more likely to have an assigned social worker [ X 2 (1) = 2.683, p = .101]. Moreover, centers were more likely to have a social worker [ X 2 (1) = 5.544, p = .019] assigned to their division if they had a renal dialysis unit at their institution, but the same was not true for pediatric psychologists assigned to division [ X 2 (2) = 2.941, p = .086]. Available services in the RDU are also reported in and did not significantly differ by center size. However, among centers with an RDU, those that were ranked in top 40 of U.S. News and World Reports were more likely to have school teachers [ X 2 (1) = 6.310, p = .012] and therapeutic recreation specialists [ X 2 (1) = 4.342, p = .037] as part of the psychosocial care team in their RDU than those that were not. The majority of respondents (60.4%) reported that health screens are performed as a part of standard of care. Further analyses indicated that the majority centers administer behavioral health screens to chronic dialysis patients (57.1% of centers administer to hemodialysis patients; 53.1% to peritoneal dialysis patients) and renal transplant patients (57.1%). Interestingly, centers ranked in the top 40 of pediatric nephrology centers by the U.S. News and World Report were more likely to administer behavioral health screens then other centers, X 2 (2) = 8.437, p = .015. Relative to psychology supports specifically, respondents were asked to estimate how many of their patients were followed by psychology in some capacity. Although several respondents did not know (20.4%), the majority (63.3%; n = 31) reported that only 0–25% of their patients were followed by psychology. Perception of psychosocial services Respondents were further queried as to the perceived importance of psychosocial supports in helping provide care in their nephrology division. Responses were largely favorable, with all indicating that psychosocial services were at least “somewhat important.” As demonstrated in , perceived importance increases with center size, r = .405, p = .005. In addition to being asked to estimate how many of their current patients are followed by psychology, respondents were also asked how many of their patients would be followed by psychology in an “ ideal world . ” Although a small number (4.1%) indicated they did not know, the majority of respondents (55.1%) reported that ideally more than half of their patients would be followed by psychology. Although no respondents reported that ≥75% of their patients are currently followed by psychology, nearly a quarter of respondents (24.5%, n = 12) reported that in an ideal world 75% would indeed be seen by a psychologist. displays the reported current percentage of patients followed by psychology versus the percent of patients that would be followed by psychology in an ideal world for each responding center, excluding the two respondents who did not provide a specific percentage of patients in their ideal world. We also asked providers to rate their satisfaction with the psychologist’s communication, specifically regarding how the patient’s emotional and behavioral health may affect their medical condition and/or treatment. Of the responses, 18.4% said that they do receive satisfactory communication, 65.3% said they sometimes receive satisfactory communication, and 12.2% said they did not receive satisfactory communication. Responses were not different based on center size X 2 (4) = 3.546, p = .471 or reported percentage of patients they would like to be followed by psychology in an ideal world, X 2 (6) = 4.125, p = .660. When asked about present access to Psychology services compared to an ideal scenario, the majority of nephrologists indicated that in an ideal scenario, psychologists would be available to a larger portion of their patients . However, nephrologists also indicated that there is room to improve effectiveness of the psychosocial services to meet the needs of nephrology patients. Just under half (47.9%) of respondents indicated that their psychosocial services were either minimally effective or less than adequate . Of note, only respondents from centers without a dialysis unit indicated psychosocial services were minimally effective. The number of nephrology providers per responding center was highly variable (physicians ranged from 1 to 21, M = 6.12; midlevel professionals ranged from 0 to 10, M = 1.60). The number of attending nephrologists per institution was used to categorize centers by size for purposes of statistical analysis. Corresponding to 25 th , 50 th , and 75 th percentile values, small centers are classified as those having ≤ 3 nephrologists, medium centers 4–7 nephrologists, and large centers ≥ 8 nephrologists. For additional characteristics of centers by size, see . All centers indicated treating a wide variety of kidney conditions; however, there was pronounced variability in treatment of metabolic bone disease, with five centers (2 small, 1 medium, and 2 large) not treating these conditions within the nephrology division. Most centers performed kidney transplants, with the likelihood of this increasing with the size of the center X 2 (2) = 7.964, p = .019. The total number of providers at a center (physicians plus midlevel professionals) was strongly associated with the number of chronic dialysis patients cared for at the center, r = .738, p < .001. Availability of psychosocial providers varied based on nephrology division size, such that as nephrology center size increased, access to various psychosocial providers increased. See for availability of psychosocial services to entire nephrology divisions, as well as the variety of psychosocial services to renal dialysis units specifically. Further analysis indicated that the larger the center, the more likely it was to have a dedicated social worker assigned to the division [ X 2 (2) = 11.688, p = .003] and the same was true for pediatric psychology [ X 2 (2) = 12.434, p = .002]. Likewise, responding centers that were ranked in top 40 of U.S. News and World Reports were more likely to have a pediatric psychologist specifically assigned to their nephrology division than those that were not [ X 2 (1) = 5.841, p = .016], but were not more likely to have an assigned social worker [ X 2 (1) = 2.683, p = .101]. Moreover, centers were more likely to have a social worker [ X 2 (1) = 5.544, p = .019] assigned to their division if they had a renal dialysis unit at their institution, but the same was not true for pediatric psychologists assigned to division [ X 2 (2) = 2.941, p = .086]. Available services in the RDU are also reported in and did not significantly differ by center size. However, among centers with an RDU, those that were ranked in top 40 of U.S. News and World Reports were more likely to have school teachers [ X 2 (1) = 6.310, p = .012] and therapeutic recreation specialists [ X 2 (1) = 4.342, p = .037] as part of the psychosocial care team in their RDU than those that were not. The majority of respondents (60.4%) reported that health screens are performed as a part of standard of care. Further analyses indicated that the majority centers administer behavioral health screens to chronic dialysis patients (57.1% of centers administer to hemodialysis patients; 53.1% to peritoneal dialysis patients) and renal transplant patients (57.1%). Interestingly, centers ranked in the top 40 of pediatric nephrology centers by the U.S. News and World Report were more likely to administer behavioral health screens then other centers, X 2 (2) = 8.437, p = .015. Relative to psychology supports specifically, respondents were asked to estimate how many of their patients were followed by psychology in some capacity. Although several respondents did not know (20.4%), the majority (63.3%; n = 31) reported that only 0–25% of their patients were followed by psychology. Respondents were further queried as to the perceived importance of psychosocial supports in helping provide care in their nephrology division. Responses were largely favorable, with all indicating that psychosocial services were at least “somewhat important.” As demonstrated in , perceived importance increases with center size, r = .405, p = .005. In addition to being asked to estimate how many of their current patients are followed by psychology, respondents were also asked how many of their patients would be followed by psychology in an “ ideal world . ” Although a small number (4.1%) indicated they did not know, the majority of respondents (55.1%) reported that ideally more than half of their patients would be followed by psychology. Although no respondents reported that ≥75% of their patients are currently followed by psychology, nearly a quarter of respondents (24.5%, n = 12) reported that in an ideal world 75% would indeed be seen by a psychologist. displays the reported current percentage of patients followed by psychology versus the percent of patients that would be followed by psychology in an ideal world for each responding center, excluding the two respondents who did not provide a specific percentage of patients in their ideal world. We also asked providers to rate their satisfaction with the psychologist’s communication, specifically regarding how the patient’s emotional and behavioral health may affect their medical condition and/or treatment. Of the responses, 18.4% said that they do receive satisfactory communication, 65.3% said they sometimes receive satisfactory communication, and 12.2% said they did not receive satisfactory communication. Responses were not different based on center size X 2 (4) = 3.546, p = .471 or reported percentage of patients they would like to be followed by psychology in an ideal world, X 2 (6) = 4.125, p = .660. When asked about present access to Psychology services compared to an ideal scenario, the majority of nephrologists indicated that in an ideal scenario, psychologists would be available to a larger portion of their patients . However, nephrologists also indicated that there is room to improve effectiveness of the psychosocial services to meet the needs of nephrology patients. Just under half (47.9%) of respondents indicated that their psychosocial services were either minimally effective or less than adequate . Of note, only respondents from centers without a dialysis unit indicated psychosocial services were minimally effective. The current study sought to describe the current landscape of pediatric nephrology center psychosocial services, inclusive of availability and satisfaction for such services, via survey distributed among participants in the Pediatric Nephrology Research Consortium. Since the PNRC represents the majority of centers in the US, this data is likely representative of the status of pediatric nephrology psychosocial services in the entire country. Collectively, pediatric nephrology centers vary widely in size (as assessed by number of physicians and providers in each program), and roughly align with workforce differences across the country which have been described previously . That is, the middle 50 percent of respondents work in programs with physician group size ranging from three to seven nephrologists, and 25 percent work within a group size of four nephrologists or less. Again, similar to previous workforce reports, the majority of nephrology programs care for the entire range of pediatric kidney conditions ; however, metabolic bone disease is a condition not ubiquitously addressed by nephrologists across programs. The majority of nephrology programs reported performing kidney transplants, although centers with small physician group sizes were less likely to do so. As to be expected, the more nephrology providers at a center, the more chronic dialysis patients were cared for at the center. Not every program had a renal dialysis unit, with less than half of smaller centers having a RDU and three quarters of medium and large centers having a RDU. The current data demonstrate that psychosocial supports are generally available within pediatric nephrology centers in the United States. However, the degree to which psychosocial services are available to, or embedded within, nephrology programs vary. Specifically, the larger the group/center size, the more likely centers were to have a social worker and psychologist specifically assigned to the nephrology division. Centers that responded and were ranked in the top 40 pediatric nephrology programs by U.S. News and World Report were more likely to have a pediatric psychologist specifically assigned to their division, and thus better equipped to provide additional support that may positively impact patient outcomes, including kidney transplant survival. The most commonly assigned psychosocial professionals in pediatric nephrology centers are social workers, with 83% of mid-sized and 100% of large centers having dedicated social workers. These professionals often serve multi-faceted roles, including helping to address social issues for patients and families, screening for behavioral issues, and providing triage and/or direct counseling services. Centers were less likely to have embedded pediatric psychologists than pediatric social workers, although all centers had access to pediatric psychologists. While having access to psychology, the majority of programs reported that one quarter or less (0–25%) of their patients are followed by psychology in some capacity. Only two (large) centers had a neuropsychologist specifically assigned to the division, whereas, some small and medium centers did not even have access to these professionals within their institution. Neuropsychologists assess cognitive and behavioral functioning that may be impacted by medical, psychological, or familial factors. A neuropsychological evaluation can help providers and families to understand specific cognitive deficits and to identify the most appropriate interventions to support the child with kidney disease. For example, understanding of a patient’s neurocognitive profile may help inform how to structure patient education or expectations for medication adherence to match with the patient’s cognitive abilities or literacy level. No centers had a psychiatrist assigned to their division, but almost all centers could access psychiatry via consultation. Unfortunately, similar to the workforce crisis in pediatric nephrology, there is a dearth and misdistribution of child psychiatrists available in the U.S. , so even if access to psychiatry exists, timely evaluations and access to ongoing follow-up care may be limited. Despite CMS requirements for regular behavioral health screening (specifically, depression and quality of life) among end stage kidney disease patients, close to 50 percent (43.9% - 47.9%) of the responding centers were not regularly screening for behavioral health indicators within chronic kidney populations. The inclusion of psychosocial team members may help to support required screening, as well as to support identification and targeted intervention for the most vulnerable nephrology patients. When examining psychosocial services available in RDUs specifically, adjunctive supports varied widely and did not appear dependent on center size. Child Life specialists were the most common service available in RDUs and thus may represent basic standard of care needs for pediatric patients receiving chronic dialysis. Massage therapists were the least common, with only two large centers reporting they were available in their RDU. Therapeutic recreation and early intervention specialists were also uncommon. Considering the average number of school hours missed due to attending hemodialysis sessions, it was notable that school teacher services were not universally available. However, school teachers were more commonly a part of RDU care when responding centers were highly ranked (top 40) by the US News and World Report, and again may represent a key service provision for highly disadvantaged chronic dialysis patients. Although availability of psychosocial supports varied, these supports appeared to be highly valued within pediatric nephrology centers, with 67.8 to 99.9 percent of centers reporting that psychosocial supports are either important or very important, with greater importance associated with larger nephrology group/center size. Although most nephrology programs indicated that few (less than a quarter) of their patients are followed by psychology, the clear majority expressed a desire (in an “ideal world”) that more of their patients were seen by psychology with most indicating they would like more than half of their patients followed by psychology in some capacity. Despite indicating the importance of psychosocial supports and highlighting an increased need for psychology supports specifically, the perceived effectiveness of psychosocial supports meeting the needs of nephrology patients varied and was often not favorable. Of note, the larger centers, with greater psychosocial resources, were more likely to indicate room for growth in order for psychosocial services to meet the needs of their patients (i.e., despite having access to more resources, the majority of larger centers reported psychosocial service effectiveness was “less than adequate”). This identified higher level of unmet need may reflect that larger nephrology centers see a high volume of patients with complex and chronic medical needs, whose health is also disproportionately impacted by key social determinants of health. These include but are not limited to their neighborhood and physical environment, educational level and health literacy, community and social support systems, economic instability, food insecurity, and overall poorer quality of life. Nephrologists, as well as the healthcare community as a whole, are identifying that these patient population characteristics require correspondingly higher psychosocial support. It may also reflect that as nephrologists become more familiar with the role of psychosocial providers and recognize the benefits (e.g., due to having a division assigned social worker and/or psychologist), they actually desire even more access to these services for their patients. It is also important to note, that centers that described psychosocial services as only “minimally effective” were all centers without an embedded RDU. Considering that CMS regulations mandate psychosocial support for patients, this may explain that association. Based on the survey results, social work is the most commonly available psychosocial discipline in pediatric nephrology centers, which may help to explain the high levels of burnout experienced by these team members , especially when other psychosocial support is not present. Nearly half of nephrology teams at small centers and all of the large pediatric nephrology centers have an embedded social worker. When a social worker is embedded in pediatric nephrology divisions, they may become the default provider for all patient psychosocial needs, creating an underutilization of other psychosocial disciplines by nephrology centers. Nephrologists and their patients are better served by having access to a full range of psychosocial providers, minimizing the potential for individual provider burnout and promoting better patient outcomes. Pediatric psychologists, who are doctoral level professionals (and often faculty within hospital systems), are well positioned to support the pediatric nephrology team as well as pediatric kidney patients. They are able to assist in identifying the most appropriate psychosocially informed medical care, directing evidenced-based interventions with patients and families, and supporting existing nephrologists and the other psychosocial team members. Given their extensive training in a broad array of treatment modalities and the intersection of physical and mental health, pediatric psychologists have unique expertise. Pediatric psychologists work to develop and implement patient interventions around adjustment to medical conditions, behavior plans to support adherence to medical regimen (including modifications for neurodiverse children), as well as promoting positive health behaviors and resiliency to prevent future injury or complications from medical conditions or comorbid psychosocial concerns. The integration of interdisciplinary teams has been introduced as a standard of care to help meet the myriad of medical and psychosocial needs of pediatric nephrology patients . While prior research has established that interdisciplinary care can vary widely across centers with variable team structure , results from our study may help define differences across centers by number of nephrology providers and patients served. The opportunity to collaborate within an interdisciplinary psychosocial provider team may help maximize best outcomes, such as improvement in medication adherence and preparation for patient transition to adulthood . Leveraging the expertise of multiple psychosocial disciplines may help alleviate the burden of care that falls on a limited number of pediatric nephrology providers. Strengths of the current study include providing an updated tally of the landscape of pediatric nephrology centers in the U.S. as well as a first look at the availability of psychosocial resources at these centers. Prior publications have examined psychosocial resources within pediatric dialysis units , but no publications to date have offered a summary of psychosocial supports, degree of availability of supports within these settings, as well as evaluation of satisfaction with and effectiveness of these supports by nephrologists. The current study is limited by the reliance on the nephrologist for self-reporting of psychosocial support resources and further lacks information from the perspective of the psychosocial providers themselves. It does not evaluate patient-related outcomes, specifically centered around key social determinants of health that we know impact this patient population . Psychosocial supports at PNRC sites in North America that did not respond to the survey ( N = 38) were not available for this project so may have skewed the results. However, responding PNRC centers were found to be similar to non-responding PNRC centers in terms of group size. Moreover, responding centers included the majority of top ranked pediatric nephrology centers. Finally, information regarding how psychosocial team members interact, service billing/reimbursement functions, and the percentage of time psychosocial team members are assigned to nephrology were not explored in the current survey. Future research should evaluate how psychosocial support services impact our understanding of social determinants of health and in turn may reduce their deleterious impacts on the pediatric nephrology population. Outcome data should also be evaluated for determination of the impact that increased psychosocial support (specifically those that are embedded in the practice) may have in reducing acute healthcare utilization. With the use of targeted interviews, research should focus on pediatric nephrology centers that have integrated psychosocial services into their practice and help elucidate how these services were established (i.e., structure of time allotment/financial funding), identify the key psychosocial providers included in interdisciplinary clinics, and undertake cost analysis supporting this integration . Future research should also explore whether psychosocial providers benefit from having a specialization in pediatric nephrology care, which addresses the complexities of disease outcomes and management, as opposed to a general psychosocial background. Emerging special interest groups and certifications highlight a role for nephrology-specific training for those who care for psychosocial aspects of patients with pediatric chronic kidney diseases. Research is needed to further assess the benefits of having a dedicated and embedded psychosocial provider in contrast to someone who may be available for consultation on an ad hoc basis. It is possible that the presence of embedded psychosocial providers, with time and expertise dedicated to this high-risk population, may result in improved preventative interventions for anticipated psychosocial concerns associated with treatment of the child’s chronic disease, as well as improve nephrologist familiarity and comfort with engaging psychosocial services. In conclusion, there is great variability in provision of psychosocial services and assignments of psychosocial professionals in pediatric nephrology centers in the US. Larger centers, with more pediatric nephrologists, are more likely to have access to a wider variety of psychosocial service professionals, and centers with a pediatric dialysis unit are more likely to have embedded social workers and psychologists. While psychosocial providers were available, the majority of respondents reported that <25% of their patients were followed by psychology, and the majority of pediatric nephrologists expressed desire for more psychosocial support for their patients. Much work remains to better understand the provision of psychosocial services for these patients at different centers and how these services are funded and utilized. Such work can help in formulation of key best practices in organizing and providing quality psychosocial care for children with kidney disorders. S1 Dataset (SAV) Click here for additional data file.
null	cedae7a2-de61-4078-bdcb-7e2a32f8ae4f	10168604	Microbiology[mh]	Eleven Salmonella outbreaks linked to NRTE breaded, stuffed chicken products (involving 187 patients) were reported in the United States during 1998–2022. Most of the products tested contained Salmonella . Products were produced by at least three establishments. Outbreaks have continued to occur despite changes made to product labels to better inform consumers and increase the percentages of persons who understand that the product is sold raw. Thus, stronger controls are needed to prevent illnesses associated with these products. NRTE breaded, stuffed chicken products can be made with various types of chicken, including comminuted, trimmings, or other parts. Certain chicken types are subject to FSIS performance standards, which are used to measure an establishment’s process control; the comminuted chicken used to make these products has the highest allowable percentage (13 of 52 [25%]) of Salmonella positives ( ). On April 28, 2023, the U.S. Department of Agriculture proposed to declare Salmonella an adulterant in NRTE breaded, stuffed chicken products, meaning that the product will be subject to regulatory action if Salmonella is detected even at very low levels ( ). Canada enacted regulations for certain breaded chicken products after investigators identified 19 Salmonella outbreaks caused by NRTE breaded chicken products during 2015–2019 ( ). These products were not stuffed; most were chicken nuggets. The government introduced four control options in nonstuffed products to reduce Salmonella to below detectable amounts in these products ( ). In 2019, the incidence of illness caused by Salmonella Enteritidis, the serotype implicated in 89% of those outbreaks, was 33% lower than it was during 2015–2018 and 7% lower than during the baseline years 2010–2014 ( ). Consumer-based interventions alone, such as improved product labels, have not been an effective solution. In recent years, labels have recommended using a conventional oven rather than a microwave and using a food thermometer ( ); however, a consumer research study found that even when consumers read the label, 12% did not realize the product was raw or partially cooked, and among consumers who owned a food thermometer, 52% reported that they typically did not use it while preparing this product ( ). Although labeling changes can help protect consumers, the questionnaire data show that some persons who knew the product was raw and followed the cooking instructions still became ill. Moreover, label changes cannot address inequities in access to recommended cooking appliances ( ). The number of patients who became ill from these products is likely much higher than that indicated from outbreak reports. Many persons regularly eat these products: in a U.S. population survey, 7.4% reported eating these products in the previous week. Although implicated products were distributed nationally, MDH officials identified all multistate outbreaks, and almost one half of outbreaks had cases identified only in Minnesota, suggesting that some outbreaks occurred but were not identified in other states. Illnesses even among persons who reported that they used a conventional oven and knew the product was raw indicate that consumer-based interventions have been insufficient. The high rate of contamination of products in outbreaks and the lack of first recognition of multistate outbreaks by a state other than Minnesota suggest that the prevalence of illness due to these products is higher than that indicated by outbreaks. Moreover, only a small proportion of all Salmonella illnesses are identified as such. Illness could be reduced by enhanced Salmonella control at the manufacturers that produce these products. The U.S. Department of Agriculture’s proposal to declare Salmonella an adulterant in NRTE breaded and stuffed chicken products will bring additional focus to this public health problem and encourage producers to better control Salmonella in the ingredients used to produce these products. Summary What is already known about this topic? Not ready-to-eat breaded, stuffed chicken products have repeatedly been a source of Salmonella outbreaks. On April 28, 2023, the U.S. Department of Agriculture proposed to declare Salmonella an adulterant in these products. What is added by this report? During 1998–2022, 11 Salmonella outbreaks linked to these products were reported; 57% of samples per outbreak from patient homes and retail stores yielded Salmonella . Outbreaks continue to occur, although a smaller percentage of patients reported cooking the product in a microwave after labeling changes. What are the implications for public health practice? Outbreaks have continued despite consumer-based interventions. Additional control measures for Salmonella contamination by manufacturers could reduce Salmonella -involved illnesses associated with these products. Not ready-to-eat breaded, stuffed chicken products have repeatedly been a source of Salmonella outbreaks. On April 28, 2023, the U.S. Department of Agriculture proposed to declare Salmonella an adulterant in these products. During 1998–2022, 11 Salmonella outbreaks linked to these products were reported; 57% of samples per outbreak from patient homes and retail stores yielded Salmonella . Outbreaks continue to occur, although a smaller percentage of patients reported cooking the product in a microwave after labeling changes. Outbreaks have continued despite consumer-based interventions. Additional control measures for Salmonella contamination by manufacturers could reduce Salmonella -involved illnesses associated with these products.
“Comparing the effectiveness, acceptability and oral hygiene status between vacuum formed retainer and Begg’s retainer”: a pilot study	4cf04c6b-e51d-433d-a6f8-cc2403b0835c	10169470	Dental[mh]	There are various goals of Orthodontic treatment. They can be expressed as achieving esthetics, stability, functional occlusion and aligned tooth. Active phase of Orthodontic movement takes one half to two years. Post the active phase, there are various changes noted in the occlusion. The negative changes are relapse and the positive changes are improved teeth interdigitation. Maintaining proper interdigitation of teeth is the most challenging stage of Orthodontic treatment. Avoiding relapse poses challenge to the Orthodontist, so thorough understanding of the factors associated with relapse is of paramount importance. In the 1960s, supra alveolar fibres were transected to prevent relapse. Orthodontically derotated teeth are more unstable. There are chances of rotational relapse even after a longer retention period. Other studies have shown that Circumferential Supracrestal Fiberotomy (CSF) could reduce dental relapse, especially of rotated teeth. Hence, it is required for the clinician to have knowledge about the various methods of reducing relapse, advantages and disadvantages of various retainers. Retention appliances are fabricated to maintain teeth alignment and arch dimensions post Orthodontic treatment. Retainer is chosen by considering various factors like efficiency, cost, patient preferences, cooperation and satisfaction . The prime concern and debate in the branch of Orthodontics is related to long term establishment of the achieved tooth movement. Commonly used retainers are the Hawley’s retainers (HRs), wrap around or Begg’s retainers, lingual bonded retainers and the newly familiarized Vacuum formed retainers (VFRs). The Hawley retainer was designed by Charles Hawley and is most popular as a removable retention appliance. The VFR was designed in 1971. Recently, VFRs are preferred as an Orthodontic retention appliance over conventional Begg’s retainers as they claimed to be aesthetic, durable, reduced failure rate, translucent , inexpensive , comfortable and simple to fabricate . There are several studies relating the comparison between VFRs and HRs . Although retention is a must factor for successful orthodontic treatment, there is less evidence regarding the most appropriate strategy. Some have conducted randomized clinical trials showing VFRs to be more effective than HRs where as other studies have shown no statistical difference in the effectiveness of both the retainers. However, very few studies have been done regarding the effectiveness, acceptability, and oral hygiene between VFRs and Begg’s retainers. Aims and objectives The aim of the present study was to compare the clinical effectiveness, oral hygiene status and acceptability of vacuum formed retainers (VFRs) and Begg’s retainers over a 12 months period post debonding, as very few cases come under less than 12 months retention period regimen. This was an experimental study with random allocation between two types of retainer. The hypothesis was that there was no comparison between the clinical effectiveness, oral hygiene status and acceptability of vacuum formed retainers ( VFRs) and Begg’s retainers over a 12 months period post debonding. The aim of the present study was to compare the clinical effectiveness, oral hygiene status and acceptability of vacuum formed retainers (VFRs) and Begg’s retainers over a 12 months period post debonding, as very few cases come under less than 12 months retention period regimen. This was an experimental study with random allocation between two types of retainer. The hypothesis was that there was no comparison between the clinical effectiveness, oral hygiene status and acceptability of vacuum formed retainers ( VFRs) and Begg’s retainers over a 12 months period post debonding. The study was approved by the Institutional Ethics Committee, Institute of Medical Sciences (IMS) and Sum Hospital, Siksha O Anusandhan University (Ref no./ DMR/IMSSH/SOA/180297). Eighty patients who completed fixed Orthodontic treatment were included. Inclusion criteria: a Patients with fixed Orthodontic appliance with or without extraction b Patients with optimum acceptable occlusion. c Patients with good oral hygiene without any periodontal disease. d Patients with age 18 years or above e Patients with no prosthetic rehabilitation. Exclusion criteria: a Patients with missing teeth b Patients who have undergone orthognathic surgery c Patients with cleft lip and palate d Patients with temporo-mandibular joint( TMD) disorders After debonding of appliance and random allocation, the retainers were delivered on the same day. 40 VFRs, depicted as Retainer 1 / R1 (Fig. ) were delivered in maxillary arch and mandibular arch. Similarly, 40 Begg’s retainers, depicted as Retainer 2 / R2 (Fig. ) were delivered in maxillary and mandibular arch. The buccal view of the retainers is represented in Fig. . The patients were instructed to wear the appliance 24 h for the first 6 months followed by 6 months of night time wear. VFRs were fabricated with Duran 1.5 mm thickness sheet, composed of Polyethyleneterephthalate Glycol (PET-G) using Ministar machine (Scheu-Dental, Iserlohn, Germany). This material allows better withstanding wear without causing bite alterations. When the thermoformed plastic is thin, the premature occlusal contacts can be easily avoided (Sheridan et.al). Also, the Essix retainer has optimum fit, is esthetic and comfortable ensuring long-term stability of the occlusal alignment . The facial and buccal surfaces of VFRs were trimmed respectively to cover the incisal one-third of the incisors and occlusal surfaces of posteriors along with providing 2-mm buccal and 3-4 mm lingual extensions( Fig. ). The retainers were extended till the last erupted molar . The Begg’s retainers were fabricated using acrylic baseplates ( DPI, heat cure acrylic, India) and labial bow with 0.9 mm stainless steel wire (Leone spa, Florence, Italy). The labial bow has U loops placed between 1 st and 2 nd premolar in non-extraction cases, and between remaining premolar and 1 st molar in extraction cases which crossed the occlusal plane distal to the last erupted molar. Effectiveness of the retainers were measured using ABO measuring gauge. Oral Hygiene Index-Simplified index (OHI-S) and Gingival Index (GI) evaluated the patient’s oral hygiene condition, and patient’s compliance was evaluated by specific questionnaire (Fig. ). All these findings were performed at T 0 (After debonding), T 1 (3 months after using retainers), T 2 (6 months after using retainers), T 3 (9 months after using retainers), T 4 (12 months after using retainers) follow up stages, except the feedback form and the breakage of retainers that was filled at T 4 stage. Patients acceptance was evaluated using a 10-cm visual analogue scale (VAS) from the data collected in questionnaires that include 4 questions i.e. teeth biting (closing teeth with retainers, not chewing food), fitting of the appliance, appearance, gingival irritation. Patients were given instructions and explanations on how to complete these questionnaires. The lowest (least favourable) score was ‘0’ and the highest (most favourable) score was ‘10’. For example, if the retainer was very comfortable, it was scored as ‘0’. Questionnaires were filled in front of the Orthodontist at T4 stage. Sample size The study envisages test of association of different characteristics (effectiveness, oral hygiene and acceptability) between retainers through chi-square test of independence. Therefore, minimum sample size is computed as per the requirement of the chi-square test of goodness of fit of contingency table. χ 2 tests— Goodness-of-fit tests: Contingency tables Analysis: A priori: Compute required sample size Input: Effect size w = 0.41 (moderate effect size was assumed)α err prob = 0.05 Power (1-β err prob) = 0.80 Df = 5 Output: Noncentrality parameter λ = 12.9437000 Critical χ 2 = 11.0704977 Total sample size = 77 Rounded off to 80. Actual power = 0.8041224 Thus, the total sample size was 80. For maxillary, mandibular arch minimum 40 R1 and R2 retainers from each were fabricated. Thus, 160 retainers were designed. Follow up: Mean retention time was 1 year Maxillary and mandibular casts were analysed at four stages: 1. T0 – Post debonding 2. T1 – After three months of using retainers 3. T2 – After six months of using retainers 4. T3—After nine months of using retainers 5. T4 – After twelve months of using retainers Irrespective of the scheduled time, patient was asked to report to the department whenever there is breakage of the appliance. We ensured that all patients stuck to the recommended schedule of retainer wear, through periodic verbal reminders. Statistical analysis Data collected on 80 cases were scrutinized, coded and entered into IBM SPSS 24.0 statistics, SPSS South Asia Pvt. Ltd. Data were analysed by the following statistical procedure. Association of teeth biting, fitting of appliances, appearance, durability, gingival irritation, comfort of Maxillary and Mandibular arch in retainer 1 and retainer 2 were analysed using frequency procedure and Chi-square test / Fisher’s Exact test. Association of oral hygiene with OHIS index and Gingival Index for maxillary and mandibular arch in retainer 1 and retainer 2 were analysed using frequency procedure and in retainer 1 and retainer 2 were analysed using frequency procedure and Chi / Fisher’s Exact test. Comparison of Alignment, Marginal Ridges and Interproximal contact of effectiveness at T1, T2, T3, T4 with reference to T0 in maxillary arch and mandibular arch in retainer 1 and retainer 2 were analysed using frequency procedure and marginal homogeneity test. The cut off value of ‘p’ for test of significance was taken as ⁢0.05. Results Both R 1 ,R 2 retainers showed improved teeth alignment in both the maxillary and mandibular arches (Table ) at subsequent stages of follow up. Considering the marginal ridge changes (Table ) with both retainers in maxillary arch; Begg’s retainers and VFRs showed 95%, 55% improved levelling at T 4 respectively ( p < 0.05). Improvement with VFRs at T 1 , T 2 , T 3 when compared to T 0 was not significant ( p > 0.05) but it was significant for Begg’s retainers ( p < 0.05). Marginal ridge changes with both retainers in mandibular arch; Begg’s and VFRs showed 87.5%, 52.5% improved levelling at T 4 respectively ( p < 0.05). At T 1 with Begg’s retainers, improvement in marginal ridge when related to T 0 was not significant ( p = 0.157) but afterwards there was significant improvement over the time T 0 ( p < 0.05), but with VFRs at T 1 , T 2 , T 3 the improvement with reference to T 0 was not significant ( p > 0.05). Interproximal contacts (Table ) in maxillary arch with VFRs and Begg’s retainers reduced to 77.5% and 60% respectively ( p < 0.05). It was also reduced to 75% at T 4 with VFRs and 67.5% with Begg’s retainers in mandibular arch that was significant ( p = 0.009) 7. Patients wearing Beggs’s retainers had significantly better ( p < 0.05) OHI-S index and GI index in comparison to VFRs (Table ). Significant differences ( p = 0.000) were observed with Begg’s retainers in teeth biting,whereas no significant difference was found with fitting of appliance ( p = 0.180) and gingival irritation ( p = 1.000). For aesthetic appearance of patients, VFRs were well accepted, which was significant ( p = 0.002). Results revealed that both the retainers were prone to breakage with subsequent follow ups but it was not significant ( p = 0.162). Overall level of comfort was good with Begg’s retainers, but was not significant ( p = 0.051) (Table ). The study envisages test of association of different characteristics (effectiveness, oral hygiene and acceptability) between retainers through chi-square test of independence. Therefore, minimum sample size is computed as per the requirement of the chi-square test of goodness of fit of contingency table. χ 2 tests— Goodness-of-fit tests: Contingency tables Analysis: A priori: Compute required sample size Input: Effect size w = 0.41 (moderate effect size was assumed)α err prob = 0.05 Power (1-β err prob) = 0.80 Df = 5 Output: Noncentrality parameter λ = 12.9437000 Critical χ 2 = 11.0704977 Total sample size = 77 Rounded off to 80. Actual power = 0.8041224 Thus, the total sample size was 80. For maxillary, mandibular arch minimum 40 R1 and R2 retainers from each were fabricated. Thus, 160 retainers were designed. Follow up: Mean retention time was 1 year Maxillary and mandibular casts were analysed at four stages: 1. T0 – Post debonding 2. T1 – After three months of using retainers 3. T2 – After six months of using retainers 4. T3—After nine months of using retainers 5. T4 – After twelve months of using retainers Irrespective of the scheduled time, patient was asked to report to the department whenever there is breakage of the appliance. We ensured that all patients stuck to the recommended schedule of retainer wear, through periodic verbal reminders. Data collected on 80 cases were scrutinized, coded and entered into IBM SPSS 24.0 statistics, SPSS South Asia Pvt. Ltd. Data were analysed by the following statistical procedure. Association of teeth biting, fitting of appliances, appearance, durability, gingival irritation, comfort of Maxillary and Mandibular arch in retainer 1 and retainer 2 were analysed using frequency procedure and Chi-square test / Fisher’s Exact test. Association of oral hygiene with OHIS index and Gingival Index for maxillary and mandibular arch in retainer 1 and retainer 2 were analysed using frequency procedure and in retainer 1 and retainer 2 were analysed using frequency procedure and Chi / Fisher’s Exact test. Comparison of Alignment, Marginal Ridges and Interproximal contact of effectiveness at T1, T2, T3, T4 with reference to T0 in maxillary arch and mandibular arch in retainer 1 and retainer 2 were analysed using frequency procedure and marginal homogeneity test. The cut off value of ‘p’ for test of significance was taken as ⁢0.05. Both R 1 ,R 2 retainers showed improved teeth alignment in both the maxillary and mandibular arches (Table ) at subsequent stages of follow up. Considering the marginal ridge changes (Table ) with both retainers in maxillary arch; Begg’s retainers and VFRs showed 95%, 55% improved levelling at T 4 respectively ( p < 0.05). Improvement with VFRs at T 1 , T 2 , T 3 when compared to T 0 was not significant ( p > 0.05) but it was significant for Begg’s retainers ( p < 0.05). Marginal ridge changes with both retainers in mandibular arch; Begg’s and VFRs showed 87.5%, 52.5% improved levelling at T 4 respectively ( p < 0.05). At T 1 with Begg’s retainers, improvement in marginal ridge when related to T 0 was not significant ( p = 0.157) but afterwards there was significant improvement over the time T 0 ( p < 0.05), but with VFRs at T 1 , T 2 , T 3 the improvement with reference to T 0 was not significant ( p > 0.05). Interproximal contacts (Table ) in maxillary arch with VFRs and Begg’s retainers reduced to 77.5% and 60% respectively ( p < 0.05). It was also reduced to 75% at T 4 with VFRs and 67.5% with Begg’s retainers in mandibular arch that was significant ( p = 0.009) 7. Patients wearing Beggs’s retainers had significantly better ( p < 0.05) OHI-S index and GI index in comparison to VFRs (Table ). Significant differences ( p = 0.000) were observed with Begg’s retainers in teeth biting,whereas no significant difference was found with fitting of appliance ( p = 0.180) and gingival irritation ( p = 1.000). For aesthetic appearance of patients, VFRs were well accepted, which was significant ( p = 0.002). Results revealed that both the retainers were prone to breakage with subsequent follow ups but it was not significant ( p = 0.162). Overall level of comfort was good with Begg’s retainers, but was not significant ( p = 0.051) (Table ). Long term studies have shown that relapse occurs in approximately 70% of patients . To avoid relapse, the respiratory, masticatory and postural functional context is very important to correct if necessary. During the process of Orthodontic treatment plan, setting the retention protocol is of utmost importance . Retention plan completes the comprehensive Orthodontic treatment . It should be the primary focus of Orthodontist to decide the best possible retention plan and retention device for individual patients. There are variety of retention appliances available in Orthodontic literature. The operator should be extra cautious to decide on the retention appliance depending on the patients need and literature evidence. Previous literature suggests Hawley’s [ , , , ], VFRs , Begg’s retainers as most efficient, effective and popular among Orthodontist. In our knowledge, this is the first ever study on the comparative assessment of compliance between two popularly used retainers over a period of one year post debonding. HRs allows more vertical settling of posterior teeth than VFRs according to previous literature . It has a demerit of wire component crossing occlusally, which has issue especially in extraction cases. The side effect of spaces opening up interdentally is evident . On this basis, Begg’s wrap around retainers have some advantage over the Hawley’s retainers, as there is absence of occlusally crossing wire components and uncovered occlusal surfaces that allow proper interdigitation of teeth. The biggest difference between VFRs and Begg’s retainers was found to be the presence of occlusal covering in VFRs, which leads to inadequate vertical settling of teeth during retention period . This aspect is beneficial in maintaining the levels of teeth that were moved in vertical direction to correct deep bite. The literature evidence comparing VFRs and Begg’s wrap around is limited and the studies done earlier, were conducted for only a 6 month retention span . The recommendation of retention period less than 12 months is very rare. Hence, this study was intended to do a thorough comparison between VFRs and Begg’s retainers focusing all important clinical aspects, over a period of 12 months. After the completion of Orthodontic treatment, debonding was done and R 1 , R 2 retainers were delivered immediately. Patients were asked to wear the retainers for at least 12 months. They were given a questionnaire to fill based on various aspect of comfort at the end of 12 months (T 4 ). Three criteria from ABO model grading system to observe the retention of teeth were adopted to evaluate the retention effectiveness of both the retainers at T 0 , T 1 , T 2 , T 3 , T 4 . Also, the durability of the retainers was assessed by considering the breakage of the appliances reported by the patients. OHI-S index and GI index were recorded to evaluate the oral hygiene of the patients associated with R 1 and R 2 retainer at T 0 , T 1 , T 2 , T 3 and T 4 . Association of effectiveness with R1 and R2 retainers When considering alignment and contact point, both show improvement over the entire period of observation (Table ), but the leveling of arches with proper marginal ridge alignment, is better with R 2 retainer (Table ). Whereas, in case of VFRs it is not improving over the observation period of 12 months. This is in agreement with previous literature by Sauget et al. (1997) showing minimal improvement of teeth by R 1 retainers in vertical direction during retention phase. Recent studies by author like Dincer and Isik Aslan (2010) , Hoybjerg et al. (2013) too showed minimal improvement in vertical direction. The explanation of this is presence of thorough adapted occlusal covering of thermoplastic sheets in VFRs. So, from the above finding it is clear that, decision of giving VFR retainers in finished Orthodontic treatment where the operator is expecting posterior settling post debonding, is contraindicated. VFRs are acceptable in cases where proper vertical positioning of teeth have been achieved before debonding. Proper posterior settling of teeth will deliver good posterior occlusal guidance resulting in distribution of occlusal forces on maximum number of inclined planes during oral functions, providing maximum periodontal support . When correlating durability between the two retainers, material thickness must be considered. Different authors have used various thickness of VFR sheets that may be of 0.75 mm, 1 mm and 1.5 mm. It was evident by Gardner et al that VFRs material are more prone to wear and tear than retainers made with acrylic. Also, Hichens et al., concluded increased number of Hawley’s retainers breakage than VFRs which could be due to thin acrylic plate. This might also be because of difference of elasticity between acrylic and thermoplastic materials used in the above discussed retainers. After 12 months of retention plan, durability of the retainers was investigated. The results indicated breakage in both the retainers which was not statistically significant (Table ). This finding is supported by the study of Sun et al . However, in our study an increased breakage in R 1 retainers was found which could be due to stresses generated by forces exerted on the covered occlusal surfaces during its continuous wear . The breakage of the R 2 retainers was more commonly due to the mishandling and negligence by the patients. Association of oral hygiene with R 1 and R 2 retainers At T 0 , immediately after debonding of fixed Orthodontic appliances, the patients had mild to moderate gingivitis based on Gingival Index (Table ). In the presence of long standing fixed Orthodontic appliance, oral cavity is prone to accumulation of plaque and calculus leading to gingivitis . It was found to be common in both the arches, but in the mandibular arch the severity was more as compared to maxillary arch. This was obvious because of the opening of mandibular salivary glands that makes mandibular anterior region more prone to accumulation of calculus . Hence higher degree of gingivitis. At T 1 , in maxillary and mandibular arch R 1 , R 2 retainers deteriorated the gingival health, but was not statistically significant. However, it was more with R 1 retainers (Table ). Over time (towards the T 4 stage), statistically significant worsening of the gingival health was observed in patients with R 1 retainers. This is in accordance with the studies done by LiciaManzon et al., that VFRs cover the entire teeth surface preventing salivary self-cleansing action intraorally, allowing growth of microorganisms resulting in poor oral hygiene. Moreover, minor inaccessible areas present in the appliance makes it cumbersome to clean, hence vulnerable to food lodgment, promoting microbial growth . Analysis of simplified oral hygiene and gingival index statistically supports the concept of VFRs favoring debris and calculus formation, resulting in worsening of gingival conditions, compared with Begg’s retainers. Many factors influence the oral health status. Oral environment can be altered due to presence of these retention appliances which may change the micro flora. The menace is also related to the design, surface roughness of retainers and physical properties of materials . Association of comfort level with R 1 and R 2 retainers The R 2 retainers showed statistically significant acceptability in terms of teeth biting than R 1 retainers (at T 4 ) in both maxillary and mandibular arch (Table ). It was apparent because in the design of R 2 retainers occlusal surfaces of the teeth are not covered, allowing better vertical settling of teeth over a period of time. Whereas, VFRs covering the incisal and occlusal surfaces retain the teeth in their debonded positions . When we evaluated fitting of the appliance with R 1 , R 2 retainers over a period of 12 months; R 1 retainers had statistically significant better fitting (Table ). Reason being the VFRs are machine made, excluding human errors and accurate adaptation of VFR sheets over the intraoral hard and soft tissues. Thus, helping in a better fit of the appliance. The R 2 retainers tend to loosen up over a period of time due to presence of malleable stainless steel wire components, which is also seen to be explained by Kumar AG, Bansal A in their study on Indian population comparing the effectiveness and acceptability between VRs and Begg’s retainer . Results of statistical analysis was significant, suggesting R 1 retainers being aesthetically more acceptable by patients than R 2 retainers (Table ). The rationale behind this is transparent plastic sheet and lighter weight appliance . However, Begg’s appliance was moderately accepted by patients and quite aesthetic as well. Mild Gingival irritation was seen with R 1 and R 2 retainers in both the arches which was statistically insignificant (Table ). Lesser gingival irritation could be due to absence of retention clasp in both R 1 and R 2 retainers. This is also supported in a single-centre, randomized control trial by Mohammed Saleh et al. on-acceptability comparison between HRs and VFRs in Orthodontic adult patients . The effect of resin based retainers on soft tissue must be understood. Studies have shown that uncured resins can leach and harm the soft tissues . We have taken adequate precaution in terms of choice of material ( heat cured), employed proper mixing technique ( vacuum mixing) to minimize the presence of uncured monomers in the fabricated Begg’s retainer. To conclude about the capability to retain the Orthodontic treatment outcome, a longer duration research is recommended. As the patients were given two different type of retainers in both maxillary and mandibular arch, the study was unable to provide insight regarding changes in occlusal contacts in vertical and sagittal direction. As per previous studies by Mufide et al , and Wenjia et al , a newer study design having two groups of subjects with VFRs and Begg’s retainers in both the arches has to be observed to conclude the same. We have taken ABO measuring gauge, which is subjected to examiner inconsistency. It would have been much better to use a digital method . Also, a mouth breathing habit can provoke gingival inflammation of the anterior teeth but this breathing parameter was not taken into account. "The outcome of our study recommends Begg’s wrap around retainers as a preferred mode of retention post Orthodontic treatment, as long as esthetics is not the prime concern.” When considering alignment and contact point, both show improvement over the entire period of observation (Table ), but the leveling of arches with proper marginal ridge alignment, is better with R 2 retainer (Table ). Whereas, in case of VFRs it is not improving over the observation period of 12 months. This is in agreement with previous literature by Sauget et al. (1997) showing minimal improvement of teeth by R 1 retainers in vertical direction during retention phase. Recent studies by author like Dincer and Isik Aslan (2010) , Hoybjerg et al. (2013) too showed minimal improvement in vertical direction. The explanation of this is presence of thorough adapted occlusal covering of thermoplastic sheets in VFRs. So, from the above finding it is clear that, decision of giving VFR retainers in finished Orthodontic treatment where the operator is expecting posterior settling post debonding, is contraindicated. VFRs are acceptable in cases where proper vertical positioning of teeth have been achieved before debonding. Proper posterior settling of teeth will deliver good posterior occlusal guidance resulting in distribution of occlusal forces on maximum number of inclined planes during oral functions, providing maximum periodontal support . When correlating durability between the two retainers, material thickness must be considered. Different authors have used various thickness of VFR sheets that may be of 0.75 mm, 1 mm and 1.5 mm. It was evident by Gardner et al that VFRs material are more prone to wear and tear than retainers made with acrylic. Also, Hichens et al., concluded increased number of Hawley’s retainers breakage than VFRs which could be due to thin acrylic plate. This might also be because of difference of elasticity between acrylic and thermoplastic materials used in the above discussed retainers. After 12 months of retention plan, durability of the retainers was investigated. The results indicated breakage in both the retainers which was not statistically significant (Table ). This finding is supported by the study of Sun et al . However, in our study an increased breakage in R 1 retainers was found which could be due to stresses generated by forces exerted on the covered occlusal surfaces during its continuous wear . The breakage of the R 2 retainers was more commonly due to the mishandling and negligence by the patients. 1 and R 2 retainers At T 0 , immediately after debonding of fixed Orthodontic appliances, the patients had mild to moderate gingivitis based on Gingival Index (Table ). In the presence of long standing fixed Orthodontic appliance, oral cavity is prone to accumulation of plaque and calculus leading to gingivitis . It was found to be common in both the arches, but in the mandibular arch the severity was more as compared to maxillary arch. This was obvious because of the opening of mandibular salivary glands that makes mandibular anterior region more prone to accumulation of calculus . Hence higher degree of gingivitis. At T 1 , in maxillary and mandibular arch R 1 , R 2 retainers deteriorated the gingival health, but was not statistically significant. However, it was more with R 1 retainers (Table ). Over time (towards the T 4 stage), statistically significant worsening of the gingival health was observed in patients with R 1 retainers. This is in accordance with the studies done by LiciaManzon et al., that VFRs cover the entire teeth surface preventing salivary self-cleansing action intraorally, allowing growth of microorganisms resulting in poor oral hygiene. Moreover, minor inaccessible areas present in the appliance makes it cumbersome to clean, hence vulnerable to food lodgment, promoting microbial growth . Analysis of simplified oral hygiene and gingival index statistically supports the concept of VFRs favoring debris and calculus formation, resulting in worsening of gingival conditions, compared with Begg’s retainers. Many factors influence the oral health status. Oral environment can be altered due to presence of these retention appliances which may change the micro flora. The menace is also related to the design, surface roughness of retainers and physical properties of materials . 1 and R 2 retainers The R 2 retainers showed statistically significant acceptability in terms of teeth biting than R 1 retainers (at T 4 ) in both maxillary and mandibular arch (Table ). It was apparent because in the design of R 2 retainers occlusal surfaces of the teeth are not covered, allowing better vertical settling of teeth over a period of time. Whereas, VFRs covering the incisal and occlusal surfaces retain the teeth in their debonded positions . When we evaluated fitting of the appliance with R 1 , R 2 retainers over a period of 12 months; R 1 retainers had statistically significant better fitting (Table ). Reason being the VFRs are machine made, excluding human errors and accurate adaptation of VFR sheets over the intraoral hard and soft tissues. Thus, helping in a better fit of the appliance. The R 2 retainers tend to loosen up over a period of time due to presence of malleable stainless steel wire components, which is also seen to be explained by Kumar AG, Bansal A in their study on Indian population comparing the effectiveness and acceptability between VRs and Begg’s retainer . Results of statistical analysis was significant, suggesting R 1 retainers being aesthetically more acceptable by patients than R 2 retainers (Table ). The rationale behind this is transparent plastic sheet and lighter weight appliance . However, Begg’s appliance was moderately accepted by patients and quite aesthetic as well. Mild Gingival irritation was seen with R 1 and R 2 retainers in both the arches which was statistically insignificant (Table ). Lesser gingival irritation could be due to absence of retention clasp in both R 1 and R 2 retainers. This is also supported in a single-centre, randomized control trial by Mohammed Saleh et al. on-acceptability comparison between HRs and VFRs in Orthodontic adult patients . The effect of resin based retainers on soft tissue must be understood. Studies have shown that uncured resins can leach and harm the soft tissues . We have taken adequate precaution in terms of choice of material ( heat cured), employed proper mixing technique ( vacuum mixing) to minimize the presence of uncured monomers in the fabricated Begg’s retainer. To conclude about the capability to retain the Orthodontic treatment outcome, a longer duration research is recommended. As the patients were given two different type of retainers in both maxillary and mandibular arch, the study was unable to provide insight regarding changes in occlusal contacts in vertical and sagittal direction. As per previous studies by Mufide et al , and Wenjia et al , a newer study design having two groups of subjects with VFRs and Begg’s retainers in both the arches has to be observed to conclude the same. We have taken ABO measuring gauge, which is subjected to examiner inconsistency. It would have been much better to use a digital method . Also, a mouth breathing habit can provoke gingival inflammation of the anterior teeth but this breathing parameter was not taken into account. "The outcome of our study recommends Begg’s wrap around retainers as a preferred mode of retention post Orthodontic treatment, as long as esthetics is not the prime concern.” Our study found that Both VFRs and Begg’s retainers are efficient enough in maintaining proper teeth alignment but better leveling of marginal ridges is found with Begg’s retainers. While considering only aesthetics, VFRs are largely preferred over Begg’s retainers. VFRs deteriorate oral hygiene of the patients progressively over a period of time as compared to Begg’s retainers. As long as durability of the retainers are considered, there is no statistically significant difference observed, but the number of breakages were more with VFRs. The outcome of our study recommends Begg’s wrap around retainers as a preferred mode of retention post Orthodontic treatment, as long as esthetics is not the prime concern.
Bioactive glass for periodontal regeneration: a systematic review	1781647c-3a03-4071-9462-0c72d13a1df4	10169491	Dental[mh]	The principal anatomical sequela of periodontitis is represented by loss of alveolar bone support, and the extent and the severity of periodontal osseous lesions are usually assessed by both clinical and radiographic means . Generally, periodontal defects are classified into three groups: suprabony (or horizontal) defects, infrabony (or vertical) defects, and interradicular (or furcation) defects. According to the classification by Goldman (1958) , suprabony defects are those in which the base of the pocket is located coronal to the alveolar crest. Infrabony defects, on the other hand, are defined when the apical end of the pocket is located below the bone crest. Specifically, an infrabony bone could be recognized as intrabony defect if subcrestal component involves the root surface of only one tooth, while we can define crater as a defect that affects two adjacent root surfaces to a similar extent. Intrabony defects have been classified with respect to the number of remaining bony walls, into three categories: the 1-wall, 2-wall and 3-wall defects [ – ]. One of the major clinical challenges of this age could be represented by the possibility to obtain a complete regeneration of infrabony defects. For the successful reconstruction of periodontal tissues, that means bone, cementum and periodontal ligament, it is fundamental to respect all the natural sequence of biological events that takes place during the periodontal healing . Currently, bone autografts represent the gold standard treatment for bone and periodontal regeneration since it provides osteogenic, osteoconductive and osteoinductive properties [ – ]. However, there are numerous disadvantages associated with bone autografts, such as limited availability and variable quality, donor site morbidity, increased operative time [ – ]. Because of these drawbacks, bone tissue engineering strategies have been developed to obtain successful bone healing . However, despite the numerous materials and different approaches developed over the past few years, periodontal regeneration still deals with many challenges, and the complete regeneration of the attachment apparatus is an unpredictable goal [ – ]. Among all biomaterials, bioglasses (BG) are one of the most interesting due to their ability to form a highly reactive carbonate hydroxyapatite layer . BG is a family of bioactive glasses composed of silicon dioxide, sodium oxide, calcium oxide, and phosphorous pentoxide. L. L. Hench discovered this material in 1969 acting as the first alternative to bioinert implants . The first material found to form a bond with bone was the original bioactive glass composition, 45S5 Bioglass (45% SiO 2 , 24.5% CaO, 24.5% Na 2 O, and 6% P 2 O 5 ). It was the first artificial material that provided bonding interface with bone as well as with soft tissues [ – ]. A very relevant effect of BG is that the release of biologically active soluble Silicon (Si 4+ ) and Calcium (Ca 2+ ) ions increases the expression of an osteoblast mitogenic growth factor and stimulates bone growth all around bone- implant interface . A very relevant effect of BG is that the release of biologically active soluble Si 4+ and Ca 2+ ions increases the expression of an osteoblast mitogenic growth factor and stimulates bone growth all around bone- implant interface . What’s more, it has figured out the angiogenetic potential of Bioglass 45S5, as it could increase the secretion of vascular endothelial growth factor in vitro and to enhance vascularization in vivo [ – ]. The aim of this systematic review is to assess the effect of BG on bone and periodontal regeneration and to perform a meta-analysis of the potential of this material for the treatment of intrabony and furcation defects in periodontal diseases. The systematic review follows the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) statement and the protocol was registered on Prospero (CRD42021254354). The proposed focused question was: “What does BG do in terms of regeneration of periodontal defects?” The focused question was established according to the PICO strategy: Population: Patients with periodontal and bone defects. Intervention: Bioactive glass. Comparison: Open flap debridement (OFD) only; different biomaterials. Outcomes: Periodontal regeneration ; Bone regenerationin terms of decrease of probing depth (PD) and gain of clinical attachment level (CAL). Search strategy This literature search was conducted to identify articles on the use of bioactive glass for periodontal and bone regeneration. In March 2021 an electronic search was performed in the following databases: MEDLINE/Pubmed, Cochrane Library, Embase and DOSS (Dentistry and Oral Sciences Source). No restriction in terms of year of publication was applied. Randomized controlled clinical trials were identified using the search terms “Bioglass” AND “periodontal defect” in the PubMed, Cochrane Library, Embase and DOSS. The search was complemented by a manual search of the references found in manuscripts to identify any additional articles of relevance. The grey literature was also scanned to broaden our search and improve the quality of the present review through title and abstract screening of all the included articles and Google Scholar database, yet retrieved no additional studies to be included. Eligibility criteria The studies were selected only if they met the following inclusion criteria: Randomized Clinical Trial (RCT), English-language publications, analysis of human teeth, use of BG for bone regeneration, at least 6 patients considered, presence of periodontal defect at the beginning of the study, at least 6 months of follow up, presence of pretreatment and post-treatment PD and/or CAL measures. In addition, the studies were required to have assessed the outcomes of interest (periodontal regeneration, bone regeneration). Studies not meeting all inclusion criteria were excluded. Also, reports based on questionnaires and interviews, hence studies without clinical examination of the patients, reviews, redundant publications and case reports were excluded. Study selection and data extraction The reviewers selected the articles included in the study considering the inclusion criteria. Disagreement was resolved by discussion with a third reviewer. The extracted data included authors, journal, year of publication, study design (randomized split-mouth design, randomized parallel trial, randomized parallel multicenter), number of participants in treatment and control groups, number of teeth in treatment and control groups, type of periodontal defect, inclusion criteria, exclusion criteria, treatment used as test, treatment used as control, follow-up period, outcomes considered and results obtained. The outcomes of interest considered in each study were periodontal and bone regeneration in terms of decrease of PD and gain of CAL. Each reviewer independently extracted all data from the finally selected articles and constructed tables on study characteristics and outcomes Authors of included studies were contacted by email to provide raw data, whenever necessary. Quality assessment Risk of bias assessment was autonomously performed by 2 reviewers (F.M. and C.M) using specific risk assessment tools depending on the study design. The overall quality of evidence at the outcome level was assessed using the Revised Cochrane Risk-of-bias Tool for Randomized Trials (RoB 2) as developed by Sterne et al. . This quality assessment is structured into the following domains: (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in the measurement of the outcome, (5) bias in the selection of the reported result. All five domains were judged as low risk of bias, some concerns or high risk of bias. Statistical analysis We computed the within-group standardized mean difference (SMD) Hedges’s g to compare the values of the same treatment group at two different time points where the time at baseline is the reference. Unbiased estimates of the sampling variances were calculated as described in Hedges . By taking the maximum value, missing standard deviations were imputed from the observed standard deviations from the same treatment group in other studies. Then the between-group difference in the two SMD values, namely SMD in the test group (SMD_T) minus SMD in the control group (SMD_C) was used as the effect size in the meta-analysis. A correlation of 0.5 was assumed to calculate the variance of the effect. This measure, SMD_T-SMD_C, indicates how much larger (or smaller) the change in the test group is (in standard deviation units) compared to the control group. Standard error of multi-arm studies has been corrected to account for effect size dependency. A network meta-analysis (NMA) was fitted, according to the graph theory methodology presented in Rücker, using a random effect model. Inconsistency between direct and indirect estimates was assessed by generalized Cochran’s Q statistics. The I^2 statistic, directly based on Cochran’s Q, was used to quantify between-study heterogeneity (i.e. the percentage of variability in the effect sizes that is not caused by sampling error). P-score was used to rank the treatments, in which a higher value indicates better performance. The comparison-adjusted funnel plot was used as a visual tool when investigating publication bias in a NMA . The horizontal axis shows the scatter of treatment effects estimated from individual studies, while the vertical axis shows standard error of the treatment estimates. If publication bias is present, as for any reporting bias, the plot will be asymmetrical. R software version 4.1.2 and R packages metafor and netmeta were used. This literature search was conducted to identify articles on the use of bioactive glass for periodontal and bone regeneration. In March 2021 an electronic search was performed in the following databases: MEDLINE/Pubmed, Cochrane Library, Embase and DOSS (Dentistry and Oral Sciences Source). No restriction in terms of year of publication was applied. Randomized controlled clinical trials were identified using the search terms “Bioglass” AND “periodontal defect” in the PubMed, Cochrane Library, Embase and DOSS. The search was complemented by a manual search of the references found in manuscripts to identify any additional articles of relevance. The grey literature was also scanned to broaden our search and improve the quality of the present review through title and abstract screening of all the included articles and Google Scholar database, yet retrieved no additional studies to be included. The studies were selected only if they met the following inclusion criteria: Randomized Clinical Trial (RCT), English-language publications, analysis of human teeth, use of BG for bone regeneration, at least 6 patients considered, presence of periodontal defect at the beginning of the study, at least 6 months of follow up, presence of pretreatment and post-treatment PD and/or CAL measures. In addition, the studies were required to have assessed the outcomes of interest (periodontal regeneration, bone regeneration). Studies not meeting all inclusion criteria were excluded. Also, reports based on questionnaires and interviews, hence studies without clinical examination of the patients, reviews, redundant publications and case reports were excluded. The reviewers selected the articles included in the study considering the inclusion criteria. Disagreement was resolved by discussion with a third reviewer. The extracted data included authors, journal, year of publication, study design (randomized split-mouth design, randomized parallel trial, randomized parallel multicenter), number of participants in treatment and control groups, number of teeth in treatment and control groups, type of periodontal defect, inclusion criteria, exclusion criteria, treatment used as test, treatment used as control, follow-up period, outcomes considered and results obtained. The outcomes of interest considered in each study were periodontal and bone regeneration in terms of decrease of PD and gain of CAL. Each reviewer independently extracted all data from the finally selected articles and constructed tables on study characteristics and outcomes Authors of included studies were contacted by email to provide raw data, whenever necessary. Risk of bias assessment was autonomously performed by 2 reviewers (F.M. and C.M) using specific risk assessment tools depending on the study design. The overall quality of evidence at the outcome level was assessed using the Revised Cochrane Risk-of-bias Tool for Randomized Trials (RoB 2) as developed by Sterne et al. . This quality assessment is structured into the following domains: (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in the measurement of the outcome, (5) bias in the selection of the reported result. All five domains were judged as low risk of bias, some concerns or high risk of bias. We computed the within-group standardized mean difference (SMD) Hedges’s g to compare the values of the same treatment group at two different time points where the time at baseline is the reference. Unbiased estimates of the sampling variances were calculated as described in Hedges . By taking the maximum value, missing standard deviations were imputed from the observed standard deviations from the same treatment group in other studies. Then the between-group difference in the two SMD values, namely SMD in the test group (SMD_T) minus SMD in the control group (SMD_C) was used as the effect size in the meta-analysis. A correlation of 0.5 was assumed to calculate the variance of the effect. This measure, SMD_T-SMD_C, indicates how much larger (or smaller) the change in the test group is (in standard deviation units) compared to the control group. Standard error of multi-arm studies has been corrected to account for effect size dependency. A network meta-analysis (NMA) was fitted, according to the graph theory methodology presented in Rücker, using a random effect model. Inconsistency between direct and indirect estimates was assessed by generalized Cochran’s Q statistics. The I^2 statistic, directly based on Cochran’s Q, was used to quantify between-study heterogeneity (i.e. the percentage of variability in the effect sizes that is not caused by sampling error). P-score was used to rank the treatments, in which a higher value indicates better performance. The comparison-adjusted funnel plot was used as a visual tool when investigating publication bias in a NMA . The horizontal axis shows the scatter of treatment effects estimated from individual studies, while the vertical axis shows standard error of the treatment estimates. If publication bias is present, as for any reporting bias, the plot will be asymmetrical. R software version 4.1.2 and R packages metafor and netmeta were used. The electronic search and other sources identified 46 records from PubMed/MEDLINE, Cochrane Library, Embase and DOSS (Dentistry and Oral Sciences Source). A total of 15 articles were duplicates and were removed. After screening the titles and abstracts, 11 articles were excluded. Therefore, following full-text reading, all the 20 articles met the defined inclusion criteria and were included in the review. A flow chart that illustrates the screening process is showed in Fig. . Authors of included studies were contacted by email to provide raw data, but no one responded to emails. According to Revised Cochrane Risk-of-bias Tool for Randomized Trials (RoB 2), the overall quality of evidence at the outcome level was assessed and displayed in Fig. . Only the studies of Leknes et al. and Sumer et al. had a low risk of bias for all domains . The method of randomization was clearly explained in 14 [ – ] out of 20 studies and it consisted in 11 cases [ – , – ] in a coin flip, in one case in the roll of a die, in another case in computer-generated random number list and in one study randomization was obtained by drawing a coded paper from a paper bag. Allocation concealment was described for 10 RCTs [ , , , , , – ] by attempting to ascertain the degree of masking. Nine [ , , , , – ] out of 20 studies described the use of an evaluator who was masked to the treatment group assignment in assessing the clinical measurements at follow-up. Eight studies [ – , – ] did not report any masking, and one study reported that all the treatments and measurements were performed by the sole investigator. The commercial name of the BG product used for the papers was clearly expressed in 14 studies [ , , , , , , – ]. In one study , the co-authors worked for the pharmaceutical company that produced the BG used in the study. Further, one study was supported by a grant from the company manufacturing the BG product analyzed, as well as one of the co-authors was employed for the same company. Only one study specifically declared: ‘‘The authors report no conflicts of interest related to this study. The study was self-funded by the authors and their institutions.’’. The 20 selected studies were published between 1997 and 2016. The main characteristics and results of the included studies are summarized in Table . In total, this review included 376 participants with 656 teeth with PD measurements and 327 participants with 558 teeth with CAL measurements. Mainly the studies considered patients with intrabony defects; four studies were of patients with furcation defects [ – , ], and one study included patients with both intrabony and furcation defects . Fourteen studies were randomized split-mouth designs [ – , , , – ]; four studies were randomized parallel trials [ , , , ] and the remaining two studies were randomized parallel multicenter studies . Papers reporting a change in PD and CAL were extremely heterogeneous since they considered different periods of follow-up and different treatments. The comparisons between different treatments included (1) BG, (2) OFD, (3) PLATELET RICH FIBRIN (PRF), (4) BG/GUIDED TISSUE REGENERATION (GTR), (5) PLATELET PELLET/GTR, (6) EMDOGAIN (EMD), (7) MEMBRANE, (8) EXPANDED POLYTETRAFLUOROETHYLENE (ePTFE), (9) Demineralized freeze-dried bone allograft (DFDBA), (10) BG + EMD, 11) AUTOGENOUS CORTICAL BONE. Each treament included OPEN FLAP DEBRIDEMENT (+ OFD). The meta-analysis examined 12 articles for PD [ – , , , , – ], and 10 for CAL [ , , , , , , – ]. All these papers reported complete data for the considered outcomes at 6 months. We have considered principally two outcomes: PD and CAL. Probing depth. The most common follow-up time was 6 months (13 studies), followed by one year (6 studies)[ , , , , , ]. Only 4[ , , , ] and 3[ , , ] studies reported data at the 3 and 9 month-time, respectively. The analysis focused on the evaluation at 6 months. Since OFD is the most common reference in most studies, it was contrasted with all available treatments, as shown in the network graph (Fig. ). Disconnected comparisons were excluded from the network analysis [ – , , , , – ]. With regard to the study performed by Rosenberg , we obtained the values of PD at 6 months by computing the available data. The heterogeneity in the network model is very high, with I 2 = 91.7% [86.1%; 95.1%]. Inconsistency is a minor concern, with the Q = 4.23 (p-value = 0.12) under the assumption of a full design-by-treatment interaction random effects model. For AUTOGENOUS CORTICAL BONE, BG and OFD, being in a closed loop in the network of evidence (i.e., there exists both direct and indirect information), the difference between the direct and indirect estimates is calculated. However, it does not appear to be statistically significant (p-value ranging from 0.18 to 0.72). Keeping in mind that a negative between-group SMD in PD favours the treatment arm, the forest plot (Fig. ) indicates that AUTOGENOUS CORTICAL BONE, BG and PRF are more efficacious than OFD. The network effect of BG is completely driven by direct evidence [ , , , , , , ], while only one and no direct estimate contributes to AUTOGENOUS CORTICAL BONE and PRF network estimate, respectively. According to the p-score, PRF (p-score = 0.96), AUTOGENOUS CORTICAL BONE (p-score = 0.72) and BG (p-score = 0.55) rank first second and third, respectively. Clinical attachment level . The most common follow-up time was 6 months (10 studies), followed by one year (5 studies)[ , , , ]. Only 2 studies considered the 3 and 9 month-time. The study performed by El-Haddad was excluded cause of inconsistencies in the dates reported for CAL at three months. As for PD, the analysis focused on the evaluation at 6 months, and OFD was contrasted with all available treatments, as shown in the network graph (Fig. ). The heterogeneity in the network model is very high, with I^2 = 91.3% [80.8%; 96.0%]. No closed loops are present in the network and inconsistency cannot be assessed. The forest plot in Fig. indicates that PRF is more efficacious than OFD alone (p-value < 0.001). The only effect completely driven by direct evidence is BG vs. OFD,[ , , , ] but it is not statistically significant (p-value = 0.39). According to the p-score, PRF (p-score = 0.99), MEMBRANE and BG (p-score = 0.56 for both) rank first and second, respectively. The visual inspection of the funnel plot shown in Fig. revealed no significant evidence of publication bias. However, the assessment through this qualitative tool is hampered by the small number of trials. Numerous CTs in the dental literature have focused on the efficacy of BG in treating periodontal bone defects. Two previous systematic reviews summarized the topic in 2002 and 2012. Trombelli et al. observed a significant weighted mean difference of 1.04 mm in CAL gain compared to the OFD, while Sohrabi et al. reported a difference between BG and controls (active or OFD) in change in PD and CAL from baseline to follow-up of 0.52 and 0.60 mm, respectively. In ten years, however, further evidence has been collected making it advisable to summarize the literature again. The present study retrieved 20 RCTs published between 1997 and 2016. As regards the PD at 6 months, AUTOGENOUS CORTICAL BONE, BG and PRF were more efficacious than OFD alone, with a statistically significant SMD equal to -1.57, -1.06 and − 2.89, respectively. As to CAL at 6 months, the effect of BG is reduced and no longer significant (SMD= -0.19, p-value = 0.4) and curiously PRF was more efficacious than OFD (SMD=-4.13, p-value < 0.001) in CAL gain, but in indirect evidence. This outcome is consistent with literature . Differently from previous reviews, this study examined the risk of bias following the Rob2 scale and pooled evidence according to a NMA approach. Only two studies had a very low risk of bias, while five [ , , , , ] were at high risk of bias. In addition, NMA allowed for a wider picture of the evidence and understanding multiple interventions’ relative merits. In fact, the use of BG was not compared only to OFD, which is considered the standard procedure to treat osseous defects, but rather a series of different comparisons were performed. NMA has advantages over conventional pairwise meta-analysis, as the technique borrows strength from indirect evidence to gain certainty about all treatment comparisons and allows for estimation of comparative effects that have not been investigated head to head in RCTs. These results are somehow less favorable to BG than the aforementioned reviews. Furthermore the only statistically relevant improvement that has been detected regards PD and no statistically significant effect was detected regarding CAL. This may depend on mixed resonsreasons. One may suggest that the different techniques tested may play a key role since different clinical situations were approached with different techniques. Many clinicians consider autografts as the gold-standard, because of their favourable biological characteristics that are osteoinduction and osteoconduction. However they present several drawbacks as, for example, the limited availability and the donor-site morbidity . Allografts and xenografts may trigger immune rejection, allow disease transmission and be less osteoinductive than autografts owing to disruptive processing . Favoured by the advances in bone tissue engineering, artificial scaffolds are very promising but often show low fusion rates due to reduced cell ingrowth and local inflammation upon degradation . However they present several drawbacks as, for example, the limited availability and the donor-site morbidity . Allografts and xenografts may trigger immune rejection, allow disease transmission and be less osteoinductive than autografts owing to disruptive processing . Favoured by the advances in bone tissue engineering, artificial scaffolds are very promising but often show low fusion rates due to reduced cell ingrowth and local inflammation upon degradation . The BG products evaluated in the summarised studies appeared biocompatible as no reports of adverse effects (i.e. allergies, other immunologic reactions, abscess formation) were made. However, the main drawback inherent in these BG materials is their brittleness hindering their use as scaffolds. To overcome this limitation, new bio-mimicking materials combining the mechanical features of tailored synthetic polymers and the bioactive element of BG were developed . Indeed, several bio-hybrid composites have been studied with positive outcomes in vitro and in vivo , but only a few have emerged successfully so far such as calcium-phosphate/poly-ε-caprolactone particles , silicon carbide/collagen scaffolds (BioSiC) , poly(N-acryloyl 2-glycine)/methacrylated gelatin hydrogels . An example of successful clinical translation which resulted in a CE-marked product currently in use is SmartBone® , “a bovine-derived mineral matrix reinforced with resorbable poly(lactic-co-caprolactone) block copolymer embedding RGD-exposing collagen fragments onto its surface” . The RCTs included in this review were heterogeneous in terms of defect types (furcation [ – , ], intrabony [ , – , – ] or both intrabony and furcation ) and control interventions (OFD alone [ , , , , , , , , , ], PRF , platelet pellet/GTR , EMD , membrane , ePTFE , DFDBA , autogenous cortical bone . Three articles [ , , ] compared the use of EMD alone or in combination with a BG. Clearly, before all the treatments analyzed always occurs a debridement of the defects, meaning that an OFD always goes with the treatment proposed. Fourteen studies were randomized split-mouth designs [ – , , , – ]; four studies were randomized parallel trials [ , , , ] and the remaining two studies were randomized parallel multicenter studies . In split-mouth RCTs, subjects are their own control, which is supposed to reduce the variability of outcome among patients from the intervention effect estimate virtually leading to an increase in statistical power. Although selection bias is avoided and masking is easier in split mouth studies, cross-over effects may be not negligible limiting the difference in outcome between interventions. The split mouth RCTs included here did not address the issue of possible carry-over effects sustained by bioglass, but according to literature they should be irrelevant . The heretogeneity sources of the studies included in this systematic review are multiple and difficult to assess: in particular the defect type, the patient features, the surgical procedures implemented, the experience of the operator, etc. Future systematic reviews would benefit greatly from studies based on protocols registered before conducting the research. The possible advantage of this choice is twofold: (1) to attain, as much as possible, homogeneous data that may be compared more easily (to reduce the remarkable aforementioned biases and the avoidable waste of data in literature) and (2) to promote the publication of whatever results may be retrieved, thus overcoming the publication bias, which is currently difficult to be taken into consideration. Data of the present review only partially support the clinical efficacy of the usage of BG in the bone regeneration treatments for periodontal purposes. Indeed, the SMD of 0.5 to 1 in PD and CAL obtained with BG compared to OFD alone seem clinically insignificant. However, the absence of the evidence of efficacy does not mean the absence of efficacy. Some sites have reported much more significant clinical and statistical changes, whereas other sites have had smaller changes or even negative results. More good quality CTs are required to provide sound evidence on the clinical efficacy of BG.
Structural mechanism of human oncochannel TRPV6 inhibition by the natural phytoestrogen genistein	6ca357bd-67be-4205-a09b-1155e9c32331	10169861	Physiology[mh]	TRPV6 is a representative of the vanilloid subfamily of transient receptor potential (TRP) channels that serves as an entry gate for capturing dietary calcium ions in the gut – . TRPV6 mutations and abnormal expression of this channel – have been linked to a range of human diseases associated with disturbed calcium homeostasis, including transient neonatal hyperparathyroidism, undermineralization, and dysplasia of the human skeleton, hypercalciuria, chronic pancreatitis, various reproductive diseases, Pendred syndrome and Crohn’s-like disease , – . Since calcium uptake is linked to cell proliferation and cancer progression, TRPV6 was also declared an oncochannel – . Indeed, TRPV6 was found to overexpress in some of the most severe human cancers, including leukemia, breast, prostate, colon, ovarian, thyroid, and endometrial cancers – , , . In addition, an ancestral variant of this oncochannel has emerged as a driver of higher incidence, higher mortality, and more aggressive forms of different cancer types in people of African descent . Inhibitors of TRPV6 are therefore in high demand. While several synthetic inhibitors of TRPV6 have been making a slow progress towards clinical trials – , – , scientific exploration and pharmaceutical exploitation of natural TRPV6 inhibitors have been largely neglected despite their apparent benefits, such as pharmacokinetics optimized by nature in the course of evolution . The natural isoflavone and phytoestrogen genistein (4′,5,7-trihydroxyisoflavone) was previously suggested to act as a TRPV6 inhibitor . Genistein is a common precursor in the biosynthesis of antimicrobial phytoalexins and phytoanticipins in legumes and as a predominant isoflavone in nutritional soy products it has a potential to be the major component of an individual’s diet – . Importantly, dietary genistein shows a range of potential health-beneficial effects, including the ability to inhibit cell invasion and metastasis in various forms of human cancer , , – . Beyond treatment of prostate, colon, kidney, pancreatic, ovarian, breast, and lung cancers , – , therapeutic potential of genistein extends to the treatment of cardiovascular diseases – , post-menopausal , and gastrointestinal ailments, and bone loss – . Genistein has been investigated in 75 clinical trials (clinicaltrials.gov), where it shows antimetastatic efficacy and positive effects in treatment of metabolic syndrome . In this study, we explore the molecular basis of human TRPV6 (hTRPV6) inhibition by the natural phytoestrogen genistein extracted from Styphnolobium japonicum . Using cryo-electron microscopy (cryo-EM) combined with calcium imaging, electrophysiology, mutagenesis, and molecular dynamics (MD) simulations we show that genistein binds in the intracellular half of the hTRPV6 pore and acts as an ion channel blocker and gating modifier. Upon binding to the open pore of hTRPV6, genistein converts it into a non-conducting conformation with a two-fold symmetrical arrangement of the pore-forming segments. This conformation is likely stabilized by two putative metal binding sites at the pore intracellular entry and has S4–S5 and S6-TRP regions, which typically include two α-helices, transformed into three helices-containing segments in two diagonal subunits. The unusual mechanism of TRPV6 inhibition by genistein lays the foundations for the development of much-needed new drugs targeting TRPV6-associated diseases, including cancers. Functional characterization of hTRPV6 inhibition by genistein TRPV6 is a constitutively open ion channel – . In response to a –100 to +70 mV voltage ramp, whole-cell patch-clamp recordings from HEK 293S cells expressing wild-type human hTRPV6 showed inward-rectified currents (Fig. ), typical for this ion channel – . In the presence of genistein, the TRPV6-mediated currents were reduced. Thus, at –60 mV membrane potential, 50 µM genistein produced 54 ± 4% (mean ± SEM, n = 13) inhibition of the hTRPV6-mediated current. Measurements of the concentration dependence of the hTRPV6-mediated current inhibition yielded the half-maximal inhibitory concentration, IC 50 = 40.7 ± 2.6 µM, and the Hill coefficient, n Hill = 1.80 ± 0.13 ( n = 7, Supplementary Fig. ). We also monitored hTRPV6 inhibition by genistein using Fura-2-based measurements of changes in intracellular Ca 2+ . Changes in the fluorescence intensity ratio at the excitation wavelengths 340 and 380 nm ( F 340 / F 380 ) evoked by addition of 10 mM Ca 2+ were measured after pre-incubation of hTRPV6-expressing HEK 293S cells in different concentrations of genistein (Fig. ). Genistein inhibited hTRPV6-mediated Ca 2+ uptake with the values of IC 50 = 113.2 ± 4.7 µM and n Hill = 0.77 ± 0.02 ( n = 3, Fig. ). Cryo-EM analysis of hTRPV6 in the presence of genistein Purified hTRPV6 protein was supplemented with 2 mM genistein and subjected to cryo-EM analysis (Methods; Supplementary Fig. ). Collected cryo-EM micrographs showed evenly dispersed particles of hTRPV6 (Supplementary Fig. ). Processing of the data revealed two distinct populations of particles, both showing diverse angular coverage. The corresponding 2D-class averages demonstrated clearly visible secondary structure elements, supporting the high quality of the collected cryo-EM data (Supplementary Fig. ). The first, smaller population of particles yielded a 2.71-Å cryo-EM map that showed four-fold rotational symmetry (C4) and represented a typical genistein-free apo state, hTRPV6 Open (Fig. , Supplementary Fig. and Supplementary Table ). The second, predominant population of particles yielded a 2.66-Å 3D reconstruction with no symmetry imposed (C1), which had two densities of the size of a genistein molecule that had never been seen in TRPV6 reconstructions before (Fig. , Supplementary Figs. , g, i and , Supplementary Table ). Further analysis showed that this second reconstruction represents a genistein-bound inhibited state of the channel, hTRPV6 GEN . For the 4-fold symmetrical hTRPV6 Open homotetramer, we built a molecular model of a single subunit, including residues 28–637 and excluding residues 1–27 (N-terminus) and 638–725 (C-terminus), which were not clearly resolved in the cryo-EM map. We also built a molecular model for each of the four subunits of the hTRPV6 GEN homotetramer, including residues 27–638 and excluding residues 1–26 (N-terminus) and 639–725 (C-terminus) that were not clearly resolved in the corresponding cryo-EM density. The transmembrane regions of hTRPV6 Open and hTRPV6 GEN were surrounded by numerous non-protein auxiliary densities, which we modeled as annular lipids (Fig. ). Due to high quality of hTRPV6 Open and hTRPV6 GEN reconstructions, three such densities per subunit were identified as representing cholesteryl hemisuccinate (CHS), which was added to the protein sample during purification to enhance protein stability (see Methods). The CHS sites are likely to bind cholesterol in vivo. hTRPV6 Open and hTRPV6 GEN structures The structures of hTRPV6 Open and hTRPV6 GEN (Fig. ) have a similar overall architecture to the previously determined structures of TRPV6 , , , . In a nutshell, hTRPV6 is assembled of four subunits and contains a transmembrane domain (TMD) with a central ion channel pore and an intracellular skirt that is mostly built of ankyrin repeat domains connected by the three-stranded β-sheets, N-terminal helices, and C-terminal hooks . Amphipathic TRP helices run nearly parallel to the membrane and interact with both the TMD and the skirt. The TMD is composed of six transmembrane helices S1–S6 and a re-entrant pore loop (P-loop) between S5 and S6. A bundle of the first four transmembrane helices represents the S1–S4 domain, which in voltage-gated ion channels forms a voltage sensor . The pore domain of each subunit includes S5, P-loop, and S6, and is packed against the S1–S4 domain of the neighboring subunit in a domain-swapped arrangement , . Despite the overall similarity (Fig. ), hTRPV6 Open and hTRPV6 GEN structures have different conformations of the ion-conducting pore (Fig. ). As in all representatives of the tetrameric ion channel family, the ion-conducting pore of TRPV6 has two narrow regions, the selectivity filter formed by the extended regions of the P-loop and the gate formed by the S6 bundle crossing, separated by the central cavity in the middle. The selectivity filter of TRPV6 is a structural element responsible for high calcium selectivity and in both structures, it adapts a conformation nearly identical to the previously published TRPV6 structures. Interestingly, however, the gate region shows drastically different conformations. In hTRPV6 Open , the pore is wide open (Fig. ) and its narrow region is lined by residues N572 and I575 (Fig. ). This conformation is the same as in the previously published open-state structures of TRPV6, independent of whether they had C-terminus truncated or not and whether the protein was purified in different types of detergents, nanodiscs or amphipols (Supplementary Fig. ), strongly supporting our initial assignment of hTRPV6 Open to the open conducting state. In contrast to hTRPV6 Open , the gate region in hTRPV6 GEN is narrow and fully sealed by the side chains of L574 and M578 (Fig. ). Similar pore narrowing was observed in the closed-state structures of TRPV6, independent of whether they were apo-state structures , , or structures obtained in the presence of inhibitors 2-APB or econazole, or ion channel blocker ruthenium red (RR) , (Fig. ). The characteristic feature of the TRPV6 closed-state is α-helical S6 (Fig. ). In the open state, S6 undergoes an α-to-π transition, which produces a π-bulge in the middle of this helix (Fig. ). Formation of the π-bulge is accompanied by a ~100° rotation of the C-terminal portion of S6, which brings a completely different set of residues to face the pore and opens it for ion conduction , . Two-fold symmetry of hTRPV6 GEN structure A closer look at the hTRPV6 GEN structure reveals non-equivalency of the two pairs of diagonal subunits, A/C and B/D (Fig. ). Overall, the structure is ~4-fold symmetrical but shows noticeable deviations from the 4-fold symmetry at the intracellular region of the ion channel pore, around the two putative sites of genistein binding (Fig. ). The exact regions that underlie the deviation from the 4-fold symmetry can be easily pinpointed by superposition of individual hTRPV6 GEN subunits (Fig. ) and include the S4–S5 linker (Fig. ) and S6-TRP helix connection (Fig. ). Indeed, the conformations of the S4–S5 linker and S6-TRP helix connection appear to be similar within A/C and B/D pairs of the diagonal subunits and different between these two pairs (Fig. and Supplementary Movie ). In the A/C pair, the S4–S5 region includes a continuous helical segment with two sections, S4–S5 linker and S5, tilted with respect to each other by ~28° at around W495 (Fig. ). This continuous helical segment is present in all previously published 4-fold symmetrical structures of TRPV6 but the tilt angle, ~51°, is much larger than in A and C subunits of hTRPV6 GEN (Fig. ). As a consequence of the smaller angular tilt, the unfolded region connecting S4 to the S4–S5 linker in A and C subunits of hTRPV6 GEN (Fig. ) is much larger than the corresponding region in the 4-fold symmetrical structures of TRPV6 (Fig. ). The conformation of the S4–S5 region in subunits B and D of hTRPV6 GEN shows an even more drastic difference from the 4-fold symmetrical structures. In this case, S5 is straight, elongated, and separated from the helical part of the S4–S5 linker by an unfolded stretch of polypeptide (Fig. ). Correspondingly, the S4–S5 region, which includes only two continuous helical segments in all published structures of TRPV6 (Fig. ), in B and D subunits of hTRPV6 GEN transforms into the region with three helical segments (Fig. and Supplementary Movie ). The S6-TRP helix region in subunits A and C of hTRPV6 GEN includes two continuous helical segments (Fig. ), similar to the corresponding region in the previously published 4-fold symmetrical structures of TRPV6 (Fig. ). However, the α-helical S6 in the closed-state structures is one-helical turn shorter than S6 in subunits A and C of hTRPV6 GEN , while the similarly long S6 in the previously published open-state structures contains the π-bulge in the middle. The elongation of S6 in subunits A and C of the closed-pore hTRPV6 GEN results in one-helical turn shortening of the TRP helix and elongation of the unfolded region connecting S6 to the TRP helix (Fig. ). In subunits B and D of hTRPV6 GEN , the S6-TRP helix region also shows a drastic difference from the corresponding region in the 4-fold symmetrical structures. In this case, the TRP helix maintains approximately the same size as in the previously published closed-state structures, while α-helical S6 splits around M578 into two helical segments, S6 and S6-TRP (Fig. and Supplementary Movie). Thus, both the S4–S5 and S6-TRP helix regions in hTRPV6 GEN adapt the conformations different from all previously published structures of TRPV6. Genistein binding sites Close inspection of the hTRPV6 GEN cryo-EM map revealed two densities of the shape of a genistein molecule at the intracellular half of the ion channel pore that were not present in the hTRPV6 Open map (Fig. , Supplementary Fig. ). When fitted into these densities, both molecules of genistein appear to be located at the central pore axis and oriented with their long axis perpendicular to the pore axis. The upper site (1) is located above the gate, at the bottom of the central cavity, and is contributed by the S6 hydrophobic residues M570, L571 and L574 (Fig. ). The bottom site (2) is right below the gate, at the intracellular entrance to the ion channel pore, and contributed by the residues M578, G579, H582, W583 and A586 (Fig. ). The cryo-EM density at site 2 is somewhat weaker than at site 1. Since genistein was purified from a natural source, we cannot exclude the possibility that the lower site represents a bound contaminant. However, we believe that such a possibility is highly unlikely given the resemblance of the density with the molecule of genistein and the high degree (~99%) of the reagent purity, which was verified using multiple methods. CHS is also an unlikely candidate to represent the density in site 2 as this density does not resemble the shape of CHS and it only appears when genistein is added to the sample, while CHS is always present in all our TRPV6 preparations. Due to the orientation of both genistein molecules perpendicular to the ion channel central axis, the two pairs of hTRPV6 GEN diagonal subunits A/C and B/D contribute a different number of residues to genistein binding. Correspondingly, different strength of genistein interaction with the two pairs of diagonal subunits is the likely cause of asymmetric pore transformation (Fig. ). This transformation appears to be stabilized by the formation of two asymmetrically located metal ion binding sites at the intracellular pore entry. Putative metal binding sites at the hTRPV6 GEN pore intracellular entry Two strong densities at the pore intracellular entrance, located two-fold symmetrically relative to the central pore axis, are observed in the cryo-EM density of hTRPV6 GEN (Supplementary Fig. ). Each of these densities is in the middle of three-histidine clusters, which include H587 of the B/D diagonal subunits pair as well as H582 and H587 of the A/C diagonal subunits pair (Fig. ). The distances from the center of the density to histidines (Supplementary Fig. ) are 2.3–2.5 Å (H587 in B/D), 2.2–2.4 Å (H582 in A/C) and 2.2 Å (H587 in A/C), which are typical for histidine coordination of Zn 2+ ions – , although other metal ions, such as Mg 2+ , Ca 2+ or Na + , can be coordinated as well – . These metal ions (M) are either the main components of our purification buffers (Na + ) or likely impurities of either the buffer components or genistein, which was extracted from Styphnolobium japonicum . To test possible contribution of the putative metal ions to TRPV6 inhibition by genistein, we introduced alanine mutations of H582, H587, or both and used Fura-2 fluorescent measurements to estimate changes in calcium uptake through the corresponding mutant channels at different genistein concentrations (Fig. ). Genistein inhibition of calcium uptake through the H582A mutant was weaker than through wild-type channels as indicated by the rightward shift of the genistein concentration-dependence. A stronger shift in the concentration-dependence was observed for H587A and H582A/H587A mutants, with a little difference between the two, consistent with the predominant contribution (two out of three) of H587 to the putative metal ion coordination (Fig. and Supplementary Fig. ). Therefore, our mutagenesis combined with functional recordings are consistent with the putative metal ion coordination by the three-histidine clusters being an additional factor that might help to stabilize the asymmetric conformation of hTRPV6 GEN . MD simulations of genistein binding sites To further validate genistein binding to sites 1 and 2, we carried out all-atom MD simulations (Fig. , Supplementary Figs. - ). Four simulations were performed for site 1 with different initial orientations of the ligand, which differed by 180° rotations around the long genistein axis and around the central pore axis. Our simulations showed that the ligand is stabilized in site 1 by two hydrogen bonds between the hydroxyl groups of genistein and the carbonyl oxygens of M570 in subunits A/C (Supplementary Fig. ). These interactions orient the ligand in either one of two symmetrical positions rotated by 180° around the central pore axis, thus demonstrating an excellent fit to the non-protein cryo-EM density in site 1 (Fig. ). Similar simulations were repeated with CHS embedded in site 1. CHS showed no specific interactions with the protein, was much more mobile than genistein, and intended to escape from site 1. The MD-predicted density for CHS formed a “band” across the pore, inconsistent with the cryo-EM density (Supplementary Fig. ). MD simulations of a single genistein molecule placed at site 2 starting with the position modeled in the cryo-EM structure (Fig. ) revealed highly dynamic behavior of this molecule, only somewhat restrained by π-stacking interactions with W583 (not shown). To stabilize the genistein molecule at this cryo-EM-like primary position, we added another genistein molecule at the secondary position between the W583 side chains below and perpendicular to the primary position (Supplementary Fig. ). We performed four MD simulations with different initial orientations of genistein molecules in the primary and secondary positions at site 2, generated in the same way as in simulations for site 1. The simulations showed that genistein in the primary position of site 2 is more mobile than genistein in site 1. However, in three out of four MD runs, genistein was stabilized in the primary position of site 2 by π-stacking interactions with W583 in subunits B/D or H582 in subunits A/C as well as by one or two hydrogen bonds with the carbonyl oxygens of the neighboring residues M578 and G579 in subunits A/C or residue I575 in subunits B/D (Supplementary Fig. ). Such a diversity of interaction partners did not allow to reveal specific positions of the ligand in the site. However, the resulting distribution of the ligand density at the primary position in site 2 averaged over the MD runs demonstrated excellent agreement with the central non-protein cryo-EM density (Fig. ). Genistein at the secondary position in site 2 was much more mobile than the genistein molecule at site 1 or in the primary position at site 2 (Supplementary Fig. ). Nevertheless, the genistein molecule at the secondary position in site 2 tended to form two to four π-stacking interactions with W583 in subunits B/D and H582 in subunits A/C as well as hydrogen bonds with the side chains of D580 in subunits B/D. In turn, these interactions imposed restraints on the genistein molecule at the primary position in site 2. The cumulative behavior of two genistein molecules at site 2 agrees well with the cryo-EM data, where a group of small non-protein densities between W583 side chains are likely to represent the ensemble of positions of the weakly interacting ligands (Supplementary Fig. ). MD simulations with CHS embedded in the primary position of site 2 in four different initial orientations revealed that the ligand is stabilized in this site much better than in site 1 (Supplementary Fig. ). Pockets between the S6 helices of neighboring subunits provide enough space for the CHS molecule, while its acidic group forms polar interactions with K484 or R589. However, being tightly packed in site 2, the CHS molecule created a much more extended density in MD simulations than the density revealed by cryo-EM (Supplementary Fig. ). Accordingly, while there is a possibility that the experimentally observed non-protein density in site 2 represents CHS or a similar size small molecule, the probability of this is low. TRPV6 is a constitutively open ion channel – . In response to a –100 to +70 mV voltage ramp, whole-cell patch-clamp recordings from HEK 293S cells expressing wild-type human hTRPV6 showed inward-rectified currents (Fig. ), typical for this ion channel – . In the presence of genistein, the TRPV6-mediated currents were reduced. Thus, at –60 mV membrane potential, 50 µM genistein produced 54 ± 4% (mean ± SEM, n = 13) inhibition of the hTRPV6-mediated current. Measurements of the concentration dependence of the hTRPV6-mediated current inhibition yielded the half-maximal inhibitory concentration, IC 50 = 40.7 ± 2.6 µM, and the Hill coefficient, n Hill = 1.80 ± 0.13 ( n = 7, Supplementary Fig. ). We also monitored hTRPV6 inhibition by genistein using Fura-2-based measurements of changes in intracellular Ca 2+ . Changes in the fluorescence intensity ratio at the excitation wavelengths 340 and 380 nm ( F 340 / F 380 ) evoked by addition of 10 mM Ca 2+ were measured after pre-incubation of hTRPV6-expressing HEK 293S cells in different concentrations of genistein (Fig. ). Genistein inhibited hTRPV6-mediated Ca 2+ uptake with the values of IC 50 = 113.2 ± 4.7 µM and n Hill = 0.77 ± 0.02 ( n = 3, Fig. ). Purified hTRPV6 protein was supplemented with 2 mM genistein and subjected to cryo-EM analysis (Methods; Supplementary Fig. ). Collected cryo-EM micrographs showed evenly dispersed particles of hTRPV6 (Supplementary Fig. ). Processing of the data revealed two distinct populations of particles, both showing diverse angular coverage. The corresponding 2D-class averages demonstrated clearly visible secondary structure elements, supporting the high quality of the collected cryo-EM data (Supplementary Fig. ). The first, smaller population of particles yielded a 2.71-Å cryo-EM map that showed four-fold rotational symmetry (C4) and represented a typical genistein-free apo state, hTRPV6 Open (Fig. , Supplementary Fig. and Supplementary Table ). The second, predominant population of particles yielded a 2.66-Å 3D reconstruction with no symmetry imposed (C1), which had two densities of the size of a genistein molecule that had never been seen in TRPV6 reconstructions before (Fig. , Supplementary Figs. , g, i and , Supplementary Table ). Further analysis showed that this second reconstruction represents a genistein-bound inhibited state of the channel, hTRPV6 GEN . For the 4-fold symmetrical hTRPV6 Open homotetramer, we built a molecular model of a single subunit, including residues 28–637 and excluding residues 1–27 (N-terminus) and 638–725 (C-terminus), which were not clearly resolved in the cryo-EM map. We also built a molecular model for each of the four subunits of the hTRPV6 GEN homotetramer, including residues 27–638 and excluding residues 1–26 (N-terminus) and 639–725 (C-terminus) that were not clearly resolved in the corresponding cryo-EM density. The transmembrane regions of hTRPV6 Open and hTRPV6 GEN were surrounded by numerous non-protein auxiliary densities, which we modeled as annular lipids (Fig. ). Due to high quality of hTRPV6 Open and hTRPV6 GEN reconstructions, three such densities per subunit were identified as representing cholesteryl hemisuccinate (CHS), which was added to the protein sample during purification to enhance protein stability (see Methods). The CHS sites are likely to bind cholesterol in vivo. Open and hTRPV6 GEN structures The structures of hTRPV6 Open and hTRPV6 GEN (Fig. ) have a similar overall architecture to the previously determined structures of TRPV6 , , , . In a nutshell, hTRPV6 is assembled of four subunits and contains a transmembrane domain (TMD) with a central ion channel pore and an intracellular skirt that is mostly built of ankyrin repeat domains connected by the three-stranded β-sheets, N-terminal helices, and C-terminal hooks . Amphipathic TRP helices run nearly parallel to the membrane and interact with both the TMD and the skirt. The TMD is composed of six transmembrane helices S1–S6 and a re-entrant pore loop (P-loop) between S5 and S6. A bundle of the first four transmembrane helices represents the S1–S4 domain, which in voltage-gated ion channels forms a voltage sensor . The pore domain of each subunit includes S5, P-loop, and S6, and is packed against the S1–S4 domain of the neighboring subunit in a domain-swapped arrangement , . Despite the overall similarity (Fig. ), hTRPV6 Open and hTRPV6 GEN structures have different conformations of the ion-conducting pore (Fig. ). As in all representatives of the tetrameric ion channel family, the ion-conducting pore of TRPV6 has two narrow regions, the selectivity filter formed by the extended regions of the P-loop and the gate formed by the S6 bundle crossing, separated by the central cavity in the middle. The selectivity filter of TRPV6 is a structural element responsible for high calcium selectivity and in both structures, it adapts a conformation nearly identical to the previously published TRPV6 structures. Interestingly, however, the gate region shows drastically different conformations. In hTRPV6 Open , the pore is wide open (Fig. ) and its narrow region is lined by residues N572 and I575 (Fig. ). This conformation is the same as in the previously published open-state structures of TRPV6, independent of whether they had C-terminus truncated or not and whether the protein was purified in different types of detergents, nanodiscs or amphipols (Supplementary Fig. ), strongly supporting our initial assignment of hTRPV6 Open to the open conducting state. In contrast to hTRPV6 Open , the gate region in hTRPV6 GEN is narrow and fully sealed by the side chains of L574 and M578 (Fig. ). Similar pore narrowing was observed in the closed-state structures of TRPV6, independent of whether they were apo-state structures , , or structures obtained in the presence of inhibitors 2-APB or econazole, or ion channel blocker ruthenium red (RR) , (Fig. ). The characteristic feature of the TRPV6 closed-state is α-helical S6 (Fig. ). In the open state, S6 undergoes an α-to-π transition, which produces a π-bulge in the middle of this helix (Fig. ). Formation of the π-bulge is accompanied by a ~100° rotation of the C-terminal portion of S6, which brings a completely different set of residues to face the pore and opens it for ion conduction , . GEN structure A closer look at the hTRPV6 GEN structure reveals non-equivalency of the two pairs of diagonal subunits, A/C and B/D (Fig. ). Overall, the structure is ~4-fold symmetrical but shows noticeable deviations from the 4-fold symmetry at the intracellular region of the ion channel pore, around the two putative sites of genistein binding (Fig. ). The exact regions that underlie the deviation from the 4-fold symmetry can be easily pinpointed by superposition of individual hTRPV6 GEN subunits (Fig. ) and include the S4–S5 linker (Fig. ) and S6-TRP helix connection (Fig. ). Indeed, the conformations of the S4–S5 linker and S6-TRP helix connection appear to be similar within A/C and B/D pairs of the diagonal subunits and different between these two pairs (Fig. and Supplementary Movie ). In the A/C pair, the S4–S5 region includes a continuous helical segment with two sections, S4–S5 linker and S5, tilted with respect to each other by ~28° at around W495 (Fig. ). This continuous helical segment is present in all previously published 4-fold symmetrical structures of TRPV6 but the tilt angle, ~51°, is much larger than in A and C subunits of hTRPV6 GEN (Fig. ). As a consequence of the smaller angular tilt, the unfolded region connecting S4 to the S4–S5 linker in A and C subunits of hTRPV6 GEN (Fig. ) is much larger than the corresponding region in the 4-fold symmetrical structures of TRPV6 (Fig. ). The conformation of the S4–S5 region in subunits B and D of hTRPV6 GEN shows an even more drastic difference from the 4-fold symmetrical structures. In this case, S5 is straight, elongated, and separated from the helical part of the S4–S5 linker by an unfolded stretch of polypeptide (Fig. ). Correspondingly, the S4–S5 region, which includes only two continuous helical segments in all published structures of TRPV6 (Fig. ), in B and D subunits of hTRPV6 GEN transforms into the region with three helical segments (Fig. and Supplementary Movie ). The S6-TRP helix region in subunits A and C of hTRPV6 GEN includes two continuous helical segments (Fig. ), similar to the corresponding region in the previously published 4-fold symmetrical structures of TRPV6 (Fig. ). However, the α-helical S6 in the closed-state structures is one-helical turn shorter than S6 in subunits A and C of hTRPV6 GEN , while the similarly long S6 in the previously published open-state structures contains the π-bulge in the middle. The elongation of S6 in subunits A and C of the closed-pore hTRPV6 GEN results in one-helical turn shortening of the TRP helix and elongation of the unfolded region connecting S6 to the TRP helix (Fig. ). In subunits B and D of hTRPV6 GEN , the S6-TRP helix region also shows a drastic difference from the corresponding region in the 4-fold symmetrical structures. In this case, the TRP helix maintains approximately the same size as in the previously published closed-state structures, while α-helical S6 splits around M578 into two helical segments, S6 and S6-TRP (Fig. and Supplementary Movie). Thus, both the S4–S5 and S6-TRP helix regions in hTRPV6 GEN adapt the conformations different from all previously published structures of TRPV6. Close inspection of the hTRPV6 GEN cryo-EM map revealed two densities of the shape of a genistein molecule at the intracellular half of the ion channel pore that were not present in the hTRPV6 Open map (Fig. , Supplementary Fig. ). When fitted into these densities, both molecules of genistein appear to be located at the central pore axis and oriented with their long axis perpendicular to the pore axis. The upper site (1) is located above the gate, at the bottom of the central cavity, and is contributed by the S6 hydrophobic residues M570, L571 and L574 (Fig. ). The bottom site (2) is right below the gate, at the intracellular entrance to the ion channel pore, and contributed by the residues M578, G579, H582, W583 and A586 (Fig. ). The cryo-EM density at site 2 is somewhat weaker than at site 1. Since genistein was purified from a natural source, we cannot exclude the possibility that the lower site represents a bound contaminant. However, we believe that such a possibility is highly unlikely given the resemblance of the density with the molecule of genistein and the high degree (~99%) of the reagent purity, which was verified using multiple methods. CHS is also an unlikely candidate to represent the density in site 2 as this density does not resemble the shape of CHS and it only appears when genistein is added to the sample, while CHS is always present in all our TRPV6 preparations. Due to the orientation of both genistein molecules perpendicular to the ion channel central axis, the two pairs of hTRPV6 GEN diagonal subunits A/C and B/D contribute a different number of residues to genistein binding. Correspondingly, different strength of genistein interaction with the two pairs of diagonal subunits is the likely cause of asymmetric pore transformation (Fig. ). This transformation appears to be stabilized by the formation of two asymmetrically located metal ion binding sites at the intracellular pore entry. GEN pore intracellular entry Two strong densities at the pore intracellular entrance, located two-fold symmetrically relative to the central pore axis, are observed in the cryo-EM density of hTRPV6 GEN (Supplementary Fig. ). Each of these densities is in the middle of three-histidine clusters, which include H587 of the B/D diagonal subunits pair as well as H582 and H587 of the A/C diagonal subunits pair (Fig. ). The distances from the center of the density to histidines (Supplementary Fig. ) are 2.3–2.5 Å (H587 in B/D), 2.2–2.4 Å (H582 in A/C) and 2.2 Å (H587 in A/C), which are typical for histidine coordination of Zn 2+ ions – , although other metal ions, such as Mg 2+ , Ca 2+ or Na + , can be coordinated as well – . These metal ions (M) are either the main components of our purification buffers (Na + ) or likely impurities of either the buffer components or genistein, which was extracted from Styphnolobium japonicum . To test possible contribution of the putative metal ions to TRPV6 inhibition by genistein, we introduced alanine mutations of H582, H587, or both and used Fura-2 fluorescent measurements to estimate changes in calcium uptake through the corresponding mutant channels at different genistein concentrations (Fig. ). Genistein inhibition of calcium uptake through the H582A mutant was weaker than through wild-type channels as indicated by the rightward shift of the genistein concentration-dependence. A stronger shift in the concentration-dependence was observed for H587A and H582A/H587A mutants, with a little difference between the two, consistent with the predominant contribution (two out of three) of H587 to the putative metal ion coordination (Fig. and Supplementary Fig. ). Therefore, our mutagenesis combined with functional recordings are consistent with the putative metal ion coordination by the three-histidine clusters being an additional factor that might help to stabilize the asymmetric conformation of hTRPV6 GEN . To further validate genistein binding to sites 1 and 2, we carried out all-atom MD simulations (Fig. , Supplementary Figs. - ). Four simulations were performed for site 1 with different initial orientations of the ligand, which differed by 180° rotations around the long genistein axis and around the central pore axis. Our simulations showed that the ligand is stabilized in site 1 by two hydrogen bonds between the hydroxyl groups of genistein and the carbonyl oxygens of M570 in subunits A/C (Supplementary Fig. ). These interactions orient the ligand in either one of two symmetrical positions rotated by 180° around the central pore axis, thus demonstrating an excellent fit to the non-protein cryo-EM density in site 1 (Fig. ). Similar simulations were repeated with CHS embedded in site 1. CHS showed no specific interactions with the protein, was much more mobile than genistein, and intended to escape from site 1. The MD-predicted density for CHS formed a “band” across the pore, inconsistent with the cryo-EM density (Supplementary Fig. ). MD simulations of a single genistein molecule placed at site 2 starting with the position modeled in the cryo-EM structure (Fig. ) revealed highly dynamic behavior of this molecule, only somewhat restrained by π-stacking interactions with W583 (not shown). To stabilize the genistein molecule at this cryo-EM-like primary position, we added another genistein molecule at the secondary position between the W583 side chains below and perpendicular to the primary position (Supplementary Fig. ). We performed four MD simulations with different initial orientations of genistein molecules in the primary and secondary positions at site 2, generated in the same way as in simulations for site 1. The simulations showed that genistein in the primary position of site 2 is more mobile than genistein in site 1. However, in three out of four MD runs, genistein was stabilized in the primary position of site 2 by π-stacking interactions with W583 in subunits B/D or H582 in subunits A/C as well as by one or two hydrogen bonds with the carbonyl oxygens of the neighboring residues M578 and G579 in subunits A/C or residue I575 in subunits B/D (Supplementary Fig. ). Such a diversity of interaction partners did not allow to reveal specific positions of the ligand in the site. However, the resulting distribution of the ligand density at the primary position in site 2 averaged over the MD runs demonstrated excellent agreement with the central non-protein cryo-EM density (Fig. ). Genistein at the secondary position in site 2 was much more mobile than the genistein molecule at site 1 or in the primary position at site 2 (Supplementary Fig. ). Nevertheless, the genistein molecule at the secondary position in site 2 tended to form two to four π-stacking interactions with W583 in subunits B/D and H582 in subunits A/C as well as hydrogen bonds with the side chains of D580 in subunits B/D. In turn, these interactions imposed restraints on the genistein molecule at the primary position in site 2. The cumulative behavior of two genistein molecules at site 2 agrees well with the cryo-EM data, where a group of small non-protein densities between W583 side chains are likely to represent the ensemble of positions of the weakly interacting ligands (Supplementary Fig. ). MD simulations with CHS embedded in the primary position of site 2 in four different initial orientations revealed that the ligand is stabilized in this site much better than in site 1 (Supplementary Fig. ). Pockets between the S6 helices of neighboring subunits provide enough space for the CHS molecule, while its acidic group forms polar interactions with K484 or R589. However, being tightly packed in site 2, the CHS molecule created a much more extended density in MD simulations than the density revealed by cryo-EM (Supplementary Fig. ). Accordingly, while there is a possibility that the experimentally observed non-protein density in site 2 represents CHS or a similar size small molecule, the probability of this is low. The superposition of hTRPV6 Open and hTRPV6 GEN structures suggests a possible mechanism of TRPV6 inhibition by genistein (Fig. and Supplementary Movie). When binding to the open TRPV6 channel perpendicularly to the central pore axis, genistein molecules pull one diagonal pair of subunits (A/C) asymmetrically towards the channel center. This motion breaks the interactions between D489 in S5 and T581 in S6 as well as Q473 in the S4–S5 linker and R589 in the TRP helix, which stabilize the energetically unfavorable α-to-π transition in S6 and the open-pore conformation, respectively . The reverse π-to-α transition in S6 is accompanied by a ~100° rotation of the intracellular portion of S6, while the stabilizing interactions with genistein molecules allow S6 in subunits A and C to stay as long as in the open state and not become shorter as in the previously published closed-state structures , , . This in turn causes shortening of the TRP helices in subunits A and C due to unwinding of their N-terminal portions (Fig. ). Upon genistein binding, the pore-forming domains in subunits B and D also change their conformations to adapt to changes in subunits A and C. This adaptation disrupts the D489-T581 and Q473-R589 interactions, reverses the unfavorable α-to-π transition in S6 but also causes the split of S6 into two helical segments, S6 and S6-TRP (Fig. ). In turn, the S4–S5 regions which stay in direct contact with the conformationally altered regions of S6-TRP helix also adapt their conformation. In subunits A and C, this involves altering the angle between S4–S5 and S5 and increasing the length of the unfolded connection between S4 and S4–S5. In contrast, the contiguous helical region that follows S4 in subunits B and D splits into two distinct helical segments, S4–S5 and S5 (Fig. d and ). Amazingly, these genistein-induced conformational changes are localized to the channel intracellular core, do not propagate beyond the S4–S5 and S6-TRP helix regions, and the rest of the TRPV6 molecule remains essentially the same (Fig. ). Currently, it is difficult to assess the relative contribution of sites 1 and 2 to the mechanism of TRPV6 inhibition by genistein. Given the more stable behavior of genistein at site 1 compared to site 2, where a more mobile genistein molecule at the primary central position might be accompanied by an even more mobile genistein molecule at the secondary position (Fig. , Supplementary Fig. ), as well as a relatively weak contribution of metal coordination in the vicinity of site 2 to inhibition of calcium influx (Fig. ), we hypothesize that site 1 represents the main site of genistein action. So far, 14 unique ligand binding sites have been identified in TRPV channels . Genistein binds to the intracellular pore entry site, which makes it a TRPV6 ion channel blocker. Given the small size of the TRPV6 selectivity filter compared to the genistein molecule, genistein likely approaches its sites in the intracellular pore by first crossing through the membrane. In fact, the relatively low affinity of TRPV6 to genistein (Fig. c, ) may be due to a slow crossing of the drug through the membrane to reach the intracellular side. Given the location of site 1 above the channel gate formed by the S6 bundle crossing and at the bottom of the central cavity, genistein is likely to reach this site only after the channel undergoes opening. If this is true, genistein can be considered an open-channel blocker. It does not, however, interact passively with the ion channel as the majority of open-channel blockers do. Instead, because of the non-equivalent positioning of genistein molecules in the pore relative to two pairs of diagonal subunits (Fig. ), they cause asymmetric transformation of the pore (Fig. , Supplementary Movie) and its closure (Fig. ). What makes genistein a unique ion channel blocker of TRPV6? Among 14 types of TRPV ligands that were characterized structurally , there are four types of ion channel blockers. TRPV6 can be blocked by trivalent cations, like Gd 3+ , which bind at the extracellular pore entry site formed by side chains of four D542 residues, each from four individual TRPV6 subunits. At this location, Gd 3+ binding occludes the pore for conductance but does not change the conformation of the selectivity filter, which is stabilized by hydrophobic interactions of the pore helix residues, including phenylalanines F531, F534, and F537, neither does it alter the conformation of the rest of TRPV6 , . RR represents the second type of ion channel blockers, which bind to the selectivity filter site located intracellularly with respect to the extracellular pore entry site . Binding of RR at this location of the TRPV6 pore does not introduce significant conformational changes in the selectivity filter either but causes a conversion of the gate region into the closed-state, presumably due to electrostatic interactions of the positive charge of RR and the electric dipole of the S6 helix . The other two types of ion channel blockers bind at the pore intracellular entry site, the region where genistein binds as well. The physiological blocker calmodulin (CaM) interacts with this site through a unique cation-π interaction by inserting the side chain of lysine K115 into a tetra-tryptophan (W583) cage at the pore’s intracellular entrance , . Similarly, (4-phenylcyclohexyl)piperazine derivatives (PCHPDs), selective nanomolar-affinity synthetic inhibitors of TRPV6 including cis-22a, plug the channel pore at the pore intracellular entry site, mimicking the action of CaM . These two types of TRPV6 ion channel blockers induce conformational rearrangements that transform the open pore into a non-conducting inactivated state, different from the closed and open conformations of TRPV6. In the inactivated state, the TRPV6 pore becomes hydrophobically sealed by the I575 side chains, while the π-bulge remains present in the middle of S6, similar to the open state , . At the same time, one of the open state-stabilizing interactions breaks (Q473-R589), but the other one remains intact (D489-T581). Similar to PCHPDs and CaM, genistein binds at the intracellular pore entry but causes drastically different structural rearrangements, accompanied by alterations of the local symmetry and secondary structure of the S4–S5 and S6-TRP regions (Fig. and Supplementary Movie). The two likely reasons for such dramatic rearrangements are (1) flexibility of the pore intracellular region, which has to easily permit gate opening and closure to maintain TRPV6 constitutive activity, and (2) the orientation of genistein molecules across the pore instead of along the pore, as in the case of PCHPDs and CaM. The latter factor allows genistein to cause more damaging mechanistic changes in the structure of TRPV6 pore compared to any other ion channel blocker studied before. Due to the unprecedented character of genistein-induced conformational transformation of the channel, the mechanism of genistein block is also unique amongst all previously described mechanisms of TRP channel inhibition by various natural and synthetic antagonists , including allosteric inhibition of hTRPV6 by 2-APB or econazole. With the potential to utilize various well-established chemical synthesis pathways for genistein and its derivatives , , , the mechanism of ion channel block described here opens new avenues for the development of drugs targeting TRPV6 in pathological conditions. Construct Full-length wild-type human TRPV6 used for cryo-EM was cloned into a pEG BacMam vector with a C-terminal thrombin cleavage site followed by a streptavidin affinity tag (WSHPQFEK). For Fura-2 AM measurements, point mutations in wild-type human TRPV6 were introduced using the standard molecular biology techniques , . Expression and purification hTRPV6 was expressed and purified based on our previously established protocols , – . Bacmids and baculoviruses were produced using standard procedures , , , . Baculovirus was made in Sf9 cells for ~72 h (Thermo Fisher Scientific, mycoplasma test negative, GIBCO #12659017) and was added to suspension-adapted HEK 293S cells lacking N-acetyl-glucosaminyltransferase I (GnTI – , mycoplasma test negative, ATCC #CRL-3022) that were maintained at 37 °C and 5% CO 2 in Freestyle 293 media (Gibco-Life Technologies #12338-018) supplemented with 2% FBS. Twenty-four hours after transduction, 10 mM sodium butyrate was added to enhance protein expression, and the temperature was reduced to 30 °C. Seventy-two hours after transduction, cells were harvested by centrifugation at 5471 × g for 15 min using a Sorvall Evolution RC centrifuge (Thermo Fisher Scientific), washed in phosphate-buffered saline pH 8.0, and pelleted by centrifugation at 3202 × g for 10 min using an Eppendorf 5810 centrifuge. The cell pellet was solubilized under constant stirring for 2 h at 4 °C in ice-cold lysis buffer containing 1% (w/v) n-dodecyl β-D-maltoside, 0.1% (w/v) CHS, 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.8 μM aprotinin, 4.3 μM leupeptin, 2 μM pepstatin A, 1 mM phenylmethylsulfonyl fluoride, and 1 mM β-mercaptoethanol (βME). The non-solubilized material was pelleted in the Eppendorf 5810 centrifuge at 3202 × g and 4 °C for 10 min. The supernatant was subjected to ultracentrifugation in a Beckman Coulter ultracentrifuge using a Beckman Coulter Type 45Ti rotor at 186,000 × g and 4 °C for 1 h to further clean up the solubilized protein. The supernatant was added to 5 ml of strep resin and rotated for 1 h at 4 °C. The resin was washed with 10 column volumes of wash buffer containing 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 1 mM βME, 0.01% (w/v) GDN, and 0.001% (w/v) CHS, and the protein was eluted with the same buffer supplemented with 2.5 mM d -Desthiobiotin. The eluted protein was concentrated using a 100 kDa NMWL centrifugal filter (MilliporeSigma Amicon) to 0.5 ml and then centrifuged in a Sorvall MTX 150 Micro-Ultracentrifuge (Thermo Fisher Scientific) using an S100AT4 rotor for 30 min at 66,000 × g and 4 °C before being injected into a size-exclusion chromatography (SEC) column. hTRPV6 was further purified using a Superose™ 6 10/300 GL SEC column attached to an AKTA FPLC (GE Healthcare) and equilibrated in 150 mM NaCl, 20 mM Tris-HCl pH 8.0, 1 mM βME, 0.01% GDN, and 0.001% CHS. The tetrameric peak fractions were pooled and concentrated using 100 kDa NMWL centrifugal filter to 3.36 mg/ml. Genistein (2 mM) was added to hTRPV6 and the resulting sample was incubated at room temperature for 120 min before grid freezing. Cryo-EM sample preparation and data collection UltrAuFoil R 1.2/1.3, Au 300 grids were used for plunge-freezing. Prior to sample application, grids were plasma treated in a PELCO easiGlow glow discharge cleaning system (0.39 mBar, 15 mA, “glow” 25 s, “hold” 10 s). A Mark IV Vitrobot (Thermo Fisher Scientific) set to 100% humidity at 4 °C was used to plunge-freeze the grids in liquid ethane after applying 3 µl of protein sample to their gold-coated side using a blot time of 5 s, a blot force of 5, and a wait time of 15 s. The grids were stored in liquid nitrogen before imaging. Images of frozen-hydrated particles of hTRPV6 in the presence of genistein were collected using Leginon – software on a Titan Krios transmission electron microscope (Thermo Fisher Scientific) operating at 300 kV and equipped with a post-column GIF Quantum energy filter and a Gatan K3 Summit direct electron detection camera (Gatan, Pleasanton, CA, USA). 3630 micrographs were collected in counting mode with a raw image pixel size of 0.83 Å across the defocus range of −0.8 to −2.0 µm. The total dose of ~60 e − Å −2 was attained by using the dose rate of ~16 e − pixel −1 s −1 across 50 frames during the 2.5-s exposure time. Image processing and 3D reconstruction Data were processed in RELION and cryoSPARC . Movie frames were aligned using the RELION’s implementation of a MotionCor2 -like algorithm. Contrast transfer function (CTF) estimation was performed on non-dose-weighted micrographs using the patch CTF estimation in cryoSPARC. Subsequent data processing was done on dose-weighted micrographs. Following CTF estimation, micrographs were manually inspected and those with outliers in defocus values, ice thickness, and astigmatism as well as micrographs with lower predicted CTF-correlated resolution (higher than 5 Å) were excluded from further processing (individually assessed for each parameter relative to the overall distribution). After several rounds of selection through 2D classification, particles were further 3D classified (heterogeneous refinement) into four classes. Particles representing the best class were re-extracted without binning (256-pixel box size) and further 3D classified. The final sets of particles for hTRPV6 GEN and hTRPV6 Open representing the best classes were subjected to homogeneous refinement. The reported resolutions of 2.66 Å and 2.71 Å for hTRPV6 GEN and hTRPV6 Open , respectively, were estimated using the gold standard Fourier shell correlation (GSFSC) (Supplementary Figs. – ). The local resolution was calculated with the resolution range estimated using the FSC = 0.143 criterion. Cryo-EM density visualization was done in UCSF Chimera and UCSF ChimeraX . Model building Models of hTRPV6 GEN and hTRPV6 Open were built in Coot , using the previously published cryo-EM structure of TRPV6 in the open state (PDB ID: 7S89) as a guide. The models were tested for overfitting by shifting their coordinates by 0.5 Å (using Shake) in Phenix , refining the shaken models against the corresponding unfiltered half maps, and generating densities from the resulting models in UCSF Chimera. Structures were visualized and figures were prepared in UCSF Chimera, UCSF ChimeraX, and Pymol . The pore radius was calculated using HOLE . Fura-2 AM measurements Full-length wild-type or mutant human TRPV6 was expressed in HEK 293S cells as described above. 48 h after transduction, the cells were harvested by centrifugation at 550 × g for 5 min, resuspended in prewarmed HEPES-buffered saline (HBS: 118 mM NaCl, 4.8 mM KCl, 1 mM MgCl 2 , 5 mM D-glucose, 10 mM HEPES pH 7.4) containing 5 µg/ml of Fura-2 AM (Life Technologies) and incubated at 37 °C for 45 min. The loaded cells were then centrifuged for 5 min at 550 × g , resuspended again in prewarmed HBS, and incubated at 37 °C for 30 min in the dark. The cells were subsequently pelleted and washed twice, then resuspended in HBS. The cells were kept on ice in the dark for a maximum of ~2 h before fluorescence measurements, which were conducted using a spectrofluorometer QuantaMaster 40 (Photon Technology International) at room temperature in a quartz cuvette under constant stirring. Intracellular Ca 2+ was measured by taking the ratio of fluorescence measurements at two excitation wavelengths (340 and 380 nm) and one emission wavelength (510 nm). The excitation wavelength was switched at 200-ms intervals. Electrophysiology DNA encoding wild-type human TRPV6 was introduced into a plasmid for expression in eukaryotic cells that was engineered to produce GFP via a downstream internal ribosome entry site . HEK 293S cells (ATCC #CRL-1573) grown on glass coverslips in 35-mm dishes were transiently transfected with 1–5 μg of plasmid DNA using Lipofectamine 2000 Reagent (Life Technologies). Recordings were made 24 h after transfection at room temperature. Currents from whole cells, typically held at a 0-mV potential, were recorded using an Axopatch 200B amplifier (MolecularDevices, LLC), filtered at 5 kHz, and digitized at 10 kHz using low-noise data acquisition system Digidata 1440 A and pCLAMP software (Molecular Devices, LLC). The external solution contained (in mM): 142 LiCl, 10 HEPES, and 10 glucose, pH 7.4. To evoke monovalent currents, 0.1–0.5 mM EGTA was added to the external solution. The internal solution contained (in mM): 100 CsAsp, 20 CsF, 10 EGTA, 3 MgCl 2 , 4 NaATP, and 20 HEPES pH 7.2, an additional 1 mM ATP was added immediately before the experiment. We used the LiCl-based extracellular solution, because, in this solution, removal of extracellular Mg 2+ and Ca 2+ does not induce endogenous currents in non-transfected cells, unlike in Na + or K + based solutions . TRPV6 currents were recorded in response to 400-ms voltage ramps from −100 mV to +70 mV applied every 5–10 s. Genistein was added directly to the aqueous buffer solutions for measurements in HEK cells. At concentrations higher than 50 μM, genistein displayed aggregation behavior and tended to clog our application system. For this reason, all experiments were carried out at genistein concentrations lower than 50 μM. Data analysis was performed using the computer program Origin 9.1.0 (OriginLab Corp.). MD simulations The model of TRPV6 GEN (residues 27–638) was inserted into a hydrated lipid bilayer with the molecular composition of 50% palmitoyloleoylphosphatidylcholine (POPC), 25% palmitoyloleoylphosphatidylethanolamine (POPE), and 25% cholesterol (about 900 lipids in total). Ca 2+ ion was embedded into the selectivity filter region (between the side chains of residues D542) and two Zn 2+ ions into the metal binding sites like in the structural model. Na + and Cl − ions were added to ensure zero net charge of the system at 0.15 M ionic concentration. Twelve replicas of the system were constructed: four replicas with a differently oriented genistein in site 1, four replicas with two differently orientated genistein molecules at the primary and secondary positions in site 2, and four replicas with two differently orientated CHS molecules in sites 1 and 2 (see the text and Supplementary Table for details). To stabilize the protein structure in the vicinity of the binding sites, intersubunit constraints were applied to the distances between C α atoms of the lower parts of the S6 helices. The distances were constrained between each residue in the region 563–578 (536–586 for the systems with genistein molecules in the primary and secondary positions in site 2) of all four subunits with the force constant of 10 kJ/(mol × Å 2 ). During MD simulations, such a network of constraints provides the necessary flexibility for protein adaptation to the ligand but retains the binding site structure close to the cryo-EM model. First, the simulated systems were equilibrated in several stages: 5 × 10 3 steps of steepest descent minimization followed by heating from 5 to 310 K during a 100 ps MD run, then 10 ns of MD run with fixed positions of the protein and genistein/CHS heavy atoms, 10 ns of MD with fixed positions of the protein backbone and genistein/CHS heavy atoms, 20 ns of MD with fixed positions of the protein Cα atoms to permit membrane and genistein/CHS relaxation after insertion. Finally, 200 ns MD-production runs were carried out. MD simulations were carried out using GROMACS 2021.4 package , CHARMM36 force field – , and the TIP3P water model . Simulations were performed with an integration time of 2 fs, constrained hydrogen-containing bond lengths by the LINCS algorithm , imposed 3D periodic boundary conditions, at constant temperature (310 K) and pressure (1 bar). A cutoff distance of 12 Å was used to evaluate nonbonded interactions and the particle-mesh Ewald method was employed to treat long-range electrostatics. The CHARMM36 topology for genistein was taken from Burendahl et al. . Reporting summary Further information on research design is available in the linked to this article. Full-length wild-type human TRPV6 used for cryo-EM was cloned into a pEG BacMam vector with a C-terminal thrombin cleavage site followed by a streptavidin affinity tag (WSHPQFEK). For Fura-2 AM measurements, point mutations in wild-type human TRPV6 were introduced using the standard molecular biology techniques , . hTRPV6 was expressed and purified based on our previously established protocols , – . Bacmids and baculoviruses were produced using standard procedures , , , . Baculovirus was made in Sf9 cells for ~72 h (Thermo Fisher Scientific, mycoplasma test negative, GIBCO #12659017) and was added to suspension-adapted HEK 293S cells lacking N-acetyl-glucosaminyltransferase I (GnTI – , mycoplasma test negative, ATCC #CRL-3022) that were maintained at 37 °C and 5% CO 2 in Freestyle 293 media (Gibco-Life Technologies #12338-018) supplemented with 2% FBS. Twenty-four hours after transduction, 10 mM sodium butyrate was added to enhance protein expression, and the temperature was reduced to 30 °C. Seventy-two hours after transduction, cells were harvested by centrifugation at 5471 × g for 15 min using a Sorvall Evolution RC centrifuge (Thermo Fisher Scientific), washed in phosphate-buffered saline pH 8.0, and pelleted by centrifugation at 3202 × g for 10 min using an Eppendorf 5810 centrifuge. The cell pellet was solubilized under constant stirring for 2 h at 4 °C in ice-cold lysis buffer containing 1% (w/v) n-dodecyl β-D-maltoside, 0.1% (w/v) CHS, 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.8 μM aprotinin, 4.3 μM leupeptin, 2 μM pepstatin A, 1 mM phenylmethylsulfonyl fluoride, and 1 mM β-mercaptoethanol (βME). The non-solubilized material was pelleted in the Eppendorf 5810 centrifuge at 3202 × g and 4 °C for 10 min. The supernatant was subjected to ultracentrifugation in a Beckman Coulter ultracentrifuge using a Beckman Coulter Type 45Ti rotor at 186,000 × g and 4 °C for 1 h to further clean up the solubilized protein. The supernatant was added to 5 ml of strep resin and rotated for 1 h at 4 °C. The resin was washed with 10 column volumes of wash buffer containing 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 1 mM βME, 0.01% (w/v) GDN, and 0.001% (w/v) CHS, and the protein was eluted with the same buffer supplemented with 2.5 mM d -Desthiobiotin. The eluted protein was concentrated using a 100 kDa NMWL centrifugal filter (MilliporeSigma Amicon) to 0.5 ml and then centrifuged in a Sorvall MTX 150 Micro-Ultracentrifuge (Thermo Fisher Scientific) using an S100AT4 rotor for 30 min at 66,000 × g and 4 °C before being injected into a size-exclusion chromatography (SEC) column. hTRPV6 was further purified using a Superose™ 6 10/300 GL SEC column attached to an AKTA FPLC (GE Healthcare) and equilibrated in 150 mM NaCl, 20 mM Tris-HCl pH 8.0, 1 mM βME, 0.01% GDN, and 0.001% CHS. The tetrameric peak fractions were pooled and concentrated using 100 kDa NMWL centrifugal filter to 3.36 mg/ml. Genistein (2 mM) was added to hTRPV6 and the resulting sample was incubated at room temperature for 120 min before grid freezing. UltrAuFoil R 1.2/1.3, Au 300 grids were used for plunge-freezing. Prior to sample application, grids were plasma treated in a PELCO easiGlow glow discharge cleaning system (0.39 mBar, 15 mA, “glow” 25 s, “hold” 10 s). A Mark IV Vitrobot (Thermo Fisher Scientific) set to 100% humidity at 4 °C was used to plunge-freeze the grids in liquid ethane after applying 3 µl of protein sample to their gold-coated side using a blot time of 5 s, a blot force of 5, and a wait time of 15 s. The grids were stored in liquid nitrogen before imaging. Images of frozen-hydrated particles of hTRPV6 in the presence of genistein were collected using Leginon – software on a Titan Krios transmission electron microscope (Thermo Fisher Scientific) operating at 300 kV and equipped with a post-column GIF Quantum energy filter and a Gatan K3 Summit direct electron detection camera (Gatan, Pleasanton, CA, USA). 3630 micrographs were collected in counting mode with a raw image pixel size of 0.83 Å across the defocus range of −0.8 to −2.0 µm. The total dose of ~60 e − Å −2 was attained by using the dose rate of ~16 e − pixel −1 s −1 across 50 frames during the 2.5-s exposure time. Data were processed in RELION and cryoSPARC . Movie frames were aligned using the RELION’s implementation of a MotionCor2 -like algorithm. Contrast transfer function (CTF) estimation was performed on non-dose-weighted micrographs using the patch CTF estimation in cryoSPARC. Subsequent data processing was done on dose-weighted micrographs. Following CTF estimation, micrographs were manually inspected and those with outliers in defocus values, ice thickness, and astigmatism as well as micrographs with lower predicted CTF-correlated resolution (higher than 5 Å) were excluded from further processing (individually assessed for each parameter relative to the overall distribution). After several rounds of selection through 2D classification, particles were further 3D classified (heterogeneous refinement) into four classes. Particles representing the best class were re-extracted without binning (256-pixel box size) and further 3D classified. The final sets of particles for hTRPV6 GEN and hTRPV6 Open representing the best classes were subjected to homogeneous refinement. The reported resolutions of 2.66 Å and 2.71 Å for hTRPV6 GEN and hTRPV6 Open , respectively, were estimated using the gold standard Fourier shell correlation (GSFSC) (Supplementary Figs. – ). The local resolution was calculated with the resolution range estimated using the FSC = 0.143 criterion. Cryo-EM density visualization was done in UCSF Chimera and UCSF ChimeraX . Models of hTRPV6 GEN and hTRPV6 Open were built in Coot , using the previously published cryo-EM structure of TRPV6 in the open state (PDB ID: 7S89) as a guide. The models were tested for overfitting by shifting their coordinates by 0.5 Å (using Shake) in Phenix , refining the shaken models against the corresponding unfiltered half maps, and generating densities from the resulting models in UCSF Chimera. Structures were visualized and figures were prepared in UCSF Chimera, UCSF ChimeraX, and Pymol . The pore radius was calculated using HOLE . Full-length wild-type or mutant human TRPV6 was expressed in HEK 293S cells as described above. 48 h after transduction, the cells were harvested by centrifugation at 550 × g for 5 min, resuspended in prewarmed HEPES-buffered saline (HBS: 118 mM NaCl, 4.8 mM KCl, 1 mM MgCl 2 , 5 mM D-glucose, 10 mM HEPES pH 7.4) containing 5 µg/ml of Fura-2 AM (Life Technologies) and incubated at 37 °C for 45 min. The loaded cells were then centrifuged for 5 min at 550 × g , resuspended again in prewarmed HBS, and incubated at 37 °C for 30 min in the dark. The cells were subsequently pelleted and washed twice, then resuspended in HBS. The cells were kept on ice in the dark for a maximum of ~2 h before fluorescence measurements, which were conducted using a spectrofluorometer QuantaMaster 40 (Photon Technology International) at room temperature in a quartz cuvette under constant stirring. Intracellular Ca 2+ was measured by taking the ratio of fluorescence measurements at two excitation wavelengths (340 and 380 nm) and one emission wavelength (510 nm). The excitation wavelength was switched at 200-ms intervals. DNA encoding wild-type human TRPV6 was introduced into a plasmid for expression in eukaryotic cells that was engineered to produce GFP via a downstream internal ribosome entry site . HEK 293S cells (ATCC #CRL-1573) grown on glass coverslips in 35-mm dishes were transiently transfected with 1–5 μg of plasmid DNA using Lipofectamine 2000 Reagent (Life Technologies). Recordings were made 24 h after transfection at room temperature. Currents from whole cells, typically held at a 0-mV potential, were recorded using an Axopatch 200B amplifier (MolecularDevices, LLC), filtered at 5 kHz, and digitized at 10 kHz using low-noise data acquisition system Digidata 1440 A and pCLAMP software (Molecular Devices, LLC). The external solution contained (in mM): 142 LiCl, 10 HEPES, and 10 glucose, pH 7.4. To evoke monovalent currents, 0.1–0.5 mM EGTA was added to the external solution. The internal solution contained (in mM): 100 CsAsp, 20 CsF, 10 EGTA, 3 MgCl 2 , 4 NaATP, and 20 HEPES pH 7.2, an additional 1 mM ATP was added immediately before the experiment. We used the LiCl-based extracellular solution, because, in this solution, removal of extracellular Mg 2+ and Ca 2+ does not induce endogenous currents in non-transfected cells, unlike in Na + or K + based solutions . TRPV6 currents were recorded in response to 400-ms voltage ramps from −100 mV to +70 mV applied every 5–10 s. Genistein was added directly to the aqueous buffer solutions for measurements in HEK cells. At concentrations higher than 50 μM, genistein displayed aggregation behavior and tended to clog our application system. For this reason, all experiments were carried out at genistein concentrations lower than 50 μM. Data analysis was performed using the computer program Origin 9.1.0 (OriginLab Corp.). The model of TRPV6 GEN (residues 27–638) was inserted into a hydrated lipid bilayer with the molecular composition of 50% palmitoyloleoylphosphatidylcholine (POPC), 25% palmitoyloleoylphosphatidylethanolamine (POPE), and 25% cholesterol (about 900 lipids in total). Ca 2+ ion was embedded into the selectivity filter region (between the side chains of residues D542) and two Zn 2+ ions into the metal binding sites like in the structural model. Na + and Cl − ions were added to ensure zero net charge of the system at 0.15 M ionic concentration. Twelve replicas of the system were constructed: four replicas with a differently oriented genistein in site 1, four replicas with two differently orientated genistein molecules at the primary and secondary positions in site 2, and four replicas with two differently orientated CHS molecules in sites 1 and 2 (see the text and Supplementary Table for details). To stabilize the protein structure in the vicinity of the binding sites, intersubunit constraints were applied to the distances between C α atoms of the lower parts of the S6 helices. The distances were constrained between each residue in the region 563–578 (536–586 for the systems with genistein molecules in the primary and secondary positions in site 2) of all four subunits with the force constant of 10 kJ/(mol × Å 2 ). During MD simulations, such a network of constraints provides the necessary flexibility for protein adaptation to the ligand but retains the binding site structure close to the cryo-EM model. First, the simulated systems were equilibrated in several stages: 5 × 10 3 steps of steepest descent minimization followed by heating from 5 to 310 K during a 100 ps MD run, then 10 ns of MD run with fixed positions of the protein and genistein/CHS heavy atoms, 10 ns of MD with fixed positions of the protein backbone and genistein/CHS heavy atoms, 20 ns of MD with fixed positions of the protein Cα atoms to permit membrane and genistein/CHS relaxation after insertion. Finally, 200 ns MD-production runs were carried out. MD simulations were carried out using GROMACS 2021.4 package , CHARMM36 force field – , and the TIP3P water model . Simulations were performed with an integration time of 2 fs, constrained hydrogen-containing bond lengths by the LINCS algorithm , imposed 3D periodic boundary conditions, at constant temperature (310 K) and pressure (1 bar). A cutoff distance of 12 Å was used to evaluate nonbonded interactions and the particle-mesh Ewald method was employed to treat long-range electrostatics. The CHARMM36 topology for genistein was taken from Burendahl et al. . Further information on research design is available in the linked to this article. Supplementary Information Peer Review File Description of additional supplementary files Supplementary Data 1 Supplementary Movie 1 Reporting Summary
A comprehensive regulatory and industry review of modeling and simulation practices in oncology clinical drug development	c6d5b182-6360-4f12-8a5e-89769623ab5b	10169901	Internal Medicine[mh]	Exposure–response (E–R) analyses of oncology drugs are an integral component in their clinical development , both for internal decision making and externally to support regulatory approval. Understanding the relationship between dose, exposure, and response allows the sponsor to demonstrate that they understand the pharmacokinetic-pharmacodynamic (PKPD) behavior of their drug. Variability in pharmacokinetics and pharmacodynamics is well known and there are certain patient subgroups that may be at increased risk for adverse events, e.g., patients with impaired renal and/or hepatic function, or decreased efficacy, e.g., ultrarapid CYP2D6 metabolizers . E–R analysis allows one to predict and/or confirm response in these patient subgroups and whether dose modification is needed in these subgroups. The term ‘exposure’ is a broad one that tries to capture how much drug a person is “exposed” to. Exposure may encompass any or all of the following: Some aspect related to dose, such as the daily dose or total cumulative dose a patient receives; Some measure of drug concentration in the body at some point in time, such as maximal concentration (C max ), average concentration (C avg ) or trough concentration at steady-state (C trough ); Time above some threshold, such as time above some minimum effective concentration, or May include integrated measures of concentration, for instance area under the curve at steady-state (AUC ss ), cumulative area under the curve, or dose at the time of some response. Indeed, there are many different possible choices for quantifying exposure over time and space (e.g., systemic or tumor). Similarly, ‘response’ is also a broad term that may encompass any or all of the following: Assessment of drug efficacy: response rate, time to progression, progression free survival (PFS) or overall survival (OS); Assessment of drug safety: whether a patient experiences nausea or vomiting, or the severity of a rash a patient may develop during treatment; or Measurement of pharmacodynamic biomarkers: degree of phosphorylation of some important protein, or level of receptor occupancy, etc. Sponsors may submit several E–R analyses in support of a regulatory submission that encompass different measures of exposure and different measures of clinical outcomes. Identifying dose/dosing regimens that provide a favorable benefit/risk profile is critical in drug development . Sponsors may also conduct E–R analyses for internal decision making for go/no go decisions or to inform dose selection for Phase 2/3 trials. In modern oncology, the old paradigm that “more is better” with the maximum tolerated dose (MTD) being used as the dose studied in late phase studies is no longer the norm. The MTD is commonly defined as the highest dose that most patients can tolerate without unacceptable side effects. However, the introduction of targeted therapies, biologics in general, and immuno-oncology is moving drug development away from the MTD concept in search of the optimal dose/dosing regimen. Recently, the Food and Drug Administration has issued a draft guidance that will require to sponsors to study a range of doses in clinical development with the goal to use the optimal dose in registration trials . As such, alternative approaches to clinical development will be required for dose optimization, facilitating the use of E–R analysis by including safety and efficacy information from more than one dose level. One of the earliest, most comprehensive, E–R analyses was presented by Houk et al. in patients with advanced solid tumors, including patients with gastrointestinal stromal tumor (GIST) and metastatic renal cell carcinoma (mRCC) treated with sunitinib. Their analyses used different measures of exposure: dose, drug systemic concentration, AUC ss , and cumulative AUC during 1 cycle of treatment. Using a combination of population PKPD modeling, repeated-measures logistic regression, and correlation analyses, authors demonstrated that increased sunitinib exposure was associated with longer time to progression, longer overall survival, and a greater chance of clinical response. They also showed that increased exposure was associated with increased blood pressure, increased incidence of fatigue, and a greater probability of neutropenia. These analyses were supportive of the recommended dosing regimen in GIST and mRCC patients that was approved at 50 mg once daily for 4 weeks every 6 weeks (4 weeks on/2 weeks off). Recently, a less intense dose regimen has been proposed for sunitinib: 50 mg for 14 days every 3 weeks (traditional “2/1 schedule”). Both regimens have the same dose intensity (4 weeks on) and both have a 2-week drug holiday period every 6 weeks. The model from Houk et al. mathematically explains why the alternate schedule (2/1) presents less toxicity, and therefore, it is better tolerated . Typically, due to lack of established standard methods, there is no one standard approach to conducting an exposure–response analysis. For example, when performing time-to-event analysis, Kaplan–Meier (KM) curves are a useful graphic assessment for the exploration of observed data. Further analysis such as Cox proportional hazards regression model, parametric time-to-event models, or an accelerated failure time model could also be considered based on the nature of the data. All of these are equally valid and have different assumptions with different pros and cons, and different prediction outputs. Furthermore, it is important to mention that E–R analyses in oncology using survival metrics, may be subject to selection bias and immortal time bias . Landmark analyses and multi-state analyses have been proposed to correct the inherent selection bias resulting in part from the fact that responders must live long enough for response to be observed . Therefore, the question arises as to which method should be used and if there is a preferred method of choice. In an attempt to address these concerns, this white paper is the output of an industry- government collaboration between scientists with broad experience in E–R modeling in oncology. The goal of this white paper is to provide guidance on the preferred methods for E–R analysis commonly used in oncology and the measures of exposure to consider in different scenarios. To make things easier for the reader, an executive summary of recommendations and comments from each of the following sections is presented in Table . When performing E–R analyses, one of the first decision points in the analysis is the exposure metric of choice. One needs to consider what type of data are available, the duration of treatment, and dose compliance. It is not uncommon in oncology for patients to experience dose modifications during the course of treatment . Often, these dose adjustments happen because of adverse events (AEs) that require either a dose holiday or dose reduction. The result of the dose adjustments could lead to lower average drug exposure for subjects with long treatment duration. Further, a high percentage of dropouts is expected due to AEs associated with concurrent exposure. Thus, under this situation, an E–R analysis for some of the most common efficacy endpoints (overall survival, OS, or progression-free survival, PFS) may suggest an inverse relationship between exposure and efficacy. To eliminate the bias introduced by the large percentage of dose reductions, an earlier exposure metric prior to any dose modification (Cycle 1) could be considered. However, this early exposure metric will have limited value in establishing any relationship between exposure and response. Under scenarios where AEs are leading to significant dose adjustments, it is important to consider if the right dose has been identified and whether there is any longitudinal model for efficacy endpoints or surrogate endpoints of efficacy that can be evaluated to better understand the E–R relationship. The oncology dose-finding workshop organized by FDA and the American Association for Cancer Research (AACR) in 2016 presented levantinib as an example of dose adjustment integrated E–R analysis (DAIER). For levantinib, FDA suggested an E–R analysis using dose-altering AEs models to evaluate different dosing regimens and efficacy . Typically in E–R analyses, drug exposure is assumed to be the cause, and response to be the outcome. However, if disease progression or remission influences pharmacokinetic (PK) parameters over time, this interaction between treatment response and PK parameters could result in artificial E–R relationships. Anti-programmed death-1 (anti-PD1) immunotherapies nivolumab and pembrolizumab exhibited time-dependent pharmacokinetics and a correlation between drug clearance changes over time and survival rates . In these situations, directly linking drug exposure at steady state to clinical outcomes in a single-dose trial may yield an over-steep E–R relationship, deviating from the true underlying relationship. Interestingly, the nivolumab baseline clearance had a strong association with survival, relative to all evaluated exposure and covariates in a multivariable E–R analysis . Wang et al. showed that baseline nivolumab clearance can be predicted by a composite of cytokine signatures using machine learning approach and the patients with predicted high nivolumab’s clearance (CL) is associated with poor survival regardless of treatment (nivolumab or chemotherapy) in patients with advanced melanoma and renal cell carcinoma (RCC) . These results support the hypothesis that nivolumab CL can be used as a prognostic marker for patient disease status. Moreover, this highlighted the importance of including more than one dose level in E–R analysis to reduce the confounding effect between exposure and CL. It has been demonstrated by Liu, et al., through simulation, that using exposure variables observed or derived from the first treatment cycle for an E–R analysis may minimize this bias . Furthermore, confounded relationships between baseline risk factors for survival and drug exposure have also been reported, complicating the choice of exposure metric to use in these circumstances and the interpretation of any observed E–R relationship . An example of confounded baseline risk factors has been reported for trastuzumab, which is indicated for the treatment of HER2 overexpressing breast cancer, metastatic gastric, and gastroesophageal junction adenocarcinoma . Yan et al. performed an exploratory analysis with simulated trough concentration in the first treatment cycle and overall survival. The Kaplan–Meier curves stratified by different exposure quartiles suggested a E–R trend based on the exposure metric. However, the unbalanced distribution of baseline disease burden across different exposure quartiles was responsible of the apparent E–R relationship. To minimize the confounding effect, a propensity matching strategy for adjusting measured confounders, which are defined by a stepwise Cox regression model, was applied. After appropriate matching, patients in the first exposure quartile of trastuzumab show no survival benefit over control . Although dose could be used as the exposure metric of choice, systemic exposure (i.e., plasma/serum/blood drug concentration) often is a more precise metric as it accounts for nonlinearities and inter-individual variability in the pharmacokinetics of the drug. Depending on the type of E–R analyses, metrics of early exposure (and its correlations with surrogates of efficacy), exposure at steady state, or the time course of drug concentration (time-varying concentration) could be used. Another important consideration when choosing the exposure metric is potential dose regimen comparisons. Prediction of response in other schedules of administration based on just one dose schedule often will lead to inaccurate outcomes. However, when information is available from more than one dose schedule, evaluating the most sensitive metric of exposure (e.g., C trough , C max , C avg ) for clinical outcomes may help to appropriately account for differences in dose schedules. In dose-escalating studies, looking at concentration or dose versus biomarker changes as surrogates of efficacy or proof of biological response could help dose selection and establish the maximum tolerable dose that will lead to maximum biological response. Further, it is worth mentioning that in order to collect sufficient information over an informative range of doses or exposure, an adaptive/Bayesian design could be a good choice. However, such study designs may cause logistical and operational challenges . Therefore, the use of the drug exposure metrics depends on the study design, the drug mechanism of action, and the nature of the relationship between exposure and response (i.e., short versus long-term effects). In oncology when there is a single read on efficacy (i.e., objective response rate, ORR), it may be more appropriate to use a simple metric that represents the drug exposure over the course of the treatment. In this case, C avg or C trough could be good metrics of choice since these provide an average measure of exposure; however for E–R safety analyses with acute AEs, the C max prior to the AE event could be explored. Cumulative AUC is confounded with time on study and careful consideration should be given to the use of this metric, as well as the nature of the endpoint under study. AUC at steady state (AUC ss ) is a valid exposure metric often associated with long-term effects. However, when looking at steady state metrics (ie., C trough , C avg , AUC ss , C max,ss ) for a given dose schedule, those may be correlated and selecting one metric versus another will often not lead to different conclusions. Another consideration is the use of model-predicted drug exposure versus observed concentration values. In the case of a drug with very high variability in exposure (i.e., > 70% residual error) and sparse PK sampling, model-predicted exposure profiles may be questionable and the use of observed C trough values over the course of the treatment could be a more reliable exposure metric. In summary, a variety of exposure metrics could be considered when performing E–R analysis; what should guide the selected drug metrics for E–R is multifactorial and includes the type of E–R analysis, endpoint under consideration, understanding of the drug pharmacokinetics and mechanism of action, the available drug exposure data, and the nature of safety and efficacy endpoints under analysis. Table provides a summary of exposure metrics used for different E–R analyses and clinical response (efficacy/safety) endpoints. Logistic regression In addition to continuous clinical endpoints, categorical or ordered categorical (ordinal) endpoints, such as graded AEs or ORR, are often considered. Depending on the granularity of data collected and the objective of the analysis, logistic regressions and Markov chain models can be used to analyze these endpoints. Logistic regression modeling is a widely used approach in E–R analysis that enables characterization of the relationship between drug exposure and ordered categorical clinical outcomes . In logistic regression analysis, a linear predictor with a link function is used. The linear regression yields a nonlinear relationship commonly through a logit predictor, but other links could be used, such as probit or complementary log–log models. Logistic regression models the probability of an outcome (binomial dependent variable) based on independent variables such as demographic factors, drug exposure, etc. This type of analysis informs about the likelihood of an event happening. The characterization of relationships between exposure and clinical outcomes of both efficacy (i.e., objective response, based on short-term tumor response) and safety (i.e., AEs of clinical interest) provides a quantitative assessment of the benefit/risk profile and is often used for dose selection in late-stage (i.e., Phase 3) oncology drug development. In some situations, these results can be utilized for dose recommendation in regulatory submissions (i.e., a rolling submission), if the treatment fits an unmet medical need and short-term clinical outcome (e.g., OR) is promising . The development of a logistic regression model generally includes 3 steps; (1) development of a base model, with an evaluation of the appropriate functional form of exposure metrics that may include interaction terms when the drug is given in combination with another drug; (2) development of a full model, including assessment of covariates of interest; (3) and development of the final model after considering the contribution of all potential variables of interest. Even though a final model can be simplified using backward elimination to achieve parsimony, the presence or absence of parameters of minimal impact in the probability will not result in model prediction differences . A minimum of 10 events per predictive variable analyzed in the logistic regression model is recommended for accurate and precise estimation of the regression coefficients. Peduzzi et al. evaluated the effect of the number of events per estimated parameters in logistic regression. They concluded that less than 10 events per predictive variable can lead to major biases, questioning the validity of the logistic regression analysis if those conditions are not fulfilled . Likewise, the number of covariates to be included in the full model should consider the total events in the analysis dataset to avoid over-parameterization. The utilization of the full model provides the benefit of avoiding biased parameter estimation by accounting for all measured covariate effects. Moreover, the full model avoids confounding among covariates and exposure metrics . A recent E–R efficacy analysis of nivolumab showed how baseline CL was a significant predictor of efficacy endpoints, OR, and OS. Subjects with higher CL had poor efficacy outcomes (i.e., higher risk of death and lower OR). Prior E–R analyses found only apparent E–R relationships that were misleading by ignoring the incorporation of baseline CL. It is important to note that baseline CL was not included as an exposure parameter but as a surrogate of disease status and other baseline confounders. In fact, exposure did not influence the outcome. Subjects within a given dose level with the lowest CL were more likely to be responders. No relationship was found comparing exposure levels and clinical response across the range of doses evaluated. The recent approval of flat dosing of nivolumab (240 mg Q2W, and 480 mg Q4W) away from originally approved body weight normalized dosing (3 mg/kg Q2W) was based on a flat E–R relationship that included both baseline CL and systemic exposure in the model, and that increasing drug exposure was unlikely to improve efficacy outcomes (thus, indicative of maximal response). This example highlights the importance of utilizing a full model approach for trials with limited dose-finding information to assess covariates' effect and exposure levels on relevant clinical outcomes . Longitudinal logistic regression models with markov elements E–R analysis for categorical endpoints commonly focuses on correlating the highest AE grades observed during the course of treatment within a patient’s measure of exposure. However, this approach ignores the time course of drug exposure and AE development, and, as such, loses valuable information. Markov models are useful when the outcomes are categorical measures that are monitored continuously over time. Markov modeling has been widely used in ordered and non-ordered categorical analyses in many therapeutic areas of drug development . The Markov models considered in this review are first-order models, Markov chains of second or higher orders are models in which the probability of one event changing depends on 2 or more preceding ones and will not be discussed here. A central concept in Markov models is the transition probability, which models the probability of one event changing to another event (or staying the same). For example, for the 2-event state, like whether a patient develops a rash after starting treatment with a drug, there are 4 transition probabilities. If ‘0’ is no rash and ‘1’ is rash, then the transition probabilities are P 00 , P 01 , P 10 , P 11 where the subscript refers to a change from the current state i to the future state j . Compared to logistic regression models where there are no transition probabilities and characterization focuses on the probability of having an event at given drug exposure, Markov modeling provides a description of longitudinal clinical data over time and assumes that (i.) the distribution of future states depends only on the current state and (ii.) is not a function of the whole history of events (first-order Markov models). Given this defined Markov property , Markov modeling can capture the onset of events, their duration, and severity as they change over time; it also allows the assessment of an exposure effect on the transition probability from one event state to another state and the severity of the event. For E–R analyses using Markov models where longitudinal categorical data are modeled, drug exposure over time would be a more appropriate exposure metric than a single exposure measurement, such as C max or AUC up to the event. Time-varying exposure would be more associated with the onset and duration of AEs when the dose was modified or discontinued due to adverse events. The effect of co-medication on transition probabilities can also be considered if co-medications were applied for the treatment of specific AEs and with the resolution of AEs. If a delay between exposure and clinical outcome was observed, effective concentrations derived from an effect compartment can be evaluated. Prognostic factors (e.g., biomarkers) could be included to assess their effects on the severity of an AE during model development. Different types of Markov models have been reported and will be briefly presented here. The discrete-time Markov model (DTMM) combines proportional odds with a transition model that allows event changes more than 1 grade higher or lower. In this model, all possible transit probabilities can be estimated, with the assumption that the transit probabilities are independent of whatever the time interval is between the two assessments, making it an ideal candidate when we have uniform time interval assessments. The continuous-time Markov model (CTMM) combines proportional odds with a transition model that prohibits event changes more than 1 grade higher or lower. For example, a transition directly from State 3 to State 1 is prohibited, whereas the transition from State 1 to State 2 and then from State 2 to State 3 is allowed, and vice-versa. With a CTMM, the influence of the previous state on the probability of the current state changes with time (usually decreasing over time as the time interval between measurements increases). Thus, this model is preferred when the observation intervals are non-uniform across patients either due to study design or missing observations. Lastly, the minimal CTMM (mCTMM), which is a simplification of the CTMM, is characterized by independent transition rates between two consecutive states and governed by a single parameter, the mean equilibration time (MET). A schematic of these models is presented in Fig. . Schindler et al. showed a few examples of model performance between DTMM and mCTMM suggesting that mCTMM had the potential in describing the data reasonably well with a more parsimonious model structure relative to the CTMM . In addition, the effect of covariates (e.g., exposure, biomarker) on the probability of each state can be described in a single relationship. Lu et al. compared the model performance between a proportional odds model, CTMM, and DTMM in which models were developed using weekly based time-course of muscle spasm AE data. Model performance showed that the odds model was influenced by time–frequency, DTMM was unable to describe unevenly spaced data, and CTMM seemed to perform well in all evaluated data frequencies (daily, weekly, and unevenly spaced) . In addition to continuous clinical endpoints, categorical or ordered categorical (ordinal) endpoints, such as graded AEs or ORR, are often considered. Depending on the granularity of data collected and the objective of the analysis, logistic regressions and Markov chain models can be used to analyze these endpoints. Logistic regression modeling is a widely used approach in E–R analysis that enables characterization of the relationship between drug exposure and ordered categorical clinical outcomes . In logistic regression analysis, a linear predictor with a link function is used. The linear regression yields a nonlinear relationship commonly through a logit predictor, but other links could be used, such as probit or complementary log–log models. Logistic regression models the probability of an outcome (binomial dependent variable) based on independent variables such as demographic factors, drug exposure, etc. This type of analysis informs about the likelihood of an event happening. The characterization of relationships between exposure and clinical outcomes of both efficacy (i.e., objective response, based on short-term tumor response) and safety (i.e., AEs of clinical interest) provides a quantitative assessment of the benefit/risk profile and is often used for dose selection in late-stage (i.e., Phase 3) oncology drug development. In some situations, these results can be utilized for dose recommendation in regulatory submissions (i.e., a rolling submission), if the treatment fits an unmet medical need and short-term clinical outcome (e.g., OR) is promising . The development of a logistic regression model generally includes 3 steps; (1) development of a base model, with an evaluation of the appropriate functional form of exposure metrics that may include interaction terms when the drug is given in combination with another drug; (2) development of a full model, including assessment of covariates of interest; (3) and development of the final model after considering the contribution of all potential variables of interest. Even though a final model can be simplified using backward elimination to achieve parsimony, the presence or absence of parameters of minimal impact in the probability will not result in model prediction differences . A minimum of 10 events per predictive variable analyzed in the logistic regression model is recommended for accurate and precise estimation of the regression coefficients. Peduzzi et al. evaluated the effect of the number of events per estimated parameters in logistic regression. They concluded that less than 10 events per predictive variable can lead to major biases, questioning the validity of the logistic regression analysis if those conditions are not fulfilled . Likewise, the number of covariates to be included in the full model should consider the total events in the analysis dataset to avoid over-parameterization. The utilization of the full model provides the benefit of avoiding biased parameter estimation by accounting for all measured covariate effects. Moreover, the full model avoids confounding among covariates and exposure metrics . A recent E–R efficacy analysis of nivolumab showed how baseline CL was a significant predictor of efficacy endpoints, OR, and OS. Subjects with higher CL had poor efficacy outcomes (i.e., higher risk of death and lower OR). Prior E–R analyses found only apparent E–R relationships that were misleading by ignoring the incorporation of baseline CL. It is important to note that baseline CL was not included as an exposure parameter but as a surrogate of disease status and other baseline confounders. In fact, exposure did not influence the outcome. Subjects within a given dose level with the lowest CL were more likely to be responders. No relationship was found comparing exposure levels and clinical response across the range of doses evaluated. The recent approval of flat dosing of nivolumab (240 mg Q2W, and 480 mg Q4W) away from originally approved body weight normalized dosing (3 mg/kg Q2W) was based on a flat E–R relationship that included both baseline CL and systemic exposure in the model, and that increasing drug exposure was unlikely to improve efficacy outcomes (thus, indicative of maximal response). This example highlights the importance of utilizing a full model approach for trials with limited dose-finding information to assess covariates' effect and exposure levels on relevant clinical outcomes . E–R analysis for categorical endpoints commonly focuses on correlating the highest AE grades observed during the course of treatment within a patient’s measure of exposure. However, this approach ignores the time course of drug exposure and AE development, and, as such, loses valuable information. Markov models are useful when the outcomes are categorical measures that are monitored continuously over time. Markov modeling has been widely used in ordered and non-ordered categorical analyses in many therapeutic areas of drug development . The Markov models considered in this review are first-order models, Markov chains of second or higher orders are models in which the probability of one event changing depends on 2 or more preceding ones and will not be discussed here. A central concept in Markov models is the transition probability, which models the probability of one event changing to another event (or staying the same). For example, for the 2-event state, like whether a patient develops a rash after starting treatment with a drug, there are 4 transition probabilities. If ‘0’ is no rash and ‘1’ is rash, then the transition probabilities are P 00 , P 01 , P 10 , P 11 where the subscript refers to a change from the current state i to the future state j . Compared to logistic regression models where there are no transition probabilities and characterization focuses on the probability of having an event at given drug exposure, Markov modeling provides a description of longitudinal clinical data over time and assumes that (i.) the distribution of future states depends only on the current state and (ii.) is not a function of the whole history of events (first-order Markov models). Given this defined Markov property , Markov modeling can capture the onset of events, their duration, and severity as they change over time; it also allows the assessment of an exposure effect on the transition probability from one event state to another state and the severity of the event. For E–R analyses using Markov models where longitudinal categorical data are modeled, drug exposure over time would be a more appropriate exposure metric than a single exposure measurement, such as C max or AUC up to the event. Time-varying exposure would be more associated with the onset and duration of AEs when the dose was modified or discontinued due to adverse events. The effect of co-medication on transition probabilities can also be considered if co-medications were applied for the treatment of specific AEs and with the resolution of AEs. If a delay between exposure and clinical outcome was observed, effective concentrations derived from an effect compartment can be evaluated. Prognostic factors (e.g., biomarkers) could be included to assess their effects on the severity of an AE during model development. Different types of Markov models have been reported and will be briefly presented here. The discrete-time Markov model (DTMM) combines proportional odds with a transition model that allows event changes more than 1 grade higher or lower. In this model, all possible transit probabilities can be estimated, with the assumption that the transit probabilities are independent of whatever the time interval is between the two assessments, making it an ideal candidate when we have uniform time interval assessments. The continuous-time Markov model (CTMM) combines proportional odds with a transition model that prohibits event changes more than 1 grade higher or lower. For example, a transition directly from State 3 to State 1 is prohibited, whereas the transition from State 1 to State 2 and then from State 2 to State 3 is allowed, and vice-versa. With a CTMM, the influence of the previous state on the probability of the current state changes with time (usually decreasing over time as the time interval between measurements increases). Thus, this model is preferred when the observation intervals are non-uniform across patients either due to study design or missing observations. Lastly, the minimal CTMM (mCTMM), which is a simplification of the CTMM, is characterized by independent transition rates between two consecutive states and governed by a single parameter, the mean equilibration time (MET). A schematic of these models is presented in Fig. . Schindler et al. showed a few examples of model performance between DTMM and mCTMM suggesting that mCTMM had the potential in describing the data reasonably well with a more parsimonious model structure relative to the CTMM . In addition, the effect of covariates (e.g., exposure, biomarker) on the probability of each state can be described in a single relationship. Lu et al. compared the model performance between a proportional odds model, CTMM, and DTMM in which models were developed using weekly based time-course of muscle spasm AE data. Model performance showed that the odds model was influenced by time–frequency, DTMM was unable to describe unevenly spaced data, and CTMM seemed to perform well in all evaluated data frequencies (daily, weekly, and unevenly spaced) . Time-course of myelosuppression Blood cell production in the bone marrow is a highly prolific process, making it susceptible to the inhibitory effects of anti-cancer agents with anti-proliferative activities or with immuno-modulatory agents directly targeting markers expressed on the surface of hematopoietic stem cell progenitors or mature blood cells. In fact, myelosuppression manifested as decreases in circulating red blood cells, white blood cells, or platelets, is among the most frequent AEs observed for anticancer therapeutics . Quantitative understanding of the drug effects on blood cell production is important for the assessment or prediction of the myelosuppression risk, as well as the optimization of the dose and regimen, to reduce myelosuppression-related AEs. The dose/E–R relationships for drug-induced myelosuppression have often been analyzed by empirical or mechanism-based modeling approaches . Empirical models are usually developed by a theoretical understanding of drug behaviors with very few assumptions of the data . Semi-mechanistic disease models use simplified biological systems to describe the available data falling between empirical models and mechanistic models. Empirical modeling approaches involve either regression-based correlation analyses between a descriptor of the blood cell change (such as maximum % decrease from baseline, often called the nadir, or incidence of particular AEs) and the dose or a particular drug exposure parameter [such as the area under the curve (AUC) and time above a threshold concentration] or by empirically linking the dynamic change in blood cell counts to a time-variant drug exposure parameter through a particular function (e.g., E max models). Semi-mechanistic and mechanism-based modeling approaches utilize differential equations to describe the physiological process of hematopoiesis and the pharmacological perturbation by the inducing agents. In practice, the choice of a specific modeling approach in evaluating drug-induced myelosuppression often depends on the purpose of the analysis and the type of data available. Empirical myelosuppression models are often expressed as the absolute or relative decrease from baseline at nadir, the maximum percentage of decrease from baseline, the duration below a threshold cell count, the area between a threshold line and the observed cell counts vs time curve, or the incidence of a graded hematological adverse event . Typical exposure parameters such as AUC, time above a threshold concentration, and C max may be explored in the analysis. The relationship between drug exposure and the myelosuppressive effects is modeled without regard to the time course of drug concentrations or blood cell counts. This type of correlation analysis may be used to determine the myelosuppression response or outcome associated with a certain dose/exposure level. This analysis is relatively easy to implement and does not require complete blood cell time course data obtained from extensive sampling. However, this type of modeling analysis has no or limited value in predicting myelosuppression time courses or responses beyond the tested dose range and regimens. Empirical longitudinal models may also describe the time course of blood cell change following drug administration . Because there is a typical delay in the myelosuppressive effect in relation to systemic drug concentrations, empirical longitudinal modeling assumes a direct drug effect from a drug exposure parameter (e.g., cumulative AUC or C average ), with or without adding a lag time parameter. This empirical modeling allows prediction of blood cell count time courses under certain conditions; however, given the often lack of physiological meaning for the PD parameters in these models, there are also challenges to extrapolate the models to untested conditions in many cases, such as cross-species translation or predicting the effects for similar compounds. Semi-mechanistic modeling of myelosuppression is based on an understanding of the hematopoiesis process and how the drug perturbs the process. Anti-cancer agents may cause bone marrow suppression through direct cytotoxicity on differentiated bone marrow cells, inhibition of progenitor or precursor cell proliferation, or disruption of growth factor signaling pathways involved in differentiation . There have been generations of mechanism-based mathematical models to describe drug-induced myelosuppression over the last two decades . The most commonly cited model is the Friberg model, which has been the basis of similar models with various modifications . The Friberg model and its related models share the following key structural components (Fig. ): (1) one or more proliferating compartments with a pool of proliferating cells that can be derived from self-renewable HSCs in the bone marrow; (2) a series of transit compartments representing nonproliferating cells at different maturation stages in the bone marrow; (3) a compartment representing circulating cells with natural turnover; (4) a negative feedback loop where circulating cells regulate the proliferation of bone marrow cells in the proliferating compartment. These structural features are represented by ordinary differential equations, with system-related parameters inherent to the body system and drug-specific parameters that vary by the inducing agents. The drug effects are incorporated into the models in a way consistent with the myelotoxicity mechanisms. The Friberg and related models have been utilized to describe various types of drug-induced myelosuppression effects including leukopenia, neutropenia, thrombocytopenia, and anemia . These models have been able to capture the delay in myelosuppressive effects relative to systemic drug concentrations, and the recovery, rebound and return to baseline for circulating blood cells upon treatment cessation. Despite the complexity of the models, the total number of parameters remains identifiable. With the mathematical representation of the physiological and pharmacological processes involved in myelosuppression, the models allow the estimation of system-related and drug-specific parameters. The semi-mechanistic basis enhances the confidence in using these models to extrapolate beyond tested conditions. These models have been used to predict the time courses of circulating blood cell profiles from different doses or regimens. Also, the models can be used in the cross-species translation of myelosuppressive effects . Integration of these semi-mechanistic models into population PK/PD modeling can assess the inter-patient variability and influential covariates of drug-induced myelosuppressive effects. Since the introduction of this semi-mechanistic model, it has become the golden standard approach to model the myelosuppressive effects of chemotherapy. However, due to the lack of clinical data characterizing the upstream processes of granulopoiesis, some assumptions must be made, such as a constant transit time between maturation compartments. In addition, since the semi-mechanistic model is largely data-driven, the predictive ability of this model might be limited compared with the full mechanistic model . Several researchers have extended the applications of this model by incorporating more complex mechanisms and relationships. For example, Quartino et al. integrated granulocyte colony-stimulating factor (G-CSF)—myelosuppression model to describe the dynamics of endogenous G-CSF and absolute neutrophil count (ANC) following chemotherapy . The final model captured both the initial rise in endogenous G-CSF concentrations following chemotherapy-induced neutropenia and the subsequent return to baseline for G-CSF and ANC. This semi-mechanistic model adequately described the time-course of ANC where the feedback mechanism of G-CSF regulated the neutrophil production and maturation in the bone marrow. QTc interval prolongation Since adoption in the ICH E14 guidance in 2015, concentration-QTc (C-QTc) analyses have rapidly replaced Thorough QT (TQT) studies for the assessment of QT prolongation risk during small molecule oncology development . However, the complexities of oncology drug development, including the quick pace of development, differences in trial design, co-medications, and risk–benefit profile in the face of life-threatening disease, pose a number of unique challenges in applying this methodology to exclude a risk of QT prolongation or accurately quantifying the effect size when the drug has a known QT liability. The scope of this section relates primarily to small molecule development; as specified by ICH E14, large targeted proteins and monoclonal antibodies have a low likelihood of direction channel interactions, and a thorough QT/QTc study (or C-QTc analysis replacing this study) is generally not necessary unless the potential for proarrhythmic risk is suggested by mechanistic considerations or data from clinical or nonclinical studies . A scientific white paper on concentration-QTc modeling published in 2018 provides clear guidance and recommendations on standardizing C-QT analyses intended for assessing QTc prolongation risk under ICH E14 in healthy volunteers . In addition to guidance on Phase 1 study design in order to support a C-QTc analysis, a pre-specified linear mixed-effects (LME) C-QTc model was proposed, including variations on this model depending on the available data, and provides guidance on exploratory and goodness-of-fit plots for model evaluation . Assumptions of the pre-specified LME C-QTc model, including lack of drug effect on heart rate (HR), adequacy of the HR correction used for the QT interval (i.e., lack of trend on QTcF vs RR plot), lack of hysteresis between concentration and QTc effects, and a linear C-QTc relationship (vs non-linear relationships) should be explored and justified during model development . Several elements that are important components of a C-QTc analysis in non-oncology drug development are typically unavailable for an oncology program—namely concentration and QTc data at a supratherapeutic exposure, and inclusion of data from placebo subjects, as is typically available during single- or multiple-dose escalation cohorts in many non-oncology programs conducted in healthy volunteers . The lack of placebo subjects in oncology trials typically requires that inferences from a C-QTc analysis are drawn based on baseline-corrected QTc (i.e., ΔQTc), rather than baseline-corrected, placebo-corrected QTc (ΔΔQTc). In addition, the lack of placebo data introduces diurnal fluctuation in QTc as a potential confounding factor for drug effect on QTc intervals . The collection of time-matched baseline (i.e., at the same time points to be collected post-dose) in order to account for diurnal fluctuation has been successfully implemented for C-QT analysis of single-arm trials ; where only a pre-dose baseline was available, the inclusion of categorical time effects in the C-QTc model has recently been proposed . Where a compound is known to substantially affect HR (i.e., mean change > 10 bpm), the inclusion of a time-matched baseline can allow a patient-specific HR correction to be calculated from baseline QT/RR data, although consensus has not been achieved on the optimal approach . Due to safety concerns as well as the differences in risk–benefit profile in oncology, the highest tested dose and exposure are frequently also the therapeutic exposure, and meeting the requirement for a supratherapeutic exposure is not possible . This may change with recent regulatory requests to greater explore the dose–response relationship in early clinical studies. Currently, however,recent draft revisions to ICH S7B and E14 may enable a greater number of oncology small molecule development programs to meet the supratherapeutic exposure requirement . When a compound meets the definition of a ‘double negative nonclinical assessment’ for QTc prolongation, a supratherapeutic exposure at the ‘high clinical exposure’ (increase in exposure under the effect of intrinsic or extrinsic factors at the maximum therapeutic dose) is required to exclude a positive control, rather than ≥ twofold the high clinical exposure as required under the previous guidance . As discussed in the ICH E14 Q&A guidance documents, in the absence of a positive control or supratherapeutic exposure, there is a reluctance to conclude a lack of effect on QT, however, if the upper bound of the two-sided 90% confidence interval around the estimated maximal effect on ΔQTc is less than 10 ms, the treatment is unlikely to have an actual mean effect as large as 20 ms . The sample size is also an important consideration for C-QTc analysis. While general guidance of 4–8 subjects on drug and 2–4 subjects on placebo across at least 4 dose cohorts has been proposed based on low false-negative and false-positive rates , these studies generally included placebo subjects and/or a supratherapeutic exposure typical of a non-oncology program . A recent review of QT prolongation risk assessment of small molecule oncology NDAs from 2011 to 2019 found that where a C-QTc analysis was performed, sample size varied greatly, but was generally smaller when the C-QTc dataset was based on data from early phase studies (~ 20–300 patients), and larger where data from later phase studies (or pooled early and late phase) was used (~ 100–800 patients) . It was noted that no clear trend was identified between sample size and the labeling recommendation category to which an NDA was assigned . Other considerations that may be more frequently encountered in oncology include pooling of data from patients with different underlying malignancies across treatment arms or trials in order to increase sample size or dose range, which may increase the risk of confounders such as differences in health status and concomitant medications, as well as in study conduct or ECG acquisition or analysis. When pooling of data is required, between-study differences and potential bias should be evaluated and justified; this may include through exploratory plots and via the inclusion of a study effect variable on key model parameters . Looking at C-QTc relationships with parent compound and/or active metabolites might be required sometimes to characterize QTc effects . Blood cell production in the bone marrow is a highly prolific process, making it susceptible to the inhibitory effects of anti-cancer agents with anti-proliferative activities or with immuno-modulatory agents directly targeting markers expressed on the surface of hematopoietic stem cell progenitors or mature blood cells. In fact, myelosuppression manifested as decreases in circulating red blood cells, white blood cells, or platelets, is among the most frequent AEs observed for anticancer therapeutics . Quantitative understanding of the drug effects on blood cell production is important for the assessment or prediction of the myelosuppression risk, as well as the optimization of the dose and regimen, to reduce myelosuppression-related AEs. The dose/E–R relationships for drug-induced myelosuppression have often been analyzed by empirical or mechanism-based modeling approaches . Empirical models are usually developed by a theoretical understanding of drug behaviors with very few assumptions of the data . Semi-mechanistic disease models use simplified biological systems to describe the available data falling between empirical models and mechanistic models. Empirical modeling approaches involve either regression-based correlation analyses between a descriptor of the blood cell change (such as maximum % decrease from baseline, often called the nadir, or incidence of particular AEs) and the dose or a particular drug exposure parameter [such as the area under the curve (AUC) and time above a threshold concentration] or by empirically linking the dynamic change in blood cell counts to a time-variant drug exposure parameter through a particular function (e.g., E max models). Semi-mechanistic and mechanism-based modeling approaches utilize differential equations to describe the physiological process of hematopoiesis and the pharmacological perturbation by the inducing agents. In practice, the choice of a specific modeling approach in evaluating drug-induced myelosuppression often depends on the purpose of the analysis and the type of data available. Empirical myelosuppression models are often expressed as the absolute or relative decrease from baseline at nadir, the maximum percentage of decrease from baseline, the duration below a threshold cell count, the area between a threshold line and the observed cell counts vs time curve, or the incidence of a graded hematological adverse event . Typical exposure parameters such as AUC, time above a threshold concentration, and C max may be explored in the analysis. The relationship between drug exposure and the myelosuppressive effects is modeled without regard to the time course of drug concentrations or blood cell counts. This type of correlation analysis may be used to determine the myelosuppression response or outcome associated with a certain dose/exposure level. This analysis is relatively easy to implement and does not require complete blood cell time course data obtained from extensive sampling. However, this type of modeling analysis has no or limited value in predicting myelosuppression time courses or responses beyond the tested dose range and regimens. Empirical longitudinal models may also describe the time course of blood cell change following drug administration . Because there is a typical delay in the myelosuppressive effect in relation to systemic drug concentrations, empirical longitudinal modeling assumes a direct drug effect from a drug exposure parameter (e.g., cumulative AUC or C average ), with or without adding a lag time parameter. This empirical modeling allows prediction of blood cell count time courses under certain conditions; however, given the often lack of physiological meaning for the PD parameters in these models, there are also challenges to extrapolate the models to untested conditions in many cases, such as cross-species translation or predicting the effects for similar compounds. Semi-mechanistic modeling of myelosuppression is based on an understanding of the hematopoiesis process and how the drug perturbs the process. Anti-cancer agents may cause bone marrow suppression through direct cytotoxicity on differentiated bone marrow cells, inhibition of progenitor or precursor cell proliferation, or disruption of growth factor signaling pathways involved in differentiation . There have been generations of mechanism-based mathematical models to describe drug-induced myelosuppression over the last two decades . The most commonly cited model is the Friberg model, which has been the basis of similar models with various modifications . The Friberg model and its related models share the following key structural components (Fig. ): (1) one or more proliferating compartments with a pool of proliferating cells that can be derived from self-renewable HSCs in the bone marrow; (2) a series of transit compartments representing nonproliferating cells at different maturation stages in the bone marrow; (3) a compartment representing circulating cells with natural turnover; (4) a negative feedback loop where circulating cells regulate the proliferation of bone marrow cells in the proliferating compartment. These structural features are represented by ordinary differential equations, with system-related parameters inherent to the body system and drug-specific parameters that vary by the inducing agents. The drug effects are incorporated into the models in a way consistent with the myelotoxicity mechanisms. The Friberg and related models have been utilized to describe various types of drug-induced myelosuppression effects including leukopenia, neutropenia, thrombocytopenia, and anemia . These models have been able to capture the delay in myelosuppressive effects relative to systemic drug concentrations, and the recovery, rebound and return to baseline for circulating blood cells upon treatment cessation. Despite the complexity of the models, the total number of parameters remains identifiable. With the mathematical representation of the physiological and pharmacological processes involved in myelosuppression, the models allow the estimation of system-related and drug-specific parameters. The semi-mechanistic basis enhances the confidence in using these models to extrapolate beyond tested conditions. These models have been used to predict the time courses of circulating blood cell profiles from different doses or regimens. Also, the models can be used in the cross-species translation of myelosuppressive effects . Integration of these semi-mechanistic models into population PK/PD modeling can assess the inter-patient variability and influential covariates of drug-induced myelosuppressive effects. Since the introduction of this semi-mechanistic model, it has become the golden standard approach to model the myelosuppressive effects of chemotherapy. However, due to the lack of clinical data characterizing the upstream processes of granulopoiesis, some assumptions must be made, such as a constant transit time between maturation compartments. In addition, since the semi-mechanistic model is largely data-driven, the predictive ability of this model might be limited compared with the full mechanistic model . Several researchers have extended the applications of this model by incorporating more complex mechanisms and relationships. For example, Quartino et al. integrated granulocyte colony-stimulating factor (G-CSF)—myelosuppression model to describe the dynamics of endogenous G-CSF and absolute neutrophil count (ANC) following chemotherapy . The final model captured both the initial rise in endogenous G-CSF concentrations following chemotherapy-induced neutropenia and the subsequent return to baseline for G-CSF and ANC. This semi-mechanistic model adequately described the time-course of ANC where the feedback mechanism of G-CSF regulated the neutrophil production and maturation in the bone marrow. Since adoption in the ICH E14 guidance in 2015, concentration-QTc (C-QTc) analyses have rapidly replaced Thorough QT (TQT) studies for the assessment of QT prolongation risk during small molecule oncology development . However, the complexities of oncology drug development, including the quick pace of development, differences in trial design, co-medications, and risk–benefit profile in the face of life-threatening disease, pose a number of unique challenges in applying this methodology to exclude a risk of QT prolongation or accurately quantifying the effect size when the drug has a known QT liability. The scope of this section relates primarily to small molecule development; as specified by ICH E14, large targeted proteins and monoclonal antibodies have a low likelihood of direction channel interactions, and a thorough QT/QTc study (or C-QTc analysis replacing this study) is generally not necessary unless the potential for proarrhythmic risk is suggested by mechanistic considerations or data from clinical or nonclinical studies . A scientific white paper on concentration-QTc modeling published in 2018 provides clear guidance and recommendations on standardizing C-QT analyses intended for assessing QTc prolongation risk under ICH E14 in healthy volunteers . In addition to guidance on Phase 1 study design in order to support a C-QTc analysis, a pre-specified linear mixed-effects (LME) C-QTc model was proposed, including variations on this model depending on the available data, and provides guidance on exploratory and goodness-of-fit plots for model evaluation . Assumptions of the pre-specified LME C-QTc model, including lack of drug effect on heart rate (HR), adequacy of the HR correction used for the QT interval (i.e., lack of trend on QTcF vs RR plot), lack of hysteresis between concentration and QTc effects, and a linear C-QTc relationship (vs non-linear relationships) should be explored and justified during model development . Several elements that are important components of a C-QTc analysis in non-oncology drug development are typically unavailable for an oncology program—namely concentration and QTc data at a supratherapeutic exposure, and inclusion of data from placebo subjects, as is typically available during single- or multiple-dose escalation cohorts in many non-oncology programs conducted in healthy volunteers . The lack of placebo subjects in oncology trials typically requires that inferences from a C-QTc analysis are drawn based on baseline-corrected QTc (i.e., ΔQTc), rather than baseline-corrected, placebo-corrected QTc (ΔΔQTc). In addition, the lack of placebo data introduces diurnal fluctuation in QTc as a potential confounding factor for drug effect on QTc intervals . The collection of time-matched baseline (i.e., at the same time points to be collected post-dose) in order to account for diurnal fluctuation has been successfully implemented for C-QT analysis of single-arm trials ; where only a pre-dose baseline was available, the inclusion of categorical time effects in the C-QTc model has recently been proposed . Where a compound is known to substantially affect HR (i.e., mean change > 10 bpm), the inclusion of a time-matched baseline can allow a patient-specific HR correction to be calculated from baseline QT/RR data, although consensus has not been achieved on the optimal approach . Due to safety concerns as well as the differences in risk–benefit profile in oncology, the highest tested dose and exposure are frequently also the therapeutic exposure, and meeting the requirement for a supratherapeutic exposure is not possible . This may change with recent regulatory requests to greater explore the dose–response relationship in early clinical studies. Currently, however,recent draft revisions to ICH S7B and E14 may enable a greater number of oncology small molecule development programs to meet the supratherapeutic exposure requirement . When a compound meets the definition of a ‘double negative nonclinical assessment’ for QTc prolongation, a supratherapeutic exposure at the ‘high clinical exposure’ (increase in exposure under the effect of intrinsic or extrinsic factors at the maximum therapeutic dose) is required to exclude a positive control, rather than ≥ twofold the high clinical exposure as required under the previous guidance . As discussed in the ICH E14 Q&A guidance documents, in the absence of a positive control or supratherapeutic exposure, there is a reluctance to conclude a lack of effect on QT, however, if the upper bound of the two-sided 90% confidence interval around the estimated maximal effect on ΔQTc is less than 10 ms, the treatment is unlikely to have an actual mean effect as large as 20 ms . The sample size is also an important consideration for C-QTc analysis. While general guidance of 4–8 subjects on drug and 2–4 subjects on placebo across at least 4 dose cohorts has been proposed based on low false-negative and false-positive rates , these studies generally included placebo subjects and/or a supratherapeutic exposure typical of a non-oncology program . A recent review of QT prolongation risk assessment of small molecule oncology NDAs from 2011 to 2019 found that where a C-QTc analysis was performed, sample size varied greatly, but was generally smaller when the C-QTc dataset was based on data from early phase studies (~ 20–300 patients), and larger where data from later phase studies (or pooled early and late phase) was used (~ 100–800 patients) . It was noted that no clear trend was identified between sample size and the labeling recommendation category to which an NDA was assigned . Other considerations that may be more frequently encountered in oncology include pooling of data from patients with different underlying malignancies across treatment arms or trials in order to increase sample size or dose range, which may increase the risk of confounders such as differences in health status and concomitant medications, as well as in study conduct or ECG acquisition or analysis. When pooling of data is required, between-study differences and potential bias should be evaluated and justified; this may include through exploratory plots and via the inclusion of a study effect variable on key model parameters . Looking at C-QTc relationships with parent compound and/or active metabolites might be required sometimes to characterize QTc effects . Time to event analysis: survival In analyzing survival data, two functions that are dependent on time are of particular interest: the survival and the hazard function. The survival function S(t) is the probability of surviving at least to time t. The hazard function h(t) is the instantaneous conditional probability of dying at time t having survived to that time. Survival curves can be estimated nonparametrically (i.e., Kaplan–Meier (KM) curves), semi-parametrically (i.e., Cox proportional hazard model), or parametrically (i.e., a lifetime model). The KM method estimates survival curves without the assumption of an underlying probability distribution in the presence of right-tailed censoring. Although easy to compute, it is hard to include covariate analysis, other than by grouping, in the models. Statistical significance between two survival curves can be made using a log-rank test, which tests the null hypothesis that there is no difference between the population survival curves (i.e., the probability of an event occurring at any time point is the same for the populations under comparison) against the alternative hypothesis that they are not the same. Cox's proportional hazards models are semi-parametric models making fewer assumptions than typical parametric methods. Cox models quantify how specific factors (covariates) influence the rate of a particular event happening at a given point in time. This rate is commonly referred to as the hazard rate. The hazard function can be written as a multiple linear regression of the logarithm of the hazard with the baseline hazard being the intercept term that varies with time: [12pt]{minimal} $$h( t ) = h_{0} ( t ) e^{{( {b_{1} x_{1} + b_{2} x_{2} + + b_{n} x_{n} } )}}$$ h t = h 0 t · e b 1 · x 1 + b 2 · x 2 + … + b n · x n where t represents the survival time, h(t) is the hazard function determined by a set of covariates where the coefficients (b 1 , b 2 …b n ) quantify the effect of covariates on the hazard, h 0 is the baseline hazard. The quantities exp(b i ) are called hazard ratios (HR). An HR > 1 increases the hazard indicating a positive effect of the covariate with event probability and therefore negatively associated with the length of survival. An HR < 1 reduces the hazard and increases the probability of survival. The main assumption of the Cox model is that the hazard is proportional: if an individual has a risk of death at some initial point that is twice as high as that of another individual, then at later times the risk of death remains twice as high, that is, the hazard ratio comparing two groups is constant over time. However, in oncology, the assumption of proportional hazard implicit into Cox’s models may not represent the ideal choice where the factor changing the hazard may vary with time. Unlike the Cox regression model which does not specify the distribution function of the hazard, there are several parametric models such as Weibull, Gompertz, exponential, log-normal, and log-logistic models where the hazard function has to be specified. These parametric models allow us to estimate the effects of covariates on the hazard function, including variables that may change over time like drug concentration/dose, age, tumor growth dynamics, or time since surgical intervention. It is important to highlight the inherent selection bias and immortal time bias in survival metrics in oncology . The TTE model approach with the estimation of a single survival function has its limitations. One alternative approach is the landmark method, which determines each patient’s response at some fixed time point, with survival estimates calculated from that time point and associated statistical tests being conditional on patients’ landmark responses excluding patients that die before the selected landmark timepoint . Another approach to address the challenges of correctly describing the hazard function over time is the use of multistate models. Beyer et al. developed a multistate model with transition hazards estimated using semiparametric models . Krishnan et al. proposed transition hazards using a parametric approach and mixture models . Five states were considered according to Response Evaluation Criteria in Solid Tumors (RECIST) and using predictions derived from a longitudinal tumor growth inhibition model. The model did not allow to move back from one state to another when a response level has been achieved and maybe changed over time. Transition rates between states were estimated and defining the different hazard distributions (see Fig. ). In oncology, analysis of the association between drug exposure levels and late-phase clinical outcomes, such as OS and PFS (or event-free survival EFS) are often limited in scope for regulatory submissions. In light of the paucity of data and time constraints, industry practice and regulatory expectations to rationalize the dose rely on performing logistic or Cox proportional hazard regression analysis of the registration-intent trial efficacy data. Even simpler exploratory approaches are sometimes considered to support the dose rationale, such as conducting a KM exploratory analysis of OS and/or PFS stratified by quartiles or tertiles of exposure levels . The general strategy is to demonstrate the absence of a relationship between exposure and the registrable endpoint within the range of exposure tested and infer from this that the dose is optimal since no substantial gain in efficacy can be achieved by increasing exposure levels further. Whether such analyses can be qualified as best practice is contentious as pragmatism is the main driver to rationalize the methodological framework used. Typically, the establishment of a recommended phase II dose is endorsed based on combined analyses of emerging safety signals and surrogate efficacy endpoints with proven independent prognostic value for OS or PFS (e.g., tumor burden or objective response obtained from earlier trial data). Confirmatory trials rarely study more than one dose, and inherently suffer from underpowered statistical consideration for E–R characterization of OS and PFS. Furthermore, the ever-increasing pace for regulatory filing driven by commercial incentives and rapid access of novel treatment to patients in unmet needs further restrains the possibility to develop a data-driven E–R model of OS or PFS for regulatory submission given the time-consuming nature of these analyses. The downside of simple analyses lies in identifying spurious relationships due to imbalance of known prognostic factors of OS at baseline in strata of exposure . For example, ramucirumab E–R showed a positive association with OS in gastric cancer patients following a Cox regression with C min,ss or C avg . However, the E–R model did not evaluate covariates such as C-reactive protein levels and tumor burden that are known prognostic factors of OS and could also impact PK as demonstrated in the case of tremelimumab , trastuzumab emtansine , and to a lesser extent with trastuzumab in the same indication . The recommendation in such a case would be to first interpret with great caution any apparent trend of E–R emerging from simple analyses of trials evaluating a single-dose regimen. To explore further the risk of potential confounding in the event of a positive E–R finding, more sophisticated methods such as the parametric hazard model and multistate models should be attempted to quantify E–R in a multivariate framework. A full model or a selection of covariates of clinical relevance for the endpoint of interest and for the PK metric chosen should then be applied to quantify the relative contribution of covariates on the underlying E–R. As the last step, exposure should be removed from the model in case its causative link with OS or PFS assumed by the model is no longer supported by the data and can be fully explained by prognostic or predictive covariates remaining in the model. The best example of this good practice would be the already mentioned nivolumab E–R analysis in which baseline CL was included in the model . Technical considerations, such as the choice of the PK metrics (C max , AUC) or when this metric is measured (first cycle vs steady-state), and data limitations (infrequent PK sampling, usually one dose tested in confirmatory trial) further constitutes challenges to a robust evaluation of E–R. Systemic exposure is consistently used in oncology as the driver of the E–R of OS and PFS while tumor penetration is often non-uniformed across solid tumors. Furthermore, some drugs, such as antibody–drug conjugates (ADC) or combination therapies, require understanding on which exposure level is assumed causally related (e.g., warhead or total antibody exposure) and would require more complex analyses to disentangle the contribution of components. Combination therapy is another challenge from a modeling standpoint since data are generally limited. Monotherapy arms are seldomly available to fully characterize each single agent E–R relationship, and assumptions on the additivity or synergistic nature of the combinatory agents are left without data to infer their validity. From a regulatory standpoint, simpler E–R approaches are customary to inform label claims justifying the dose as long as several conditions are fulfilled . First, clinical data should demonstrate meaningful benefit-risk for patients for the registrable endpoint and a statistically compelling primary analysis of trial-level data at the given dose, with minimal dose reduction, delays, or omission after treatment initiation. Second, the regulatory strategy implies pre-specifying the model-based analysis plan to reinforce its confirmatory nature. If a 2-stage analysis is selected, individual post-hoc estimates from the popPK model will be incorporated as part of the analysis dataset and use to calculate the metrics of exposure used in the TTE analysis. Alternatively, a simultaneous modeling of PK and response could be carried out with longer running times, more intensive computational power requirements, and in many cases not a clear benefit in final model results unless the response is affecting the drug disposition and the proposed model is accounting for it. Ideally, data from Phase 2 trials with more than one dose level would be ideal to start model development and be prepared to streamline the critical path activities for filing regulatory dossiers. The reality is that due to the life-threating condition of this therapeutic area and the possibility of expedited pathways for approval, phase 2 trials or expansions of Phase 1 trials are often the basis for initial approvals. Therefore, there is usually no possibility of externally validating these analyses with independent (test) data sets. Thus, the analyst is generally not expected to perform external model validation. Fortunately, more complex models relating OS or PFS with systemic exposure levels, tumor growth dynamics and accounting for dropout, and dose reduction/delays or interruptions are developed once the regulatory submission timeline pressure unwinds . These multivariate tools are far more valuable in their demonstrated track records of impacting the drug development strategy and post-marketing study designs. Leveraging the ability to integrate data from multiple studies and extrapolation/interpolation intrinsic properties , these models are used to bridge subcutaneous vs. intravenous dosing, convert flat vs. weight-based or BSA-based , adults vs. pediatrics, ethnicity considerations, extend the dose interval , redefine therapeutic window for earlier line of therapies, inform patients treatment strategy , integrate historical data and other advanced analytic framework . Tumor growth dynamics RECIST is the current standard for determining how well a patient’s tumor responds to treatment using assessments of growth/shrinkage captured in on-study X-rays, computerized tomography (CT), or magnetic resonance imaging (MRI) scans. RECIST is broadly accepted by oncology practitioners and regulatory bodies, and nearly all clinical trial treatment assessments for solid malignancies apply this framework. Still, standard RECIST methodology and its criteria for declaring treatment ‘response’ versus ‘non-response’ based on certain % of tumor shrinkage in the original reported tumor lesions, is often critiqued as inadequately representing overall disease burden , e.g. at times penalizing a deeper response with shorter time to a progression from nadir . At the center of all RECIST-based assessments, including newer versions such as irRECIST , is a practice of categorization of data-rich longitudinal tumor size information into response strata of Progressive Disease (PD), Stable Disease (SD), Partial Response (PR), and Complete Response (CR). These categories are often further dichotomized into binary assignments of responders (PR + CR) versus non-responders (SD + PD) subgroups summarized at the population level as an ORR %. Hematologic malignancy studies use similar categorical response criteria to classify continuous assessments of disease burden, such as percent blasts in Acute Myeloid Leukemia (AML) , BCRABL/BCR ratio in Chronic Myeloid Leukemia (CML) , and M-protein in multiple myeloma (MM) . In all cases, this categorization leads to loss of statistical power and is insensitive to both time dependencies and depth of response (or non-response) information captured in underlying time-course data. Indeed, if, as in MM, a discretized response spectrum has over six categories including “Very Good Partial Response (VGPR),” it may be a sign that the limits of discretization are being over-stretched to describe an underlying continuum! For this reason, longitudinal tumor burden modeling has become an increasingly applied tool for describing efficacy outcomes in clinical trials and relating tumor dynamics to predictive factors, including treatment dose/exposure . Longitudinal modeling allows for a better understanding of the entirety of a patient’s tumor burden growth/shrinkage time-course to assess the possible impact of dose or schedule selection on disease response. Such analyses are attractive also because they permit derivation of simple secondary parameters to describe features of the profile (e.g. time to re-growth/nadir, depth of nadir, etc.) through interpolation—and sometimes extrapolation—of the observable data. Secondary parameters may be more intuitively linked with survival outcomes in TTE analyses, and thus play an important role in communicating modeling results with a clinical audience. Previously described tumor burden models have been employed in a wide array of drug development applications, and a significant number of these models successfully applied in late-stage development are simple, empirical models with a minimal combination of linear or exponential primary parameters describing growth (e.g., Kg) or shrinkage (e.g., Kd) of target tumor(s). Details on many of these kinetic tumor burden models have been previously published including several excellent review articles . Central to the notion of longitudinal tumor burden modeling is the incorporation of multiple time-point observations per patient to describe an overall disease trajectory. The type of data incorporated will inherently impact model fidelity/interpretability, and hence, the utility of its application. In oncology, where ethical/logistical considerations dictate the availability of tumor assessments before treatment initiation and after discontinuation, the influence of underlying data structure on tumor model inferences deserves particularly close attention . Often, when a patient’s disease progresses due to tumor burden growth, the patient will discontinue study medication and contribute little or no more data beyond the treatment discontinuation date. Conversely, patients with responding disease tend to remain on study longer, thus contributing more and longer duration of scan data. Those patients with responsive disease, therefore, tend to receive more cumulative therapy and are more likely to experience safety-related dose modifications, which are common in oncology clinical trials due to the long-term systemic toxicities of many antineoplastic treatments. All these factors influence E–R interpretations and require careful consideration. The emphasis here, and for any astute longitudinal modeler, should always be on handling selective missingness of data, or informative censoring. In particular, this can problematically impact multiple aspects of E–R tumor burden modeling . Informative censoring often contributes to issues of tumor model parameter identifiability in cases where very little on-study data was collected in one or more patient subgroups. In particular, the time-truncation of data from patients with rapidly progressing disease may limit the availability of data to describe accurate rates of tumor (re-)growth. As such, typical models with simple growth and decay terms to describe tumor kinetic profiles tend to be more empirical than mechanistic in nature, and any biological interpretation of a given parameter with regard to cancer cell replication, treatment resistance, and cell killing is often confounded. In non-linear mixed-effects models, this could manifest as high shrinkage of the estimated between-subject variability of one or more parameters. Parameter variability and parameter estimates could very well be appropriately estimated when the levels of shrinkage are high. However, graphical exploration of covariates using empirical Bayes estimates (EBEs) will not be able to guide covariate search. To mitigate on-study parameter identifiability issues, ideally, one would 1) acquire more data throughout the course of the trial, especially scans following disease progression, or 2) incorporate measurements of pre-treatment tumor growth into the tumor burden model (Fig. ). Such assessments could be an immensely valuable decision tool as they allow each trial participants’ pre-treatment trajectory to serve as internal control—assuming care is taken in assessing the same set of lesions as the 'future' RECIST target lesions. However, incorporating when “baseline tumor” was collected is relevant information as it is in general days to weeks before the treatment starts. In indications where PFS events are more often related to tumor burden growth (as opposed to survival), this approach then allows projection of control arm PFS using a single (active) arm study . However, due to the nature of most sponsor-initiated studies and focus on the assessment of on-study treatment effects, obtaining and properly digitizing pre-study scans requires additional effort/resourcing and therefore has been rarely implemented. Another common impact of informative censoring, particularly relevant to E–R applications of tumor burden modeling, involves potentially biased estimates of the E–R relationship when an aggregate exposure summary (i.e. steady-state or cumulative) is applied. A responding individual who remains on the study will necessarily have a higher aggregate/cumulative exposure even if there is no “true” E–R relationship simply because they have had more time on-study to accumulate exposure. A simple best practice in such analyses is to apply an instantaneous, time-varying, or early milestone/baseline exposure metric as the longitudinal model input. Careful consideration of data structure is still required in E–R analyses based on a steady-state exposure/dose intensity which may also be confounded as a result of time-varying dosing due to possible dose reductions and delays. Such analyses could actually imply an inverse exposure–efficacy relationship due to the commonly encountered correlation of time on-study, favorable treatment response, and corresponding increased likelihood of safety-related dose/exposure reductions from the longer duration of exposure. Hence, a general awareness of some of the key aspects of typical oncology trial conduct and these multiple potential confounding phenomena is key in both study design and data analysis to avoid possible misinterpretation of spurious E–R relationships. Similar considerations of informative censoring should also factor into the use of on-study (post-baseline) covariates. Zhen et al. model survival data and longitudinal changes in target lesions accounting for correlation between dropout and response. In this trial with locally advanced or metastatic urothelial carcinoma (UC) patients treated with durvalumab, at the time of the data cut, 100 of 186 subjects had dropped out of the study (67 of them were due to death). As is typical of oncology trials, the risk of patient dropout was strongly influenced by treatment response. Patients with rapid tumor progression dropped out early, whereas those whose disease improved had longer follow-up times. Similarly, Hansson et al. and Schindler et al. incorporated a dropout model enabling prospective simulations of tumor response over time as dropouts were not considered at random. Tumor burden model structure selection and verification for a particular application should be driven by multiple considerations, including general goodness-of-fit, parsimony considerations, and whether the model is able to adequately describe the data for pre-stated objectives. Prior knowledge on the kinetics of disease burden for a given treatment modality may also factor into model selection. For example, cytotoxic chemotherapy treatments, which differ from immunomodulating treatments in their mechanisms of action, can be expected to show unique kinetics of disease response and progression . In general, immunotherapy efficacy has been associated with delayed but durable responses that contrast with the more rapid but transient responses seen with cytotoxic agents. Studies have also shown initial tumor “pseudoprogression” followed by delayed response in some patients treated with immunotherapy . The simpler two or three-parameter empirical tumor models derived primarily from clinical experience with chemotherapy agents often will not adequately capture this type of response pattern. Previous publications describing mixture models have been used to account for and categorize patients with hyper progressive disease as well as those with a delayed, durable response . When choosing a model, careful evaluation of extrapolation bias is recommended given that many existing tumor size models include unbounded exponential growth terms that fail to adequately extrapolate without significant prediction bias . It is therefore critical to examine model performance in the extrapolation setting and to investigate the relationship between follow-up duration and extrapolation bias. Simulation bias tests should generally be performed when evaluating base model structures by first estimating models with time-truncated data, and then assessing the general ability of the model to generate unbiased predictions of ‘future’ data. When assessing model GOF diagnostics, informative censoring and extrapolation bias may also impact interpretation. Visual inspection of trends in conditional weighted residuals versus population predictions, for example, will tend to obscure model misspecification or biased prediction of observations from participants with rapidly growing tumors, which comprise a smaller proportion of the total data set than data from participants with shrinking tumors. VPCs may therefore be a valuable tool for model verification, but again, may also be impacted by extrapolation bias since exponential growth in the post-discontinuation phase can drive anomalous prediction intervals. It is crucial that VPCs also include censoring rules or a drop-out model which approximates clinical practice implemented in the corresponding trial protocols, e.g. truncation of simulated tumor size profiles after 20% growth from nadir—a typical rule governing RECIST progression of target lesions in solid tumor indications. Advances in radiomics, which applies informatics, machine learning, and other big-data approaches to imaging data, have led to a growth in the number and types of features that may be captured for tumor burden modeling . With appropriate application, this information holds great potential to enhance the clinical relevance of inferences drawn from tumor burden modeling. In alignment with the conventions of RECIST, the majority of published tumor burden modeling studies have been conducted using the summary metric “sum of longest diameters” (SLD) from radiologist-selected target lesions. However, several reports have indicated that volumetric data may be more informative . Hierarchical modeling of individual lesion dynamics with between-tumor variability may yield even deeper insights into disease heterogeneity, as the homogenizing effect of combining lesion information from different anatomic sites for SLD inherently reduces the information available to the modeler. A key shortcoming of nearly all of the aforementioned tumor burden models, including individual lesion models, is that they rely on information from only pre-specified target lesions—which may or may not be adequately indicative of an overall disease burden. Per RECIST guidelines, individual target lesions are chosen as representative of a patient’s tumor burden for monitoring a relative treatment effect, but a more precise understanding of the tumor size to survival relationships may be established by accounting for the entire tumor/metabolic disease burden . Irrespective of whether unidimensional, bidimensional, or volumetric radiographic assessments are employed, the clinical appropriateness of any given tumor size descriptor should be re-evaluated in different treatment settings and tumor types. While construction of quantitative models linking longitudinal tumor burden with instantaneous survival risk is a relatively recent endeavor, it is already clear that these relationships may vary by nature of different cancer types, anatomical locations of the lesions, and potentially by treatment modality. Hence, the choice of appropriate tumor burden descriptors is likely to be case-dependent and may involve either one or more derived primary or secondary tumor parameters. Special considerations in modeling hematologic malignancies With the exception of lesion-level modeling, all recommendations in the above sections can be applied to hematologic malignancies, where total target tumor size is replaced by the appropriate continuous tumor burden metric for that particular malignancy: M-protein levels in secretory multiple myeloma patients, percent blasts in AML, and BCRABL/ABL ratio in CML. In the latter two, care must be taken to correctly transform the raw disease burden to a bounded assessment value based on the nature of the measurement. For example, in CML, assuming that mRNA levels are proportional to the number of genes in a cell, the BCRABL/ABL ratio can be represented as the ratio of a number of malignant cells to the weighted sum of normal cells (which have two copies of ABL) and malignant cells (which have one copy of ABL) . Modeling lymphoma data may present additional complexities as response assessments are based on both lesion size (sum of products of diameters, SPD) and metabolic activity (FDG-PET avidity) . If raw scans are available, assessment of metabolic tumor volume (MTV) , which is the total number of FDG-PET avid voxels in the scanned region of the patient’s body, are preferable to the dichotomized criteria or sum of products of diameters alone, which does not consider whether the nodes in question are actually metabolically inactive (dead). An additional feature of many hematologic malignancies is the concept of minimal residual disease (MRD) , which typically refers to technology with higher sensitivity to low disease burden than the conventional metrics. For example, 6-color flow cytometry in multiple myeloma can detect down to 0.01% levels of myeloma-transformed plasma cells in the bone marrow, as opposed to M-protein levels in peripheral blood which reach the lower limit of detection of 0.1 g/dL at underlying disease burdens ranging from 0.001 to 1%. Six-color MRD, which in the case of MM has been shown to predict incremental survival benefit with every tenfold decrease in MRD , are nonetheless dichotomized into MRD positive or negative categories, despite that these definitions may change yearly as more sensitive assays are developed . For this reason and many aforementioned benefits above, we recommend fitting the tumor burden dynamic model to both the conventional continuous metric (e.g. M-protein levels in g/dL) and the continuous MRD metric (e.g. number of cancer cells per ml of sample) simultaneously, which provides greater identifiability particularly when M-protein is below the limit of quantitation (BLQ), which is often the only time MRD is assessed . In analyzing survival data, two functions that are dependent on time are of particular interest: the survival and the hazard function. The survival function S(t) is the probability of surviving at least to time t. The hazard function h(t) is the instantaneous conditional probability of dying at time t having survived to that time. Survival curves can be estimated nonparametrically (i.e., Kaplan–Meier (KM) curves), semi-parametrically (i.e., Cox proportional hazard model), or parametrically (i.e., a lifetime model). The KM method estimates survival curves without the assumption of an underlying probability distribution in the presence of right-tailed censoring. Although easy to compute, it is hard to include covariate analysis, other than by grouping, in the models. Statistical significance between two survival curves can be made using a log-rank test, which tests the null hypothesis that there is no difference between the population survival curves (i.e., the probability of an event occurring at any time point is the same for the populations under comparison) against the alternative hypothesis that they are not the same. Cox's proportional hazards models are semi-parametric models making fewer assumptions than typical parametric methods. Cox models quantify how specific factors (covariates) influence the rate of a particular event happening at a given point in time. This rate is commonly referred to as the hazard rate. The hazard function can be written as a multiple linear regression of the logarithm of the hazard with the baseline hazard being the intercept term that varies with time: [12pt]{minimal} $$h( t ) = h_{0} ( t ) e^{{( {b_{1} x_{1} + b_{2} x_{2} + + b_{n} x_{n} } )}}$$ h t = h 0 t · e b 1 · x 1 + b 2 · x 2 + … + b n · x n where t represents the survival time, h(t) is the hazard function determined by a set of covariates where the coefficients (b 1 , b 2 …b n ) quantify the effect of covariates on the hazard, h 0 is the baseline hazard. The quantities exp(b i ) are called hazard ratios (HR). An HR > 1 increases the hazard indicating a positive effect of the covariate with event probability and therefore negatively associated with the length of survival. An HR < 1 reduces the hazard and increases the probability of survival. The main assumption of the Cox model is that the hazard is proportional: if an individual has a risk of death at some initial point that is twice as high as that of another individual, then at later times the risk of death remains twice as high, that is, the hazard ratio comparing two groups is constant over time. However, in oncology, the assumption of proportional hazard implicit into Cox’s models may not represent the ideal choice where the factor changing the hazard may vary with time. Unlike the Cox regression model which does not specify the distribution function of the hazard, there are several parametric models such as Weibull, Gompertz, exponential, log-normal, and log-logistic models where the hazard function has to be specified. These parametric models allow us to estimate the effects of covariates on the hazard function, including variables that may change over time like drug concentration/dose, age, tumor growth dynamics, or time since surgical intervention. It is important to highlight the inherent selection bias and immortal time bias in survival metrics in oncology . The TTE model approach with the estimation of a single survival function has its limitations. One alternative approach is the landmark method, which determines each patient’s response at some fixed time point, with survival estimates calculated from that time point and associated statistical tests being conditional on patients’ landmark responses excluding patients that die before the selected landmark timepoint . Another approach to address the challenges of correctly describing the hazard function over time is the use of multistate models. Beyer et al. developed a multistate model with transition hazards estimated using semiparametric models . Krishnan et al. proposed transition hazards using a parametric approach and mixture models . Five states were considered according to Response Evaluation Criteria in Solid Tumors (RECIST) and using predictions derived from a longitudinal tumor growth inhibition model. The model did not allow to move back from one state to another when a response level has been achieved and maybe changed over time. Transition rates between states were estimated and defining the different hazard distributions (see Fig. ). In oncology, analysis of the association between drug exposure levels and late-phase clinical outcomes, such as OS and PFS (or event-free survival EFS) are often limited in scope for regulatory submissions. In light of the paucity of data and time constraints, industry practice and regulatory expectations to rationalize the dose rely on performing logistic or Cox proportional hazard regression analysis of the registration-intent trial efficacy data. Even simpler exploratory approaches are sometimes considered to support the dose rationale, such as conducting a KM exploratory analysis of OS and/or PFS stratified by quartiles or tertiles of exposure levels . The general strategy is to demonstrate the absence of a relationship between exposure and the registrable endpoint within the range of exposure tested and infer from this that the dose is optimal since no substantial gain in efficacy can be achieved by increasing exposure levels further. Whether such analyses can be qualified as best practice is contentious as pragmatism is the main driver to rationalize the methodological framework used. Typically, the establishment of a recommended phase II dose is endorsed based on combined analyses of emerging safety signals and surrogate efficacy endpoints with proven independent prognostic value for OS or PFS (e.g., tumor burden or objective response obtained from earlier trial data). Confirmatory trials rarely study more than one dose, and inherently suffer from underpowered statistical consideration for E–R characterization of OS and PFS. Furthermore, the ever-increasing pace for regulatory filing driven by commercial incentives and rapid access of novel treatment to patients in unmet needs further restrains the possibility to develop a data-driven E–R model of OS or PFS for regulatory submission given the time-consuming nature of these analyses. The downside of simple analyses lies in identifying spurious relationships due to imbalance of known prognostic factors of OS at baseline in strata of exposure . For example, ramucirumab E–R showed a positive association with OS in gastric cancer patients following a Cox regression with C min,ss or C avg . However, the E–R model did not evaluate covariates such as C-reactive protein levels and tumor burden that are known prognostic factors of OS and could also impact PK as demonstrated in the case of tremelimumab , trastuzumab emtansine , and to a lesser extent with trastuzumab in the same indication . The recommendation in such a case would be to first interpret with great caution any apparent trend of E–R emerging from simple analyses of trials evaluating a single-dose regimen. To explore further the risk of potential confounding in the event of a positive E–R finding, more sophisticated methods such as the parametric hazard model and multistate models should be attempted to quantify E–R in a multivariate framework. A full model or a selection of covariates of clinical relevance for the endpoint of interest and for the PK metric chosen should then be applied to quantify the relative contribution of covariates on the underlying E–R. As the last step, exposure should be removed from the model in case its causative link with OS or PFS assumed by the model is no longer supported by the data and can be fully explained by prognostic or predictive covariates remaining in the model. The best example of this good practice would be the already mentioned nivolumab E–R analysis in which baseline CL was included in the model . Technical considerations, such as the choice of the PK metrics (C max , AUC) or when this metric is measured (first cycle vs steady-state), and data limitations (infrequent PK sampling, usually one dose tested in confirmatory trial) further constitutes challenges to a robust evaluation of E–R. Systemic exposure is consistently used in oncology as the driver of the E–R of OS and PFS while tumor penetration is often non-uniformed across solid tumors. Furthermore, some drugs, such as antibody–drug conjugates (ADC) or combination therapies, require understanding on which exposure level is assumed causally related (e.g., warhead or total antibody exposure) and would require more complex analyses to disentangle the contribution of components. Combination therapy is another challenge from a modeling standpoint since data are generally limited. Monotherapy arms are seldomly available to fully characterize each single agent E–R relationship, and assumptions on the additivity or synergistic nature of the combinatory agents are left without data to infer their validity. From a regulatory standpoint, simpler E–R approaches are customary to inform label claims justifying the dose as long as several conditions are fulfilled . First, clinical data should demonstrate meaningful benefit-risk for patients for the registrable endpoint and a statistically compelling primary analysis of trial-level data at the given dose, with minimal dose reduction, delays, or omission after treatment initiation. Second, the regulatory strategy implies pre-specifying the model-based analysis plan to reinforce its confirmatory nature. If a 2-stage analysis is selected, individual post-hoc estimates from the popPK model will be incorporated as part of the analysis dataset and use to calculate the metrics of exposure used in the TTE analysis. Alternatively, a simultaneous modeling of PK and response could be carried out with longer running times, more intensive computational power requirements, and in many cases not a clear benefit in final model results unless the response is affecting the drug disposition and the proposed model is accounting for it. Ideally, data from Phase 2 trials with more than one dose level would be ideal to start model development and be prepared to streamline the critical path activities for filing regulatory dossiers. The reality is that due to the life-threating condition of this therapeutic area and the possibility of expedited pathways for approval, phase 2 trials or expansions of Phase 1 trials are often the basis for initial approvals. Therefore, there is usually no possibility of externally validating these analyses with independent (test) data sets. Thus, the analyst is generally not expected to perform external model validation. Fortunately, more complex models relating OS or PFS with systemic exposure levels, tumor growth dynamics and accounting for dropout, and dose reduction/delays or interruptions are developed once the regulatory submission timeline pressure unwinds . These multivariate tools are far more valuable in their demonstrated track records of impacting the drug development strategy and post-marketing study designs. Leveraging the ability to integrate data from multiple studies and extrapolation/interpolation intrinsic properties , these models are used to bridge subcutaneous vs. intravenous dosing, convert flat vs. weight-based or BSA-based , adults vs. pediatrics, ethnicity considerations, extend the dose interval , redefine therapeutic window for earlier line of therapies, inform patients treatment strategy , integrate historical data and other advanced analytic framework . RECIST is the current standard for determining how well a patient’s tumor responds to treatment using assessments of growth/shrinkage captured in on-study X-rays, computerized tomography (CT), or magnetic resonance imaging (MRI) scans. RECIST is broadly accepted by oncology practitioners and regulatory bodies, and nearly all clinical trial treatment assessments for solid malignancies apply this framework. Still, standard RECIST methodology and its criteria for declaring treatment ‘response’ versus ‘non-response’ based on certain % of tumor shrinkage in the original reported tumor lesions, is often critiqued as inadequately representing overall disease burden , e.g. at times penalizing a deeper response with shorter time to a progression from nadir . At the center of all RECIST-based assessments, including newer versions such as irRECIST , is a practice of categorization of data-rich longitudinal tumor size information into response strata of Progressive Disease (PD), Stable Disease (SD), Partial Response (PR), and Complete Response (CR). These categories are often further dichotomized into binary assignments of responders (PR + CR) versus non-responders (SD + PD) subgroups summarized at the population level as an ORR %. Hematologic malignancy studies use similar categorical response criteria to classify continuous assessments of disease burden, such as percent blasts in Acute Myeloid Leukemia (AML) , BCRABL/BCR ratio in Chronic Myeloid Leukemia (CML) , and M-protein in multiple myeloma (MM) . In all cases, this categorization leads to loss of statistical power and is insensitive to both time dependencies and depth of response (or non-response) information captured in underlying time-course data. Indeed, if, as in MM, a discretized response spectrum has over six categories including “Very Good Partial Response (VGPR),” it may be a sign that the limits of discretization are being over-stretched to describe an underlying continuum! For this reason, longitudinal tumor burden modeling has become an increasingly applied tool for describing efficacy outcomes in clinical trials and relating tumor dynamics to predictive factors, including treatment dose/exposure . Longitudinal modeling allows for a better understanding of the entirety of a patient’s tumor burden growth/shrinkage time-course to assess the possible impact of dose or schedule selection on disease response. Such analyses are attractive also because they permit derivation of simple secondary parameters to describe features of the profile (e.g. time to re-growth/nadir, depth of nadir, etc.) through interpolation—and sometimes extrapolation—of the observable data. Secondary parameters may be more intuitively linked with survival outcomes in TTE analyses, and thus play an important role in communicating modeling results with a clinical audience. Previously described tumor burden models have been employed in a wide array of drug development applications, and a significant number of these models successfully applied in late-stage development are simple, empirical models with a minimal combination of linear or exponential primary parameters describing growth (e.g., Kg) or shrinkage (e.g., Kd) of target tumor(s). Details on many of these kinetic tumor burden models have been previously published including several excellent review articles . Central to the notion of longitudinal tumor burden modeling is the incorporation of multiple time-point observations per patient to describe an overall disease trajectory. The type of data incorporated will inherently impact model fidelity/interpretability, and hence, the utility of its application. In oncology, where ethical/logistical considerations dictate the availability of tumor assessments before treatment initiation and after discontinuation, the influence of underlying data structure on tumor model inferences deserves particularly close attention . Often, when a patient’s disease progresses due to tumor burden growth, the patient will discontinue study medication and contribute little or no more data beyond the treatment discontinuation date. Conversely, patients with responding disease tend to remain on study longer, thus contributing more and longer duration of scan data. Those patients with responsive disease, therefore, tend to receive more cumulative therapy and are more likely to experience safety-related dose modifications, which are common in oncology clinical trials due to the long-term systemic toxicities of many antineoplastic treatments. All these factors influence E–R interpretations and require careful consideration. The emphasis here, and for any astute longitudinal modeler, should always be on handling selective missingness of data, or informative censoring. In particular, this can problematically impact multiple aspects of E–R tumor burden modeling . Informative censoring often contributes to issues of tumor model parameter identifiability in cases where very little on-study data was collected in one or more patient subgroups. In particular, the time-truncation of data from patients with rapidly progressing disease may limit the availability of data to describe accurate rates of tumor (re-)growth. As such, typical models with simple growth and decay terms to describe tumor kinetic profiles tend to be more empirical than mechanistic in nature, and any biological interpretation of a given parameter with regard to cancer cell replication, treatment resistance, and cell killing is often confounded. In non-linear mixed-effects models, this could manifest as high shrinkage of the estimated between-subject variability of one or more parameters. Parameter variability and parameter estimates could very well be appropriately estimated when the levels of shrinkage are high. However, graphical exploration of covariates using empirical Bayes estimates (EBEs) will not be able to guide covariate search. To mitigate on-study parameter identifiability issues, ideally, one would 1) acquire more data throughout the course of the trial, especially scans following disease progression, or 2) incorporate measurements of pre-treatment tumor growth into the tumor burden model (Fig. ). Such assessments could be an immensely valuable decision tool as they allow each trial participants’ pre-treatment trajectory to serve as internal control—assuming care is taken in assessing the same set of lesions as the 'future' RECIST target lesions. However, incorporating when “baseline tumor” was collected is relevant information as it is in general days to weeks before the treatment starts. In indications where PFS events are more often related to tumor burden growth (as opposed to survival), this approach then allows projection of control arm PFS using a single (active) arm study . However, due to the nature of most sponsor-initiated studies and focus on the assessment of on-study treatment effects, obtaining and properly digitizing pre-study scans requires additional effort/resourcing and therefore has been rarely implemented. Another common impact of informative censoring, particularly relevant to E–R applications of tumor burden modeling, involves potentially biased estimates of the E–R relationship when an aggregate exposure summary (i.e. steady-state or cumulative) is applied. A responding individual who remains on the study will necessarily have a higher aggregate/cumulative exposure even if there is no “true” E–R relationship simply because they have had more time on-study to accumulate exposure. A simple best practice in such analyses is to apply an instantaneous, time-varying, or early milestone/baseline exposure metric as the longitudinal model input. Careful consideration of data structure is still required in E–R analyses based on a steady-state exposure/dose intensity which may also be confounded as a result of time-varying dosing due to possible dose reductions and delays. Such analyses could actually imply an inverse exposure–efficacy relationship due to the commonly encountered correlation of time on-study, favorable treatment response, and corresponding increased likelihood of safety-related dose/exposure reductions from the longer duration of exposure. Hence, a general awareness of some of the key aspects of typical oncology trial conduct and these multiple potential confounding phenomena is key in both study design and data analysis to avoid possible misinterpretation of spurious E–R relationships. Similar considerations of informative censoring should also factor into the use of on-study (post-baseline) covariates. Zhen et al. model survival data and longitudinal changes in target lesions accounting for correlation between dropout and response. In this trial with locally advanced or metastatic urothelial carcinoma (UC) patients treated with durvalumab, at the time of the data cut, 100 of 186 subjects had dropped out of the study (67 of them were due to death). As is typical of oncology trials, the risk of patient dropout was strongly influenced by treatment response. Patients with rapid tumor progression dropped out early, whereas those whose disease improved had longer follow-up times. Similarly, Hansson et al. and Schindler et al. incorporated a dropout model enabling prospective simulations of tumor response over time as dropouts were not considered at random. Tumor burden model structure selection and verification for a particular application should be driven by multiple considerations, including general goodness-of-fit, parsimony considerations, and whether the model is able to adequately describe the data for pre-stated objectives. Prior knowledge on the kinetics of disease burden for a given treatment modality may also factor into model selection. For example, cytotoxic chemotherapy treatments, which differ from immunomodulating treatments in their mechanisms of action, can be expected to show unique kinetics of disease response and progression . In general, immunotherapy efficacy has been associated with delayed but durable responses that contrast with the more rapid but transient responses seen with cytotoxic agents. Studies have also shown initial tumor “pseudoprogression” followed by delayed response in some patients treated with immunotherapy . The simpler two or three-parameter empirical tumor models derived primarily from clinical experience with chemotherapy agents often will not adequately capture this type of response pattern. Previous publications describing mixture models have been used to account for and categorize patients with hyper progressive disease as well as those with a delayed, durable response . When choosing a model, careful evaluation of extrapolation bias is recommended given that many existing tumor size models include unbounded exponential growth terms that fail to adequately extrapolate without significant prediction bias . It is therefore critical to examine model performance in the extrapolation setting and to investigate the relationship between follow-up duration and extrapolation bias. Simulation bias tests should generally be performed when evaluating base model structures by first estimating models with time-truncated data, and then assessing the general ability of the model to generate unbiased predictions of ‘future’ data. When assessing model GOF diagnostics, informative censoring and extrapolation bias may also impact interpretation. Visual inspection of trends in conditional weighted residuals versus population predictions, for example, will tend to obscure model misspecification or biased prediction of observations from participants with rapidly growing tumors, which comprise a smaller proportion of the total data set than data from participants with shrinking tumors. VPCs may therefore be a valuable tool for model verification, but again, may also be impacted by extrapolation bias since exponential growth in the post-discontinuation phase can drive anomalous prediction intervals. It is crucial that VPCs also include censoring rules or a drop-out model which approximates clinical practice implemented in the corresponding trial protocols, e.g. truncation of simulated tumor size profiles after 20% growth from nadir—a typical rule governing RECIST progression of target lesions in solid tumor indications. Advances in radiomics, which applies informatics, machine learning, and other big-data approaches to imaging data, have led to a growth in the number and types of features that may be captured for tumor burden modeling . With appropriate application, this information holds great potential to enhance the clinical relevance of inferences drawn from tumor burden modeling. In alignment with the conventions of RECIST, the majority of published tumor burden modeling studies have been conducted using the summary metric “sum of longest diameters” (SLD) from radiologist-selected target lesions. However, several reports have indicated that volumetric data may be more informative . Hierarchical modeling of individual lesion dynamics with between-tumor variability may yield even deeper insights into disease heterogeneity, as the homogenizing effect of combining lesion information from different anatomic sites for SLD inherently reduces the information available to the modeler. A key shortcoming of nearly all of the aforementioned tumor burden models, including individual lesion models, is that they rely on information from only pre-specified target lesions—which may or may not be adequately indicative of an overall disease burden. Per RECIST guidelines, individual target lesions are chosen as representative of a patient’s tumor burden for monitoring a relative treatment effect, but a more precise understanding of the tumor size to survival relationships may be established by accounting for the entire tumor/metabolic disease burden . Irrespective of whether unidimensional, bidimensional, or volumetric radiographic assessments are employed, the clinical appropriateness of any given tumor size descriptor should be re-evaluated in different treatment settings and tumor types. While construction of quantitative models linking longitudinal tumor burden with instantaneous survival risk is a relatively recent endeavor, it is already clear that these relationships may vary by nature of different cancer types, anatomical locations of the lesions, and potentially by treatment modality. Hence, the choice of appropriate tumor burden descriptors is likely to be case-dependent and may involve either one or more derived primary or secondary tumor parameters. With the exception of lesion-level modeling, all recommendations in the above sections can be applied to hematologic malignancies, where total target tumor size is replaced by the appropriate continuous tumor burden metric for that particular malignancy: M-protein levels in secretory multiple myeloma patients, percent blasts in AML, and BCRABL/ABL ratio in CML. In the latter two, care must be taken to correctly transform the raw disease burden to a bounded assessment value based on the nature of the measurement. For example, in CML, assuming that mRNA levels are proportional to the number of genes in a cell, the BCRABL/ABL ratio can be represented as the ratio of a number of malignant cells to the weighted sum of normal cells (which have two copies of ABL) and malignant cells (which have one copy of ABL) . Modeling lymphoma data may present additional complexities as response assessments are based on both lesion size (sum of products of diameters, SPD) and metabolic activity (FDG-PET avidity) . If raw scans are available, assessment of metabolic tumor volume (MTV) , which is the total number of FDG-PET avid voxels in the scanned region of the patient’s body, are preferable to the dichotomized criteria or sum of products of diameters alone, which does not consider whether the nodes in question are actually metabolically inactive (dead). An additional feature of many hematologic malignancies is the concept of minimal residual disease (MRD) , which typically refers to technology with higher sensitivity to low disease burden than the conventional metrics. For example, 6-color flow cytometry in multiple myeloma can detect down to 0.01% levels of myeloma-transformed plasma cells in the bone marrow, as opposed to M-protein levels in peripheral blood which reach the lower limit of detection of 0.1 g/dL at underlying disease burdens ranging from 0.001 to 1%. Six-color MRD, which in the case of MM has been shown to predict incremental survival benefit with every tenfold decrease in MRD , are nonetheless dichotomized into MRD positive or negative categories, despite that these definitions may change yearly as more sensitive assays are developed . For this reason and many aforementioned benefits above, we recommend fitting the tumor burden dynamic model to both the conventional continuous metric (e.g. M-protein levels in g/dL) and the continuous MRD metric (e.g. number of cancer cells per ml of sample) simultaneously, which provides greater identifiability particularly when M-protein is below the limit of quantitation (BLQ), which is often the only time MRD is assessed . Disease progression modeling is utilized to describe the time course of disease status and track disease severity over time. These models usually incorporate biomarker data and clinical outcomes to characterize natural disease progression . In prostate cancer (PCa), prostate-specific antigen (PSA) has been recognized as a biomarker for diagnosis, prognosis, and monitoring of disease activity . PCa is usually characterized as either low-risk non-aggressive (indolent) or high-risk aggressive tumors. While indolent PCa is benign prostatic hyperplasia (BPH) in general, aggressive PCa may lead to cancer-specific morbidity . Identification of the most aggressive PCa cases among all patients diagnosed with PCa could help in the selection of the patients who might benefit from radical therapy. de Charry et al. developed a semi-mechanistic model of PSA longitudinal growth to help differentiate aggressive and indolent PCa at diagnosis . The individual preoperative PSA data from patients with PCa and those with benign prostatic hyperplasia were analyzed using a population kinetic approach and a semi-mechanistic nonlinear mixed-effects model . This analysis demonstrated a greater PSA increase rate by cancer cells than by non-cancer cells, while PSA production rate was greater by benign tissue than by malignant tissue. A significant relationship between the PSA production rate by cancer cells and the probability of D’Amico high-risk group was also identified with logistic regression. Moreover, multivariate tests demonstrated that the PSA production rate by cancer cells, Gleason score, and positive surgical margin status were all significant independent predictive factors regarding relapse-free survival (RFS). This semi-mechanistic model provided a possible means to determine whether a patient is likely to have aggressive PCa before surgery. In addition to PSA, the count of circulating tumor cells (CTCs) has also emerged as a promising surrogate marker in patients with metastatic castration-resistant prostate cancer (mCRPC) . In 2004, the FDA approved the use of the CellSearch® system for detecting CTCs in cancer patients. It is to date the only approved laboratory test for CTCs and is being used to enumerate the number of CTCs of epithelial origin in a 7.5 mL blood sample . Wilbaux et al. developed a semi-mechanistic model to quantify the dynamic relationships between the kinetics of CTC counts and PSA concentrations during treatment in patients with mCRPC . Their joint model incorporated drug effect kinetics for chemotherapy and hormonal therapy through two different K-PD compartments as no drug concentrations data were available. The treatment effects on both PSA and CTCs were assumed to be mediated through a common latent variable that was interpreted as tumor burden. PSA kinetics were described by an indirect response model, while the CTC kinetics in the total body blood was modeled by a cell lifespan model assuming that the rate of cell loss was equal to the rate of cell production delayed by the lifespan. The dynamic change of CTC counts was considered as a random sampling from a negative binomial distribution. By simulating the kinetics of PSA, CTC counts, and the tumor burden, CTC counts turned out to be more sensitive to the variation of the tumor burden. This model was the first to quantify the dynamic links between the kinetics of PSA and CTC counts during treatment in patients with mCRPC. Although limitations exist, this model demonstrated the potential of using CTC counts as a predictor of treatment response or disease progression in patients with mCRPC. In recent years, biological products, such as monoclonal antibodies and other therapeutic proteins, have been widely used for the treatment of various cancers. Occasionally, the administered biological product may provoke an immune response (known as immunogenicity) in some patients receiving repeated dosing which may lead to excessive cytokine release and/or formation of anti-drug antibodies (ADAs) . Some unwanted adverse events, including anaphylaxis, cytokine release syndrome, infusion reaction, and other non-acute reactions, may occur as the result of the immune response. The cross-reaction between ADAs and their endogenous counterparts may interfere with certain physiological processes, causing additional safety concerns. Furthermore, the immunological reaction may compromise the drug effect. For example, the presence of ADAs usually affects the drug clearance and decreases drug exposure. Additionally, neutralizing ADAs may interfere with the interaction between the therapeutic protein and its target. Therefore, using modeling approaches to quantitatively characterize the clinical consequences of product immunogenicity may provide insights into the risk and benefit profiles in patient subgroups (e.g., ADA positive and negative) and guide the optimal use of the product . Appropriate quantification of the clinical impact of immunogenicity relies on the accurate acquisition of the data. ADAs titers in serum/plasma have been considered as the major biomarker to track patients receiving treatment with a biological product. Time to the first appearance of ADA (i.e., onset), sustained duration, and level of ADA (i.e., duration) may vary among individuals following chronic treatment. Unfortunately, the ADA level cannot be monitored on a continuous basis. The sampling schedule, therefore, becomes a critical component to characterize the potential ADA changes and to link the ADA changes to clinical consequences. A sensitive, specific, and selective bioassay for ADA is another key factor to ensure data quality. In general, ADA is detected by immunoassay using an ADA-antibody, which may interfere with the coexisting therapeutic protein and endogenous substances in biological matrices. An assay’s drug tolerance, which measures the assay sensitivity in the presence of the therapeutic proteins, as compared to typical concentration levels in patients is important to understand the reliability of the detected ADA . The Global Bioanalysis Consortium (GBC) set up an international team to explore the impact of immunogenicity on PK assessments. The result of the work was a white paper where they presented strategies to assess if changes in drug concentration are due to ADA-mediated changes in clearance or instead a consequence of ADA interference with the bioanalytical assay . There is a multitude of factors that could influence the immunogenicity of biologics. These factors could be classified into disease-, patient-, or product-related factors . Immunogenicity as a consequence of a biologic therapeutic administration may result in multiple polyclonal antibodies against multiple epitopes circulating in serum. Each ADA species has its own specificity and binding affinity. As a result, these different treatment-emergent ADA responses may have different effects on the PK/PD of the drug therapeutic: a neutralizing effect on activity by interfering with the drug’s ability to bind to its pharmacologic target, a non-neutralizing effect on activity paired with an effect on the PK, or a non-neutralizing effect on activity with an enhanced elimination of the drug therapeutic . The findings on clinical consequences of immunogenicity for a biological product may be misleading if an inappropriate modeling approach is used. Randomized, well-controlled clinical trials are routinely used to characterize the efficacy and safety profiles of a biological product before it gains marketing authorization. An analysis, for instance, may be conducted to compare overall survival in patients with (i.e., ADA positive group) or without ADA (i.e., ADA negative group) detected anytime during the treatment. However, this analysis may lead to misleading outcomes. The time to the first occurrence of ADA and duration for detectable ADA sustained in plasma appears to be random among individuals during the treatment phase. It should be noted that some patients in the ADA-positive group are the patients with the late-occurring ADA formation, who must already survive long enough with continuous treatment. Thus, the true ADA effect on overall survival may be attenuated (i.e., biased), if the overall survival is directly compared between the two groups. Rather, a landmark analysis comparing patients with or without early (e.g., within 4 weeks after the treatment is initiated) detectable ADA may provide a relatively better angle to characterize the potential impact of ADA formation on drug effect, if adequate data are available. The analysis performed for atezolizumab provides a good example. As demonstrated in OAK study in locally advanced and metastatic non-small cell lung cancer patients, 21% of patients were tested positive for ADA by week 4. The analysis suggested that the presence of ADA early after the treatment initiation (i.e., 4 weeks) may significantly affect the overall survival in this patient population . However, careful consideration must be given to any post-randomization variable, including ADA formation. Kong et al. studied the potential impact of ADA formation on long term benefit in a randomized controlled trial with atezolizumab, in which ADA was not observed in the control arm . ADA status was only observable in atezolizumab-treated patients, and variables that are a consequence of the treatment either preclude observation or affect the interpretation of the clinical endpoint of interest. The authors propose a weighted approach for estimating effects based on Weight Placebo Patients (WPP) approach where the post-baseline strata are observed for every subject. The idea behind this is if ADA can be organized in categories, is the treatment benefit clinically meaningful for all categories of ADA? and is the treatment effect similar between certain or all categories of ADA?. Atezolizumab treated ADA-positive patients showed worse OS relative to ADA-negative patients. Furthermore, ADA-positive patients also showed differences in several prognostic baseline variables compared to ADA-negative patients. The authors concluded after correcting the data for the prognostic baseline variables that hazard ratios were similar for ADA-positive and ADA-negative patient populations. Cell and Gene Therapy (CGT) have now entered a new area of treating diseases. Although scientific efforts have been in progress for over a century, recent advancements in the last decade have significantly shifted the treatment paradigm with several regulatory approvals . As of February 2023, there were 27 CGT products that were approved by FDA . The proven hypothesis, that either alteration of certain cells (Cell therapy) or certain genes into cells (Gene therapy) may lead to treatment, has made CGT very effective in individualized treatment. CGT is formulated differently from traditional large batch production of a small or large molecule (antibody) drug, and in many instances, CGT is designed specifically for individual patients (for example, autologous cell therapy). Among several CGT modalities, Chimeric Antigen Receptor (CAR) T cell-based therapy has shown clinical benefits in different cancer types . In this treatment, a patient’s T-cells are modified in the laboratory and specific CARs are introduced to the surface of the T-cell in order to target very specific malignant cancer cells. For the purpose of this review, we will focus on current advancements in the clinical pharmacology of autologous CAR-T therapy only. The PK of CGT drugs is characterized using quantitative polymerase chain reaction (qPCR) assay for viral products and using flow cytometry or qPCR for autologous cell therapies. A typical small or large molecule drug after IV administration exhibits maximum drug concentration (C max ) at the end of infusion, whereas for CAR-T cells, the C max is generally achieved after a few days of IV administration due to cellular proliferation and expansion upon interaction with the target antigens in circulation (or at the site of action in tumor or bone marrow). The apparent kinetics of CAR-T may consist of four phases including distribution, expansion, contraction, and persistence , see Fig. A and B. In humans, the expansion/proliferation of CAR-T cells is found to be highly variable and primarily driven by an individual's immune system activity. Cilta-cel (Carvykti™) CAR transgene concentrations showed maximum peripheral expansion at a median of 12·7 days (range 8·7–54·6) with observed persistence lasting over > 100 days in peripheral blood . Among patients with 6 months follow-up, most had cilta-cel CAR transgene concentration below the level of quantification (< 50 CAR gene copies per μg DNA) in peripheral blood. Several factors, such as prior lines of therapy, tumor burden, and disease status, may impact an individual's immune system activity. In addition, CAR-T cell drug product characteristics, such as CD4:CD8 ratio, %CAR + T cells, the constitution of effector, and memory cells may also contribute to cellular expansion/proliferation, resulting in highly variable exposure parameters such as C max and area under the curve (AUC) with inter-individual variability as large as 165–190% . Since expansion and persistence of CAR-T cells are dependent on the cellular composition of the individual donor’s original T cells and the individual’s immune system, it is difficult to extrapolate or predict the extent of inter-individual PK exposure (cellular expansion/persistence) in humans. In addition, given the limitation of pre-clinical models in mimicking the complexity of the human immune system, allometric-based principles cannot be applied in translating CAR-T exposure from preclinical species to humans. There have been recent attempts to develop translational models , but the lack of relevant preclinical species limits such preclinical to clinical predictions. Current regulatory guidance on gene and cell therapy also states that unlike for small and large molecules, allometric-based scaling cannot be applied to predict starting dose in humans. Typical CAR-T starting dose selection has been based on prior knowledge of cell therapy and target expression. Since CAR-T exposure exhibits high inter-individual variability, depending on the range of dose levels studied, and the number of subjects at each dose level, dose dependence on PK parameters and impact on response may be difficult to identify. A population cellular kinetic model was developed using clinical data of Kymriah™ where the dose range administered ranged from 0.2 to 5.4 Mcells/kg. The authors did not observe any relationship between dose and exposure. Even though an increase in C max was found to increase the incidence and severity of key safety outcomes (cytokine release syndrome, CRS), due to lack of a clear dose–exposure relationship, such correlations could not be used to identify an optimal dosing regimen. In another study using Abecma® , the authors proposed a dose-dependent increase in expansion/persistence based on a prospectively designed Phase 1/2 study where dose escalation was conducted in cohorts at dose levels of 50, 150, 450, and 800 Mcells. Despite a dose-dependent increase in exposure, a plateau in exposure was apparent at 450–800 Mcells. Abecma® transgene levels were positively associated with objective tumor response (partial response or better), where responders achieved ~ 4.6-fold higher C max and ~ 5.6-fold higher AUC (0–28 days) in comparison to non-responders . It is unclear if high inter-individual variability with overlapping exposures and lack of a clear dose–exposure relationship influenced the approved dose range of 300–460 Mcells for Abecma®. In summary, drug exposure after CAR-T therapy has shown high inter-individual variability with overlapping exposures between dose levels. Although higher exposure may be associated with higher responses, the lack of a clear dose–exposure relationship limits the determination of optimal dose levels. For future trials, a dose-escalation study should be considered in early development with rich PK collection throughout the study to inform better characterization of the dose/exposure–response relationship. Adequate care needs to be taken in performing E–R analyses using data collected from late-phase oncology clinical trials. The majority of Phase 2 and 3 oncology clinical trials are carried out with one dose level, having minimal E–R information available from the dose-escalation part of the Phase 1 study, often in refractory patients where at the most we can get an idea of E–R for safety endpoints, but we don’t expect any efficacy as the target population is normally not the intended population for the projected filing indication. With only a single dose, the range of exposures is often limited, and may make detection of E–R relationships difficult. FDA has long been pushing for a more thoughtful approach to dosing. With the recent release of a draft guidance requesting sponsors to study a range of doses in early development, this may make E–R modeling more sensitive at detecting such relationships. Performing a comprehensive pharmacometrics analysis of a drug product at the time of NDA/BLA filing is often challenging but probably the most cost-efficient approach when considering subsequent filings in subsequent indications and special populations. We have seen repeatedly with many therapies that disease burden as characterized by multiple risk factors, such as Eastern Cooperative Oncology Group (ECOG) performance score (PS), percentage of prior surgery, and increased number of metastatic sites, may vary significantly among enrolled patients at baseline and usually changes in the course of the treatment. In late-stage cancer, patients with a high disease burden may be associated with increased clearance for some compounds. Usually, these patients tend to show more aggressive disease progression and shorter survival times. Thus, the disease burden becomes a confounder that affects both drug exposure and survival. All these factors in combination may distort the underlying causal relationship associated with the treatment effect. Inappropriate modeling approaches are more likely to happen when very little information has been collected during early-stage development or no input has been requested from the pharmacometrician, which could lead to an erroneous presentation of the underlying E–R relationship when the pivotal trial reads out. Often pharmacometricians play “catch-up” by having limited time to perform the retrospective analyses required to comply with regulatory agencies’ recommendations with very little influence on requesting adequately powered dose-finding studies during the drug development program. Usually, the modeling comes at the end, when perhaps the necessary assessments were not collected or incompletely collected, and the information available is insufficient to perform an in-depth analysis of the safety and efficacy endpoints and their correlations with exposure. The pharmacometrics community has the tools, the knowledge, and well-described models summarized in this review that can help characterize E–R in drug development and hence better understand the drug behavior and therapy management. However, tools, knowledge, and models are only half of the work, well-designed trials, and informative data collections are the other half and it is very challenging to do both parts in isolation. A collaborative interaction is required among clinicians, statisticians, pharmacologists, pharmacometricians, and decision-makers to succeed in prospectively defining a clinical development strategy tailored towards the identification of the optimal dose regimen and benefit/risk ratio of a given drug product for the intended-to-treat population of patients.
Association of Patient, Physician, and Practice-Level Factors with Uptake of Payer-Led Oncology Clinical Pathways	955bc481-c7ef-4cab-94f1-7724d928cb4d	10170335	Internal Medicine[mh]	Both cost and complexity have increased in the practice of oncology. Evidence suggests that cancer drug prescriptions account for the largest portion of spending on cancer care and the greatest variation in practice. Given this, there has been growing interest among both payers and clinicians in the use of tools that would promote quality by introducing a degree of standardization to drug prescribing and decrease costs by reducing inappropriate non–evidence-based care. One such approach is to implement clinical pathways programs, a subset of evidence-based guidelines, such as those from the National Comprehensive Cancer Network, that help clarify decisions when multiple treatment options exist, typically along 3 priorities in the following order: efficacy, safety, and cost. When efficacy and safety profiles are similar between evidence-based treatment regimens, the pathways program favors lower-priced regimens over higher-priced regimens. As pathways take cost into account, the hope is that their use will help manage drug use in a dynamic treatment landscape by nudging clinicians into selecting higher-value choices. , With this value proposition, multiple national payers have developed and introduced oncology clinical pathways programs. However, execution is critical to the success of the premise. Payers have historically found that compliance is a key challenge, with studies showing that compliance with pathways can be as low as 50% to 70%. , , For example, UnitedHealthcare piloted a clinician-led pathway program with 5 volunteer medical oncology groups, and later discovered a 3-fold variation in drug costs between groups. Further analysis revealed that compliance with the pathway was less than 50%, and this noncompliance was responsible for the variance. Understanding the factors associated with pathway compliance is therefore a key implementation question. In this retrospective cohort study, we used claims data from a large insurer and administrative data from a company that develops cancer treatment pathways to identify factors associated with compliance that incorporates a wide range of patient-, physician-, and practice-level factors. Program Overview Claims data were obtained from Elevance Health (formerly Anthem Inc), a national insurer that launched an oncology clinical pathway program, the Cancer Care Quality Program (CCQP), through its subsidiary AIM Specialty Health in 2014. An external panel of oncology experts meets on a quarterly basis and reviews, approves, and updates regimens over time to reflect current evidence. To incentivize selection of pathway-compliant regimens, Elevance Health pays practices an additional reimbursement of $350 per patient each month. Treatment requests made by oncologists for regimens not included in the pathways program are reviewed and still authorized without delay or requirement for peer-to-peer interaction if they are determined to be medically necessary pursuant to medical policies and clinical guidelines. Study Design and Data We obtained longitudinal medical and pharmacy claims and eligibility files from Elevance Health care plans across the US for information about health care utilization, medical spending, and health plan enrollment. Our data were from 3 types of health plan members: Medicare Advantage (MA), fully insured members, and members of plans through self-insured employers for whom Elevance Health offers administrative services only (ASO). CCQP files contained data from prior authorization requests, which included information on requested drug regimen, line of therapy, cancer type, cancer stage, body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), Eastern Cooperative Oncology Group (ECOG) performance score, National Provider Identifier, and tax identification number of the chemotherapy-prescribing clinicians. We obtained data about prescribing oncologists—clinician sex and years from medical school graduation—from Enclarity’s Provider Data Masterfile (formerly LexisNexis). We used data from the American Community Survey for insights on neighborhood-level social determinants of health. We derived Oncology Care Model (OCM) participation data by practices from the Centers for Medicare & Medicaid Services program website. This study was conducted by researchers using a limited data set for analysis, which was devoid of individual patient identifiers and complied with all relevant provisions of the Health Insurance Portability and Accountability Act (HIPAA) and the HIPAA Privacy Rule (45 CFR 164.514(e)). Findings are reported using the Strengthening the Reporting of Observational Studies in Epidemiology ( STROBE ) guideline for cohort studies. Study Population From CCQP data, we identified adult patients with metastatic cancer of the following cancer types: breast, lung, colorectal, pancreatic, melanoma, kidney, bladder, gastric, and uterine cancer, who were prescribed initial anti-cancer drug therapy between July 1, 2018, and October 31, 2021. These cancer diagnoses represented the most common solid tumor malignant neoplasms among the insurer cohort. We excluded prostate cancer due to preponderance of hormonal therapy prescribed for this indication. Index date was defined as the first request data of the first-line anti-cancer drug therapy after a metastatic diagnosis. Six months of continuous insurance coverage prior to first treatment date were required to assess health care characteristics for patients and clinicians during this baseline period. We limited our evaluation to index cancer drugs prescribed in the first-line setting for metastatic cancer because evidence supporting the choice of initial regimens is stronger than for later-line regimens. Line of therapy was obtained from prior authorization data submitted by oncologists. Outcome The primary outcome was use of a pathway program–endorsed regimen among patients with metastatic cancer treated in the first-line setting. A patient’s pathway status was designated as on-pathway, off-pathway, or unresolved. Unresolved pathway status was present when regimen requests were sufficient for authorization but clinical data provided (such as biomarkers) were inadequate to determine on-pathway or off-pathway status. Patients with unresolved pathway status were excluded from the main analysis. Factors Patient characteristics included age (categorized into ages 18 to 44, 45 to 49, 50 to 64, 65 to 74, and 75 plus years), sex, BMI, Deyo-Charlson Comorbidity index (DCI) score, cancer type, performance status as measured on the ECOG scale, insurance type (commercial fully insured, commercial ASO, or MA), census region (Northeast, Midwest, South, West), urban or rural residence (inside or outside of a US Census Bureau metropolitan statistical area), distance of patient to treatment facility, third party prescription coverage, and health care utilization and cost during the baseline period (eg, inpatient visits, emergency department (ED) visits, total medical cost, total pharmacy cost). Demographic covariates were assessed on index date, and comorbidities and patient history were assessed from claims during the 6 months prior to index date. We obtained the following social determinants of health variables, drawn at the census block group level from the American Community Survey—median family income, Black race, Hispanic ethnicity, education level, and socioeconomic status index scores. Clinician and practice covariates included clinician sex, years since medical school graduation, exposure time to the CCQP pathway program (calculated as months between the time when a clinician first appeared in CCQP data and index date), plan member patient volume per physician (number of distinct patients associated with a clinician during 6 months prior to index date), and participation in the OCM. Statistical Analysis We used a logistic regression model to identify factors associated with receipt of pathway-compliant drug regimens. Stepwise backward selection was used to retain covariates based on Akaike information criteria at P < .157. Continuous variables were dichotomized based on median values for easier interpretation; these variables included median family income, years since physicians’ medical school graduation, months of exposure of physician to the pathways program, and number of plan member patients per physician. Log 2 transformation was performed on total medical cost and total pharmacy cost to remove skewness—this implies that the coefficient should be interpreted as the change in pathway compliance rate when doubling costs during the baseline period. Imputation was not performed given small numbers for data missingness within variables such as geographic region (0.6% to 0.7%), residence in urban area (0.6% to 0.7%), clinician covariates including sex and years since medical school graduation (1.3% to 2.6%), socioeconomic status variables (4.2% to 4.8%) and ECOG status (3.3% to 4.7%). Independent variables were tested for multicollinearity, and excluded if the pairwise correlation coefficient was greater than 0.7. Index year of drug prescribing was included to account for secular trends. We present the results of logistic regression as adjusted odds ratios (aORs). We conducted a sensitivity analysis in which we grouped patients with unresolved pathway status with those on-pathway, with the assumption that an unresolved status would more likely include on-pathway rather than off-pathway regimens, as the regimen requests were otherwise sufficient for authorization. Statistical significance was set at P < .05 in 2-sided tests. Analyses were conducted using SAS Enterprise Guide version 7.15 (SAS Institute Inc). Claims data were obtained from Elevance Health (formerly Anthem Inc), a national insurer that launched an oncology clinical pathway program, the Cancer Care Quality Program (CCQP), through its subsidiary AIM Specialty Health in 2014. An external panel of oncology experts meets on a quarterly basis and reviews, approves, and updates regimens over time to reflect current evidence. To incentivize selection of pathway-compliant regimens, Elevance Health pays practices an additional reimbursement of $350 per patient each month. Treatment requests made by oncologists for regimens not included in the pathways program are reviewed and still authorized without delay or requirement for peer-to-peer interaction if they are determined to be medically necessary pursuant to medical policies and clinical guidelines. We obtained longitudinal medical and pharmacy claims and eligibility files from Elevance Health care plans across the US for information about health care utilization, medical spending, and health plan enrollment. Our data were from 3 types of health plan members: Medicare Advantage (MA), fully insured members, and members of plans through self-insured employers for whom Elevance Health offers administrative services only (ASO). CCQP files contained data from prior authorization requests, which included information on requested drug regimen, line of therapy, cancer type, cancer stage, body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), Eastern Cooperative Oncology Group (ECOG) performance score, National Provider Identifier, and tax identification number of the chemotherapy-prescribing clinicians. We obtained data about prescribing oncologists—clinician sex and years from medical school graduation—from Enclarity’s Provider Data Masterfile (formerly LexisNexis). We used data from the American Community Survey for insights on neighborhood-level social determinants of health. We derived Oncology Care Model (OCM) participation data by practices from the Centers for Medicare & Medicaid Services program website. This study was conducted by researchers using a limited data set for analysis, which was devoid of individual patient identifiers and complied with all relevant provisions of the Health Insurance Portability and Accountability Act (HIPAA) and the HIPAA Privacy Rule (45 CFR 164.514(e)). Findings are reported using the Strengthening the Reporting of Observational Studies in Epidemiology ( STROBE ) guideline for cohort studies. From CCQP data, we identified adult patients with metastatic cancer of the following cancer types: breast, lung, colorectal, pancreatic, melanoma, kidney, bladder, gastric, and uterine cancer, who were prescribed initial anti-cancer drug therapy between July 1, 2018, and October 31, 2021. These cancer diagnoses represented the most common solid tumor malignant neoplasms among the insurer cohort. We excluded prostate cancer due to preponderance of hormonal therapy prescribed for this indication. Index date was defined as the first request data of the first-line anti-cancer drug therapy after a metastatic diagnosis. Six months of continuous insurance coverage prior to first treatment date were required to assess health care characteristics for patients and clinicians during this baseline period. We limited our evaluation to index cancer drugs prescribed in the first-line setting for metastatic cancer because evidence supporting the choice of initial regimens is stronger than for later-line regimens. Line of therapy was obtained from prior authorization data submitted by oncologists. The primary outcome was use of a pathway program–endorsed regimen among patients with metastatic cancer treated in the first-line setting. A patient’s pathway status was designated as on-pathway, off-pathway, or unresolved. Unresolved pathway status was present when regimen requests were sufficient for authorization but clinical data provided (such as biomarkers) were inadequate to determine on-pathway or off-pathway status. Patients with unresolved pathway status were excluded from the main analysis. Patient characteristics included age (categorized into ages 18 to 44, 45 to 49, 50 to 64, 65 to 74, and 75 plus years), sex, BMI, Deyo-Charlson Comorbidity index (DCI) score, cancer type, performance status as measured on the ECOG scale, insurance type (commercial fully insured, commercial ASO, or MA), census region (Northeast, Midwest, South, West), urban or rural residence (inside or outside of a US Census Bureau metropolitan statistical area), distance of patient to treatment facility, third party prescription coverage, and health care utilization and cost during the baseline period (eg, inpatient visits, emergency department (ED) visits, total medical cost, total pharmacy cost). Demographic covariates were assessed on index date, and comorbidities and patient history were assessed from claims during the 6 months prior to index date. We obtained the following social determinants of health variables, drawn at the census block group level from the American Community Survey—median family income, Black race, Hispanic ethnicity, education level, and socioeconomic status index scores. Clinician and practice covariates included clinician sex, years since medical school graduation, exposure time to the CCQP pathway program (calculated as months between the time when a clinician first appeared in CCQP data and index date), plan member patient volume per physician (number of distinct patients associated with a clinician during 6 months prior to index date), and participation in the OCM. We used a logistic regression model to identify factors associated with receipt of pathway-compliant drug regimens. Stepwise backward selection was used to retain covariates based on Akaike information criteria at P < .157. Continuous variables were dichotomized based on median values for easier interpretation; these variables included median family income, years since physicians’ medical school graduation, months of exposure of physician to the pathways program, and number of plan member patients per physician. Log 2 transformation was performed on total medical cost and total pharmacy cost to remove skewness—this implies that the coefficient should be interpreted as the change in pathway compliance rate when doubling costs during the baseline period. Imputation was not performed given small numbers for data missingness within variables such as geographic region (0.6% to 0.7%), residence in urban area (0.6% to 0.7%), clinician covariates including sex and years since medical school graduation (1.3% to 2.6%), socioeconomic status variables (4.2% to 4.8%) and ECOG status (3.3% to 4.7%). Independent variables were tested for multicollinearity, and excluded if the pairwise correlation coefficient was greater than 0.7. Index year of drug prescribing was included to account for secular trends. We present the results of logistic regression as adjusted odds ratios (aORs). We conducted a sensitivity analysis in which we grouped patients with unresolved pathway status with those on-pathway, with the assumption that an unresolved status would more likely include on-pathway rather than off-pathway regimens, as the regimen requests were otherwise sufficient for authorization. Statistical significance was set at P < .05 in 2-sided tests. Analyses were conducted using SAS Enterprise Guide version 7.15 (SAS Institute Inc). Cohort Characteristics The cohort consisted of 17 293 patients, of whom 11 071 (64.0%) were treated with pathway-compliant regimens, 6222 (36.0%) were treated with off-pathway regimens . On-pathway and off-pathway groups had similar mean (SD) age (60.8 [11.1] years vs 60.5 years [11.6]; P = .08), with fewer women in the on-pathway group (5759 of 11 071 [52.0%] vs 3424 of 6222 [55.0%]; P < .001). The top 3 cancer types in both groups were breast, lung, and colorectal cancer. The majority of patients were from the Midwest (5871 [34.0%]), had commercial ASO plans (9659 [55.9%]), and resided in urban areas (13 248 [76.6%]). Mean (SD) BMI was 28.0 (6.8), mean Deyo-Charlson Comorbidity Index score was 6.9 (2.1), and 1535 patients (8.9%) had an ECOG value of 2 or greater. Results from Logistic Regression The odds of pathway compliance decreased over time. Compared with the reference year 2018, in which the pathway compliance rate was 74.4%, the compliance rate was 70% in 2019 (aOR, 0.73; 95% CI, 0.65-0.82), 55.8% in 2020 (aOR, 0.36; 95% CI, 0.32-0.41), and 59.4% in 2021 (aOR, 0.43; 95% CI, 0.38-0.49) (all P < .001) . We found heterogeneity in the level of compliance between cancer types. For example, compared with the reference lung cancer, gastric and bladder cancer had lower odds of being pathway compliant (gastric cancer: aOR, 0.50; 95% CI, 0.40-0.64; P < .001; bladder cancer: aOR, 0.76; 95% CI, 0.61-0.94; P < .001), while pancreatic cancer had much higher odds of being pathway compliant (aOR, 6.06; 95% CI, 5.14-7.14; P < .001). Patients with MA plans had lower odds of being pathway compliant compared with those with ASO plans (aOR, 0.84; 95% CI, 0.75-0.96; P = .008). The presence of inpatient hospitalizations or ED visits during the baseline 6-month period compared with no utilization was associated with higher odds of pathway compliance (inpatient visit in last 6 months: aOR, 1.32; 95% CI, 1.22-1.43; P < .001; ED visit in last 6 months: aOR, 1.21; 95% CI, 1.12-1.31; P < .001) while higher total medical costs in the baseline period was associated with lower odds for pathway compliance (aOR, 0.86; 95% CI, 0.83-0.88; P < .001). One clinician characteristic that was significantly associated with pathway compliance after adjusting for other factors was the patient volume seen by the prescribing clinician. Prescribing clinicians who treated an above average number of patients were more likely to prescribe pathway-compliant regimens (aOR, 1.12; 95% CI, 1.04-1.20; P = .002). Practices that participated in the Oncology Care Model were positively associated with prescribing pathway-compliant regimens (aOR, 1.13; 95% CI, 1.04-1.23; P = .004). In the sensitivity analysis, we grouped together 2794 patients with unresolved pathway status with those on-pathway and found that almost all factors remained significant, with the exception of the OCM participation as a factor losing significance. The cohort consisted of 17 293 patients, of whom 11 071 (64.0%) were treated with pathway-compliant regimens, 6222 (36.0%) were treated with off-pathway regimens . On-pathway and off-pathway groups had similar mean (SD) age (60.8 [11.1] years vs 60.5 years [11.6]; P = .08), with fewer women in the on-pathway group (5759 of 11 071 [52.0%] vs 3424 of 6222 [55.0%]; P < .001). The top 3 cancer types in both groups were breast, lung, and colorectal cancer. The majority of patients were from the Midwest (5871 [34.0%]), had commercial ASO plans (9659 [55.9%]), and resided in urban areas (13 248 [76.6%]). Mean (SD) BMI was 28.0 (6.8), mean Deyo-Charlson Comorbidity Index score was 6.9 (2.1), and 1535 patients (8.9%) had an ECOG value of 2 or greater. The odds of pathway compliance decreased over time. Compared with the reference year 2018, in which the pathway compliance rate was 74.4%, the compliance rate was 70% in 2019 (aOR, 0.73; 95% CI, 0.65-0.82), 55.8% in 2020 (aOR, 0.36; 95% CI, 0.32-0.41), and 59.4% in 2021 (aOR, 0.43; 95% CI, 0.38-0.49) (all P < .001) . We found heterogeneity in the level of compliance between cancer types. For example, compared with the reference lung cancer, gastric and bladder cancer had lower odds of being pathway compliant (gastric cancer: aOR, 0.50; 95% CI, 0.40-0.64; P < .001; bladder cancer: aOR, 0.76; 95% CI, 0.61-0.94; P < .001), while pancreatic cancer had much higher odds of being pathway compliant (aOR, 6.06; 95% CI, 5.14-7.14; P < .001). Patients with MA plans had lower odds of being pathway compliant compared with those with ASO plans (aOR, 0.84; 95% CI, 0.75-0.96; P = .008). The presence of inpatient hospitalizations or ED visits during the baseline 6-month period compared with no utilization was associated with higher odds of pathway compliance (inpatient visit in last 6 months: aOR, 1.32; 95% CI, 1.22-1.43; P < .001; ED visit in last 6 months: aOR, 1.21; 95% CI, 1.12-1.31; P < .001) while higher total medical costs in the baseline period was associated with lower odds for pathway compliance (aOR, 0.86; 95% CI, 0.83-0.88; P < .001). One clinician characteristic that was significantly associated with pathway compliance after adjusting for other factors was the patient volume seen by the prescribing clinician. Prescribing clinicians who treated an above average number of patients were more likely to prescribe pathway-compliant regimens (aOR, 1.12; 95% CI, 1.04-1.20; P = .002). Practices that participated in the Oncology Care Model were positively associated with prescribing pathway-compliant regimens (aOR, 1.13; 95% CI, 1.04-1.23; P = .004). In the sensitivity analysis, we grouped together 2794 patients with unresolved pathway status with those on-pathway and found that almost all factors remained significant, with the exception of the OCM participation as a factor losing significance. We found that despite participation in an oncology clinical pathway program that included financial incentives to encourage clinicians to prescribe on-pathway treatment regimens, only 64% of patients with newly diagnosed metastatic cancer received on-pathway treatments. This figure is consistent with prior studies. , Variable compliance threatens the internal validity of the use of pathways compliance as an indicator of quality prescribing. While the goal is not to achieve a 100% compliance rate given that a certain portion of pathway deviations are not only expected but likely appropriate, previously published experiences of operationalizing cancer clinical pathways have targeted between 70% to 80% compliance as a threshold for success. In order to boost compliance rates, Elevance Health pays practices an additional reimbursement of $350 per patient each month when they adhere to pathways, which is not insignificant. As a benchmark, the additional monthly per-patient payments in the OCM were $160. Like other pathway programs, adherence reports are also provided for their clinic customers. Many factors likely influence pathway compliance, such as patient factors, individual physician’s beliefs and routines, peer effects, etc. In at least 1 notable study looking at uptake of guideline-directed care, the most important variance in drug prescribing was associated with physician characteristics, not patient or disease characteristics. Given that many older adults are excluded from clinical trials, geriatric oncology research has shown that value-based assessments often lead to deviations from standards of care. Other studies have found that pathway adherence can range from 53% to 70% depending on disease site, and tends to decrease with increasing pathway complexity. , We would be remiss not to mention the impact of the COVID-19 pandemic on treatment modifications and pathway deviations, particularly in the early, prevaccine days of the pandemic. One additional and important possibility is related to clinic workflow: payer-led pathways are not routinely incorporated into the electronic treatment ordering system. This can lead to oncologists not being aware of which treatments are on- or off-pathway at the moment of clinical decision-making, and requires that oncologists asynchronously study pathways for any differences in their plan of care and selection of regimen. Executing on pathways under these circumstances involves creating prior awareness and socialization of the program and the pathway choices, and effective communication is needed between the clinical staff and the administrative staff that enters the regimen and clinical scenario details into the electronic portal. We found that the rate of pathway compliance decreased over time. This was a surprising finding, as one might expect that program maturity and physician familiarity would tilt the results in the other direction. Hypotheses to explain this finding include an increasing multiplicity of treatment options with time, lag time from regulatory approval of a new regimen to inclusion on pathway, the aforementioned implementation challenges related to workflow, increased biomarker reporting requirements for pathway adjudication that increase risk of missing data, and pathway deviations in the setting of the pandemic, particularly with the disruption of the flow of information in the practices with the displacement of administrative staff. Two findings are noteworthy to policy makers or program architects seeking to influence clinician prescribing behavior. First, the positive association between plan member patient volume within a physician’s panel and prescribing of on-pathway regimens suggests that increasing familiarity with the pathways program is likely to influence compliance. In its State of Cancer Care in America 2017 report, ASCO reported a 42% increase in the preceding 2 years in practices reporting implementation of a pathway program. As such, it remains to be seen if this hypothesis bears out in practice, as physician exposure to pathways programs increases over time. Second, practice participation in Centers for Medicare & Medicaid Services’s flagship value-based payment model for cancer—the OCM—was also modestly associated with higher likelihood of pathway compliance. The OCM is a total cost-of-care model, and as such, it is focused on the outcome of cost. Pathway programs focus on the process, or the actual selecting of drug regimens. The synergy between these 2 approaches should be of interest to policy makers looking to iterate on the next value-based care model. We were surprised by the lack of association between ECOG status and the Deyo-Charlson Comorbidity Index on pathway compliance. Given the frequent intersection between complexity and high health care utilization, we studied patients’ health care utilization (dichotomized to any use or no use) and cost during the baseline period. While health care utilization during the baseline period was positively associated with compliance, higher total medical costs in the baseline period led to lower odds of pathway compliance. These results are discordant; we would note though that both are imperfect measures of complexity. In future studies, we would consider creating distinct classes of health care utilization along a gradation in order to describe profiles of health complexity and utilization with more nuance. Limitations Limitations of this study include the fact that this was a retrospective analysis. There was likely confounding by indication in the odds of whether a patient receives pathway-endorsed drug regimens or not. We were not able to distinguish the reasons for off-pathway prescribing, such as when there is a delay between when evidence is generated and inclusion of that regimen into the pathways program (which would not be indicative of poor-quality care). Administrative data lack clinical details, which could have informed some of the counterintuitive findings such as the association between clinical complexity and likelihood of pathway compliance. We did not have 100% capture rate of on- or off-pathway status, with unresolved cases due to missing data. While we have information about intended treatment from prior authorizations, we do not have information on whether patients actually started their prescribed regimens. We tried to adjust for selection bias by limiting our analysis by stage and line of therapy, and by adjusting for patient characteristics including age, sex, and comorbidity score. Program execution was only alluded to in covariates such as patient volume and payer type, but was likely a critical factor in the success of this program. Limitations of this study include the fact that this was a retrospective analysis. There was likely confounding by indication in the odds of whether a patient receives pathway-endorsed drug regimens or not. We were not able to distinguish the reasons for off-pathway prescribing, such as when there is a delay between when evidence is generated and inclusion of that regimen into the pathways program (which would not be indicative of poor-quality care). Administrative data lack clinical details, which could have informed some of the counterintuitive findings such as the association between clinical complexity and likelihood of pathway compliance. We did not have 100% capture rate of on- or off-pathway status, with unresolved cases due to missing data. While we have information about intended treatment from prior authorizations, we do not have information on whether patients actually started their prescribed regimens. We tried to adjust for selection bias by limiting our analysis by stage and line of therapy, and by adjusting for patient characteristics including age, sex, and comorbidity score. Program execution was only alluded to in covariates such as patient volume and payer type, but was likely a critical factor in the success of this program. This cohort study on factors associated with compliance with a national payer-led oncology clinical pathway program yielded several expected and unexpected results. Despite significant financial incentives, compliance with the pathways program remained at previously reported rates, and furthermore, compliance appeared to trend down over the years. There is synergy between this program and other value-based payment programs such as the OCM. The contribution of patient complexity toward physician treatment decisions remains poorly understood. A forthcoming companion study will examine associations of pathway compliance with several cost and quality outcome measures. These data will be useful to architects of value-based payment models seeking best practices.
Genomic investigations of unexplained acute hepatitis in children	0e36b962-13c9-42a6-a8d7-1c18802110fc	10170458	Anatomy[mh]	In March 2022, the report of five cases of severe hepatitis of unknown aetiology led to the UK Health Security Agency (UKHSA) identifying 278 cases in total as of 30 September 2022 . Cases, defined as acute non-A–E hepatitis with serum transaminases of more than 500 IU in children under 10 years of age, were found to have been occurring since January 2022 . In the UK, 196 cases required hospitalization, 69 were admitted to intensive care and 13 required liver transplantation . Case numbers have declined since April 2022 . UKHSA investigations identified HAdV to be commonly associated with the unexplained paediatric hepatitis, with 64.7% (156 of 241) testing positive in one or more samples from whole blood (the most sensitive sample type ) or mucosal swabs. HAdVs from the blood of 35 of 77 patients were typed as F41. Seven of eight patients in England who required liver transplantation tested HAdV positive in blood samples, with F41 found in five of five genotyped . SARS-CoV-2 infection was detected in 8.9% (15 of 169) of UK and 12.8% (16 of 125) of English cases . Given the uncertainty around the aetiology of this outbreak, and the potential that HAdV-F41, if implicated (Fig. ), could be a new or recombinant variant, we undertook untargeted metagenomic and metatranscriptomic sequencing of liver biopsies from five liver transplant cases and whole blood from five non-transplanted cases (Table and Fig. ). The results were further verified by confirmatory PCRs of liver, blood, stool and nasopharyngeal samples from a total of 38 cases for which there was sufficient residual material. We compared our results with those from 13 healthy children and 52 previously healthy children presenting to hospital with other febrile illness, including HAdV, hepatitis unrelated to the current outbreak or a critical illness requiring admission to the intensive care unit. We also tested blood and liver biopsies from 17 profoundly immunosuppressed children with hepatitis who were not part of the current outbreak, in whom reactivation of latent infections might be expected. We received samples from 38 children meeting the case definition (Table ). All cases were less than 10 years of age and 22 of 23 cases previously tested were positive by HAdV PCR (Table , Extended Data Table and Supplementary Table ). A summary of the samples received from these cases and the investigations carried out on them are shown in Fig. . Pre-existing conditions, autoimmune, toxic and other infectious causes of hepatitis were excluded in 12 transplanted (cases 1–5, 28, 29, 31–34 and 36) and four non-transplanted (cases 30, 35, 37 and 38) children, investigated at two liver transplant units (Supplementary Table ). The 12 transplanted cases reported gastrointestinal symptoms (nausea, vomiting and diarrhoea) preceding transplant by a median of 20 days (range 8–42 days). All 12 transplanted children survived, whereas the four children who did not receive liver transplants recovered without sequelae or evidence of chronic liver-related conditions. Five of the remaining 22 cases referred by Health Security Agencies, for whom this information was available, recovered without sequelae (Table and Supplementary Table ). We performed metagenomic and metatranscriptomic sequencing on samples of frozen explanted liver tissue from five cases who received liver transplants (median age of 3 years) and six blood samples from five non-transplanted hepatitis cases (median age of 5 years) (Table and Fig. ). The liver samples had uniform and consistently high sequencing depth both for DNA sequencing (DNA-seq) and RNA-seq, whereas the blood samples had variable sequencing depth particularly for RNA-seq (Supplementary Table ). We detected abundant AAV2 reads in DNA-seq from five of five explanted livers and four of five blood samples from non-transplant cases (7–42 and 1.2–42 reads per million, respectively) (Table ). Lower levels of HHV-6B were present in DNA-seq of all explanted liver samples (0.09–4 reads per million) but not in the six blood samples (Table ). HAdV was detected (five reads) in one blood sample (Table ). Metatranscriptomics revealed AAV2, but not HHV-6B or HAdV, RNA reads, in liver and blood samples (0.7–10 and 0–7.8 reads per million, respectively). Mapping liver RNA-seq data to the RefSeq AAV2 genome ( NC_001401.2 ) identified high expression of the Cap open reading frame, particularly at the 3′ end of the capsid, suggesting viral replication (Extended Data Fig. ), whereas reverse transcription (RT)–PCR of two livers confirmed the presence of AAV2 mRNA from the Cap open reading frame (Extended Data Fig. ). In the blood samples, which had not been treated to preserve RNA, we detected low levels of AAV2 RNA reads mapping throughout the genome (Extended Data Fig. ). Ligation-based untargeted nanopore sequencing was applied to DNA from four of five frozen liver samples. All four samples were initially sequenced at a lower depth (average N50 of 8.37 kb). Six to sixteen AAV2 reads were obtained from each sample (5.57–22.24 million total reads; Supplementary Table ). Mapping revealed concatenation of the 4-kb genome, compatible with active AAV2 replication . We observed alternating and head-to-tail concatemers, which could be consistent with both HAdV and human herpesvirus-mediated rolling hairpin and rolling circle replication, respectively . Two of these samples were sequenced more deeply, resulting in 52 and 178 AAV2 reads in 82.9 and 122 million total (N50 of 4.40–8.52 kb) (Supplementary Table ). Of the reads in the more deeply sequenced datasets, 42–48% comprised randomly linked, truncated and rearranged genomes, with few that were intact and of full length (Extended Data Fig. ). The remaining reads were less than 3,000 bp long and may represent sections either of monomeric genomes or of more complex structures. There was some evidence of AAV2 integration by deeper nanopore sequencing of explanted livers (Supplementary Table ); however, none of the integration sites was confirmed by Illumina metagenomic or targeted AAV2 sequencing. The results are likely to represent artefacts of this library preparation method; chimeric reads have been described to occur in 1.7–3% of reads , . Given the number of human reads (72–120 million), we might expect to see this artefact occurring most commonly between AAV2 and human than between AAV2 reads. Where sufficient residual material was available, PCR tests were performed for AAV2 (28 of 38 cases), HAdV (31 of 38 cases) and HHV-6B (23 of 38 cases). The results confirmed high levels (cycle threshold (Ct) values: 17–21) of AAV2 DNA in all five frozen explanted livers that had undergone metagenomics (Table and Fig. ), and lower levels of HHV-6B and HAdV DNA (Ct values: 27–32 and 37–42, respectively). AAV2 DNA was also detected (Ct values: 19–25) in blood samples from four of five cases that had undergone metagenomics, whereas HAdV, at levels too low to genotype, and HHV-6B were detected in two of four and three of four cases, respectively (one case had insufficient material) (Table ). One of the blood metagenomics cases (case 9, JBB1) with insufficient material to test for HAdV and HHV-6B, tested positive for both viruses in the referring laboratory. The AAV2-negative blood sample (case 10, JBB15) was also negative for HAdV but positive for HHV-6B (Table ). A further ten of ten blood samples tested from cases were positive for HAdV by PCR. Sufficient material was available for AAV2 PCR in six of these (all positive; Ct values: 20–23) and HHV-6B PCR in two (one positive Ct value: 37) (Extended Data Table ). AAV2 PCR was positive in nine formalin-fixed paraffin-embedded (FFPE) liver samples, including seven from transplanted cases (Ct values: 23–25) and two from non-transplanted cases (Ct values: 34–36; Extended Data Table ). HHV-6B PCR was positive in six of seven FFPE samples (not case 32) from transplanted (Ct values: 30–37) and zero of two from non-transplanted (cases 30 and 35) cases, with positive HAdV (Ct values: 40–44) in four of nine cases. Three transplanted (cases 32, 34 and 36) and three non-transplanted (cases 35, 37 and 38) cases had serum available for testing. All were AAV2 positive (Ct values: 27–32) and HHV-6B negative, with one transplanted case and one non-transplanted case testing HAdV positive (Extended Data Table ). Together, 27 of 28 cases tested were AAV2 PCR positive, 23 of 31 were HAdV positive and 16 of 23 were HHV-6B positive. When results from referring laboratories were included, 33 of 38 cases were positive for HAdV and 19 of 26 cases were positive for HHV-6B (Table and Extended Data Table ). To better contextualize the findings in cases with unexplained hepatitis, we selected control groups of children who were not part of the outbreak. Whole blood from 65 immunocompetent children matched by age to cases (median age of 3.8 years) ( Fig. , Extended Data Table and Supplementary Table ) who were healthy, or had HAdV infection, hepatitis or critical illness, including requiring critical care, were selected from the PERFORM (personalised risk assessment in febrile illness to optimise real-life management; www.perform2020.org ) and DIAMONDS (diagnosis and management of febrile illness using RNA personalised molecular signature diagnosis study; www.diamonds2020.eu ) studies. Both studies recruited children presenting to hospital with an acute-onset febrile illness between 2017 and 2020 (PERFORM) and July 2020 to October 2021, during the COVID-19 pandemic (DIAMONDS) (Supplementary Table ). Of the PERFORM–DIAMONDS control whole-blood samples, 6 of 65 (9.2%) were AAV2 PCR positive (Supplementary Table ), compared with 10 of 11 (91%) whole-blood samples from cases (Fig. ; P = 8.466 × 10 −8 , Fisher’s exact test). AAV2 DNA levels were significantly higher in whole-blood samples from cases than from controls (Fig. ; P = 2.747 × 10 − 11 , Mann–Whitney test ) . One participant with an HAdV-F41-positive blood sample, originally thought to have unexplained paediatric hepatitis, was later found to have a previous condition that explained the hepatitis and was therefore reclassified as a control (referred to as ‘reclassified control’ or CONB40) (Supplementary Table ). This blood sample was negative for AAV2 by PCR (Supplementary Table ). Frozen liver biopsy material from four immunocompromised children (median age of 10 years) (CONL1–4) who had been investigated for other forms of hepatitis was also tested (Fig. and Extended Data Table ). In three children, liver enzyme levels were raised (Supplementary Table ); no results were available for CONL4. AAV2 was detected in CONL3 (Ct value: 39) and HHV-6B was detected in CONL2 (Ct value: 34), whereas HAdV was negative (Fig. and Supplementary Table ). We also tested immunocompromised children who are more likely to reactivate latent viruses. Whole-blood samples from 17 immunocompromised children (median age of 1 year) with raised levels of liver transaminases (AST/ALT of more than 500 IU) and viraemia (HAdV or cytomegalovirus), all sampled in 2022 (Fig. ), were tested for AAV2, HHV-6B and HAdV (Extended Data Table and Supplementary Table ). The majority had received human stem cell or solid organ transplants, and none was linked to the recent hepatitis outbreak (Extended Data Table ). Five of 15 (33%) whole-blood samples were positive for HHV-6B, whereas 6 of 17 (35%) were positive for AAV2, significantly fewer than in cases ( P = 0.005957, Fisher’s exact test) and at significantly lower Ct levels ( P = 6.517 × 10 − 5 , Mann–Whitney test) (Fig. and Supplementary Table ). One HAdV-positive and AAV2-positive immunocompromised comparator (CONB23) was also positive for HHV-6B (Supplementary Table ). Four of the six AAV2-positive children from the PERFORM–DIAMONDS cohort (Fig. and Supplementary Table ) and all six of the AAV2-positive immunocompromised children ( Fig. and Supplementary Table ) were also HAdV positive. One full HAdV-F41 genome sequence from the stool of one case (OP174926, case 22) (Supplementary Table ) clustered phylogenetically with the HAdV-F41 sequence obtained from the reclassified control (CONB40) and with other HAdV-F41 sequences collected between 2015 and 2022, including 23 contemporaneous stool samples from children without the unexplained paediatric hepatitis (Figs. and ). Sequencing and k -mer analysis of HAdV from 13 cases with partial sequences identified the genotype HAdV-F41 in 12 cases (Supplementary Tables and ). The partial sequences showed most similarity to the control sequence OP047699 (Supplementary Table ), mapping across the entire viral genome, thus further excluding a recombinant virus. Single-nucleotide polymorphisms were largely shared between the single HAdV-positive stool from a case (OP174926) and control whole-genome sequences (Extended Data Fig. ). Given reported mutation rates for HAdV-F41 and other adenoviruses , , any differences are likely to have arisen before the outbreak. No new or unique amino acid substitutions were noted in HAdV sequences from cases with only two substitutions overall (Extended Data Fig. ) and none in proteins critical for AAV2 replication. AAV2 sequences from 15 cases, including five from the explanted livers and ten from whole blood from non-transplanted cases, clustered phylogenetically with control AAV2 sequences obtained from four immunocompromised HAdV-positive children with elevated levels of ALT in the comparator group (Extended Data Table ) and two healthy children with recent HAdV-F41 diarrhoea (Fig. and Supplementary Table ). The degree of diversity and lack of a unique common ancestor between case AAV2 genomes suggest that these are not specific to the hepatitis outbreak, but instead reflect the current viral diversity of the general population. Although comparison of the AAV2 sequences showed no difference between cases and controls, contemporary AAV2 sequences showed changes in the capsid compared with historic AAV2 (Extended Data Fig. ). None of these changes was shared with the hepatotropic AAV7 and AAV8 viruses (Extended Data Fig. ). The majority of the contemporary AAV2 genomes in cases and controls (20 of 21) contained a stop codon in the X gene, which is involved in viral replication , whereas historic AAV2 genomes contained this less frequently (11 of 35). The significance, if any, of this is currently unknown. Although mean read depths for four HHV-6B genomes recovered from explanted livers were low (×5–10) (Supplementary Table ), phylogeny (Fig. ) confirmed that all were different. Using a recombinant AAV2 (rAAV2) vector with a VP1 sequence (Extended Data Fig. ) containing the consensus amino acid sequence from AAV2 cases (AAV2Hepcase) (Extended Data Fig. ), we generated functional rAAV particles that transduced Huh-7 cells with comparable efficacy to both canonical AAV2 and the synthetic liver-tropic LK03 AAV vector . Unlike canonical AAV2, the AAV2Hepcase capsid, which contains mutations (R585S and R588T) that potentially affect the heparin sulfate proteoglycan (HSPG)-binding domain, was unaffected by heparin competition, a feature that is associated with increased hepatotropism , (Extended Data Fig. ). Histological examination of the 12 liver explants and two liver biopsies showed nonspecific features of acute hepatitis with ballooning hepatocytes, disrupted liver architecture with varying degrees of perivenular, bridging or pan-acinar necrosis. There was no evidence of fibrosis suggestive of an underlying chronic liver disease. The appearances were similar to historic cases of seronegative hepatitis of unknown cause in children. There were no typical histological features of autoimmune hepatitis, notably no evidence of portal-based plasma cell-rich infiltrates. A cellular infiltrate was present in all cases, which on staining appeared to be predominantly of CD8 + T cells but also included CD20 + B cells. More widespread staining with the CD79a pan-B cell lineage, which also identifies plasma cells, was also observed (Extended Data Fig. ). Macrophage lineage cells showed some C4d complement staining, whereas staining for immunoglobulins was nonspecific with disruption of the normal canalicular staining seen in controls due to the architectural collapse. MHC class I and class II staining, although increased in cases, was nonspecific and associated with sinusoid-containing blood cells and necrotic tissue (Extended Data Fig. ). No viral inclusions were observed and there were no features suggestive of direct viral cytopathic effect. Immunohistochemistry was negative for adenovirus. Staining of the five explanted livers with AAV2 antibodies demonstrated evidence of nonspecific ingested debris but not the nuclear staining seen in the positive AAV2-infected cell lines and infected mouse tissue (Extended Data Fig. ). All five liver explants showed positive staining of macrophage-derived cells with antibody to HHV-6B, with no staining of negative control serial sections (Extended Data Fig. ). No specific HHV-6B staining was observed in 13 control liver biopsies from patients (including three children less than 18 years of age) with other viral hepatitis, toxic liver necrosis, autoimmune and other hepatitis, and normal liver. The control set was also negative for HAdV and AAV2 by immunohistochemistry. Liver sections were morphologically suboptimal for electron microscopy, but no viral particles were identified in hepatocytes, blood vessel endothelial cells and Kupffer cells. We quantified functional cytokine activity by expression of independently derived cytokine-inducible transcriptional signatures of cell-mediated immunity (Supplementary Table ) in bulk genome-wide transcriptional profiles from four of the frozen explanted livers. Results were compared with published data from normal adult livers ( n = 10) and adult hepatitis B-associated acute liver failure ( n = 17) ( GSE96851 ) . Data from the unexplained hepatitis cases revealed increased expression of diverse cytokines and pathways compared with normal liver. These pathways included prototypic cytokines associated with T cell responses, including IFNγ, IL-2, CD40LG, IL-4, IL-5, IL-7, IL-13 and IL-15 (Fig. and Supplementary Table ), as well as some evidence of innate immune type I interferon responses. Many of these responses showed substantially greater activity in unexplained hepatitis than in fulminant hepatitis B virus disease. The most striking enrichment was for TNF expression, and included other canonical pro-inflammatory cytokines including IL-1 and IL-6 (Extended Data Fig. ). These data are consistent with an inflammatory process involving multiple pathways. Proteomic analysis of the five frozen explanted livers did not detect AAV2 or HAdV proteins. Expression of HHV-6B U4, a protein of unknown function, was found in four of five cases; U43, part of the helicase primase complex, was found in two of five cases; and U84, a homologue of cytomegalovirus UL117, implicated in HHV-6B nuclear replication, was found in two of five cases (Extended Data Fig. ). The human proteome from the five frozen liver explants was compared with publicly available data from seven control ‘normal’ livers, taken from two different studies , . Both protein and peptide analyses (Fig. and Supplementary Tables and ) found increased expression in unexplained hepatitis cases of HLA class II proteins and peptides (for example, HLADRB1 and HLADRB4), multiple peptides from variable regions of the heavy and light chains of immunoglobulin, complement proteins (such as C1q) and intracellular and extracellular released proteins from neutrophils and macrophages (MMP8 and MPO). There was no evidence of HAdV, AAV2 or HHV-6B in any of the control livers. Despite reports implicating HAdV-F41 as causing the recent outbreak of unexplained paediatric hepatitis, we found very low levels of HAdV DNA, no proteins, inclusions or viral particles, including in explanted liver tissue from affected cases and no evidence of a change in the virus. By contrast, metagenomic and PCR analysis of liver tissue and blood identified high levels of DNA from AAV2, a member of the Dependoparvovirus genus, which has not been previously associated with clinical disease, in 27 of 28 cases. Replication of AAV2 requires co-infection with a helper virus, such as HAdV, herpesviruses or papillomavirus , and can also be triggered in the laboratory by cellular damage , raising the possibility that the AAV2 detected was a bystander of previous HAdV-F41 infection and/or liver damage. Against this, we found little or no AAV2 in blood from age-matched, immunocompetent children including those with HAdV infection, hepatitis or critical illness (Fig. ). AAV2 has been reported to establish latency in the liver ; however, even in critically ill immunosuppressed children with hepatitis in whom reactivation might occur, we detected AAV2 infrequently and at significantly lower levels in the blood or in liver biopsies (Fig. ). RNA transcriptomic and real-time PCR data from explanted livers point to active AAV2 infection, although we did not detect AAV2 proteins by immunohistochemistry (Extended Data Fig. ) or proteomics (Extended Data Fig. ) or any viral particles. The abundant AAV2 genomes in the explanted liver are concatenated with many complex and abnormal configurations. AAV genome concatenation may occur during AAV2 replication , whereas abnormal AAV2 DNA complexes and rearrangements have been observed in the liver following AAV gene therapy . Hepatitis following AAV gene therapy has been well described – , with deaths occurring, albeit rarely . The pattern of complexes typify both HAdV and herpesvirus (including HHV-6B)-mediated AAV2 DNA replication . The presence of HHV-6B DNA in 11 of 12 explanted livers, but not in livers (0 of 2) of non-transplanted children, or control livers as well as the expression, in 5 of 5 cases tested, of HHV-6B proteins, including U43, a homologue of the HSV1 helicase primase UL52, which is known to aid AAV2 replication, highlight a possible role for HHV-6B as well as HAdV in the pathogenesis of AAV2 hepatitis, particularly in severe cases. Although AAV2 is also capable of chromosomal integration – , we found little evidence of this by long read sequencing, computational analysis of metagenomics data or examination of unmapped reads, although further confirmatory studies may be required. Although the pathogenesis of unexplained paediatric hepatitis and the role of AAV2 remain to be determined, our results point strongly to an immune-mediated process. Transcriptomic and proteomic data from the five explant livers identified significant immune dysregulation involving genes and proteins that are strongly associated with activation of B cells and T cells, neutrophils and macrophages as well as innate pathways. The findings are supported by immunohistochemical staining showing infiltration into liver tissue of CD8 + , B cell and B cell lineage cells. Upregulation of canonical pro-inflammatory cytokines including lL-15, which has also been seen in a mouse model of AAV hepatitis , IL-4 and TNF occurred at levels greater even than are seen in fulminant liver failure following infection with hepatitis B virus. Increased levels in the same immunoglobulin variable region peptides and corresponding proteins from both immunoglobulin heavy and light chains across all five livers point to specific antibody involvement . HLA-DRB1*04:01 (12 of 13 cases tested) (Supplementary Table ) among children in our study supports the same genetic predisposition as mooted in a parallel study conducted in Scotland . An immune-mediated process is consistent with studies of hepatitis following AAV gene therapy, in which raised AAV2 IgG and capsid specific cytotoxic T lymphocytes are observed in the affected patients; however, whether these directly mediate hepatitis remains unclear , . Although we did not find that AAV2 sequences in cases differed from those in AAV2 occurring as co-infections in HAdV-F41-positive stool collected from control children during the contemporary HAdV-F41 gastroenteritis outbreak (Fig. ), rAAV capsid expressing a consensus capsid sequence from the unexplained hepatitis cases (AAV2Hepcase) showed reduced HSPG dependency, compared with canonical AAV2 (Extended Data Fig. ), while retaining hepatocyte transduction ability. This points to likely greater in vivo hepatotropism of currently circulating AAV2 than has hitherto been assumed from data on canonical AAV2 (ref. ). Another member of the parvovirus family, equine parvovirus-hepatitis, has also been associated with acute hepatitis in horses (Theiler’s disease) . There are several limitations to our study. Although other known infectious, autoimmune, toxic and metabolic aetiologies have been excluded including by other studies , , the number of cases investigated here is small, the study is retrospective, the immunocompromised controls were not perfectly age-matched, and only one immunocompetent and 17 immunocompromised controls were sampled during exactly the same period as the outbreak. Age-matched, immunocompetent controls contemporaneous with the outbreak from the DIAMONDS study, although few in number, were however found to be AAV2 negative in a separate study carried out in Scotland . Finally, our data alone are not sufficient on their own to rule out a contribution from SARS-CoV-2 Omicron, the appearance of which preceded the outbreak of unexplained hepatitis (Supplementary Table ). We did not detect SARS-CoV-2 metagenomically even in three participants who tested positive on admission. Moreover, although seropositivity was higher in our cases (15 of 20) than in controls (3 of 10), this was not the case for another UK cohort (38%) or in preliminary data from a UKHSA case–control study , which showed similar SARS-CoV-2 antibody prevalence between unexplained hepatitis cases and population controls (less than 5 years of age: 60.5% versus 46.3%, respectively; 5–10 years of age: 66.7% versus 69.6%, respectively). In line with UK national recommendations at the time, none of the children had received a COVID vaccine. Although we found little evidence for SARS-CoV-2 directly causing the hepatitis outbreak, we cannot exclude the effect of the COVID-19 pandemic on child mixing and infection patterns. The contemporaneous development of unexplained paediatric hepatitis with a national outbreak of HAdV-F41 (ref. ) and the finding of HAdV-F41 in many cases suggest that the two are linked. Enteric HAdV infection is most common in those younger than 5 years of age , and infection is influenced by mixing and hygiene . Few cases of HAdV-F41 occurred between 2020 and 2022 and no major outbreaks were recorded . The current HAdV outbreak followed relaxation of restrictions due to the pandemic and represented one of many infections, including other enteric pathogens that occurred in UK children following return to normal mixing . Under normal circumstances, the levels of AAV2 antibodies are high at birth, subsequently declining to reach their lowest point at 7–11 months of age, increasing thereafter through childhood and adolescence . AAV2 is known to spread with respiratory HAdVs, infections that declined during the COVID-19 pandemic, and has not been detected by us in over 30 SARS-CoV-2-positive nasopharyngeal aspirates (data not shown). We also found AAV2 DNA to be present in HAdV-F41-positive stool from both cases and controls (Supplementary Table ). With loss of child mixing during the COVID-19 pandemic, reduced spread of common respiratory and enteric viral infections and no evidence of AAV2 in SARS-CoV-2-positive nasopharyngeal swabs, it is likely that immunity to both HAdV-F41 and AAV2 declined sharply in the age group affected by this unexplained hepatitis outbreak. Pre-existing antibody is known to reduce levels of AAV DNA in the liver of non-human primates following infusion of AAV gene therapy vectors . The possibility that, in the absence of protective immunity, excessive replication of HAdV-F41 and AAV2 with accumulation of AAV2 DNA in the liver led to immune-mediated hepatic disease in genetically predisposed individuals needs further investigation. Evaluation of drugs that inhibit TNF and other cytokines massively elevated in this condition may identify important therapeutic options for future cases. Ethics Metagenomic analysis and HAdV sequencing were carried out by the routine diagnostic service at Great Ormond Street Hospital (GOSH). Additional PCRs, immunohistochemistry and proteomics on samples received for metagenomics are part of the GOSH protocol for confirmation of new and unexpected pathogens. The use for research of anonymized laboratory request data, diagnostic results and residual material from any specimen received in the GOSH diagnostic laboratory, including all cases received from Birmingham’s Children Hospital UKHSA, Public Health Wales, Public Health Scotland as well as non-case samples from UKHSA, Public Health Scotland and GOSH research was approved by UCL Partners Pathogen Biobank under ethical approval granted by the NRES Committee London-Fulham (REC reference: 17/LO/1530). Children undergoing liver transplant were consented for additional research under the International Severe Acute Respiratory and Emerging Infection Consortium (ISARIC) WHO Clinical Characterisation Protocol UK (CCP-UK) (ISRCTN 66726260) (RQ3001-0591, RQ301-0594, RQ301-0596, RQ301-0597 and RQ301-0598). Ethical approval for the ISARIC CCP-UK study was given by the South Central–Oxford Research Ethics Committee in England (13/SC/0149), the Scotland A Research Ethics Committee (20/SS/0028) and the WHO Ethics Review Committee (RPC571 and RPC572). The UKHSA has legal permission, provided by regulation 3 of The Health Service (Control of Patient Information) Regulations 2002, to process patient confidential information for national surveillance of communicable diseases and, as such, individual patient consent is not required. Control participants from the EU Horizon 2020 research and innovation program DIAMONDS–PERFORM (grant agreement nos. 668303 and 848196) were recruited according to the approved enrolment procedures of each study, and with the informed consent of parents or guardians: DIAMONDS (London-Dulwich Research Ethics Committee: 20/HRA/1714) and PERFORM (London-Central Research Ethics Committee: 16/LO/1684). The sample IDs for the cases and controls are anonymized IDs that cannot reveal the identity of the study participants and are not known to anyone outside the research group, such as the patients or the hospital staff. Samples Initial diagnostic testing by metagenomics and PCR was performed at GOSH Microbiology and Virology clinical laboratories. Further WGS and characterization were performed at UCL. Cases Birmingham Children’s Hospital provided us with explanted liver tissue from five biopsy sites from five cases, five whole blood 500 µl from four cases and serum plasma from one case (Table and Fig. ). These were used in metagenomics testing (Table ), followed by HAdV, HHV-6 and AAV2 testing by PCR and, depending on the Ct value, WGS (Supplementary Tables , and ). We subsequently received 25 additional specimens from UKHSA, Public Health Wales and Public Health Scotland/Edinburgh Royal Infirmary, including 16 additional blood samples, four respiratory specimens and five stool samples, for HAdV WGS and, depending on residual material, for AAV2 PCR testing followed by sequencing (Tables and , Fig. and Supplementary Tables , and ). We also received ten FFPE liver biopsy samples and six serum samples from 11 cases from King’s College Hospital (Table ). Of these cases, seven had received liver transplants. Controls from DIAMONDS and PERFORM PERFORM recruited children from ten EU countries (2016–2020). PERFORM was funded by the European Union’s Horizon 2020 programme under GA no. 668303. DIAMONDS is funded by the European Union Horizon 2020 programme grant number 848196. Recruitment commenced in 2020 and is ongoing. Both studies recruited children presenting with suspected infection or inflammation and assigned them to diagnostic groups according to a standardized algorithm. Controls from GOSH for PCR Blood samples from 17 patients not linked to the non-A–E hepatitis outbreak were tested by real-time PCR targeting AAV2 (Extended Data Table ). These comparators were patients with ALT/AST of more than 500 and HAdV or cytomegalovirus viraemia. These were purified DNA from residual diagnostic specimens received in the GOSH microbiology and virology laboratory in the previous year. All residual specimens were stored at −80 °C before testing and pseudo-anonymized at the point of processing and analysis. Viraemia was initially detected using targeted real-time PCR during routine diagnostic testing with UKAS-accredited laboratory-developed assays that conform to ISO:15189 standards. In addition to the blood samples, four residual liver biopsies from four control patients referred for investigation of infection were tested by AAV2 and HHV-6B PCR. The liver biopsies were submitted to the GOSH microbiology laboratory for routine diagnosis by bacterial broad-range 16S rRNA gene PCR or metagenomics testing in 2021 and 2022. Three of four control patients were known to have elevated levels of liver enzymes. Two adult frozen liver samples previously tested by metagenomics were negative for AAV2 and positive for HHV-6B (Supplementary Table ). Controls from UKHSA We received a blood sample from one patient with elevated levels of liver enzymes and HAdV infection. We also received one control stool sample from Public Health Scotland/Edinburgh Royal Infirmary and 22 control stool samples for sequencing. Controls from King’s College Hospital A single FFPE liver biopsy control of normal marginal tissue from a hepatoblastoma from a child was negative for AAV2 and HAdV, but positive for HHV-6B (Ct = 37). Controls from Queen Mary University of London We received FFPE liver control samples from ten adults and three children (under 18 years of age) with other viral hepatitis, toxic liver necrosis, autoimmune and other hepatitis, and normal liver, from Queen Mary University of London. PCR gave valid results for samples from two children and eight adults, all of which were negative by PCR for AAV2 and HHV-6, apart from one adult sample, which was positive for HHV-6 at a high Ct value (Supplementary Table ). Metagenomic sequencing Nucleic acid purification Frozen liver biopsies were infused overnight at −20 °C with RNAlater-ICE. Up to 20 mg biopsy was lysed with 1.4-mm ceramic, 0.1-mm silica and 4-mm glass beads, before DNA and RNA purification using the Qiagen AllPrep DNA/RNA Mini kit as per the manufacturer’s instructions, with a 30 µl elution volume for RNA and 50 µl for DNA. Up to 400 µl whole blood was lysed with 0.5-mm and 0.1-mm glass beads before DNA and RNA purification on a Qiagen EZ1 instrument with an EZ1 virus mini kit as per the manufacturer’s instructions, with a 60 µl elution volume. For quality assurance, every batch of samples was accompanied by a control sample containing feline calicivirus RNA and cowpox DNA, which was processed alongside clinical specimens, from nucleic acid purification through to sequencing. All specimens and controls were spiked with MS2 phage RNA internal control before nucleic acid purification. Library preparation and sequencing RNA from whole-blood samples with an RNA yield of more than 2.5 ng µl −1 and from biopsies underwent ribosomal RNA depletion and library preparation with KAPA RNA HyperPrep kit with RiboErase, according to the manufacturer’s instructions. RNA from whole blood with an RNA yield of less than 2.5 ng µl −1 did not undergo rRNA depletion before library preparation. DNA from whole-blood samples with a DNA yield of more than 1 ng µl −1 and from biopsies underwent depletion of CpG-methylated DNA using the NEBNext Microbiome DNA Enrichment Kit, followed by library preparation with the NEBNext Ultra II FS DNA Library Prep Kit for Illumina, according to manufacturer’s instructions. DNA from whole blood with a DNA yield of less than 1 ng µl −1 did not undergo depletion of CpG-methylated DNA before library preparation. Sequencing was performed with a NextSeq High output 150 cycle kit with a maximum of 12 libraries pooled per run, including controls. Metagenomics data analysis Pre-processing pipeline An initial quality control step was performed by trimming adapters and low-quality ends from the reads (Trim Galore! 0.3.7). Human sequences were then removed using the human reference GRCH38 p.9 (Bowtie2 (ref. ), version 2.4.1) followed by removal of low-quality and low-complexity sequences (PrinSeq , version 0.20.3). An additional step of human sequences removal followed (megaBLAST , version 2.9.0). For RNA-seq, rRNA sequences were also removed using a similar two-step approach (Bowtie2 and megaBLAST). Finally, nucleotide similarity and protein similarity searches were performed (megaBLAST and DIAMOND (version 0.9.30), respectively) against custom reference databases that consisted of nucleotide and protein sequences of the RefSeq collections (downloaded March 2020) for viruses, bacteria, fungi, parasites and human. Taxonomic classification DNA and RNA sequence data were analysed with metaMix (version 0.4) nucleotide and protein analysis pipelines. metaMix resolves metagenomics mixtures using Bayesian mixture models and a parallel Markov chain Monte Carlo search of the potential species space to infer the most likely species profile. metaMix considers all reads simultaneously to infer relative abundances and probabilistically assign the reads to the species most likely to be present. It uses an ‘unknown’ category to capture the fact that some reads cannot be assigned to any species. The resulting metagenomic profile includes posterior probabilities of species presence as well as Bayes factor for presence versus absence of specific species. There are two modes: metaMix-protein, which is optimal for RNA virus detection, and metaMix-nucl, which is best for speciation of DNA microorganisms. Both modes were used for RNA-seq, whereas metaMix-nucl was used for DNA-seq. For sequence results to be valid, MS2 phage RNA had to be detected in every sample and feline calicivirus RNA and cowpox DNA, with no additional unexpected organisms, detected in the controls. Confirmatory mapping of AAV2 The RNA-seq reads were mapped to the AAV2 reference genome (NCBI reference sequence NC_001401 ) using Bowtie2, with the –very-sensitive option. Samtools (version 1.9) and Picard (version 2.26.9; http://broadinstitute.github.io/picard/ ) were used to sort, deduplicate and index the alignments, and to create a depth file, which was plotted using a custom script in R. De novo assembly of unclassified reads We performed a de novo assembly step with metaSPADES (v3.15.5), using all the reads with no matches to the nucleotide database that we used for our similarity search. A search using megaBLAST with the standard nucleotide collection was carried out on all resulting contigs over 1,000 bp in length. All of the contigs longer than 1,000 bp matched to human, except two that mapped to Torque Teno virus. Nanopore sequencing DNA from up to 20 mg of liver was purified using the Qiagen DNeasy Blood & Tissue kit as per the manufacturer’s instructions. Samples with limited amount of DNA were fragmented to an average size of 10 kb using a Megaruptor 3 (Diagenode) to reach an optimal molar concentration for library preparation. Quality control was perform using a Femto Pulse System (Agilent Technologies) and a Qubit fluorometer (Invitrogen). Samples were prepared for Nanopore sequencing using the ligation sequencing kit SQK-LSK110. DNA was sequenced on a PromethION using R9.4.1 flowcells (Oxford Nanopore Technologies). Samples were run for 72 h including a washing and reload step after 24 h and 48 h. All library preparation and sequencing were performed by the UCL Long Read Sequencing facility. Passed reads from Minknow were mapped to the reference AAV2 genome ( NC_001401 ) using minimap2 (ref. ) using the default parameters. Reads were trimmed of adapters using Porechop v0.2.4 ( https://github.com/rrwick/Porechop/ ), with the sequences of the adapters used added to adapters.py, and using an adapter threshold of 85. Reads that also mapped by minimap to the human genome (Ensemble GRCh38_v107), which could be ligation artefacts, were excluded from further analysis. The passed reads were also classified using Kraken2 (ref. ) with the PlusPF database (17 May 2021). The data relating to AAV2 reads in Supplementary Table refer to reads that were classified as AAV2 by both minimap2 and Kraken2 (version 2.0.8-beta), as the results from both methods were similar. Four reads across all four lower-depth samples were classified as HHV-6B by the EPI2ME WIMP pipeline. No reads were classified as HAdV or HHV-6B by Kraken2 in the two higher-depth samples. Alignment dot plots were created for the AAV2 reads using redotable (version 1.1) , with a window size of 20. These were manually classified into possible complex and monomeric structures. Integration analysis of Illumina data We investigated potential integrations of AAV2 and HHV-6 viruses into the genome using the Illumina metagenomics data for five liver transplant cases. We first processed the pair-end reads (average sequence coverage per genome = 5×), quality checking using FastQC , with barcode and adaptor sequence trimmed by TrimGalore (phred-score = 20). Potential viral integrations were investigated with Vseq-Toolkit (mode 3 with default settings except for high stringency levels). Predicted genomic integrations were visualized with IGV , requiring at least three reads supporting an integration site, spanning both human and viral sequences. Predicted integrations were supported by only one read, thus not fulfilling the algorithm criteria. Sequencing was performed at a lower depth than optimal for integration analysis, but no evidence was found for AAV2 or HHV-6B integration into the genomes of cases. PCR Real-time PCR targeting a 62-nt region of the AAV2 inverted terminal repeat sequence was performed using primers and probes previously described . This assay has been predicted to amplify AAV2 and AAV6. The Qiagen QuantiNova probe PCR kit (PERFORM and DIAMONDS controls) or the Qiagen Quantifast probe PCR kit (all other samples) were used. Each 25-µl reaction consisted of 0.1 µM forward primer, 0.34 µM reverse primer and 0.1 µM probe with 5 µl template DNA. Real-time PCR targeting a 74-bp region of the HHV-6 DNA polymerase gene was performed using primers and probes previously described multiplexed with an internal positive control targeting mouse ( mus ) DNA spiked into each sample during DNA purification, as previously described . In brief, each 25-µl reaction consisted of 0.5 µM of each primer, 0.3 µM HHV-6 probe, 0.12 µM of each mus primer, 0.08 µM mus probe and 12.5 µl Qiagen Quantifast Fast mastermix with 10 µl template DNA. Real-time PCR targeting a 132-bp region of the HAdV hexon gene was performed using primers and probes previously described multiplexed with an internal positive control targeting mouse ( mus ) DNA spiked into each sample during DNA purification, as previously described . In brief, each 25-µl reaction consisted of 0.6 µM of each HHV-6 primer, 0.4 µM HHV-6 probe, 0.12 µM of each mus primer, 0.08 µM mus probe and 12.5 µl Qiagen Quantifast Fast mastermix with 10 µl template DNA. PCR cycling for all targets, apart from the controls from the PERFORM and DIAMONDS studies, was performed on an ABI 7500 Fast thermocycler and consisted of 95 °C for 5 min followed by 45 cycles of 95 °C for 30 s and 60 °C for 30 s. For the PERFORM and DIAMONDS controls, PCR was performed on a StepOnePlus Real-Time PCR System and consisted of 95 °C for 2 min followed by 45 cycles of 95 °C for 5 s and 60 °C for 10 s. Each PCR run included a no template control and a DNA-positive control for each target. Neat DNA extracts of the FFPE material were inhibitory to PCR, so PCR results shown were performed following a 1 in 10 dilution. AAV2 quantitative PCR with reverse transcription RNA samples were treated with the Turbo-DNA free kit (Thermo) to remove residual genomic DNA. Complementary DNA (cDNA) was synthesized using the QuantiTect Reverse Transcription kit. In brief, 12 µl of RNA was mixed with 2 µl of genomic DNA Wipeout buffer and incubated at 42 °C for 2 min and transferred to ice. For reverse transcription, 6 µl mastermix was used and incubated at 42 °C for 20 min followed by 3 min at 95 °C. Real-time PCR targeting a 120-nt region of the AAV2 cap open reading frame sequence was performed using primers AAV2_cap _Fw- ATCCTTCGACCACCTTCAGT, AAV2_cap _Rv-GATT CCAGCGTTTGCTGTT and the probe AAV2_cap _Pr FAM-ACACAGTAT/ZEN/TCC ACGG GACAGGT-IBFQ. This assay has been predicted to amplify AAV2 and AAV6. The Qiagen QuantiNova probe PCR kit was used. Each 25-µl reaction consisted of 0.1 µM forward primer, 0.1 µM reverse primer and 0.2 µM probe with 2.5 µl template cDNA. PCR was performed on a StepOnePlus Real-Time PCR System and consisted of incubation at 95 °C for 2 min followed by 45 cycles of 95 °C for 5 s and 60 °C for 10 s. Each PCR run included a no template control, a DNA-positive control and a RNA control from each sample to verify efficient removal of genomic DNA. Immunohistochemistry All immunohistochemistry was done on FFPE tissue cut at a thickness of 3 µm. Adenovirus AdV immunohistochemistry was carried out using the Ventana Benchmark ULTRA, Optiview Detection Kit, PIER with protease 1 for 4 min and antibody incubation for 32 min (AdV clone 2/6 and 20/11, Roche, 760-4870, pre-diluted). The positive control was a known HAdV-positive gastrointestinal surgical case. Preparation of AAV2-positive controls The plasmid used for transfection was pAAV2/2 (addgene, plasmid #104963; https://www.addgene.org/104963/ ), which expresses the genes encoding Rep/Cap of AAV2. This was delivered by tail-vein hydrodynamic injection into albino C57BL/6 mice (5 mg in 2 ml PBS). Negative controls received PBS alone. At 48 h, mice were terminally exsanguinated and perfused by PBS. Livers were collected into 10% neutral buffered formalin (CellPath UK). This was performed under Home Office License PAD4E6357. AAV2 immunohistochemistry was carried out with four commercially available antibodies: Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 30 min and antibody incubation for 30 min (anti-AAV VP1/VP2/VP3 clone B1, PROGEN, 690058S, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1/VP2/VP3 rabbit polyclonal, OriGene, BP5024, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1 clone A1, OriGene, BM5013, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1/VP2 clone A69, OriGene, BM5014, 1:100). HHV-6 immunohistochemistry straining was carried out with: Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, PIER with Bond Enzyme 1 Kit for 10 min and antibody incubation for 30 min (mouse monoclonal antibody (C3108-103) to HHV-6, ABCAM, ab128404, 1:100). Negative reagent control slides were stained using the same antigen retrieval conditions and staining protocol incubation times using only BondTM Primary Antibody Diluent #AR9352 for the antibody incubation. Electron microscopy Samples of liver were fixed in 2.5% glutaraldehyde in 0.1 M cacodylate buffer followed by secondary fixation in 1.0% osmium tetroxide. Tissues were dehydrated in graded ethanol, transferred to an intermediate reagent, propylene oxide and then infiltrated and embedded in Agar 100 epoxy resin. Polymerization was undertaken at 60 °C for 48 h. Ultrathin sections of 90 nm were cut using a Diatome diamond knife on a Leica UC7 ultramicrotome. Sections were transferred to copper grids and stained with alcoholic urynal acetate and Reynold’s lead citrate. The samples were examined using a JEOL 1400 transmission electron microscope. Images were captured on an AMT XR80 digital camera. WGS Bait design To produce the capture probes for hybridization, biotinylated RNA oligonucleotides (baits) used in the SureSelectXT protocols for HAdV and HHV-6 WGS were designed in-house using Agilent community design baits with part numbers 5191-6711 and 5191-6713, respectively. They were synthesized by Agilent Technologies (2021) (available through Agilent’s Community Designs programme: SSXT CD Pan Adenovirus and SSXT CD Pan HHV-6 and used previously , ). Library preparation and sequencing For WGS of HAdV and HHV-6B, DNA (bulked with male human genomic DNA (Promega) if required) was sheared using a Covaris E220 focused ultrasonication system (PIP 75, duty factor of 10, 1,000 cycles per burst). End-repair, non-templated addition of 3′ poly A, adapter ligation, hybridization, PCR (pre-capture cycles dependent on DNA input and post-capture cycles dependent on viral load) and all post-reaction clean-up steps were performed according to either the SureSelectXT Low Input Target Enrichment for Illumina Paired-End Multiplexed Sequencing protocol (version A0), the SureSelectXT Target Enrichment for Illumina Paired-End Multiplexed Sequencing protocol (version C3) or the SureSelectXTHS Target Enrichment using the Magnis NGS Prep System protocol (version A0) (Agilent Technologies). Quality control steps were performed on the 4200 TapeStation (Agilent Technologies). Samples were sequenced using the Illumina MiSeq platform. Base calling and sample demultiplexing were performed as standard for the MiSeq platform, generating paired FASTQ files for each sample. A negative control was included on each processing run. A targeted enrichment approach was used due to the predicted high variability of the HHV-6 and HAdV genomes. For AAV2 WGS, an AAV2 primer scheme was designed using primalscheme with 17 AAV2 sequences from NCBI and one AAV2 sequence provided by GOSH from metagenomic sequencing of a liver biopsy DNA extract as the reference material. These primers amplify 15 overlapping 400-bp amplicons. Primers were supplied by Merck. Two multiplex PCRs were prepared using Q5 Hot Start High-Fidelity 2X Master Mix, with a 65 °C, 3 min annealing/extension temperature. Pools 1 and 2 multiplex PCRs were run for 35 cycles. Of each PCR, 10 µl was combined and 20 µl nuclease-free water was added. Libraries were prepared either manually or on the Agilent Bravo NGS workstation option B, following a reduced-scale version of the Illumina DNA protocol as used in the CoronaHiT protocol . Equal volumes of the final libraries were pooled, bead purified and sequenced on the Illumina MiSeq. A negative control was included on each processing run. All library preparation and sequencing were performed by UCL Genomics. AAV2 sequence analysis The raw fastq reads were adapted, trimmed and low-quality reads were removed. The reads were mapped to the NC_001401 reference sequence and then the amplicon primers regions were trimmed using the location provided in a bed file. Consensus sequences were then called at a minimum of 10× coverage. The entire processing of raw reads to consensus was carried out using the nf-core/viralrecon pipeline ( https://nf-co.re/viralrecon/2.4.1 ; 10.5281/zenodo.3901628). Basic quality metrics for the samples sequenced are in Supplementary Table . All samples that gave 10× genome coverage over 90% were then used for further phylogenetic analysis. Samples were aligned along with known reference strains from GenBank using MAFFT (version v7.271), and the trees were built with IQ-TREE (multicore version 1.6.12) with 1,000 rapid bootstraps and approximate likelihood-ratio test support. The samples were then labelled based on type and provider on the trees (Fig. ). For each AAV2 sample, we aligned the consensus nucleotide sequence to the AAV2 reference sequence. From these alignments, the exact coordinates of the sample capsid were determined. We then used the coordinates to extract the corresponding nucleotide sequence and translated it to find the amino acid sequence. Next, we compared each sample to the reference to identify amino acid changes. Amino acid sequences from AAV capsid sequences were retrieved from GenBank for AAV1 to AAV12. Amino acid sequences of capsid constructs designed to be more hepatotropic were retrieved from refs. , . These sequence sets were then aligned to the AAV2 reference sequence using MAFFT . We then compared each construct to the AAV2 reference to identify amino acid changes present, while retaining the AAV2 coordinate set. HAdV and HHV-6B sequence analysis Raw data quality control was performed using trim-galore (v.0.6.7) on the raw FASTQ files. For HHV-6B, short reads were mapped with BWA mem (0.7.17-r1188) using the RefSeq reference NC_000898 . For HAdV, genotyping is performed using AYUKA (version 22-111). This novel tool is used to confidently assign one or more HAdV genotypes to a sample of interest, assessing inter-genotype recombination if more than one genotype is detected. The results from this screening step guide which downstream analyses are performed and which reference genome (or genomes) is used. If mixed infection is suspected, reads are separated using bbsplit ( https://sourceforge.net/projects/bbmap/ ), and each genotype is analysed independently as normal. If recombination is suspected, a more detailed analysis is performed using Recombination Detection Program (RDP) and the sample is excluded from phylogenetic analysis. After genotyping, the cleaned read data are mapped using BWA to the relevant reference sequence (or sequences), and SNPs and small insertions and deletions are called using bcftool (version1.15.1, https://github.com/samtools/bcftools ) and a consensus sequence is generated also with bcftools, masking with Ns positions that do not have enough read support (15× by default). Consensus sequences generated with the pipeline are then concatenated to previously sequenced samples and a multiple sequence alignment is performed using the G-INS-I algorithm in the MAFFT software (MAFFT G-INS-I v7.481). The multiple sequence alignment is then used for phylogenetic analysis with IQ-TREE (IQ-TREE 2 2.2.0), using modelfinder and performing 1,000 rapid bootstraps. Proteomics data generation Liver explant tissue from cases was homogenized in lysis buffer, 100 mM Tris (pH 8.5), 5% sodium dodecyl sulfate, 5 mM tris(2-carboxyethyl)phosphine and 20 mM chloroacetamide then heated at 95 °C for 10 min and sonicated in an ultrasonic bath for another 10 min. The lysed proteins were quantified with NanoDrop 2000 (Thermo Fisher Scientific). One-hundred micrograms was precipitated with the methanol/chloroform protocol and then protein pellets were reconstituted in 100 mM Tris (pH 8.5) and 4% sodium deoxycholate (SDC). The proteins were subjected to proteolysis with 1:50 trypsin overnight at 37 °C with constant shaking. Digestion was stopped by adding 1% trifluoroacetic acid to a final concentration of 0.5%. Precipitated SDC was removed by centrifugation at 10,000 g for 5 min, and the supernatant containing digested peptides was desalted on an SOLAµ HRP (Thermo Fisher Scientific). Of the desalted peptide, 50 µg was then fractionated on Vanquish HPLC (Thermo Fisher Scientific) using a Acquity BEH C18 column (2.1 × 50 mm with 1.7-µm particles from Waters): buffer A was 10 mM ammonium formiate at pH 10, whereas buffer B was 80% acetonitrile and the flow was set to 500 µl per minute. We used a gradient of 8 min to collect 24 fractions that were then concatenated to obtain 12 fractions. These 12 fractions were dried and dissolved in 2% formic acid before liquid chromatography–tandem mass spectrometry analysis. An estimated total of 2,000 ng from each fraction was analysed using an Ultimate3000 high-performance liquid chromatography system coupled online to an Eclipse mass spectrometer (Thermo Fisher Scientific). Buffer A consisted of water acidified with 0.1% formic acid, whereas buffer B was 80% acetonitrile and 20% water with 0.1% formic acid. The peptides were first trapped for 1 min at 30 μl per minute with 100% buffer A on a trap (0.3 mm × 5 mm with PepMap C18, 5 μm, 100 Å; Thermo Fisher Scientific); after trapping, the peptides were separated by a 50-cm analytical column (Acclaim PepMap, 3 μm; Thermo Fisher Scientific). The gradient was 9–35% buffer B for 103 min at 300 nl per minute. Buffer B was then raised to 55% in 2 min and increased to 99% for the cleaning step. Peptides were ionized using a spray voltage of 2.1 kV and a capillary heated at 280 °C. The mass spectrometer was set to acquire full-scan mass spectrometry spectra (350:1,400 mass:charge ratio) for a maximum injection time set to auto at a mass resolution of 120,000 and an automated gain control target value of 100%. For a second, the most intense precursor ions were selected for tandem mass spectrometry. Higher energy collisional dissocation (HCD) fragmentation was performed in the HCD cell, with the readout in the Orbitrap mass analyser at a resolution of 15,000 (isolation window of 3 Th) and an automated gain control target value of 200% with a maximum injection time set to auto and a normalized collision energy of 30%. All raw files were analysed by MaxQuant v2.1 software using the integrated Andromeda search engine and searched against the Human UniProt Reference Proteome (February release with 79,057 protein sequences) together with UniProt-reported AAV proteins and specific fasta created using EMBOSS Sixpack translating patient’s virus genome. MaxQuant was used with the standard parameters with only the addition of deamidation (N) as variable modification. Data analysis was then carried out with Perseus v2.05: proteins reported in the file ‘proteinGroups.txt’ were filtered for reverse and potential contaminants. Figures were created using Origin pro version 2022b. Transduction of AAV2 capsid mutants A transgene sequence containing enhanced green fluorescent protein (eGFP) was packaged into rAAV2 particles to track their expression in transduced cells, compared with rAAV capsids derived from canonical AAV2, AAV9 and a synthetic liver-tropic AAV vector called LK03 (ref. ). rAAV vector particles were delivered to Huh-7 hepatocytes at a multiplicity of infection of 100,000 vector genomes per cell before analysing eGFP expression by flow cytometry 72 h later. Recombinant AAV capsid sequence The VP1 sequence was generated by generating a consensus sequence from a multiple sequence alignment of sequenced AAV2 genomes derived from patient samples, using the Biopython package AlignIO. The designed VP1 sequence was then synthesized as a ‘gBlock’ (Integrated DNA Technologies) and incorporated into an AAV2 RepCap plasmid (AAV2/2 was a gift from M. Fan, Addgene plasmid #104963) between the SwaI and XmaI restriction sites, using InFusion cloning reagent (product 638948, Clontech). AAV vector production rAAV particles were generated by transient transfection of HEK 293T cells as previously described . In brief, 1.8 × 10 7 cells were plated in 15-cm dishes before transfecting the pAAV-CAG-eGFP transgene plasmid (a gift from E. Boyden, Addgene plasmid #37825), the relevant RepCap plasmid and the pAdDeltaF6 helper plasmid (a gift from J. M. Wilson, Addgene plasmid #112867), at a ratio of 10.5 µg, 10.5 µg and 30.5 µg, respectively, using PEIPro transfection reagent (PolyPlus) at a ratio of 1 µl per 1 µg DNA. Seventy-two hours post-transfection, cell pellets and supernatant were harvested and rAAV particles were purified using an Akta HPLC platform. rAAV particle genome copy numbers were calculated by quantitative PCR targeting the vector transgene region. The rAAV2 vector used in this study was purchased as ready-to-use AAV2 particles from Addgene (Addgene viral prep #37825-AAV2). Analysis of rAAV transduction Huh-7 hepatocytes (a gift from J. Baruteau, UCL) were plated in DMEM medium supplemented with 10% FBS and 1% penicillin–streptomycin supplement. The cell line was validated by testing for glypican-3 and was not tested for mycoplasma contamination. Cells were plated at a density of 1.5 × 10 3 cells per square centimetre and transduced with 1 × 10 5 viral genomes per cell. Transductions were performed in the presence or absence of 400 µg ml −1 heparin, which was supplemented directly to cell media. Seventy-two hours after transduction, cells were analysed by microscopy using an EVOS Cell Imaging System (Thermo Fisher Scientific) before quantifying eGFP expression by flow cytometry using a Cytoflex Flow Cytometer (Beckman). eGFP-positive cells were determined by gating the live-cell population and quantifying the level of eGFP signal versus untransduced controls. Human short-read data analysis Cytokine transcriptomics analysis Cytokine inducible gene expression modules were derived from previously published bulk tissue genome-wide transcriptomes of the tuberculin skin test that have been shown to reflect canonical human in vivo cell-mediated immune pathways using a validated bioinformatic approach . Cytokine regulators of genes enriched in the tuberculin skin test (ArrayExpress accession number E-MTAB-6816 ) were identified using Ingenuity Pathway Analysis (Qiagen). Average correlation of log 2 -transformed transcripts per million data for every gene pair in each of the target gene modules were compared with 100 iterations of randomly selected gene modules of the same size, to select cytokine-inducible modules that showed significantly greater co-correlation (adjusted P < 0.05), representing co-regulated transcriptional networks for each 59 cytokines. We then used the average log 2 -transformed transcripts per million expression of all the genes in each of these co-regulated modules to quantify the biological activity of the associated upstream cytokine within bulk genome-wide transcriptional profiles from AAV2-associated hepatitis ( n = 4) obtained in the present study, compared with published log 2 -transformed and normalized microarray data from normal adult liver ( n = 10) and hepatitis B adult liver ( n = 17) (Gene Expression Omnibus accession number GSE96851 ) . To enable comparison across the datasets, we transformed average gene expression values for each cytokine-inducible module to standardized ( Z scores) using mean and standard deviation of randomly selected gene sets of the same size within each individual dataset. Statistically significant differences in Z scores between groups were identified by Student’s t -tests with multiple testing correction (adjusted P < 0.05). Proteomics differential expression To compare the proteomics data from the explanted livers of cases with data from healthy livers, we downloaded the raw files from two studies , from PRIDE. The raw files were searched together with our files using the same settings and databases. We performed differential expression analyses at the protein level and peptide level using a hybrid approach including statistical inference on the abundance (quantitative approach), as well as the presence or absence (binary approach) of proteins or peptides. DEP R package version 1.18.0 was used for quantitative analysis . Proteins or peptides were filtered for those detected in all replicates of at least one group (case or control). The data were background corrected and variance was normalized using variance-stabilizing transformation. Missing intensity values were not distributed randomly and were biased to specific samples (either cases or controls). Therefore, for imputing the missing data, we applied random draws from a manually defined left-shifted Gaussian distribution using the DEP impute function with parameters fun:“man”, shift:1.8 and scale:0.3. The test_diff function based on linear models and the empirical Bayes method was used for testing differential expressions between the case and control samples. HLA typing methods Typing was undertaken in the liver centre units. Next-generation sequencing (sequencing by synthesis (Illumina) using AllType kits (VHBio/OneLambda), a high-resolution HLA typing method, was used. Statistical analysis Fisher’s exact test and two-sided Wilcoxon (Mann–Whitney) non-parametric rank sum test were used for differences between case and control groups. Where multiple groups were compared, Kruskal–Wallis tests followed by Wilcoxon pairwise tests using a Benjamini–Hochberg correction were performed. All analysis were performed in R version 4.2.0. Reporting summary Further information on research design is available in the linked to this article. Metagenomic analysis and HAdV sequencing were carried out by the routine diagnostic service at Great Ormond Street Hospital (GOSH). Additional PCRs, immunohistochemistry and proteomics on samples received for metagenomics are part of the GOSH protocol for confirmation of new and unexpected pathogens. The use for research of anonymized laboratory request data, diagnostic results and residual material from any specimen received in the GOSH diagnostic laboratory, including all cases received from Birmingham’s Children Hospital UKHSA, Public Health Wales, Public Health Scotland as well as non-case samples from UKHSA, Public Health Scotland and GOSH research was approved by UCL Partners Pathogen Biobank under ethical approval granted by the NRES Committee London-Fulham (REC reference: 17/LO/1530). Children undergoing liver transplant were consented for additional research under the International Severe Acute Respiratory and Emerging Infection Consortium (ISARIC) WHO Clinical Characterisation Protocol UK (CCP-UK) (ISRCTN 66726260) (RQ3001-0591, RQ301-0594, RQ301-0596, RQ301-0597 and RQ301-0598). Ethical approval for the ISARIC CCP-UK study was given by the South Central–Oxford Research Ethics Committee in England (13/SC/0149), the Scotland A Research Ethics Committee (20/SS/0028) and the WHO Ethics Review Committee (RPC571 and RPC572). The UKHSA has legal permission, provided by regulation 3 of The Health Service (Control of Patient Information) Regulations 2002, to process patient confidential information for national surveillance of communicable diseases and, as such, individual patient consent is not required. Control participants from the EU Horizon 2020 research and innovation program DIAMONDS–PERFORM (grant agreement nos. 668303 and 848196) were recruited according to the approved enrolment procedures of each study, and with the informed consent of parents or guardians: DIAMONDS (London-Dulwich Research Ethics Committee: 20/HRA/1714) and PERFORM (London-Central Research Ethics Committee: 16/LO/1684). The sample IDs for the cases and controls are anonymized IDs that cannot reveal the identity of the study participants and are not known to anyone outside the research group, such as the patients or the hospital staff. Initial diagnostic testing by metagenomics and PCR was performed at GOSH Microbiology and Virology clinical laboratories. Further WGS and characterization were performed at UCL. Birmingham Children’s Hospital provided us with explanted liver tissue from five biopsy sites from five cases, five whole blood 500 µl from four cases and serum plasma from one case (Table and Fig. ). These were used in metagenomics testing (Table ), followed by HAdV, HHV-6 and AAV2 testing by PCR and, depending on the Ct value, WGS (Supplementary Tables , and ). We subsequently received 25 additional specimens from UKHSA, Public Health Wales and Public Health Scotland/Edinburgh Royal Infirmary, including 16 additional blood samples, four respiratory specimens and five stool samples, for HAdV WGS and, depending on residual material, for AAV2 PCR testing followed by sequencing (Tables and , Fig. and Supplementary Tables , and ). We also received ten FFPE liver biopsy samples and six serum samples from 11 cases from King’s College Hospital (Table ). Of these cases, seven had received liver transplants. PERFORM recruited children from ten EU countries (2016–2020). PERFORM was funded by the European Union’s Horizon 2020 programme under GA no. 668303. DIAMONDS is funded by the European Union Horizon 2020 programme grant number 848196. Recruitment commenced in 2020 and is ongoing. Both studies recruited children presenting with suspected infection or inflammation and assigned them to diagnostic groups according to a standardized algorithm. Blood samples from 17 patients not linked to the non-A–E hepatitis outbreak were tested by real-time PCR targeting AAV2 (Extended Data Table ). These comparators were patients with ALT/AST of more than 500 and HAdV or cytomegalovirus viraemia. These were purified DNA from residual diagnostic specimens received in the GOSH microbiology and virology laboratory in the previous year. All residual specimens were stored at −80 °C before testing and pseudo-anonymized at the point of processing and analysis. Viraemia was initially detected using targeted real-time PCR during routine diagnostic testing with UKAS-accredited laboratory-developed assays that conform to ISO:15189 standards. In addition to the blood samples, four residual liver biopsies from four control patients referred for investigation of infection were tested by AAV2 and HHV-6B PCR. The liver biopsies were submitted to the GOSH microbiology laboratory for routine diagnosis by bacterial broad-range 16S rRNA gene PCR or metagenomics testing in 2021 and 2022. Three of four control patients were known to have elevated levels of liver enzymes. Two adult frozen liver samples previously tested by metagenomics were negative for AAV2 and positive for HHV-6B (Supplementary Table ). We received a blood sample from one patient with elevated levels of liver enzymes and HAdV infection. We also received one control stool sample from Public Health Scotland/Edinburgh Royal Infirmary and 22 control stool samples for sequencing. A single FFPE liver biopsy control of normal marginal tissue from a hepatoblastoma from a child was negative for AAV2 and HAdV, but positive for HHV-6B (Ct = 37). We received FFPE liver control samples from ten adults and three children (under 18 years of age) with other viral hepatitis, toxic liver necrosis, autoimmune and other hepatitis, and normal liver, from Queen Mary University of London. PCR gave valid results for samples from two children and eight adults, all of which were negative by PCR for AAV2 and HHV-6, apart from one adult sample, which was positive for HHV-6 at a high Ct value (Supplementary Table ). Nucleic acid purification Frozen liver biopsies were infused overnight at −20 °C with RNAlater-ICE. Up to 20 mg biopsy was lysed with 1.4-mm ceramic, 0.1-mm silica and 4-mm glass beads, before DNA and RNA purification using the Qiagen AllPrep DNA/RNA Mini kit as per the manufacturer’s instructions, with a 30 µl elution volume for RNA and 50 µl for DNA. Up to 400 µl whole blood was lysed with 0.5-mm and 0.1-mm glass beads before DNA and RNA purification on a Qiagen EZ1 instrument with an EZ1 virus mini kit as per the manufacturer’s instructions, with a 60 µl elution volume. For quality assurance, every batch of samples was accompanied by a control sample containing feline calicivirus RNA and cowpox DNA, which was processed alongside clinical specimens, from nucleic acid purification through to sequencing. All specimens and controls were spiked with MS2 phage RNA internal control before nucleic acid purification. Library preparation and sequencing RNA from whole-blood samples with an RNA yield of more than 2.5 ng µl −1 and from biopsies underwent ribosomal RNA depletion and library preparation with KAPA RNA HyperPrep kit with RiboErase, according to the manufacturer’s instructions. RNA from whole blood with an RNA yield of less than 2.5 ng µl −1 did not undergo rRNA depletion before library preparation. DNA from whole-blood samples with a DNA yield of more than 1 ng µl −1 and from biopsies underwent depletion of CpG-methylated DNA using the NEBNext Microbiome DNA Enrichment Kit, followed by library preparation with the NEBNext Ultra II FS DNA Library Prep Kit for Illumina, according to manufacturer’s instructions. DNA from whole blood with a DNA yield of less than 1 ng µl −1 did not undergo depletion of CpG-methylated DNA before library preparation. Sequencing was performed with a NextSeq High output 150 cycle kit with a maximum of 12 libraries pooled per run, including controls. Frozen liver biopsies were infused overnight at −20 °C with RNAlater-ICE. Up to 20 mg biopsy was lysed with 1.4-mm ceramic, 0.1-mm silica and 4-mm glass beads, before DNA and RNA purification using the Qiagen AllPrep DNA/RNA Mini kit as per the manufacturer’s instructions, with a 30 µl elution volume for RNA and 50 µl for DNA. Up to 400 µl whole blood was lysed with 0.5-mm and 0.1-mm glass beads before DNA and RNA purification on a Qiagen EZ1 instrument with an EZ1 virus mini kit as per the manufacturer’s instructions, with a 60 µl elution volume. For quality assurance, every batch of samples was accompanied by a control sample containing feline calicivirus RNA and cowpox DNA, which was processed alongside clinical specimens, from nucleic acid purification through to sequencing. All specimens and controls were spiked with MS2 phage RNA internal control before nucleic acid purification. RNA from whole-blood samples with an RNA yield of more than 2.5 ng µl −1 and from biopsies underwent ribosomal RNA depletion and library preparation with KAPA RNA HyperPrep kit with RiboErase, according to the manufacturer’s instructions. RNA from whole blood with an RNA yield of less than 2.5 ng µl −1 did not undergo rRNA depletion before library preparation. DNA from whole-blood samples with a DNA yield of more than 1 ng µl −1 and from biopsies underwent depletion of CpG-methylated DNA using the NEBNext Microbiome DNA Enrichment Kit, followed by library preparation with the NEBNext Ultra II FS DNA Library Prep Kit for Illumina, according to manufacturer’s instructions. DNA from whole blood with a DNA yield of less than 1 ng µl −1 did not undergo depletion of CpG-methylated DNA before library preparation. Sequencing was performed with a NextSeq High output 150 cycle kit with a maximum of 12 libraries pooled per run, including controls. Pre-processing pipeline An initial quality control step was performed by trimming adapters and low-quality ends from the reads (Trim Galore! 0.3.7). Human sequences were then removed using the human reference GRCH38 p.9 (Bowtie2 (ref. ), version 2.4.1) followed by removal of low-quality and low-complexity sequences (PrinSeq , version 0.20.3). An additional step of human sequences removal followed (megaBLAST , version 2.9.0). For RNA-seq, rRNA sequences were also removed using a similar two-step approach (Bowtie2 and megaBLAST). Finally, nucleotide similarity and protein similarity searches were performed (megaBLAST and DIAMOND (version 0.9.30), respectively) against custom reference databases that consisted of nucleotide and protein sequences of the RefSeq collections (downloaded March 2020) for viruses, bacteria, fungi, parasites and human. Taxonomic classification DNA and RNA sequence data were analysed with metaMix (version 0.4) nucleotide and protein analysis pipelines. metaMix resolves metagenomics mixtures using Bayesian mixture models and a parallel Markov chain Monte Carlo search of the potential species space to infer the most likely species profile. metaMix considers all reads simultaneously to infer relative abundances and probabilistically assign the reads to the species most likely to be present. It uses an ‘unknown’ category to capture the fact that some reads cannot be assigned to any species. The resulting metagenomic profile includes posterior probabilities of species presence as well as Bayes factor for presence versus absence of specific species. There are two modes: metaMix-protein, which is optimal for RNA virus detection, and metaMix-nucl, which is best for speciation of DNA microorganisms. Both modes were used for RNA-seq, whereas metaMix-nucl was used for DNA-seq. For sequence results to be valid, MS2 phage RNA had to be detected in every sample and feline calicivirus RNA and cowpox DNA, with no additional unexpected organisms, detected in the controls. Confirmatory mapping of AAV2 The RNA-seq reads were mapped to the AAV2 reference genome (NCBI reference sequence NC_001401 ) using Bowtie2, with the –very-sensitive option. Samtools (version 1.9) and Picard (version 2.26.9; http://broadinstitute.github.io/picard/ ) were used to sort, deduplicate and index the alignments, and to create a depth file, which was plotted using a custom script in R. De novo assembly of unclassified reads We performed a de novo assembly step with metaSPADES (v3.15.5), using all the reads with no matches to the nucleotide database that we used for our similarity search. A search using megaBLAST with the standard nucleotide collection was carried out on all resulting contigs over 1,000 bp in length. All of the contigs longer than 1,000 bp matched to human, except two that mapped to Torque Teno virus. Nanopore sequencing DNA from up to 20 mg of liver was purified using the Qiagen DNeasy Blood & Tissue kit as per the manufacturer’s instructions. Samples with limited amount of DNA were fragmented to an average size of 10 kb using a Megaruptor 3 (Diagenode) to reach an optimal molar concentration for library preparation. Quality control was perform using a Femto Pulse System (Agilent Technologies) and a Qubit fluorometer (Invitrogen). Samples were prepared for Nanopore sequencing using the ligation sequencing kit SQK-LSK110. DNA was sequenced on a PromethION using R9.4.1 flowcells (Oxford Nanopore Technologies). Samples were run for 72 h including a washing and reload step after 24 h and 48 h. All library preparation and sequencing were performed by the UCL Long Read Sequencing facility. Passed reads from Minknow were mapped to the reference AAV2 genome ( NC_001401 ) using minimap2 (ref. ) using the default parameters. Reads were trimmed of adapters using Porechop v0.2.4 ( https://github.com/rrwick/Porechop/ ), with the sequences of the adapters used added to adapters.py, and using an adapter threshold of 85. Reads that also mapped by minimap to the human genome (Ensemble GRCh38_v107), which could be ligation artefacts, were excluded from further analysis. The passed reads were also classified using Kraken2 (ref. ) with the PlusPF database (17 May 2021). The data relating to AAV2 reads in Supplementary Table refer to reads that were classified as AAV2 by both minimap2 and Kraken2 (version 2.0.8-beta), as the results from both methods were similar. Four reads across all four lower-depth samples were classified as HHV-6B by the EPI2ME WIMP pipeline. No reads were classified as HAdV or HHV-6B by Kraken2 in the two higher-depth samples. Alignment dot plots were created for the AAV2 reads using redotable (version 1.1) , with a window size of 20. These were manually classified into possible complex and monomeric structures. Integration analysis of Illumina data We investigated potential integrations of AAV2 and HHV-6 viruses into the genome using the Illumina metagenomics data for five liver transplant cases. We first processed the pair-end reads (average sequence coverage per genome = 5×), quality checking using FastQC , with barcode and adaptor sequence trimmed by TrimGalore (phred-score = 20). Potential viral integrations were investigated with Vseq-Toolkit (mode 3 with default settings except for high stringency levels). Predicted genomic integrations were visualized with IGV , requiring at least three reads supporting an integration site, spanning both human and viral sequences. Predicted integrations were supported by only one read, thus not fulfilling the algorithm criteria. Sequencing was performed at a lower depth than optimal for integration analysis, but no evidence was found for AAV2 or HHV-6B integration into the genomes of cases. PCR Real-time PCR targeting a 62-nt region of the AAV2 inverted terminal repeat sequence was performed using primers and probes previously described . This assay has been predicted to amplify AAV2 and AAV6. The Qiagen QuantiNova probe PCR kit (PERFORM and DIAMONDS controls) or the Qiagen Quantifast probe PCR kit (all other samples) were used. Each 25-µl reaction consisted of 0.1 µM forward primer, 0.34 µM reverse primer and 0.1 µM probe with 5 µl template DNA. Real-time PCR targeting a 74-bp region of the HHV-6 DNA polymerase gene was performed using primers and probes previously described multiplexed with an internal positive control targeting mouse ( mus ) DNA spiked into each sample during DNA purification, as previously described . In brief, each 25-µl reaction consisted of 0.5 µM of each primer, 0.3 µM HHV-6 probe, 0.12 µM of each mus primer, 0.08 µM mus probe and 12.5 µl Qiagen Quantifast Fast mastermix with 10 µl template DNA. Real-time PCR targeting a 132-bp region of the HAdV hexon gene was performed using primers and probes previously described multiplexed with an internal positive control targeting mouse ( mus ) DNA spiked into each sample during DNA purification, as previously described . In brief, each 25-µl reaction consisted of 0.6 µM of each HHV-6 primer, 0.4 µM HHV-6 probe, 0.12 µM of each mus primer, 0.08 µM mus probe and 12.5 µl Qiagen Quantifast Fast mastermix with 10 µl template DNA. PCR cycling for all targets, apart from the controls from the PERFORM and DIAMONDS studies, was performed on an ABI 7500 Fast thermocycler and consisted of 95 °C for 5 min followed by 45 cycles of 95 °C for 30 s and 60 °C for 30 s. For the PERFORM and DIAMONDS controls, PCR was performed on a StepOnePlus Real-Time PCR System and consisted of 95 °C for 2 min followed by 45 cycles of 95 °C for 5 s and 60 °C for 10 s. Each PCR run included a no template control and a DNA-positive control for each target. Neat DNA extracts of the FFPE material were inhibitory to PCR, so PCR results shown were performed following a 1 in 10 dilution. AAV2 quantitative PCR with reverse transcription RNA samples were treated with the Turbo-DNA free kit (Thermo) to remove residual genomic DNA. Complementary DNA (cDNA) was synthesized using the QuantiTect Reverse Transcription kit. In brief, 12 µl of RNA was mixed with 2 µl of genomic DNA Wipeout buffer and incubated at 42 °C for 2 min and transferred to ice. For reverse transcription, 6 µl mastermix was used and incubated at 42 °C for 20 min followed by 3 min at 95 °C. Real-time PCR targeting a 120-nt region of the AAV2 cap open reading frame sequence was performed using primers AAV2_cap _Fw- ATCCTTCGACCACCTTCAGT, AAV2_cap _Rv-GATT CCAGCGTTTGCTGTT and the probe AAV2_cap _Pr FAM-ACACAGTAT/ZEN/TCC ACGG GACAGGT-IBFQ. This assay has been predicted to amplify AAV2 and AAV6. The Qiagen QuantiNova probe PCR kit was used. Each 25-µl reaction consisted of 0.1 µM forward primer, 0.1 µM reverse primer and 0.2 µM probe with 2.5 µl template cDNA. PCR was performed on a StepOnePlus Real-Time PCR System and consisted of incubation at 95 °C for 2 min followed by 45 cycles of 95 °C for 5 s and 60 °C for 10 s. Each PCR run included a no template control, a DNA-positive control and a RNA control from each sample to verify efficient removal of genomic DNA. Immunohistochemistry All immunohistochemistry was done on FFPE tissue cut at a thickness of 3 µm. Adenovirus AdV immunohistochemistry was carried out using the Ventana Benchmark ULTRA, Optiview Detection Kit, PIER with protease 1 for 4 min and antibody incubation for 32 min (AdV clone 2/6 and 20/11, Roche, 760-4870, pre-diluted). The positive control was a known HAdV-positive gastrointestinal surgical case. Preparation of AAV2-positive controls The plasmid used for transfection was pAAV2/2 (addgene, plasmid #104963; https://www.addgene.org/104963/ ), which expresses the genes encoding Rep/Cap of AAV2. This was delivered by tail-vein hydrodynamic injection into albino C57BL/6 mice (5 mg in 2 ml PBS). Negative controls received PBS alone. At 48 h, mice were terminally exsanguinated and perfused by PBS. Livers were collected into 10% neutral buffered formalin (CellPath UK). This was performed under Home Office License PAD4E6357. AAV2 immunohistochemistry was carried out with four commercially available antibodies: Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 30 min and antibody incubation for 30 min (anti-AAV VP1/VP2/VP3 clone B1, PROGEN, 690058S, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1/VP2/VP3 rabbit polyclonal, OriGene, BP5024, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1 clone A1, OriGene, BM5013, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1/VP2 clone A69, OriGene, BM5014, 1:100). HHV-6 immunohistochemistry straining was carried out with: Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, PIER with Bond Enzyme 1 Kit for 10 min and antibody incubation for 30 min (mouse monoclonal antibody (C3108-103) to HHV-6, ABCAM, ab128404, 1:100). Negative reagent control slides were stained using the same antigen retrieval conditions and staining protocol incubation times using only BondTM Primary Antibody Diluent #AR9352 for the antibody incubation. Electron microscopy Samples of liver were fixed in 2.5% glutaraldehyde in 0.1 M cacodylate buffer followed by secondary fixation in 1.0% osmium tetroxide. Tissues were dehydrated in graded ethanol, transferred to an intermediate reagent, propylene oxide and then infiltrated and embedded in Agar 100 epoxy resin. Polymerization was undertaken at 60 °C for 48 h. Ultrathin sections of 90 nm were cut using a Diatome diamond knife on a Leica UC7 ultramicrotome. Sections were transferred to copper grids and stained with alcoholic urynal acetate and Reynold’s lead citrate. The samples were examined using a JEOL 1400 transmission electron microscope. Images were captured on an AMT XR80 digital camera. An initial quality control step was performed by trimming adapters and low-quality ends from the reads (Trim Galore! 0.3.7). Human sequences were then removed using the human reference GRCH38 p.9 (Bowtie2 (ref. ), version 2.4.1) followed by removal of low-quality and low-complexity sequences (PrinSeq , version 0.20.3). An additional step of human sequences removal followed (megaBLAST , version 2.9.0). For RNA-seq, rRNA sequences were also removed using a similar two-step approach (Bowtie2 and megaBLAST). Finally, nucleotide similarity and protein similarity searches were performed (megaBLAST and DIAMOND (version 0.9.30), respectively) against custom reference databases that consisted of nucleotide and protein sequences of the RefSeq collections (downloaded March 2020) for viruses, bacteria, fungi, parasites and human. DNA and RNA sequence data were analysed with metaMix (version 0.4) nucleotide and protein analysis pipelines. metaMix resolves metagenomics mixtures using Bayesian mixture models and a parallel Markov chain Monte Carlo search of the potential species space to infer the most likely species profile. metaMix considers all reads simultaneously to infer relative abundances and probabilistically assign the reads to the species most likely to be present. It uses an ‘unknown’ category to capture the fact that some reads cannot be assigned to any species. The resulting metagenomic profile includes posterior probabilities of species presence as well as Bayes factor for presence versus absence of specific species. There are two modes: metaMix-protein, which is optimal for RNA virus detection, and metaMix-nucl, which is best for speciation of DNA microorganisms. Both modes were used for RNA-seq, whereas metaMix-nucl was used for DNA-seq. For sequence results to be valid, MS2 phage RNA had to be detected in every sample and feline calicivirus RNA and cowpox DNA, with no additional unexpected organisms, detected in the controls. Confirmatory mapping of AAV2 The RNA-seq reads were mapped to the AAV2 reference genome (NCBI reference sequence NC_001401 ) using Bowtie2, with the –very-sensitive option. Samtools (version 1.9) and Picard (version 2.26.9; http://broadinstitute.github.io/picard/ ) were used to sort, deduplicate and index the alignments, and to create a depth file, which was plotted using a custom script in R. De novo assembly of unclassified reads We performed a de novo assembly step with metaSPADES (v3.15.5), using all the reads with no matches to the nucleotide database that we used for our similarity search. A search using megaBLAST with the standard nucleotide collection was carried out on all resulting contigs over 1,000 bp in length. All of the contigs longer than 1,000 bp matched to human, except two that mapped to Torque Teno virus. The RNA-seq reads were mapped to the AAV2 reference genome (NCBI reference sequence NC_001401 ) using Bowtie2, with the –very-sensitive option. Samtools (version 1.9) and Picard (version 2.26.9; http://broadinstitute.github.io/picard/ ) were used to sort, deduplicate and index the alignments, and to create a depth file, which was plotted using a custom script in R. We performed a de novo assembly step with metaSPADES (v3.15.5), using all the reads with no matches to the nucleotide database that we used for our similarity search. A search using megaBLAST with the standard nucleotide collection was carried out on all resulting contigs over 1,000 bp in length. All of the contigs longer than 1,000 bp matched to human, except two that mapped to Torque Teno virus. DNA from up to 20 mg of liver was purified using the Qiagen DNeasy Blood & Tissue kit as per the manufacturer’s instructions. Samples with limited amount of DNA were fragmented to an average size of 10 kb using a Megaruptor 3 (Diagenode) to reach an optimal molar concentration for library preparation. Quality control was perform using a Femto Pulse System (Agilent Technologies) and a Qubit fluorometer (Invitrogen). Samples were prepared for Nanopore sequencing using the ligation sequencing kit SQK-LSK110. DNA was sequenced on a PromethION using R9.4.1 flowcells (Oxford Nanopore Technologies). Samples were run for 72 h including a washing and reload step after 24 h and 48 h. All library preparation and sequencing were performed by the UCL Long Read Sequencing facility. Passed reads from Minknow were mapped to the reference AAV2 genome ( NC_001401 ) using minimap2 (ref. ) using the default parameters. Reads were trimmed of adapters using Porechop v0.2.4 ( https://github.com/rrwick/Porechop/ ), with the sequences of the adapters used added to adapters.py, and using an adapter threshold of 85. Reads that also mapped by minimap to the human genome (Ensemble GRCh38_v107), which could be ligation artefacts, were excluded from further analysis. The passed reads were also classified using Kraken2 (ref. ) with the PlusPF database (17 May 2021). The data relating to AAV2 reads in Supplementary Table refer to reads that were classified as AAV2 by both minimap2 and Kraken2 (version 2.0.8-beta), as the results from both methods were similar. Four reads across all four lower-depth samples were classified as HHV-6B by the EPI2ME WIMP pipeline. No reads were classified as HAdV or HHV-6B by Kraken2 in the two higher-depth samples. Alignment dot plots were created for the AAV2 reads using redotable (version 1.1) , with a window size of 20. These were manually classified into possible complex and monomeric structures. We investigated potential integrations of AAV2 and HHV-6 viruses into the genome using the Illumina metagenomics data for five liver transplant cases. We first processed the pair-end reads (average sequence coverage per genome = 5×), quality checking using FastQC , with barcode and adaptor sequence trimmed by TrimGalore (phred-score = 20). Potential viral integrations were investigated with Vseq-Toolkit (mode 3 with default settings except for high stringency levels). Predicted genomic integrations were visualized with IGV , requiring at least three reads supporting an integration site, spanning both human and viral sequences. Predicted integrations were supported by only one read, thus not fulfilling the algorithm criteria. Sequencing was performed at a lower depth than optimal for integration analysis, but no evidence was found for AAV2 or HHV-6B integration into the genomes of cases. Real-time PCR targeting a 62-nt region of the AAV2 inverted terminal repeat sequence was performed using primers and probes previously described . This assay has been predicted to amplify AAV2 and AAV6. The Qiagen QuantiNova probe PCR kit (PERFORM and DIAMONDS controls) or the Qiagen Quantifast probe PCR kit (all other samples) were used. Each 25-µl reaction consisted of 0.1 µM forward primer, 0.34 µM reverse primer and 0.1 µM probe with 5 µl template DNA. Real-time PCR targeting a 74-bp region of the HHV-6 DNA polymerase gene was performed using primers and probes previously described multiplexed with an internal positive control targeting mouse ( mus ) DNA spiked into each sample during DNA purification, as previously described . In brief, each 25-µl reaction consisted of 0.5 µM of each primer, 0.3 µM HHV-6 probe, 0.12 µM of each mus primer, 0.08 µM mus probe and 12.5 µl Qiagen Quantifast Fast mastermix with 10 µl template DNA. Real-time PCR targeting a 132-bp region of the HAdV hexon gene was performed using primers and probes previously described multiplexed with an internal positive control targeting mouse ( mus ) DNA spiked into each sample during DNA purification, as previously described . In brief, each 25-µl reaction consisted of 0.6 µM of each HHV-6 primer, 0.4 µM HHV-6 probe, 0.12 µM of each mus primer, 0.08 µM mus probe and 12.5 µl Qiagen Quantifast Fast mastermix with 10 µl template DNA. PCR cycling for all targets, apart from the controls from the PERFORM and DIAMONDS studies, was performed on an ABI 7500 Fast thermocycler and consisted of 95 °C for 5 min followed by 45 cycles of 95 °C for 30 s and 60 °C for 30 s. For the PERFORM and DIAMONDS controls, PCR was performed on a StepOnePlus Real-Time PCR System and consisted of 95 °C for 2 min followed by 45 cycles of 95 °C for 5 s and 60 °C for 10 s. Each PCR run included a no template control and a DNA-positive control for each target. Neat DNA extracts of the FFPE material were inhibitory to PCR, so PCR results shown were performed following a 1 in 10 dilution. RNA samples were treated with the Turbo-DNA free kit (Thermo) to remove residual genomic DNA. Complementary DNA (cDNA) was synthesized using the QuantiTect Reverse Transcription kit. In brief, 12 µl of RNA was mixed with 2 µl of genomic DNA Wipeout buffer and incubated at 42 °C for 2 min and transferred to ice. For reverse transcription, 6 µl mastermix was used and incubated at 42 °C for 20 min followed by 3 min at 95 °C. Real-time PCR targeting a 120-nt region of the AAV2 cap open reading frame sequence was performed using primers AAV2_cap _Fw- ATCCTTCGACCACCTTCAGT, AAV2_cap _Rv-GATT CCAGCGTTTGCTGTT and the probe AAV2_cap _Pr FAM-ACACAGTAT/ZEN/TCC ACGG GACAGGT-IBFQ. This assay has been predicted to amplify AAV2 and AAV6. The Qiagen QuantiNova probe PCR kit was used. Each 25-µl reaction consisted of 0.1 µM forward primer, 0.1 µM reverse primer and 0.2 µM probe with 2.5 µl template cDNA. PCR was performed on a StepOnePlus Real-Time PCR System and consisted of incubation at 95 °C for 2 min followed by 45 cycles of 95 °C for 5 s and 60 °C for 10 s. Each PCR run included a no template control, a DNA-positive control and a RNA control from each sample to verify efficient removal of genomic DNA. All immunohistochemistry was done on FFPE tissue cut at a thickness of 3 µm. AdV immunohistochemistry was carried out using the Ventana Benchmark ULTRA, Optiview Detection Kit, PIER with protease 1 for 4 min and antibody incubation for 32 min (AdV clone 2/6 and 20/11, Roche, 760-4870, pre-diluted). The positive control was a known HAdV-positive gastrointestinal surgical case. The plasmid used for transfection was pAAV2/2 (addgene, plasmid #104963; https://www.addgene.org/104963/ ), which expresses the genes encoding Rep/Cap of AAV2. This was delivered by tail-vein hydrodynamic injection into albino C57BL/6 mice (5 mg in 2 ml PBS). Negative controls received PBS alone. At 48 h, mice were terminally exsanguinated and perfused by PBS. Livers were collected into 10% neutral buffered formalin (CellPath UK). This was performed under Home Office License PAD4E6357. AAV2 immunohistochemistry was carried out with four commercially available antibodies: Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 30 min and antibody incubation for 30 min (anti-AAV VP1/VP2/VP3 clone B1, PROGEN, 690058S, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1/VP2/VP3 rabbit polyclonal, OriGene, BP5024, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1 clone A1, OriGene, BM5013, 1:100). Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, HIER with Bond Epitope Retrieval Solution 1 (citrate based pH 6) for 40 min and antibody incubation for 30 min (anti-AAV VP1/VP2 clone A69, OriGene, BM5014, 1:100). HHV-6 immunohistochemistry straining was carried out with: Leica Bond-III, Bond Polymer Refine Detection Kit with DAB Enhancer, PIER with Bond Enzyme 1 Kit for 10 min and antibody incubation for 30 min (mouse monoclonal antibody (C3108-103) to HHV-6, ABCAM, ab128404, 1:100). Negative reagent control slides were stained using the same antigen retrieval conditions and staining protocol incubation times using only BondTM Primary Antibody Diluent #AR9352 for the antibody incubation. Samples of liver were fixed in 2.5% glutaraldehyde in 0.1 M cacodylate buffer followed by secondary fixation in 1.0% osmium tetroxide. Tissues were dehydrated in graded ethanol, transferred to an intermediate reagent, propylene oxide and then infiltrated and embedded in Agar 100 epoxy resin. Polymerization was undertaken at 60 °C for 48 h. Ultrathin sections of 90 nm were cut using a Diatome diamond knife on a Leica UC7 ultramicrotome. Sections were transferred to copper grids and stained with alcoholic urynal acetate and Reynold’s lead citrate. The samples were examined using a JEOL 1400 transmission electron microscope. Images were captured on an AMT XR80 digital camera. Bait design To produce the capture probes for hybridization, biotinylated RNA oligonucleotides (baits) used in the SureSelectXT protocols for HAdV and HHV-6 WGS were designed in-house using Agilent community design baits with part numbers 5191-6711 and 5191-6713, respectively. They were synthesized by Agilent Technologies (2021) (available through Agilent’s Community Designs programme: SSXT CD Pan Adenovirus and SSXT CD Pan HHV-6 and used previously , ). Library preparation and sequencing For WGS of HAdV and HHV-6B, DNA (bulked with male human genomic DNA (Promega) if required) was sheared using a Covaris E220 focused ultrasonication system (PIP 75, duty factor of 10, 1,000 cycles per burst). End-repair, non-templated addition of 3′ poly A, adapter ligation, hybridization, PCR (pre-capture cycles dependent on DNA input and post-capture cycles dependent on viral load) and all post-reaction clean-up steps were performed according to either the SureSelectXT Low Input Target Enrichment for Illumina Paired-End Multiplexed Sequencing protocol (version A0), the SureSelectXT Target Enrichment for Illumina Paired-End Multiplexed Sequencing protocol (version C3) or the SureSelectXTHS Target Enrichment using the Magnis NGS Prep System protocol (version A0) (Agilent Technologies). Quality control steps were performed on the 4200 TapeStation (Agilent Technologies). Samples were sequenced using the Illumina MiSeq platform. Base calling and sample demultiplexing were performed as standard for the MiSeq platform, generating paired FASTQ files for each sample. A negative control was included on each processing run. A targeted enrichment approach was used due to the predicted high variability of the HHV-6 and HAdV genomes. For AAV2 WGS, an AAV2 primer scheme was designed using primalscheme with 17 AAV2 sequences from NCBI and one AAV2 sequence provided by GOSH from metagenomic sequencing of a liver biopsy DNA extract as the reference material. These primers amplify 15 overlapping 400-bp amplicons. Primers were supplied by Merck. Two multiplex PCRs were prepared using Q5 Hot Start High-Fidelity 2X Master Mix, with a 65 °C, 3 min annealing/extension temperature. Pools 1 and 2 multiplex PCRs were run for 35 cycles. Of each PCR, 10 µl was combined and 20 µl nuclease-free water was added. Libraries were prepared either manually or on the Agilent Bravo NGS workstation option B, following a reduced-scale version of the Illumina DNA protocol as used in the CoronaHiT protocol . Equal volumes of the final libraries were pooled, bead purified and sequenced on the Illumina MiSeq. A negative control was included on each processing run. All library preparation and sequencing were performed by UCL Genomics. AAV2 sequence analysis The raw fastq reads were adapted, trimmed and low-quality reads were removed. The reads were mapped to the NC_001401 reference sequence and then the amplicon primers regions were trimmed using the location provided in a bed file. Consensus sequences were then called at a minimum of 10× coverage. The entire processing of raw reads to consensus was carried out using the nf-core/viralrecon pipeline ( https://nf-co.re/viralrecon/2.4.1 ; 10.5281/zenodo.3901628). Basic quality metrics for the samples sequenced are in Supplementary Table . All samples that gave 10× genome coverage over 90% were then used for further phylogenetic analysis. Samples were aligned along with known reference strains from GenBank using MAFFT (version v7.271), and the trees were built with IQ-TREE (multicore version 1.6.12) with 1,000 rapid bootstraps and approximate likelihood-ratio test support. The samples were then labelled based on type and provider on the trees (Fig. ). For each AAV2 sample, we aligned the consensus nucleotide sequence to the AAV2 reference sequence. From these alignments, the exact coordinates of the sample capsid were determined. We then used the coordinates to extract the corresponding nucleotide sequence and translated it to find the amino acid sequence. Next, we compared each sample to the reference to identify amino acid changes. Amino acid sequences from AAV capsid sequences were retrieved from GenBank for AAV1 to AAV12. Amino acid sequences of capsid constructs designed to be more hepatotropic were retrieved from refs. , . These sequence sets were then aligned to the AAV2 reference sequence using MAFFT . We then compared each construct to the AAV2 reference to identify amino acid changes present, while retaining the AAV2 coordinate set. HAdV and HHV-6B sequence analysis Raw data quality control was performed using trim-galore (v.0.6.7) on the raw FASTQ files. For HHV-6B, short reads were mapped with BWA mem (0.7.17-r1188) using the RefSeq reference NC_000898 . For HAdV, genotyping is performed using AYUKA (version 22-111). This novel tool is used to confidently assign one or more HAdV genotypes to a sample of interest, assessing inter-genotype recombination if more than one genotype is detected. The results from this screening step guide which downstream analyses are performed and which reference genome (or genomes) is used. If mixed infection is suspected, reads are separated using bbsplit ( https://sourceforge.net/projects/bbmap/ ), and each genotype is analysed independently as normal. If recombination is suspected, a more detailed analysis is performed using Recombination Detection Program (RDP) and the sample is excluded from phylogenetic analysis. After genotyping, the cleaned read data are mapped using BWA to the relevant reference sequence (or sequences), and SNPs and small insertions and deletions are called using bcftool (version1.15.1, https://github.com/samtools/bcftools ) and a consensus sequence is generated also with bcftools, masking with Ns positions that do not have enough read support (15× by default). Consensus sequences generated with the pipeline are then concatenated to previously sequenced samples and a multiple sequence alignment is performed using the G-INS-I algorithm in the MAFFT software (MAFFT G-INS-I v7.481). The multiple sequence alignment is then used for phylogenetic analysis with IQ-TREE (IQ-TREE 2 2.2.0), using modelfinder and performing 1,000 rapid bootstraps. Proteomics data generation Liver explant tissue from cases was homogenized in lysis buffer, 100 mM Tris (pH 8.5), 5% sodium dodecyl sulfate, 5 mM tris(2-carboxyethyl)phosphine and 20 mM chloroacetamide then heated at 95 °C for 10 min and sonicated in an ultrasonic bath for another 10 min. The lysed proteins were quantified with NanoDrop 2000 (Thermo Fisher Scientific). One-hundred micrograms was precipitated with the methanol/chloroform protocol and then protein pellets were reconstituted in 100 mM Tris (pH 8.5) and 4% sodium deoxycholate (SDC). The proteins were subjected to proteolysis with 1:50 trypsin overnight at 37 °C with constant shaking. Digestion was stopped by adding 1% trifluoroacetic acid to a final concentration of 0.5%. Precipitated SDC was removed by centrifugation at 10,000 g for 5 min, and the supernatant containing digested peptides was desalted on an SOLAµ HRP (Thermo Fisher Scientific). Of the desalted peptide, 50 µg was then fractionated on Vanquish HPLC (Thermo Fisher Scientific) using a Acquity BEH C18 column (2.1 × 50 mm with 1.7-µm particles from Waters): buffer A was 10 mM ammonium formiate at pH 10, whereas buffer B was 80% acetonitrile and the flow was set to 500 µl per minute. We used a gradient of 8 min to collect 24 fractions that were then concatenated to obtain 12 fractions. These 12 fractions were dried and dissolved in 2% formic acid before liquid chromatography–tandem mass spectrometry analysis. An estimated total of 2,000 ng from each fraction was analysed using an Ultimate3000 high-performance liquid chromatography system coupled online to an Eclipse mass spectrometer (Thermo Fisher Scientific). Buffer A consisted of water acidified with 0.1% formic acid, whereas buffer B was 80% acetonitrile and 20% water with 0.1% formic acid. The peptides were first trapped for 1 min at 30 μl per minute with 100% buffer A on a trap (0.3 mm × 5 mm with PepMap C18, 5 μm, 100 Å; Thermo Fisher Scientific); after trapping, the peptides were separated by a 50-cm analytical column (Acclaim PepMap, 3 μm; Thermo Fisher Scientific). The gradient was 9–35% buffer B for 103 min at 300 nl per minute. Buffer B was then raised to 55% in 2 min and increased to 99% for the cleaning step. Peptides were ionized using a spray voltage of 2.1 kV and a capillary heated at 280 °C. The mass spectrometer was set to acquire full-scan mass spectrometry spectra (350:1,400 mass:charge ratio) for a maximum injection time set to auto at a mass resolution of 120,000 and an automated gain control target value of 100%. For a second, the most intense precursor ions were selected for tandem mass spectrometry. Higher energy collisional dissocation (HCD) fragmentation was performed in the HCD cell, with the readout in the Orbitrap mass analyser at a resolution of 15,000 (isolation window of 3 Th) and an automated gain control target value of 200% with a maximum injection time set to auto and a normalized collision energy of 30%. All raw files were analysed by MaxQuant v2.1 software using the integrated Andromeda search engine and searched against the Human UniProt Reference Proteome (February release with 79,057 protein sequences) together with UniProt-reported AAV proteins and specific fasta created using EMBOSS Sixpack translating patient’s virus genome. MaxQuant was used with the standard parameters with only the addition of deamidation (N) as variable modification. Data analysis was then carried out with Perseus v2.05: proteins reported in the file ‘proteinGroups.txt’ were filtered for reverse and potential contaminants. Figures were created using Origin pro version 2022b. Transduction of AAV2 capsid mutants A transgene sequence containing enhanced green fluorescent protein (eGFP) was packaged into rAAV2 particles to track their expression in transduced cells, compared with rAAV capsids derived from canonical AAV2, AAV9 and a synthetic liver-tropic AAV vector called LK03 (ref. ). rAAV vector particles were delivered to Huh-7 hepatocytes at a multiplicity of infection of 100,000 vector genomes per cell before analysing eGFP expression by flow cytometry 72 h later. Recombinant AAV capsid sequence The VP1 sequence was generated by generating a consensus sequence from a multiple sequence alignment of sequenced AAV2 genomes derived from patient samples, using the Biopython package AlignIO. The designed VP1 sequence was then synthesized as a ‘gBlock’ (Integrated DNA Technologies) and incorporated into an AAV2 RepCap plasmid (AAV2/2 was a gift from M. Fan, Addgene plasmid #104963) between the SwaI and XmaI restriction sites, using InFusion cloning reagent (product 638948, Clontech). AAV vector production rAAV particles were generated by transient transfection of HEK 293T cells as previously described . In brief, 1.8 × 10 7 cells were plated in 15-cm dishes before transfecting the pAAV-CAG-eGFP transgene plasmid (a gift from E. Boyden, Addgene plasmid #37825), the relevant RepCap plasmid and the pAdDeltaF6 helper plasmid (a gift from J. M. Wilson, Addgene plasmid #112867), at a ratio of 10.5 µg, 10.5 µg and 30.5 µg, respectively, using PEIPro transfection reagent (PolyPlus) at a ratio of 1 µl per 1 µg DNA. Seventy-two hours post-transfection, cell pellets and supernatant were harvested and rAAV particles were purified using an Akta HPLC platform. rAAV particle genome copy numbers were calculated by quantitative PCR targeting the vector transgene region. The rAAV2 vector used in this study was purchased as ready-to-use AAV2 particles from Addgene (Addgene viral prep #37825-AAV2). Analysis of rAAV transduction Huh-7 hepatocytes (a gift from J. Baruteau, UCL) were plated in DMEM medium supplemented with 10% FBS and 1% penicillin–streptomycin supplement. The cell line was validated by testing for glypican-3 and was not tested for mycoplasma contamination. Cells were plated at a density of 1.5 × 10 3 cells per square centimetre and transduced with 1 × 10 5 viral genomes per cell. Transductions were performed in the presence or absence of 400 µg ml −1 heparin, which was supplemented directly to cell media. Seventy-two hours after transduction, cells were analysed by microscopy using an EVOS Cell Imaging System (Thermo Fisher Scientific) before quantifying eGFP expression by flow cytometry using a Cytoflex Flow Cytometer (Beckman). eGFP-positive cells were determined by gating the live-cell population and quantifying the level of eGFP signal versus untransduced controls. To produce the capture probes for hybridization, biotinylated RNA oligonucleotides (baits) used in the SureSelectXT protocols for HAdV and HHV-6 WGS were designed in-house using Agilent community design baits with part numbers 5191-6711 and 5191-6713, respectively. They were synthesized by Agilent Technologies (2021) (available through Agilent’s Community Designs programme: SSXT CD Pan Adenovirus and SSXT CD Pan HHV-6 and used previously , ). For WGS of HAdV and HHV-6B, DNA (bulked with male human genomic DNA (Promega) if required) was sheared using a Covaris E220 focused ultrasonication system (PIP 75, duty factor of 10, 1,000 cycles per burst). End-repair, non-templated addition of 3′ poly A, adapter ligation, hybridization, PCR (pre-capture cycles dependent on DNA input and post-capture cycles dependent on viral load) and all post-reaction clean-up steps were performed according to either the SureSelectXT Low Input Target Enrichment for Illumina Paired-End Multiplexed Sequencing protocol (version A0), the SureSelectXT Target Enrichment for Illumina Paired-End Multiplexed Sequencing protocol (version C3) or the SureSelectXTHS Target Enrichment using the Magnis NGS Prep System protocol (version A0) (Agilent Technologies). Quality control steps were performed on the 4200 TapeStation (Agilent Technologies). Samples were sequenced using the Illumina MiSeq platform. Base calling and sample demultiplexing were performed as standard for the MiSeq platform, generating paired FASTQ files for each sample. A negative control was included on each processing run. A targeted enrichment approach was used due to the predicted high variability of the HHV-6 and HAdV genomes. For AAV2 WGS, an AAV2 primer scheme was designed using primalscheme with 17 AAV2 sequences from NCBI and one AAV2 sequence provided by GOSH from metagenomic sequencing of a liver biopsy DNA extract as the reference material. These primers amplify 15 overlapping 400-bp amplicons. Primers were supplied by Merck. Two multiplex PCRs were prepared using Q5 Hot Start High-Fidelity 2X Master Mix, with a 65 °C, 3 min annealing/extension temperature. Pools 1 and 2 multiplex PCRs were run for 35 cycles. Of each PCR, 10 µl was combined and 20 µl nuclease-free water was added. Libraries were prepared either manually or on the Agilent Bravo NGS workstation option B, following a reduced-scale version of the Illumina DNA protocol as used in the CoronaHiT protocol . Equal volumes of the final libraries were pooled, bead purified and sequenced on the Illumina MiSeq. A negative control was included on each processing run. All library preparation and sequencing were performed by UCL Genomics. The raw fastq reads were adapted, trimmed and low-quality reads were removed. The reads were mapped to the NC_001401 reference sequence and then the amplicon primers regions were trimmed using the location provided in a bed file. Consensus sequences were then called at a minimum of 10× coverage. The entire processing of raw reads to consensus was carried out using the nf-core/viralrecon pipeline ( https://nf-co.re/viralrecon/2.4.1 ; 10.5281/zenodo.3901628). Basic quality metrics for the samples sequenced are in Supplementary Table . All samples that gave 10× genome coverage over 90% were then used for further phylogenetic analysis. Samples were aligned along with known reference strains from GenBank using MAFFT (version v7.271), and the trees were built with IQ-TREE (multicore version 1.6.12) with 1,000 rapid bootstraps and approximate likelihood-ratio test support. The samples were then labelled based on type and provider on the trees (Fig. ). For each AAV2 sample, we aligned the consensus nucleotide sequence to the AAV2 reference sequence. From these alignments, the exact coordinates of the sample capsid were determined. We then used the coordinates to extract the corresponding nucleotide sequence and translated it to find the amino acid sequence. Next, we compared each sample to the reference to identify amino acid changes. Amino acid sequences from AAV capsid sequences were retrieved from GenBank for AAV1 to AAV12. Amino acid sequences of capsid constructs designed to be more hepatotropic were retrieved from refs. , . These sequence sets were then aligned to the AAV2 reference sequence using MAFFT . We then compared each construct to the AAV2 reference to identify amino acid changes present, while retaining the AAV2 coordinate set. Raw data quality control was performed using trim-galore (v.0.6.7) on the raw FASTQ files. For HHV-6B, short reads were mapped with BWA mem (0.7.17-r1188) using the RefSeq reference NC_000898 . For HAdV, genotyping is performed using AYUKA (version 22-111). This novel tool is used to confidently assign one or more HAdV genotypes to a sample of interest, assessing inter-genotype recombination if more than one genotype is detected. The results from this screening step guide which downstream analyses are performed and which reference genome (or genomes) is used. If mixed infection is suspected, reads are separated using bbsplit ( https://sourceforge.net/projects/bbmap/ ), and each genotype is analysed independently as normal. If recombination is suspected, a more detailed analysis is performed using Recombination Detection Program (RDP) and the sample is excluded from phylogenetic analysis. After genotyping, the cleaned read data are mapped using BWA to the relevant reference sequence (or sequences), and SNPs and small insertions and deletions are called using bcftool (version1.15.1, https://github.com/samtools/bcftools ) and a consensus sequence is generated also with bcftools, masking with Ns positions that do not have enough read support (15× by default). Consensus sequences generated with the pipeline are then concatenated to previously sequenced samples and a multiple sequence alignment is performed using the G-INS-I algorithm in the MAFFT software (MAFFT G-INS-I v7.481). The multiple sequence alignment is then used for phylogenetic analysis with IQ-TREE (IQ-TREE 2 2.2.0), using modelfinder and performing 1,000 rapid bootstraps. Liver explant tissue from cases was homogenized in lysis buffer, 100 mM Tris (pH 8.5), 5% sodium dodecyl sulfate, 5 mM tris(2-carboxyethyl)phosphine and 20 mM chloroacetamide then heated at 95 °C for 10 min and sonicated in an ultrasonic bath for another 10 min. The lysed proteins were quantified with NanoDrop 2000 (Thermo Fisher Scientific). One-hundred micrograms was precipitated with the methanol/chloroform protocol and then protein pellets were reconstituted in 100 mM Tris (pH 8.5) and 4% sodium deoxycholate (SDC). The proteins were subjected to proteolysis with 1:50 trypsin overnight at 37 °C with constant shaking. Digestion was stopped by adding 1% trifluoroacetic acid to a final concentration of 0.5%. Precipitated SDC was removed by centrifugation at 10,000 g for 5 min, and the supernatant containing digested peptides was desalted on an SOLAµ HRP (Thermo Fisher Scientific). Of the desalted peptide, 50 µg was then fractionated on Vanquish HPLC (Thermo Fisher Scientific) using a Acquity BEH C18 column (2.1 × 50 mm with 1.7-µm particles from Waters): buffer A was 10 mM ammonium formiate at pH 10, whereas buffer B was 80% acetonitrile and the flow was set to 500 µl per minute. We used a gradient of 8 min to collect 24 fractions that were then concatenated to obtain 12 fractions. These 12 fractions were dried and dissolved in 2% formic acid before liquid chromatography–tandem mass spectrometry analysis. An estimated total of 2,000 ng from each fraction was analysed using an Ultimate3000 high-performance liquid chromatography system coupled online to an Eclipse mass spectrometer (Thermo Fisher Scientific). Buffer A consisted of water acidified with 0.1% formic acid, whereas buffer B was 80% acetonitrile and 20% water with 0.1% formic acid. The peptides were first trapped for 1 min at 30 μl per minute with 100% buffer A on a trap (0.3 mm × 5 mm with PepMap C18, 5 μm, 100 Å; Thermo Fisher Scientific); after trapping, the peptides were separated by a 50-cm analytical column (Acclaim PepMap, 3 μm; Thermo Fisher Scientific). The gradient was 9–35% buffer B for 103 min at 300 nl per minute. Buffer B was then raised to 55% in 2 min and increased to 99% for the cleaning step. Peptides were ionized using a spray voltage of 2.1 kV and a capillary heated at 280 °C. The mass spectrometer was set to acquire full-scan mass spectrometry spectra (350:1,400 mass:charge ratio) for a maximum injection time set to auto at a mass resolution of 120,000 and an automated gain control target value of 100%. For a second, the most intense precursor ions were selected for tandem mass spectrometry. Higher energy collisional dissocation (HCD) fragmentation was performed in the HCD cell, with the readout in the Orbitrap mass analyser at a resolution of 15,000 (isolation window of 3 Th) and an automated gain control target value of 200% with a maximum injection time set to auto and a normalized collision energy of 30%. All raw files were analysed by MaxQuant v2.1 software using the integrated Andromeda search engine and searched against the Human UniProt Reference Proteome (February release with 79,057 protein sequences) together with UniProt-reported AAV proteins and specific fasta created using EMBOSS Sixpack translating patient’s virus genome. MaxQuant was used with the standard parameters with only the addition of deamidation (N) as variable modification. Data analysis was then carried out with Perseus v2.05: proteins reported in the file ‘proteinGroups.txt’ were filtered for reverse and potential contaminants. Figures were created using Origin pro version 2022b. A transgene sequence containing enhanced green fluorescent protein (eGFP) was packaged into rAAV2 particles to track their expression in transduced cells, compared with rAAV capsids derived from canonical AAV2, AAV9 and a synthetic liver-tropic AAV vector called LK03 (ref. ). rAAV vector particles were delivered to Huh-7 hepatocytes at a multiplicity of infection of 100,000 vector genomes per cell before analysing eGFP expression by flow cytometry 72 h later. The VP1 sequence was generated by generating a consensus sequence from a multiple sequence alignment of sequenced AAV2 genomes derived from patient samples, using the Biopython package AlignIO. The designed VP1 sequence was then synthesized as a ‘gBlock’ (Integrated DNA Technologies) and incorporated into an AAV2 RepCap plasmid (AAV2/2 was a gift from M. Fan, Addgene plasmid #104963) between the SwaI and XmaI restriction sites, using InFusion cloning reagent (product 638948, Clontech). rAAV particles were generated by transient transfection of HEK 293T cells as previously described . In brief, 1.8 × 10 7 cells were plated in 15-cm dishes before transfecting the pAAV-CAG-eGFP transgene plasmid (a gift from E. Boyden, Addgene plasmid #37825), the relevant RepCap plasmid and the pAdDeltaF6 helper plasmid (a gift from J. M. Wilson, Addgene plasmid #112867), at a ratio of 10.5 µg, 10.5 µg and 30.5 µg, respectively, using PEIPro transfection reagent (PolyPlus) at a ratio of 1 µl per 1 µg DNA. Seventy-two hours post-transfection, cell pellets and supernatant were harvested and rAAV particles were purified using an Akta HPLC platform. rAAV particle genome copy numbers were calculated by quantitative PCR targeting the vector transgene region. The rAAV2 vector used in this study was purchased as ready-to-use AAV2 particles from Addgene (Addgene viral prep #37825-AAV2). Huh-7 hepatocytes (a gift from J. Baruteau, UCL) were plated in DMEM medium supplemented with 10% FBS and 1% penicillin–streptomycin supplement. The cell line was validated by testing for glypican-3 and was not tested for mycoplasma contamination. Cells were plated at a density of 1.5 × 10 3 cells per square centimetre and transduced with 1 × 10 5 viral genomes per cell. Transductions were performed in the presence or absence of 400 µg ml −1 heparin, which was supplemented directly to cell media. Seventy-two hours after transduction, cells were analysed by microscopy using an EVOS Cell Imaging System (Thermo Fisher Scientific) before quantifying eGFP expression by flow cytometry using a Cytoflex Flow Cytometer (Beckman). eGFP-positive cells were determined by gating the live-cell population and quantifying the level of eGFP signal versus untransduced controls. Cytokine transcriptomics analysis Cytokine inducible gene expression modules were derived from previously published bulk tissue genome-wide transcriptomes of the tuberculin skin test that have been shown to reflect canonical human in vivo cell-mediated immune pathways using a validated bioinformatic approach . Cytokine regulators of genes enriched in the tuberculin skin test (ArrayExpress accession number E-MTAB-6816 ) were identified using Ingenuity Pathway Analysis (Qiagen). Average correlation of log 2 -transformed transcripts per million data for every gene pair in each of the target gene modules were compared with 100 iterations of randomly selected gene modules of the same size, to select cytokine-inducible modules that showed significantly greater co-correlation (adjusted P < 0.05), representing co-regulated transcriptional networks for each 59 cytokines. We then used the average log 2 -transformed transcripts per million expression of all the genes in each of these co-regulated modules to quantify the biological activity of the associated upstream cytokine within bulk genome-wide transcriptional profiles from AAV2-associated hepatitis ( n = 4) obtained in the present study, compared with published log 2 -transformed and normalized microarray data from normal adult liver ( n = 10) and hepatitis B adult liver ( n = 17) (Gene Expression Omnibus accession number GSE96851 ) . To enable comparison across the datasets, we transformed average gene expression values for each cytokine-inducible module to standardized ( Z scores) using mean and standard deviation of randomly selected gene sets of the same size within each individual dataset. Statistically significant differences in Z scores between groups were identified by Student’s t -tests with multiple testing correction (adjusted P < 0.05). Proteomics differential expression To compare the proteomics data from the explanted livers of cases with data from healthy livers, we downloaded the raw files from two studies , from PRIDE. The raw files were searched together with our files using the same settings and databases. We performed differential expression analyses at the protein level and peptide level using a hybrid approach including statistical inference on the abundance (quantitative approach), as well as the presence or absence (binary approach) of proteins or peptides. DEP R package version 1.18.0 was used for quantitative analysis . Proteins or peptides were filtered for those detected in all replicates of at least one group (case or control). The data were background corrected and variance was normalized using variance-stabilizing transformation. Missing intensity values were not distributed randomly and were biased to specific samples (either cases or controls). Therefore, for imputing the missing data, we applied random draws from a manually defined left-shifted Gaussian distribution using the DEP impute function with parameters fun:“man”, shift:1.8 and scale:0.3. The test_diff function based on linear models and the empirical Bayes method was used for testing differential expressions between the case and control samples. HLA typing methods Typing was undertaken in the liver centre units. Next-generation sequencing (sequencing by synthesis (Illumina) using AllType kits (VHBio/OneLambda), a high-resolution HLA typing method, was used. Cytokine inducible gene expression modules were derived from previously published bulk tissue genome-wide transcriptomes of the tuberculin skin test that have been shown to reflect canonical human in vivo cell-mediated immune pathways using a validated bioinformatic approach . Cytokine regulators of genes enriched in the tuberculin skin test (ArrayExpress accession number E-MTAB-6816 ) were identified using Ingenuity Pathway Analysis (Qiagen). Average correlation of log 2 -transformed transcripts per million data for every gene pair in each of the target gene modules were compared with 100 iterations of randomly selected gene modules of the same size, to select cytokine-inducible modules that showed significantly greater co-correlation (adjusted P < 0.05), representing co-regulated transcriptional networks for each 59 cytokines. We then used the average log 2 -transformed transcripts per million expression of all the genes in each of these co-regulated modules to quantify the biological activity of the associated upstream cytokine within bulk genome-wide transcriptional profiles from AAV2-associated hepatitis ( n = 4) obtained in the present study, compared with published log 2 -transformed and normalized microarray data from normal adult liver ( n = 10) and hepatitis B adult liver ( n = 17) (Gene Expression Omnibus accession number GSE96851 ) . To enable comparison across the datasets, we transformed average gene expression values for each cytokine-inducible module to standardized ( Z scores) using mean and standard deviation of randomly selected gene sets of the same size within each individual dataset. Statistically significant differences in Z scores between groups were identified by Student’s t -tests with multiple testing correction (adjusted P < 0.05). To compare the proteomics data from the explanted livers of cases with data from healthy livers, we downloaded the raw files from two studies , from PRIDE. The raw files were searched together with our files using the same settings and databases. We performed differential expression analyses at the protein level and peptide level using a hybrid approach including statistical inference on the abundance (quantitative approach), as well as the presence or absence (binary approach) of proteins or peptides. DEP R package version 1.18.0 was used for quantitative analysis . Proteins or peptides were filtered for those detected in all replicates of at least one group (case or control). The data were background corrected and variance was normalized using variance-stabilizing transformation. Missing intensity values were not distributed randomly and were biased to specific samples (either cases or controls). Therefore, for imputing the missing data, we applied random draws from a manually defined left-shifted Gaussian distribution using the DEP impute function with parameters fun:“man”, shift:1.8 and scale:0.3. The test_diff function based on linear models and the empirical Bayes method was used for testing differential expressions between the case and control samples. Typing was undertaken in the liver centre units. Next-generation sequencing (sequencing by synthesis (Illumina) using AllType kits (VHBio/OneLambda), a high-resolution HLA typing method, was used. Fisher’s exact test and two-sided Wilcoxon (Mann–Whitney) non-parametric rank sum test were used for differences between case and control groups. Where multiple groups were compared, Kruskal–Wallis tests followed by Wilcoxon pairwise tests using a Benjamini–Hochberg correction were performed. All analysis were performed in R version 4.2.0. Further information on research design is available in the linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-023-06003-w. Reporting Summary Peer Review File Supplementary Table 1 Clinical details of cases. 12 transplanted cases, 26 non-transplanted. Median age of cases where exact age is known is 3 years, with range 1.5-9. Case 10 was 9 years old. All other cases were aged 7 or under. Of 22 cases where the gender is known, 12 cases were female and 10 were male. Where known, all cases were of white ethnicity other than two. Supplementary Table 2 Metagenomics summary statistics: raw read counts, human filtered and other findings. F: sequencing failed. Supplementary Table 3 Nanopore sequencing. All four samples were sequenced to a lower depth. Case 3 and 5 underwent a second round of deeper sequencing. N50s across all sequencing runs were similar. Supplementary Table 4 Clinical details for PERFORM/DIAMONDS immunocompetent controls and microbiological testing by referring laboratory for DIAMONDS controls. P: positive, N: negative, IC: inconclusive, VL: viral load. Supplementary Table 5 PCR Results. Not all samples were tested for all viruses due to lack of remaining material. LLP: low level positive (Ct value > 38 and < 45), ND: not determined (negative PCR results), NA: not tested due to lack of material. Supplementary Table 6 Clinical details of liver controls and comparators Supplementary Table 7 HAdV whole genome sequencing. OTR: on target reads, MRD: mean read depth, Coverage 1X: percentage of genome covered at depth 1, Coverage 30X: percentage of genome covered at depth 30. Supplementary Table 8 Mapping of HAdV partial sequences. Supplementary Table 9 AAV2 whole genome sequencing: OTR: on target reads, MRD: mean read depth, Coverage 1X: percentage of genome covered at depth 1, Coverage 10X: percentage of genome covered at depth 10. Supplementary Table 10 HHV-6B whole genome sequencing from liver case samples. OTR: on target reads, MRD: mean read depth, Coverage 1X: percentage of genome covered at depth 1, Coverage 10X: percentage of genome covered at depth 10. Supplementary Table 11 Cytokine modules - cytokine-inducible transcriptional signatures of cell-mediated immunity. Supplementary Table 12 Summary statistics for multiple comparisons of cytokine transcriptional modules. Two-tailed unpaired t-tests with Holm Sidak multiple testing correction for adjusted p values were performed. Supplementary Table 13 List of differentially expressed proteins between 5 cases and 7 controls. The p-values were calculated by applying two-tailed empirical Bayes moderated t-statistics on protein-wise linear models. P values were not adjusted for multiple comparisons. Supplementary Table 14 List of differentially expressed peptides between 5 cases and 7 controls. The p-values were calculated by applying two-tailed empirical Bayes moderated t-statistics on peptide-wise linear models. P values were not adjusted for multiple comparisons. Supplementary Table 15 HLA allele frequency of cases.
null	e801c76e-65d9-4a93-a81e-15bab1f2bad6	10170724	Microbiology[mh]	Campylobacter species are gram-negative bacteria with different morphologies (from spiral to curved, or rod-shaped) . They have single polar flagellum, bipolar flagella, or no flagellum, depending on the species. It has been reported that at least 12 species of Campylobacter cause human disease, the most common of which are Campylobacter jejuni and Campylobacter coli . Many countries around the world recognize C. jejuni (~ 90%) and C. coli (~ 10%) as the major causative agents of human campylobacteriosis whose symptoms include diarrhea that occasionally is bloody, abdominal pain, and fever . Rarely, serious long-term complications occur such as peripheral neuropathies, reactive arthritis, and Miller Fisher syndrome. Infection caused by C. jejuni is the most common reason of neurological sequelae . Campylobacter is a zoonotic pathogen and its most common source is poultry . In addition, contaminated water and food products, such as unpasteurized milk and contaminated fresh produce, are also known as other sources of Campylobacter infections . Campylobacter infection can also occur from direct contact with infected animals, which usually carry the bacteria asymptomatically . According to recent data, there has been a rise in the global incidence of campylobacteriosis in most countries, although there is incomplete data from Asia, and the Middle East . There is no comprehensive data on the prevalence of Campylobacter at the national level. This systematic review was conducted to provide comprehensive evidence on the prevalence of Campylobacter in human, animal, and food in Iran by using a systematic review and meta-analysis based method. Results of this study will serve as data that can be used for the prevention and control of Campylobacter infections in the country as well as guide to identify the research gaps. Overall a total of 536 articles were identified through PubMed, Scopus, and Web of Science, and 72 additional articles were identified through Google scholar, SID, and hand-based searching for the prevalence of Campylobacter species. Figure illustrates the method applied for selecting eligible studies. 582 articles remained after removing duplicates. Based on the eligibility criteria, 457 articles were excluded. A further 5 full-text articles were excluded due to the following reasons Review (1), Case report (1), Abstract (1), confused text/incomprehensible data and duplicate data (1), Non-available full-text (1). Finally, 119 articles were included in the quantitative synthesis. Table presents the detailed characteristics of every included study. Prevalence/proportion of Campylobacter spp. in meat/animal products and environment of Iran An overview showing the pooled Campylobacter spp. prevalence data generated from Iranian meat (92 studies), environment (6 studies), fecal (79 studies) and animal product sample (44 studies) categories generated using the random effects model is provided in Fig. . The highest prevalence of Campylobacter spp. has been observed in white meat (43.9%) from 55 studies among the meat and animal products that was reported in different studies from 0 to 90%. Campylobacter spp. prevalence in white meat was higher for chicken (48.6%) than other types of poultry meat (33.9%). Within the red meat category by 37 studies, Campylobacter spp. was detected at an overall pooled prevalence of 7.9% (Table ), which was reported from 0 to 24% in the literature. Campylobacter contamination in this category was mostly prevalent in buffalo (13.5%), followed by goat and sheep (8.6%), cattle (8.4%) and camel (2.5%) meat. While among animal products eggs were found to have a 5.5% prevalence of Campylobacter spp. contamination, with a high rate of contamination prevalence being observed for chicken eggs (9.9%) in eight studies compared to eggs of other types of poultry (4.2%) from 24 studies. The prevalence of Campylobacter spp. contamination detected among environmental samples was 14.9%. Vegetables were constituted environmental samples that showed highest prevalence (19.4%) of Campylobacter contamination. Water and sewage samples had prevalence of 15.4% and 7.4%, respectively. As the I2 heterogeneity index was more than 50, there was heterogeneity in the included studies. Prevalence/proportion of Campylobacter spp. in fecal samples Literature review of 79 studies that investigated the fecal samples in animal and human revealed that pooled proportion of Campylobacter spp. was 18.7% in fecal samples. Among food animals, poultry had the highest contamination of fecal samples (46.8%). Domestic and wild animal had 21% and 14.1% contamination of Campylobacter spp. (Table ). A proportion of 8.4% of human samples were positive regarding Campylobacter spp. Prevalence/proportion of Campylobacter spp. by place of sampling Table presents an overview from the meta-analysis of Campylobacter spp. prevalence from Iran based on sampling places. Poultry feces (61.9%) and white meat (47.2%) were determined to have the highest Campylobacter spp. prevalence at the slaughterhouse. This was followed by white meat at market (42.6%) and farm (40%) levels. The lowest pooled prevalence of Campylobacter spp. was observed for milk sampled at farm (1%) and market (3.3%) levels, eggs sampled at market (5.4%) and red meat sampled at slaughterhouse (6.2%) levels. Campylobacter spp. prevalence in white and red meat, and milk samples at markets (sampled from retails, supermarkets and butcher’s) was higher than at farms (Table ). Considerable proportions of wild animal (prevalence of 25.4%) and dog and cat feces (prevalence of 20.4%), were found to be contaminated with Campylobacter spp.. Prevalence/proportion of C. jejuni and C. coli As the C. jejuni and C. coli are the main causative agents of human campylobacteriosis, the pooled prevalence of these two species was determined in Iran samples. Most of the studies reported the prevalence of C. jejuni and C. coli in their samples. C. jejuni had higher pooled prevalence/proportion than C. coli in all of the obtained samples except for those derived from vegetables. Sewage (100%) (one study), milk (86.6%) (7 studies), human feces (83.3%) (33 studies) and water (82.8%) (3 studies) samples had the most frequent contaminations with C. jejuni (Fig. ) . Pooled C. jejuni prevalence in white meat (54 studies), egg (28 studies), poultry feces (19 studies) and red meat (35 studies) was 68.7%, 65.5%, 65.2% and 62.7%, respectively. Vegetable (2 studies) samples had the least pooled prevalence of C. jejuni (28%). On the other hand the highest pooled prevalence of C. coli was reported in vegetable samples (72%) followed by egg (33%) and red meat (24.1%) samples. Pooled prevalence of C. coli was zero (95%CI: 0–84.2%) in sewage samples (Fig. ). Pooled proportion of virulence genes in Campylobacter spp. Despite the high number of studies that reported the prevalence of Campylobacter spp., a limited number of them investigated the virulence genes required for pathogenesis. CdtA, cdtB, cdtC, cadF and pldA had the highest number of investigated studies. Figure shows the proportion of virulence genes in Campylobacter spp. cadF (97%) had the highest pooled prevalence in Campylobacter spp. in 28 studies, followed by racR (93.8%) (3 studies) and flaA (91.3%) (17 studies). VirB11 had the least prevalence (0%) in the Campylobacter spp. in 11 investigated studies. A total of 31% of Campylobacter spp. contained wlaN in 7 studies. With the sensitivity analysis, it was found that one of the studies pulls the results towards itself. The virB11 gene has the greatest impact on heterogeneity. Campylobacter spp. in meat/animal products and environment of Iran An overview showing the pooled Campylobacter spp. prevalence data generated from Iranian meat (92 studies), environment (6 studies), fecal (79 studies) and animal product sample (44 studies) categories generated using the random effects model is provided in Fig. . The highest prevalence of Campylobacter spp. has been observed in white meat (43.9%) from 55 studies among the meat and animal products that was reported in different studies from 0 to 90%. Campylobacter spp. prevalence in white meat was higher for chicken (48.6%) than other types of poultry meat (33.9%). Within the red meat category by 37 studies, Campylobacter spp. was detected at an overall pooled prevalence of 7.9% (Table ), which was reported from 0 to 24% in the literature. Campylobacter contamination in this category was mostly prevalent in buffalo (13.5%), followed by goat and sheep (8.6%), cattle (8.4%) and camel (2.5%) meat. While among animal products eggs were found to have a 5.5% prevalence of Campylobacter spp. contamination, with a high rate of contamination prevalence being observed for chicken eggs (9.9%) in eight studies compared to eggs of other types of poultry (4.2%) from 24 studies. The prevalence of Campylobacter spp. contamination detected among environmental samples was 14.9%. Vegetables were constituted environmental samples that showed highest prevalence (19.4%) of Campylobacter contamination. Water and sewage samples had prevalence of 15.4% and 7.4%, respectively. As the I2 heterogeneity index was more than 50, there was heterogeneity in the included studies. Campylobacter spp. in fecal samples Literature review of 79 studies that investigated the fecal samples in animal and human revealed that pooled proportion of Campylobacter spp. was 18.7% in fecal samples. Among food animals, poultry had the highest contamination of fecal samples (46.8%). Domestic and wild animal had 21% and 14.1% contamination of Campylobacter spp. (Table ). A proportion of 8.4% of human samples were positive regarding Campylobacter spp. Campylobacter spp. by place of sampling Table presents an overview from the meta-analysis of Campylobacter spp. prevalence from Iran based on sampling places. Poultry feces (61.9%) and white meat (47.2%) were determined to have the highest Campylobacter spp. prevalence at the slaughterhouse. This was followed by white meat at market (42.6%) and farm (40%) levels. The lowest pooled prevalence of Campylobacter spp. was observed for milk sampled at farm (1%) and market (3.3%) levels, eggs sampled at market (5.4%) and red meat sampled at slaughterhouse (6.2%) levels. Campylobacter spp. prevalence in white and red meat, and milk samples at markets (sampled from retails, supermarkets and butcher’s) was higher than at farms (Table ). Considerable proportions of wild animal (prevalence of 25.4%) and dog and cat feces (prevalence of 20.4%), were found to be contaminated with Campylobacter spp.. C. jejuni and C. coli As the C. jejuni and C. coli are the main causative agents of human campylobacteriosis, the pooled prevalence of these two species was determined in Iran samples. Most of the studies reported the prevalence of C. jejuni and C. coli in their samples. C. jejuni had higher pooled prevalence/proportion than C. coli in all of the obtained samples except for those derived from vegetables. Sewage (100%) (one study), milk (86.6%) (7 studies), human feces (83.3%) (33 studies) and water (82.8%) (3 studies) samples had the most frequent contaminations with C. jejuni (Fig. ) . Pooled C. jejuni prevalence in white meat (54 studies), egg (28 studies), poultry feces (19 studies) and red meat (35 studies) was 68.7%, 65.5%, 65.2% and 62.7%, respectively. Vegetable (2 studies) samples had the least pooled prevalence of C. jejuni (28%). On the other hand the highest pooled prevalence of C. coli was reported in vegetable samples (72%) followed by egg (33%) and red meat (24.1%) samples. Pooled prevalence of C. coli was zero (95%CI: 0–84.2%) in sewage samples (Fig. ). Campylobacter spp. Despite the high number of studies that reported the prevalence of Campylobacter spp., a limited number of them investigated the virulence genes required for pathogenesis. CdtA, cdtB, cdtC, cadF and pldA had the highest number of investigated studies. Figure shows the proportion of virulence genes in Campylobacter spp. cadF (97%) had the highest pooled prevalence in Campylobacter spp. in 28 studies, followed by racR (93.8%) (3 studies) and flaA (91.3%) (17 studies). VirB11 had the least prevalence (0%) in the Campylobacter spp. in 11 investigated studies. A total of 31% of Campylobacter spp. contained wlaN in 7 studies. With the sensitivity analysis, it was found that one of the studies pulls the results towards itself. The virB11 gene has the greatest impact on heterogeneity. Campylobacter spp. are regarded as the commonest cause of bacterial human gastroenteritis around the world . In the present study, we tried to determine the prevalence of Campylobacter spp. in the food, animal and human samples of Iran based on systematic review of studies published from the country. Our findings showed that in Iran, white meat including, chicken and poultry accounts for the highest pooled prevalence of Campylobacter spp. These results are consistent with high average Campylobacter contamination prevalence that has also been observed for broiler chicken (36.7%) and turkey (11.0%) meat in Europe as reported by the European Food Safety Authority . Campylobacter spp. (33.3%) represented the second most prevalent bacterial contamination of poultry meat based on a systematic review of European surveys . As much as 48.6% of chicken and 23% of other poultry meat samples in Europe were contaminated with Campylobacter spp. . Frequency of Campylobacter spp. contamination in chicken was reported as 99.5% in Italy, 93.7% in Northern Ireland, 84% in Ireland, 82% in Switzerland, 56% in Turkey, 53% in Spain, 51% in Austria, 50% in Poland, 14.9% in Sweden, and 9.7% in Romania . In Portugal 40.3% of fresh broiler meat samples were reported to be contaminated with Campylobacter spp. . Our analysis in this review shows that about 76% of broiler flocks in Shiraz, Iran were positive for Campylobacter. C. jejuni accounted for 22% whereas C. coli for 32% of the Campylobacter positive chicken samples . The current study revealed a higher prevalence of C. jejuni than C. coli in white meat of Iran. Poultry carcasses had 35.37% and 19.82% prevalence of C. jejuni and C. coli contaminations, respectively from the slaughterhouses of Jahrom-Iran . Campylobacter was recovered from 49.2% of poultry liver, 42.8% of gizzard 33.3% of heart and 25.4% of meat from poultry slaughterhouses at West Azerbaijan, Iran . The quail meat had the highest contamination (68.4%) with Campylobacter spp. followed by chicken (56.1%), turkey (27.4%) and ostrich meat (11.7%). The high contamination of quail meat could be due to handling in slaughtering and packaging procedure that leads to higher cross–contamination . The total prevalence of Campylobacter spp. in poultry meat sampled from Isfahan was 47.1% . Meanwhile about 55.4% of hen carcasses sampled in processing plant of Ahvaz, Iran, were contaminated with Campylobacter spp. . Turkey samples had contamination with Campylobacter spp. (62.1%) . Duck samples were more contaminated (39.2%) than goose samples (26.1%) . Hen liver had the highest frequency of Campylobacter spp. (63.6%), then was turkey (40%) and ostrich liver (16.7%) . Liver was more contaminated with Campylobacter spp. than meat . Recovery of Campylobacter was more in chicken (63%) than beef (10%) . Sheep meat (3.10%) was the most contaminated in the meat samples followed by chicken (2.40%), beef (1.80%), and buffalo meat (1.10%) from Khuzestan. 81.30% of the isolates were C . jejuni and 18.70% were C. coli . Campylobacter was detected in 49.5% of chicken and 8% of beef samples . Lamb meat had the highest prevalence (12%) of Campylobacter spp. followed by goat (9.4%), beef (2.4%) and camel meat (0.9%) in Isfahan and Yazd, which was according to the present study. Higher contamination of lamb and goat meat revealed the effect of manual skinning, evisceration and processing in abattoir and inadequate hygiene in transport, storage and cutting of meat in local butcheries. Lower rate of contamination of camel milk may be related to high number of homogenic bacteria in rumen of camel and H2 accumulation that leads to destroying of campylobacter . In a study that examined individual unpasteurized bovine and ovine milk samples from Zanjan, Iran, Haghi et al. detected no Campylobacter contamination, which was in contrast to most of other studies covered in the current meta-analysis and it could be due to that other studies examined bulk milk, but Haghi et al. investigated individual milk. Campylobacter spp. isolated from 2.5% to 12.5% of milk samples in Mazandaran, Isfahan and Mashhad. C. jejuni was detected in 2.5% to 13.88% of these milk samples [ , , , ]. Results of the current study showed 5.5% detection of Campylobacter spp. in eggs. Another study showed 7% contamination of eggshell of hen, 5% of duck’s eggshell, 3.3% of goose, 2.5% of ostrich, 4.2% of partridge, 5% of quail and 3.8% of turkey’s eggshell to Campylobacter spp. . Prevalence of C. jejuni (6.3%) was more than C. coli (1.3%) in avian eggs which was according to present study. Safaei et al. observed no C. jejuni in table eggs. 18.67% to 31.6% of eggshell were contaminated with Campylobacter spp. . Examination of cecal contents of poultry conducted in Kurdistan revealed that 55% of samples were contaminated with Campylobacter spp. that included C. jejuni (86.2%) and C. coli (13.7%) . Similar prevalence levels have also been reported in Iran based on literature reviewed here that found C. jejuni is more frequent than C. coli in poultry feces. Khoshbakht et al. reported 67.8% of Campylobacter spp. in cattle and sheep fecal samples of Shiraz, which was higher than current study. C. jejuni and C. coli were seen in 78.5% of the samples simultaneously. Moreover, 2.9% and 12.6% of the samples were positive for C. coli and C. jejuni , respectively . Prevalence studies conducted in Isfahan detected Campylobacter spp. in 10%, 8%, 5.3% and 4% of sheep, goat, cattle and camel feces . Salari et al. (2020) observed no C. jejuni in Crested lark . About 33% of pet bird feces were contaminated with Campylobacter spp. . C. jejuni was detected in 48.62% of bird feces . 52.3% of Persian fallow deer fecal samples which were collected from Dasht-e-Arzhan located in southwest of Iran, were contaminated with Campylobacter spp. , which was higher than the present study. Most of the studies reported higher prevalence of C. jejuni than C. coli in the foodstuffs [ , , , , , , , , , , , , , , , , , , , ] and fecal samples [ , , , , , , ]. Among environmental samples examined from northern Iran, the prevalence of Campylobacter spp. was higher in river water (36.92%) than fecal samples of poultry (34.88%), cow (28.57%), horse (20%) and sheep (9%) origin. The lowest contaminated environmental samples were those of sewage (7.4%) origin . A study that have examined Caspian Sea’s water reported a Campylobacter spp. contamination prevalence of 2.66% . In the investigation of vegetable samples, 15% of mushrooms in Shahrekord had Campylobacter spp. contamination . Campylobacter spp. was detected in 3.5% of leafy vegetables marketed in Tehran . These different reported rate of contamination could be due to the difference of geographical location and season of sampling, type and number of the samples, method of isolation, and different sanitary situation on farms and slaughterhouses . Our current study found that human diarrheal samples examined from Iran had a pooled Campylobacter spp. prevalence of 8.4%. Studies from central Iran reported that 33% of infectious diarrheal samples were positive for C. jejuni . Among acute diarrhea samples examined in Tehran, Campylobacter spp. were detected in 8.6% of the samples of which 69.5% were C. jejuni and 24.5% was C. coli . Jafari et al., studied the prevalence of Campylobacter spp. in children under five years of age with acute diarrhea in Tehran. They found campylobacter in 5.5% of patients, equal to 10.8% of all isolated bacteria. In Shiraz ~ 9.6% of acute diarrhea samples were positive for C. jejuni . 4% of fecal samples were contaminated with Campylobacter spp. . 9.8% of diarrheic children was positive for C. jejuni . C. jejuni was the major species recovered from human samples . Pathogenesis of Campylobacter was associated with some virulence genes. cadF, flaA, and ciaB genes are essential virulence factors for adhesion and colonization of Campylobacter to epithelial cells in human intestine . Some studies observed 100% prevalence of cadF virulence gene in C. jejuni [ , , , , ] and C. coli isolates which was agreed with the current study. The CDT toxin leads to cell cycle arrest and promotes DNA damage; so, its presence is related with the severity of the campylobacteriosis . Prevalence of cdtA, cdtB, cdtC, pldA, and iamA genes were 97%, 97%, 96%, 72%, and 60%, respectively in the isolates , which was higher than the current study. Prevalence of cdtA, cdtB, cdtC, racR and pldA was observed 100% in some studies [ , , , , , ]. VirB11 gene was not detected in any of the strains that was according to present study and could be related to the plasmid nature of this gene . Guillain–Barre’ and Miller-Fischer syndromes are associated with wlaN, cgtB genes and waaC gene . Prevalence of other genes including iamA , and wlaN , was reported as 81.11%, and 82.22%, respectively , which was higher than current meta-analysis. Frequency of cgtB genes was observed as 22.22% that was lower than present study. Frequency of ciaB was reported in 76.92% of poultry, 55.56% of cow and 100% of sheep fecal samples . pldA and cgtB were detected in raw chicken Campylobacter isolates in Shiraz as 65.4% and 15.4%, respectively . Prevalence of dnaJ was from 11 to 100% in different samples . WaaC was detected in 100% of food isolates of C. jejuni and 75.6% of C. coli . Campylobacter food isolates carried most of the virulence genes essential for pathogenesis that shows the high risk of these isolates for human. Prevalence of Campylobacter spp. contamination was higher at market than farm level in Iran as determined in the present study, which is similar to observations from previous studies conducted in other countries . Gonçalves-Tenório et al. reported higher prevalence of Campylobacter spp. (44.3%) contamination at retail level than at the end-processing (30.7%) stage in poultry meat. Campylobacter spp. are able to colonize and attach to tissues of poultry during processing . Carcass processing in the slaughterhouse including, scalding, washing and cooling was found not to decrease the level of Campylobacter spp. contamination of poultry meat . Freezing significantly decreased chicken contamination with Campylobacter spp. during processing of poultry carcasses from 80 to 30% . Washing reduced the contamination of sheep carcass from 10% after hiding to 8% after washing . Since farms are considered as the initial site of contamination with Campylobacter, most preventive strategies must therefore be implemented at farm level by increasing of biosecurity and enhancing monitoring . The higher contamination observed at market level may be due to uncontrolled temperature during transport of meat . Poultry are regarded as a major source of this organism due to their carriage of Campylobacter spp. in the intestinal tract . Similarly we also found here that poultry samples in Iran including meat and feces are associated with higher Campylobacter spp. contamination. The handling and preparation of broiler meat led to cross-contamination of poultry meat and is considered as contributing cause for one-third of human campylobacter infection in Europe while the remaining cases are related to the self-contamination of chicken with Campylobacter as the reservoir of the organism . Establishing if such a link also exists in Iran is rather difficult due to the fact that there is currently neither notification nor investigation of food vehicles of human campylobacteriosis. In conclusion the current systematic review and meta-analysis of Campylobacter prevalence shows that chicken has great concern for Campylobacter carriage in Iran. This must be considered in preparation of undercooked poultry such as barbecue. Most of the isolated Campylobacter carried virulence associated genes that show their potential pathogenicity. Since our analysis showed that the gastrointestinal tract and slaughtering facilities are among the main sources of Campylobacter contamination for poultry meat in Iran, implementing preventive and corrective actions at several stages mainly at farm level is very vital. Implementing control strategies specifically for this pathogen will have a remarkable impact on its incidence and production of safer meat for consumers. Moreover, consumer education in hand hygiene, sanitation of surfaces prior to and after handling meat, separation of raw and cooked meat and checking the temperature of refrigerator is also needed to reduce contamination and infections with this pathogen. Search strategy A systematic search was performed in PubMed, Scopus, and Web of Science electronic databases in papers that were published from November of 2021 to the end of January 2022. The search keyword was “ Campylobacter coli “ or “ Campylobacter jejuni ” combined with the following terms: “Food”, “Animal”, “Chicken”, “Poultry”, “Meat”, “Beef”, “Lamb”, “Fish”, “Milk”, “Dairy”, “Egg”, “Sheep”, “Goat”, “Avian”, “Cow”, “Cattle”, “Human”, “Feces”, “Diarrhea”, “Gastroenteritis “ and “Iran” ( ). Handmade search was performed in Google Scholar and scientific information database (SID). PRISMA guidelines were used to perform the systematic reviews. Selection criteria and quality assessment Selection of studies were performed by these inclusion criteria: research studies including original article either published or in press; studies with a cross-sectional design to detect Campylobacter on the samples based on culture or PCR; had a known sample size; and studies with available full-text. Title and abstracts of the searched papers were assessed to identify articles that matched with the inclusion criteria. In some circumstances full texts were evaluated. The exclusion criteria include articles that did not follow standard methods, duplicate articles and reports, studies with unclear or incomprehensible text and analysis, articles that did not report the exact sample size and number /percent of Campylobacter . Positive samples Reviews; letters or editorial articles without original data were also excluded. Quality assessment of the eligible studies were performed by Joanna Briggs Institute . Articles which gained 6 score (from 10) were eligible for data extraction. When two reviewers (EA and TZ) were disagreed about an article, seek the opinion of third reviewer (PS). Duplicates articles were removed by help of Endnote reference manager and also some of them were found by manual check. Data extraction Data extraction forms were designed in Microsoft Excel. Articles that obtained more than 60% of quality score were eventually included in the analysis as they were meet 6 out of 10 criteria of Joanna Briggs checklist. Following information was collected from the included studies: the first author’s name, date of publication, study design, study location, number of samples, source of samples (animal, human and environment), sample group (meat, food product?, feces and environment) and type of samples (human, domestic animal, wild animal, poultry, white meat, red meat, milk, egg, water, sewage, vegetable), sample species (chicken, poultry white meat, cattle, goat, sheep, camel and other red meat, hen egg and poultry egg), place of sampling (hospital, pet clinic, slaughterhouse, farm, market and environment), diagnostic technique (Culture, PCR, culture and PCR), prevalence of Campylobacter spp., C. jejuni , C. coli, virulence factors and quality score. Statistical analysis In this study, the data analysis was done with STATA 14 software (STATA Corp., College Station, Texas) with metaprop command. A random effect model was applied to determine the pooled prevalence and 95% Confidence interval of Campylobacter spp.. A forest plot was used to calculate the pooled prevalence with 95% confidence intervals. Statistical heterogeneity among studies was evaluated by computing I 2 , Cochran’s Q. 25%, 50%, and 75% of I 2 values are classified as low, medium, and high heterogeneity, respectively. A subgroup analysis, sensitivity analysis, and meta-regression were performed on the basis of publication year, and type of sampling to evaluate sources of heterogeneity. A systematic search was performed in PubMed, Scopus, and Web of Science electronic databases in papers that were published from November of 2021 to the end of January 2022. The search keyword was “ Campylobacter coli “ or “ Campylobacter jejuni ” combined with the following terms: “Food”, “Animal”, “Chicken”, “Poultry”, “Meat”, “Beef”, “Lamb”, “Fish”, “Milk”, “Dairy”, “Egg”, “Sheep”, “Goat”, “Avian”, “Cow”, “Cattle”, “Human”, “Feces”, “Diarrhea”, “Gastroenteritis “ and “Iran” ( ). Handmade search was performed in Google Scholar and scientific information database (SID). PRISMA guidelines were used to perform the systematic reviews. Selection of studies were performed by these inclusion criteria: research studies including original article either published or in press; studies with a cross-sectional design to detect Campylobacter on the samples based on culture or PCR; had a known sample size; and studies with available full-text. Title and abstracts of the searched papers were assessed to identify articles that matched with the inclusion criteria. In some circumstances full texts were evaluated. The exclusion criteria include articles that did not follow standard methods, duplicate articles and reports, studies with unclear or incomprehensible text and analysis, articles that did not report the exact sample size and number /percent of Campylobacter . Positive samples Reviews; letters or editorial articles without original data were also excluded. Quality assessment of the eligible studies were performed by Joanna Briggs Institute . Articles which gained 6 score (from 10) were eligible for data extraction. When two reviewers (EA and TZ) were disagreed about an article, seek the opinion of third reviewer (PS). Duplicates articles were removed by help of Endnote reference manager and also some of them were found by manual check. Data extraction forms were designed in Microsoft Excel. Articles that obtained more than 60% of quality score were eventually included in the analysis as they were meet 6 out of 10 criteria of Joanna Briggs checklist. Following information was collected from the included studies: the first author’s name, date of publication, study design, study location, number of samples, source of samples (animal, human and environment), sample group (meat, food product?, feces and environment) and type of samples (human, domestic animal, wild animal, poultry, white meat, red meat, milk, egg, water, sewage, vegetable), sample species (chicken, poultry white meat, cattle, goat, sheep, camel and other red meat, hen egg and poultry egg), place of sampling (hospital, pet clinic, slaughterhouse, farm, market and environment), diagnostic technique (Culture, PCR, culture and PCR), prevalence of Campylobacter spp., C. jejuni , C. coli, virulence factors and quality score. In this study, the data analysis was done with STATA 14 software (STATA Corp., College Station, Texas) with metaprop command. A random effect model was applied to determine the pooled prevalence and 95% Confidence interval of Campylobacter spp.. A forest plot was used to calculate the pooled prevalence with 95% confidence intervals. Statistical heterogeneity among studies was evaluated by computing I 2 , Cochran’s Q. 25%, 50%, and 75% of I 2 values are classified as low, medium, and high heterogeneity, respectively. A subgroup analysis, sensitivity analysis, and meta-regression were performed on the basis of publication year, and type of sampling to evaluate sources of heterogeneity. Additional file 1.
Endotracheal suture through extending tracheostoma for post-tracheostomy tracheal laceration: a case report	19e5f310-2096-4fa1-a2e3-e98adade37a9	10170781	Suturing[mh]	Tracheal laceration is a rare complication that can occur during intubation or tracheostomy . Most lacerations occur longitudinally, at a membranous portion (a connection between cartilage rings) of the trachea . The incidence of tracheal laceration is approximately 1 per 2000 intubations; however, the exact statistical probability of all tracheal lacerations cannot be defined [ , , ]. If tracheal laceration is suspected, emergency bronchoscopy should be performed for diagnosis . Treatment is tailored according to the location and size of the torn tracheal wall . Unless early correction through surgical repair is performed, tracheal laceration can progress to various catastrophic situations . Herein, we report a case of post-tracheostomy tracheal laceration that was successfully managed using extended tracheostomy. We report the case of a 30-year-old Asian woman, with no past medical history, diagnosed with locked-in syndrome due to multiple postpartum infarctions. Intracerebral hemorrhage and intraventricular hemorrhage were confirmed during follow-up, after which she was admitted to the intensive care unit. Long-term mechanical ventilation care was expected after intubation; the patient was thus referred to the thoracic surgery department for tracheostomy. Tracheostomy was performed in the usual manner (Fig. ). In this case, complications such as subcutaneous emphysema or pneumo-mediastinum did not occur; however, tidal volume did not shift to the lung once the tube was connected to a mechanical ventilator. After noting that the ventilator was not working appropriately, the endotracheal tube was reinserted through the oral orifice. Then, through bronchoscopy, we found that the posterior tracheal wall was torn. The tearing wound was 5–6 cm in length, from the middle to distal parts of the trachea (approximately, third to fifth tracheal rings). At this point, following general guidelines, we should have considered right thoracotomy for management. However, we performed extending tracheostoma technique and endotracheal suture. After examining the torn tracheal wall via bronchoscopy (Fig. ), the endotracheal (ETT) tube was lifted from tracheostoma level, and the tracheostoma was extended on both sides to expose the laceration area. After the area was confirmed, under direct vision, suturing was started from the superior part, with vicryl 4–0, using an open needle holder. While lifting the thread, the trachea under the opening level was exposed as much as possible, and continuous suture was performed downward. After suturing was complete, the tracheostomy tube was placed slightly below the lesion, facilitating mechanical ventilation care. After 7 days, antibiotics for pneumonia were discontinued. There were no postoperative complications (Fig. ). On day 13, the patient was moved to the general ward, and her first tracheostomy tube change was performed 2 weeks later. No oxygen requirements were present. She was then transferred to a rehabilitation hospital. Tracheal laceration is a rare complication that occurs during the process of intubation or tracheostomy. It can have fatal consequences . Tracheal laceration can be diagnosed via bronchoscopy, which can confirm the location and size of the tear wound in the tracheal wall [ – ]. Even if the lesion cannot be visually confirmed by bronchoscopy, a lesion should be suspected if subcutaneous emphysema, pneumothorax, or pneumo-mediastinum suddenly occurs [ – ]. If mechanical ventilation is ineffective, despite a well-placed tracheostomy tube (as in our case), examinations must be performed to identify whether the tracheal wall has been torn. If tracheal laceration management is not performed in a timely manner, laceration may progress to descending mediastinitis and worsen the patient’s clinical condition . Treatment of tracheal lacerations is divided into conservative versus surgical treatments, depending on the size and location of the laceration involved [ , , ]. Conservative treatment can be used if the longitudinal length of the laceration is shorter than 2 cm and the patient’s vital signs are stable. Timely surgical treatment should be performed if the longitudinal length of the laceration is longer than 2 cm and the patient’s vital signs are unstable . Since the surgical approach is dependent on the size and location of the tear wound, if the upper and middle parts of the trachea are affected, left cervical incision or trans-tracheal repair is chosen; if the distal trachea or main bronchus are affected, right thoracotomy is generally chosen [ , , ]. In this report, making a new incision was not necessary due to an already existing tracheal opening. The tracheostoma was extended bilaterally, exposing the lacerated tracheal wall, and endotracheal suture was performed. During the suturing process, clashes between instruments could occur and it might be difficult to expose the lesion. However, by extending the previous incision, the tearing point can be repaired under direct vision. In addition, tagging of the first suture site helped us maintain tension, secure the previous suture, and easily suture the lesion continuously with a “no touch” technique. The minimally invasive method performed in this case might be a useful treatment for patients who are expected to have long-term mechanical ventilation care. Endotracheal suture through tracheostoma could reduce the incidence of additional morbidity or mortality caused by general anesthesia, broadening treatment options in cases of post-tracheostomy tracheal laceration.
Treatment of FIGO 2018 stage IIIC cervical cancer with different local tumor factors	52b1362f-3708-41e8-8c6d-89792b91e278	10170857	Internal Medicine[mh]	CC is the predominant genital-tract malignancy in women. Tumor staging reflects both prognosis and survival, and is crucial for the management of tumor and treatment guidance. International Federation of Gynecology and Obstetrics (FIGO) staging is widely used for CC. In 2018, FIGO released a new staging process , which was revised in 2019 . It was mainly updated to add stage IIIC, referring to involvement of the pelvic lymph nodes(LNs) and/or para-aortic LNs regardless of tumor size and extent of spread. Certain scholars dispute stage IIIC prognosis, which is solely based on the lymph-node-metastasis stage, without considering the local invasion scope, tumor size, and other factors . The “2022 NCCN Cervical Cancer Clinical Practice Guidelines (1st Edition)” recommend R-CT for stage IIIC ; therefore, patients with relatively localized tumors are denied surgery. Although FIGO staging is recommended by most medical guidelines, the European Society of Gynecologic Oncology, European Society of Radiotherapy and Oncology, and European Society of Pathology guidelines recommend tumor-node-metastasis (TNM) staging. Therefore, certain scholars have raised the following question: “Using T-staging, are the oncological outcomes of different treatments in patients with FIGO 2018 stage IIIC consistent?” To date, only a small number of studies have focused on this stage, using small sample sizes and few stratified comparisons. Therefore, based on the 1538 project database, our study screened FIGO 2018 stage IIIC cases using T-staging stratification, that is, according to T1, T2a, T2b, and T3 stratification (corresponding to FIGO 2009 stages I, IIA, IIB, and III cases with LN metastasis), to investigate the survival outcomes of ARH, NACT, and R-CT for the selection of appropriate treatment strategies for FIGO 2018 stage IIIC CC. Data collection Chinese Cervical Cancer Clinical (Four-C) study is a multi-center, retrospective cohort study, including 63,926 patients with various stages of CC who were hospitalized in 47 Chinese hospitals during 2004–2018. This study was approved by the Ethics Committee of Nanfang Hospital, Southern Medical University (ethical number: NFEC-2017–135, Clinical trial Registration Number: CHiCTR1800017778). We collect the following data by reviewing electronic medical records: general clinical data, preoperative laboratory test results, preoperative pathological results, relevant surgical data, preoperative adjuvant treatment data, postoperative adjuvant therapy data, postoperative pathological results, and follow-up data. We phoned to follow up and obtained messages of survival, recurrence, and complications, if failed, we obtained information from inpatient and outpatient medical records. All original data were reviewed and validated by two independent gynecologists to ensure the accuracy. The details of case collection in the Four-C study database is as shown in our previous studies . In accordance with the journal’s guidelines, we will provide our data for the reproducibility of this study in other centers if such is requested. Inclusion and exclusion criteria TNM-staging tumor factors were used to divide IIIC-stage patients into four groups: T1, T2a, T2b, and T3. The following criteria were for inclusion: age ≥ 18 years; CC diagnosed by cervical biopsy; squamous cell carcinoma(SCC), adenocarcinoma(AC), or adenosquamous cell carcinoma(ASC) based on histology; FIGO 2018 stage IIIC (patients who underwent radiation therapy defined lymph node status depended on CT, MRI or/and PET-CT before treatment, and those who underwent radical surgery defined lymph node status depended on pathological examinations after surgery); ARH-group patients who underwent radical surgery (Q-M type-B or type-C radical hysterectomy pelvic lymphadenectomy ± para-aortic lymphadenectomy); NACT-group patients who underwent neoadjuvant therapy and radical surgery (Q-M type-B or type-C radical hysterectomy pelvic lymphadenectomy ± para-aortic lymphadenectomy); R-CT-group patients who underwent radiation therapy(RT),and RT dose ≥ 45 Gy; with follow-up outcomes. The following criteria were for exclusion: gestation, cervical stump cancer, CC complicated with other malignant tumors, loss to follow-up, and failure to satisfy the inclusion criteria. Outcome measurement The main outcome measures were overall survival (OS) and disease-free survival (DFS) in all subgroups of FIGO 2018 stage IIIC, with the cut-off point 5 years after treatment. The concept of OS is the final time from diagnosis to effective follow-up or death from any cause, moreover the concept of DFS is the final time from diagnosis to effective follow-up, recurrence or death. Statistics No data were missing among the included cases. Mean ± standard deviation (x ± s) was used to represent continuous data, furthermore percentage was used to represent counting data. Fisher's exact test or Chi-square test were used to compare categorical variables. Kaplan–Meier curves were used to describe changes in survival outcomes. Cox proportional risk regression models were used to adjust variables and evaluate the HRs and 95% CI of stratification for 5-year OS and DFS. SPSS 26.0 (SPSS, Inc., Chicago, IL, USA) was used for statistical analyses, and statistical significance was set at P < 0.05. Chinese Cervical Cancer Clinical (Four-C) study is a multi-center, retrospective cohort study, including 63,926 patients with various stages of CC who were hospitalized in 47 Chinese hospitals during 2004–2018. This study was approved by the Ethics Committee of Nanfang Hospital, Southern Medical University (ethical number: NFEC-2017–135, Clinical trial Registration Number: CHiCTR1800017778). We collect the following data by reviewing electronic medical records: general clinical data, preoperative laboratory test results, preoperative pathological results, relevant surgical data, preoperative adjuvant treatment data, postoperative adjuvant therapy data, postoperative pathological results, and follow-up data. We phoned to follow up and obtained messages of survival, recurrence, and complications, if failed, we obtained information from inpatient and outpatient medical records. All original data were reviewed and validated by two independent gynecologists to ensure the accuracy. The details of case collection in the Four-C study database is as shown in our previous studies . In accordance with the journal’s guidelines, we will provide our data for the reproducibility of this study in other centers if such is requested. TNM-staging tumor factors were used to divide IIIC-stage patients into four groups: T1, T2a, T2b, and T3. The following criteria were for inclusion: age ≥ 18 years; CC diagnosed by cervical biopsy; squamous cell carcinoma(SCC), adenocarcinoma(AC), or adenosquamous cell carcinoma(ASC) based on histology; FIGO 2018 stage IIIC (patients who underwent radiation therapy defined lymph node status depended on CT, MRI or/and PET-CT before treatment, and those who underwent radical surgery defined lymph node status depended on pathological examinations after surgery); ARH-group patients who underwent radical surgery (Q-M type-B or type-C radical hysterectomy pelvic lymphadenectomy ± para-aortic lymphadenectomy); NACT-group patients who underwent neoadjuvant therapy and radical surgery (Q-M type-B or type-C radical hysterectomy pelvic lymphadenectomy ± para-aortic lymphadenectomy); R-CT-group patients who underwent radiation therapy(RT),and RT dose ≥ 45 Gy; with follow-up outcomes. The following criteria were for exclusion: gestation, cervical stump cancer, CC complicated with other malignant tumors, loss to follow-up, and failure to satisfy the inclusion criteria. The main outcome measures were overall survival (OS) and disease-free survival (DFS) in all subgroups of FIGO 2018 stage IIIC, with the cut-off point 5 years after treatment. The concept of OS is the final time from diagnosis to effective follow-up or death from any cause, moreover the concept of DFS is the final time from diagnosis to effective follow-up, recurrence or death. No data were missing among the included cases. Mean ± standard deviation (x ± s) was used to represent continuous data, furthermore percentage was used to represent counting data. Fisher's exact test or Chi-square test were used to compare categorical variables. Kaplan–Meier curves were used to describe changes in survival outcomes. Cox proportional risk regression models were used to adjust variables and evaluate the HRs and 95% CI of stratification for 5-year OS and DFS. SPSS 26.0 (SPSS, Inc., Chicago, IL, USA) was used for statistical analyses, and statistical significance was set at P < 0.05. Based on the inclusion criteria, 4,086 CC cases (including 1,117, 1,019, 869, and 1,081 cases in the T1, T2a, T2b, and T3 groups, respectively) were included. Data screening process is shown in Fig. . Clinicopathological characteristics of each group Clinicopathological characteristics of each group are shown in Table . Clear distinctions were shown among the median age of the patients using the three treatments in the T1, T2a, T2b, and T3 groups ( P < 0.05 for all). No sharp distinctions were observed in the proportion of histologic types between the T1 and T2a groups ( P > 0.05 for all); however, a statistically significant distinction was shown between the T2b and T3 groups ( P < 0.001 for all). Comparison of oncological outcomes of the three treatment methods in each group Comparison of oncological outcomes of the three treatment methods in the T1 group In the T1 group, 1,117 cases were divided into R-CT ( n = 58), ARH ( n = 838), and NACT ( n = 221) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year OS (76.5% vs . 81.7% vs . 75.3%, P = 0.015) and DFS ( 73.6% vs .74.3% vs . 60.1%, P = 0.001) among the three subgroups (Fig. A, B). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 1.040, 95% CI: 0.481–2.249, P = 0.921) or DFS (HR = 1.227, 95% CI: 0.642–2.344, P = 0.535) than R-CT. Furthermore, ARH was not correlated with 5-year OS (HR = 0.637, 95% CI: 0.307–1.322, P = 0.226) and 5-year DFS (HR = 0.741, 95% CI: 0.400–1.375, P = 0.343) than R-CT (Table ). However, NACT was correlated with a decrease in OS (HR = 1.631, 95% CI: 1.150–2.315, P = 0.006) and DFS (HR = 1.665, 95% CI: 1.255–2.182, P < 0.001) than ARH (Table ). AC was associated with a decrease in 5-year OS (HR = 1.734, 95% CI: 1.092–2.573, P = 0.020) and 5-year DFS (HR = 1.786, 95% CI: 1.241–2.570, P = 0.002) than SCC; nonetheless, ASC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Moreover, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Tables and ). Comparison of oncological outcomes of the three treatments in the T2a group In the T2a group, 1,019 cases were divided into R-CT ( n = 157), ARH ( n = 608), and NACT ( n = 254) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year DFS (69.1% vs . 64.9% vs . 54.5%, P = 0.005) among the three subgroups; however, there was no distinctions in 5-year OS ( 76.3% vs . 75.8% vs . 69.5%, P = 0.079) (Fig. C, D). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 1.339,95% CI:0.800–2.243, P = 0.267) or DFS (HR = 1.425,95% CI:0.950–2.139, P = 0.087) than R-CT. Furthermore, ARH was not correlated with 5-year OS (HR = 0.921, 95% CI: 0.572–1.484, P = 0.736) and 5-year DFS (HR = 0.932, 95% CI: 0.640–1.357, P = 0.714) than R-CT (Table ). Nevertheless, NACT was correlated with a decrease in OS (HR = 1.454, 95% CI: 1.057–2.000, P = 0.021) and DFS (HR = 1.529, 95% CI: 1.185–1.974, P = 0.001) than ARH (Table ). ASC was associated with a decrease in 5-year OS (HR = 3.182, 95% CI: 1.803–5.617, P < 0.001) and 5-year DFS (HR = 2.210, 95% CI: 1.288–3.792, P = 0.004) than SCC; nonetheless, AC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Moreover, age was not associated with 5-year OS and DFS ( P > 0.05 for all) (Tables and ). Comparison of the oncological outcomes of the three treatments in the T2b group In the T2b group, 869 cases were divided into R-CT ( n = 645), ARH ( n = 73), and NACT ( n = 151) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year OS (73.9% vs . 86.3% vs . 68.5%, P = 0.011) and DFS (72.6% vs . 64.0% vs . 47.6%, P < 0.001) among the three subgroups (Fig. E, F). Cox proportional hazards regression analysis revealed NACT correlated with a decrease in 5-year DFS (HR = 1.847, 95% CI: 1.347–2.532, P < 0.001) than R-CT; however, there was no correlation with OS (HR = 1.353, 95% CI: 0.901–2.032, P = 0.146). ARH was not correlated with 5-year OS (HR = 0.490, 95% CI: 0.236–1.017, P = 0.056) and DFS (HR = 1.101, 95% CI 0.702–1.725, P = 0.676). AC was correlated with a decrease in 5-year OS (HR = 1.999, 95% CI: 1.092–3.660, P = 0.025) and DFS (HR = 2.098, 95% CI: 1.298–3.392, P = 0.003) than SCC. However, ASC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Furthermore, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Table ). Comparison of the oncological outcomes of the three treatments in the T3 group In the T3 group, 1,081 cases were divided into R-CT ( n = 1,053), ARH ( n = 10), and NACT ( n = 18) subgroups. Kaplan–Meier analysis revealed no statistically sharp distinctions in 5-year OS (64.7% vs . 67.5% vs . 53.1%, P = 0.941) and DFS (61.1% vs . 68.6% vs . 45.5%, P = 0.761) among the three subgroups (Fig. G, H). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 0.780, 95% CI: 0.284–2.139, P = 0.629) and DFS (HR = 1.075, 95% CI: 0.499–2.316, P = 0.853) compared with R-CT. ARH was not correlated with 5-year OS (HR = 0.879, 95% CI: 0.281–2.746, P = 0.824) and DFS (HR = 0.747, 95% CI: 0.239–2.328, P = 0.614). ASC was correlated with a decrease in 5-year OS (HR = 2.733, 95% CI: 1.109–6.733, P = 0.029) than SCC; nevertheless, it was not correlated with 5-year DFS (HR = 2.156, 95% CI: 0.946–4.916, P = 0.068), and AC was not correlated with 5-year OS and DFS ( P > 0.05 for all). In addition, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Table ). Clinicopathological characteristics of each group are shown in Table . Clear distinctions were shown among the median age of the patients using the three treatments in the T1, T2a, T2b, and T3 groups ( P < 0.05 for all). No sharp distinctions were observed in the proportion of histologic types between the T1 and T2a groups ( P > 0.05 for all); however, a statistically significant distinction was shown between the T2b and T3 groups ( P < 0.001 for all). Comparison of oncological outcomes of the three treatment methods in the T1 group In the T1 group, 1,117 cases were divided into R-CT ( n = 58), ARH ( n = 838), and NACT ( n = 221) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year OS (76.5% vs . 81.7% vs . 75.3%, P = 0.015) and DFS ( 73.6% vs .74.3% vs . 60.1%, P = 0.001) among the three subgroups (Fig. A, B). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 1.040, 95% CI: 0.481–2.249, P = 0.921) or DFS (HR = 1.227, 95% CI: 0.642–2.344, P = 0.535) than R-CT. Furthermore, ARH was not correlated with 5-year OS (HR = 0.637, 95% CI: 0.307–1.322, P = 0.226) and 5-year DFS (HR = 0.741, 95% CI: 0.400–1.375, P = 0.343) than R-CT (Table ). However, NACT was correlated with a decrease in OS (HR = 1.631, 95% CI: 1.150–2.315, P = 0.006) and DFS (HR = 1.665, 95% CI: 1.255–2.182, P < 0.001) than ARH (Table ). AC was associated with a decrease in 5-year OS (HR = 1.734, 95% CI: 1.092–2.573, P = 0.020) and 5-year DFS (HR = 1.786, 95% CI: 1.241–2.570, P = 0.002) than SCC; nonetheless, ASC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Moreover, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Tables and ). Comparison of oncological outcomes of the three treatments in the T2a group In the T2a group, 1,019 cases were divided into R-CT ( n = 157), ARH ( n = 608), and NACT ( n = 254) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year DFS (69.1% vs . 64.9% vs . 54.5%, P = 0.005) among the three subgroups; however, there was no distinctions in 5-year OS ( 76.3% vs . 75.8% vs . 69.5%, P = 0.079) (Fig. C, D). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 1.339,95% CI:0.800–2.243, P = 0.267) or DFS (HR = 1.425,95% CI:0.950–2.139, P = 0.087) than R-CT. Furthermore, ARH was not correlated with 5-year OS (HR = 0.921, 95% CI: 0.572–1.484, P = 0.736) and 5-year DFS (HR = 0.932, 95% CI: 0.640–1.357, P = 0.714) than R-CT (Table ). Nevertheless, NACT was correlated with a decrease in OS (HR = 1.454, 95% CI: 1.057–2.000, P = 0.021) and DFS (HR = 1.529, 95% CI: 1.185–1.974, P = 0.001) than ARH (Table ). ASC was associated with a decrease in 5-year OS (HR = 3.182, 95% CI: 1.803–5.617, P < 0.001) and 5-year DFS (HR = 2.210, 95% CI: 1.288–3.792, P = 0.004) than SCC; nonetheless, AC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Moreover, age was not associated with 5-year OS and DFS ( P > 0.05 for all) (Tables and ). Comparison of the oncological outcomes of the three treatments in the T2b group In the T2b group, 869 cases were divided into R-CT ( n = 645), ARH ( n = 73), and NACT ( n = 151) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year OS (73.9% vs . 86.3% vs . 68.5%, P = 0.011) and DFS (72.6% vs . 64.0% vs . 47.6%, P < 0.001) among the three subgroups (Fig. E, F). Cox proportional hazards regression analysis revealed NACT correlated with a decrease in 5-year DFS (HR = 1.847, 95% CI: 1.347–2.532, P < 0.001) than R-CT; however, there was no correlation with OS (HR = 1.353, 95% CI: 0.901–2.032, P = 0.146). ARH was not correlated with 5-year OS (HR = 0.490, 95% CI: 0.236–1.017, P = 0.056) and DFS (HR = 1.101, 95% CI 0.702–1.725, P = 0.676). AC was correlated with a decrease in 5-year OS (HR = 1.999, 95% CI: 1.092–3.660, P = 0.025) and DFS (HR = 2.098, 95% CI: 1.298–3.392, P = 0.003) than SCC. However, ASC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Furthermore, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Table ). Comparison of the oncological outcomes of the three treatments in the T3 group In the T3 group, 1,081 cases were divided into R-CT ( n = 1,053), ARH ( n = 10), and NACT ( n = 18) subgroups. Kaplan–Meier analysis revealed no statistically sharp distinctions in 5-year OS (64.7% vs . 67.5% vs . 53.1%, P = 0.941) and DFS (61.1% vs . 68.6% vs . 45.5%, P = 0.761) among the three subgroups (Fig. G, H). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 0.780, 95% CI: 0.284–2.139, P = 0.629) and DFS (HR = 1.075, 95% CI: 0.499–2.316, P = 0.853) compared with R-CT. ARH was not correlated with 5-year OS (HR = 0.879, 95% CI: 0.281–2.746, P = 0.824) and DFS (HR = 0.747, 95% CI: 0.239–2.328, P = 0.614). ASC was correlated with a decrease in 5-year OS (HR = 2.733, 95% CI: 1.109–6.733, P = 0.029) than SCC; nevertheless, it was not correlated with 5-year DFS (HR = 2.156, 95% CI: 0.946–4.916, P = 0.068), and AC was not correlated with 5-year OS and DFS ( P > 0.05 for all). In addition, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Table ). In the T1 group, 1,117 cases were divided into R-CT ( n = 58), ARH ( n = 838), and NACT ( n = 221) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year OS (76.5% vs . 81.7% vs . 75.3%, P = 0.015) and DFS ( 73.6% vs .74.3% vs . 60.1%, P = 0.001) among the three subgroups (Fig. A, B). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 1.040, 95% CI: 0.481–2.249, P = 0.921) or DFS (HR = 1.227, 95% CI: 0.642–2.344, P = 0.535) than R-CT. Furthermore, ARH was not correlated with 5-year OS (HR = 0.637, 95% CI: 0.307–1.322, P = 0.226) and 5-year DFS (HR = 0.741, 95% CI: 0.400–1.375, P = 0.343) than R-CT (Table ). However, NACT was correlated with a decrease in OS (HR = 1.631, 95% CI: 1.150–2.315, P = 0.006) and DFS (HR = 1.665, 95% CI: 1.255–2.182, P < 0.001) than ARH (Table ). AC was associated with a decrease in 5-year OS (HR = 1.734, 95% CI: 1.092–2.573, P = 0.020) and 5-year DFS (HR = 1.786, 95% CI: 1.241–2.570, P = 0.002) than SCC; nonetheless, ASC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Moreover, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Tables and ). In the T2a group, 1,019 cases were divided into R-CT ( n = 157), ARH ( n = 608), and NACT ( n = 254) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year DFS (69.1% vs . 64.9% vs . 54.5%, P = 0.005) among the three subgroups; however, there was no distinctions in 5-year OS ( 76.3% vs . 75.8% vs . 69.5%, P = 0.079) (Fig. C, D). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 1.339,95% CI:0.800–2.243, P = 0.267) or DFS (HR = 1.425,95% CI:0.950–2.139, P = 0.087) than R-CT. Furthermore, ARH was not correlated with 5-year OS (HR = 0.921, 95% CI: 0.572–1.484, P = 0.736) and 5-year DFS (HR = 0.932, 95% CI: 0.640–1.357, P = 0.714) than R-CT (Table ). Nevertheless, NACT was correlated with a decrease in OS (HR = 1.454, 95% CI: 1.057–2.000, P = 0.021) and DFS (HR = 1.529, 95% CI: 1.185–1.974, P = 0.001) than ARH (Table ). ASC was associated with a decrease in 5-year OS (HR = 3.182, 95% CI: 1.803–5.617, P < 0.001) and 5-year DFS (HR = 2.210, 95% CI: 1.288–3.792, P = 0.004) than SCC; nonetheless, AC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Moreover, age was not associated with 5-year OS and DFS ( P > 0.05 for all) (Tables and ). In the T2b group, 869 cases were divided into R-CT ( n = 645), ARH ( n = 73), and NACT ( n = 151) subgroups. Kaplan–Meier analysis revealed statistically sharp distinctions in 5-year OS (73.9% vs . 86.3% vs . 68.5%, P = 0.011) and DFS (72.6% vs . 64.0% vs . 47.6%, P < 0.001) among the three subgroups (Fig. E, F). Cox proportional hazards regression analysis revealed NACT correlated with a decrease in 5-year DFS (HR = 1.847, 95% CI: 1.347–2.532, P < 0.001) than R-CT; however, there was no correlation with OS (HR = 1.353, 95% CI: 0.901–2.032, P = 0.146). ARH was not correlated with 5-year OS (HR = 0.490, 95% CI: 0.236–1.017, P = 0.056) and DFS (HR = 1.101, 95% CI 0.702–1.725, P = 0.676). AC was correlated with a decrease in 5-year OS (HR = 1.999, 95% CI: 1.092–3.660, P = 0.025) and DFS (HR = 2.098, 95% CI: 1.298–3.392, P = 0.003) than SCC. However, ASC was not correlated with 5-year OS and DFS ( P > 0.05 for all). Furthermore, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Table ). In the T3 group, 1,081 cases were divided into R-CT ( n = 1,053), ARH ( n = 10), and NACT ( n = 18) subgroups. Kaplan–Meier analysis revealed no statistically sharp distinctions in 5-year OS (64.7% vs . 67.5% vs . 53.1%, P = 0.941) and DFS (61.1% vs . 68.6% vs . 45.5%, P = 0.761) among the three subgroups (Fig. G, H). Cox proportional hazards regression analysis revealed NACT was not correlated with 5-year OS (HR = 0.780, 95% CI: 0.284–2.139, P = 0.629) and DFS (HR = 1.075, 95% CI: 0.499–2.316, P = 0.853) compared with R-CT. ARH was not correlated with 5-year OS (HR = 0.879, 95% CI: 0.281–2.746, P = 0.824) and DFS (HR = 0.747, 95% CI: 0.239–2.328, P = 0.614). ASC was correlated with a decrease in 5-year OS (HR = 2.733, 95% CI: 1.109–6.733, P = 0.029) than SCC; nevertheless, it was not correlated with 5-year DFS (HR = 2.156, 95% CI: 0.946–4.916, P = 0.068), and AC was not correlated with 5-year OS and DFS ( P > 0.05 for all). In addition, age was not correlated with 5-year OS and DFS ( P > 0.05 for all) (Table ). Summary of main results A key change to the FIGO 2018 staging system was LN metastasis inclusion, which indicates its importance in tumor progression and prognosis. Thus, this group is treated differently, which is helpful for clinical research and medical intervention. However, the 2018 FIGO Cervical Cancer guidelines have no stratified treatment recommendations for these patients. Since the release of the 2018 FIGO staging system, a wave of research on new staging treatment strategies and prognosis has emerged. Based on the 1538 project database, this study conducted a real-world study on FIGO 2018 stage IIIC CC using T-staging to investigate the oncological outcomes of ARH, NACT, and R-CT. The results demonstrate that different treatments affect the oncological outcomes of patients with T1, T2a, T2b, and T3 CC in FIGO 2018 stage IIIC. Mortality and recurrence/death risks were higher in NACT than in ARH in the T1 and T2a group, while recurrence/death risk was higher in NACT than in R-CT in the T2b group. Among the T1, T2a, T2b, and T3 groups, no statistically significant differences in death and recurrence/death risks were noted between ARH and R-CT. ARH may be an alternative initial treatment for patients with FIGO 2018 stage IIIC CC. NACT is not recommended for stage T1, T2a and T2b. The number of cases in the three subgroups of the T3 stage varied greatly, therefore further research is required to confirm the results. Results in the context of published literature TNM-staging Valid staging systems are characterized by intragroup homogeneity, that is, same-staged patients essentially exhibit minimal prognostic differences. Previous related studies suggested that although LN metastasis is an important prognostic factor for patients with CC, including all patients with LN metastasis in the same stage leads to high patient heterogeneity . Certain studies found that CC prognosis was related to T-staging ; these results may be related to stage IIIC patient heterogeneity. In a retrospective study by Matsuo K et al. , 733 patients with stage IIIC1 were divided into T1, T2, and T3 groups based on T-staging. T3b had a lower survival rate, revealing a significant OS difference based on T-staging among stage IIIC1 patients; however, this study exclusively included stage IIIC1 patients. Evidently, the prognosis of stage IIIC patients is also affected by local tumor factors, with a significantly different OS rate. The vast heterogeneity among these patients not only affects prognostic prediction but also clinical decision-making. Treating stage IIIC patients, with sole consideration of LN status, without stratifying local tumor factors and extent of spread, may be inappropriate. Therefore, our study stratified stage IIIC cases using T-staging and compared the oncological outcomes of ARH, NACT and R-CT. Treatment strategies CC treatment should be implemented in a planned and sequential stage-based manner, adjusting according to surgical results and post-radiotherapy tumor regression. Due to the advantages of preserving ovarian function, tissue elasticity, and reproductive function, surgical-treatment is increasing . RT is predominant for advanced cervical cancer. Chemotherapy is used as an adjuvant therapy for RT sensitization. Radical surgical treatment can remove metastatic pelvic LN to minish burthen of tumor and determine LN status, thus guiding postoperative supplementary treatment selection. Avoiding excessive treatment is important. Currently, radical surgery and pelvic lymphadenectomy are preferred for early CC. A study of FIGO 2009 IB1 and IIA1 CC by Wu et al. suggested that no significant oncological-outcome difference existed between ARH and R-CT. Landoni’s study of 19/343 IB1 and IIA1 CC cases also concurred that ARH and R-CT had similar effects. Our study agreed with these findings. However, a study on FIGO 2009 IB1-IIA2 CC concluded that ARH oncological outcomes were superior to those of R-CT; nevertheless, it exclusively included SCC cases. A seven-study meta-analysis by Yan et al. revealed that ARH had obvious advantages over chemoradiotherapy for IB2-IIA CC. Jang et al. conducted a study on FIGO 2009 IB1-IIA CC and found that the oncological prognosis of ARH was significantly superior to that of concurrent chemoradiotherapy (CCRT). Bansal et al. analyzed 4,885 cases of FIGO 2009 IB1-IIA CC and found that when the tumor diameter is < 6 cm, ARH potentially benefits patient survival more than R-CT; when it is > 6 cm, the two are equivalent. For FIGO 2018 stage IIIC, the oncological prognosis of R-CT does not prevail over that of ARH, considering that R-CT has a series of complications, such as weakened ovarian function , vaginal constriction, and dry intercourse , which seriously affect patients’ quality of life. Hence, ARH is recommended as an initial treatment. NACT is the predominant pre-operative adjuvant therapy for CC, while FIGO guidelines recommend it for clinical trials or areas where radiotherapy equipment is lacking. NCCN guidelines only recommend it for small-cell neuroendocrine carcinoma of the cervix. At present, whether NACT can improve the prognosis of patients with CC remains controversial . In the T1 and T2a group of ours, NACT was associated with a decrease in 5-year OS and 5-year DFS than ARH. Meantime, relevant studies found worse prognosis or no difference between post-NACT surgery and radical radiotherapy for patients with CC. A single-center, phase III, randomized controlled trial by Gupta et al. found that for FIGO 2009 stage IB2-IIB cervical SCC, the group undergoing post-NACT surgery had a statistically difference in 5-year DFS compared with the CCRT group; however no sharp distinction in 5-year OS. In the T2b group of ours, NACT was correlated with a decrease in 5-year DFS than R-CT; nevertheless, it was not correlated with 5-year OS. Duenas-Gonzalez et al. did not detect any difference in response and viability between post-chemotherapy surgery and standard cisplatin-based chemoradiotherapy. Similar results were obtained in the T1, T2a and T3 groups in ours, and no statistical distinction in 5-year OS and DFS was noted between NACT and R-CT. Evidently, in FIGO 2018 stage IIIC, patients in the T1 and T2a groups potentially benefited more from ARH than NACT; meanwhile, in the T2b group, NACT was not efficaciously advantageous. Therefore, NACT should be used with caution in stage T1,T2a and T2b. Previous studies predominantly used FIGO 2009 staging, included cases with positive and negative LN, and did not exclude patients treated with laparoscopic surgery. Differences in adjuvant-treatment regimens and treatment courses also existed. Several studies advocate that LN metastasis is an important factor affecting CC prognosis . Results of LACC study in 2018 suggested that laparoscopic surgery had adverse oncology outcomes in patients with CC. Therefore, ours only included LN-positive cases and was limited to laparotomy-treated patients, thus potentially justifying the inconsistencies between this and previous studies. Strengths and weaknesses Ours is a large-scale research based on FIGO 2018 stage IIIC CC cases. Notably, it innovatively stratified patients basing on local tumor conditions to compare the prognosis of different treatment methods. Notwithstanding, this study also had some shortcomings. First, this was a real-world, retrospective analysis, resulting in unbalanced data between groups. Second, the numbers of cases among the T3 subgroup were significantly different, thus potentially affecting the results’ reliability. Third, there is a limitation on a distinction between staging modalities. Patients who underwent radical surgery in the ARH and NACT groups defined LN status depended on pathological examinations after surgery; nevertheless, those who underwent radical chemoradiotherapy in the R-CT group defined lymph node status depended on CT or/and MRI before surgery. There is a difference in false positive rates between the two methods. Fourth, no further stratification based on N stage was conducted in this study due to insufficient information on the status of para-aortic lymph nodes. Implications for practice and future research In conclusion, for patients with FIGO 2018 stage IIIC CC, different treatment strategies impact oncological outcomes. When selecting a treatment strategy for these patients, T-staging is required. Patients with stages T1, T2a, T2b, and T3 can select ARH for initial treatment. Because of the huge differences in the number of cases among the T3 subgroups, our results require confirmation through further research. NACT is not recommended for patients with stage T1, T2a and T2b. Evidently, the recommended treatment methods for patients with stage IIIC CC in the guidelines are debatable, and more prospective studies are warranted. A key change to the FIGO 2018 staging system was LN metastasis inclusion, which indicates its importance in tumor progression and prognosis. Thus, this group is treated differently, which is helpful for clinical research and medical intervention. However, the 2018 FIGO Cervical Cancer guidelines have no stratified treatment recommendations for these patients. Since the release of the 2018 FIGO staging system, a wave of research on new staging treatment strategies and prognosis has emerged. Based on the 1538 project database, this study conducted a real-world study on FIGO 2018 stage IIIC CC using T-staging to investigate the oncological outcomes of ARH, NACT, and R-CT. The results demonstrate that different treatments affect the oncological outcomes of patients with T1, T2a, T2b, and T3 CC in FIGO 2018 stage IIIC. Mortality and recurrence/death risks were higher in NACT than in ARH in the T1 and T2a group, while recurrence/death risk was higher in NACT than in R-CT in the T2b group. Among the T1, T2a, T2b, and T3 groups, no statistically significant differences in death and recurrence/death risks were noted between ARH and R-CT. ARH may be an alternative initial treatment for patients with FIGO 2018 stage IIIC CC. NACT is not recommended for stage T1, T2a and T2b. The number of cases in the three subgroups of the T3 stage varied greatly, therefore further research is required to confirm the results. TNM-staging Valid staging systems are characterized by intragroup homogeneity, that is, same-staged patients essentially exhibit minimal prognostic differences. Previous related studies suggested that although LN metastasis is an important prognostic factor for patients with CC, including all patients with LN metastasis in the same stage leads to high patient heterogeneity . Certain studies found that CC prognosis was related to T-staging ; these results may be related to stage IIIC patient heterogeneity. In a retrospective study by Matsuo K et al. , 733 patients with stage IIIC1 were divided into T1, T2, and T3 groups based on T-staging. T3b had a lower survival rate, revealing a significant OS difference based on T-staging among stage IIIC1 patients; however, this study exclusively included stage IIIC1 patients. Evidently, the prognosis of stage IIIC patients is also affected by local tumor factors, with a significantly different OS rate. The vast heterogeneity among these patients not only affects prognostic prediction but also clinical decision-making. Treating stage IIIC patients, with sole consideration of LN status, without stratifying local tumor factors and extent of spread, may be inappropriate. Therefore, our study stratified stage IIIC cases using T-staging and compared the oncological outcomes of ARH, NACT and R-CT. Treatment strategies CC treatment should be implemented in a planned and sequential stage-based manner, adjusting according to surgical results and post-radiotherapy tumor regression. Due to the advantages of preserving ovarian function, tissue elasticity, and reproductive function, surgical-treatment is increasing . RT is predominant for advanced cervical cancer. Chemotherapy is used as an adjuvant therapy for RT sensitization. Radical surgical treatment can remove metastatic pelvic LN to minish burthen of tumor and determine LN status, thus guiding postoperative supplementary treatment selection. Avoiding excessive treatment is important. Currently, radical surgery and pelvic lymphadenectomy are preferred for early CC. A study of FIGO 2009 IB1 and IIA1 CC by Wu et al. suggested that no significant oncological-outcome difference existed between ARH and R-CT. Landoni’s study of 19/343 IB1 and IIA1 CC cases also concurred that ARH and R-CT had similar effects. Our study agreed with these findings. However, a study on FIGO 2009 IB1-IIA2 CC concluded that ARH oncological outcomes were superior to those of R-CT; nevertheless, it exclusively included SCC cases. A seven-study meta-analysis by Yan et al. revealed that ARH had obvious advantages over chemoradiotherapy for IB2-IIA CC. Jang et al. conducted a study on FIGO 2009 IB1-IIA CC and found that the oncological prognosis of ARH was significantly superior to that of concurrent chemoradiotherapy (CCRT). Bansal et al. analyzed 4,885 cases of FIGO 2009 IB1-IIA CC and found that when the tumor diameter is < 6 cm, ARH potentially benefits patient survival more than R-CT; when it is > 6 cm, the two are equivalent. For FIGO 2018 stage IIIC, the oncological prognosis of R-CT does not prevail over that of ARH, considering that R-CT has a series of complications, such as weakened ovarian function , vaginal constriction, and dry intercourse , which seriously affect patients’ quality of life. Hence, ARH is recommended as an initial treatment. NACT is the predominant pre-operative adjuvant therapy for CC, while FIGO guidelines recommend it for clinical trials or areas where radiotherapy equipment is lacking. NCCN guidelines only recommend it for small-cell neuroendocrine carcinoma of the cervix. At present, whether NACT can improve the prognosis of patients with CC remains controversial . In the T1 and T2a group of ours, NACT was associated with a decrease in 5-year OS and 5-year DFS than ARH. Meantime, relevant studies found worse prognosis or no difference between post-NACT surgery and radical radiotherapy for patients with CC. A single-center, phase III, randomized controlled trial by Gupta et al. found that for FIGO 2009 stage IB2-IIB cervical SCC, the group undergoing post-NACT surgery had a statistically difference in 5-year DFS compared with the CCRT group; however no sharp distinction in 5-year OS. In the T2b group of ours, NACT was correlated with a decrease in 5-year DFS than R-CT; nevertheless, it was not correlated with 5-year OS. Duenas-Gonzalez et al. did not detect any difference in response and viability between post-chemotherapy surgery and standard cisplatin-based chemoradiotherapy. Similar results were obtained in the T1, T2a and T3 groups in ours, and no statistical distinction in 5-year OS and DFS was noted between NACT and R-CT. Evidently, in FIGO 2018 stage IIIC, patients in the T1 and T2a groups potentially benefited more from ARH than NACT; meanwhile, in the T2b group, NACT was not efficaciously advantageous. Therefore, NACT should be used with caution in stage T1,T2a and T2b. Previous studies predominantly used FIGO 2009 staging, included cases with positive and negative LN, and did not exclude patients treated with laparoscopic surgery. Differences in adjuvant-treatment regimens and treatment courses also existed. Several studies advocate that LN metastasis is an important factor affecting CC prognosis . Results of LACC study in 2018 suggested that laparoscopic surgery had adverse oncology outcomes in patients with CC. Therefore, ours only included LN-positive cases and was limited to laparotomy-treated patients, thus potentially justifying the inconsistencies between this and previous studies. Valid staging systems are characterized by intragroup homogeneity, that is, same-staged patients essentially exhibit minimal prognostic differences. Previous related studies suggested that although LN metastasis is an important prognostic factor for patients with CC, including all patients with LN metastasis in the same stage leads to high patient heterogeneity . Certain studies found that CC prognosis was related to T-staging ; these results may be related to stage IIIC patient heterogeneity. In a retrospective study by Matsuo K et al. , 733 patients with stage IIIC1 were divided into T1, T2, and T3 groups based on T-staging. T3b had a lower survival rate, revealing a significant OS difference based on T-staging among stage IIIC1 patients; however, this study exclusively included stage IIIC1 patients. Evidently, the prognosis of stage IIIC patients is also affected by local tumor factors, with a significantly different OS rate. The vast heterogeneity among these patients not only affects prognostic prediction but also clinical decision-making. Treating stage IIIC patients, with sole consideration of LN status, without stratifying local tumor factors and extent of spread, may be inappropriate. Therefore, our study stratified stage IIIC cases using T-staging and compared the oncological outcomes of ARH, NACT and R-CT. CC treatment should be implemented in a planned and sequential stage-based manner, adjusting according to surgical results and post-radiotherapy tumor regression. Due to the advantages of preserving ovarian function, tissue elasticity, and reproductive function, surgical-treatment is increasing . RT is predominant for advanced cervical cancer. Chemotherapy is used as an adjuvant therapy for RT sensitization. Radical surgical treatment can remove metastatic pelvic LN to minish burthen of tumor and determine LN status, thus guiding postoperative supplementary treatment selection. Avoiding excessive treatment is important. Currently, radical surgery and pelvic lymphadenectomy are preferred for early CC. A study of FIGO 2009 IB1 and IIA1 CC by Wu et al. suggested that no significant oncological-outcome difference existed between ARH and R-CT. Landoni’s study of 19/343 IB1 and IIA1 CC cases also concurred that ARH and R-CT had similar effects. Our study agreed with these findings. However, a study on FIGO 2009 IB1-IIA2 CC concluded that ARH oncological outcomes were superior to those of R-CT; nevertheless, it exclusively included SCC cases. A seven-study meta-analysis by Yan et al. revealed that ARH had obvious advantages over chemoradiotherapy for IB2-IIA CC. Jang et al. conducted a study on FIGO 2009 IB1-IIA CC and found that the oncological prognosis of ARH was significantly superior to that of concurrent chemoradiotherapy (CCRT). Bansal et al. analyzed 4,885 cases of FIGO 2009 IB1-IIA CC and found that when the tumor diameter is < 6 cm, ARH potentially benefits patient survival more than R-CT; when it is > 6 cm, the two are equivalent. For FIGO 2018 stage IIIC, the oncological prognosis of R-CT does not prevail over that of ARH, considering that R-CT has a series of complications, such as weakened ovarian function , vaginal constriction, and dry intercourse , which seriously affect patients’ quality of life. Hence, ARH is recommended as an initial treatment. NACT is the predominant pre-operative adjuvant therapy for CC, while FIGO guidelines recommend it for clinical trials or areas where radiotherapy equipment is lacking. NCCN guidelines only recommend it for small-cell neuroendocrine carcinoma of the cervix. At present, whether NACT can improve the prognosis of patients with CC remains controversial . In the T1 and T2a group of ours, NACT was associated with a decrease in 5-year OS and 5-year DFS than ARH. Meantime, relevant studies found worse prognosis or no difference between post-NACT surgery and radical radiotherapy for patients with CC. A single-center, phase III, randomized controlled trial by Gupta et al. found that for FIGO 2009 stage IB2-IIB cervical SCC, the group undergoing post-NACT surgery had a statistically difference in 5-year DFS compared with the CCRT group; however no sharp distinction in 5-year OS. In the T2b group of ours, NACT was correlated with a decrease in 5-year DFS than R-CT; nevertheless, it was not correlated with 5-year OS. Duenas-Gonzalez et al. did not detect any difference in response and viability between post-chemotherapy surgery and standard cisplatin-based chemoradiotherapy. Similar results were obtained in the T1, T2a and T3 groups in ours, and no statistical distinction in 5-year OS and DFS was noted between NACT and R-CT. Evidently, in FIGO 2018 stage IIIC, patients in the T1 and T2a groups potentially benefited more from ARH than NACT; meanwhile, in the T2b group, NACT was not efficaciously advantageous. Therefore, NACT should be used with caution in stage T1,T2a and T2b. Previous studies predominantly used FIGO 2009 staging, included cases with positive and negative LN, and did not exclude patients treated with laparoscopic surgery. Differences in adjuvant-treatment regimens and treatment courses also existed. Several studies advocate that LN metastasis is an important factor affecting CC prognosis . Results of LACC study in 2018 suggested that laparoscopic surgery had adverse oncology outcomes in patients with CC. Therefore, ours only included LN-positive cases and was limited to laparotomy-treated patients, thus potentially justifying the inconsistencies between this and previous studies. Ours is a large-scale research based on FIGO 2018 stage IIIC CC cases. Notably, it innovatively stratified patients basing on local tumor conditions to compare the prognosis of different treatment methods. Notwithstanding, this study also had some shortcomings. First, this was a real-world, retrospective analysis, resulting in unbalanced data between groups. Second, the numbers of cases among the T3 subgroup were significantly different, thus potentially affecting the results’ reliability. Third, there is a limitation on a distinction between staging modalities. Patients who underwent radical surgery in the ARH and NACT groups defined LN status depended on pathological examinations after surgery; nevertheless, those who underwent radical chemoradiotherapy in the R-CT group defined lymph node status depended on CT or/and MRI before surgery. There is a difference in false positive rates between the two methods. Fourth, no further stratification based on N stage was conducted in this study due to insufficient information on the status of para-aortic lymph nodes. In conclusion, for patients with FIGO 2018 stage IIIC CC, different treatment strategies impact oncological outcomes. When selecting a treatment strategy for these patients, T-staging is required. Patients with stages T1, T2a, T2b, and T3 can select ARH for initial treatment. Because of the huge differences in the number of cases among the T3 subgroups, our results require confirmation through further research. NACT is not recommended for patients with stage T1, T2a and T2b. Evidently, the recommended treatment methods for patients with stage IIIC CC in the guidelines are debatable, and more prospective studies are warranted. R-CT oncological outcomes were not entirely superior to those of NACT or ARH under different local tumor factors with stage IIIC. NACT is not suitable for stage T1, T2a and T2b. Nevertheless ARH is potentially applicable to stage T1, T2a, T2b and T3. Stage T3 results require confirmation through further research due to disparity in case numbers in each subgroup. Additional file 1.
Calculating the optimal number of nurses based on nursing intensity by patient classification groups in general units in South Korea: A cross‐sectional study	885b76e3-83ed-46dd-a480-900e7fb0d15f	10170926	Internal Medicine[mh]	BACKGROUND Suboptimal nurse staffing in hospitals can affect the quality of patient care and nurses' health and well‐being (Nantsupawat et al., ). Therefore, if nursing intensity based on nursing needs can be calculated by using a more sophisticated patient classification tool and the calculation criteria for nurse staffing can be elaborately designed, realistic standards for nursing unit operations can be established. Currently, the optimal number of nurses per nursing unit can be calculated using a method based on nursing time by patient classification groups (Jang et al., ). This approach involves quantifying nursing time taken according to patients' nursing needs; moreover, it is limited in considering the differences in qualitative characteristics, such as nurses' qualifications and competencies. As it cannot realistically reflect nursing time taken according to the severity of inpatients' diseases and does not record time‐consuming nursing activities, this approach does not reflect the reality of clinical nursing practice (Ko & Park, ). Most previous studies measuring nursing time have used direct measurement methods such as the work sampling method and time‐and‐motion studies (Kim et al., ). Since nurse staffing is calculated based on the required nursing time per patient, the required nursing time should be measured and predicted using tools (Kim et al., ). Traditional time studies use a direct measurement method that requires much time and effort and does not comprehensively reveal the attributes of nursing activities and the essence of nursing care. Since nursing activities are complex, with several activities occurring simultaneously, there are practical limitations to their estimation by breaking them down and measuring them individually (Fagerström et al., ). Therefore, there is a need to develop a new method that can efficiently measure nursing time based on a comprehensive approach. Santos et al. suggested that working time should be described in three dimensions: the duration, distribution and intensity. Working time refers to the total time spent on work. The distribution refers to how it is spread across days, weeks, months and years; it includes intermittent work, compensatory time, home office or remote work and all the free time used to obtain qualifications to work more competently. To determine nurse staffing levels, managers need to understand the underlying determinants—patient (patients' nursing needs based on their acuity and dependency levels), ward (patient throughput) and nursing staff (number and skill level) factors (National Institute for Health and Care Excellence, ). A challenge faced by managers responsible for staffing is trying to understand the influence of the multiple factors that make up each individual care environment (Driscoll et al., ). Meanwhile, Fagerström et al. in Finland defined nursing intensity per patient as the sum of the means of the weighting coefficients per patient classification group divided by the number of nurses, presenting a method for concisely quantifying nursing intensity. They believed that this method would enable comparisons of nursing intensity between various nursing units and the accumulated data can be used as a basis for resource allocation in hospitals. 1.1 Objectives This study aimed to calculate the total daily nursing workload per nursing unit and the optimal number of nurses based on the nursing intensity and direct nursing time per inpatient. Our findings can be used as basic data for calculating the optimal number of nurses required in nursing units, establishing effective nurse staffing strategies, and enabling better management of human resources. The specific objectives were as follows: To calculate direct and non‐direct nursing time by investigating the total working time, non‐direct nursing time and personal time perceived by nurses in each nursing unit (internal medicine, surgical and comprehensive nursing care units); To calculate and compare nursing intensity in nursing units based on patient classification scores; To calculate nursing intensity and direct nursing time taken per patient‐by‐patient classification groups in the nursing units; and To calculate the total daily nursing workload and the optimal number of nurses per nursing unit based on the nursing intensity and nursing time by patient classification groups. Objectives This study aimed to calculate the total daily nursing workload per nursing unit and the optimal number of nurses based on the nursing intensity and direct nursing time per inpatient. Our findings can be used as basic data for calculating the optimal number of nurses required in nursing units, establishing effective nurse staffing strategies, and enabling better management of human resources. The specific objectives were as follows: To calculate direct and non‐direct nursing time by investigating the total working time, non‐direct nursing time and personal time perceived by nurses in each nursing unit (internal medicine, surgical and comprehensive nursing care units); To calculate and compare nursing intensity in nursing units based on patient classification scores; To calculate nursing intensity and direct nursing time taken per patient‐by‐patient classification groups in the nursing units; and To calculate the total daily nursing workload and the optimal number of nurses per nursing unit based on the nursing intensity and nursing time by patient classification groups. METHODS 2.1 Research design This study adopted a descriptive research design. The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist for cross‐sectional studies was followed to strengthen this study's methodological rigour. 2.2 Data collection Data were collected for this study using the following four steps. The data collection procedure is presented in Figure . 2.3 Participants To classify patients according to their nursing needs and calculate nursing intensity, three nursing units (internal medicine, surgical and comprehensive) at a single tertiary general hospital in J Province, South Korea, were selected as the study units. At the time of the investigation, 20, 28 and 22 nurses belonged to the internal medicine, surgical and comprehensive wards, respectively (Table ). According to nursing needs, patient classification and nursing time in each unit were investigated over 10 days between September 1 and 30 2018, excluding weekends. Night shift nurses in charge of the three nursing units performed patient classification of all patients admitted to their units using the Korean Patient Classification System on Nursing Needs for intensive care units (KPCSNI), a tool developed by Ko and Park (Ko et al., ). The total number of patients who were subjected to classification comprised 473, 278 and 143 patient‐days in the internal medicine, surgical and comprehensive nursing care units, respectively. The nursing time of all nurses in the nursing units who performed patient classification were investigated on that same day in the relevant nursing unit. Nurses who worked day, evening or night shifts were instructed to complete a self‐report questionnaire immediately after the end of their working hours, thereby minimizing recall errors. The number of nurses who participated in the survey of nursing time comprised 87, 125 and 77 person‐days in the internal medicine, surgical and comprehensive nursing care units, respectively, totalling 289 person‐days. 2.4 Definitions of terms 2.4.1 Patient classification The patient classification system is a method of classifying patients according to the amount and complexity of nursing care provided to them over a certain period (Park, ). Here, it refers to classifying patients admitted to the units into groups 1–4 using the KPCSNI. As the patient classification group number increases from 1 to 4, the total score for each item increases, indicating that patients' nursing needs are higher. 2.4.2 Nursing intensity Nursing intensity refers to direct and non‐direct nursing activities related to patients; it includes patients' dependency, severity of the disease, complexity of nursing care and time required for nursing as factors directly affecting such nursing activities (Hoi et al., ). To calculate nursing intensity, patient classification scores were calculated using a tool that was modified and supplemented based on the one developed by Ko and Park (Ko et al., ). This tool comprises 50 direct nursing activities covering eight domains (symptom management infection control, nutrition and medication, personal hygiene and secretion, activity, sleep and rest, guidance in nursing/emotional support, nursing activity planning and coordination) and 11 indirect activities. Based on the calculated patient classification scores, the weighting coefficient per nursing unit, that is nursing intensity, was calculated following the method used by Fagerström et al. . 2.4.3 Personal time Personal time excludes direct and non‐direct nursing activity times during working time and includes meal and rest times. 2.4.4 Non‐direct nursing time Non‐direct nursing activities include managing the necessary items and environment for nursing and maintaining the operation of nursing units except for direct nursing care for patients (Park & Song, ). Non‐direct nursing time refers to the sum of the nursing time required for handover, making rounds, work delay, recording, patient‐related calls and deliveries, administrative affairs, cognitive workload, education/supervision, research, etc., as measured using the patient classification tool developed by Ko and Park (Ko et al., ). 2.4.5 Direct nursing time Direct nursing time refers to nursing time (Park & Song, ) for providing direct nursing care to patients and preparing and organizing nursing care. Here, it refers to total working time after subtracting personal and indirect nursing times. 2.5 Measurement 2.5.1 Patient classification and calculation of nursing intensity for nursing units based on nursing needs Patient classification based on nursing needs was conducted using the KPCSNI. This tool is a factor‐type classification tool and includes scores for the clinical features of patients in addition to scores for nursing needs when calculating patient classification scores. It comprises 8 domains and 18 sub‐domains covering 50 nursing activities. After the tool was reviewed by the researchers, its content validity was tested in consultation with six nursing professors. The average daily value of the calculated patient classification scores for each date was calculated. As a result, a patient classification score of 1–30 points was classified as Group 1, a score of 31–60 points was classified as Group 2, a score of 61–90 points was classified as Group 3 and a score of 91 points or more was classified as Group 4 based on the results of the study by Ko and Park (Ko et al., ). After setting the patient classification score in Group 1 as the reference value of “1,” the patient classification scores in Groups 2–4 were divided by the patient classification score in Group 1 to calculate nursing intensity weighting coefficients for the groups. Nursing intensity scores for nursing units were calculated by multiplying the weighting coefficient for each group by the number of patients in each group and then aggregating the values. Fagerström et al. propounded the “Professional Assessment of Optimal Nursing Care Intensity Level,” a new method that goes beyond the traditional time study methodology and could establish optimal nursing intensity levels for individual units. It calculates nursing intensity based on patient classification results and assesses nursing intensity for nursing units by reflecting statistical estimations and expert opinions. 2.5.2 Calculating nursing time Nursing time was measured using a questionnaire developed by this study's researchers with reference to non‐direct nursing activities in a tool developed by Ko and Park (Ko et al., ). This questionnaire comprises 28 items, including total working, break and non‐direct nursing times of the day. Total working time was calculated based on the time at which nurses logged into work and left for the day, while the break time was calculated by summing up the meal and rest times. Non‐direct nursing time was calculated by summing the time for each subdomain of the three domains (nursing activity planning and coordination, non‐direct activity and break time). Direct nursing time was calculated by subtracting the non‐direct nursing time including leisure time from the total working time (Formula in Appendix ). Six nursing professors and one expert reviewed the validity of the direct nursing calculation method. 2.5.3 Calculating direct nursing time per inpatient by patient classification groups To calculate direct nursing time by patient classification groups, direct nursing time per nursing intensity point was calculated (Formula in Appendix ). This value was then multiplied by the weighting coefficient for each patient classification group to calculate direct nursing time per inpatient by patient classification groups (Formula in Appendix ). 2.5.4 Calculating the optimal number of nurses The optimal number of nurses in the internal medicine, surgical, and comprehensive nursing care units was estimated by applying the calculated nursing time results to Formulas (4–6) in Appendix . After non‐direct nursing time was estimated using the ratio (20%) of non‐direct nursing time to the total nursing working time—calculated with the nursing time analysis results—the total nursing time was measured (Formula in Appendix ). The optimal number of nurses was calculated by adding 40% to the value obtained by dividing the total nursing work time by the mean daily work hours (Formula in Appendix ). The total number of annual holidays in the current clinical reality is estimated to be about 134 days, considering weekly holidays: 52 weeks × 2 (Saturday and Sunday) based on an average of 20 working days per month, plus 15 legal holidays (excluding Sundays), 15 basic annual holidays and additional annual holidays according to the nurses' professional positions. Although 1.4 can be assigned as an additive value owing to the number of holidays by rounding off 1.37 [(134 + 365)/365], a constant of 1.6 was used in this study following previous studies (Cho et al., ; Lee et al., ). 2.6 Data analysis The collected data were analysed using the Microsoft Excel program. The participants' general characteristics, direct nursing time and nursing intensity for each date by patient classification groups were analysed using descriptive statistics such as frequency, percentage and average. Direct nursing time among the nursing unit nurses, direct nursing time per patient classification point or nursing intensity point and direct nursing time per patient were calculated using Microsoft Office Excel 2017. 2.7 Ethical considerations This study was conducted after explaining its purpose to the head of the nursing department at the study hospital. The researchers visited the nursing units to explain this study's purpose. Nurses who agreed to participate, in writing were selected. This study was approved by the Institutional Review Board at a university to which the author belongs (Approval No: 1040271‐201808‐HR‐026). A study on the calculation of the optimal number of nurses based on nursing intensity in the intensive care unit using the same research model has been published in the Korean Journal of Hospital Management (Ko & Park, ). Research design This study adopted a descriptive research design. The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist for cross‐sectional studies was followed to strengthen this study's methodological rigour. Data collection Data were collected for this study using the following four steps. The data collection procedure is presented in Figure . Participants To classify patients according to their nursing needs and calculate nursing intensity, three nursing units (internal medicine, surgical and comprehensive) at a single tertiary general hospital in J Province, South Korea, were selected as the study units. At the time of the investigation, 20, 28 and 22 nurses belonged to the internal medicine, surgical and comprehensive wards, respectively (Table ). According to nursing needs, patient classification and nursing time in each unit were investigated over 10 days between September 1 and 30 2018, excluding weekends. Night shift nurses in charge of the three nursing units performed patient classification of all patients admitted to their units using the Korean Patient Classification System on Nursing Needs for intensive care units (KPCSNI), a tool developed by Ko and Park (Ko et al., ). The total number of patients who were subjected to classification comprised 473, 278 and 143 patient‐days in the internal medicine, surgical and comprehensive nursing care units, respectively. The nursing time of all nurses in the nursing units who performed patient classification were investigated on that same day in the relevant nursing unit. Nurses who worked day, evening or night shifts were instructed to complete a self‐report questionnaire immediately after the end of their working hours, thereby minimizing recall errors. The number of nurses who participated in the survey of nursing time comprised 87, 125 and 77 person‐days in the internal medicine, surgical and comprehensive nursing care units, respectively, totalling 289 person‐days. Definitions of terms 2.4.1 Patient classification The patient classification system is a method of classifying patients according to the amount and complexity of nursing care provided to them over a certain period (Park, ). Here, it refers to classifying patients admitted to the units into groups 1–4 using the KPCSNI. As the patient classification group number increases from 1 to 4, the total score for each item increases, indicating that patients' nursing needs are higher. 2.4.2 Nursing intensity Nursing intensity refers to direct and non‐direct nursing activities related to patients; it includes patients' dependency, severity of the disease, complexity of nursing care and time required for nursing as factors directly affecting such nursing activities (Hoi et al., ). To calculate nursing intensity, patient classification scores were calculated using a tool that was modified and supplemented based on the one developed by Ko and Park (Ko et al., ). This tool comprises 50 direct nursing activities covering eight domains (symptom management infection control, nutrition and medication, personal hygiene and secretion, activity, sleep and rest, guidance in nursing/emotional support, nursing activity planning and coordination) and 11 indirect activities. Based on the calculated patient classification scores, the weighting coefficient per nursing unit, that is nursing intensity, was calculated following the method used by Fagerström et al. . 2.4.3 Personal time Personal time excludes direct and non‐direct nursing activity times during working time and includes meal and rest times. 2.4.4 Non‐direct nursing time Non‐direct nursing activities include managing the necessary items and environment for nursing and maintaining the operation of nursing units except for direct nursing care for patients (Park & Song, ). Non‐direct nursing time refers to the sum of the nursing time required for handover, making rounds, work delay, recording, patient‐related calls and deliveries, administrative affairs, cognitive workload, education/supervision, research, etc., as measured using the patient classification tool developed by Ko and Park (Ko et al., ). 2.4.5 Direct nursing time Direct nursing time refers to nursing time (Park & Song, ) for providing direct nursing care to patients and preparing and organizing nursing care. Here, it refers to total working time after subtracting personal and indirect nursing times. Patient classification The patient classification system is a method of classifying patients according to the amount and complexity of nursing care provided to them over a certain period (Park, ). Here, it refers to classifying patients admitted to the units into groups 1–4 using the KPCSNI. As the patient classification group number increases from 1 to 4, the total score for each item increases, indicating that patients' nursing needs are higher. Nursing intensity Nursing intensity refers to direct and non‐direct nursing activities related to patients; it includes patients' dependency, severity of the disease, complexity of nursing care and time required for nursing as factors directly affecting such nursing activities (Hoi et al., ). To calculate nursing intensity, patient classification scores were calculated using a tool that was modified and supplemented based on the one developed by Ko and Park (Ko et al., ). This tool comprises 50 direct nursing activities covering eight domains (symptom management infection control, nutrition and medication, personal hygiene and secretion, activity, sleep and rest, guidance in nursing/emotional support, nursing activity planning and coordination) and 11 indirect activities. Based on the calculated patient classification scores, the weighting coefficient per nursing unit, that is nursing intensity, was calculated following the method used by Fagerström et al. . Personal time Personal time excludes direct and non‐direct nursing activity times during working time and includes meal and rest times. Non‐direct nursing time Non‐direct nursing activities include managing the necessary items and environment for nursing and maintaining the operation of nursing units except for direct nursing care for patients (Park & Song, ). Non‐direct nursing time refers to the sum of the nursing time required for handover, making rounds, work delay, recording, patient‐related calls and deliveries, administrative affairs, cognitive workload, education/supervision, research, etc., as measured using the patient classification tool developed by Ko and Park (Ko et al., ). Direct nursing time Direct nursing time refers to nursing time (Park & Song, ) for providing direct nursing care to patients and preparing and organizing nursing care. Here, it refers to total working time after subtracting personal and indirect nursing times. Measurement 2.5.1 Patient classification and calculation of nursing intensity for nursing units based on nursing needs Patient classification based on nursing needs was conducted using the KPCSNI. This tool is a factor‐type classification tool and includes scores for the clinical features of patients in addition to scores for nursing needs when calculating patient classification scores. It comprises 8 domains and 18 sub‐domains covering 50 nursing activities. After the tool was reviewed by the researchers, its content validity was tested in consultation with six nursing professors. The average daily value of the calculated patient classification scores for each date was calculated. As a result, a patient classification score of 1–30 points was classified as Group 1, a score of 31–60 points was classified as Group 2, a score of 61–90 points was classified as Group 3 and a score of 91 points or more was classified as Group 4 based on the results of the study by Ko and Park (Ko et al., ). After setting the patient classification score in Group 1 as the reference value of “1,” the patient classification scores in Groups 2–4 were divided by the patient classification score in Group 1 to calculate nursing intensity weighting coefficients for the groups. Nursing intensity scores for nursing units were calculated by multiplying the weighting coefficient for each group by the number of patients in each group and then aggregating the values. Fagerström et al. propounded the “Professional Assessment of Optimal Nursing Care Intensity Level,” a new method that goes beyond the traditional time study methodology and could establish optimal nursing intensity levels for individual units. It calculates nursing intensity based on patient classification results and assesses nursing intensity for nursing units by reflecting statistical estimations and expert opinions. 2.5.2 Calculating nursing time Nursing time was measured using a questionnaire developed by this study's researchers with reference to non‐direct nursing activities in a tool developed by Ko and Park (Ko et al., ). This questionnaire comprises 28 items, including total working, break and non‐direct nursing times of the day. Total working time was calculated based on the time at which nurses logged into work and left for the day, while the break time was calculated by summing up the meal and rest times. Non‐direct nursing time was calculated by summing the time for each subdomain of the three domains (nursing activity planning and coordination, non‐direct activity and break time). Direct nursing time was calculated by subtracting the non‐direct nursing time including leisure time from the total working time (Formula in Appendix ). Six nursing professors and one expert reviewed the validity of the direct nursing calculation method. 2.5.3 Calculating direct nursing time per inpatient by patient classification groups To calculate direct nursing time by patient classification groups, direct nursing time per nursing intensity point was calculated (Formula in Appendix ). This value was then multiplied by the weighting coefficient for each patient classification group to calculate direct nursing time per inpatient by patient classification groups (Formula in Appendix ). 2.5.4 Calculating the optimal number of nurses The optimal number of nurses in the internal medicine, surgical, and comprehensive nursing care units was estimated by applying the calculated nursing time results to Formulas (4–6) in Appendix . After non‐direct nursing time was estimated using the ratio (20%) of non‐direct nursing time to the total nursing working time—calculated with the nursing time analysis results—the total nursing time was measured (Formula in Appendix ). The optimal number of nurses was calculated by adding 40% to the value obtained by dividing the total nursing work time by the mean daily work hours (Formula in Appendix ). The total number of annual holidays in the current clinical reality is estimated to be about 134 days, considering weekly holidays: 52 weeks × 2 (Saturday and Sunday) based on an average of 20 working days per month, plus 15 legal holidays (excluding Sundays), 15 basic annual holidays and additional annual holidays according to the nurses' professional positions. Although 1.4 can be assigned as an additive value owing to the number of holidays by rounding off 1.37 [(134 + 365)/365], a constant of 1.6 was used in this study following previous studies (Cho et al., ; Lee et al., ). Patient classification and calculation of nursing intensity for nursing units based on nursing needs Patient classification based on nursing needs was conducted using the KPCSNI. This tool is a factor‐type classification tool and includes scores for the clinical features of patients in addition to scores for nursing needs when calculating patient classification scores. It comprises 8 domains and 18 sub‐domains covering 50 nursing activities. After the tool was reviewed by the researchers, its content validity was tested in consultation with six nursing professors. The average daily value of the calculated patient classification scores for each date was calculated. As a result, a patient classification score of 1–30 points was classified as Group 1, a score of 31–60 points was classified as Group 2, a score of 61–90 points was classified as Group 3 and a score of 91 points or more was classified as Group 4 based on the results of the study by Ko and Park (Ko et al., ). After setting the patient classification score in Group 1 as the reference value of “1,” the patient classification scores in Groups 2–4 were divided by the patient classification score in Group 1 to calculate nursing intensity weighting coefficients for the groups. Nursing intensity scores for nursing units were calculated by multiplying the weighting coefficient for each group by the number of patients in each group and then aggregating the values. Fagerström et al. propounded the “Professional Assessment of Optimal Nursing Care Intensity Level,” a new method that goes beyond the traditional time study methodology and could establish optimal nursing intensity levels for individual units. It calculates nursing intensity based on patient classification results and assesses nursing intensity for nursing units by reflecting statistical estimations and expert opinions. Calculating nursing time Nursing time was measured using a questionnaire developed by this study's researchers with reference to non‐direct nursing activities in a tool developed by Ko and Park (Ko et al., ). This questionnaire comprises 28 items, including total working, break and non‐direct nursing times of the day. Total working time was calculated based on the time at which nurses logged into work and left for the day, while the break time was calculated by summing up the meal and rest times. Non‐direct nursing time was calculated by summing the time for each subdomain of the three domains (nursing activity planning and coordination, non‐direct activity and break time). Direct nursing time was calculated by subtracting the non‐direct nursing time including leisure time from the total working time (Formula in Appendix ). Six nursing professors and one expert reviewed the validity of the direct nursing calculation method. Calculating direct nursing time per inpatient by patient classification groups To calculate direct nursing time by patient classification groups, direct nursing time per nursing intensity point was calculated (Formula in Appendix ). This value was then multiplied by the weighting coefficient for each patient classification group to calculate direct nursing time per inpatient by patient classification groups (Formula in Appendix ). Calculating the optimal number of nurses The optimal number of nurses in the internal medicine, surgical, and comprehensive nursing care units was estimated by applying the calculated nursing time results to Formulas (4–6) in Appendix . After non‐direct nursing time was estimated using the ratio (20%) of non‐direct nursing time to the total nursing working time—calculated with the nursing time analysis results—the total nursing time was measured (Formula in Appendix ). The optimal number of nurses was calculated by adding 40% to the value obtained by dividing the total nursing work time by the mean daily work hours (Formula in Appendix ). The total number of annual holidays in the current clinical reality is estimated to be about 134 days, considering weekly holidays: 52 weeks × 2 (Saturday and Sunday) based on an average of 20 working days per month, plus 15 legal holidays (excluding Sundays), 15 basic annual holidays and additional annual holidays according to the nurses' professional positions. Although 1.4 can be assigned as an additive value owing to the number of holidays by rounding off 1.37 [(134 + 365)/365], a constant of 1.6 was used in this study following previous studies (Cho et al., ; Lee et al., ). Data analysis The collected data were analysed using the Microsoft Excel program. The participants' general characteristics, direct nursing time and nursing intensity for each date by patient classification groups were analysed using descriptive statistics such as frequency, percentage and average. Direct nursing time among the nursing unit nurses, direct nursing time per patient classification point or nursing intensity point and direct nursing time per patient were calculated using Microsoft Office Excel 2017. Ethical considerations This study was conducted after explaining its purpose to the head of the nursing department at the study hospital. The researchers visited the nursing units to explain this study's purpose. Nurses who agreed to participate, in writing were selected. This study was approved by the Institutional Review Board at a university to which the author belongs (Approval No: 1040271‐201808‐HR‐026). A study on the calculation of the optimal number of nurses based on nursing intensity in the intensive care unit using the same research model has been published in the Korean Journal of Hospital Management (Ko & Park, ). RESULTS 3.1 Calculation of patient classification and nursing intensity based on nursing needs The average daily number of patients was 39.7, 67.7 and 20.4 while patient classification scores per patient were 64.1, 54.7 and 51.0 points in the internal medicine, surgical and comprehensive nursing care units, respectively. Nursing intensity scores were 5.4, 5.7 and 2.4 in the internal medicine, surgical and comprehensive nursing care units, respectively (Table ). 3.2 Calculation of nursing time The proportion of nurses' non‐direct nursing time and break time out of total working time was 41.1%, 46.9% and 74.4% and 4.1%, 3.0% and 5.1% in the internal medicine, surgical and comprehensive nursing care units, respectively. The proportion of direct nursing time was calculated to be 54.7%, 50.0% and 20.5% in the internal medicine, surgical and comprehensive nursing care units, respectively. Specifically, the proportion of mealtime was 2.5%, 2.1% and 3.3% and when converted into time required per nurse, it was 15.5, 12.7 and 18.8 min for the internal medicine, surgical and comprehensive nursing care units, respectively. The average handover time per nurse was 30.5, 44.4 and 58.8 min in the internal medicine, surgical and comprehensive nursing care units, respectively (Table ). 3.3 Direct nursing time per patient‐by‐patient classification groups The direct nursing time per patient calculated through patient classification groups in internal medicine unit was 1.0, 1.5, 2.2 and 2.9 h for Groups 1, 2, 3 and 4, respectively; and in the surgical unit was 0.9, 1.4, 2.1 and 2.6 h for Groups 1, 2, 3 and 4, respectively. Furthermore, in the comprehensive nursing care unit it was 0.8, 1.2, 1.7 and 2.2 h for Groups 1, 2, 3 and 4, respectively (Table ). 3.4 Calculation of optimal number of nurses in the internal medicine unit The optimal number of nurses in the internal medicine, surgical and comprehensive nursing care units was 25 ( n = 24.6), 37 ( n = 36.9) and 22 ( n = 21.2), respectively. At the time of data collection, 20, 28 and 22 nurses were assigned in these units, respectively. Therefore, if the number of previously assigned nurses was subtracted from the calculated optimal number, 5 and 9 additional nurses would be needed in the internal medicine and surgical units, respectively; however, the number of nurses in the comprehensive nursing care unit was deemed appropriate (Table ). Calculation of patient classification and nursing intensity based on nursing needs The average daily number of patients was 39.7, 67.7 and 20.4 while patient classification scores per patient were 64.1, 54.7 and 51.0 points in the internal medicine, surgical and comprehensive nursing care units, respectively. Nursing intensity scores were 5.4, 5.7 and 2.4 in the internal medicine, surgical and comprehensive nursing care units, respectively (Table ). Calculation of nursing time The proportion of nurses' non‐direct nursing time and break time out of total working time was 41.1%, 46.9% and 74.4% and 4.1%, 3.0% and 5.1% in the internal medicine, surgical and comprehensive nursing care units, respectively. The proportion of direct nursing time was calculated to be 54.7%, 50.0% and 20.5% in the internal medicine, surgical and comprehensive nursing care units, respectively. Specifically, the proportion of mealtime was 2.5%, 2.1% and 3.3% and when converted into time required per nurse, it was 15.5, 12.7 and 18.8 min for the internal medicine, surgical and comprehensive nursing care units, respectively. The average handover time per nurse was 30.5, 44.4 and 58.8 min in the internal medicine, surgical and comprehensive nursing care units, respectively (Table ). Direct nursing time per patient‐by‐patient classification groups The direct nursing time per patient calculated through patient classification groups in internal medicine unit was 1.0, 1.5, 2.2 and 2.9 h for Groups 1, 2, 3 and 4, respectively; and in the surgical unit was 0.9, 1.4, 2.1 and 2.6 h for Groups 1, 2, 3 and 4, respectively. Furthermore, in the comprehensive nursing care unit it was 0.8, 1.2, 1.7 and 2.2 h for Groups 1, 2, 3 and 4, respectively (Table ). Calculation of optimal number of nurses in the internal medicine unit The optimal number of nurses in the internal medicine, surgical and comprehensive nursing care units was 25 ( n = 24.6), 37 ( n = 36.9) and 22 ( n = 21.2), respectively. At the time of data collection, 20, 28 and 22 nurses were assigned in these units, respectively. Therefore, if the number of previously assigned nurses was subtracted from the calculated optimal number, 5 and 9 additional nurses would be needed in the internal medicine and surgical units, respectively; however, the number of nurses in the comprehensive nursing care unit was deemed appropriate (Table ). DISCUSSION This study calculated the optimal number of nurses in each nursing unit using a comprehensive approach based on the total working time, indirect nursing time and nursing intensity, considering all nurses in each nursing unit. Whether nursing activities that occur simultaneously with nursing time or only key nursing activities should be reflected in nursing time measurements is a topic of debate. According to the data from the Labor Union of Korean Healthcare Service, the average working time per week among hospital nurses in 2016 was 46.6 h, with 12% of them working longer than 52 h per week (Statistics Korea, ).In this study, the average working time per week was 51, 51 and 47.6 h in the internal medicine, surgical and comprehensive nursing care units, respectively, indicating a slightly longer duration than that reported by the Labor Union of Korean Healthcare Service (Statistics Korea, ).The meal timeout of the total working time was the longest in the comprehensive nursing care unit; however, the mealtime in all three units indicated that nurses hurriedly eat meals in limited time (20 min). The Labor Standards Act, Article 54 (Korea Ministry of Government Legislation, ) stipulates that workers should be given a rest break of 30 min or longer or 1 h or longer if they work 4 or 8 h, respectively. Break times including meal and personal times were very small, comprising 4% of the total working time, suggesting that there is high working intensity in clinical practice. Therefore, it is essential to secure a minimum mealtime, provide break opportunities to enhance nurses' well‐being and create a nursing working environment that encourages nurses to serve effectively for long periods (Hwang & Bae, ). Nursing intensity includes nurses' technical and physical efforts, mental efforts, and judgement and nursing time. However, this study calculated a representative value of nursing intensity in each nursing unit using weighting coefficients by patient classification groups (Fagerström et al., ). The results revealed that the nursing intensity in the surgical unit was the highest. This nursing intensity index makes it possible to directly compare nursing intensity scores and can be effectively used for nurse staffing at hospitals. The total nursing intensity score in the nursing units was 67.49, 101.55 and 26.52 points in the internal medicine, surgical and comprehensive nursing care units, respectively. Compared with the results by Fagerström and Rauhala ( –2002), where the average nursing intensity per patient was found to be 13.4 points in 86 nursing units of 14 Finnish hospitals, the nursing intensity in this study was high. Although it is difficult to directly compare the two studies, such a difference in nursing intensity may be attributed to differences in working environments, data collection period and study participants. Park et al. estimated nursing costs by patient classification groups in a general nursing unit and revealed that the proportion of direct nursing time was 44.1% of the average total nursing time per patient per day, whereas that of non‐direct nursing time was 55.9%. The results of this study revealed that the proportion of non‐direct nursing time in the internal medicine and surgical units was less than 50%, whereas that of non‐direct nursing time was 74.4%, indicating a distinct difference according to the units' characteristics. To determine the appropriate nurse staffing in nursing units, factors related to a particular patient's nursing needs should be identified (Padilha et al., ). Fagerström et al. claimed that personal characteristics of inpatients, such as gender and health status, affected their nursing needs, thereby affecting nurses' workload and nursing intensity. It is also important to distinguish between patient‐ and unit‐related workload when measuring nurses' workload by a patient classification system based on the measurement of nursing intensity. When measuring nursing intensity, only nurses' workload related to patients was measured, whereas workload related to the unit to which nurses belonged was not included (Morris et al., ). Thus, although nurses' workload may increase owing to factors related to the respective units, their unit‐related workload may be overlooked and may thus be underestimated. For example, some patients are unstable and require continuous observation and intensive care; therefore, although the severity is high, their dependence may be low because they can complete their activities of daily living by themselves. However, patients in the rehabilitation/recovery phase may be stable; therefore, although their severity is low, their dependence may be high as they cannot complete their daily activities by themselves. Therefore, it is more reasonable to independently evaluate and typify the two areas of severity and dependence rather than evaluating the need for nursing as one dimension (Hoi et al., ). To determine whether the nurse staffing level is appropriate for patient care, an accurate basis for nursing needs with a detailed patient classification tool to calculate the optimal number of nurses is required (Kim et al., ). However, if additional indicators or scales are added, it will be an added burden for nurses to observe patients and record data every day as in the current system, with the accuracy of data being lower than it is at present (Kim et al., ). The results of this study revealed that the optimal number of nurses in each of the nursing units corresponded to the nurse staffing grade 1 level. This proves that applying the nurse staffing grade 1 for general units to the respective nursing units is an appropriate staffing level. South Korea's nursing workforce problem is not a shortage of licensed nurses, but one of an active nursing workforce, owing to working conditions. To create an environment in which experienced nurses can continue to work effectively, it is necessary to reflect an experienced nurse retaining index in the reimbursement policies or to develop policies that offer economic rewards for experienced nurses and build stable working environments. Our study's limitations are as follows: Since it involved only three nursing units at a single hospital, its generalizability may be limited to South Korea. Repetitive studies are needed to verify whether the research method used is applicable even in situations where the size and type of medical institutions, nursing units' characteristics, and the level of securing nursing personnel are different. Despite these limitations, this study's results are significant. First, previous studies have measured nursing needs to calculate nursing workload. However, this study measured nursing intensity in the nursing units using patient classification scores based on nursing needs, presenting it as coefficients in the nursing units. This can be a useful indicator for the allocation of resources such as workforce at hospitals by comparing nursing intensity between nursing units. Second, this study proposed a new approach by calculating optimal nurse staffing. CONCLUSIONS This study attempted to calculate not only the direct nursing time per patient and nursing intensity per nurse in nursing units using data from a survey on total working time and indirect nursing time, but also the direct nursing time by patient classification group and the optimal number of nurses in nursing units to provide basic data for calculating the required number of nursing personnel. The nursing intensity score per nurse in each nursing unit was the highest in the surgical unit, followed by the internal medicine and the comprehensive nursing care units. The study results revealed that additional nurses were needed in the internal medicine and surgical nursing units and break time was not properly guaranteed during working hours. Therefore, it is necessary to secure an optimal number of nurses through additional recruitment and to create a working environment encouraging nurses to work effectively for long periods. IMPLICATIONS FOR NURSING MANAGEMENT It is essential to provide nurses a work environment with adequate nursing staff to optimize their professional skills. When planning labour requirements for each nursing unit, the time required for direct and indirect nursing activities should be considered, and nursing appointments should satisfy patient needs. Recommendations for nursing management include objectively evaluating the extent of nursing activities and providing placement criteria according to the patient's level of nursing need. These measures can help increase nurses' motivation and functioning. YK contributed to make study design, investigation, supervision, project administration, funding acquisition and writing. BP contributed to software utilization, formal analysis and data curation. Both authors read and approved the final manuscript for publication. There are no conflict of interest to declare. ETHICAL APPROVAL This study was approved by the Institutional Review Board at a university to which the author belongs (Approval No: 1040271‐201808‐HR‐026).
Supporting recovery in persons with stress-related disorders: A reflective life world research study of health care professionals in primary health care in Sweden	26997d65-8d67-4cfe-b324-b5ca343d3299	10171129	Patient-Centered Care[mh]	Stress-related disorders constitute a major public health problem globally, and they are the most common reason for sick leave in Sweden (Försäkringskassan, ). Stress-related disorders have a major impact on health and lives, and require resources for beneficial care to support recovery. Primary health care has an important role in this respect, and in this study we focus on the lived experiences of primary health care professionals supporting recovery in persons with stress-related disorders. The concept of stress can be operationalized in various ways, and several diagnoses are used. Selye ( ) defines stress as a general reaction to a stimulus that may pose a challenge and/or threat. Lazarus and Folkman ( ) describe stress in the “transactional model” as an imbalance between perceived demands and the availability of the resources required to handle a situation. Prolonged and persistent stress without sufficient rest can lead to a gradual development of stress-related disorders (Danielsson et al., ; Glise, ). Such conditions often include exhaustion, cognitive dysfunction, sleep disturbance, reduced tolerance to further stress and somatic symptoms (Bernardsson et al., ). A common work-related, non-diagnostic term internationally is “burnout” (Maslach & Leiter, ). Several diagnoses are related to stress-related disorders, e.g., adjustment disorder and reaction to severe stress. In Sweden, the diagnosis of exhaustion disorder is commonly used (SE, ). Understanding stress-related disorders and care for persons on sick leave is nevertheless complex, and a purely diagnostic perspective limits a holistic view of the person. When a person experiences illness or suffering, such as in a stress-related disorder, this affects the body and therefore one’s whole existence (K. Dahlberg & Segesten, ). Jingrot and Rosberg ( ) describe the progress of stress-related disorders as an increasing loss of “homelikeness” in the body and the familiar world. Furthermore, suffering from stress-related disorders has consequences for personal identity, and brings about existential concerns (Alsén et al., ) as well as a search for meaning (Engebretsen & Bjorbaekmo, ). Persons with stress-related disorders have also been found to feel that their interactions with significant others have been disrupted, threatening their family role and relationships (Engebretsen & Bjorbaekmo, ). These existential challenges, and long-term sick leave, affect the person and his/her family. The need for effective care that can support the recovery process is crucial. The notion of recovery has developed in recent decades as an alternative to the patient/illness focus of psychiatric medicine (Buchanan-Barker & Barker, ). Recovery can be understood as a personal process or journey to regain power and live a fulfilling life, in which hope, optimism, meaning and togetherness are important components (Anthony, ; Topor et al., ; van Weeghel et al., ). The initial meaning of recovery integrates persons in their social context but has over the years been diluted and taken a more individualistic form. In recent years explanation of recovery as only a personal journey has re-evolved to include the person´s social context were social aspects as relationships and living conditions are considered as important for recovery (Topor et al., , ). In a modern recovery-oriented practice, it is important as a health care professional to have a holistic approach. This means to support the person´s own thoughts, experiences and opinions and simultaneously support the person in a social context. A crucial aspect in this approach is a working relationship built on reciprocity and hope (Le Boutillier et al., ; Topor et al., ). To support recovery, care for persons with stress-related disorders commonly occurs within primary health care (Wiegner et al., ). Sufficient evidence is lacking regarding the effect of rehabilitation and treatment for stress-related disorders, especially with a focus on primary health care (Swedish Agency for Health Technology Assessment and Assessment of Social Services, ; Wallensten et al., ). Various interventions have been found to have some effect, including workplace interventions, sleep-improvement interventions, and aerobic and cognitive training. Cognitive behavioural therapy, nature-based rehabilitation and multimodal intervention seem to reduce stress symptoms while the intervention is ongoing but lack effect upon earlier return to work (Wallensten et al., ). Engebretsen and Bjorbaekmo ( ) suggest a need for humanistic interventions, and Jingrot and Rosberg ( ) suggest interventions that help a person to regain attachment to the body and the world, with a focus on bodily experiences and habitual stress-related patterns. These unconclusive results and recommendations leave health care professionals lacking guidance regarding treatment of stress-related disorders, and a deeper understanding how they actually support recovery is needed. Previous studies have found that persons suffering from stress-related disorders experience contact with primary health care as a conflict involving mistrust, and which results in feelings of shame that negatively influence the recovery process (Engebretsen & Bjorbaekmo, ). Support from staff, peers and family and friends has been described as beneficial for recovery (Salminen et al., ). Arman et al. ( ) conveys that seeking to reach beyond the defences of a person with a stress-related disordered, and persistently supporting their exploration of an existential understanding of life, is a way to help. The humanistic, existential, relational and supportive aspects of care seem important for recovery from stress-related disorders, but little is known regarding the health care professional’s experiences of supporting such persons in a primary health care context. Wiegner et al. ( ) explores how health care professionals practiced and perceived their task as “care managers” in primary health care. This entailed facilitating effective, person-centred treatment for stress-related disorders concordant with evidence-based guidelines. The health care professionals experienced increased continuity, early detection of deterioration, providing self-to-self help, being sensitive to the person’s needs and grounding contact in an alliance to be important (Wiegner et al., ). Yet, the care-manager role may be different from the support given by primary health care professionals. To our knowledge, no previous study has studied this phenomenon from the health care professional’s perspective. Health care professionals who have cared for persons with stress-related disorders in primary health care possess valuable experiences that could be used to gain knowledge, improve care and support the recovery process. The study aimed to describe primary health care professionals lived experiences of supporting recovery in persons with stress-related disorders. Design This study is based on a phenomenological approach known as reflective lifeworld research (RLR). This approach is founded Husserl’s philosophical theory of the lifeworld, the theory of intentionality and Mearleu-Ponty’s theory of the lived body (K. Dahlberg et al., ). To describe the phenomenon “supporting recovery in persons with stress-related disorders in primary health care”, the RLR methodological principles of openness, flexibility and bridling were used. This can be understood as being open, flexible and to have a reflective attitude towards one’s understanding of the phenomenon. These principles are used throughout the research process in order to avoid to grasp and describe the specific phenomenon to quickly and unreflected (K. Dahlberg et al., ). Participants and settings Health care in Sweden is managed by decentralized regions or municipalities. Both public and private facilities exist, and basic health and medical care is referred to as primary health care (National Board of Health and Welfare, . In primary health care centres, multidisciplinary teams work together and comprise various occupations including physicians, registered nurses, specialized registered nurses, occupational therapists, psychologists and physiotherapists (National Board of Health and Welfare, ). Participants were recruited partly via email to primary health care centres in the southern part of Sweden and partly by advertising nationally in social media. Invitations to participate and a written information letter were sent to potential participants. The inclusion criteria were being a health care professional in a primary health care context and having experience with caring for persons with stress-related disorders. Interested participants contacted the research group, and a purposive sample was used for sample variation in terms of, e.g., sex, age, occupation and professional experience. Included participants were contacted, and the time, place and interview means (in-person, telephone, Skype, etc.) were decided. Informed consent was provided by the participant in connection with the interview. In total, 32 persons contacted the research group to participate, of whom were 17 included to ensure variation and a manageable data quantity. One request to participate came via regular post to the university which, due to COVID-19 restrictions, was not discovered until data collection was complete, and was therefore not included. The 17 included health care professionals comprised 13 women and 4 men, aged 31–65. The range of primary health care experience was 1–30 years (see for characteristics of the participants). Data collection The lifeworld interviews were conducted in autumn and winter 2020/2021. Five interviews were conducted in person at the primary health care centres, eleven interviews via digital platforms and one via telephone. The interviews began with an introductory question: “Can you tell me about how you support recovery in persons with stress-related disorders?”. The introductory question was seen as an opportunity to invite descriptions of lived experiences. When participants touched on the phenomenon, follow-up questions were asked to help the participant deepen the reasoning about his/her lived experiences of the phenomenon. Follow-up included “Can you describe a situation?”, and “Can you give one or a few examples?” All interviews were conducted by the first author. The interviews were audio-recorded, lasted between 47–104 minutes and were transcribed verbatim. Data analysis Interview transcripts were analysed in accordance with phenomenological RLR principles. In RLR, it is crucial to be open minded, flexible and patient in understanding the phenomenon (H. Dahlberg & Dahlberg, ; K. Dahlberg et al., ). First, transcripts were read repeatedly to get to know the material. This process was characterized by a movement between “the whole—the parts—the whole” in order to reach a new wholeness. Analysis began with the whole being broken down into parts, and searching for section of text that carried the meaning of the phenomenon “health care professionals lived experiences of supporting recovery in persons with stress-related disorder in primary health care”. Meanings were compared based on differences and similarities, and grouped into clusters of meanings. Analysis continued with the search for patterns of meanings. To find these patterns, it is important to search the “in between”, the place between the researcher and the phenomenon, the in-between world connecting us with other subjects and objects in the world (K. Dahlberg, ). With an in-between perspective, the analysis continued with “figure and background,” putting contrasting individual meanings and cluster-groups, and comparing meanings to cluster groups and vice versa, while constantly trying to read between the lines and discover underlying patterns between meanings. One must adopt a bridled attitude, which requires the researcher to reflect on his/her own pre-conceptions and slow down the process of understanding, to permit the phenomenon to remain indefinite for as long as possible and the phenomenon’s essential structure to emerge. During the analysis process, the image of an essential structure of meaning gradually became clearer and clearer. The essential structure is the most abstract description of the phenomenon and how meanings relate to each other, and when the essential structure is clarified one can describe its constituents. The constituents describe variations of the phenomenon and are described individually. The essential structure and the constituents form together a new whole (K. Dahlberg et al., ). The analysis process was led by the first author and then repeatedly discussed by all authors. The authors had a lengthy reflection process where their own work with the data alternated with joint meetings. New findings were critical discussed and questioned in a bridled attitude throughout the process, always with their preunderstanding in mind. The first author has experience of supporting persons with stress related disorders and works part time as a specialized psychiatric nurse in a primary health care centre. The other authors work as experienced researchers with a background as nurses in different parts of the health care system (see biographical notes for more information). Ethical considerations The study was approved by the Swedish Ethical Review Authority (Reg. No. 2020 03163). Decisions have been made pursuant to the Declaration of Helsinki (WMA, ) and the Swedish Ethical Review Act (SFS 2003:460). The respondents were informed in writing prior to the interview, and verbally at the time of the interview, regarding the aim of the study, their freedom to withdraw at any time and without explanation and that confidentiality applied throughout the research process. All participants provided informed consent before the interview. Due to a relative high number of willing participants during recruitment, we had to turn some down. For ethical reasons, it was important for us to explain all details to these potential participants and thank them for replying and for their willingness to support research. None expressed any discomfort for not being included. This study is based on a phenomenological approach known as reflective lifeworld research (RLR). This approach is founded Husserl’s philosophical theory of the lifeworld, the theory of intentionality and Mearleu-Ponty’s theory of the lived body (K. Dahlberg et al., ). To describe the phenomenon “supporting recovery in persons with stress-related disorders in primary health care”, the RLR methodological principles of openness, flexibility and bridling were used. This can be understood as being open, flexible and to have a reflective attitude towards one’s understanding of the phenomenon. These principles are used throughout the research process in order to avoid to grasp and describe the specific phenomenon to quickly and unreflected (K. Dahlberg et al., ). Health care in Sweden is managed by decentralized regions or municipalities. Both public and private facilities exist, and basic health and medical care is referred to as primary health care (National Board of Health and Welfare, . In primary health care centres, multidisciplinary teams work together and comprise various occupations including physicians, registered nurses, specialized registered nurses, occupational therapists, psychologists and physiotherapists (National Board of Health and Welfare, ). Participants were recruited partly via email to primary health care centres in the southern part of Sweden and partly by advertising nationally in social media. Invitations to participate and a written information letter were sent to potential participants. The inclusion criteria were being a health care professional in a primary health care context and having experience with caring for persons with stress-related disorders. Interested participants contacted the research group, and a purposive sample was used for sample variation in terms of, e.g., sex, age, occupation and professional experience. Included participants were contacted, and the time, place and interview means (in-person, telephone, Skype, etc.) were decided. Informed consent was provided by the participant in connection with the interview. In total, 32 persons contacted the research group to participate, of whom were 17 included to ensure variation and a manageable data quantity. One request to participate came via regular post to the university which, due to COVID-19 restrictions, was not discovered until data collection was complete, and was therefore not included. The 17 included health care professionals comprised 13 women and 4 men, aged 31–65. The range of primary health care experience was 1–30 years (see for characteristics of the participants). The lifeworld interviews were conducted in autumn and winter 2020/2021. Five interviews were conducted in person at the primary health care centres, eleven interviews via digital platforms and one via telephone. The interviews began with an introductory question: “Can you tell me about how you support recovery in persons with stress-related disorders?”. The introductory question was seen as an opportunity to invite descriptions of lived experiences. When participants touched on the phenomenon, follow-up questions were asked to help the participant deepen the reasoning about his/her lived experiences of the phenomenon. Follow-up included “Can you describe a situation?”, and “Can you give one or a few examples?” All interviews were conducted by the first author. The interviews were audio-recorded, lasted between 47–104 minutes and were transcribed verbatim. Interview transcripts were analysed in accordance with phenomenological RLR principles. In RLR, it is crucial to be open minded, flexible and patient in understanding the phenomenon (H. Dahlberg & Dahlberg, ; K. Dahlberg et al., ). First, transcripts were read repeatedly to get to know the material. This process was characterized by a movement between “the whole—the parts—the whole” in order to reach a new wholeness. Analysis began with the whole being broken down into parts, and searching for section of text that carried the meaning of the phenomenon “health care professionals lived experiences of supporting recovery in persons with stress-related disorder in primary health care”. Meanings were compared based on differences and similarities, and grouped into clusters of meanings. Analysis continued with the search for patterns of meanings. To find these patterns, it is important to search the “in between”, the place between the researcher and the phenomenon, the in-between world connecting us with other subjects and objects in the world (K. Dahlberg, ). With an in-between perspective, the analysis continued with “figure and background,” putting contrasting individual meanings and cluster-groups, and comparing meanings to cluster groups and vice versa, while constantly trying to read between the lines and discover underlying patterns between meanings. One must adopt a bridled attitude, which requires the researcher to reflect on his/her own pre-conceptions and slow down the process of understanding, to permit the phenomenon to remain indefinite for as long as possible and the phenomenon’s essential structure to emerge. During the analysis process, the image of an essential structure of meaning gradually became clearer and clearer. The essential structure is the most abstract description of the phenomenon and how meanings relate to each other, and when the essential structure is clarified one can describe its constituents. The constituents describe variations of the phenomenon and are described individually. The essential structure and the constituents form together a new whole (K. Dahlberg et al., ). The analysis process was led by the first author and then repeatedly discussed by all authors. The authors had a lengthy reflection process where their own work with the data alternated with joint meetings. New findings were critical discussed and questioned in a bridled attitude throughout the process, always with their preunderstanding in mind. The first author has experience of supporting persons with stress related disorders and works part time as a specialized psychiatric nurse in a primary health care centre. The other authors work as experienced researchers with a background as nurses in different parts of the health care system (see biographical notes for more information). The study was approved by the Swedish Ethical Review Authority (Reg. No. 2020 03163). Decisions have been made pursuant to the Declaration of Helsinki (WMA, ) and the Swedish Ethical Review Act (SFS 2003:460). The respondents were informed in writing prior to the interview, and verbally at the time of the interview, regarding the aim of the study, their freedom to withdraw at any time and without explanation and that confidentiality applied throughout the research process. All participants provided informed consent before the interview. Due to a relative high number of willing participants during recruitment, we had to turn some down. For ethical reasons, it was important for us to explain all details to these potential participants and thank them for replying and for their willingness to support research. None expressed any discomfort for not being included. Supporting recovery in persons with stress-related disorders in primary health care means to meet the person where, in relation to recovery, he or she describes his/her lifeworld and experience of the stress-related disorder. By remaining aware, the health care professional tailors the caring interaction and support to the needs of the person and their narrative. A mutual, interpersonal communication, as well as new ways of understanding lifeworld narratives, is essential for recovery support. Encounters between health care professionals and the person create a unique shared interpersonal platform for care that moves between the person’s life narrative and the health care professional’s understanding of that narrative. This platform creates a shared space for reflection regarding the prevailing imbalance in the person’s unsustainable life situation and the lack of sufficient rest. Recovery support is permeated by an alliance with the person being cared for. This alliance constitutes the “interpersonal glue” of the care relationship, with the care relationship as the external structure of the alliance. Supporting recovery means that the person has space to participate in their care, and this space is created by sensing and accepting the person’s mood, and keeping the alliance in focus. It means looking beyond instrumental parts of caring to encounter the person’s existential situation. A lingering and flexible approach supporting growing insight into a sustainable life balance is needed to support recovery. Given the unpredictability of a recovery journey, care requires continuity and flexibility. Supporting the recovery process can be challenging, but can also result in feeling able to make a difference by supporting a person. Supporting recovery also means having a guiding approach that actively supports the person in learning to rediscover and consider his/her own needs for a mode of life that is sustainable in the long term. The following constituents further highlight the meaning of the phenomenon “supporting recovery in persons with stress-related disorders in primary health care”: The caring alliance, The interpersonal platform, The lingering and flexible approach and The guiding approach . The caring alliance Supporting recovery of persons with stress-related disorders in primary health care means using a caring alliance to instil courage that allows the person to dare to start exploring and changing unhealthy modes of life. The formation of an alliance occurs parallel to the care process. Health care professionals experience this as both difficult and beneficial in relation to the challenges of the recovery process. A continuous, non-judgemental and attentive approach, where the person is taken seriously, lays the foundation for an alliance in which the person can feel trust and security, and can start taking responsibility for his/her recovery journey, first at the primary health care centre and then in their own life. “First of all, I think it’s important, or it is really the starting point, that the patient feels safe in the care relationship, in the care meeting. And if you feel safe, there is often an greater calm”. (2) A supporting alliance provides a basis for interaction with a permissive atmosphere where the person cared for has space to express his/her feelings without fear of being judged. The alliance allows the person to feel safe describing in depth how they feel. It is constantly revised and continuously calibrated, an approach aimed at an “alliance-promoting” atmosphere with both verbal and non-verbal communication. “It’s about creating that sense of security in the consulting room that allows talking about … that I can get a patient to discuss difficult things and ask questions, and that we can work with the resistance which sometimes arises. This creates the opportunity for a confidential conversation”. (11) The “alliance-promoting” atmosphere can also be maintained with humour and through sharing personal experiences in care. It also entails that the person feels allied with the interprofessional team at the primary health care centre. The interpersonal platform When encountering persons with stress-related disorders, an interpersonal interaction is sought, where the person is gradually supported in reflecting on and learning from his/her life narrative. This forms an interpersonal platform, and care takes place and is created within the relationship with the person. The primary health care centre’s interprofessional resources are used as necessary. Different professions can have different theoretical bases and different methods for understanding the person’s narrative. The person’s narrative is inventoried, and unhealthy stress-related patterns are identified and considered. The person’s ability to recover is confirmed and new recovery strategies are formulated. The interpersonal interaction is the framework, while the result of the interaction—the gradually extended narrative—stabilizes the platform as a supporting element. The platform is co-created by health care professionals as they strive to understand the person’s life narrative. In the caring encounter, verbal language, body language and intuition are used to try to understand the person. “On the one hand, I’m very empathic. I check in. Then it involves asking: what does your everyday life look like? What do you like to do? What’s important to you? What do you need—both in terms of this conversation, but also a question you ask yourself daily, what do I need right now? Do you touch base in that way sometimes?” (1) Listening and intuitively feeling the person captures different expressions of the stress-related disorder and their ability to recover. The discrepancy between what the person conveys, and what the health care professional understands, creates a space for a caring conversation. This space is gradually filled with content through questions and considerations regarding what the person expresses, while at the same time ensuring correct understanding. The interaction gives rise to further consideration of and questions about the person’s life. “You need to have the confidence that I’m still here. I am listening to you, I am affirming you, I ensure that I have understood you correctly so as not to pass judgment or imagine I understand something when you didn’t mean it that way at all. Without me touching base, as it were, in the conversation—is this what you mean when you say this, do you mean this, have I understood you correctly now? That touching base makes us a team.” (1) Switching between the encounter with the person and the narrative occurs simultaneously using a so-called “helicopter perspective” of everything explored and understood earlier during the care process. New information about the person’s life situation is contrasted with the narrative constantly being revised. This involves a shift between proximity to the person’s existential reflections and reviewing the person’s described context. Thus he or she is supported in considering imbalances in his/her life situation related to a stress-related disorder, in understanding how different parts are connected and in a search for balance and recovery. On the interpersonal platform, the person is supported in understanding the different expressions of the stress-related disorder, their effect on the person’s life and the importance of recovery. The interpersonal platform provides opportunities for personal growth and change. The lingering and flexible approach The person returns to the primary health care centre in encounters with health care professionals who use a lingering and flexible approach to adapt their attitude to the person and their journey. The person is involved and receives gradually support in assuming greater responsibility for rediscovering his/her ability to initiate a recovery process and recover. “I think that time helps you to, like—I don’t like the word process, it’s an awful word really but for something like that to start, I mean if you’re talking about a person with exhaustion, an exhaustion disorder or other concerns that are really severe and stress-related, then you need to process this. Yes. You need to work around it in some way, and get closer to the idea that this is how it is, this is how I am, this is how I function, so what should I do to feel well?.” (11) The lingering and flexible approach is characterized by waiting for the person’s imminent insight and, further, waiting for his/her deeper understanding of how to manage unsustainable stress-related patterns. Beginning by recognizing and considering bodily signals and personal boundaries, as a way start to changing such patterns and developing and maintaining a necessary recovery process, changes life in a way that naturally results in recovery. This takes place in a changing process over time, and the person is encouraged to continue even after the care process has ended. “You can’t expect to get well in a month. It may not take an eternity, but it might take a while, and that’s something that I prepare patients for. We have to work together so that you get well, so that you get out of this. And it will take time, it will be demanding, but you won’t need to be on sick leave for years if you can tackle stress management and change your attitude towards different stressors over the course of several years until you end up, until you have everything ….in place.” (17) This process involves various aspects of the person’s life and is both relational and individual, personal and professional. In primary health care, supporting recovery must be a “safe haven” to which the person can return over time, to reflect, learn and rehearse, and then take on daily challenges with new knowledge and insight. The goal is that the recovery process gradually become more dominant in the person’s life. The guiding approach Supporting recovery with a guiding approach aims to help the person understand the difference between a state of stress and a state of recovery. The guiding approach means supporting the person in finding “clues” to a sustainable recovery process over time. The person receives guidance in using his/her senses, slowing down the body’s pace, anchoring themselves in their bodies and thus listening to themselves and considering his/her need for recovery. To strengthen these “clues” leading to a sustainable balance, care is based on providing verbally, visually or physically descriptive examples of the body’s need for balance. This may involve verbal metaphors, illustrative images, body awareness or encouraging a focus on sensing one’s surroundings, e.g., in nature. “Then we try … we try together, of course, with the patient. It’s better if the patient comes up with things that they want to change, but otherwise you have to provide help and maybe indicate some things you see are not sustainable, you can make suggestions.” (7) This guiding approach receives additional dimensions when provided for several persons with similar experiences who meet in a group. Supporting recovery in these group contexts means to providing an interconnecting link and being responsible for the structure and content of the encounters. Sharing with others reinforces the effects of the guiding approach and provides synergistic effects, such as recognizing oneself in others, feeling that one is not alone in one’s troubles and feeling understood. “The group is tremendously beneficial for all its members, they contribute a lot that you miss when you meet individually, so group treatment is ….it’s not possible to do it as well individually, I think.” (7) Another guiding approach is the involvement of significant others. This is done by providing information to increase understanding of the person’s situation and their needs in, e.g., family life. Significant others are involved as a support, either primarily in care through direct contact, or secondarily through supporting the person to communicate with his/her relatives. Supporting recovery of persons with stress-related disorders in primary health care means using a caring alliance to instil courage that allows the person to dare to start exploring and changing unhealthy modes of life. The formation of an alliance occurs parallel to the care process. Health care professionals experience this as both difficult and beneficial in relation to the challenges of the recovery process. A continuous, non-judgemental and attentive approach, where the person is taken seriously, lays the foundation for an alliance in which the person can feel trust and security, and can start taking responsibility for his/her recovery journey, first at the primary health care centre and then in their own life. “First of all, I think it’s important, or it is really the starting point, that the patient feels safe in the care relationship, in the care meeting. And if you feel safe, there is often an greater calm”. (2) A supporting alliance provides a basis for interaction with a permissive atmosphere where the person cared for has space to express his/her feelings without fear of being judged. The alliance allows the person to feel safe describing in depth how they feel. It is constantly revised and continuously calibrated, an approach aimed at an “alliance-promoting” atmosphere with both verbal and non-verbal communication. “It’s about creating that sense of security in the consulting room that allows talking about … that I can get a patient to discuss difficult things and ask questions, and that we can work with the resistance which sometimes arises. This creates the opportunity for a confidential conversation”. (11) The “alliance-promoting” atmosphere can also be maintained with humour and through sharing personal experiences in care. It also entails that the person feels allied with the interprofessional team at the primary health care centre. When encountering persons with stress-related disorders, an interpersonal interaction is sought, where the person is gradually supported in reflecting on and learning from his/her life narrative. This forms an interpersonal platform, and care takes place and is created within the relationship with the person. The primary health care centre’s interprofessional resources are used as necessary. Different professions can have different theoretical bases and different methods for understanding the person’s narrative. The person’s narrative is inventoried, and unhealthy stress-related patterns are identified and considered. The person’s ability to recover is confirmed and new recovery strategies are formulated. The interpersonal interaction is the framework, while the result of the interaction—the gradually extended narrative—stabilizes the platform as a supporting element. The platform is co-created by health care professionals as they strive to understand the person’s life narrative. In the caring encounter, verbal language, body language and intuition are used to try to understand the person. “On the one hand, I’m very empathic. I check in. Then it involves asking: what does your everyday life look like? What do you like to do? What’s important to you? What do you need—both in terms of this conversation, but also a question you ask yourself daily, what do I need right now? Do you touch base in that way sometimes?” (1) Listening and intuitively feeling the person captures different expressions of the stress-related disorder and their ability to recover. The discrepancy between what the person conveys, and what the health care professional understands, creates a space for a caring conversation. This space is gradually filled with content through questions and considerations regarding what the person expresses, while at the same time ensuring correct understanding. The interaction gives rise to further consideration of and questions about the person’s life. “You need to have the confidence that I’m still here. I am listening to you, I am affirming you, I ensure that I have understood you correctly so as not to pass judgment or imagine I understand something when you didn’t mean it that way at all. Without me touching base, as it were, in the conversation—is this what you mean when you say this, do you mean this, have I understood you correctly now? That touching base makes us a team.” (1) Switching between the encounter with the person and the narrative occurs simultaneously using a so-called “helicopter perspective” of everything explored and understood earlier during the care process. New information about the person’s life situation is contrasted with the narrative constantly being revised. This involves a shift between proximity to the person’s existential reflections and reviewing the person’s described context. Thus he or she is supported in considering imbalances in his/her life situation related to a stress-related disorder, in understanding how different parts are connected and in a search for balance and recovery. On the interpersonal platform, the person is supported in understanding the different expressions of the stress-related disorder, their effect on the person’s life and the importance of recovery. The interpersonal platform provides opportunities for personal growth and change. The person returns to the primary health care centre in encounters with health care professionals who use a lingering and flexible approach to adapt their attitude to the person and their journey. The person is involved and receives gradually support in assuming greater responsibility for rediscovering his/her ability to initiate a recovery process and recover. “I think that time helps you to, like—I don’t like the word process, it’s an awful word really but for something like that to start, I mean if you’re talking about a person with exhaustion, an exhaustion disorder or other concerns that are really severe and stress-related, then you need to process this. Yes. You need to work around it in some way, and get closer to the idea that this is how it is, this is how I am, this is how I function, so what should I do to feel well?.” (11) The lingering and flexible approach is characterized by waiting for the person’s imminent insight and, further, waiting for his/her deeper understanding of how to manage unsustainable stress-related patterns. Beginning by recognizing and considering bodily signals and personal boundaries, as a way start to changing such patterns and developing and maintaining a necessary recovery process, changes life in a way that naturally results in recovery. This takes place in a changing process over time, and the person is encouraged to continue even after the care process has ended. “You can’t expect to get well in a month. It may not take an eternity, but it might take a while, and that’s something that I prepare patients for. We have to work together so that you get well, so that you get out of this. And it will take time, it will be demanding, but you won’t need to be on sick leave for years if you can tackle stress management and change your attitude towards different stressors over the course of several years until you end up, until you have everything ….in place.” (17) This process involves various aspects of the person’s life and is both relational and individual, personal and professional. In primary health care, supporting recovery must be a “safe haven” to which the person can return over time, to reflect, learn and rehearse, and then take on daily challenges with new knowledge and insight. The goal is that the recovery process gradually become more dominant in the person’s life. Supporting recovery with a guiding approach aims to help the person understand the difference between a state of stress and a state of recovery. The guiding approach means supporting the person in finding “clues” to a sustainable recovery process over time. The person receives guidance in using his/her senses, slowing down the body’s pace, anchoring themselves in their bodies and thus listening to themselves and considering his/her need for recovery. To strengthen these “clues” leading to a sustainable balance, care is based on providing verbally, visually or physically descriptive examples of the body’s need for balance. This may involve verbal metaphors, illustrative images, body awareness or encouraging a focus on sensing one’s surroundings, e.g., in nature. “Then we try … we try together, of course, with the patient. It’s better if the patient comes up with things that they want to change, but otherwise you have to provide help and maybe indicate some things you see are not sustainable, you can make suggestions.” (7) This guiding approach receives additional dimensions when provided for several persons with similar experiences who meet in a group. Supporting recovery in these group contexts means to providing an interconnecting link and being responsible for the structure and content of the encounters. Sharing with others reinforces the effects of the guiding approach and provides synergistic effects, such as recognizing oneself in others, feeling that one is not alone in one’s troubles and feeling understood. “The group is tremendously beneficial for all its members, they contribute a lot that you miss when you meet individually, so group treatment is ….it’s not possible to do it as well individually, I think.” (7) Another guiding approach is the involvement of significant others. This is done by providing information to increase understanding of the person’s situation and their needs in, e.g., family life. Significant others are involved as a support, either primarily in care through direct contact, or secondarily through supporting the person to communicate with his/her relatives. This study is, to our knowledge, the first that has focused on the phenomenon of supporting recovery in persons with stress-related disorders in primary health care. The results show that the health care professionals experienced supporting recovery as a complex process requiring a tailored approach where relational aspects, such as a caring alliance and collaboration, are foundations of support, regardless of profession. The results also show the importance of meeting the person where they are, starting from their narratives about their life and, from there, shaping an interpersonal platform where reflection and learning are crucial. With this platform, the health care professional provides space for existential reflection regarding the person’s life, and the health care professional can have a guiding approach to supporting the person’s rediscovery and consideration of their own needs. Providing professional support to these persons means being lingering and flexible in following the person’s journey through entering recovery process, achieving sustainable recovery and promoting life balance and health. The results of this study show that health care professionals express a caring alliance as a- crucial factor for the tailored support described in our result. This alliance was needed to make the person feel safe to express themselves, reflect on and deal with concerns during the primary health care encounter, between health care visits and after completion of care. According to Topor et al. ( ), a reciprocal relationship with a health care professional is important for recovery. Specific treatment methods are of less importance than the all-important working alliance with the professional. Our results also showed that the interpersonal nature of care is evident, and it was described as an interpersonal platform shifting between the person’s narrative and the health care professional’s attempts to understand that narrative. The efforts of health care professionals to understand correspond with the description by Todres et al. ( ) of important insights when trying to understand the “insiderness” of another person. They point out that the process of “reaching towards” understanding another person is more important than understanding exactly. The findings in our study describe learning as central for supporting recovery in persons with stress-related disorders. Health care professionals encourage the persons’ insights about personal identity and life situation through reflection and encourage them to expand their narratives about their lifeworld. The importance of learning and insight have been found in previous research. Andersson et al. ( ) found, e.g., how a health care professional’s changed approach—from informing about disease, illness and treatment to supporting a person to a learning about their entire health situation—permits change and reprioritisation. Another example is the group intervention ReDO-10, adapted to primary health care, where learning about life was found important, based on learning through an activity perspective (Fox et al., ; Olsson et al., ). Further, the findings of our study describe an overarching interprofessional tailored caring approach that unites beyond theories and methods and may be crucial in supporting recovery on the interpersonal platform. The health care professionals described the importance of communication, attentive listening, taking the person and their feelings seriously, a non-judgemental attitude, offering continuity and safety, being open to humour and sharing personal examples. Topor et al. ( ) describe such approaches in a professional relationship in a recovery-orientated practice as “small things” which are often overlooked but which play an important role for the person to improve their sense of self. These small things can be expressed in words, gestures or actions by the professional but require a reciprocity with the person. The findings of this study further stress the importance of “a lingering and flexible approach” by primary health care professionals towards persons with stress-related disorders. This approach creates space for reflection and awaiting the person’s imminent insight regarding their ability to recover and, further, supports a gradually increasing insight into how life should be handled and embraced to achieve sustainable balance and recovery. Different studies (Alsén et al., ; Arman et al., ; Jingrot & Rosberg, ) describe such crucial turning points or perspective shifts in persons with stress-related disorders as a “crossroad”, where they can choose to engage in a recovery process and find a way forward. Onken et al. ( ) gives credence to this means of supporting recovery, as described in our results, by explaining the notion of change that is incorporated in different important elements of recovery, both individual and interactional/societal. Our findings show that recovery can be supported by adapting “the guiding approach” with various instructive or creative tools and existential reflections including metaphors, experiences in nature and reflections in group settings. Creative and nature-based initiatives have also been found beneficial in other studies vis-à-vis stress-related disorders. Gunnarsson et al. ( ) found, e.g., that taking photographs of situations, places or settings linked to well-being, was a good starting point for a conversation about life and health with persons with stress-related disorders. Krantz et al. ( ) found in their autobiographical study that spending time in nature thinking and reflecting undisturbed, was an important part of the recovery process. Hörberg et al. ( ) describe “unconditional beingness” as a crucial existential space for well-being and recovery when living with stress-related disorder. This space can be achieved, e.g., in nature, painting or in undemanding situations together with others where you can be yourself. It seems important that health care professionals, when suitable, provide for such “unconditional beingness” in life. In the findings of our study, the professionals describe guiding persons with stress-related disorder with clues to existential reflections about life as a way to support changing unhealthy patterns. Arman et al. ( ) finds that the caring act can benefit from health care professionals supporting existential reflection to awaken longing and creativity. This indicate that persons with stress-related disorders may benefit from existential elements of care to recover. Care leaders are encouraged to initiate and support existential aspects of care together with health care professionals in primary health care. Further research is needed to understand the importance and effects of existential efforts in primary health care. The results of our study show that, to support recovery, it is important that health care professionals find the persons wherever they may be in their journey with stress related disorders, and start from their own narrative about their life situation to support recovery. This result is contrary to study of persons with burnout who felt mistrust, misunderstood and rejected in health care encounters, by Engebretsen and Bjorbaekmo ( ). The differences with these results may result from person-centred care being overshadowed by overly result-oriented, time-limited and depersonalized care, as indicated in a study with general practitioners (Derksen et al., ). Furthermore, previous research based on the person’s own perspectives highlights the importance of humanistic needs where the persons whole lifeworld is considered (Alsén et al., , ; Arman et al., ; Engebretsen & Bjorbaekmo, ; Jingrot & Rosberg, ). This speaks to the importance of a more person-centred approach, where the person and his/her narrative are in focus, which also seems important in our findings from a health care professional perspective. The results of our study show that supporting recovery involves an overall attitude similar to the recovery-focused approach described in the literature (Onken et al., ; Topor et al., ; van Weeghel et al., ). Buchanan-Barker and Barker ( ) describe such a recovery-focused approach as different from the more accepted medical and diagnostic approach presently dominant in health care. Recovery is not the same as a cure or a return to a state prior to illness with focus on the person’s symptoms. It is a complex and ongoing healing process, or “journey,” with a holistic view of the person (Jacob, ; Onken et al., ; Tew et al., ). It’s reasonable to assume that the recovery-focused approach found in our study might be unconscious or assumed. Raising awareness of the tacit knowledge they possess, and consciously using this knowledge in their work with persons with stress-related disorders, might beneficially support recovery. Methodological considerations Due to the COVID-19 pandemic, the strategies for recruitment and data collection in this study had to be adjusted. Major re-prioritizations made recruitment directly from the primary health care centres difficult. Increased restrictions in society made it impossible to conduct interviews in person, as planned. We therefore conducted some of the meetings digitally. A study by Archibald et al. ( ) found that researchers and participants describe digital meetings as a convenient platform for collecting qualitative interview data, due to its ease of use, security and interactivity. Interviewing on a digital platform is a good complement to in-person meetings and gives more qualitative dimensions than telephone meetings. However, the in-person meetings of this study provided interpersonal dimensions that may be missed in digital interviews, especially when conducting lifeworld interviews. K. Dahlberg et al. ( ) conveys that an atmosphere of presence in an interview situation increases the possibility of a necessary openness that is important to grasp in-depth data. As a researcher, this means simultaneously being present in a “bridled” way with the phenomenon and the participant. It is likely easier, in an in-person meeting, to notice different verbal and nonverbal expressions when participants describe lived experiences related to the phenomenon. The necessary adoption of digital meetings by this study is therefore a limitation. On the other hand, digital meetings permitted a greater variation geographically, likely to be important depending on the division of Swedish primary health care into heterogeneous regions of varying structure (National Board of Health and Welfare, ). Furthermore, the number of the interviews, their length and the variation in the sample resulted in rich data, which we value as important for the validity and reliability of the results. This study followed the RLR methodological principles of openness, flexibility and bridling throughout the research process (K. Dahlberg et al., ). An open, flexible and bridled attitude permeated both the interview and analysis process by being delaying and a little slow, zooming in and zooming out the focus on understanding the phenomenon and in a flexibly and reflectively way grasping different nuances of the data. This attitude was intended to avoid taking any unknowns for granted. To maintain objectivity throughout interviews and analysis, van Wijngaarden et al. ( ) suggest using such a phenomenological attitude. Validity in phenomenological research is connected with meaning. The underlaying meaning of the participants’ lifeworld descriptions differs from the participants’ statements, i.e., the content (van Wijngaarden et al., ). The first author has searched the transcribed lifeworld interviews for meanings related to the studied phenomenon. We have, within an expanded research group, reflected over and discussed the meaning of these units and the patterns of meaning that emerged. We have also discussed, whenever possible, how our individual preunderstanding affects our understanding of the phenomenon and the analysis of the text, in order to enhance objectivity in the study (van Wijngaarden et al., ). The transferability of the results from this study, the structure of meaning in the essence and the constituents, could probably be generalized to other care contexts with similar interprofessional teams, such as psychiatric outpatient care. Due to the COVID-19 pandemic, the strategies for recruitment and data collection in this study had to be adjusted. Major re-prioritizations made recruitment directly from the primary health care centres difficult. Increased restrictions in society made it impossible to conduct interviews in person, as planned. We therefore conducted some of the meetings digitally. A study by Archibald et al. ( ) found that researchers and participants describe digital meetings as a convenient platform for collecting qualitative interview data, due to its ease of use, security and interactivity. Interviewing on a digital platform is a good complement to in-person meetings and gives more qualitative dimensions than telephone meetings. However, the in-person meetings of this study provided interpersonal dimensions that may be missed in digital interviews, especially when conducting lifeworld interviews. K. Dahlberg et al. ( ) conveys that an atmosphere of presence in an interview situation increases the possibility of a necessary openness that is important to grasp in-depth data. As a researcher, this means simultaneously being present in a “bridled” way with the phenomenon and the participant. It is likely easier, in an in-person meeting, to notice different verbal and nonverbal expressions when participants describe lived experiences related to the phenomenon. The necessary adoption of digital meetings by this study is therefore a limitation. On the other hand, digital meetings permitted a greater variation geographically, likely to be important depending on the division of Swedish primary health care into heterogeneous regions of varying structure (National Board of Health and Welfare, ). Furthermore, the number of the interviews, their length and the variation in the sample resulted in rich data, which we value as important for the validity and reliability of the results. This study followed the RLR methodological principles of openness, flexibility and bridling throughout the research process (K. Dahlberg et al., ). An open, flexible and bridled attitude permeated both the interview and analysis process by being delaying and a little slow, zooming in and zooming out the focus on understanding the phenomenon and in a flexibly and reflectively way grasping different nuances of the data. This attitude was intended to avoid taking any unknowns for granted. To maintain objectivity throughout interviews and analysis, van Wijngaarden et al. ( ) suggest using such a phenomenological attitude. Validity in phenomenological research is connected with meaning. The underlaying meaning of the participants’ lifeworld descriptions differs from the participants’ statements, i.e., the content (van Wijngaarden et al., ). The first author has searched the transcribed lifeworld interviews for meanings related to the studied phenomenon. We have, within an expanded research group, reflected over and discussed the meaning of these units and the patterns of meaning that emerged. We have also discussed, whenever possible, how our individual preunderstanding affects our understanding of the phenomenon and the analysis of the text, in order to enhance objectivity in the study (van Wijngaarden et al., ). The transferability of the results from this study, the structure of meaning in the essence and the constituents, could probably be generalized to other care contexts with similar interprofessional teams, such as psychiatric outpatient care. We conclude that care to support recovery may benefit from being genuinely adapted to the specific person and his/her lifeworld. Space should exist for existential questions and learning based on the person’s own narrative, and the alliance and the interpersonal relational aspect are considered as crucial. Regarding the tailored approach of health care professionals described in this study, one important finding is the necessity to anticipate and capture the person’s imminent insight into their own capacity for recovery. The healthcare professionals must provide space and support in finding strategies and meaning in life and recovery. Implementing more recovery-focused models in primary health care could help to raise awareness regarding effective person-centred strategies already applied by health care professionals. In order to support recovery, specific interventions and rehabilitation strategies may be adapted to the person’s individual journey to recovery from a stress-related disorder, and not the other way around. We recommend that elements of existential care be highlighted and given greater space in research and care.
A survey of Canadian adult rheumatologists’ knowledge, comfort level, and barriers in assessing psychosocial needs of young adults with rheumatic diseases	a25d441b-f247-4dd2-aac3-1121d83cde20	10171159	Internal Medicine[mh]	Chronic rheumatic diseases diagnosed in childhood often require long-term medical management, and the transition period from pediatric to adult rheumatology care is an important and often vulnerable time in patients’ lives . Adult care is more patient focused than family focused and requires more independence. Inadequate preparation for adult care combined with the vulnerable state patients can lead to gaps in care and loss to follow-up . Existing resources for transition include the American College of Rheumatology (ACR) pediatric transition toolkit and the European Alliance of Associations for Rheumatology (EULAR) Standards and Recommendations for the Transitional Care of Young People with Juvenile-Onset Rheumatic Diseases . However, in a recent publication, only 31% of both adult and pediatric health care providers reported using these resources to help with the transition process . For youth, gaps in care and loss to follow-up may occur for a multitude of reasons, many of which may be personal including psychosocial concerns, and important life milestones as well as disease relapses, medication side effects and difficulties accessing care . Disruptions in medical care may also be related to their initial experiences with their adult practitioner . There have been limited studies reporting the perspectives of adult rheumatologists in transition . A recent survey of adult rheumatologists, based out of the United States, self-identified as having inadequate training in transition issues of young adults, specifically psychosocial concerns . This contrasts with pediatric healthcare providers who have familiarity working with adolescent psychosocial concerns and behaviors . Adult rheumatologists reported a lack of comfort in managing patients with pediatric-onset disease and endorsed less familiarity with transition guidelines compared to pediatric healthcare providers . The objectives for this Canadian survey of adult rheumatologists and adult rheumatology trainees were to determine the comfort level and barriers to caring for young adult patients with pediatric-onset rheumatic disease after transfer to adult care. In addition, the management of psychosocial needs in the adolescent and young adult population including comfort level and frequency of discussions around these topics were assessed. A combination of literature review, building on prior work from the United States-based study by Zisman et al. and information provided by a needs assessment conducted by the Canadian Rheumatology Association (CRA), formed the basis of our survey . The survey incorporated entrustable professional activities (EPAs) designed by the Royal College of Canada for Rheumatology trainees which are specific tasks that trainees can be trusted to perform independently in different contexts . Specifically, Core EPA #12 P: Supporting adolescents/young adults with rheumatologic disease in the transition from the pediatric to adult care setting and its individual milestones were used for each question. This EPA focuses on developmental readiness and the risks that can occur as patients develop autonomy regarding their health . The initial survey was drafted by two pediatric rheumatologists, a pediatric resident and a clinical researcher. It was then reviewed by two adult rheumatologists whose revisions were incorporated prior to sharing it with four adult rheumatologist members of the CRA Transition Working Group for final edits, comments, and feedback. The Consensus-based Checklist for reporting of Survey Studies (CROSS) and EQUATOR guidelines were followed during survey preparation and reporting . The survey collected information on demographics, current practice, comfort in managing psychosocial aspects of transition care and perceived barriers to providing care to this population. Responses to questions were graded on a five-point Likert scale. The project was approved by the Hamilton Integrated Research Ethics Board (Project #13568) on September 16th, 2021. The survey was distributed by email via the CRA using the Alchemer platform ( www.alchemer.com ) on November 3rd, 2021 to adult rheumatologists and trainees who were CRA members. Pediatric rheumatologists and trainees were not included. Data collection occurred until March 18 th , 2022 with monthly reminders sent by the CRA. Statistical analysis Descriptive statistics (frequencies and proportions) summarizing survey responses were determined using SPSS v.28 (IBM SPSS Statistics, USA). Responses from the five-point Likert scale were then translated to a dichotomous outcome to better analyze data due to the relatively small sample size in the study. For example, those who answered questions regarding comfort level with agree or strongly agree were designated as comfortable and those who answered strongly disagree, disagree, and neutral were designated not comfortable. Descriptive statistics (frequencies and proportions) summarizing survey responses were determined using SPSS v.28 (IBM SPSS Statistics, USA). Responses from the five-point Likert scale were then translated to a dichotomous outcome to better analyze data due to the relatively small sample size in the study. For example, those who answered questions regarding comfort level with agree or strongly agree were designated as comfortable and those who answered strongly disagree, disagree, and neutral were designated not comfortable. Demographics The survey was sent to 490 rheumatologists (451 in English and 39 in French) and 15.2% ( n = 69) completed the survey. More than half were female with ≥ 10 years of experience in practice and had trained in Canada. Most respondents were practicing in Ontario and had either a university-based practice or a combined university- and community-based practice (Table ). Approximately half reported involvement in providing transition care. Fifty-one rheumatologists (74%) reported that they had no formal training in transition care (Table ). Of those who had received training, it was primarily in fellowship and by continuing medical education at their own institution. Current transition practices Providers most commonly saw patients being transferred to them between 16 and 18 years of age and were referred by pediatric rheumatologists. The ideal age for transfer was felt to be between 16 and 20 years old. About 45% had a multidisciplinary team to help support transitioned patients which included nursing, physiotherapy, social work, and occupational therapy. On average, 40% had resources in their office for patients such as an orientation to the adult rheumatologists’ practice, self-management skills assessments and tools (pamphlets, phone apps, etc.). More than half of the participants reported no formal transition policy at their institution with one-third reporting an informal policy was being followed. Approximately 40% of individuals felt comfortable discussing psychosocial concerns such as gender identity, sexuality, contraception, drug and alcohol use, vaping, depression, and anxiety (Table ); however, about half reported discussing vaping and gender identity not at all or rarely. Barriers to transition More than two-thirds of respondents reported insufficient skills to address transition-related concerns while only 13% reported having sufficient resources and personnel to adequately address these concerns. Greater than 75% reported lack of time and renumeration for providing transition care. The most frequently reported barriers to providing optimal care to patients transitioning to adult care were: (1) lack of primary care providers, (2) lack of allied health support, (3) inadequate training in caring for this age group, and (4) not being able to be reimbursed or having time listed to provide transition care. Despite this, the majority (75%) expressed an interest in learning more about providing transition care to adolescents. Three quarters of respondents believed that the family physician should be the provider to discuss psychosocial concerns such as mental health, body image, drug use, gender identity and sexuality, and educational goals (Table ). About 40% felt that tele-rheumatology can be a helpful resource in transition care. The survey was sent to 490 rheumatologists (451 in English and 39 in French) and 15.2% ( n = 69) completed the survey. More than half were female with ≥ 10 years of experience in practice and had trained in Canada. Most respondents were practicing in Ontario and had either a university-based practice or a combined university- and community-based practice (Table ). Approximately half reported involvement in providing transition care. Fifty-one rheumatologists (74%) reported that they had no formal training in transition care (Table ). Of those who had received training, it was primarily in fellowship and by continuing medical education at their own institution. Providers most commonly saw patients being transferred to them between 16 and 18 years of age and were referred by pediatric rheumatologists. The ideal age for transfer was felt to be between 16 and 20 years old. About 45% had a multidisciplinary team to help support transitioned patients which included nursing, physiotherapy, social work, and occupational therapy. On average, 40% had resources in their office for patients such as an orientation to the adult rheumatologists’ practice, self-management skills assessments and tools (pamphlets, phone apps, etc.). More than half of the participants reported no formal transition policy at their institution with one-third reporting an informal policy was being followed. Approximately 40% of individuals felt comfortable discussing psychosocial concerns such as gender identity, sexuality, contraception, drug and alcohol use, vaping, depression, and anxiety (Table ); however, about half reported discussing vaping and gender identity not at all or rarely. More than two-thirds of respondents reported insufficient skills to address transition-related concerns while only 13% reported having sufficient resources and personnel to adequately address these concerns. Greater than 75% reported lack of time and renumeration for providing transition care. The most frequently reported barriers to providing optimal care to patients transitioning to adult care were: (1) lack of primary care providers, (2) lack of allied health support, (3) inadequate training in caring for this age group, and (4) not being able to be reimbursed or having time listed to provide transition care. Despite this, the majority (75%) expressed an interest in learning more about providing transition care to adolescents. Three quarters of respondents believed that the family physician should be the provider to discuss psychosocial concerns such as mental health, body image, drug use, gender identity and sexuality, and educational goals (Table ). About 40% felt that tele-rheumatology can be a helpful resource in transition care. The key findings from this survey highlight that most adult rheumatologists lack training in caring for the adolescent population and express a lack of supportive medical care, both family physicians and allied health, to help them care for this population. Further, while they may feel comfortable—specifically discussing psychosocial concerns—the majority believe that it is the family physician’s responsibility to do so. Compared to previous studies, there were similarities regarding the overall transition process and the barriers to a successful transition. For example, approximately one-third of respondents in our survey had no formal transition policy at their institution but an informal procedure that is followed, which is comparable to the findings in the Childhood Arthritis and Rheumatology Research Alliance (CARRA) survey completed in 2010 and the survey in 2014 by Zisman et al. . In addition, the most frequently encountered barriers such as lack of training in caring for this age group, allied health providers, primary care providers, not being able to be renumerated appropriately and time constraints were common themes that have been found in previous studies . Almost half of adult rheumatologists reported feeling comfortable discussing psychosocial concerns with patients which is higher than previous studies (Table ) . However, this was self-reported and patient perspectives were not sought. Topics such as alcohol and tobacco use, contraception, and fertility were reported to be discussed either most or all the time by the majority of rheumatologists. However, 45% reported discussing vaping either “not at all” or “rarely” despite the same number feeling comfortable discussing the topic. Given the increase in incidence and the detrimental effects on lung function in youth with chronic conditions, this topic should be addressed consistently in this population . Despite half of the participants reporting comfort in discussing psychosocial topics, more than 75% believed that the family physician should be the provider to discuss these issues (Table ). Many of these concerns are related to the patient’s rheumatic disease and/or medications and side effects leading to difficulties with body image, self-esteem and, ultimately, to mental health concerns . This can place patients in a difficult position as both health care providers feel as though it is the other’s responsibility to address these concerns. Unfortunately, many patients do not have a family doctor due to an overall shortage of primary care physicians in Canada or may have difficulty accessing their primary care provider . In addition, many adolescents with chronic rheumatic disease are otherwise generally healthy and medically stable, and their rheumatologist may be the only physician they see regularly. Ideally, conversations about psychosocial issues would happen jointly among health care providers and the primary care provider should be included in the transition process. The majority of transition research is targeted toward the pediatric healthcare provider’s care and perspective. Following a patient from childhood into adulthood establishes strong relationships between patients, their families, and the pediatric care team ; however, adult healthcare providers will ultimately spend a far longer time with patients over their life course. It is, therefore, essential to gain an understanding of their current practices and perceptions of gaps in care to ensure that patients receive optimal care. This survey adds to the body of research and provides a Canadian perspective on this topic. This survey design is unique in that it incorporated the competency by design framework. As more programs shift toward a competency by design curriculum, these milestones are now mandatory for trainees to complete their training . The questions are not only applicable to rheumatology but can be used to assess transition in other pediatric sub-specialities with chronic conditions. It also looked at multiple aspects of psychosocial care, which have not been previously evaluated, such as vaping and gender identity. Despite reminder emails, our study was limited by the relatively low response rate (15.2%) and small numbers of individuals in certain demographic groups (e.g., physicians 46–55 years old, practicing 6–15 years) making subgroup comparisons infeasible. In addition, many rheumatologists do not provide transition care and may not have completed the survey as a result. It is also probable that the survey was completed by adult rheumatologists who were already interested in transition care, leading to response bias and a limited perspective, thereby affecting the generalizability of results to the general adult rheumatology community. Although the survey was released in 2021, the impact of the COVID-19 pandemic upon transitional care and psychosocial concerns of patients was not included in the survey. These results demonstrate that adult rheumatologists would benefit from more training, multidisciplinary support, resources, and appropriate renumeration to optimize the provision of transition care to young adult patients. Knowledge and treatment of psychosocial factors related to the young adult population should be emphasized during training. Educational initiatives such as workshops and peer teaching sessions with patient engagement should be co-created to improve the transition knowledge and skills for adult health care providers. Tele-rheumatology is a resource that can be valuable for this population as it can be used when patients are away for university/college for follow-up appointments and make it easier to have joint appointments with health care providers. A transition navigator or coach may be a valuable resource to bridge the divide between pediatric and adult care, provide psychosocial support to young adult patients, and provide continuity to patients during this challenging time . Transition navigators have been found to improve adherence to medication, attendance at clinic, and reduce acute complications of disease in other chronic disease populations . Future research should be aimed toward assessing the impact of educational initiatives in subspecialty training of rheumatologists in their comfort supporting the young adult patient population with pediatric-onset disease. Furthermore, the effectiveness of implementing the above resources such as transition navigators upon the transition process from the patient, primary care provider and subspecialist perspective should also be obtained. Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 217 KB)
Learnable latent embeddings for joint behavioural and neural analysis	304d1a21-1045-475a-af36-bfb2c2062a24	10172131	Physiology[mh]	A central quest in neuroscience is the neural origin of behaviour , . Nevertheless, we are still limited in both the number of neurons and length of time we can record from behaving animals in a session. Therefore, we need new methods that can combine data across animals and sessions with minimal assumptions, thereby generating interpretable neural embedding spaces , . Current tools for representation learning are either linear or, if nonlinear, typically rely on generative models and they do not yield consistent embeddings across animals (or repeated runs of the algorithm). Here, we combine recent advances in nonlinear disentangled representation learning and self-supervised learning to develop a new dimensionality reduction method that can be applied jointly to behavioural and neural recordings to show meaningful lower-dimensional neural population dynamics – . From data visualization (clustering) to discovery of latent spaces that explain neural variance, dimensionality reduction of behaviour or neural data has been impactful in neuroscience. For example, complex three-dimensional (3D) forelimb reaching can be reduced to between only eight and twelve dimensions , , and low-dimensional embeddings show some robust aspects of movements (for example, principal component analysis (PCA)-based manifolds in which the neural state space can easily be constrained and is stable across time – ). Linear methods such as PCA are often used to increase interpretability, but this comes at the cost of performance . Uniform manifold approximation and projection (UMAP) and t -distributed stochastic neighbour embedding ( t -SNE) are excellent nonlinear methods but they lack the ability to explicitly use time information, which is always available in neural recordings, and they are not as directly interpretable as PCA. Nonlinear methods are desirable for use in high-performance decoding but often lack identifiability—the desirable property that true model parameters can be determined, up to a known indeterminacy , . This is critical because it ensures that the learned representations are uniquely determined and thus facilitates consistency across animals and/or sessions. There is recent evidence that label-guided variational auto-encoders (VAEs) could improve interpretability , , . Namely, by using behavioural variables, such algorithms can learn to project future behaviour onto past neural activity , or explicitly to use label priors to shape the embedding . However, these methods still have restrictive explicit assumptions on the underlying statistics of the data and they do not guarantee consistent neural embeddings across animals , , , which limits both their generalizability and interpretability (and thereby affects accurate decoding across animals). We address these open challenges with CEBRA, a new self-supervised learning algorithm for obtaining interpretable, consistent embeddings of high-dimensional recordings using auxiliary variables. Our method combines ideas from nonlinear independent component analysis (ICA) with contrastive learning , – , a powerful self-supervised learning scheme, to generate latent embeddings conditioned on behaviour (auxiliary variables) and/or time. CEBRA uses a new data-sampling scheme to train a neural network encoder with a contrastive optimization objective to shape the embedding space. It can also generate embeddings across multiple subjects and cope with distribution shifts among experimental sessions, subjects and recording modalities. Importantly, our method relies on neither data augmentation (as does SimCLR ) nor a specific generative model, which would limit its range of use. We propose a framework for jointly trained latent embeddings. CEBRA leverages user-defined labels (supervised, hypothesis-driven) or time-only labels (self-supervised, discovery-driven; Fig. and Supplementary Note ) to obtain consistent embeddings of neural activity that can be used for both visualization of data and downstream tasks such as decoding. Specifically, it is an instantiation of nonlinear ICA based on contrastive learning . Contrastive learning is a technique that leverages contrasting samples (positive and negative) against each other to find attributes in common and those that separate them. We can use discrete and continuous variables and/or time to shape the distribution of positive and negative pairs, and then use a nonlinear encoder (here, a convolutional neural network but can be another type of model) trained with a new contrastive learning objective. The encoder features form a low-dimensional embedding of the data (Fig. ). Generation of consistent embeddings is highly desirable and closely linked to identifiability in nonlinear ICA , . Theoretical work has shown that the use of contrastive learning with auxiliary variables is identifiable for bijective neural networks using a noise contrastive estimation (NCE) loss , and that with an InfoNCE loss this bijectivity assumption can sometimes be removed (see also our theoretical generalization in Supplementary Note ). InfoNCE minimization can be viewed as a classification problem such that, given a reference sample, the correct positive sample needs to be distinguished from multiple negative samples. CEBRA optimizes neural networks f , f ′ that map neural activity to an embedding space of a defined dimension (Fig. ). Pairs of data ( x , y ) are mapped to this embedding space and then compared with a similarity measure ϕ (⋅,⋅). Abbreviating this process with [12pt]{minimal} $$ / $$ ψ x , y = φ f x , f ′ y / τ and a temperature hyperparameter, τ , the full criterion for optimization is [12pt]{minimal} $$}}_{} p,\,{{}}_{+} p\\ {{}}_{1}, ,{{}}_{n} q+ _{i=1}^{n}{e}^{ ({},{{}}_{i})}],$$ E x ∼ p ( x ) , y + ∼ p ( y \| x ) y 1 , … , y n ∼ q ( y \| x ) [ − ψ ( x , y + ) + log ∑ i = 1 n e ψ ( x , y i ) ] , which, depending on the dataset size, can be optimized with algorithms for either batch or stochastic gradient descent. In contrast to other contrastive learning algorithms, the positive-pair distribution p and negative-pair distribution q can be systematically designed and allow the use of time, behaviour and other auxiliary information to shape the geometry of the embedding space. If only discrete labels are used, this training scheme is conceptually similar to supervised contrastive learning . CEBRA can leverage continuous behavioural (kinematics, actions) as well as other discrete variables (trial ID, rewards, brain-area ID and so on). Additionally, user-defined information about desired invariances in the embedding is used (across animals, sessions and so on), allowing for flexibility in data analysis. We group this information into task-irrelevant and -relevant variables, and these can be leveraged in different contexts. For example, to investigate trial-to-trial variability or learning across trials, information such as a trial ID would be considered a task-relevant variable. On the contrary, if we aim to build a robust brain machine interface that should be invariant to such short-term changes, we would include trial information as a task-irrelevant variable and obtain an embedding space that no longer carries this information. Crucially, this allows inference of latent embeddings without explicit modelling of the data-generating process (as done in pi-VAE and latent factor analysis via dynamical systems (LFADS) ). Omitting the generative model and replacing it by a contrastive learning algorithm facilitates broader applicability without modifications. We first demonstrate that CEBRA significantly outperforms t -SNE, UMAP, automatic LFADS (autoLFADS) and pi-VAE (the latter was shown to outperform PCA, LFADS, demixed PCA and PfLDS (Poisson feed-forward neural network linear dynamical system) on some tasks) in the reconstruction of ground truth synthetic data (one-way analysis of variance (ANOVA), F (4, 495) = 251, P = 1.12 × 10 −117 ; Fig. and Extended Data Fig. ). We then turned to a hippocampus dataset that was used to benchmark neural embedding algorithms , (Extended Data Fig. and Supplementary Note ). Of note, we first significantly improved pi-VAE by the addition of a convolutional neural network (conv-pi-VAE), thereby allowing this model to leverage multiple time steps, and used this for further benchmarking (Extended Data Fig. ). To test our methods, we first considered the correlation of the resulting embedding space across subjects (does it produce similar latent spaces?), and the correlation across repeated runs of the algorithm (how consistent are the results?). We found that CEBRA significantly outperformed other algorithms in the production of consistent embeddings, and it produced visually informative embeddings (Fig. and Extended Data Figs. and ; for each embedding a single point represents the neural population activity over a specified time bin). When using CEBRA-Behaviour, the consistency of the resulting embedding space across subjects is significantly higher compared with autoLFADS and conv-pi-VAE, with or without test-time labels (one-way ANOVA F (25.4) P = 1.92 × 10 −16 ; Supplementary Table and Fig. ). Qualitatively, it can be appreciated that both CEBRA-Behaviour and -Time have similar output embeddings whereas the latents from conv-pi-VAE, either with label priors or without labels, are not consistent (CEBRA does not need test-time labels), suggesting that the label prior strongly shapes the output embedding structure of conv-pi-VAE. We also considered correlations across repeated runs of the algorithm, and found higher consistency and lower variability with CEBRA (Extended Data Fig. ). Among the advantages of CEBRA are its collective flexibility, limited assumptions, and ability to test hypotheses. For the hippocampus, one can hypothesize that these neurons represent space , and therefore the behavioural label could be either position or velocity (Fig. ). In addition, considering structure in only the behavioural data (with CEBRA) could help refine which behavioural labels to use jointly with neural data (Fig. ). Conversely, for the sake of argument, we could have an alternative hypothesis: that the hippocampus does not map space, but simply maps the direction of travel or some other feature. Using the same model but hypothesis free, and using time for selection of contrastive pairs, is also possible, and/or a hybrid thereof (Fig. ). We trained hypothesis-guided (supervised), time-only (self-supervised) and hybrid models across a range of input dimensions and embedded the neural latents into a 3D space for visualization. Qualitatively, we find that the position-based model produces a highly smooth embedding that shows the position of the animal—namely, there is a continuous ‘loop’ of latent dynamics around the track (Fig. ). This is consistent with what is known about the hippocampus and shows the topology of the linear track with direction specificity whereas shuffling the labels, which breaks the correlation between neural activity and direction and position, produces an unstructured embedding (Fig. ). CEBRA-Time produces an embedding that more closely resembles that of position (Fig. ). This also suggests that time contrastive learning captured the major latent space structure, independent of any label input, reinforcing the idea that CEBRA can serve both discovery- and hypothesis-driven questions (and that running both variants can be informative). The hybrid design, whose goal is to disentangle the latent to subspaces that are relevant to the given behavioural and residual temporal variance and noise, showed a structured embedding space similar to behaviour (Fig. ). To quantify how CEBRA can disentangle which variable had the largest influence on embedding, we tested for encoding position, direction and combinations thereof (Fig. ). We find that position plus direction is the most informative label (Fig. and Extended Data Fig. ). This is evident both in the embedding and the value of the loss function on convergence, which serves as a ‘goodness of fit’ metric to select the best labels—that is, which label(s) produce the lowest loss at the same point in training (Extended Data Fig. ). Note that erroneous (shuffled) labels converge to considerably higher loss values. To measure performance, we consider how well we could decode behaviour from the embeddings. As an additional baseline we performed linear dimensionality reduction with PCA. We used a k -nearest-neighbour (kNN) decoder for position and direction and measured the reconstruction error. We find that CEBRA-Behaviour has significantly better decoding performance (Fig. and Supplementary Video ) compared with both pi-VAE and our conv-pi-VAE (one-way ANOVA, F = 131, P = 3.6 × 10 −24 ), and also CEBRA-Time compared with unsupervised methods (autoLFADS, t -SNE, UMAP and PCA; one-way ANOVA, F = 1,983, P = 6 × 10 −50 ; Supplementary Table ). Zhou and Wei reported a median absolute decoding error of 12 cm error whereas we achieved approximately 5 cm (Fig. ). CEBRA therefore allows for high-performance decoding and also ensures consistent embeddings. Although CEBRA can be trained across a range of dimensions, and models can be selected based on decoding, goodness of fit and consistency, we also sought to find a principled approach to verify the robustness of embeddings that might yield insight into neural computations , (Fig. ). We used algebraic topology to measure the persistent cohomology as a comparison in regard to whether learned latent spaces are equivalent. Although it is not required to project embeddings onto a sphere, this has the advantage that there are default Betti numbers (for a d -dimensional uniform embedding, [12pt]{minimal} $${H}^{0}=1,{H}^{1}=0, ,{H}^{d-1}=1$$ H 0 = 1 , H 1 = 0 , ⋯ , H d − 1 = 1 —that is, 1,0,1 for the two-sphere). We used the distance from the unity line (and threshold based on a computed null shuffled distribution in Births versus Deaths to compute Betti numbers; Extended Data Fig. ). Using CEBRA-Behaviour or -Time we find a ring topology (1,1,0; Fig. ), as one would expect from a linear track for place cells. We then computed the Eilenberg–MacLane coordinates for the identified cocycle (H 1 ) for each model , —this allowed us to map each time point to topology-preserving coordinates—and indeed we find that the ring topology for the CEBRA models matches space (position) across dimensions (Fig. and Extended Data Fig. ). Note that this topology differs from (1,0,1)—that is, Betti numbers for a uniformly covered sphere—which in our setting would indicate a random embedding as found by shuffling (Fig. ). CEBRA can also be used to jointly train across sessions and different animals, which can be highly advantageous when there is limited access to simultaneously recorded neurons or when looking for animal-invariant features in the neural data. We trained CEBRA across animals within each multi-animal dataset and find that this joint embedding allows for even more consistent embeddings across subjects (Extended Data Fig. ; one-sided, paired t -tests; Allen data: t = −5.80, P = 5.99 × 10 −5 ; hippocampus: t = −2.22, P = 0.024). Although consistency increased, it is not a priori clear that decoding from ‘pseudosubjects’ would be equally good because there could be session- or animal-specific information that is lost in pseudodecoding (because decoding is usually performed within the session). Alternatively, if this joint latent space was as high performance as the single subject, that would suggest that CEBRA is able to produce robust latent spaces across subjects. Indeed, we find no loss in decoding performance (Extended Data Fig. ). It is also possible to rapidly decode from a new session that is unseen during training, which is an attractive setting for brain machine interface deployment. We show that, by pretraining on a subset of the subjects, we can apply and rapidly adapt CEBRA-Behaviour on unseen data (that is, it runs at 50–100 steps s –1 , and positional decoding error already decreased by 10 cm after adapting the pretrained network for one step). Lastly, we can achieve a lower error more rapidly compared with training fully on the unseen individual (Extended Data Fig. ). Collectively, this shows that CEBRA can rapidly produce high-performance, consistent and robust latent spaces. We next consider an eight-direction ‘centre-out’ reaching task paired with electrophysiology recordings in primate somatosensory cortex (S1) (Fig. ). The monkey performed many active movements, and in a subset of trials experienced randomized bumps that caused passive limb movement. CEBRA produced highly informative visualizations of the data compared with other methods (Fig. ), and CEBRA-Behaviour can be used to test the encoding properties of S1. Using either position or time information showed embeddings with clear positional encoding (Fig. and Extended Data Fig. ). To test how directional information and active versus passive movements influence population dynamics in S1 (refs. – ), we trained embedding spaces with directional information and then either separated the trials into active and passive for training (Fig. ) or trained jointly and post hoc plotted separately (Fig. ). We find striking similarities suggesting that active versus passive strongly influences the neural latent space: the embeddings for active trials show a clear start and stop whereas for passive trials they show a continuous trajectory through the embedding, independently of how they are trained. This finding is confirmed in embeddings that used only the continuous position of the end effector as the behavioural label (Fig. ). Notably, direction is a less prominent feature (Fig. ) although they are entangled parameters in this task. As the position and active or passive trial type appear robust in the embeddings, we further explored the decodability of the embeddings. Both position and trial type were readily decodable from 8D+ embeddings with a kNN decoder trained on position only, but directional information was not as decodable (Fig. ). Here too, the loss function value is informative for goodness of fit during hypothesis testing (Extended Data Fig. ). Notably, we could recover the hand trajectory with R 2 = 88% (concatenated across 26 held-out test trials; Fig. ) using a 16D CEBRA-Behaviour model trained on position (Fig. ). For comparison, an L1 regression using all neurons achieved R 2 = 74% and 16D conv-pi-VAE achieved R 2 = 82%. We also tested CEBRA on an additional monkey dataset (mc-maze) presented in the Neural Latent Benchmark , in which it achieved state-of-the-art behaviour (velocity) decoding performance (Extended Data Fig. ). Although CEBRA is agnostic to the recording modality of neural data, do different modalities produce similar latent embeddings? Understanding the relationship of calcium signalling and electrophysiology is a debated topic, yet an underlying assumption is that they inherently represent related, yet not identical, information. Although there is a wealth of excellent tools aimed at inferring spike trains from calcium data, currently the pseudo- R 2 of algorithms on paired spiking and calcium data tops out at around 0.6 (ref. ). Nonetheless, it is clear that recording with either modality has led to similar global conclusions—for example, grid cells can be uncovered in spiking or calcium signals , , reward prediction errors can be found in dopamine neurons across species and recording modalities – , and visual cortex shows orientation tuning across species and modalities – . We aimed to formally study whether CEBRA could capture the same neural population dynamics either from spikes or calcium imaging. We utilized a dataset from the Allen Brain Observatory where mice passively watched three videos repeatedly. We focused on paired data from ten repeats of ‘Natural Movie 1’ where neural data were recorded with either Neuropixels (NP) probes or calcium imaging with a two-photon (2P) microscope (from separate mice) , . Note that, although the data we have considered thus far have goal-driven actions of the animals (such as running down a linear track or reaching for targets), this visual cortex dataset was collected during passive viewing (Fig. ). We used the video features as ‘behaviour’ labels by extracting high-level visual features from the video on a frame-by-frame basis with DINO, a powerful vision transformer model . These were then used to sample the neural data with feature-labels (Fig. ). Next, we used either Neuropixels or 2P data (each with multi-session training) to generate (from 8D to 128D) latent spaces from varying numbers of neurons recorded from primary visual cortex (V1) (Fig. ). Visualization of CEBRA-Behaviour showed trajectories that smoothly capture the video of either modality with an increasing number of neurons. This is reflected quantitatively in the consistency metric (Fig. ). Strikingly, CEBRA-Time efficiently captured the ten repeats of the video (Extended Data Fig. ), which was not captured by other methods. This result demonstrates that there is a highly consistent latent space independent of the recording method. Next, we stacked neurons from different mice and modalities and then sampled random subsets of V1 neurons to construct a pseudomouse. We did not find that joint training lowered consistency within modality (Extended Data Fig. ) and, overall, we found considerable improvement in consistency with joint training (Fig. ). Using CEBRA-Behaviour or -Time, we trained models on five higher visual areas and measured consistency with and without joint training, and within or across areas. Our results show that, with joint training, intra-area consistency is higher compared with other areas (Fig. ), suggesting that CEBRA is not removing biological differences across areas, which have known differences in decodability and feature representations , . Moreover, we tested within modality and find a similar effect for CEBRA-Behaviour and -Time within recording modality (Extended Data Fig. . We performed V1 decoding analysis using CEBRA models that are either joint-modality trained, single-modality trained or with a baseline population vector paired with a simple kNN or naive Bayes decoder. We aimed to determine whether we could decode, on a frame-by-frame basis, the natural video watched by the mice. We used the final video repeat as a held-out test set and nine repeats as the training set. We achieved greater than 95% decoding accuracy, which is significantly better than baseline decoding methods (naive Bayes or kNN) for Neuropixels recordings, and joint-training CEBRA outperformed Neuropixels-only CEBRA-based training (single frame: one-way ANOVA, F (3,197) = 5.88, P = 0.0007; Supplementary Tables – , Fig. and Extended Data Fig. ). Accuracy was defined by either the fraction of correct frames within a 1 s window or identification of the correct scene. Frame-by-frame results also showed reduced frame ID errors (one-way ANOVA, F (3,16) = 20.22, P = 1.09 × 10 −5 , n = 1,000 neurons; Supplementary Table ), which can be seen in Fig. , Extended Data Fig. and Supplementary Video . The DINO features themselves did not drive performance, because shuffling of features showed poor decoding (Extended Data Fig. ). Lastly, we tested decoding from other higher visual areas using DINO features. Overall, decoding from V1 had the highest performance and VISrl the lowest (Fig. and Extended Data Fig. ). Given the high decoding performance of CEBRA, we tested whether there was a particular V1 layer that was most informative. We leveraged CEBRA-Behaviour by training models on each category and found that layers 2/3 and 5/6 showed significantly higher decoding performance compared with layer 4 (one-way ANOVA, F (2,12) = 9.88, P = 0.003; Fig. ). Given the known cortical connectivity, this suggests that the nonthalamic input layers render frame information more explicit, perhaps via feedback or predictive processing. CEBRA is a nonlinear dimensionality reduction method newly developed to explicitly leverage auxiliary (behaviour) labels and/or time to discover latent features in time series data—in this case, latent neural embeddings. The unique property of CEBRA is the extension and generalization of the standard InfoNCE objective by introduction of a variety of different sampling strategies tuned for usage of the algorithm in the experimental sciences and for analysis of time series datasets, and it can also be used for supervised and self-supervised analysis, thereby directly facilitating hypothesis- and discovery-driven science. It produces both consistent embeddings across subjects (thus showing common structure) and can find the dimensionality of neural spaces that are topologically robust. Although there remains a gap in our understanding of how these latent spaces map to neural-level computations, we believe this tool provides an advance in our ability to map behaviour to neural populations. Moreover, because pretrained CEBRA models can be used for decoding in new animals within tens of steps (milliseconds), we can thereby obtain equal or better performance compared with training on the unseen animal alone. Dimensionality reduction is often tightly linked to data visualization, and here we make an empirical argument that ultimately this is useful only when obtaining consistent results and discovering robust features. Unsupervised t -SNE and UMAP are examples of algorithms widely used in life sciences for discovery-based analysis. However, they do not leverage time and, for neural recordings, this is always available and can be used. Even more critical is that concatenation of data from different animals can lead to shifted clusters with t -SNE or UMAP due to inherent small changes across animals or in how the data were collected. CEBRA allows the user to remove this unwanted variance and discover robust latents that are invariant to animal ID, sessions or any-other-user-defined nuisance variable. Collectively we believe that CEBRA will become a complement to (or replacement for) these methods such that, at minimum, the structure of time in the neural code is leveraged and robustness is prioritized. Datasets Artificial spiking dataset The synthetic spiking data used for benchmarking in Fig. were adopted from Zhou and Wei . The continuous 1D behaviour variable [12pt]{minimal} $$c [0,2 )$$ c ∈ [ 0 , 2 π ) was sampled uniformly in the interval [12pt]{minimal} $$[0,2 )$$ [ 0 , 2 π ) . The true 2D latent variable [12pt]{minimal} $${} {{}}^{2}$$ z ∈ R 2 was then sampled from a Gaussian distribution [12pt]{minimal} $${}( (c), (c))$$ N μ c , Σ c with mean [12pt]{minimal} $$ (c)={(c,2 c)}^{ }$$ μ c = c , 2 sin c ⊤ and covariance [12pt]{minimal} $$ (c)={}(0.6-0.3\| c\|,0.3\| c\|)$$ Σ c = diag 0.6 − 0.3 sin c , 0.3 sin c . After sampling, the 2D latent variable [12pt]{minimal} $${}$$ z was mapped to the spiking rates of 100 neurons by the application of four randomly initialized RealNVP blocks. Poisson noise was then applied to map firing rates onto spike counts. The final dataset consisted of 1.5 × 10 4 data points for 100 neurons ([number of samples, number of neurons]) and was split into train (80%) and validation (20%) sets. We quantified consistency across the entire dataset for all methods. Additional synthetic data, presented in Extended Data Fig. , were generated by varying noise distribution in the above generative process. Beside Poisson noise, we used additive truncated ([0,1000]) Gaussian noise with s.d. = 1 and additive uniform noise defined in [0,2], which was applied to the spiking rate. We also adapted Poisson spiking by simulating neurons with a refractory period. For this, we scaled the spiking rates to an average of 110 Hz. We sampled interspike intervals from an exponential distribution with the given rate and added a refractory period of 10 ms. Rat hippocampus dataset We used the dataset presented in Grosmark and Buzsáki . In brief, bilaterally implanted silicon probes recorded multicellular electrophysiological data from CA1 hippocampus areas from each of four male Long–Evans rats. During a given session, each rat independently ran on a 1.6-m-long linear track where they were rewarded with water at each end of the track. The numbers of recorded putative pyramidal neurons for each rat ranged between 48 and 120. Here, we processed the data as in Zhou and Wei . Specifically, the spikes were binned into 25 ms time windows. The position and running direction (left or right) of the rat were encoded into a 3D vector, which consisted of the continuous position value and two binary values indicating right or left direction. Recordings from each rat were parsed into trials (a round trip from one end of the track as a trial) and then split into train, validation and test sets with a k = 3 nested cross-validation scheme for the decoding task. Macaque dataset We used the dataset presented in Chowdhury et al. In brief, electrophysiological recordings were performed in Area 2 of somatosensory cortex (S1) in a rhesus macaque (monkey) during a centre-out reaching task with a manipulandum. Specifically, the monkey performed an eight-direction reaching task in which on 50% of trials it actively made centre-out movements to a presented target. The remaining trials were ‘passive’ trials in which an unexpected 2 Newton force bump was given to the manipulandum towards one of the eight target directions during a holding period. The trials were aligned as in Pei et al. , and we used the data for −100 and 500 ms from movement onset. We used 1 ms time bins and convolved the data using a Gaussian kernel with s.d. = 40 ms. Mouse visual cortex datasets We utilized the Allen Institute two-photon calcium imaging and Neuropixels data recorded from five mouse visual and higher visual cortical areas (VISp, VISl, VISal, VISam, VISpm and VISrl) during presentation of a monochrome video with 30 Hz frame rate, as presented previously , , . For calcium imaging (2P) we used the processed dataset from Vries et al. with a sampling rate of 30 Hz, aligned to the video frames. We considered the recordings from excitatory neurons (Emx1-IRES-Cre, Slc17a7-IRES2-Cre, Cux2-CreERT2, Rorb-IRES2-Cre, Scnn1a-Tg3-Cre, Nr5a1-Cre, Rbp4-Cre_KL100, Fezf2-CreER and Tlx3-Cre_PL56) in the ‘Visual Coding-2P’ dataset. Ten repeats of the first video (Movie 1) were shown in all session types (A, B and C) for each mouse and we used those neurons that were recorded in all three session types, found via cell registration . The Neuropixels recordings were obtained from the ‘Brain Observatory 1.1’ dataset . We used the preprocessed spike timings and binned them to a sampling frequency of 120 Hz, aligned with the video timestamps (exactly four bins aligned with each frame). The dataset contains recordings for ten repeats, and we used the same video (Movie 1) that was used for the 2P recordings. For analysis of consistency across the visual cortical areas we used a disjoint set of neurons for each seed, to avoid higher intraconsistency due to overlapping neuron identities. We made three disjoint sets of neurons by considering only neurons from session A (for 2P data) and nonoverlapping random sampling for each seed. CEBRA model framework Notation We will use x , y as general placeholder variables and denote the multidimensional, time-varying signal as s t , parameterized by time t . The multidimensional, continuous context variable c t contains additional information about the experimental condition and additional recordings, similar to the discrete categorical variable k t . The exact composition of s , c and k depends on the experimental context. CEBRA is agnostic to exact signal types; with the default parameterizations, s t and c t can have up to an order of hundreds or thousands of dimensions. For even higher-dimensional datasets (for example, raw video, audio and so on) other optimized deep learning tools can be used for feature extraction before the application of CEBRA. Applicable problem setup We refer to [12pt]{minimal} $${} X$$ x ∈ X as the reference sample and to [12pt]{minimal} $${} Y$$ y ∈ Y as a corresponding positive or negative sample. Together, ( x , y ) form a positive or negative pair based on the distribution from which y is sampled. We denote the distribution and density function of x as p ( x ), the conditional distribution and density of the positive sample y given x as [12pt]{minimal} $$p$$ p ( y ∣ x ) and the conditional distribution and density of the negative sample y given x as [12pt]{minimal} $$q$$ q y ∣ x . After sampling—and irrespective of whether we are considering a positive or negative pair—samples [12pt]{minimal} $${} {{}}^{D}$$ x ∈ R D and [12pt]{minimal} $${} {{}}^{{D}^{{} }}$$ y ∈ R D ′ are encoded by feature extractors [12pt]{minimal} $${}:X Z$$ f : X ↦ Z and [12pt]{minimal} $${{}}^{{} }:Y Z$$ f ′ : Y ↦ Z . The feature extractors map both samples from signal space [12pt]{minimal} $$X {{}}^{D},Y {{}}^{{D}^{{} }}$$ X ⊆ R D , Y ⊆ R D ′ into a common embedding space [12pt]{minimal} $$Z {{}}^{E}$$ Z ⊆ R E . The design and parameterization of the feature extractor are chosen by the user of the algorithm. Note that spaces X and Y and their corresponding feature extractors can be the same (which is the case for single-session experiments in this work), but that this is not a strict requirement within the CEBRA framework (for example, in multi-session training across animals or modalities, X and Y are selected as signals from different mice or modalities, respectively). It is also possible to include the context variable (for example, behaviour) into X , or to set x to the context variable and y to the signal variable. Given two encoded samples, a similarity measure [12pt]{minimal} $$ :Z Z {}$$ φ : Z × Z ↦ R assigns a score to a pair of embeddings. The similarity measure needs to assign a higher score to more similar pairs of points, and to have an upper bound. For this work we consider the dot product between normalized feature vectors, [12pt]{minimal} $$ ={{}}^{{}}{{}}^{{} }/ $$ φ ( z , z ′ ) = z ⊤ z ′ / τ , in most analyses (latents on a hypersphere) or the negative mean squared error, [12pt]{minimal} $$ =-\, {}-{{}}^{{} }{ }^{2}/ $$ φ ( z , z ′ ) = − ∥ z − z ′ ∥ 2 / τ (latents in Euclidean space). Both metrics can be scaled by a temperature parameter τ that is either fixed or jointly learned with the network. Other L p norms and other similarity metrics, or even a trainable neural network (a so-called projection head commonly used in contrastive learning algorithms , ), are possible choices within the CEBRA software package. The exact choice of ϕ shapes the properties of the embedding space and encodes assumptions about distributions p and q . The technique requires paired data recordings—for example, as is common in aligned time series. The signal s t , continuous context c t and discrete context k t are synced in their time point t . How the reference, positive and negative samples are constructed from these available signals is a configuration choice made by the algorithm user, and depends on the scientific question under investigation. Optimization Given the feature encoders f and f ′ for the different sample types, as well as the similarity measure ϕ , we introduce the shorthand [12pt]{minimal} $$ $$ ψ x , y = φ f x , f ′ y . The objective function can then be compactly written as: 1 [12pt]{minimal} $$_{{} {}}{}{}{}}].$$ ∫ x ∈ X d x p ( x ) [ − ∫ y ∈ Y d y p ( y \| x ) ψ ( x , y ) + log ∫ y ∈ Y d y q ( y \| x ) e ψ ( x , y ) ] . We approximate this objective by drawing a single positive example y + , and multiple negative examples y i from the distributions outlined above, and minimize the loss function 2 [12pt]{minimal} $$}}_{} p,{{}}_{+} p\\ {{}}_{1}, ,{{}}_{n} q+\, _{i=1}^{n}{e}^{ ({},{{}}_{i})}],$$ E x ∼ p ( x ) , y + ∼ p ( y \| x ) y 1 , … , y n ∼ q ( y \| x ) [ − ψ ( x , y + ) + log ∑ i = 1 n e ψ ( x , y i ) ] , with a gradient-based optimization algorithm. The number of negative samples is a hyperparameter of the algorithm, and larger batch sizes are generally preferable. For sufficiently small datasets, as used in this paper, both positive and negative samples are drawn from all available samples in the dataset. This is in contrast to the common practice in many contrastive learning frameworks in which a minibatch of samples is drawn first, which are then grouped into positive and negative pairs. Allowing access to the whole dataset to form pairs gives a better approximation of the respective distributions [12pt]{minimal} $$p$$ p y ∣ x and [12pt]{minimal} $$q$$ q y ∣ x , and considerably improves the quality of the obtained embeddings. If the dataset is sufficiently small to fit into the memory, CEBRA can be optimized with batch gradient descent—that is, using the whole dataset at each optimizer step. Goodness of fit Comparing the loss value—at both absolute and relative values across models at the same point in training time— can be used to determine goodness of fit. In practical terms, this means that one can find which hypothesis best fits one’s data, in the case of using CEBRA-Behaviour. Specifically, let us denote the objective in equation ) as L asympt and its approximation in equation with a batch size of n as L n . In the limit of many samples, the objective converges up to a constant, [12pt]{minimal} $${L}_{{}}={{}}_{n }[{L}_{n}- n]$$ L asympt = lim n → ∞ L n − log n (Supplementaty Note and ref. ). The objective also has two trivial solutions: the first is obtained for a constant [12pt]{minimal} $$ = $$ ψ x , y = ψ , which yields a value of L n = log n . This solution can be obtained when the labels are not related to the signal (e.g., with shuffled labels). It is typically not obtained during regular training because the network is initialized randomly, causing the initial embedding points to be randomly distributed in space. If the embedding points are distributed uniformly in space and ϕ is selected such that [12pt]{minimal} $${}=0$$ E φ x , y = 0 , we will also get a value that is approximately L n = log n . This value can be readily estimated by computing [12pt]{minimal} $$ $$ φ u , v for randomly distributed points. The minimizer of equation is also clearly defined as [12pt]{minimal} $${-D}_{{}}(p q)$$ − D KL p ∥ q and depends on the positive and negative distribution. For discovery-driven (time contrastive) learning, this value is impossible to estimate because it would require access to the underlying conditional distribution of the latents. However, for training with predefined positive and negative distributions, this quantity can be again numerically estimated. Interesting values of the loss function when fitting a CEBRA model are therefore 3 [12pt]{minimal} $$-{D}_{{}}(p q) {L}_{n}- n\, 0$$ − D KL p ∥ q ≤ L n − log n ≤ 0 where L n – log n is the goodness of fit (lower is better) of the CEBRA model. Note that the metric is independent of the batch size used for training. Sampling Selection of the sampling scheme is CEBRA’s key feature in regard to adapting embedding spaces to different datasets and recording setups. The conditional distributions [12pt]{minimal} $$p$$ p y ∣ x for positive samples and [12pt]{minimal} $$q$$ q y ∣ x for negative samples, as well as the marginal distribution p ( x ) for reference samples, are specified by the user. CEBRA offers a set of predefined sampling techniques but customized variants can be specified to implement additional, domain-specific distributions. This form of training allows the use of context variables to shape the properties of the embedding space, as outlined in the graphical model in Supplementary Note . Through the choice of sampling technique, various use cases can be built into the algorithm. For instance, by forcing positive and negative distributions to sample uniformly across a factor, the model will become invariant to this factor because its inclusion would yield a suboptimal value of the objective function. When considering different sampling mechanisms we distinguish between single- and multi-session datasets: a single-session dataset consists of samples s t associated to one or more context variables c t and/or k t . These context variables allow imposition of the structure on the marginal and conditional distribution used for obtaining the embedding. Multi-session datasets consist of multiple, single-session datasets. The dimension of context variables c t and/or k t must be shared across all sessions whereas the dimension of the signal s t can vary. In such a setting, CEBRA allows learning of a shared embedding space for signals from all sessions. For single-session datasets, sampling is done in two steps. First, based on a specified ‘index’ (the user-defined context variable c t and/or k t ), locations t are sampled for reference, positive and negative samples. The algorithm differentiates between categorical ( k ) and continuous ( c ) variables for this purpose. In the simplest case, negative sampling ( q ) returns a random sample from the empirical distribution by returning a randomly chosen index from the dataset. Optionally, with a categorical context variable [12pt]{minimal} $${k}_{t} [K]$$ k t ∈ K , negative sampling can be performed to approximate a uniform distribution of samples over this context variable. If this is performed for both negative and positive samples, the resulting embedding will become invariant with respect to the variable k t . Sampling is performed in this case by computing the cumulative histogram of k t and sampling uniformly over k using the transformation theorem for probability densities. For positive pairs, different options exist based on the availability of continuous and discrete context variables. For a discrete context variable [12pt]{minimal} $${k}_{t} [K]$$ k t ∈ K with K possible values, sampling from the conditional distribution is done by filtering the whole dataset for the value k t of the reference sample, and uniformly selecting a positive sample with the same value. For a continuous context variable c t we can use a set of time offsets Δ to specify the distribution. Given the time offsets, the empirical distribution [12pt]{minimal} $$P({{}}_{t+ }\,{}\,{{}}_{t})$$ P c t + τ ∣ c t for a particular choice of [12pt]{minimal} $$ $$ τ ∈ Δ can be computed from the dataset: we build up a set [12pt]{minimal} $$D=\{t [T], :{{}}_{t+ }-{{}}_{t}\}$$ D = { t ∈ T , τ ∈ Δ : c t + τ − c t } , sample a d uniformly from D and obtain the sample that is closest to the reference sample’s context variable modified by this distance ( c + d ) from the dataset. It is possible to combine a continuous variable c t with a categorical variable k t for mixed sampling. On top of the continual sampling step above, it is ensured that both samples in the positive pair share the same value of k t . It is crucial that the context samples c and the norm used in the algorithm match in some way; for simple context variables with predictable conditional distributions (for example, a 1D or 2D position of a moving animal, which can most probably be well described by a Gaussian conditional distribution based on the previous sample), the positive sample distribution can also be specified directly, for example, as a normal distribution centred around c t . An additional alternative is to use CEBRA also to preprocess the original context samples c and use the embedded context samples with the metric used for CEBRA training. This scheme is especially useful for higher-dimensional behavioural data, or even for complex inputs such as video. We next consider the multi-session case in which signals [12pt]{minimal} $${{}}_{t}^{(i)} {{}}^{{n}_{i}}$$ s t i ∈ R n i come from N different sessions [12pt]{minimal} $$i [N]$$ i ∈ N with session-dependent dimensionality n i . Importantly, the corresponding continuous context variables [12pt]{minimal} $${{}}_{t}^{(i)} {{}}^{m}$$ c t i ∈ R m share the same dimensionality m , which makes it possible to relate samples across sessions. The multi-session setup is similar to mixed-session sampling (if we treat the session ID as a categorical variable [12pt]{minimal} $${k}_{t}^{(i)}:\,=i$$ k t ( i ) : = i for all time steps t in session i ). The conditional distribution for both negative and positive pairs is uniformly sampled across sessions, irrespective of session length. Multi-session mixed or discrete sampling can be implemented analogously. CEBRA is sufficiently flexible to incorporate more specialized sampling schemes beyond those outlined above. For instance, mixed single-session sampling could be extended additionally to incorporate a dimension to which the algorithm should become invariant; this would add an additional step of uniform sampling with regard to this desired discrete variable (for example, via ancestral sampling). Choice of reference, positive and negative samples Depending on the exact application, the contrastive learning step can be performed by explicitly including or excluding the context variable. The reference sample x can contain information from the signal s t , but also from the experimental conditions, behavioural recordings or other context variables. The positive and negative samples y are set to the signal variable s t . Theoretical guarantees for linear identifiability of CEBRA models Identifiability describes the property of an algorithm to give a consistent estimate for the model parameters given that the data distributions match. We here apply the relaxed notion of linear identifiability that was previously discussed and used , . After training two encoder models f and f ′, the models are linearly identifiable if f ( x ) = L f( x ), where L is a linear map. When applying CEBRA, three cases are of potential interest. (1) When applying discovery-driven CEBRA, will two models estimated on comparable experimental data agree in their inferred representation? (2) Under which assumptions about the data will we be able to discover the true latent distribution? (3) In the hypothesis-driven or hybrid application of CEBRA, is the algorithm guaranteed to give a meaningful (nonstandard) latent space when we can find signal within the data? For the first case, we note that the CEBRA objective with a cosine similarity metric follows the canonical discriminative form for which Roeder et al. showed linear identifiability: for sufficiently diverse datasets, two CEBRA models trained to convergence on the same dataset will be consistent up to linear transformations. Note that the consistency of CEBRA is independent of the exact data distribution: it is merely required that the embeddings of reference samples across multiple positive pairs, and the embeddings of negative samples across multiple negative pairs, vary in sufficiently numerous linearly independent directions. Alternatively, we can derive linear identifiability from assumptions about data distribution: if the ground truth latents are sufficiently diverse (that is, vary in all latent directions under distributions p and q ), and the model is sufficiently parameterized to fit the data, we will also obtain consistency up to a linear transformation. See Supplementary Note for a full formal discussion and proof. For the second case, additional assumptions are required regarding the exact form of data-generating distributions. Within the scope of this work we consider ground truth latents distributed on the hypersphere or Euclidean space. The metric then needs to match assumptions about the variation of ground truth latents over time. In discovery-driven CEBRA, using the dot product as the similarity measure then encodes the assumption that latents vary according to a von Mises–Fisher distribution whereas the (negative) mean squared error encodes an assumption that latents vary according to a normal distribution. More broadly, if we assume that the latents have a uniform marginal distribution (which can be ensured by designing unbiased experiments), the similarity measure should be chosen as the log-likelihood of conditional distribution over time. In this case, CEBRA identifies the latents up to an affine transformation (in the most general case). This result also explains the empirically high performance of CEBRA for decoding applications: if trained for decoding (using the variable to decode for informing the conditional distribution), it is trivial to select matching conditional distributions because both quantities are directly selected by the user. CEBRA then ‘identifies’ the context variable up to an affine transformation. For the third case, we are interested in hypothesis-testing capabilities. We can show that if a mapping exists between the context variable and the signal space, CEBRA will recover this relationship and yield a meaningful embedding, which is also decodable. However, if such a mapping does not exist we can show that CEBRA will not learn a structured embedding. CEBRA models We chose X = Y as the neural signal, with varying levels of recorded neurons and channels based on the dataset. We used three types of encoder model based on the required receptive field: a receptive field of one sample was used for the synthetic dataset experiments (Fig. ) and a receptive field of ten samples in all other experiments (rat, monkey, mouse) except for the Neuropixels dataset, in which a receptive field of 40 samples was used due to the fourfold higher sampling rate of the dataset. All feature encoders were parameterized by the number of neurons (input dimension), a hidden dimension used to control model size and capacity, as well as by their output (embedding) dimension. For the model with the receptive field of one, a four-layer MLP was used. The first and second layers map their respective inputs to the hidden dimension whereas the third introduces a bottleneck and maps to half the hidden dimension. The final layer maps to the requested output dimension. For the model with a receptive field of ten, a convolutional network with five time-convolutional layers was used. The first layer had a kernel size of two, and the next three had a kernel size of three and used skip connections. The final layer had a kernel size of three and mapped hidden dimensions to the output dimension. For the model with receptive field 40, we first preprocessed the signal by concatenating a 2× downsampled version of the signal with a learnable downsample operation implemented as a convolutional layer with kernel size four and stride two, directly followed (without activation function between) by another convolutional layer with kernel size three and stride two. After these first layers, the signal was subsampled by a factor of four. Afterwards, similar to the receptive field ten model, we applied three layers with kernel size three and skip connections and a final layer with kernel size three. In all models, Gaussian error linear unit activation functions were applied after each layer except the last. The feature vector was normalized after the last layer unless a mean squared error-based similarity metric was used (as shown in Extended Data Fig. ). Our implementation of the InfoNCE criterion received a minibatch (or the full dataset) of size n × d for each of the reference, positive and negative samples. n dot-product similarities were computed between reference and positive samples and n × n dot-product similarities were computed between reference and negative samples. Similarities were scaled with the inverse of the temperature parameter τ : from torch import einsum, logsumexp, no_grad def info_nce(ref, pos, neg, tau = 1.0): pos_dist = einsum(“nd,nd–>n”, ref, pos)/tauneg_dist = einsum(“nd,md–>nm”, ref, neg)/tauwith no_grad:c, _ = neg_dist.max(dim=1)pos_dist = pos_dist – c.detachneg_dist = neg_dist – c.detachpos_loss = –pos_dist.meanneg_loss = logsumexp(neg_dist, dim = 1).meanreturn pos_loss + neg_loss Alternatively, a learnable temperature can be used. For a numerically stable implementation we store the log inverse temperature [12pt]{minimal} $$ =-\, $$ α = − log τ as a parameter of loss function. At each step we scale the distances in loss function with [12pt]{minimal} $$ ( ,\,1/{ }_{ })$$ min exp α , 1 / τ min . The additional parameter τ min is a lower bound on the temperature. The inverse temperature used for scaling the distances in the loss will hence lie in [12pt]{minimal} $$(0,1/{ }_{ }]$$ ( 0 , 1 / τ min ] . CEBRA model parameters used In the main figures we have used the default parameters ( https://cebra.ai/docs/api.html ) for fitting CEBRA unless otherwise stated in the text (such as dimension, which varied and is noted in figure legends), or below: Synthetic data: model_architecture= ‘offset1-model-mse’, conditional= ‘delta’, delta=0.1, distance= ‘euclidean’, batch_size=512, learning_rate=1e-4. Rat hippocampus neural data: model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Rat behavioural data: model_architecture= ‘offset10-model-mse’, distance= ‘euclidean’, time_offsets=10, batch_size=512. Primate S1 neural data: model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Allen datasets (2P): model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Allen datasets (NP): model_architecture= ‘offset40-model-4x-subsample’, time_offsets=10, batch_size=512. CEBRA API and example usage The Python implementation of CEBRA is written in PyTorch and NumPy and provides an application programming interface (API) that is fully compatible with scikit-learn , a package commonly used for machine learning. This allows the use of scikit-learn tools for hyperparameter selection and downstream processing of the embeddings—for example, decoding. CEBRA can be used as a dropin replacement in existing data pipelines for algorithms such as t -SNE, UMAP, PCA or FastICA. Both CPU and GPU implementations are available. Using the previously introduced notations, suppose we have a dataset containing signals s t , continuous context variables c t and discrete context variables k t for all time steps t , import numpy as npN = 500s = np.zeros((N, 55), dtype = float)k = np.zeros((N,), dtype = int)c = np.zeros((N, 10), dtype = float) along with a second session of data, s2 = np.zeros((N, 75), dtype = float)c2 = np.zeros((N, 10), dtype = float) assert c2.shape[1] == c.shape[1]:note that both the number of samples and the dimension in s ′ does not need to match s . Session alignment leverages the fact that the second dimensions of c and c ′ match. With this dataset in place, different variants of CEBRA can be applied as follows: import cebramodel = cebra.CEBRA(output_dimension=8,num_hidden_units=32,batch_size=1024,learning_rate=3e-4,max_iterations=1000) The training mode to use is determined automatically based on what combination of data is passed to the algorithm: # time contrastive learning model.fit(s) # discrete behaviour contrastive learning model.fit(s, k) # continuous behaviour contrastive learning model.fit(s, c) # mixed behaviour contrastive learning model.fit(s, c, k) # multi-session training model.fit([s, s2], [c, c2]) # adapt to new session model.fit(s, c) model.fit(s2, c2, adapt = True) Because CEBRA is a parametric method training a neural network internally, it is possible to embed new data points after fitting the model: s_test = np.zeros((N, 55), dtype=float)# obtain and plot embeddingz = model.transform(s_test)plt.scatter(z[:, 0], z[:, 1])plt.show Besides this simple-to-use API for end users, our implementation of CEBRA is a modular software library that includes a plugin system, allowing more advanced users to readily add additional model implementations, similarity functions, datasets and data loaders and distributions for sampling positive and negative pairs. Consistency of embeddings across runs, subjects, sessions, recording modalities and areas To measure the consistency of the embeddings we used the R 2 score of linear regression (including an intercept) between embeddings from different subjects (or sessions). Secondly, pi-VAE, which we benchmarked and improved (Extended Data Fig. ), demonstrated a theoretical guarantee that it can reconstruct the true latent space up to an affine transformation. Across runs, we measured the R 2 score of linear regression between embeddings across ten runs of the algorithms, yielding 90 comparisons. These runs were done with the same hyperparameters, model and training setup. For the rat hippocampus data, the numbers of neurons recorded were different across subjects. The behaviour setting was the same: the rats moved along a 1.6-meter-long track and, for analysis, behaviour data were binned into 100 bins of equal size for each direction (leftwards, rightwards). We computed averaged feature vectors for each bin by averaging all normalized CEBRA embeddings for a given bin and renormalized the average to lie on the hypersphere. If a bin did not contain any sample, it was filled by samples from the two adjacent bins. CEBRA was trained with latent dimension three (the minimum) such that it was constrained to lie only on a two-sphere (making this ‘3D’ space equivalent to a 2D Euclidean space). All other methods were trained with two latent dimensions in Euclidean space. Note that n + 1 dimensions of CEBRA are equivalent to n dimensions of other methods that we compared, because the feature space of CEBRA is normalized (that is, the feature vectors are normalized to have unit length). For Allen visual data in which the number of behavioural data points is the same across different sessions (that is, fixed length of video stimuli), we directly computed the R 2 score of linear regression between embeddings from different sessions and modalities. We surveyed three, four, eight, 32, 64 and 128 latent dimensions with CEBRA. To compare the consistency of embeddings between or within the areas considered, we computed intra- and interarea consistency within the same recording modality (2P or NP). Within the same modality we sampled 400 neurons from each area. We trained one CEBRA model per area and computed linear consistency between all pairs of embeddings. For intra-area comparison we sampled an additional 400 disjoint neurons. For each area we trained two CEBRA models on these two sets of neurons and computed their linear consistency. We repeated this process three times. For comparisons across modalities (2P and NP) we sampled 400 neurons from each modality (which are disjoint, as above, because one set was sampled from 2P recordings and the other from NP recordings). We trained a multi-session CEBRA model with one encoder for 2P and one for NP in the same embedding space. For intra-area comparison we computed linear consistency between NP and 2P decoders from the same area. For interarea comparison we computed linear consistency between the NP encoder from one area and the 2P encoder from another and again considered all combinations of areas. We repeated this process three times. For comparison of single- and multi-session training (Extended Data Fig. ) we computed embeddings using encoder models with eight, 16, … , 128 hidden units to vary the model size, and benchmarked eight, 16, … , 128 latent dimensions. Hyperparameters, except for number of optimization steps, were selected according to either validation set decoding R 2 (rat) or accuracy (Allen). Consistency was reported as the point in training at which position decoding error was less than 7 cm for the first rat in the hippocampus dataset, and a decoding accuracy of 60% in the Allen dataset. For single-session training, four embeddings were trained independently on each individual animal whereas for multi-session training the embeddings were trained jointly on all sessions. For multi-session training, the same number of samples was drawn from each session to learn an embedding invariant to the session ID. The consistency versus decoding error trade-off (Extended Data Fig. ) was reported as the average consistency across all 12 comparisons (Extended Data Fig. ) versus average decoding performance across all rats and data splits. Model comparisons pi-VAE parameter selection and modifications to pi-VAE Because the original implementation of pi-VAE used a single time bin spiking rate as an input, we therefore modified their code to allow for larger time bin inputs and found that time window input with a receptive field of ten time bins (250 ms) gave higher consistency across subjects and better preserved the qualitative structure of the embedding (thereby outperforming the results presented by Zhou and Wei ; Extended Data Fig. ). To do this we used the same encoder neural network architecture as that for CEBRA and modified the decoder to a 2D output (we call our modified version conv-pi-VAE). Note, we used this modified pi-VAE for all experiments except for the synthetic setting, for which there is no time dimension and thus the original implementation is sufficient. The original implementation reported a median absolute error of 12 cm for rat 1 (the individual considered most in that work), and our implementation of time-windowed input with ten bins resulted in a median absolute error of 11 cm (Fig. ). For hyperparameters we tested different epochs between 600 (the published value used) and 1,000, and learning rate between 1.0 × 10 −6 and 5.0 × 10 −4 via a grid search. We fixed hyperparameters as those that gave the highest consistency across subjects, which were training epochs of 1,000 and learning rate 2.5 × 10 −4 . All other hyperparameters were retained as in the original implementation . Note that the original paper demonstrated that pi-VAE is fairly robust across different hyperparameters. For decoding (Fig. ) we considered both a simple kNN decoder (that we use for CEBRA) and the computationally more expensive Monte Carlo sampling method originally proposed for pi-VAE . Our implementation of conv-pi-VAE can be found at https://github.com/AdaptiveMotorControlLab/CEBRA . autoLFADS parameter selection AutoLFADS includes a hyperparameter selection and tuning protocol, which we used, and we also used the original implementation ( https://github.com/snel-repo/autolfads-tf2/ , https://github.com/neurallatents/nlb_tools/tree/main/examples/baselines/autolfads ). For the rat hippocampus dataset we chopped the continuous spiking rate (25 ms bin size) into 250-ms-long segments with 225 ms overlap between segments to match the training setup for CEBRA, UMAP, t -SNE and pi-VAE. We used population-based training (PBT) for hyperparameter searches and constrained the search range to default values given in the original script (initial learning rate between 1.0 × 10 −5 and 5.0 × 10 −3 , dropout rate 0–0.6, coordinated dropout rate 0.01–0.70, L2 generator weight between 1.0 × 10 −4 and 1.0, L2 controller weight between 1.0 × 10 −4 and 1.0, KL controller weight between 1.0 × 10 −6 and 1.0 × 10 −4 and KL initial condition weight between 1.0 × 10 −6 and 1.0 × 10 –3 ). The negative log-likelihood metric was used to select the best hyperparameters. Each generation of PBT consisted of 25 training epochs and we trained for a maximum of 5,000 epochs of batch size 100 while executing early stopping after awaiting 50 epochs. The PBT search was done using 20 parallel workers on each rat. UMAP parameter selection For UMAP , following the parameter guide ( umap-learn.readthedocs.io/ ), we focused on tuning the number of neighbours ( n _ neighbors ) and minimum distance ( min _ dist ). The n _ components parameter was fixed to 2 and we used a cosine metric to make a fair comparison with CEBRA, which also used the cosine distance metric for learning. We performed a grid search with 100 total hyperparameter values in the range [2, 200] for n _ neighbors and in the range [0.0001, 0.99] for min _ dist . The highest consistency across runs in the rat hippocampus dataset was achieved with min _ dist of 0.0001 and n _ neighbors of 24. For the other datasets in Extended Data Fig. we used the default value of n _ neighbors as 15 and min _ dist as 0.1. t -SNE parameter selection For t -SNE we used the implementation in openTSNE . We performed a sweep on perplexity in the range [5, 50] and early _ exaggeration in the range [12, 32] following the parameter guide, while fixing n _ components as 2 and used a cosine metric for fair comparison with UMAP and CEBRA. We used PCA initialization to improve the run consistency of t -SNE . The highest consistency across runs in the rat hippocampus dataset was achieved with perplexity of ten and early _ exaggeration of 16.44. For the other datasets in Extended Data Fig. we used the default value for perplexity of 30 and for early _ exaggeration of 12. Decoding analysis We primarily used a simple kNN algorithm, a nonparametric supervised learning method, as a decoding method for CEBRA. We used the implementation in scikit-learn . We used a kNN regressor for continuous value regression and a kNN classifier for discrete label classification. For embeddings obtained with cosine metrics we used cosine distance metrics for kNN, and Euclidean distance metrics for those obtained in Euclidean space. For the rat hippocampus data a kNN regressor, as implemented in scikit-learn , was used to decode position and a kNN classifier to decode direction. The number of neighbours was searched over the range [1, 4, 9, 16, 25] and we used the cosine distance metric. We used the R 2 score of predicted position and direction vector on the validation set as a metric to choose the best n_neighbours parameter. We report the median absolute error for positional decoding on the test set. For pi-VAE, we additionally evaluated decoding quality using the originally proposed decoding method based on Monte Carlo sampling, with the settings from the original article . For autoLFADS, use of their default Ridge regression decoder performed worse than our kNN decoder, which is why we reported all results for the kNN decoder. Note that UMAP, t -SNE and CEBRA-Time were trained using the full dataset without label information when learning the embedding, and we used the above split only for training and cross-validation of the decoder. For direction decoding within the monkey dataset we used a Ridge classifier as a baseline. The regularization hyperparameter was searched over [10 −6 , 10 2 ]. For CEBRA we used a kNN classifier for decoding direction with k searched over the range [1, 2500]. For conv-pi-VAE we searched for the best learning rate over [1.0 × 10 −5 , 1.0 × 10 −3 ]. For position decoding we used Lasso as a baseline. The regularization hyperparameter was searched over [10 −6 , 10 2 ]. For conv-pi-VAE we used 600 epochs and searched for the best learning rates over [5 × 10 −4 , 2.5 × 10 −4 , 0.125 × 10 −4 , 5 × 10 −5 ] via a grid of ( x , y ) space in 1 cm bins for each axis as the sampling process for decoding. For CEBRA we used kNN regression, and the number of neighbours k was again searched over [1, 2500]. For the Allen Institute datasets we performed decoding (frame number or scene classification) for each frame from Video 1. Here we used a kNN classifier with a population vector kNN as a baseline, similar to the decoding of orientation grating performed in ref. . For CEBRA we used the same kNN classifier method as on CEBRA features. In both cases the number of neighbours, k , was searched over a range [1, 100] in an exponential fashion. We used neural data recorded during the first eight repeats as the train set, the ninth repeat for validation in choosing the hyperparameter and the last repeat as the test set to report decoding accuracy. We also used a Gaussian naive Bayes decoder to test linear decoding from the CEBRA model and neural population vector. Here we assumed uniform priors over frame number and searched over the range [10 −10 , 10 3 ] in an exponential manner for the var_smoothing hyperparameter. For layer-specific decoding we used data from excitatory neurons in area VISp: layers 2/3 [Emx1-IRES-Cre, Slc17a7-IRES2-Cre]; layer 4 [Cux2-CreERT2, Rorb-IRES2-Cre, Scnn1a-Tg3-Cre]; and layers 5/6 [Nr5a1-Cre, Rbp4-Cre_KL100, Fezf2-CreER, Tlx3-Cre_PL56, Ntrsr1-cre]. Neural Latents Benchmark We tested CEBRA on the mc-maze 20 ms task from the Neural Latents Benchmark ( https://eval.ai/web/challenges/challenge-page/1256/leaderboard/3183 ). We trained the offset10-model with 48 output dimensions and [128, 256, 512] hidden units, as presented throughout the paper. We trained, in total, 48 models by additionally varying the temperature in [0.0001, 0.004] and time offsets from {1, 2}. We performed smoothing of input neural data using a Gaussian kernel with 50 ms s.d. Lastly, we took 45 embeddings from the trained models picked by the validation score, aligned the embeddings (using the Procrustes method ) and averaged them. Topological analysis For the persistent cohomology analysis we utilized ripser.py . For the hippocampus dataset we used 1,000 randomly sampled points from CEBRA-Behaviour trained with temperature 1, time offset 10 and minibatch size 512 for 10,000 training steps on the full dataset and then analysed up to 2D cohomology. Maximum distance considered for filtration was set to infinity. To determine the number of cocycles in each cohomology dimension with a significant lifespan we trained 500 CEBRA embeddings with shuffled labels, similar to the approach taken in ref. . We took the maximum lifespan of each dimension across these 500 runs as a threshold to determine robust Betti numbers, then surveyed the Betti numbers of CEBRA embeddings across three, eight, 16, 32 and 64 latent dimensions. Next we used DREiMac to obtain topology-preserving circular coordinates (radial angle) of the first cocycle (H 1 ) from the persistent cohomology analysis. Similar to above, we used 1,000 randomly sampled points from the CEBRA-Behaviour models of embedding dimensions 3, 8, 16, 32 and 64. Behaviour embeddings for video datasets High-dimensional inputs, such as videos, need further preprocessing for effective use with CEBRA. First we used the recently presented DINO model to embed video frames into a 768D feature space. Specifically we used the pretrained ViT/8 vision transformer model, which was trained by a self-supervised learning objective on the ImageNet database. This model is particularly well suited for video analysis and among the state-of-the-art models available for embedding natural images into a space appropriate for a kNN search , a desired property when making the dataset compatible with CEBRA. We obtained a normalized feature vector for each video frame, which was then used as the continuous behaviour variable for all further CEBRA experiments. For scene labels, three individuals labelled each video frame using eight candidate descriptive labels allowing multilabel classes. We took the majority vote of these three individuals to determine the label of each frame. In the case of multilabels we considered this as a new class label. The above procedure resulted in ten classes of frame annotation. Reporting summary Further information on research design is available in the linked to this article. Artificial spiking dataset The synthetic spiking data used for benchmarking in Fig. were adopted from Zhou and Wei . The continuous 1D behaviour variable [12pt]{minimal} $$c [0,2 )$$ c ∈ [ 0 , 2 π ) was sampled uniformly in the interval [12pt]{minimal} $$[0,2 )$$ [ 0 , 2 π ) . The true 2D latent variable [12pt]{minimal} $${} {{}}^{2}$$ z ∈ R 2 was then sampled from a Gaussian distribution [12pt]{minimal} $${}( (c), (c))$$ N μ c , Σ c with mean [12pt]{minimal} $$ (c)={(c,2 c)}^{ }$$ μ c = c , 2 sin c ⊤ and covariance [12pt]{minimal} $$ (c)={}(0.6-0.3\| c\|,0.3\| c\|)$$ Σ c = diag 0.6 − 0.3 sin c , 0.3 sin c . After sampling, the 2D latent variable [12pt]{minimal} $${}$$ z was mapped to the spiking rates of 100 neurons by the application of four randomly initialized RealNVP blocks. Poisson noise was then applied to map firing rates onto spike counts. The final dataset consisted of 1.5 × 10 4 data points for 100 neurons ([number of samples, number of neurons]) and was split into train (80%) and validation (20%) sets. We quantified consistency across the entire dataset for all methods. Additional synthetic data, presented in Extended Data Fig. , were generated by varying noise distribution in the above generative process. Beside Poisson noise, we used additive truncated ([0,1000]) Gaussian noise with s.d. = 1 and additive uniform noise defined in [0,2], which was applied to the spiking rate. We also adapted Poisson spiking by simulating neurons with a refractory period. For this, we scaled the spiking rates to an average of 110 Hz. We sampled interspike intervals from an exponential distribution with the given rate and added a refractory period of 10 ms. Rat hippocampus dataset We used the dataset presented in Grosmark and Buzsáki . In brief, bilaterally implanted silicon probes recorded multicellular electrophysiological data from CA1 hippocampus areas from each of four male Long–Evans rats. During a given session, each rat independently ran on a 1.6-m-long linear track where they were rewarded with water at each end of the track. The numbers of recorded putative pyramidal neurons for each rat ranged between 48 and 120. Here, we processed the data as in Zhou and Wei . Specifically, the spikes were binned into 25 ms time windows. The position and running direction (left or right) of the rat were encoded into a 3D vector, which consisted of the continuous position value and two binary values indicating right or left direction. Recordings from each rat were parsed into trials (a round trip from one end of the track as a trial) and then split into train, validation and test sets with a k = 3 nested cross-validation scheme for the decoding task. Macaque dataset We used the dataset presented in Chowdhury et al. In brief, electrophysiological recordings were performed in Area 2 of somatosensory cortex (S1) in a rhesus macaque (monkey) during a centre-out reaching task with a manipulandum. Specifically, the monkey performed an eight-direction reaching task in which on 50% of trials it actively made centre-out movements to a presented target. The remaining trials were ‘passive’ trials in which an unexpected 2 Newton force bump was given to the manipulandum towards one of the eight target directions during a holding period. The trials were aligned as in Pei et al. , and we used the data for −100 and 500 ms from movement onset. We used 1 ms time bins and convolved the data using a Gaussian kernel with s.d. = 40 ms. Mouse visual cortex datasets We utilized the Allen Institute two-photon calcium imaging and Neuropixels data recorded from five mouse visual and higher visual cortical areas (VISp, VISl, VISal, VISam, VISpm and VISrl) during presentation of a monochrome video with 30 Hz frame rate, as presented previously , , . For calcium imaging (2P) we used the processed dataset from Vries et al. with a sampling rate of 30 Hz, aligned to the video frames. We considered the recordings from excitatory neurons (Emx1-IRES-Cre, Slc17a7-IRES2-Cre, Cux2-CreERT2, Rorb-IRES2-Cre, Scnn1a-Tg3-Cre, Nr5a1-Cre, Rbp4-Cre_KL100, Fezf2-CreER and Tlx3-Cre_PL56) in the ‘Visual Coding-2P’ dataset. Ten repeats of the first video (Movie 1) were shown in all session types (A, B and C) for each mouse and we used those neurons that were recorded in all three session types, found via cell registration . The Neuropixels recordings were obtained from the ‘Brain Observatory 1.1’ dataset . We used the preprocessed spike timings and binned them to a sampling frequency of 120 Hz, aligned with the video timestamps (exactly four bins aligned with each frame). The dataset contains recordings for ten repeats, and we used the same video (Movie 1) that was used for the 2P recordings. For analysis of consistency across the visual cortical areas we used a disjoint set of neurons for each seed, to avoid higher intraconsistency due to overlapping neuron identities. We made three disjoint sets of neurons by considering only neurons from session A (for 2P data) and nonoverlapping random sampling for each seed. The synthetic spiking data used for benchmarking in Fig. were adopted from Zhou and Wei . The continuous 1D behaviour variable [12pt]{minimal} $$c [0,2 )$$ c ∈ [ 0 , 2 π ) was sampled uniformly in the interval [12pt]{minimal} $$[0,2 )$$ [ 0 , 2 π ) . The true 2D latent variable [12pt]{minimal} $${} {{}}^{2}$$ z ∈ R 2 was then sampled from a Gaussian distribution [12pt]{minimal} $${}( (c), (c))$$ N μ c , Σ c with mean [12pt]{minimal} $$ (c)={(c,2 c)}^{ }$$ μ c = c , 2 sin c ⊤ and covariance [12pt]{minimal} $$ (c)={}(0.6-0.3\| c\|,0.3\| c\|)$$ Σ c = diag 0.6 − 0.3 sin c , 0.3 sin c . After sampling, the 2D latent variable [12pt]{minimal} $${}$$ z was mapped to the spiking rates of 100 neurons by the application of four randomly initialized RealNVP blocks. Poisson noise was then applied to map firing rates onto spike counts. The final dataset consisted of 1.5 × 10 4 data points for 100 neurons ([number of samples, number of neurons]) and was split into train (80%) and validation (20%) sets. We quantified consistency across the entire dataset for all methods. Additional synthetic data, presented in Extended Data Fig. , were generated by varying noise distribution in the above generative process. Beside Poisson noise, we used additive truncated ([0,1000]) Gaussian noise with s.d. = 1 and additive uniform noise defined in [0,2], which was applied to the spiking rate. We also adapted Poisson spiking by simulating neurons with a refractory period. For this, we scaled the spiking rates to an average of 110 Hz. We sampled interspike intervals from an exponential distribution with the given rate and added a refractory period of 10 ms. We used the dataset presented in Grosmark and Buzsáki . In brief, bilaterally implanted silicon probes recorded multicellular electrophysiological data from CA1 hippocampus areas from each of four male Long–Evans rats. During a given session, each rat independently ran on a 1.6-m-long linear track where they were rewarded with water at each end of the track. The numbers of recorded putative pyramidal neurons for each rat ranged between 48 and 120. Here, we processed the data as in Zhou and Wei . Specifically, the spikes were binned into 25 ms time windows. The position and running direction (left or right) of the rat were encoded into a 3D vector, which consisted of the continuous position value and two binary values indicating right or left direction. Recordings from each rat were parsed into trials (a round trip from one end of the track as a trial) and then split into train, validation and test sets with a k = 3 nested cross-validation scheme for the decoding task. We used the dataset presented in Chowdhury et al. In brief, electrophysiological recordings were performed in Area 2 of somatosensory cortex (S1) in a rhesus macaque (monkey) during a centre-out reaching task with a manipulandum. Specifically, the monkey performed an eight-direction reaching task in which on 50% of trials it actively made centre-out movements to a presented target. The remaining trials were ‘passive’ trials in which an unexpected 2 Newton force bump was given to the manipulandum towards one of the eight target directions during a holding period. The trials were aligned as in Pei et al. , and we used the data for −100 and 500 ms from movement onset. We used 1 ms time bins and convolved the data using a Gaussian kernel with s.d. = 40 ms. We utilized the Allen Institute two-photon calcium imaging and Neuropixels data recorded from five mouse visual and higher visual cortical areas (VISp, VISl, VISal, VISam, VISpm and VISrl) during presentation of a monochrome video with 30 Hz frame rate, as presented previously , , . For calcium imaging (2P) we used the processed dataset from Vries et al. with a sampling rate of 30 Hz, aligned to the video frames. We considered the recordings from excitatory neurons (Emx1-IRES-Cre, Slc17a7-IRES2-Cre, Cux2-CreERT2, Rorb-IRES2-Cre, Scnn1a-Tg3-Cre, Nr5a1-Cre, Rbp4-Cre_KL100, Fezf2-CreER and Tlx3-Cre_PL56) in the ‘Visual Coding-2P’ dataset. Ten repeats of the first video (Movie 1) were shown in all session types (A, B and C) for each mouse and we used those neurons that were recorded in all three session types, found via cell registration . The Neuropixels recordings were obtained from the ‘Brain Observatory 1.1’ dataset . We used the preprocessed spike timings and binned them to a sampling frequency of 120 Hz, aligned with the video timestamps (exactly four bins aligned with each frame). The dataset contains recordings for ten repeats, and we used the same video (Movie 1) that was used for the 2P recordings. For analysis of consistency across the visual cortical areas we used a disjoint set of neurons for each seed, to avoid higher intraconsistency due to overlapping neuron identities. We made three disjoint sets of neurons by considering only neurons from session A (for 2P data) and nonoverlapping random sampling for each seed. Notation We will use x , y as general placeholder variables and denote the multidimensional, time-varying signal as s t , parameterized by time t . The multidimensional, continuous context variable c t contains additional information about the experimental condition and additional recordings, similar to the discrete categorical variable k t . The exact composition of s , c and k depends on the experimental context. CEBRA is agnostic to exact signal types; with the default parameterizations, s t and c t can have up to an order of hundreds or thousands of dimensions. For even higher-dimensional datasets (for example, raw video, audio and so on) other optimized deep learning tools can be used for feature extraction before the application of CEBRA. Applicable problem setup We refer to [12pt]{minimal} $${} X$$ x ∈ X as the reference sample and to [12pt]{minimal} $${} Y$$ y ∈ Y as a corresponding positive or negative sample. Together, ( x , y ) form a positive or negative pair based on the distribution from which y is sampled. We denote the distribution and density function of x as p ( x ), the conditional distribution and density of the positive sample y given x as [12pt]{minimal} $$p$$ p ( y ∣ x ) and the conditional distribution and density of the negative sample y given x as [12pt]{minimal} $$q$$ q y ∣ x . After sampling—and irrespective of whether we are considering a positive or negative pair—samples [12pt]{minimal} $${} {{}}^{D}$$ x ∈ R D and [12pt]{minimal} $${} {{}}^{{D}^{{} }}$$ y ∈ R D ′ are encoded by feature extractors [12pt]{minimal} $${}:X Z$$ f : X ↦ Z and [12pt]{minimal} $${{}}^{{} }:Y Z$$ f ′ : Y ↦ Z . The feature extractors map both samples from signal space [12pt]{minimal} $$X {{}}^{D},Y {{}}^{{D}^{{} }}$$ X ⊆ R D , Y ⊆ R D ′ into a common embedding space [12pt]{minimal} $$Z {{}}^{E}$$ Z ⊆ R E . The design and parameterization of the feature extractor are chosen by the user of the algorithm. Note that spaces X and Y and their corresponding feature extractors can be the same (which is the case for single-session experiments in this work), but that this is not a strict requirement within the CEBRA framework (for example, in multi-session training across animals or modalities, X and Y are selected as signals from different mice or modalities, respectively). It is also possible to include the context variable (for example, behaviour) into X , or to set x to the context variable and y to the signal variable. Given two encoded samples, a similarity measure [12pt]{minimal} $$ :Z Z {}$$ φ : Z × Z ↦ R assigns a score to a pair of embeddings. The similarity measure needs to assign a higher score to more similar pairs of points, and to have an upper bound. For this work we consider the dot product between normalized feature vectors, [12pt]{minimal} $$ ={{}}^{{}}{{}}^{{} }/ $$ φ ( z , z ′ ) = z ⊤ z ′ / τ , in most analyses (latents on a hypersphere) or the negative mean squared error, [12pt]{minimal} $$ =-\, {}-{{}}^{{} }{ }^{2}/ $$ φ ( z , z ′ ) = − ∥ z − z ′ ∥ 2 / τ (latents in Euclidean space). Both metrics can be scaled by a temperature parameter τ that is either fixed or jointly learned with the network. Other L p norms and other similarity metrics, or even a trainable neural network (a so-called projection head commonly used in contrastive learning algorithms , ), are possible choices within the CEBRA software package. The exact choice of ϕ shapes the properties of the embedding space and encodes assumptions about distributions p and q . The technique requires paired data recordings—for example, as is common in aligned time series. The signal s t , continuous context c t and discrete context k t are synced in their time point t . How the reference, positive and negative samples are constructed from these available signals is a configuration choice made by the algorithm user, and depends on the scientific question under investigation. Optimization Given the feature encoders f and f ′ for the different sample types, as well as the similarity measure ϕ , we introduce the shorthand [12pt]{minimal} $$ $$ ψ x , y = φ f x , f ′ y . The objective function can then be compactly written as: 1 [12pt]{minimal} $$_{{} {}}{}{}{}}].$$ ∫ x ∈ X d x p ( x ) [ − ∫ y ∈ Y d y p ( y \| x ) ψ ( x , y ) + log ∫ y ∈ Y d y q ( y \| x ) e ψ ( x , y ) ] . We approximate this objective by drawing a single positive example y + , and multiple negative examples y i from the distributions outlined above, and minimize the loss function 2 [12pt]{minimal} $$}}_{} p,{{}}_{+} p\\ {{}}_{1}, ,{{}}_{n} q+\, _{i=1}^{n}{e}^{ ({},{{}}_{i})}],$$ E x ∼ p ( x ) , y + ∼ p ( y \| x ) y 1 , … , y n ∼ q ( y \| x ) [ − ψ ( x , y + ) + log ∑ i = 1 n e ψ ( x , y i ) ] , with a gradient-based optimization algorithm. The number of negative samples is a hyperparameter of the algorithm, and larger batch sizes are generally preferable. For sufficiently small datasets, as used in this paper, both positive and negative samples are drawn from all available samples in the dataset. This is in contrast to the common practice in many contrastive learning frameworks in which a minibatch of samples is drawn first, which are then grouped into positive and negative pairs. Allowing access to the whole dataset to form pairs gives a better approximation of the respective distributions [12pt]{minimal} $$p$$ p y ∣ x and [12pt]{minimal} $$q$$ q y ∣ x , and considerably improves the quality of the obtained embeddings. If the dataset is sufficiently small to fit into the memory, CEBRA can be optimized with batch gradient descent—that is, using the whole dataset at each optimizer step. Goodness of fit Comparing the loss value—at both absolute and relative values across models at the same point in training time— can be used to determine goodness of fit. In practical terms, this means that one can find which hypothesis best fits one’s data, in the case of using CEBRA-Behaviour. Specifically, let us denote the objective in equation ) as L asympt and its approximation in equation with a batch size of n as L n . In the limit of many samples, the objective converges up to a constant, [12pt]{minimal} $${L}_{{}}={{}}_{n }[{L}_{n}- n]$$ L asympt = lim n → ∞ L n − log n (Supplementaty Note and ref. ). The objective also has two trivial solutions: the first is obtained for a constant [12pt]{minimal} $$ = $$ ψ x , y = ψ , which yields a value of L n = log n . This solution can be obtained when the labels are not related to the signal (e.g., with shuffled labels). It is typically not obtained during regular training because the network is initialized randomly, causing the initial embedding points to be randomly distributed in space. If the embedding points are distributed uniformly in space and ϕ is selected such that [12pt]{minimal} $${}=0$$ E φ x , y = 0 , we will also get a value that is approximately L n = log n . This value can be readily estimated by computing [12pt]{minimal} $$ $$ φ u , v for randomly distributed points. The minimizer of equation is also clearly defined as [12pt]{minimal} $${-D}_{{}}(p q)$$ − D KL p ∥ q and depends on the positive and negative distribution. For discovery-driven (time contrastive) learning, this value is impossible to estimate because it would require access to the underlying conditional distribution of the latents. However, for training with predefined positive and negative distributions, this quantity can be again numerically estimated. Interesting values of the loss function when fitting a CEBRA model are therefore 3 [12pt]{minimal} $$-{D}_{{}}(p q) {L}_{n}- n\, 0$$ − D KL p ∥ q ≤ L n − log n ≤ 0 where L n – log n is the goodness of fit (lower is better) of the CEBRA model. Note that the metric is independent of the batch size used for training. Sampling Selection of the sampling scheme is CEBRA’s key feature in regard to adapting embedding spaces to different datasets and recording setups. The conditional distributions [12pt]{minimal} $$p$$ p y ∣ x for positive samples and [12pt]{minimal} $$q$$ q y ∣ x for negative samples, as well as the marginal distribution p ( x ) for reference samples, are specified by the user. CEBRA offers a set of predefined sampling techniques but customized variants can be specified to implement additional, domain-specific distributions. This form of training allows the use of context variables to shape the properties of the embedding space, as outlined in the graphical model in Supplementary Note . Through the choice of sampling technique, various use cases can be built into the algorithm. For instance, by forcing positive and negative distributions to sample uniformly across a factor, the model will become invariant to this factor because its inclusion would yield a suboptimal value of the objective function. When considering different sampling mechanisms we distinguish between single- and multi-session datasets: a single-session dataset consists of samples s t associated to one or more context variables c t and/or k t . These context variables allow imposition of the structure on the marginal and conditional distribution used for obtaining the embedding. Multi-session datasets consist of multiple, single-session datasets. The dimension of context variables c t and/or k t must be shared across all sessions whereas the dimension of the signal s t can vary. In such a setting, CEBRA allows learning of a shared embedding space for signals from all sessions. For single-session datasets, sampling is done in two steps. First, based on a specified ‘index’ (the user-defined context variable c t and/or k t ), locations t are sampled for reference, positive and negative samples. The algorithm differentiates between categorical ( k ) and continuous ( c ) variables for this purpose. In the simplest case, negative sampling ( q ) returns a random sample from the empirical distribution by returning a randomly chosen index from the dataset. Optionally, with a categorical context variable [12pt]{minimal} $${k}_{t} [K]$$ k t ∈ K , negative sampling can be performed to approximate a uniform distribution of samples over this context variable. If this is performed for both negative and positive samples, the resulting embedding will become invariant with respect to the variable k t . Sampling is performed in this case by computing the cumulative histogram of k t and sampling uniformly over k using the transformation theorem for probability densities. For positive pairs, different options exist based on the availability of continuous and discrete context variables. For a discrete context variable [12pt]{minimal} $${k}_{t} [K]$$ k t ∈ K with K possible values, sampling from the conditional distribution is done by filtering the whole dataset for the value k t of the reference sample, and uniformly selecting a positive sample with the same value. For a continuous context variable c t we can use a set of time offsets Δ to specify the distribution. Given the time offsets, the empirical distribution [12pt]{minimal} $$P({{}}_{t+ }\,{}\,{{}}_{t})$$ P c t + τ ∣ c t for a particular choice of [12pt]{minimal} $$ $$ τ ∈ Δ can be computed from the dataset: we build up a set [12pt]{minimal} $$D=\{t [T], :{{}}_{t+ }-{{}}_{t}\}$$ D = { t ∈ T , τ ∈ Δ : c t + τ − c t } , sample a d uniformly from D and obtain the sample that is closest to the reference sample’s context variable modified by this distance ( c + d ) from the dataset. It is possible to combine a continuous variable c t with a categorical variable k t for mixed sampling. On top of the continual sampling step above, it is ensured that both samples in the positive pair share the same value of k t . It is crucial that the context samples c and the norm used in the algorithm match in some way; for simple context variables with predictable conditional distributions (for example, a 1D or 2D position of a moving animal, which can most probably be well described by a Gaussian conditional distribution based on the previous sample), the positive sample distribution can also be specified directly, for example, as a normal distribution centred around c t . An additional alternative is to use CEBRA also to preprocess the original context samples c and use the embedded context samples with the metric used for CEBRA training. This scheme is especially useful for higher-dimensional behavioural data, or even for complex inputs such as video. We next consider the multi-session case in which signals [12pt]{minimal} $${{}}_{t}^{(i)} {{}}^{{n}_{i}}$$ s t i ∈ R n i come from N different sessions [12pt]{minimal} $$i [N]$$ i ∈ N with session-dependent dimensionality n i . Importantly, the corresponding continuous context variables [12pt]{minimal} $${{}}_{t}^{(i)} {{}}^{m}$$ c t i ∈ R m share the same dimensionality m , which makes it possible to relate samples across sessions. The multi-session setup is similar to mixed-session sampling (if we treat the session ID as a categorical variable [12pt]{minimal} $${k}_{t}^{(i)}:\,=i$$ k t ( i ) : = i for all time steps t in session i ). The conditional distribution for both negative and positive pairs is uniformly sampled across sessions, irrespective of session length. Multi-session mixed or discrete sampling can be implemented analogously. CEBRA is sufficiently flexible to incorporate more specialized sampling schemes beyond those outlined above. For instance, mixed single-session sampling could be extended additionally to incorporate a dimension to which the algorithm should become invariant; this would add an additional step of uniform sampling with regard to this desired discrete variable (for example, via ancestral sampling). Choice of reference, positive and negative samples Depending on the exact application, the contrastive learning step can be performed by explicitly including or excluding the context variable. The reference sample x can contain information from the signal s t , but also from the experimental conditions, behavioural recordings or other context variables. The positive and negative samples y are set to the signal variable s t . Theoretical guarantees for linear identifiability of CEBRA models Identifiability describes the property of an algorithm to give a consistent estimate for the model parameters given that the data distributions match. We here apply the relaxed notion of linear identifiability that was previously discussed and used , . After training two encoder models f and f ′, the models are linearly identifiable if f ( x ) = L f( x ), where L is a linear map. When applying CEBRA, three cases are of potential interest. (1) When applying discovery-driven CEBRA, will two models estimated on comparable experimental data agree in their inferred representation? (2) Under which assumptions about the data will we be able to discover the true latent distribution? (3) In the hypothesis-driven or hybrid application of CEBRA, is the algorithm guaranteed to give a meaningful (nonstandard) latent space when we can find signal within the data? For the first case, we note that the CEBRA objective with a cosine similarity metric follows the canonical discriminative form for which Roeder et al. showed linear identifiability: for sufficiently diverse datasets, two CEBRA models trained to convergence on the same dataset will be consistent up to linear transformations. Note that the consistency of CEBRA is independent of the exact data distribution: it is merely required that the embeddings of reference samples across multiple positive pairs, and the embeddings of negative samples across multiple negative pairs, vary in sufficiently numerous linearly independent directions. Alternatively, we can derive linear identifiability from assumptions about data distribution: if the ground truth latents are sufficiently diverse (that is, vary in all latent directions under distributions p and q ), and the model is sufficiently parameterized to fit the data, we will also obtain consistency up to a linear transformation. See Supplementary Note for a full formal discussion and proof. For the second case, additional assumptions are required regarding the exact form of data-generating distributions. Within the scope of this work we consider ground truth latents distributed on the hypersphere or Euclidean space. The metric then needs to match assumptions about the variation of ground truth latents over time. In discovery-driven CEBRA, using the dot product as the similarity measure then encodes the assumption that latents vary according to a von Mises–Fisher distribution whereas the (negative) mean squared error encodes an assumption that latents vary according to a normal distribution. More broadly, if we assume that the latents have a uniform marginal distribution (which can be ensured by designing unbiased experiments), the similarity measure should be chosen as the log-likelihood of conditional distribution over time. In this case, CEBRA identifies the latents up to an affine transformation (in the most general case). This result also explains the empirically high performance of CEBRA for decoding applications: if trained for decoding (using the variable to decode for informing the conditional distribution), it is trivial to select matching conditional distributions because both quantities are directly selected by the user. CEBRA then ‘identifies’ the context variable up to an affine transformation. For the third case, we are interested in hypothesis-testing capabilities. We can show that if a mapping exists between the context variable and the signal space, CEBRA will recover this relationship and yield a meaningful embedding, which is also decodable. However, if such a mapping does not exist we can show that CEBRA will not learn a structured embedding. We will use x , y as general placeholder variables and denote the multidimensional, time-varying signal as s t , parameterized by time t . The multidimensional, continuous context variable c t contains additional information about the experimental condition and additional recordings, similar to the discrete categorical variable k t . The exact composition of s , c and k depends on the experimental context. CEBRA is agnostic to exact signal types; with the default parameterizations, s t and c t can have up to an order of hundreds or thousands of dimensions. For even higher-dimensional datasets (for example, raw video, audio and so on) other optimized deep learning tools can be used for feature extraction before the application of CEBRA. We refer to [12pt]{minimal} $${} X$$ x ∈ X as the reference sample and to [12pt]{minimal} $${} Y$$ y ∈ Y as a corresponding positive or negative sample. Together, ( x , y ) form a positive or negative pair based on the distribution from which y is sampled. We denote the distribution and density function of x as p ( x ), the conditional distribution and density of the positive sample y given x as [12pt]{minimal} $$p$$ p ( y ∣ x ) and the conditional distribution and density of the negative sample y given x as [12pt]{minimal} $$q$$ q y ∣ x . After sampling—and irrespective of whether we are considering a positive or negative pair—samples [12pt]{minimal} $${} {{}}^{D}$$ x ∈ R D and [12pt]{minimal} $${} {{}}^{{D}^{{} }}$$ y ∈ R D ′ are encoded by feature extractors [12pt]{minimal} $${}:X Z$$ f : X ↦ Z and [12pt]{minimal} $${{}}^{{} }:Y Z$$ f ′ : Y ↦ Z . The feature extractors map both samples from signal space [12pt]{minimal} $$X {{}}^{D},Y {{}}^{{D}^{{} }}$$ X ⊆ R D , Y ⊆ R D ′ into a common embedding space [12pt]{minimal} $$Z {{}}^{E}$$ Z ⊆ R E . The design and parameterization of the feature extractor are chosen by the user of the algorithm. Note that spaces X and Y and their corresponding feature extractors can be the same (which is the case for single-session experiments in this work), but that this is not a strict requirement within the CEBRA framework (for example, in multi-session training across animals or modalities, X and Y are selected as signals from different mice or modalities, respectively). It is also possible to include the context variable (for example, behaviour) into X , or to set x to the context variable and y to the signal variable. Given two encoded samples, a similarity measure [12pt]{minimal} $$ :Z Z {}$$ φ : Z × Z ↦ R assigns a score to a pair of embeddings. The similarity measure needs to assign a higher score to more similar pairs of points, and to have an upper bound. For this work we consider the dot product between normalized feature vectors, [12pt]{minimal} $$ ={{}}^{{}}{{}}^{{} }/ $$ φ ( z , z ′ ) = z ⊤ z ′ / τ , in most analyses (latents on a hypersphere) or the negative mean squared error, [12pt]{minimal} $$ =-\, {}-{{}}^{{} }{ }^{2}/ $$ φ ( z , z ′ ) = − ∥ z − z ′ ∥ 2 / τ (latents in Euclidean space). Both metrics can be scaled by a temperature parameter τ that is either fixed or jointly learned with the network. Other L p norms and other similarity metrics, or even a trainable neural network (a so-called projection head commonly used in contrastive learning algorithms , ), are possible choices within the CEBRA software package. The exact choice of ϕ shapes the properties of the embedding space and encodes assumptions about distributions p and q . The technique requires paired data recordings—for example, as is common in aligned time series. The signal s t , continuous context c t and discrete context k t are synced in their time point t . How the reference, positive and negative samples are constructed from these available signals is a configuration choice made by the algorithm user, and depends on the scientific question under investigation. Given the feature encoders f and f ′ for the different sample types, as well as the similarity measure ϕ , we introduce the shorthand [12pt]{minimal} $$ $$ ψ x , y = φ f x , f ′ y . The objective function can then be compactly written as: 1 [12pt]{minimal} $$_{{} {}}{}{}{}}].$$ ∫ x ∈ X d x p ( x ) [ − ∫ y ∈ Y d y p ( y \| x ) ψ ( x , y ) + log ∫ y ∈ Y d y q ( y \| x ) e ψ ( x , y ) ] . We approximate this objective by drawing a single positive example y + , and multiple negative examples y i from the distributions outlined above, and minimize the loss function 2 [12pt]{minimal} $$}}_{} p,{{}}_{+} p\\ {{}}_{1}, ,{{}}_{n} q+\, _{i=1}^{n}{e}^{ ({},{{}}_{i})}],$$ E x ∼ p ( x ) , y + ∼ p ( y \| x ) y 1 , … , y n ∼ q ( y \| x ) [ − ψ ( x , y + ) + log ∑ i = 1 n e ψ ( x , y i ) ] , with a gradient-based optimization algorithm. The number of negative samples is a hyperparameter of the algorithm, and larger batch sizes are generally preferable. For sufficiently small datasets, as used in this paper, both positive and negative samples are drawn from all available samples in the dataset. This is in contrast to the common practice in many contrastive learning frameworks in which a minibatch of samples is drawn first, which are then grouped into positive and negative pairs. Allowing access to the whole dataset to form pairs gives a better approximation of the respective distributions [12pt]{minimal} $$p$$ p y ∣ x and [12pt]{minimal} $$q$$ q y ∣ x , and considerably improves the quality of the obtained embeddings. If the dataset is sufficiently small to fit into the memory, CEBRA can be optimized with batch gradient descent—that is, using the whole dataset at each optimizer step. Comparing the loss value—at both absolute and relative values across models at the same point in training time— can be used to determine goodness of fit. In practical terms, this means that one can find which hypothesis best fits one’s data, in the case of using CEBRA-Behaviour. Specifically, let us denote the objective in equation ) as L asympt and its approximation in equation with a batch size of n as L n . In the limit of many samples, the objective converges up to a constant, [12pt]{minimal} $${L}_{{}}={{}}_{n }[{L}_{n}- n]$$ L asympt = lim n → ∞ L n − log n (Supplementaty Note and ref. ). The objective also has two trivial solutions: the first is obtained for a constant [12pt]{minimal} $$ = $$ ψ x , y = ψ , which yields a value of L n = log n . This solution can be obtained when the labels are not related to the signal (e.g., with shuffled labels). It is typically not obtained during regular training because the network is initialized randomly, causing the initial embedding points to be randomly distributed in space. If the embedding points are distributed uniformly in space and ϕ is selected such that [12pt]{minimal} $${}=0$$ E φ x , y = 0 , we will also get a value that is approximately L n = log n . This value can be readily estimated by computing [12pt]{minimal} $$ $$ φ u , v for randomly distributed points. The minimizer of equation is also clearly defined as [12pt]{minimal} $${-D}_{{}}(p q)$$ − D KL p ∥ q and depends on the positive and negative distribution. For discovery-driven (time contrastive) learning, this value is impossible to estimate because it would require access to the underlying conditional distribution of the latents. However, for training with predefined positive and negative distributions, this quantity can be again numerically estimated. Interesting values of the loss function when fitting a CEBRA model are therefore 3 [12pt]{minimal} $$-{D}_{{}}(p q) {L}_{n}- n\, 0$$ − D KL p ∥ q ≤ L n − log n ≤ 0 where L n – log n is the goodness of fit (lower is better) of the CEBRA model. Note that the metric is independent of the batch size used for training. Selection of the sampling scheme is CEBRA’s key feature in regard to adapting embedding spaces to different datasets and recording setups. The conditional distributions [12pt]{minimal} $$p$$ p y ∣ x for positive samples and [12pt]{minimal} $$q$$ q y ∣ x for negative samples, as well as the marginal distribution p ( x ) for reference samples, are specified by the user. CEBRA offers a set of predefined sampling techniques but customized variants can be specified to implement additional, domain-specific distributions. This form of training allows the use of context variables to shape the properties of the embedding space, as outlined in the graphical model in Supplementary Note . Through the choice of sampling technique, various use cases can be built into the algorithm. For instance, by forcing positive and negative distributions to sample uniformly across a factor, the model will become invariant to this factor because its inclusion would yield a suboptimal value of the objective function. When considering different sampling mechanisms we distinguish between single- and multi-session datasets: a single-session dataset consists of samples s t associated to one or more context variables c t and/or k t . These context variables allow imposition of the structure on the marginal and conditional distribution used for obtaining the embedding. Multi-session datasets consist of multiple, single-session datasets. The dimension of context variables c t and/or k t must be shared across all sessions whereas the dimension of the signal s t can vary. In such a setting, CEBRA allows learning of a shared embedding space for signals from all sessions. For single-session datasets, sampling is done in two steps. First, based on a specified ‘index’ (the user-defined context variable c t and/or k t ), locations t are sampled for reference, positive and negative samples. The algorithm differentiates between categorical ( k ) and continuous ( c ) variables for this purpose. In the simplest case, negative sampling ( q ) returns a random sample from the empirical distribution by returning a randomly chosen index from the dataset. Optionally, with a categorical context variable [12pt]{minimal} $${k}_{t} [K]$$ k t ∈ K , negative sampling can be performed to approximate a uniform distribution of samples over this context variable. If this is performed for both negative and positive samples, the resulting embedding will become invariant with respect to the variable k t . Sampling is performed in this case by computing the cumulative histogram of k t and sampling uniformly over k using the transformation theorem for probability densities. For positive pairs, different options exist based on the availability of continuous and discrete context variables. For a discrete context variable [12pt]{minimal} $${k}_{t} [K]$$ k t ∈ K with K possible values, sampling from the conditional distribution is done by filtering the whole dataset for the value k t of the reference sample, and uniformly selecting a positive sample with the same value. For a continuous context variable c t we can use a set of time offsets Δ to specify the distribution. Given the time offsets, the empirical distribution [12pt]{minimal} $$P({{}}_{t+ }\,{}\,{{}}_{t})$$ P c t + τ ∣ c t for a particular choice of [12pt]{minimal} $$ $$ τ ∈ Δ can be computed from the dataset: we build up a set [12pt]{minimal} $$D=\{t [T], :{{}}_{t+ }-{{}}_{t}\}$$ D = { t ∈ T , τ ∈ Δ : c t + τ − c t } , sample a d uniformly from D and obtain the sample that is closest to the reference sample’s context variable modified by this distance ( c + d ) from the dataset. It is possible to combine a continuous variable c t with a categorical variable k t for mixed sampling. On top of the continual sampling step above, it is ensured that both samples in the positive pair share the same value of k t . It is crucial that the context samples c and the norm used in the algorithm match in some way; for simple context variables with predictable conditional distributions (for example, a 1D or 2D position of a moving animal, which can most probably be well described by a Gaussian conditional distribution based on the previous sample), the positive sample distribution can also be specified directly, for example, as a normal distribution centred around c t . An additional alternative is to use CEBRA also to preprocess the original context samples c and use the embedded context samples with the metric used for CEBRA training. This scheme is especially useful for higher-dimensional behavioural data, or even for complex inputs such as video. We next consider the multi-session case in which signals [12pt]{minimal} $${{}}_{t}^{(i)} {{}}^{{n}_{i}}$$ s t i ∈ R n i come from N different sessions [12pt]{minimal} $$i [N]$$ i ∈ N with session-dependent dimensionality n i . Importantly, the corresponding continuous context variables [12pt]{minimal} $${{}}_{t}^{(i)} {{}}^{m}$$ c t i ∈ R m share the same dimensionality m , which makes it possible to relate samples across sessions. The multi-session setup is similar to mixed-session sampling (if we treat the session ID as a categorical variable [12pt]{minimal} $${k}_{t}^{(i)}:\,=i$$ k t ( i ) : = i for all time steps t in session i ). The conditional distribution for both negative and positive pairs is uniformly sampled across sessions, irrespective of session length. Multi-session mixed or discrete sampling can be implemented analogously. CEBRA is sufficiently flexible to incorporate more specialized sampling schemes beyond those outlined above. For instance, mixed single-session sampling could be extended additionally to incorporate a dimension to which the algorithm should become invariant; this would add an additional step of uniform sampling with regard to this desired discrete variable (for example, via ancestral sampling). Depending on the exact application, the contrastive learning step can be performed by explicitly including or excluding the context variable. The reference sample x can contain information from the signal s t , but also from the experimental conditions, behavioural recordings or other context variables. The positive and negative samples y are set to the signal variable s t . Identifiability describes the property of an algorithm to give a consistent estimate for the model parameters given that the data distributions match. We here apply the relaxed notion of linear identifiability that was previously discussed and used , . After training two encoder models f and f ′, the models are linearly identifiable if f ( x ) = L f( x ), where L is a linear map. When applying CEBRA, three cases are of potential interest. (1) When applying discovery-driven CEBRA, will two models estimated on comparable experimental data agree in their inferred representation? (2) Under which assumptions about the data will we be able to discover the true latent distribution? (3) In the hypothesis-driven or hybrid application of CEBRA, is the algorithm guaranteed to give a meaningful (nonstandard) latent space when we can find signal within the data? For the first case, we note that the CEBRA objective with a cosine similarity metric follows the canonical discriminative form for which Roeder et al. showed linear identifiability: for sufficiently diverse datasets, two CEBRA models trained to convergence on the same dataset will be consistent up to linear transformations. Note that the consistency of CEBRA is independent of the exact data distribution: it is merely required that the embeddings of reference samples across multiple positive pairs, and the embeddings of negative samples across multiple negative pairs, vary in sufficiently numerous linearly independent directions. Alternatively, we can derive linear identifiability from assumptions about data distribution: if the ground truth latents are sufficiently diverse (that is, vary in all latent directions under distributions p and q ), and the model is sufficiently parameterized to fit the data, we will also obtain consistency up to a linear transformation. See Supplementary Note for a full formal discussion and proof. For the second case, additional assumptions are required regarding the exact form of data-generating distributions. Within the scope of this work we consider ground truth latents distributed on the hypersphere or Euclidean space. The metric then needs to match assumptions about the variation of ground truth latents over time. In discovery-driven CEBRA, using the dot product as the similarity measure then encodes the assumption that latents vary according to a von Mises–Fisher distribution whereas the (negative) mean squared error encodes an assumption that latents vary according to a normal distribution. More broadly, if we assume that the latents have a uniform marginal distribution (which can be ensured by designing unbiased experiments), the similarity measure should be chosen as the log-likelihood of conditional distribution over time. In this case, CEBRA identifies the latents up to an affine transformation (in the most general case). This result also explains the empirically high performance of CEBRA for decoding applications: if trained for decoding (using the variable to decode for informing the conditional distribution), it is trivial to select matching conditional distributions because both quantities are directly selected by the user. CEBRA then ‘identifies’ the context variable up to an affine transformation. For the third case, we are interested in hypothesis-testing capabilities. We can show that if a mapping exists between the context variable and the signal space, CEBRA will recover this relationship and yield a meaningful embedding, which is also decodable. However, if such a mapping does not exist we can show that CEBRA will not learn a structured embedding. We chose X = Y as the neural signal, with varying levels of recorded neurons and channels based on the dataset. We used three types of encoder model based on the required receptive field: a receptive field of one sample was used for the synthetic dataset experiments (Fig. ) and a receptive field of ten samples in all other experiments (rat, monkey, mouse) except for the Neuropixels dataset, in which a receptive field of 40 samples was used due to the fourfold higher sampling rate of the dataset. All feature encoders were parameterized by the number of neurons (input dimension), a hidden dimension used to control model size and capacity, as well as by their output (embedding) dimension. For the model with the receptive field of one, a four-layer MLP was used. The first and second layers map their respective inputs to the hidden dimension whereas the third introduces a bottleneck and maps to half the hidden dimension. The final layer maps to the requested output dimension. For the model with a receptive field of ten, a convolutional network with five time-convolutional layers was used. The first layer had a kernel size of two, and the next three had a kernel size of three and used skip connections. The final layer had a kernel size of three and mapped hidden dimensions to the output dimension. For the model with receptive field 40, we first preprocessed the signal by concatenating a 2× downsampled version of the signal with a learnable downsample operation implemented as a convolutional layer with kernel size four and stride two, directly followed (without activation function between) by another convolutional layer with kernel size three and stride two. After these first layers, the signal was subsampled by a factor of four. Afterwards, similar to the receptive field ten model, we applied three layers with kernel size three and skip connections and a final layer with kernel size three. In all models, Gaussian error linear unit activation functions were applied after each layer except the last. The feature vector was normalized after the last layer unless a mean squared error-based similarity metric was used (as shown in Extended Data Fig. ). Our implementation of the InfoNCE criterion received a minibatch (or the full dataset) of size n × d for each of the reference, positive and negative samples. n dot-product similarities were computed between reference and positive samples and n × n dot-product similarities were computed between reference and negative samples. Similarities were scaled with the inverse of the temperature parameter τ : from torch import einsum, logsumexp, no_grad def info_nce(ref, pos, neg, tau = 1.0): pos_dist = einsum(“nd,nd–>n”, ref, pos)/tauneg_dist = einsum(“nd,md–>nm”, ref, neg)/tauwith no_grad:c, _ = neg_dist.max(dim=1)pos_dist = pos_dist – c.detachneg_dist = neg_dist – c.detachpos_loss = –pos_dist.meanneg_loss = logsumexp(neg_dist, dim = 1).meanreturn pos_loss + neg_loss Alternatively, a learnable temperature can be used. For a numerically stable implementation we store the log inverse temperature [12pt]{minimal} $$ =-\, $$ α = − log τ as a parameter of loss function. At each step we scale the distances in loss function with [12pt]{minimal} $$ ( ,\,1/{ }_{ })$$ min exp α , 1 / τ min . The additional parameter τ min is a lower bound on the temperature. The inverse temperature used for scaling the distances in the loss will hence lie in [12pt]{minimal} $$(0,1/{ }_{ }]$$ ( 0 , 1 / τ min ] . CEBRA model parameters used In the main figures we have used the default parameters ( https://cebra.ai/docs/api.html ) for fitting CEBRA unless otherwise stated in the text (such as dimension, which varied and is noted in figure legends), or below: Synthetic data: model_architecture= ‘offset1-model-mse’, conditional= ‘delta’, delta=0.1, distance= ‘euclidean’, batch_size=512, learning_rate=1e-4. Rat hippocampus neural data: model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Rat behavioural data: model_architecture= ‘offset10-model-mse’, distance= ‘euclidean’, time_offsets=10, batch_size=512. Primate S1 neural data: model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Allen datasets (2P): model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Allen datasets (NP): model_architecture= ‘offset40-model-4x-subsample’, time_offsets=10, batch_size=512. CEBRA API and example usage The Python implementation of CEBRA is written in PyTorch and NumPy and provides an application programming interface (API) that is fully compatible with scikit-learn , a package commonly used for machine learning. This allows the use of scikit-learn tools for hyperparameter selection and downstream processing of the embeddings—for example, decoding. CEBRA can be used as a dropin replacement in existing data pipelines for algorithms such as t -SNE, UMAP, PCA or FastICA. Both CPU and GPU implementations are available. Using the previously introduced notations, suppose we have a dataset containing signals s t , continuous context variables c t and discrete context variables k t for all time steps t , import numpy as npN = 500s = np.zeros((N, 55), dtype = float)k = np.zeros((N,), dtype = int)c = np.zeros((N, 10), dtype = float) along with a second session of data, s2 = np.zeros((N, 75), dtype = float)c2 = np.zeros((N, 10), dtype = float) assert c2.shape[1] == c.shape[1]:note that both the number of samples and the dimension in s ′ does not need to match s . Session alignment leverages the fact that the second dimensions of c and c ′ match. With this dataset in place, different variants of CEBRA can be applied as follows: import cebramodel = cebra.CEBRA(output_dimension=8,num_hidden_units=32,batch_size=1024,learning_rate=3e-4,max_iterations=1000) The training mode to use is determined automatically based on what combination of data is passed to the algorithm: # time contrastive learning model.fit(s) # discrete behaviour contrastive learning model.fit(s, k) # continuous behaviour contrastive learning model.fit(s, c) # mixed behaviour contrastive learning model.fit(s, c, k) # multi-session training model.fit([s, s2], [c, c2]) # adapt to new session model.fit(s, c) model.fit(s2, c2, adapt = True) Because CEBRA is a parametric method training a neural network internally, it is possible to embed new data points after fitting the model: s_test = np.zeros((N, 55), dtype=float)# obtain and plot embeddingz = model.transform(s_test)plt.scatter(z[:, 0], z[:, 1])plt.show Besides this simple-to-use API for end users, our implementation of CEBRA is a modular software library that includes a plugin system, allowing more advanced users to readily add additional model implementations, similarity functions, datasets and data loaders and distributions for sampling positive and negative pairs. In the main figures we have used the default parameters ( https://cebra.ai/docs/api.html ) for fitting CEBRA unless otherwise stated in the text (such as dimension, which varied and is noted in figure legends), or below: Synthetic data: model_architecture= ‘offset1-model-mse’, conditional= ‘delta’, delta=0.1, distance= ‘euclidean’, batch_size=512, learning_rate=1e-4. Rat hippocampus neural data: model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Rat behavioural data: model_architecture= ‘offset10-model-mse’, distance= ‘euclidean’, time_offsets=10, batch_size=512. Primate S1 neural data: model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Allen datasets (2P): model_architecture= ‘offset10-model’, time_offsets=10, batch_size=512. Allen datasets (NP): model_architecture= ‘offset40-model-4x-subsample’, time_offsets=10, batch_size=512. The Python implementation of CEBRA is written in PyTorch and NumPy and provides an application programming interface (API) that is fully compatible with scikit-learn , a package commonly used for machine learning. This allows the use of scikit-learn tools for hyperparameter selection and downstream processing of the embeddings—for example, decoding. CEBRA can be used as a dropin replacement in existing data pipelines for algorithms such as t -SNE, UMAP, PCA or FastICA. Both CPU and GPU implementations are available. Using the previously introduced notations, suppose we have a dataset containing signals s t , continuous context variables c t and discrete context variables k t for all time steps t , import numpy as npN = 500s = np.zeros((N, 55), dtype = float)k = np.zeros((N,), dtype = int)c = np.zeros((N, 10), dtype = float) along with a second session of data, s2 = np.zeros((N, 75), dtype = float)c2 = np.zeros((N, 10), dtype = float) assert c2.shape[1] == c.shape[1]:note that both the number of samples and the dimension in s ′ does not need to match s . Session alignment leverages the fact that the second dimensions of c and c ′ match. With this dataset in place, different variants of CEBRA can be applied as follows: import cebramodel = cebra.CEBRA(output_dimension=8,num_hidden_units=32,batch_size=1024,learning_rate=3e-4,max_iterations=1000) The training mode to use is determined automatically based on what combination of data is passed to the algorithm: # time contrastive learning model.fit(s) # discrete behaviour contrastive learning model.fit(s, k) # continuous behaviour contrastive learning model.fit(s, c) # mixed behaviour contrastive learning model.fit(s, c, k) # multi-session training model.fit([s, s2], [c, c2]) # adapt to new session model.fit(s, c) model.fit(s2, c2, adapt = True) Because CEBRA is a parametric method training a neural network internally, it is possible to embed new data points after fitting the model: s_test = np.zeros((N, 55), dtype=float)# obtain and plot embeddingz = model.transform(s_test)plt.scatter(z[:, 0], z[:, 1])plt.show Besides this simple-to-use API for end users, our implementation of CEBRA is a modular software library that includes a plugin system, allowing more advanced users to readily add additional model implementations, similarity functions, datasets and data loaders and distributions for sampling positive and negative pairs. To measure the consistency of the embeddings we used the R 2 score of linear regression (including an intercept) between embeddings from different subjects (or sessions). Secondly, pi-VAE, which we benchmarked and improved (Extended Data Fig. ), demonstrated a theoretical guarantee that it can reconstruct the true latent space up to an affine transformation. Across runs, we measured the R 2 score of linear regression between embeddings across ten runs of the algorithms, yielding 90 comparisons. These runs were done with the same hyperparameters, model and training setup. For the rat hippocampus data, the numbers of neurons recorded were different across subjects. The behaviour setting was the same: the rats moved along a 1.6-meter-long track and, for analysis, behaviour data were binned into 100 bins of equal size for each direction (leftwards, rightwards). We computed averaged feature vectors for each bin by averaging all normalized CEBRA embeddings for a given bin and renormalized the average to lie on the hypersphere. If a bin did not contain any sample, it was filled by samples from the two adjacent bins. CEBRA was trained with latent dimension three (the minimum) such that it was constrained to lie only on a two-sphere (making this ‘3D’ space equivalent to a 2D Euclidean space). All other methods were trained with two latent dimensions in Euclidean space. Note that n + 1 dimensions of CEBRA are equivalent to n dimensions of other methods that we compared, because the feature space of CEBRA is normalized (that is, the feature vectors are normalized to have unit length). For Allen visual data in which the number of behavioural data points is the same across different sessions (that is, fixed length of video stimuli), we directly computed the R 2 score of linear regression between embeddings from different sessions and modalities. We surveyed three, four, eight, 32, 64 and 128 latent dimensions with CEBRA. To compare the consistency of embeddings between or within the areas considered, we computed intra- and interarea consistency within the same recording modality (2P or NP). Within the same modality we sampled 400 neurons from each area. We trained one CEBRA model per area and computed linear consistency between all pairs of embeddings. For intra-area comparison we sampled an additional 400 disjoint neurons. For each area we trained two CEBRA models on these two sets of neurons and computed their linear consistency. We repeated this process three times. For comparisons across modalities (2P and NP) we sampled 400 neurons from each modality (which are disjoint, as above, because one set was sampled from 2P recordings and the other from NP recordings). We trained a multi-session CEBRA model with one encoder for 2P and one for NP in the same embedding space. For intra-area comparison we computed linear consistency between NP and 2P decoders from the same area. For interarea comparison we computed linear consistency between the NP encoder from one area and the 2P encoder from another and again considered all combinations of areas. We repeated this process three times. For comparison of single- and multi-session training (Extended Data Fig. ) we computed embeddings using encoder models with eight, 16, … , 128 hidden units to vary the model size, and benchmarked eight, 16, … , 128 latent dimensions. Hyperparameters, except for number of optimization steps, were selected according to either validation set decoding R 2 (rat) or accuracy (Allen). Consistency was reported as the point in training at which position decoding error was less than 7 cm for the first rat in the hippocampus dataset, and a decoding accuracy of 60% in the Allen dataset. For single-session training, four embeddings were trained independently on each individual animal whereas for multi-session training the embeddings were trained jointly on all sessions. For multi-session training, the same number of samples was drawn from each session to learn an embedding invariant to the session ID. The consistency versus decoding error trade-off (Extended Data Fig. ) was reported as the average consistency across all 12 comparisons (Extended Data Fig. ) versus average decoding performance across all rats and data splits. pi-VAE parameter selection and modifications to pi-VAE Because the original implementation of pi-VAE used a single time bin spiking rate as an input, we therefore modified their code to allow for larger time bin inputs and found that time window input with a receptive field of ten time bins (250 ms) gave higher consistency across subjects and better preserved the qualitative structure of the embedding (thereby outperforming the results presented by Zhou and Wei ; Extended Data Fig. ). To do this we used the same encoder neural network architecture as that for CEBRA and modified the decoder to a 2D output (we call our modified version conv-pi-VAE). Note, we used this modified pi-VAE for all experiments except for the synthetic setting, for which there is no time dimension and thus the original implementation is sufficient. The original implementation reported a median absolute error of 12 cm for rat 1 (the individual considered most in that work), and our implementation of time-windowed input with ten bins resulted in a median absolute error of 11 cm (Fig. ). For hyperparameters we tested different epochs between 600 (the published value used) and 1,000, and learning rate between 1.0 × 10 −6 and 5.0 × 10 −4 via a grid search. We fixed hyperparameters as those that gave the highest consistency across subjects, which were training epochs of 1,000 and learning rate 2.5 × 10 −4 . All other hyperparameters were retained as in the original implementation . Note that the original paper demonstrated that pi-VAE is fairly robust across different hyperparameters. For decoding (Fig. ) we considered both a simple kNN decoder (that we use for CEBRA) and the computationally more expensive Monte Carlo sampling method originally proposed for pi-VAE . Our implementation of conv-pi-VAE can be found at https://github.com/AdaptiveMotorControlLab/CEBRA . autoLFADS parameter selection AutoLFADS includes a hyperparameter selection and tuning protocol, which we used, and we also used the original implementation ( https://github.com/snel-repo/autolfads-tf2/ , https://github.com/neurallatents/nlb_tools/tree/main/examples/baselines/autolfads ). For the rat hippocampus dataset we chopped the continuous spiking rate (25 ms bin size) into 250-ms-long segments with 225 ms overlap between segments to match the training setup for CEBRA, UMAP, t -SNE and pi-VAE. We used population-based training (PBT) for hyperparameter searches and constrained the search range to default values given in the original script (initial learning rate between 1.0 × 10 −5 and 5.0 × 10 −3 , dropout rate 0–0.6, coordinated dropout rate 0.01–0.70, L2 generator weight between 1.0 × 10 −4 and 1.0, L2 controller weight between 1.0 × 10 −4 and 1.0, KL controller weight between 1.0 × 10 −6 and 1.0 × 10 −4 and KL initial condition weight between 1.0 × 10 −6 and 1.0 × 10 –3 ). The negative log-likelihood metric was used to select the best hyperparameters. Each generation of PBT consisted of 25 training epochs and we trained for a maximum of 5,000 epochs of batch size 100 while executing early stopping after awaiting 50 epochs. The PBT search was done using 20 parallel workers on each rat. UMAP parameter selection For UMAP , following the parameter guide ( umap-learn.readthedocs.io/ ), we focused on tuning the number of neighbours ( n _ neighbors ) and minimum distance ( min _ dist ). The n _ components parameter was fixed to 2 and we used a cosine metric to make a fair comparison with CEBRA, which also used the cosine distance metric for learning. We performed a grid search with 100 total hyperparameter values in the range [2, 200] for n _ neighbors and in the range [0.0001, 0.99] for min _ dist . The highest consistency across runs in the rat hippocampus dataset was achieved with min _ dist of 0.0001 and n _ neighbors of 24. For the other datasets in Extended Data Fig. we used the default value of n _ neighbors as 15 and min _ dist as 0.1. t -SNE parameter selection For t -SNE we used the implementation in openTSNE . We performed a sweep on perplexity in the range [5, 50] and early _ exaggeration in the range [12, 32] following the parameter guide, while fixing n _ components as 2 and used a cosine metric for fair comparison with UMAP and CEBRA. We used PCA initialization to improve the run consistency of t -SNE . The highest consistency across runs in the rat hippocampus dataset was achieved with perplexity of ten and early _ exaggeration of 16.44. For the other datasets in Extended Data Fig. we used the default value for perplexity of 30 and for early _ exaggeration of 12. Because the original implementation of pi-VAE used a single time bin spiking rate as an input, we therefore modified their code to allow for larger time bin inputs and found that time window input with a receptive field of ten time bins (250 ms) gave higher consistency across subjects and better preserved the qualitative structure of the embedding (thereby outperforming the results presented by Zhou and Wei ; Extended Data Fig. ). To do this we used the same encoder neural network architecture as that for CEBRA and modified the decoder to a 2D output (we call our modified version conv-pi-VAE). Note, we used this modified pi-VAE for all experiments except for the synthetic setting, for which there is no time dimension and thus the original implementation is sufficient. The original implementation reported a median absolute error of 12 cm for rat 1 (the individual considered most in that work), and our implementation of time-windowed input with ten bins resulted in a median absolute error of 11 cm (Fig. ). For hyperparameters we tested different epochs between 600 (the published value used) and 1,000, and learning rate between 1.0 × 10 −6 and 5.0 × 10 −4 via a grid search. We fixed hyperparameters as those that gave the highest consistency across subjects, which were training epochs of 1,000 and learning rate 2.5 × 10 −4 . All other hyperparameters were retained as in the original implementation . Note that the original paper demonstrated that pi-VAE is fairly robust across different hyperparameters. For decoding (Fig. ) we considered both a simple kNN decoder (that we use for CEBRA) and the computationally more expensive Monte Carlo sampling method originally proposed for pi-VAE . Our implementation of conv-pi-VAE can be found at https://github.com/AdaptiveMotorControlLab/CEBRA . AutoLFADS includes a hyperparameter selection and tuning protocol, which we used, and we also used the original implementation ( https://github.com/snel-repo/autolfads-tf2/ , https://github.com/neurallatents/nlb_tools/tree/main/examples/baselines/autolfads ). For the rat hippocampus dataset we chopped the continuous spiking rate (25 ms bin size) into 250-ms-long segments with 225 ms overlap between segments to match the training setup for CEBRA, UMAP, t -SNE and pi-VAE. We used population-based training (PBT) for hyperparameter searches and constrained the search range to default values given in the original script (initial learning rate between 1.0 × 10 −5 and 5.0 × 10 −3 , dropout rate 0–0.6, coordinated dropout rate 0.01–0.70, L2 generator weight between 1.0 × 10 −4 and 1.0, L2 controller weight between 1.0 × 10 −4 and 1.0, KL controller weight between 1.0 × 10 −6 and 1.0 × 10 −4 and KL initial condition weight between 1.0 × 10 −6 and 1.0 × 10 –3 ). The negative log-likelihood metric was used to select the best hyperparameters. Each generation of PBT consisted of 25 training epochs and we trained for a maximum of 5,000 epochs of batch size 100 while executing early stopping after awaiting 50 epochs. The PBT search was done using 20 parallel workers on each rat. For UMAP , following the parameter guide ( umap-learn.readthedocs.io/ ), we focused on tuning the number of neighbours ( n _ neighbors ) and minimum distance ( min _ dist ). The n _ components parameter was fixed to 2 and we used a cosine metric to make a fair comparison with CEBRA, which also used the cosine distance metric for learning. We performed a grid search with 100 total hyperparameter values in the range [2, 200] for n _ neighbors and in the range [0.0001, 0.99] for min _ dist . The highest consistency across runs in the rat hippocampus dataset was achieved with min _ dist of 0.0001 and n _ neighbors of 24. For the other datasets in Extended Data Fig. we used the default value of n _ neighbors as 15 and min _ dist as 0.1. -SNE parameter selection For t -SNE we used the implementation in openTSNE . We performed a sweep on perplexity in the range [5, 50] and early _ exaggeration in the range [12, 32] following the parameter guide, while fixing n _ components as 2 and used a cosine metric for fair comparison with UMAP and CEBRA. We used PCA initialization to improve the run consistency of t -SNE . The highest consistency across runs in the rat hippocampus dataset was achieved with perplexity of ten and early _ exaggeration of 16.44. For the other datasets in Extended Data Fig. we used the default value for perplexity of 30 and for early _ exaggeration of 12. We primarily used a simple kNN algorithm, a nonparametric supervised learning method, as a decoding method for CEBRA. We used the implementation in scikit-learn . We used a kNN regressor for continuous value regression and a kNN classifier for discrete label classification. For embeddings obtained with cosine metrics we used cosine distance metrics for kNN, and Euclidean distance metrics for those obtained in Euclidean space. For the rat hippocampus data a kNN regressor, as implemented in scikit-learn , was used to decode position and a kNN classifier to decode direction. The number of neighbours was searched over the range [1, 4, 9, 16, 25] and we used the cosine distance metric. We used the R 2 score of predicted position and direction vector on the validation set as a metric to choose the best n_neighbours parameter. We report the median absolute error for positional decoding on the test set. For pi-VAE, we additionally evaluated decoding quality using the originally proposed decoding method based on Monte Carlo sampling, with the settings from the original article . For autoLFADS, use of their default Ridge regression decoder performed worse than our kNN decoder, which is why we reported all results for the kNN decoder. Note that UMAP, t -SNE and CEBRA-Time were trained using the full dataset without label information when learning the embedding, and we used the above split only for training and cross-validation of the decoder. For direction decoding within the monkey dataset we used a Ridge classifier as a baseline. The regularization hyperparameter was searched over [10 −6 , 10 2 ]. For CEBRA we used a kNN classifier for decoding direction with k searched over the range [1, 2500]. For conv-pi-VAE we searched for the best learning rate over [1.0 × 10 −5 , 1.0 × 10 −3 ]. For position decoding we used Lasso as a baseline. The regularization hyperparameter was searched over [10 −6 , 10 2 ]. For conv-pi-VAE we used 600 epochs and searched for the best learning rates over [5 × 10 −4 , 2.5 × 10 −4 , 0.125 × 10 −4 , 5 × 10 −5 ] via a grid of ( x , y ) space in 1 cm bins for each axis as the sampling process for decoding. For CEBRA we used kNN regression, and the number of neighbours k was again searched over [1, 2500]. For the Allen Institute datasets we performed decoding (frame number or scene classification) for each frame from Video 1. Here we used a kNN classifier with a population vector kNN as a baseline, similar to the decoding of orientation grating performed in ref. . For CEBRA we used the same kNN classifier method as on CEBRA features. In both cases the number of neighbours, k , was searched over a range [1, 100] in an exponential fashion. We used neural data recorded during the first eight repeats as the train set, the ninth repeat for validation in choosing the hyperparameter and the last repeat as the test set to report decoding accuracy. We also used a Gaussian naive Bayes decoder to test linear decoding from the CEBRA model and neural population vector. Here we assumed uniform priors over frame number and searched over the range [10 −10 , 10 3 ] in an exponential manner for the var_smoothing hyperparameter. For layer-specific decoding we used data from excitatory neurons in area VISp: layers 2/3 [Emx1-IRES-Cre, Slc17a7-IRES2-Cre]; layer 4 [Cux2-CreERT2, Rorb-IRES2-Cre, Scnn1a-Tg3-Cre]; and layers 5/6 [Nr5a1-Cre, Rbp4-Cre_KL100, Fezf2-CreER, Tlx3-Cre_PL56, Ntrsr1-cre]. Neural Latents Benchmark We tested CEBRA on the mc-maze 20 ms task from the Neural Latents Benchmark ( https://eval.ai/web/challenges/challenge-page/1256/leaderboard/3183 ). We trained the offset10-model with 48 output dimensions and [128, 256, 512] hidden units, as presented throughout the paper. We trained, in total, 48 models by additionally varying the temperature in [0.0001, 0.004] and time offsets from {1, 2}. We performed smoothing of input neural data using a Gaussian kernel with 50 ms s.d. Lastly, we took 45 embeddings from the trained models picked by the validation score, aligned the embeddings (using the Procrustes method ) and averaged them. We tested CEBRA on the mc-maze 20 ms task from the Neural Latents Benchmark ( https://eval.ai/web/challenges/challenge-page/1256/leaderboard/3183 ). We trained the offset10-model with 48 output dimensions and [128, 256, 512] hidden units, as presented throughout the paper. We trained, in total, 48 models by additionally varying the temperature in [0.0001, 0.004] and time offsets from {1, 2}. We performed smoothing of input neural data using a Gaussian kernel with 50 ms s.d. Lastly, we took 45 embeddings from the trained models picked by the validation score, aligned the embeddings (using the Procrustes method ) and averaged them. For the persistent cohomology analysis we utilized ripser.py . For the hippocampus dataset we used 1,000 randomly sampled points from CEBRA-Behaviour trained with temperature 1, time offset 10 and minibatch size 512 for 10,000 training steps on the full dataset and then analysed up to 2D cohomology. Maximum distance considered for filtration was set to infinity. To determine the number of cocycles in each cohomology dimension with a significant lifespan we trained 500 CEBRA embeddings with shuffled labels, similar to the approach taken in ref. . We took the maximum lifespan of each dimension across these 500 runs as a threshold to determine robust Betti numbers, then surveyed the Betti numbers of CEBRA embeddings across three, eight, 16, 32 and 64 latent dimensions. Next we used DREiMac to obtain topology-preserving circular coordinates (radial angle) of the first cocycle (H 1 ) from the persistent cohomology analysis. Similar to above, we used 1,000 randomly sampled points from the CEBRA-Behaviour models of embedding dimensions 3, 8, 16, 32 and 64. High-dimensional inputs, such as videos, need further preprocessing for effective use with CEBRA. First we used the recently presented DINO model to embed video frames into a 768D feature space. Specifically we used the pretrained ViT/8 vision transformer model, which was trained by a self-supervised learning objective on the ImageNet database. This model is particularly well suited for video analysis and among the state-of-the-art models available for embedding natural images into a space appropriate for a kNN search , a desired property when making the dataset compatible with CEBRA. We obtained a normalized feature vector for each video frame, which was then used as the continuous behaviour variable for all further CEBRA experiments. For scene labels, three individuals labelled each video frame using eight candidate descriptive labels allowing multilabel classes. We took the majority vote of these three individuals to determine the label of each frame. In the case of multilabels we considered this as a new class label. The above procedure resulted in ten classes of frame annotation. Further information on research design is available in the linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-023-06031-6. Supplementary Information Attached in a single PDF are Supplementary Notes 1 and 2, which provide extended discussions on identifiability and the theoretical guarantees of CEBRA, respectively, and Tables 1–6, which provide statistical support to the conclusions drawn in the main manuscript. Reporting Summary Supplementary Video 1 Corresponding to Fig. 2d. CEBRA-Behaviour trained with position and direction on rat 1. Video is in 2× real time. Supplementary Video 2 Corresponding to Fig. 5b. The left-hand panels show example calcium traces from two-photon imaging (top) and spikes from Neuropixels recording (bottom) of primary visual cortex while the video was shown to mice (here, we randomly picked neurons to visualize the pseudomouse). The centre panel shows an embedding space constructed by jointly training a CEBRA-Behaviour model with two-photon and Neuropixels recordings using DINO frame features as labels. The trace is an embedding of a held-out test repeat from the Neuropixels recording. The colour map indicates the frame number of the 30-s-long video (30 Hz). The final panels show true video (top) and predicted frame sequence (bottom) using a kNN decoder on CEBRA-Behaviour embedding from the test set. Video is in real time.
Cortical suspensory button fixation has superior biomechanical properties to knotless anchor suture in anterior cruciate ligament repair: a biomechanical study	e4f237b3-6e67-4659-bb04-9b683066ce45	10172190	Suturing[mh]	Anterior cruciate ligament (ACL) tears are the most common knee injury associated with sports. Annually, between 100,000 and 200,000 ACL injuries occur in the United States, with football, skiing, and gymnastics being the most prevalent sources of injuries . Historically, ACL injuries were treated with primary ACL repair. More recent research has shown that ACL reconstruction is more successful than primary ACL repair. ACL reconstruction is now the gold standard for ACL injury treatment – . In 1895, Mayo Robson reported the first recorded case of anterior cruciate ligament (ACL) repair in a 41-year-old male who underwent open primary repair of bilateral ACL tears at the femoral attachment site. In 1976, Feagin and Curl conducted a study on athletes who underwent open primary ACL repair, and found that most patients were able to return to sports. However, after a five-year follow-up, the failure rate was high with 94% of patients experiencing instability, 53% experiencing reinjury, and 34% requiring a second surgical treatment. Sherman et al. introduced a four-level classification system in 1991. Type 1 involves a complete tear of the anterior cruciate ligament (ACL) from the femoral attachment, with no remaining connection to the femur. Type 2 includes injuries where less than 20% of ligaments remain attached to the femoral attachment. Type 3 tears occur when less than 33% of ligaments are connected to the femoral attachment. Type 4 refers to a mid-substance tear. With a reported follow-up of 61 months following open primary ACL repair, patients over the age of 22 with a ski injury, a Type 1 tear, good tissue quality, and a low-grade pivot had a favorable result. Compared to primary ACL repair, there are several disadvantages of ACL reconstruction including the loss of the native knee kinematics, loss of proprioceptive sensation, inability to prevent osteoarthritis, and an increasing difficulty in subsequent surgeries . Arthroscopic surgery has increased in popularity over the past decade with advancements in surgical equipment. Arthroscopic ACL primary repair has received increased attention and has been reported to have favorable surgical outcomes in the short- and medium-term, particularly in patients with Sherman Type 1 ACL tear , . Each orthopedic surgeon employs a unique approach and suture equipment when doing arthroscopic primary ACL repair , – . Cortical suspensory button and knotless anchor suture are the most frequently used implants. The cortical suspensory button has been employed by certain surgeons to reestablish the connection between a ruptured ligament and the femoral footprint , , . Heusdens et al. have recently reported on the results of a two-year follow-up investigation of anterior cruciate ligament (ACL) reconstruction utilizing the cortical suspensory button, and have identified substantial enhancement in clinical outcomes . Meanwhile, some surgeons also utilize the knotless suture anchor , , , and have found promising results in their procedures , . However, there are no biomechanical comparisons between an ACL repair with a cortical suspensory button and a knotless anchor suture to evaluate whether they are stronger and robust enough to resist the forces applied on the knee during postoperative rehabilitation. To date, there has been no biomechanical study completed on human cadaveric knees. There is only one study using fresh frozen porcine knees . The objective of this study was to determine the load-to-failure (N), stiffness (N/mm), gap formation (mm), and failure mechanism for ACL repair with both cortical suspensory button and knotless anchor suture repair techniques. Sample size calculation Calculations of sample size are based on results from research performed on porcine ligaments . Epitools (Ausvet, Australia) was used to calculate a total sample size of 6 and a sample size of 3 per group with confidence level of 0.95, and power = 0.8. To avoid missing or incomplete data, a 33% increase in sample size was added to the total sample size of 8 or the sample size of 4 per group. Inclusions The study was approved by the Chulalongkorn University Faculty of Medicine's Institutional Review Board (IRB No. 632/64). Informed consent was obtained from all subjects and/or their legal guardians for using cadaveric samples used in the study. Eight paired cadaveric knees were taken from four Thiel’s embalmed cadavers – . Specimens with altered knee anatomy from any pathology were excluded. Cadaver demographic data, such as age, weight, height, and gender, has been obtained from the Chula soft cadaver surgical training center's registry. The present study was conducted in accordance with the tenets of the Declaration of Helsinki, 1975, as revised in 2013. Study procedures Eight Thiel’s embalmed paired cadaveric knees were prepared – . The femur and tibia were cut fifteen centimeters from the joint line. The collateral and posterior cruciate ligaments were peeled off, leaving only the anterior cruciate ligament connected between femur and tibia. A type 1 ACL tear was created in each cadaveric knee. An ACL is peeled off the femoral attachment (Fig. A) and is repaired using a cortical suspensory button (CSB) or knotless anchor suture (KAS). Repair with cortical suspensory button technique (CSB) The ACL was sutured using #2 HiFi suture (CONMED, Utica, NY) with a single loop stitch. After stitching the ACL, a 4.5-mm guide-reamer was used to create the ACL femoral footprint. All sutures were threaded through the femoral footprint holes. The cortical suspensory button (XO button, CONMED, Utica, NY) was placed at the lateral femoral cortex with a surgical knot and five half-hitches with the ACL remnant tension in the semi-extension position (Fig. B). Repair with knotless anchor suture technique (KAS) To repair the ACL using the knotless anchor suture technique, stitch ACL was performed using the cortical suspensory button technique was completed and followed by drill and tap of 4.5 mm × 20 mm holes in the femoral footprint of the ACL. Threading of all the sutured limbs through the eyelet of the knotless anchor suture (4.5-mm PopLok, CONMED, Utica, NY) and insertion of the knotless anchor suture into the prepared ACL femoral footprint with the ACL remnant tension in the semi-extension position was performed. The end of the sutures were then cut-off (Fig. C). Model for testing Biomechanical testing was performed on specimens using a mechanical testing machine (E10000, Instron, Canton, MA) with the tibia attached to the base stationary portion and the femur attached to a servohydrolic testing system in a 180-degree knee-upright position (semi-extension) (Fig. ). While exerting force on the femur, the Tibia bone remained at rest. The Servohydrolic testing system (E10000, Instron, Canton, MA) replicates cyclic loading in a position-controlled mode. The testing started with 500 cycles at 0.75 Hz and a peak elongation of 1 mm. Determination of the gap formation (plastic deformity) after 500 cycles was completed by increasing peak elongation from 1 to 3 mm. Peak elongation was then increased to 5 mm (1500 cyclic loading cycles) and the gap formation was monitored every 500 cycles throughout the cyclic loading. Traction was applied at a rate of 50 mm/min until failure. The load-to-failure and stiffness between the pull-to-failure and primary modes of failure in both groups were measured. Outcome measurement Biomechanical evaluation of the ACL repair techniques using a cortical suspensory button and a knotless anchor suture. Key outcomes measured were load-to-failure (N), stiffness (N/mm), gap formation (mm). Statistical analysis Statistical analysis was performed using SPSS 22.0 (IBM, USA) for Windows. The student's t-test was used to compare the load-to-failure (N), stiffness (N/mm), and gap formation (mm) values between the two groups. The 95% confidence interval was also calculated for both groups. The significance level was set at p value ≤ 0.05. Calculations of sample size are based on results from research performed on porcine ligaments . Epitools (Ausvet, Australia) was used to calculate a total sample size of 6 and a sample size of 3 per group with confidence level of 0.95, and power = 0.8. To avoid missing or incomplete data, a 33% increase in sample size was added to the total sample size of 8 or the sample size of 4 per group. The study was approved by the Chulalongkorn University Faculty of Medicine's Institutional Review Board (IRB No. 632/64). Informed consent was obtained from all subjects and/or their legal guardians for using cadaveric samples used in the study. Eight paired cadaveric knees were taken from four Thiel’s embalmed cadavers – . Specimens with altered knee anatomy from any pathology were excluded. Cadaver demographic data, such as age, weight, height, and gender, has been obtained from the Chula soft cadaver surgical training center's registry. The present study was conducted in accordance with the tenets of the Declaration of Helsinki, 1975, as revised in 2013. Eight Thiel’s embalmed paired cadaveric knees were prepared – . The femur and tibia were cut fifteen centimeters from the joint line. The collateral and posterior cruciate ligaments were peeled off, leaving only the anterior cruciate ligament connected between femur and tibia. A type 1 ACL tear was created in each cadaveric knee. An ACL is peeled off the femoral attachment (Fig. A) and is repaired using a cortical suspensory button (CSB) or knotless anchor suture (KAS). The ACL was sutured using #2 HiFi suture (CONMED, Utica, NY) with a single loop stitch. After stitching the ACL, a 4.5-mm guide-reamer was used to create the ACL femoral footprint. All sutures were threaded through the femoral footprint holes. The cortical suspensory button (XO button, CONMED, Utica, NY) was placed at the lateral femoral cortex with a surgical knot and five half-hitches with the ACL remnant tension in the semi-extension position (Fig. B). To repair the ACL using the knotless anchor suture technique, stitch ACL was performed using the cortical suspensory button technique was completed and followed by drill and tap of 4.5 mm × 20 mm holes in the femoral footprint of the ACL. Threading of all the sutured limbs through the eyelet of the knotless anchor suture (4.5-mm PopLok, CONMED, Utica, NY) and insertion of the knotless anchor suture into the prepared ACL femoral footprint with the ACL remnant tension in the semi-extension position was performed. The end of the sutures were then cut-off (Fig. C). Biomechanical testing was performed on specimens using a mechanical testing machine (E10000, Instron, Canton, MA) with the tibia attached to the base stationary portion and the femur attached to a servohydrolic testing system in a 180-degree knee-upright position (semi-extension) (Fig. ). While exerting force on the femur, the Tibia bone remained at rest. The Servohydrolic testing system (E10000, Instron, Canton, MA) replicates cyclic loading in a position-controlled mode. The testing started with 500 cycles at 0.75 Hz and a peak elongation of 1 mm. Determination of the gap formation (plastic deformity) after 500 cycles was completed by increasing peak elongation from 1 to 3 mm. Peak elongation was then increased to 5 mm (1500 cyclic loading cycles) and the gap formation was monitored every 500 cycles throughout the cyclic loading. Traction was applied at a rate of 50 mm/min until failure. The load-to-failure and stiffness between the pull-to-failure and primary modes of failure in both groups were measured. Biomechanical evaluation of the ACL repair techniques using a cortical suspensory button and a knotless anchor suture. Key outcomes measured were load-to-failure (N), stiffness (N/mm), gap formation (mm). Statistical analysis was performed using SPSS 22.0 (IBM, USA) for Windows. The student's t-test was used to compare the load-to-failure (N), stiffness (N/mm), and gap formation (mm) values between the two groups. The 95% confidence interval was also calculated for both groups. The significance level was set at p value ≤ 0.05. Eight of the specimens, one knee from each cadaver in each group, came from four cadavers. The cadavers had an average age of 68.00 ± 17.40 years, average body weight of 65.00 ± 10.80 kg, and average height of 171.25 ± 12.50 cm. The specimens were 3 males and 1 female. The details of the demographic data are shown in Table . At 1-, 3-, and 5-mm peak elongation, the mean gap formation in the CSB group was 0.96 ± 0.18 mm, 2.14 ± 0.71 mm, and 3.92 ± 0.41 mm, respectively. In comparison, the mean gap formation of KAS group at 1, 3, and 5 mm was 1.03 ± 1.09 mm, 2.51 ± 0.75 mm, and 4.41 ± 0.71 mm, respectively. There is no significant difference between the CSB and KAS groups in the cyclic data at the end of each cycle up to 5 mm (Table ). During cycle loading, none of the specimens failed. As a result, all of the specimens were put through a final pull-to-failure test. The CSB group had a mean load-to-failure and stiffness of 212.96 ± 54.57 N and 34.83 ± 9.40 N/mm, respectively. The mean load-to-failure and stiffness of the KAS group was 44.57 ± 20.80 N and 28.76 ± 14.48 N/mm. There was a significant difference in load-to-failure between the CSB and KAS groups ( p value < 0.01) (Table ). The entire CSB group failed due to knot slippage at the button. Three specimens of the KAS group failed due to suture slippage from the anchor. Another failed due to a mid-substance ACL tear (Table ). A published meta-analysis found minimal complications and high functional scores following ACL repair, but it is limited by short follow-up and a significant risk of selection and publication bias . Some studies have also found that adolescents have a higher risk of re-rupture due to high activity and an early return to sports , . We performed the study to determine load-to-failure, the stiffness, gap formation, and failure mechanism for ACL repair with cortical suspensory button and knotless anchor suture. With values of 212.96 ± 54.57 N, the CSB group had the highest load-to-failure. The stiffness of the CSB group was 34.83 ± 9.40 N/mm. The KAS group’s mean load-to-failure and stiffness were 44.57 ± 20.80 N and 28.76 ± 14.48 N/mm, respectively. Between the CSB and KAS groups, there was a significant difference in the mean load-to-failure ( p value < 0.01). Bachmaier et al. published a biomechanical study on fresh frozen porcine knees showing that adjustable single-cinch cortex button fixation had the highest ultimate strength when compared to knotless anchor suture, double cinch-fixed loop cortical button, and single cinch-fixed loop cortical button fixation . According to our findings in human cadaveric knees, most of construct failures were caused by knot slippage at the button and suture slippage from the anchor. It is possible to sew the ACL with 1 or 2 loops because failure is not related to the repair site. The present biomechanical found no significant difference in gap formation after cyclic loading between CSB and KAS groups. Although the cadavers used in our study were of advanced age and concerns regarding bone quality were present, our experimental results revealed that none of the specimens failed due to bone breakage or anchor pull-out. This implies that the bone quality was still adequate. Morrison analyzed ACL loads during activities of daily living and found that normal level walking created 169 N of force, while descending stairs generated 445 N of force due to the activation of the knee extensor mechanism. Ascending stairs, on the other hand, produced forces of less than 100 N – . Based on the findings of the present study, the load-to-failure values for the cortical suspensory button and knotless anchor suture femoral fixation were found to be 212.96 ± 54.57 N and 44.57 ± 20.80 N, respectively. According to a biomechanical study of ACL repair with and without internal brace augmentation by Massey et al., ACL repair with internal brace augmentation had a load-to-failure of 693 ± 248 N and a load-to-failure of 279 ± 91 N without augmentation . Kuptniratsaikul et al. , reported a surgical technique to augment the ACL with multiple high strength sutures which is comparable to internal bracing with suture tape. As a result, if the ACL is repaired with femoral fixation using a cortical suspensory button or knotless anchor suture, reinforcement with synthetic sutures or protection with an internal brace is recommended. Heusdens et al. reported the 2-year follow-up results of a novel technique for repairing acute, proximal ACL tears using a cortical suspensory button and suture tape augmentation. The study included 42 patients with good ACL tissue quality and excluded those with poor tissue quality, retracted ACL remnants, or multiple ligament injuries. The results showed significant improvements in the Knee Injury and Osteoarthritis Outcome Score (KOOS), the Visual Analogue Pain Scale, and the Veterans RAND 12-Item Health Survey physical score, with meaningful changes in the KOOS sport and recreation subscale. However, the Marx activity scale decreased significantly, and two patients (4.8%) reported ACL rupture. Jonkergouw et al. examined the outcomes of arthroscopic primary repair using knotless suture anchor of proximal ACL tears in a cohort of 56 patients, followed up for a minimum of 3.2 years. Comparing internal brace (27 patients) and without internal brace (29 patients), they found that arthroscopic primary repair using knotless suture anchor with or without internal brace resulted in good objective and subjective outcomes and similar outcomes. Vermeijden et al. reported the study of the same cohort, which aimed to compare the extent to which patients forget about their operative knee joint following arthroscopic primary repair versus reconstruction of the ACL. Patients who underwent primary repair reported less daily awareness of their operated knee compared to those who underwent reconstruction. These findings were more significant in patients who were older than 30 years, male, and had a body mass index greater than 25. It is currently believed that in the short-term, the outcomes of ACL repair are similar to those of repair or reconstruction, regardless of the type of repair technique used. However, there is limited evidence regarding long-term outcomes and complications. All of the examples in CSB failed because the knot slipped on the femoral site. This means that the construction of the knot is essential if the cortical button is chosen for ACL repair. On the other hand, KAS placed more reliance on the stability of the attachment between the suture and the anchor. However, neither method is strong enough for everyday tasks. Therefore, it is advised to wear external support for the first few weeks following the operation; alternatively, an internal brace can be performed in conjunction with ACL repair. Limitations This study has several limitations. First, the loads were pulled vertically along the longitudinal axis, resembling the worst-case scenario rather than anterior translation or pivot-shifting. Second, Thiel's embalmed cadavers were used in this study, rather than fresh frozen cadavers, which have the same elasticity, color, and flexibility as in vivo ligaments. Studies have shown that the Thiel embalming method is effective for preserving ligaments for research purposes , . Third, our cadavers' average had wide range, with an average age of 68.00 ± 17.40 years which does not properly represent the younger population for whom ACL repair operations are commonly performed, and the quality of the bones and ligaments may deteriorate as a result of age. The results of our study suggest that despite concerns about the bone quality of the cadavers used, the absence of failures due to bone breakage or anchor pull-out implies that the bone quality was still adequate. This finding has important implications for the use of cadaveric specimens in biomechanical studies, particularly in cases where concerns about bone quality may limit their use. However, further research is needed to confirm these findings and explore the potential impact of bone quality on biomechanical outcomes in other contexts. Fourth, this study had a small sample size which limited finding significant differences. Finally, only one manufacturer (CONMED, Utica, NY) of knotless anchor suture and cortical suspensory button was tested in this study. This study has several limitations. First, the loads were pulled vertically along the longitudinal axis, resembling the worst-case scenario rather than anterior translation or pivot-shifting. Second, Thiel's embalmed cadavers were used in this study, rather than fresh frozen cadavers, which have the same elasticity, color, and flexibility as in vivo ligaments. Studies have shown that the Thiel embalming method is effective for preserving ligaments for research purposes , . Third, our cadavers' average had wide range, with an average age of 68.00 ± 17.40 years which does not properly represent the younger population for whom ACL repair operations are commonly performed, and the quality of the bones and ligaments may deteriorate as a result of age. The results of our study suggest that despite concerns about the bone quality of the cadavers used, the absence of failures due to bone breakage or anchor pull-out implies that the bone quality was still adequate. This finding has important implications for the use of cadaveric specimens in biomechanical studies, particularly in cases where concerns about bone quality may limit their use. However, further research is needed to confirm these findings and explore the potential impact of bone quality on biomechanical outcomes in other contexts. Fourth, this study had a small sample size which limited finding significant differences. Finally, only one manufacturer (CONMED, Utica, NY) of knotless anchor suture and cortical suspensory button was tested in this study. This study showed a higher load-to-failure for the ACL repair with cortical suspensory button compared to ACL repair with knotless anchor sutures. However, the load-to-failure in both cortical suspensory button and knotless anchor suture are below a regular daily activity load. An internal brace or external support is recommended during rehabilitation.
Does Consulting an Occupational Medicine Specialist Decrease Time to Return to Work Among Total Knee Arthroplasty Patients? A 12-Month Prospective Multicenter Cohort Study	13489bbb-434b-4986-bb8e-715f78af3cab	10172284	Preventive Medicine[mh]	Worldwide there is a steep rising demand for total knee arthroplasty (TKA), especially among patients of working age. By 2030–2035 the majority of TKA patients in the US and UK will already be of working age . Return to work (RTW) rates among these TKA patients vary between 40 and 98% with a mean time to return to work between 8 and 17 weeks . Time to RTW in TKA patients is often retrospectively measured and therefore prone to recall bias . In the most recent prospective cohort study in the Netherlands only 24% of TKA patients returned to work completely at 3 months, which was 51% at 6 and 71% at 12 months . These percentages are in contrast with orthopedic guidelines advising RTW within 3 months, starting gradually if needed . Moreover patients who receive TKA have the greatest productivity and income loss when compared to other types of common surgery . To decrease time to RTW among these TKA patients, attention should be paid to the beneficial and hindering factors for RTW within health care as well as occupational health. Known prognostic factors effecting RTW are patient characteristics, such as age, gender, Body Mass Index (BMI) and physical function score, as well as work-related characteristics, such as sense of urgency to RTW, having a handicap accessible workplace, preoperative sick leave, and having knee-straining work . Active referral by the orthopedic surgeon to an occupational health expert is expected to enhance RTW . The occupational health experts in the Netherlands are the Occupational Medicine Specialists (OMS) who are physicians with four years post-graduate training in Occupational Medicine. Every employee has direct access to an OMS, on account of the employer. Self-employed patients can consult an OMS on their own account, though this consultation is not familiar to patients and health care professionals and thereby not frequently used. By their specialized training the OMS has insight into the patient’s work demands in relation to the patient’s work ability. Subsequently the OMS can advise and support the RTW process, for instance by advising adjustment of working hours, working tasks (modified duties) or workplace adaptations using ergonomic principles. Patients who have limited access to these kinds of work adjustments and have difficulty performing work-related knee-straining activities may also be referred by their OMS to work rehabilitation . Moreover, an OMS is able to discuss and stimulate the sense of urgency to RTW given the value of work in life . Currently however, no evidence is available on whether time to RTW after TKA can be decreased by consulting an OMS. The aim of this study is to investigate whether TKA patients who consult an OMS within 3 months after surgery, return to work sooner than patients who do not consult an OMS. Study Design and Population A multi-center prospective cohort study among TKA patients was performed . Patients were included from nine surgeons, working in seven hospitals, in five Dutch regions, to minimize selection bias. The hospitals varied from general hospitals, large teaching hospitals to tertiary university hospitals. Inclusion criteria were (1) patients undergoing TKA between June 2014 and March 2018, (2) aged 18 to 65 years (working age), (3) having a paid job, (4) self-reported intend to RTW after surgery and (5) provided information about consulting an OMS within 3 months after TKA or not. Data were collected before TKA and at 3, 6 and 12 months after surgery using a self-report questionnaire in Dutch. Patients could choose a paper or electronic version of the questionnaire to avoid selection bias based on patients’ computer literacy. At each measurement, non-responding patients were reminded up to two times after 2 weeks. Patients who were willing to participate but missed the preoperative measurement due to logistical reasons, were included in the follow-up measurements. Patient Characteristics Patient characteristics were registered, such as date of birth, gender, body height and body mass. The latter two were used to calculate BMI. Patients were asked whether they had other diseases that were limiting their activities at work, with three categories: (1) No, (2) Yes, one disease that is limiting my activities at work, and (3) Yes, more than one disease that is limiting my activities at work. These categories were dichotomized into either ‘no’ (No) or ‘one or more other disease(s) that limits my activities at work’ (Yes), which was defined as comorbidity. Knee Injury and Osteoarthritis Outcome Score (KOOS) subscales on pain, symptoms and quality of live were filled out by the patients. All KOOS subscales are validated in Dutch and range from 0, representing extreme knee problems, to 100 representing no knee problems . The Work Osteoarthritis or joint-Replacement Questionnaire (WORQ) was used for the work-related physical functioning score, and is also validated in Dutch . The WORQ score consists of 13 items on work-related knee-straining activities, such as lifting and working with hands below knee-height. These 13 activities are assessed on a 5-point Likert scale, from 0 (extreme difficulty or unable to perform) to 4 (no difficulty at all), resulting in a converted total score between 0 (extreme difficulties) to 100 (no problems at all). Work-Related Characteristics Work-related characteristics were also self-reported: being a bread winner (yes/no); being self-employed (yes/no); having a handicap accessible workplace (yes/no); and preoperative sick leave (yes/no). Patients’ preoperative expectation regarding work ability at 6 months after surgery was reported by using the single item work ability score (WAS). This score ranges from 0, at which score a patient expect no work ability at all at 6 months postoperative, to 10, an expected work ability as it was at lifetime best . Expected WAS was dichotomized with a cut-off point of 8 or higher defined as high expectations of postoperative work ability and lower than 8 as low expectations of postoperative work ability . Having a knee-straining job was defined by patients who reported that they have to perform at least one of the following five activities ‘often’ or ‘always’: crouching, kneeling, clambering, taking the stairs or lifting . Resumed working hours at 6 and 12 months after surgery were self-reported and calculated as a percentage of self-reported regular working hours of each individual TKA patient. TKA patients reported their actual work ability on the single item Work Ability Score (WAS) again at 6 and 12 months. Furthermore, satisfaction with their physical work ability regarding the operated knee was reported on a single item score from 0, not satisfied at all to 10, totally satisfied. Occupational Medicine Specialist Whether or not an OMS was consulted by TKA patients within 3 months after surgery was self-reported at the 3-month postoperative measurement. Potential Confounders for RTW Potential confounders for RTW were: age; gender; BMI; pain related to the knee (KOOS pain); symptoms related to the knee (KOOS symptoms); quality of life related to the knee (KOOS quality of life); perceived difficulty with work-related knee-straining activities (WORQ score); having a knee-straining job; preoperative sick leave; being self-employed; availability of a handicap accessible workplace and preoperative expected work ability (expected WAS) at 6 months postoperative . Study Size A study size of 160 patients was deemed to be needed. This was based on an expected inclusion of six dichotomous or continuous variables in multiple linear regression analyses. Thereby we took into account that for every variable in the multiple analyses a minimum of 10 patients are required . Finally, we assumed that 60 patients would be included without consult of an OMS and 100 patients with consult of an OMS after TKA. Statistics Firstly, descriptive statistics were used to describe patient and work-related characteristics of TKA patients who did and did not consult an OMS. Differences between these groups were statistically tested at a significance level of p < 0.05. Secondly, a Mann Whitney U non-parametric test (MWU) and a Kaplan Meier survival analysis were performed. The Kaplan Meier survival analysis was performed with the Wilcoxon (Breslow) test to give more weight to the first phase of RTW. This was done because of the importance of an early RTW and in line with the Dutch orthopedic guideline recommendation of RTW within 3 months. Thirdly, to answer the question of whether or not consulting an OMS decreases (median) time to RTW after TKA, multiple linear regression analysis was performed in order to adjust for confounding and effect modification. To meet the assumptions of linear regression, the outcome measure, time to RTW, was transformed by taking the square root. Potential confounders were included in the multiple linear regression analysis that correlated with time to RTW at a significance level of p = 0.05 and showed a collinearity of < 0.7 using Spearman's Rank correlation coefficient with other potential confounders. This linear regression analysis was performed using the comprehensive method for association models . First, OMS consultation was entered into the model. Initially, effect modifiers were identified because this meant that the presence of this factor differently affected (the square root of) time to RTW, if an OMS was consulted or not. A significance level of p < 0.10 was used to prevent potential relevant effect modifiers from being opted out. If effect modification was observed, a multiple linear regression analysis was performed within each stratum of the effect modifier to secure that the effect on RTW could be attributed to this specific factor. Subsequently, a confounder analysis was performed by adding one potential confounder at the time to the regression model, if needed per stratum of the effect modifier. The predictor variable with the largest effect on the regression coefficient of the OMS, and with an effect of at least a 10% change in the coefficient of the OMS, was then added to the model. These steps were repeated until none of the remaining potential confounders had an effect of at least a 10% change on the coefficient of the OMS, the number of cases was less than ten per variable or if no variables were left. Assumptions for applying linear regression analysis were checked on the final association model for homoscedasticity of errors, independency of errors using the Durbin-Watson test and normal distribution of errors. Additionally, as a secondary outcome, working hours, experienced WAS and satisfaction with work ability at 6 and 12 months was assessed for differences between patients who did and did not consult an OMS. All statistical analysis were performed using SPSS for Windows (Version 26.0; IBM Corp, Armonk, NY, USA). A multi-center prospective cohort study among TKA patients was performed . Patients were included from nine surgeons, working in seven hospitals, in five Dutch regions, to minimize selection bias. The hospitals varied from general hospitals, large teaching hospitals to tertiary university hospitals. Inclusion criteria were (1) patients undergoing TKA between June 2014 and March 2018, (2) aged 18 to 65 years (working age), (3) having a paid job, (4) self-reported intend to RTW after surgery and (5) provided information about consulting an OMS within 3 months after TKA or not. Data were collected before TKA and at 3, 6 and 12 months after surgery using a self-report questionnaire in Dutch. Patients could choose a paper or electronic version of the questionnaire to avoid selection bias based on patients’ computer literacy. At each measurement, non-responding patients were reminded up to two times after 2 weeks. Patients who were willing to participate but missed the preoperative measurement due to logistical reasons, were included in the follow-up measurements. Patient characteristics were registered, such as date of birth, gender, body height and body mass. The latter two were used to calculate BMI. Patients were asked whether they had other diseases that were limiting their activities at work, with three categories: (1) No, (2) Yes, one disease that is limiting my activities at work, and (3) Yes, more than one disease that is limiting my activities at work. These categories were dichotomized into either ‘no’ (No) or ‘one or more other disease(s) that limits my activities at work’ (Yes), which was defined as comorbidity. Knee Injury and Osteoarthritis Outcome Score (KOOS) subscales on pain, symptoms and quality of live were filled out by the patients. All KOOS subscales are validated in Dutch and range from 0, representing extreme knee problems, to 100 representing no knee problems . The Work Osteoarthritis or joint-Replacement Questionnaire (WORQ) was used for the work-related physical functioning score, and is also validated in Dutch . The WORQ score consists of 13 items on work-related knee-straining activities, such as lifting and working with hands below knee-height. These 13 activities are assessed on a 5-point Likert scale, from 0 (extreme difficulty or unable to perform) to 4 (no difficulty at all), resulting in a converted total score between 0 (extreme difficulties) to 100 (no problems at all). Work-related characteristics were also self-reported: being a bread winner (yes/no); being self-employed (yes/no); having a handicap accessible workplace (yes/no); and preoperative sick leave (yes/no). Patients’ preoperative expectation regarding work ability at 6 months after surgery was reported by using the single item work ability score (WAS). This score ranges from 0, at which score a patient expect no work ability at all at 6 months postoperative, to 10, an expected work ability as it was at lifetime best . Expected WAS was dichotomized with a cut-off point of 8 or higher defined as high expectations of postoperative work ability and lower than 8 as low expectations of postoperative work ability . Having a knee-straining job was defined by patients who reported that they have to perform at least one of the following five activities ‘often’ or ‘always’: crouching, kneeling, clambering, taking the stairs or lifting . Resumed working hours at 6 and 12 months after surgery were self-reported and calculated as a percentage of self-reported regular working hours of each individual TKA patient. TKA patients reported their actual work ability on the single item Work Ability Score (WAS) again at 6 and 12 months. Furthermore, satisfaction with their physical work ability regarding the operated knee was reported on a single item score from 0, not satisfied at all to 10, totally satisfied. Whether or not an OMS was consulted by TKA patients within 3 months after surgery was self-reported at the 3-month postoperative measurement. Potential Confounders for RTW Potential confounders for RTW were: age; gender; BMI; pain related to the knee (KOOS pain); symptoms related to the knee (KOOS symptoms); quality of life related to the knee (KOOS quality of life); perceived difficulty with work-related knee-straining activities (WORQ score); having a knee-straining job; preoperative sick leave; being self-employed; availability of a handicap accessible workplace and preoperative expected work ability (expected WAS) at 6 months postoperative . Study Size A study size of 160 patients was deemed to be needed. This was based on an expected inclusion of six dichotomous or continuous variables in multiple linear regression analyses. Thereby we took into account that for every variable in the multiple analyses a minimum of 10 patients are required . Finally, we assumed that 60 patients would be included without consult of an OMS and 100 patients with consult of an OMS after TKA. Potential confounders for RTW were: age; gender; BMI; pain related to the knee (KOOS pain); symptoms related to the knee (KOOS symptoms); quality of life related to the knee (KOOS quality of life); perceived difficulty with work-related knee-straining activities (WORQ score); having a knee-straining job; preoperative sick leave; being self-employed; availability of a handicap accessible workplace and preoperative expected work ability (expected WAS) at 6 months postoperative . Study Size A study size of 160 patients was deemed to be needed. This was based on an expected inclusion of six dichotomous or continuous variables in multiple linear regression analyses. Thereby we took into account that for every variable in the multiple analyses a minimum of 10 patients are required . Finally, we assumed that 60 patients would be included without consult of an OMS and 100 patients with consult of an OMS after TKA. A study size of 160 patients was deemed to be needed. This was based on an expected inclusion of six dichotomous or continuous variables in multiple linear regression analyses. Thereby we took into account that for every variable in the multiple analyses a minimum of 10 patients are required . Finally, we assumed that 60 patients would be included without consult of an OMS and 100 patients with consult of an OMS after TKA. Firstly, descriptive statistics were used to describe patient and work-related characteristics of TKA patients who did and did not consult an OMS. Differences between these groups were statistically tested at a significance level of p < 0.05. Secondly, a Mann Whitney U non-parametric test (MWU) and a Kaplan Meier survival analysis were performed. The Kaplan Meier survival analysis was performed with the Wilcoxon (Breslow) test to give more weight to the first phase of RTW. This was done because of the importance of an early RTW and in line with the Dutch orthopedic guideline recommendation of RTW within 3 months. Thirdly, to answer the question of whether or not consulting an OMS decreases (median) time to RTW after TKA, multiple linear regression analysis was performed in order to adjust for confounding and effect modification. To meet the assumptions of linear regression, the outcome measure, time to RTW, was transformed by taking the square root. Potential confounders were included in the multiple linear regression analysis that correlated with time to RTW at a significance level of p = 0.05 and showed a collinearity of < 0.7 using Spearman's Rank correlation coefficient with other potential confounders. This linear regression analysis was performed using the comprehensive method for association models . First, OMS consultation was entered into the model. Initially, effect modifiers were identified because this meant that the presence of this factor differently affected (the square root of) time to RTW, if an OMS was consulted or not. A significance level of p < 0.10 was used to prevent potential relevant effect modifiers from being opted out. If effect modification was observed, a multiple linear regression analysis was performed within each stratum of the effect modifier to secure that the effect on RTW could be attributed to this specific factor. Subsequently, a confounder analysis was performed by adding one potential confounder at the time to the regression model, if needed per stratum of the effect modifier. The predictor variable with the largest effect on the regression coefficient of the OMS, and with an effect of at least a 10% change in the coefficient of the OMS, was then added to the model. These steps were repeated until none of the remaining potential confounders had an effect of at least a 10% change on the coefficient of the OMS, the number of cases was less than ten per variable or if no variables were left. Assumptions for applying linear regression analysis were checked on the final association model for homoscedasticity of errors, independency of errors using the Durbin-Watson test and normal distribution of errors. Additionally, as a secondary outcome, working hours, experienced WAS and satisfaction with work ability at 6 and 12 months was assessed for differences between patients who did and did not consult an OMS. All statistical analysis were performed using SPSS for Windows (Version 26.0; IBM Corp, Armonk, NY, USA). Patient and Work-Related Characteristics One hundred eighty-two (182) TKA patients were included (Fig. ) with a median age at 3 months postoperative of 59 years [IQR 54–62], 87 men (48%), a median BMI of 29 [IQR 26–32], and a median KOOS symptoms score of 61 [IQR 46–71] (Table ). Patients with and without an OMS consult did not differ regarding their personal and work-related characteristics such as comorbidity, KOOS symptoms and preoperative sick leave, except for being self-employed. Patients who consulted an OMS were less often self-employed (2%) than those not consulting an OMS (30%, p < 0.01). Preoperative KOOS subscales and WORQ scores were also not statistically different between TKA patients who did and did not consult an OMS. OMS Consultation and Effect on Time to Return to Work The Kaplan Meier survival curve of patients consulting an OMS (thick solid line) shows a later RTW compared to patients without consult of an OMS (dashed line) within 3 months postoperative (n = 182, p = 0.03 Fig. ). Patients, regardless whether or not they visited an OMS within 3 months postoperative, returned to work with a median of 72 days [IQR 47–108]. Patients who consulted an OMS returned to work later (median 78 [61–111]) than patients who did not consult an OMS (median 62 [34–102], MWU p < 0.01; Table ). The effect of consulting an OMS on time to RTW within 3 months postoperatively was modified by two variables in linear regression analysis, namely expected WAS (p = 0.054) and being self-employed (p = 0.062). Therefore, confounder analysis was performed in patients with high and low expected WAS and in patients with paid employment and being self-employed (Table ). After controlling for confounding the following results were found. Only among patients who expected good postoperative work ability a significant association was found between the (square root of) days to RTW and a consult with an OMS when adjusted for having a knee- straining job and their KOOS pain level at 3 months postoperative (√days to RTW = √(7.875 + (1.404OMS) + (2.524Knee-straining job) + (− 0.027KOOS-pain))). This association resulted in an R square of 0.274, meaning 27% of (the variation in) time needed to RTW was explained by (the variation in) a consult with an OMS, a knee straining job and pain level. TKA patients with high expected WAS returned to work 24 days later when an OMS was consulted compared to patients who did not consult an OMS (p = 0.03). TKA patients with low expected WAS returned to work at the same time whether an OMS was consulted or not (p = 0.69). TKA patients with paid employment returned to work at the same time with or without consulting an OMS (p = 0.14). Patients being self-employed seemed to return to work later when an OMS was consulted, however due to the low number of these patients (n = 2) this analysis could not be performed. Therefore, it could not be confirmed that employment was an effect modifier or confounder. Working Hours and Work Ability In patients with high expected WAS without consulting an OMS the number of working hours (median 32 [18–40]), experienced WAS (median 8 [IQR 7–8] and satisfaction with work ability (median 8 [IQR7-9]) at 6 months postoperative did not differ from patients with high expected WAS consulting an OMS (respectively 32 [IQR 16–40], 7 [IQR 7–8] and 8 [IQR 7–8]). The same was true at 12 months postoperative. Preoperative Non-responders Patients who did not respond to preoperative measurements did not differ significantly (at a p < 0.05 significance level) from patients who responded to preoperative measurements regarding patient and work-related characteristics (Supplementary data, Table 3). One hundred eighty-two (182) TKA patients were included (Fig. ) with a median age at 3 months postoperative of 59 years [IQR 54–62], 87 men (48%), a median BMI of 29 [IQR 26–32], and a median KOOS symptoms score of 61 [IQR 46–71] (Table ). Patients with and without an OMS consult did not differ regarding their personal and work-related characteristics such as comorbidity, KOOS symptoms and preoperative sick leave, except for being self-employed. Patients who consulted an OMS were less often self-employed (2%) than those not consulting an OMS (30%, p < 0.01). Preoperative KOOS subscales and WORQ scores were also not statistically different between TKA patients who did and did not consult an OMS. The Kaplan Meier survival curve of patients consulting an OMS (thick solid line) shows a later RTW compared to patients without consult of an OMS (dashed line) within 3 months postoperative (n = 182, p = 0.03 Fig. ). Patients, regardless whether or not they visited an OMS within 3 months postoperative, returned to work with a median of 72 days [IQR 47–108]. Patients who consulted an OMS returned to work later (median 78 [61–111]) than patients who did not consult an OMS (median 62 [34–102], MWU p < 0.01; Table ). The effect of consulting an OMS on time to RTW within 3 months postoperatively was modified by two variables in linear regression analysis, namely expected WAS (p = 0.054) and being self-employed (p = 0.062). Therefore, confounder analysis was performed in patients with high and low expected WAS and in patients with paid employment and being self-employed (Table ). After controlling for confounding the following results were found. Only among patients who expected good postoperative work ability a significant association was found between the (square root of) days to RTW and a consult with an OMS when adjusted for having a knee- straining job and their KOOS pain level at 3 months postoperative (√days to RTW = √(7.875 + (1.404OMS) + (2.524Knee-straining job) + (− 0.027KOOS-pain))). This association resulted in an R square of 0.274, meaning 27% of (the variation in) time needed to RTW was explained by (the variation in) a consult with an OMS, a knee straining job and pain level. TKA patients with high expected WAS returned to work 24 days later when an OMS was consulted compared to patients who did not consult an OMS (p = 0.03). TKA patients with low expected WAS returned to work at the same time whether an OMS was consulted or not (p = 0.69). TKA patients with paid employment returned to work at the same time with or without consulting an OMS (p = 0.14). Patients being self-employed seemed to return to work later when an OMS was consulted, however due to the low number of these patients (n = 2) this analysis could not be performed. Therefore, it could not be confirmed that employment was an effect modifier or confounder. Working Hours and Work Ability In patients with high expected WAS without consulting an OMS the number of working hours (median 32 [18–40]), experienced WAS (median 8 [IQR 7–8] and satisfaction with work ability (median 8 [IQR7-9]) at 6 months postoperative did not differ from patients with high expected WAS consulting an OMS (respectively 32 [IQR 16–40], 7 [IQR 7–8] and 8 [IQR 7–8]). The same was true at 12 months postoperative. Preoperative Non-responders Patients who did not respond to preoperative measurements did not differ significantly (at a p < 0.05 significance level) from patients who responded to preoperative measurements regarding patient and work-related characteristics (Supplementary data, Table 3). In patients with high expected WAS without consulting an OMS the number of working hours (median 32 [18–40]), experienced WAS (median 8 [IQR 7–8] and satisfaction with work ability (median 8 [IQR7-9]) at 6 months postoperative did not differ from patients with high expected WAS consulting an OMS (respectively 32 [IQR 16–40], 7 [IQR 7–8] and 8 [IQR 7–8]). The same was true at 12 months postoperative. Patients who did not respond to preoperative measurements did not differ significantly (at a p < 0.05 significance level) from patients who responded to preoperative measurements regarding patient and work-related characteristics (Supplementary data, Table 3). Consulting an OMS did not show an earlier RTW among TKA patients. Moreover, in the group of TKA patients with high expectations an earlier RTW was seen in patients that had not consulted an OMS. Regarding no earlier RTW in patients who consulted an OMS, four possible explanations can be given. First, it might be that OMSs advise a more conservative RTW trajectory than needed, to secure a safe recovery and sustainable RTW without increasing a risk of complications. At the moment no occupational health guideline regarding RTW advice for these patients is available in the Netherlands and other countries. The only guidance given is the practice-based recommendation in Dutch orthopedic TKA guideline stating that ‘RTW is possible within 3 months and should start gradually if needed . Also the recently published Dutch and American physiotherapy guidelines, advice early and personalized progression of physical activity for TKA patients but lack recommendations regarding return to work . Recently the Dutch multidisciplinary practice guideline for occupational health professionals was developed for patients with low back pain and lumbosacral radicular syndrome . Following this example a multidisciplinary occupational health guideline on prevention and work participation of knee osteoarthritis patients, as well as pre- and postoperative care in TKA patients can be of help for informed decision making and alleviate the burden of knee OA and TKA on patients, employers, health care and society . A second explanation might be that the timing of the consult with an OMS is not early enough to establish a decrease in time to RTW. In the Netherlands not every patient, for instance a self-employed patient, has free access to an OMS. If an employed patient can consult an OMS the waiting time for an appointment is often around 6 weeks which is when a consultation is mandatory to comply with the Dutch Gatekeeper law. This law states that a problem analysis regarding RTW has to be made by an OMS within the first 6 weeks of sick leave and this analysis should be used to make an RTW-plan by the employer and employee in the first 8 weeks of sick leave. However, managing the RTW-process for TKA patients should start earlier and preferably before surgery to have an effect on timely RTW after TKA. This is especially the case if hindering factors need to be managed like perceived difficulty with knee-straining activities, having a knee-straining job or a workplace that is not handicap accessible. A third explanation might be that the previously mentioned waiting time for an OMS appointment can potentially lead to a wait-and-see attitude regarding RTW in patients who do not prefer or feel secure to RTW on their own accord. This would possibly be true in patients with less self-management skills, less confidence in their TKA recovery, less urgency to return to work or other reasons for a wait-and-see attitude regarding RTW. This possible wait-and-see attitude caused by the time to an OMS appointment is therefore also related to the fourth possible explanation, namely a possible selection bias. This fourth possible explanation for no decrease in time to RTW among patients who consult an OMS might be selection bias based on psychosocial factors we did not measure, or so called reverse causation. In terms of the biopsychosocial model this study does not confirm selection bias based on biological factors, such as comorbidity, knee pain and symptoms (KOOS) or difficulty to perform work-related knee-straining activities (WORQ). However, selection bias based on psychosocial factors could be addressed more properly. To limit the number of questions, we prioritized prognostic variables for RTW regarding TKA patients described in literature. Remarkably, self-efficacy was not one of these variables. Patients that have less possibilities for personal job development or have less work recognition are recently found to have an increased time to RTW after TKA . It seems plausible that these characteristics would be more present among patients consulting an OMS because of a possible need for psychosocial support. Patients Who Expect Good Postoperative Work Ability A mixture of the aforementioned reasons 1, 2 and 3 can probably explain the later RTW in patients with high expected WAS that consulted an OMS compared to patients that did not consult an OMS. Regarding a probably conservative RTW trajectory by OMSs to secure a safe recovery and sustainable RTW, we could assess in our data that the early RTW in patients without consulting an OMS did not result in a worse outcome. The number of hours at work, the work ability of patients and the satisfaction with work ability did not differ between patients who did and did not consult an OMS. Another explanation of an earlier RTW in patients with high expected WAS that did not consult an OMS might be a high self-efficacy which increases the probability of RTW on their own accord instead of waiting for an appointment with an OMS. Patients with high expected WAS are probably also patients with more possibilities for personal job development or more work recognition . Moreover, expectations of work ability after surgery might (partly) be based on realistic insights in their physical ability in relation to physical job demands by these patients. If that is true then it can be argued that patients with high expected WAS would be able to safely RTW early and on their own accord in contrast to patients with low expected WAS given their worse physical ability in relation to their physical job demands. Qualitative findings also support the hypothesis that patients’ needs are partially based on having access to work adjustments and tools . In line with these findings, patients in our study who have high expectations and do consult an OMS, might probably experience a lack of supportive interventions and for that reason consult an OMS. Future Directions First of all, we would recommend to better inform OMSs regarding facilitating and hindering factors for RTW and corresponding median times for RTW regardless whether this is partial or full. This information might empower OMSs and thereby they can reassure their patients that resuming work in a timely manner has better prognosis for RTW . As yet we do not know which exercise-based therapy and integrated care interventions are effective for soon, safe and sustainable RTW in TKA patients . Moreover, in line with the hierarchy of risk management it is more ethical to start with adjusting the work to the patient’s needs, especially if the work is knee-straining. Recent studies have shown that interventions supporting RTW in arthroplasty patients based on knowledge for safe recovery, also regarding work-related activities, and sustainable RTW are being developed . Managing too high patient expectations has also been suggested in TKA patients for better patient-reported outcome as well as RTW . The present study provides another perspective on the relevance of patients’ expectations, given that patients with high expectations who do not consult an OMS have a more timely RTW than their counterparts who consulted an OMS. OMSs might also pay attention to other patient needs than addressed by the potential confounders in our study because these only explained 24% of the later RTW in patients with high work ability expectations. Independent of what advice or guidance supports an earlier RTW, it is also important to know whether this earlier RTW can be considered safe recovery and sustainable RTW. As said, TKA patients with high expectations without consulting an OMS returned to work earlier and had the same outcome of work ability at 6 and 12 months as their counterparts. We foresee, based on these findings, that an earlier RTW seems safe and sustainable, especially when patients have a high physical ability given their physical job demands and return to work on their own accord. Of course, more prospective studies are needed to confirm these findings. Promising Interventions Hindering factors for RTW need to be addressed. Based on earlier research potential effective interventions might be work-directed rehabilitation in patients with a knee-straining job, facilitating arrangements to improve access to the workplace, preoperatively managing patient expectations regarding work ability after TKA and introducing ergonomic measures to decrease the physical workload . These measures could be initiated or coordinated by an OMS, even before surgery, so that the proposed interventions could be implemented in time . Moreover, patients receiving a positive advice regarding RTW by their OMS as well as their orthopedic surgeon said that this was beneficial for their RTW . When the orthopedic surgeon refers patients with hindering factors for RTW to their OMS, preferably before surgery, this might enhance a timely RTW if the OMS is able to act accordingly. RTW advise probably should also be individualized and needs involvement of the employer, as has been found in an intervention mapping approach to develop a clinical occupational advice intervention for knee arthroplasty patients . M/-eHealth could be another promising add-on intervention, since it can provide personalized and frequent advice for TKA patients regarding timely performance of activities based on for instance activity trackers, self-reported recovery and algorithms . Another possible effective element might be setting specific work activity goals in rehabilitation, since this resulted in an increased satisfaction with performing work-related activities in TKA patients . Physiotherapists, especially those specialized in occupational health and ergonomics, could also add value because of their knowledge of the physical recovery of the patient and of assessing and (temporarily) adjusting physical job demands . This dual approach of work directed care and corresponding adjustment of job demands might have potential to enhance (time to) RTW. Above all, given the lack of evidence, studies are needed that evaluate the effectiveness of physical rehabilitation for RTW, given that this care is received by most patients post-surgery . Strengths and Limitations A strength of this study is its prospective multicenter design, in which seven hospitals throughout the Netherlands participated, resulting in a large number of working age TKA patients intending to RTW (n = 182). Another strength is that, as far as we know, this is the first study addressing the potential added value of consulting an OMS regarding RTW after TKA. Moreover, this study incorporated a priori chosen potential confounders for RTW although other confounders might still be present. And lastly, our multiple linear regression analysis, also focusing on modifiers and confounders of the effect of an OMS consult on RTW, can also be considered a strength of this study. The most important limitation of our study is that selection bias is still possible due to the study design, being a prospective cohort study and not an intervention study. Another major limitation of our study is that we only had self-reported data and did not have information regarding the content and exact timing of the OMS consult. Also, the RTW process is complex and can be influenced by factors not measured, e.g. psychosocial factors such as support from a patients supervisor or colleagues at work . Another limitation of the study is that it was originally designed as a cross-sectional preoperative measurement and pending approval for the postoperative measurements by the Medical Ethics Review Committee, the first consecutively 49 patients could not be invited for their 3-months measurement. Therefore, these patients were not eligible but we expect these to be random and not subject to whether or not an OMS was consulted within 3 months. A mixture of the aforementioned reasons 1, 2 and 3 can probably explain the later RTW in patients with high expected WAS that consulted an OMS compared to patients that did not consult an OMS. Regarding a probably conservative RTW trajectory by OMSs to secure a safe recovery and sustainable RTW, we could assess in our data that the early RTW in patients without consulting an OMS did not result in a worse outcome. The number of hours at work, the work ability of patients and the satisfaction with work ability did not differ between patients who did and did not consult an OMS. Another explanation of an earlier RTW in patients with high expected WAS that did not consult an OMS might be a high self-efficacy which increases the probability of RTW on their own accord instead of waiting for an appointment with an OMS. Patients with high expected WAS are probably also patients with more possibilities for personal job development or more work recognition . Moreover, expectations of work ability after surgery might (partly) be based on realistic insights in their physical ability in relation to physical job demands by these patients. If that is true then it can be argued that patients with high expected WAS would be able to safely RTW early and on their own accord in contrast to patients with low expected WAS given their worse physical ability in relation to their physical job demands. Qualitative findings also support the hypothesis that patients’ needs are partially based on having access to work adjustments and tools . In line with these findings, patients in our study who have high expectations and do consult an OMS, might probably experience a lack of supportive interventions and for that reason consult an OMS. First of all, we would recommend to better inform OMSs regarding facilitating and hindering factors for RTW and corresponding median times for RTW regardless whether this is partial or full. This information might empower OMSs and thereby they can reassure their patients that resuming work in a timely manner has better prognosis for RTW . As yet we do not know which exercise-based therapy and integrated care interventions are effective for soon, safe and sustainable RTW in TKA patients . Moreover, in line with the hierarchy of risk management it is more ethical to start with adjusting the work to the patient’s needs, especially if the work is knee-straining. Recent studies have shown that interventions supporting RTW in arthroplasty patients based on knowledge for safe recovery, also regarding work-related activities, and sustainable RTW are being developed . Managing too high patient expectations has also been suggested in TKA patients for better patient-reported outcome as well as RTW . The present study provides another perspective on the relevance of patients’ expectations, given that patients with high expectations who do not consult an OMS have a more timely RTW than their counterparts who consulted an OMS. OMSs might also pay attention to other patient needs than addressed by the potential confounders in our study because these only explained 24% of the later RTW in patients with high work ability expectations. Independent of what advice or guidance supports an earlier RTW, it is also important to know whether this earlier RTW can be considered safe recovery and sustainable RTW. As said, TKA patients with high expectations without consulting an OMS returned to work earlier and had the same outcome of work ability at 6 and 12 months as their counterparts. We foresee, based on these findings, that an earlier RTW seems safe and sustainable, especially when patients have a high physical ability given their physical job demands and return to work on their own accord. Of course, more prospective studies are needed to confirm these findings. Hindering factors for RTW need to be addressed. Based on earlier research potential effective interventions might be work-directed rehabilitation in patients with a knee-straining job, facilitating arrangements to improve access to the workplace, preoperatively managing patient expectations regarding work ability after TKA and introducing ergonomic measures to decrease the physical workload . These measures could be initiated or coordinated by an OMS, even before surgery, so that the proposed interventions could be implemented in time . Moreover, patients receiving a positive advice regarding RTW by their OMS as well as their orthopedic surgeon said that this was beneficial for their RTW . When the orthopedic surgeon refers patients with hindering factors for RTW to their OMS, preferably before surgery, this might enhance a timely RTW if the OMS is able to act accordingly. RTW advise probably should also be individualized and needs involvement of the employer, as has been found in an intervention mapping approach to develop a clinical occupational advice intervention for knee arthroplasty patients . M/-eHealth could be another promising add-on intervention, since it can provide personalized and frequent advice for TKA patients regarding timely performance of activities based on for instance activity trackers, self-reported recovery and algorithms . Another possible effective element might be setting specific work activity goals in rehabilitation, since this resulted in an increased satisfaction with performing work-related activities in TKA patients . Physiotherapists, especially those specialized in occupational health and ergonomics, could also add value because of their knowledge of the physical recovery of the patient and of assessing and (temporarily) adjusting physical job demands . This dual approach of work directed care and corresponding adjustment of job demands might have potential to enhance (time to) RTW. Above all, given the lack of evidence, studies are needed that evaluate the effectiveness of physical rehabilitation for RTW, given that this care is received by most patients post-surgery . A strength of this study is its prospective multicenter design, in which seven hospitals throughout the Netherlands participated, resulting in a large number of working age TKA patients intending to RTW (n = 182). Another strength is that, as far as we know, this is the first study addressing the potential added value of consulting an OMS regarding RTW after TKA. Moreover, this study incorporated a priori chosen potential confounders for RTW although other confounders might still be present. And lastly, our multiple linear regression analysis, also focusing on modifiers and confounders of the effect of an OMS consult on RTW, can also be considered a strength of this study. The most important limitation of our study is that selection bias is still possible due to the study design, being a prospective cohort study and not an intervention study. Another major limitation of our study is that we only had self-reported data and did not have information regarding the content and exact timing of the OMS consult. Also, the RTW process is complex and can be influenced by factors not measured, e.g. psychosocial factors such as support from a patients supervisor or colleagues at work . Another limitation of the study is that it was originally designed as a cross-sectional preoperative measurement and pending approval for the postoperative measurements by the Medical Ethics Review Committee, the first consecutively 49 patients could not be invited for their 3-months measurement. Therefore, these patients were not eligible but we expect these to be random and not subject to whether or not an OMS was consulted within 3 months. Consulting an OMS within 3 months after surgery did not result in an earlier RTW in TKA patients. TKA patients with preoperative high expectations of work ability did even RTW earlier if they had not consulted an OMS compared to their counterparts. Given the increasing number of working age TKA patients worldwide, these findings strengthen the plea for more research on interventions to decrease time to RTW, incorporating the positive effects of high expectations on RTW. Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 35 kb)
Mobile health can be patient-centered and help solve inequality issues in Brazil's Unified Health System	e014adbd-d624-44b4-9933-bd4da0e290b3	10172931	Patient-Centered Care[mh]	The use of mobile health (mHealth) is on the rise in many fields ( ). The Global Observatory for eHealth of the World Health Organization defines mHealth as “medical and public health practice supported by mobile devices, such as mobile phones, patient monitoring devices, personal digital assistants, and other wireless devices ( ).” The desire to overcome obstacles imposed by the pandemic made mHealth more popular than ever in the past years ( ). Ethical issues, however, stimulate discussions about the theme ( , ). In Brazil, one of the most challenging public health problems is attending to the universality principle of the country's Unified Health System or, in Portuguese, Sistema Único de Saúde (SUS). SUS is the world's largest health care system run by a government. The challenge is not only due to the almost 200 million users of the system, but also due to the continental dimensions of the country. Therefore, although SUS is offered to the everyone in the country, including foreigners, some areas have better access to health service than others. But, in face of that problem, some mHealth initiatives have started to help solve this inequality issue ( ). With an estimated 6 billion users in 2022, “mobile phones have become nearly ubiquitous both at the margins and the centers of capitalism ( ).” That means mobile phones can serve as instruments to help solve the problem of inequality in public healthcare access worldwide, since many studies have proven mHealth can be an efficient way to deliver health services at lower costs ( , , , ). In fact, The Universal Declaration on Bioethics and Human Rights presented by the United Nations Educational, Scientific, and Cultural Organization (UNESCO) has stated the necessity to rapidly share new therapeutic modalities or products stemming from research with countries in the developing world ( ). In spite of its potential, mHealth assistance carries disadvantages over in-person care. It is no secret that mHealth deprives patients and healthcare professionals (HCP) of essential elements of interpersonal communication (e.g., full body language) ( ). But it also has advantages. Having the chance to communicate with the HCP and access other health services using a mobile phone in the comfort of home has proven to be cost-effective and to have, in many cases, satisfactory outcomes in patient self-efficacy, quality of relationship between HCP and patient and overall health outcomes ( , , , ). While mobile health industry may inadvertently convey the idea that HCP can be replaced by artificial intelligence, nothing is equivalent to or better than the bond between HCP and patients. Unless a mobile application is facilitating that relationship, it is probably not as good as it could be if it created a space for that rapport to flourish. Many studies have demonstrated patient-centered care has positive outcomes in satisfaction and self-management ( ). The vulnerability of a human being in need of help is not better covered by artificial intelligence than by another human being's sensitivity. Studies have shown technology works at best when it facilitates and strengthens the relationship between the parts involved in healthcare. A meaningful relationship in the context is not only important for patients, but for HCP as well, being recently rated as the most significant source of professional satisfaction among physicians ( ). Since mHealth can serve as a channel of relationship via synchronized and unsynchronized communication, it can serve as an environment for humanized patient-centered healthcare. Adjusting to users' preferences and enabling a meaningful experience for both sides can help in the context. Instead of rapidness and depersonalization, which characteristics associated to mobile applications, mHealth can provide unhurried, deep healthcare experiences. The field of mHealth may offer promising tools against health inequalities in Brazil and other parts of the developing world. They may have quality problems, as in person healthcare does. But what determines whether healthcare is patient-centered is the focus on the patient and his or her needs and not the environment where it occurs. Different environments, including those mediated by technology, can serve as scenarios where human dignity is promoted. In fact, many solutions in mHealth have good outcomes, equivalent to those in-person ( , , – ). Therefore, due to its vast capillarity, mHealth may help attenuate the problem of inequality in healthcare assistance in public health in Brazil and other parts of the developing world and should be the target of substantial investment and research. The author confirms being the sole contributor of this work and has approved it for publication.
Organic fertilization enhances the resistance and resilience of soil microbial communities under extreme drought	4762e82b-767e-4cd3-861d-ed868c429b21	10173193	Microbiology[mh]	Soils are crucial for human wellbeing by providing food, feed and medicine . The highly diverse soil microbiome is underlying those functions, as well as determining biogeochemical cycling, plant productivity and the performance of soil-borne diseases , . But human practices, such as the overuse of chemical fertilizers , have led to large-scale soil degradation including reductions in the microbiome , . Thus, alternative management forms that aim at enhancing ecosystem multifunctionality including those based on enhancing the soil microbiome were introduced, such as organic and conservation agriculture , , . However, soil microbiome functions are not constant over time and change depending on external conditions, such as being influenced by the increasing incidences of extreme events in the ongoing climate change . Climate extremes impact soil microbial communities and their functioning constantly, such as increased incidences of drought , , , extreme precipitation events and resulting floodings . Under these stresses, microorganisms and their functions are depending on distinct taxon- and community-specific differences in resistance and resilience. Resistance is commonly defined as the ability of a system to withstand a stress, while resilience is the speed of system’s recovery towards its pre-disturbance state or a new stable state , . Among the soil microbiome, bacteria seem more responsive to changes than fungi . But there are differences among bacterial taxa to changing conditions as taxa exist that can be sensitive, tolerant or opportunistic to extreme stress , , . Sensitive microorganisms are damaged during drought stress and can hardly recover , ; some microbes that can remain active during water limitation are tolerators, and opportunistic microorganisms can colonize empty environmental niches first and influence the chronology of the ensuing microbial species to rebound , . Stress, as induced by drought, can also interactively impact microorganisms with additional external factors like agricultural practices, such as chemical and organic fertilization regimes. However, it remains unknown how management might impact the resistance and resilience of soil microbial communities and their functions. Another issue is that most contemporary studies focused on only one component, resistance or resilience, also neglecting, complex multi-step process of recovery that can differ between biotic taxa and communities , rendering our understanding of ecological stability impossible . In this study, we evaluated the impact of different fertilizer practices and the potential positive impact of organic inputs on bacterial communities and their function under drought to evaluate resistance, and rewetting to study resilience. For that, we established a series of experiments to track different responses of bacterial microbiomes and their multidimensional ecosystem stability in organic vs chemical fertilizer input soils under extreme drought stress and rewetting. We hypothesized that long-term organic inputs establish a distinct microbiota compared with conventional inputs, which possess (1) higher bacterial diversity and resistance to drought stress; (2) higher resilience and recovery under rewetting; (3) increased microbial functions after recovery from extreme drought. Soil collection, inoculation and cultivation Experimental soils were collected from the long-term field experiment conducted for tomato cultivation in a plastic greenhouse located at the Nanjing Institute of Vegetable Science, Nanjing, China (31°43′N, 118°46′E). The long-term field experiment was performed from 2013 and included the following treatments: (1) CF, soil amended with chemical fertilizer; and (2) OF, soil amended with organic fertilizer (chicken manure compost). The amount of 120 kg ha −1 nitrogen (N, in form of urea), 180 kg ha −1 phosphorous (P, in form of calcium superphosphate) and 120 kg ha −1 potassium (K, in form of potassium sulphate) mineral fertilizers were applied to soils in the sampling season in CF treatment. The amount of 7500 kg ha −1 organic fertilizer (chicken manure compost) was applied in OF treatment. In addition, we compensated for the nutrient differences among all treatments with mineral fertilizer in each season. Bulk soil was collected in October 2018, two months after seedling transplantation, by first removing tomato plants and then taking soil cores to a depth of 10 cm. Soils of three replicates were combined and subsequently sieved (2-mm mesh size) . Additionally, we collected soil cores with 0–15 cm depth from the surface of grassland soil near this site. To dilute the difference in physicochemical properties between the two fertilized soils, the tested soils were inoculated into sterilized grassland soil and named NCF (for chemical fertilizer input soil) and NOF (for organic fertilizer input soil). In detail, grassland soil was sterilized by Co75 γ-ray irradiation twice at Nanjing Xiyue Technology Co., ltd, Nanjing, China. Then, 14 g of non-sterilized field soil was dissolved in 20 ml sterilized water and shaken (170 rpm) for 30 minutes to obtain a soil suspension. Subsequently, 186 g of sterilized grassland soil was inoculated with 20 ml soil suspension and stored in plastic tissue culture bottles (350 ml) that had a 0.22-μm filter membrane to prevent cross contamination with bacteria while allowing gas and water vapour exchange. The water holding capacity (WHC) of the tested soils was measured as 35.36 %. All the bottles were randomly placed and cultured in the condition simulating growing season with 28 °C and constant moisture (50 % of water holding capacity) for eight weeks. Experimental design Mesocosm experiments were performed from August 2019 to April 2020. Mesocosms were constructed from the bottles described above. Two stress levels of no stress (ambient) and extreme drought (drought) were established with each stress consisting of four replicates. All the bottles were randomly placed in the greenhouse with an average temperature of 28 °C. Drought treatment bottles without lids were air-dried by electric fans to achieve a drought period while moisture of ambient treatment was kept constant at about 17 % (50 % of water holding capacity). After 80 days of drought stress, sterilized deionized water was added in October 2019 to adjust the water content of the soils recovering to the original level, and all the bottles were cultured for another 170 days. The graphical representation of our experimental design and soil moisture contents during the whole experiment period is shown in and . Soil samples were collected before drought (time initial), at 5, 10, 20, 50, and 80 days during the drought period (times S5, S10, S20, S50 and S80) and at 2, 20, 40, and 170 days during the rewetting period (times R2, R20, R40 and R170). Five grams of soil were sampled in each bottle, and 152 samples were collected (2 fertilization treatments × 2 stress level treatments × 9 time points × 4 replicates + 8 initial samples). After sieving through a sterile 2-mm sieve, all soil samples were stored at −80 °C for soil DNA extraction. DNA extraction and illumina MiSeq sequencing Total soil genomic DNA was extracted from 0.5 g soil using the PowerSoil DNA Isolation Kit (Mobio Laboratories, Carlsbad, CA, USA) according to the manufacturer’s instructions. The quality and quantity of DNA were determined using a NanoDrop 2000 spectrophotometer (Thermo Scientific, Waltham, MA, USA). Soil DNA was subjected to MiSeq sequencing at Magigene Biotechnology Co., ltd. (Guangdong, China). Bacterial sequencing libraries were constructed according to previously described protocols , . Amplification of the V4-V5 hypervariable regions of the 16S rRNA genes was performed using the general bacterial primers 515F (5′- GTGCCAGCMGCCGCGGTAA-3′) and 907R (5′- CCGTCAATTCMTTTRAGTTT-3′) . Bioinformatic analysis The raw split sequences were merged using USEARCH (version 11.0) . After trimming the adaptors and primer sequences, quality filtering was performed using VSEARCH (version 2.15.0) . Sequences with expected errors > 0.5 and length < 300 bp were discarded. Singletons were removed, and the remaining reads were clustered into operational taxonomic units (OTUs) at a 97 % similarity identity level. Finally, a representative sequence for each OTU was selected and classified using the RDP classifier against the RDP 16S rRNA database . Sequences were randomly subsampled to 67,899 reads per sample for 16S rRNA gene sequences. Phylogenetic trees were constructed with FastTree tools . The relative abundance of a given taxonomic group per sample was calculated as the number of sequences affiliated with that group divided by the total number of sequences. Alpha diversity was analysed by Faith’s phylogenetic diversity (Faith’s PD) index. Principal coordinate analysis (PCoA) based on a Bray-Curtis dissimilarity matrix was performed and plotted using the R vegan package to explore differences in bacterial community structures across all soil samples . Permutational multivariate analysis of variance (PERMANOVA) was conducted to evaluate the effect of fertilization treatment and stress treatment on the whole soil bacterial community using the R vegan package . To identify OTUs significantly enriched within the drought treatment, indicator species analysis was conducted within the R library labdsv . Drought response indicators were defined as the relative abundance and frequency of OTUs in the drought treatment being higher than those in the ambient treatment. Early recovery/late recovery-related indicators were defined as the relative abundance and frequency of OTUs in drought-treated R2/R170 samples being higher than those in drought-treated S80 samples. The individual score for each gene was produced according to de Vries et al. (2018). In brief, the relative frequency and relative average abundance within each treatment produced individual scores, and significance was calculated through random reassignment of groups (1000 permutations). Only indicator OTUs that were significant (P < 0.05) and present at > 0.1 % relative abundance were involved in our subsequent analysis. Full tables of indicator scores are provided in the . To calculate indicator taxa, all taxa with fewer than ten reads across all samples were removed. Quantitative real-time PCR analysis Quantitative real-time PCR amplification (qPCR) was used to estimate the abundance of bacteria and Fusarium oxysporum f.sp . lycopersici (FOL) in soil, according to previously described protocols . The abundance of bacteria was quantified with primers Eub338F (5′- ACTCCTACGGGAGGCAGCAG-3′)/Eub518R (5′- ATTACCGCGGCTGCTGG-3′), according to Fierer . Standard curves were generated using 10-fold serial dilutions of a plasmid containing a full-length copy of the 16S rRNA gene from Escherichia coli , and gene copy numbers were calculated according to standard curve equations. The abundance of FOL was determined using a SYBR Green assay (Takara Bio Inc., Japan) with the primers sp1–2f (5′- GCTGGCGGATCTGACACTGT-3′) and sp1–2r (5′- CCTAAACCACATATCTCGTCCAAA-3′), targeting the rDNA intergenic spacer (IGS) . A serial dilution from 10 10 to 10 2 gene copies μl −1 of the IGS gene was used as a standard. Soil respiration measuring Soil basal respiration was measured using the MicroResp™ method described by Creamer . Briefly, colorimetric gel detector plates were created using cresol red indicator solution to be read at an optical density of 570 nm (OD570). Water was added to the substrate plates, which contained 0.25 g testing soil, for basal respiration measurements. Initial colorimetry values were read from the indicator plate at 570 nm before the system was sealed and incubated at 25 °C for 6 h. Following the 6 h incubation, the colorimetric detector plate was re-read on the plate reader at 570 nm to provide the final absorption data. Respiration rates (μg CO 2 ·g −1 ·h −1 ) were calculated from adsorption data minus the blank sample (average values for each plate calculated from initial colorimetric values). Plant growth promotion and pathogen suppression determination Pot experiments were performed to examine two plant-related functions of the soil microbial community in a growth chamber (28 °C average temperature, 60 % relative humidity, 16 h light/8h dark) in October 2019 and April 2020, corresponding to time R2 and R170, respectively. Soil inoculum (30 g) was taken from each mesocosm bottle and mixed with 170 g autoclaved vermiculite and silica sand as a nutritionally defined medium . Two copies of soil were sampled from each bottle: one for the plant growth promotion experiment and another for the pathogen suppression experiment. Each treatment at each time point contained four replicates, which resembled those in our mesocosm experiments. Tomato tissue culture seedlings were cultivated using the plug seeding method and watered with sterile deionized water. After transplanting the seedlings to pots (Volume of 250 ml and filled with preinoculated medium) and watering with 1/2 sterile Hoagland solution for 30 days, tomato biomass in the plant growth promotion experiment was measured. The tomatoes in the pathogen suppression experiment were then inoculated with Fusarium oxysporum f.sp. lycopersici (FOL) spore suspension (final density of 1 × 10 5 spores/g of the substrate as described above). Rhizosphere soil was sampled ten days after FOL inoculation as described by Fu . Total genomic DNA of rhizosphere soil samples was extracted, and the abundance of FOL was quantified by qPCR analysis. Microbial trait value estimation According to the revised microbial life history theory by Malik , the yield, resource acquisition and stress tolerance (Y-A-S) framework was used to explain the soil microbial reaction to drought events. Microbial traits during initial, drought and rewetting were estimated to correspond to the time of initiation, S80 and R170. The rRNA operon copy number has been used as a proxy for a microorganism’s heterotrophic strategy because of the rapidity of its response to resources . Therefore, the abundance-mean weighted ribosomal operon count across all taxa (if available) in each sample was estimated through the rrnDB database to present a high yield (Y) strategy value . PICRUSt2 was used to reconstruct metagenome content and queried for sporulation-related genes . Functional annotations were assigned using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to generate a sample × functional count table . The KEGG functional pathways categorized as stress tolerance (S) strategies are listed in the . Eco-microplates (Biolog Inc., Hayward, Calif) were used to determine the difference in carbon resource acquisition (R) ability of the microbial communities. In brief, microplates containing 96 wells with 31 carbon sources were inoculated with different diluted sample suspensions and incubated for 144 h at 27 °C. Colour formation was measured at 590 nm and 750 nm twice a day. The raw data were transformed: raw difference (RD) = X − X 0 , where X was the mean of the same three wells per plate and X 0 was the mean of the water blanks per plate; average well colour development AWCD = Σ(RD 590 -RD 750 )/31. The ternary phase diagram was analysed according to standardized values of three estimated microbial traits. Statistical analysis All statistical analyses were performed in R (Version 3.6.0). All statistical tests performed in this study were considered significant at P < 0.05. To determine significant differences, unpaired t-tests and two-way analysis of variance (ANOVA) were performed. We quantified six components of compositional stability as previously described definitions (as proxies for the Bray-Curtis dissimilarity between ambient and drought treatments following the experimental design) , ( ). The trend lines were fitted using “geom_smooth” with the “lm” function in ggplot2 . Spearman’s rank correlation coefficients between the relative abundance of OTUs and bacterial abundance/soil respiration/plant biomass/pathogen abundance were calculated in R. P value adjustments for multiple comparisons were performed using false discovery rate (FDR) correction . Heat map analysis of the recovery-related indicators linked to community functions across all fertilized soils was carried out with the “pheatmap” package in R. Experimental soils were collected from the long-term field experiment conducted for tomato cultivation in a plastic greenhouse located at the Nanjing Institute of Vegetable Science, Nanjing, China (31°43′N, 118°46′E). The long-term field experiment was performed from 2013 and included the following treatments: (1) CF, soil amended with chemical fertilizer; and (2) OF, soil amended with organic fertilizer (chicken manure compost). The amount of 120 kg ha −1 nitrogen (N, in form of urea), 180 kg ha −1 phosphorous (P, in form of calcium superphosphate) and 120 kg ha −1 potassium (K, in form of potassium sulphate) mineral fertilizers were applied to soils in the sampling season in CF treatment. The amount of 7500 kg ha −1 organic fertilizer (chicken manure compost) was applied in OF treatment. In addition, we compensated for the nutrient differences among all treatments with mineral fertilizer in each season. Bulk soil was collected in October 2018, two months after seedling transplantation, by first removing tomato plants and then taking soil cores to a depth of 10 cm. Soils of three replicates were combined and subsequently sieved (2-mm mesh size) . Additionally, we collected soil cores with 0–15 cm depth from the surface of grassland soil near this site. To dilute the difference in physicochemical properties between the two fertilized soils, the tested soils were inoculated into sterilized grassland soil and named NCF (for chemical fertilizer input soil) and NOF (for organic fertilizer input soil). In detail, grassland soil was sterilized by Co75 γ-ray irradiation twice at Nanjing Xiyue Technology Co., ltd, Nanjing, China. Then, 14 g of non-sterilized field soil was dissolved in 20 ml sterilized water and shaken (170 rpm) for 30 minutes to obtain a soil suspension. Subsequently, 186 g of sterilized grassland soil was inoculated with 20 ml soil suspension and stored in plastic tissue culture bottles (350 ml) that had a 0.22-μm filter membrane to prevent cross contamination with bacteria while allowing gas and water vapour exchange. The water holding capacity (WHC) of the tested soils was measured as 35.36 %. All the bottles were randomly placed and cultured in the condition simulating growing season with 28 °C and constant moisture (50 % of water holding capacity) for eight weeks. Mesocosm experiments were performed from August 2019 to April 2020. Mesocosms were constructed from the bottles described above. Two stress levels of no stress (ambient) and extreme drought (drought) were established with each stress consisting of four replicates. All the bottles were randomly placed in the greenhouse with an average temperature of 28 °C. Drought treatment bottles without lids were air-dried by electric fans to achieve a drought period while moisture of ambient treatment was kept constant at about 17 % (50 % of water holding capacity). After 80 days of drought stress, sterilized deionized water was added in October 2019 to adjust the water content of the soils recovering to the original level, and all the bottles were cultured for another 170 days. The graphical representation of our experimental design and soil moisture contents during the whole experiment period is shown in and . Soil samples were collected before drought (time initial), at 5, 10, 20, 50, and 80 days during the drought period (times S5, S10, S20, S50 and S80) and at 2, 20, 40, and 170 days during the rewetting period (times R2, R20, R40 and R170). Five grams of soil were sampled in each bottle, and 152 samples were collected (2 fertilization treatments × 2 stress level treatments × 9 time points × 4 replicates + 8 initial samples). After sieving through a sterile 2-mm sieve, all soil samples were stored at −80 °C for soil DNA extraction. Total soil genomic DNA was extracted from 0.5 g soil using the PowerSoil DNA Isolation Kit (Mobio Laboratories, Carlsbad, CA, USA) according to the manufacturer’s instructions. The quality and quantity of DNA were determined using a NanoDrop 2000 spectrophotometer (Thermo Scientific, Waltham, MA, USA). Soil DNA was subjected to MiSeq sequencing at Magigene Biotechnology Co., ltd. (Guangdong, China). Bacterial sequencing libraries were constructed according to previously described protocols , . Amplification of the V4-V5 hypervariable regions of the 16S rRNA genes was performed using the general bacterial primers 515F (5′- GTGCCAGCMGCCGCGGTAA-3′) and 907R (5′- CCGTCAATTCMTTTRAGTTT-3′) . The raw split sequences were merged using USEARCH (version 11.0) . After trimming the adaptors and primer sequences, quality filtering was performed using VSEARCH (version 2.15.0) . Sequences with expected errors > 0.5 and length < 300 bp were discarded. Singletons were removed, and the remaining reads were clustered into operational taxonomic units (OTUs) at a 97 % similarity identity level. Finally, a representative sequence for each OTU was selected and classified using the RDP classifier against the RDP 16S rRNA database . Sequences were randomly subsampled to 67,899 reads per sample for 16S rRNA gene sequences. Phylogenetic trees were constructed with FastTree tools . The relative abundance of a given taxonomic group per sample was calculated as the number of sequences affiliated with that group divided by the total number of sequences. Alpha diversity was analysed by Faith’s phylogenetic diversity (Faith’s PD) index. Principal coordinate analysis (PCoA) based on a Bray-Curtis dissimilarity matrix was performed and plotted using the R vegan package to explore differences in bacterial community structures across all soil samples . Permutational multivariate analysis of variance (PERMANOVA) was conducted to evaluate the effect of fertilization treatment and stress treatment on the whole soil bacterial community using the R vegan package . To identify OTUs significantly enriched within the drought treatment, indicator species analysis was conducted within the R library labdsv . Drought response indicators were defined as the relative abundance and frequency of OTUs in the drought treatment being higher than those in the ambient treatment. Early recovery/late recovery-related indicators were defined as the relative abundance and frequency of OTUs in drought-treated R2/R170 samples being higher than those in drought-treated S80 samples. The individual score for each gene was produced according to de Vries et al. (2018). In brief, the relative frequency and relative average abundance within each treatment produced individual scores, and significance was calculated through random reassignment of groups (1000 permutations). Only indicator OTUs that were significant (P < 0.05) and present at > 0.1 % relative abundance were involved in our subsequent analysis. Full tables of indicator scores are provided in the . To calculate indicator taxa, all taxa with fewer than ten reads across all samples were removed. Quantitative real-time PCR amplification (qPCR) was used to estimate the abundance of bacteria and Fusarium oxysporum f.sp . lycopersici (FOL) in soil, according to previously described protocols . The abundance of bacteria was quantified with primers Eub338F (5′- ACTCCTACGGGAGGCAGCAG-3′)/Eub518R (5′- ATTACCGCGGCTGCTGG-3′), according to Fierer . Standard curves were generated using 10-fold serial dilutions of a plasmid containing a full-length copy of the 16S rRNA gene from Escherichia coli , and gene copy numbers were calculated according to standard curve equations. The abundance of FOL was determined using a SYBR Green assay (Takara Bio Inc., Japan) with the primers sp1–2f (5′- GCTGGCGGATCTGACACTGT-3′) and sp1–2r (5′- CCTAAACCACATATCTCGTCCAAA-3′), targeting the rDNA intergenic spacer (IGS) . A serial dilution from 10 10 to 10 2 gene copies μl −1 of the IGS gene was used as a standard. Soil basal respiration was measured using the MicroResp™ method described by Creamer . Briefly, colorimetric gel detector plates were created using cresol red indicator solution to be read at an optical density of 570 nm (OD570). Water was added to the substrate plates, which contained 0.25 g testing soil, for basal respiration measurements. Initial colorimetry values were read from the indicator plate at 570 nm before the system was sealed and incubated at 25 °C for 6 h. Following the 6 h incubation, the colorimetric detector plate was re-read on the plate reader at 570 nm to provide the final absorption data. Respiration rates (μg CO 2 ·g −1 ·h −1 ) were calculated from adsorption data minus the blank sample (average values for each plate calculated from initial colorimetric values). Pot experiments were performed to examine two plant-related functions of the soil microbial community in a growth chamber (28 °C average temperature, 60 % relative humidity, 16 h light/8h dark) in October 2019 and April 2020, corresponding to time R2 and R170, respectively. Soil inoculum (30 g) was taken from each mesocosm bottle and mixed with 170 g autoclaved vermiculite and silica sand as a nutritionally defined medium . Two copies of soil were sampled from each bottle: one for the plant growth promotion experiment and another for the pathogen suppression experiment. Each treatment at each time point contained four replicates, which resembled those in our mesocosm experiments. Tomato tissue culture seedlings were cultivated using the plug seeding method and watered with sterile deionized water. After transplanting the seedlings to pots (Volume of 250 ml and filled with preinoculated medium) and watering with 1/2 sterile Hoagland solution for 30 days, tomato biomass in the plant growth promotion experiment was measured. The tomatoes in the pathogen suppression experiment were then inoculated with Fusarium oxysporum f.sp. lycopersici (FOL) spore suspension (final density of 1 × 10 5 spores/g of the substrate as described above). Rhizosphere soil was sampled ten days after FOL inoculation as described by Fu . Total genomic DNA of rhizosphere soil samples was extracted, and the abundance of FOL was quantified by qPCR analysis. According to the revised microbial life history theory by Malik , the yield, resource acquisition and stress tolerance (Y-A-S) framework was used to explain the soil microbial reaction to drought events. Microbial traits during initial, drought and rewetting were estimated to correspond to the time of initiation, S80 and R170. The rRNA operon copy number has been used as a proxy for a microorganism’s heterotrophic strategy because of the rapidity of its response to resources . Therefore, the abundance-mean weighted ribosomal operon count across all taxa (if available) in each sample was estimated through the rrnDB database to present a high yield (Y) strategy value . PICRUSt2 was used to reconstruct metagenome content and queried for sporulation-related genes . Functional annotations were assigned using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to generate a sample × functional count table . The KEGG functional pathways categorized as stress tolerance (S) strategies are listed in the . Eco-microplates (Biolog Inc., Hayward, Calif) were used to determine the difference in carbon resource acquisition (R) ability of the microbial communities. In brief, microplates containing 96 wells with 31 carbon sources were inoculated with different diluted sample suspensions and incubated for 144 h at 27 °C. Colour formation was measured at 590 nm and 750 nm twice a day. The raw data were transformed: raw difference (RD) = X − X 0 , where X was the mean of the same three wells per plate and X 0 was the mean of the water blanks per plate; average well colour development AWCD = Σ(RD 590 -RD 750 )/31. The ternary phase diagram was analysed according to standardized values of three estimated microbial traits. All statistical analyses were performed in R (Version 3.6.0). All statistical tests performed in this study were considered significant at P < 0.05. To determine significant differences, unpaired t-tests and two-way analysis of variance (ANOVA) were performed. We quantified six components of compositional stability as previously described definitions (as proxies for the Bray-Curtis dissimilarity between ambient and drought treatments following the experimental design) , ( ). The trend lines were fitted using “geom_smooth” with the “lm” function in ggplot2 . Spearman’s rank correlation coefficients between the relative abundance of OTUs and bacterial abundance/soil respiration/plant biomass/pathogen abundance were calculated in R. P value adjustments for multiple comparisons were performed using false discovery rate (FDR) correction . Heat map analysis of the recovery-related indicators linked to community functions across all fertilized soils was carried out with the “pheatmap” package in R. Bacterial community diversity Bacterial diversity and community structure were strongly affected by drought under both fertilization regimes. shows that the average index of Faith’s PD diversity in NOF during the whole drought period was 397 whereas 322 in NCF. The Faith’s PD index of NOF was significantly higher than NCF at the S10, S20, S50 and S80 timepoints ( P < 0.05, t -test). Principal coordinate analysis (PCoA) revealed significant differences between fertilization and drought treatments ( P (fertilization treatment) < 0.001, P (stress treatment) < 0.001, PERMANOVA) ( a). Overall, the bacterial community compositions of the NCF and NOF treatments were clearly distinct along the first axis (PCoA1). The drought and ambient treatments showed significant differences along the second axis (PCoA2), while the initial and ambient samples grouped together. Time also had a significant effect in the drought treatment during drought and rewetting period ( P < 0.05, PERMANOVA, b). The fluctuation in the relative abundance of dominant phyla is shown in a stream graph ( ). During the stress stage, the most increased phyla were Actinomyces and Firmicutes, with the relative abundances of Actinomyces in NCF raising from 11.64 % to 32.60 % and Firmicutes increasing by approximately 15 % in both treatments. After rewetting, the community composition changed mainly because the relative abundance of Proteobacteria sharply increased by about 25 % immediately in R2. Multiple dimensions of ecological stability To explore the response patterns of the NCF and NOF soil microbial community structures, we choose the Bray-Curtis dissimilarity between drought and ambient treatment as a proxy to calculate ecological stabilities ( ). No significant difference among fertilization patterns in spatial and temporal variability (removing potentially confounding effects of change over the duration of the experiment) was observed. The trend line of Bray-Curtis dissimilarity represents the divergence from the ambient bacterial community. The whole trend line of NOF was more curved than that of NCF ( c). Because dissimilarity continuously decreased during rewetting, resilience (slope of regression lines) was a negative value. In rewetting period, the resilience of the NOF treatment was significantly lower than that of the NCF treatment ( P < 0.05, t -test) ( d), which indicated a greater recovery rate. Finally, the average of recovery (Bray-Curtis dissimilarity between drought and ambient treatment in R170) of the NCF treatment was 0.46, while it was 0.40 in NOF ( P < 0.05, t -test) ( d). The patterns and ecological stability based on the weighted UniFrac dissimilarity were mostly the same as the results based on the Bray-Curtis dissimilarity ( ). Responsive OTU indicators Responsive drought OTUs (species enriched under drought) were mainly from the phyla Actinomyces and Firmicutes , while responsive rewetting OTUs belonged to Proteobacteria and Bacteroidetes ( a, c). During drought, the relative abundance of responsive OTUs in NOF was significantly higher than in NCF at S5, S10, S20 ( P < 0.05, t -test). After rewetting for two days, the relative abundance of responsive OTUs slightly increased to 51 % and 61 % in the NCF and NOF treatments, respectively. During R2 to R170, the responsive OTU proportion of NCF decreased gradually to 15 % in R170, while NOF maintained a high proportion until R40 and then decreased rapidly to 12 % in R170. According to the Venn diagram ( b, d), the most enriched indicator OTUs at the end of stress (S80), early recovery stage (R2) and late recovery stage (R170) were different. We classified these indicator OTUs into seven groups to capture taxa that may contribute to different metrics in multiple dimensions of stability ( e). Group 1 (always enriched at three timepoints) and Group 2 were higher in the NOF treatment than in the NCF treatment ( P < 0.05, t -test). Groups 5/6/7 are abundant groups, whose relative abundance exceeds ten percent. Group 5 (enriched at the end of stress and early recovery) showed higher relative abundance in NCF ( P < 0.05, t -test). Group 6 (drought resistant) showed no difference in relative abundance between the two fertilization treatments, which consisted of 369 OTUs in NOF and 340 OTUs in NCF. Most drought resistant taxa belonged to Actinomyces and Firmicutes ( ). The relative abundance of Group 7 (opportunists) in NOF reached with 49.27 % a higher value than that in NCF in the early recovery stage ( P < 0.05, t -test). 202 OTUs in NCF and 191 OTUs in NOF of the opportunists were classified as Proteobacteria. Soil ecosystem functions Pot experiments showed that all the late recovery bacterial communities (induced by chemical and organic fertilization) increased fresh weight of tomato by about 35 % more than early recovery communities. In plant-based pathogen inhibition assays, the Fusarium abundance in the rhizosphere of NOF was significantly lower than in NCF at the R170 timepoint ( P < 0.05, t -test) ( a), while there was no difference between the two treatments in the ambient treatment ( ). The bacterial abundance in NCF was with 8.25 log/g dry soil at the R170 timepoint higher than bacteria in NOF soil with 7.24 log/g dry soil ( P < 0.05, t -test). The heatmap revealed potential links between recovery-related species and soil ecosystem functions (recovery-related OTUs in R2 and R170 compared with S80 samples of drought treatment are listed in the ). Soil respiration was always positively correlated with most of the late recovery OTUs (r > 0, Spearman correlation), while microbial biomass had negative relations with soil respiration ( b). For plant-related functions, both growth-promoting and pathogen inhibition function had positive relationships with late recovery OTUs (r > 0, Spearman correlation). Organic fertilization enhanced the recovery of pathogen inhibiting function in R170, because the recovery-related OTUs showed a stronger relationship (higher Spearman r value) than in NCF. Microbial trait values Our results on trait-based microbial strategies to explore why organic input enhances bacterial community stability under arid conditions revealed that the NCF microbial community characteristics seemed close to the triangle vertices, while the NOF community characteristic points were near the centre of the three-phase diagram ( a). Significantly higher average rRNA copy number were detected in NOF during drought and rewetting ( P < 0.05, t -test), and the value of rRNA copy numbers were highest during rewetting, indicating a high community growth rate ( b). As for sporulation-related gene numbers, which were predicted by PICRUSt2, the value in NCF was higher than NOF ( P < 0.05, t -test) during drought ( c). The average well development (AWCD) result of BIOLOG showed that long-term organic input strengthened the carbon resource acquisition ability in NOF, which was significantly higher than that of NCF ( P < 0.05, t -test) at the initial and the end of the recovery ( d). Bacterial diversity and community structure were strongly affected by drought under both fertilization regimes. shows that the average index of Faith’s PD diversity in NOF during the whole drought period was 397 whereas 322 in NCF. The Faith’s PD index of NOF was significantly higher than NCF at the S10, S20, S50 and S80 timepoints ( P < 0.05, t -test). Principal coordinate analysis (PCoA) revealed significant differences between fertilization and drought treatments ( P (fertilization treatment) < 0.001, P (stress treatment) < 0.001, PERMANOVA) ( a). Overall, the bacterial community compositions of the NCF and NOF treatments were clearly distinct along the first axis (PCoA1). The drought and ambient treatments showed significant differences along the second axis (PCoA2), while the initial and ambient samples grouped together. Time also had a significant effect in the drought treatment during drought and rewetting period ( P < 0.05, PERMANOVA, b). The fluctuation in the relative abundance of dominant phyla is shown in a stream graph ( ). During the stress stage, the most increased phyla were Actinomyces and Firmicutes, with the relative abundances of Actinomyces in NCF raising from 11.64 % to 32.60 % and Firmicutes increasing by approximately 15 % in both treatments. After rewetting, the community composition changed mainly because the relative abundance of Proteobacteria sharply increased by about 25 % immediately in R2. To explore the response patterns of the NCF and NOF soil microbial community structures, we choose the Bray-Curtis dissimilarity between drought and ambient treatment as a proxy to calculate ecological stabilities ( ). No significant difference among fertilization patterns in spatial and temporal variability (removing potentially confounding effects of change over the duration of the experiment) was observed. The trend line of Bray-Curtis dissimilarity represents the divergence from the ambient bacterial community. The whole trend line of NOF was more curved than that of NCF ( c). Because dissimilarity continuously decreased during rewetting, resilience (slope of regression lines) was a negative value. In rewetting period, the resilience of the NOF treatment was significantly lower than that of the NCF treatment ( P < 0.05, t -test) ( d), which indicated a greater recovery rate. Finally, the average of recovery (Bray-Curtis dissimilarity between drought and ambient treatment in R170) of the NCF treatment was 0.46, while it was 0.40 in NOF ( P < 0.05, t -test) ( d). The patterns and ecological stability based on the weighted UniFrac dissimilarity were mostly the same as the results based on the Bray-Curtis dissimilarity ( ). Responsive drought OTUs (species enriched under drought) were mainly from the phyla Actinomyces and Firmicutes , while responsive rewetting OTUs belonged to Proteobacteria and Bacteroidetes ( a, c). During drought, the relative abundance of responsive OTUs in NOF was significantly higher than in NCF at S5, S10, S20 ( P < 0.05, t -test). After rewetting for two days, the relative abundance of responsive OTUs slightly increased to 51 % and 61 % in the NCF and NOF treatments, respectively. During R2 to R170, the responsive OTU proportion of NCF decreased gradually to 15 % in R170, while NOF maintained a high proportion until R40 and then decreased rapidly to 12 % in R170. According to the Venn diagram ( b, d), the most enriched indicator OTUs at the end of stress (S80), early recovery stage (R2) and late recovery stage (R170) were different. We classified these indicator OTUs into seven groups to capture taxa that may contribute to different metrics in multiple dimensions of stability ( e). Group 1 (always enriched at three timepoints) and Group 2 were higher in the NOF treatment than in the NCF treatment ( P < 0.05, t -test). Groups 5/6/7 are abundant groups, whose relative abundance exceeds ten percent. Group 5 (enriched at the end of stress and early recovery) showed higher relative abundance in NCF ( P < 0.05, t -test). Group 6 (drought resistant) showed no difference in relative abundance between the two fertilization treatments, which consisted of 369 OTUs in NOF and 340 OTUs in NCF. Most drought resistant taxa belonged to Actinomyces and Firmicutes ( ). The relative abundance of Group 7 (opportunists) in NOF reached with 49.27 % a higher value than that in NCF in the early recovery stage ( P < 0.05, t -test). 202 OTUs in NCF and 191 OTUs in NOF of the opportunists were classified as Proteobacteria. Pot experiments showed that all the late recovery bacterial communities (induced by chemical and organic fertilization) increased fresh weight of tomato by about 35 % more than early recovery communities. In plant-based pathogen inhibition assays, the Fusarium abundance in the rhizosphere of NOF was significantly lower than in NCF at the R170 timepoint ( P < 0.05, t -test) ( a), while there was no difference between the two treatments in the ambient treatment ( ). The bacterial abundance in NCF was with 8.25 log/g dry soil at the R170 timepoint higher than bacteria in NOF soil with 7.24 log/g dry soil ( P < 0.05, t -test). The heatmap revealed potential links between recovery-related species and soil ecosystem functions (recovery-related OTUs in R2 and R170 compared with S80 samples of drought treatment are listed in the ). Soil respiration was always positively correlated with most of the late recovery OTUs (r > 0, Spearman correlation), while microbial biomass had negative relations with soil respiration ( b). For plant-related functions, both growth-promoting and pathogen inhibition function had positive relationships with late recovery OTUs (r > 0, Spearman correlation). Organic fertilization enhanced the recovery of pathogen inhibiting function in R170, because the recovery-related OTUs showed a stronger relationship (higher Spearman r value) than in NCF. Our results on trait-based microbial strategies to explore why organic input enhances bacterial community stability under arid conditions revealed that the NCF microbial community characteristics seemed close to the triangle vertices, while the NOF community characteristic points were near the centre of the three-phase diagram ( a). Significantly higher average rRNA copy number were detected in NOF during drought and rewetting ( P < 0.05, t -test), and the value of rRNA copy numbers were highest during rewetting, indicating a high community growth rate ( b). As for sporulation-related gene numbers, which were predicted by PICRUSt2, the value in NCF was higher than NOF ( P < 0.05, t -test) during drought ( c). The average well development (AWCD) result of BIOLOG showed that long-term organic input strengthened the carbon resource acquisition ability in NOF, which was significantly higher than that of NCF ( P < 0.05, t -test) at the initial and the end of the recovery ( d). In this study, we observed that organic fertilization (NOF) established a bacterial community that possesses higher bacterial Faith’s PD index during drought and higher resilience of composition during rewetting compared with conventional fertilizers (NCF). The recovery of bacterial community further enhanced the pathogen-inhibiting function recovery of soil ecosystems in NOF compared to NCF. Generally, drought induced shifts in bacterial diversity and composition. Our results support part of our first hypothesis that NOF supports a higher bacterial diversity under drought than NCF, which is in line with previous studies , . Higher diversity is assumed to be beneficial to cope with extreme stress due to higher metabolic capacities of the entire community . Our results revealed that the relative abundances of Gram-negative bacterial phyla, such as Proteobacteria and Bacteroides , were reduced by drought, while relative abundances of Gram-positive, oligotrophic Actinomyces and Firmicutes were increased. Actinomyces that are plant beneficial under drought , were additionally enhanced in NOF, showing the positive impact induced by organic fertilizers. However, no significant difference of the reactivity and resistance of community composition between NOF and NCF were observed, which contradicts our first hypothesis. On short time scales, the resistance of microorganisms to this dramatic alteration in environmental conditions is determined by specific “response traits” that protect against desiccation , such as a thick peptidoglycan cell wall, osmolyte production, sporulation, and dormancy. On long time scales in our experiment, we speculated that the exhausting of microorganisms’ energy for adapting the 80 days drought affect the resistance of microorganisms in NOF. Our results are consistent with the second hypothesis that higher resilience and recovery of bacterial communities prevail in NOF than in NCF. Interestingly, we observed a sudden decline of bacterial diversity after rewetting the NOF treatment for two days with compositional recovery to the initial community composition only taking place in the late recovery period (from R40 to R170). This finding can be linked to sudden increases in microbial activity after rewetting a dry soil found in a previous study , a phenomenon called the Birch effect . This Birch effect and the findings in our study might be explained by a burst of opportunistic bacteria after reconnecting the aqueous habitat in rewetted soils that result in a diversity decrease of the community . The enrichment of Group 7 indicators in NOF further explained this result at the population level. The dynamics of late recovery in NOF can be explained by the fact that the input of organic fertilizer can increase K-strategist populations, which possess a relatively slow growth rate and plays an important role in moderating the recovery patterns of the soil microbial community . We believe that these microorganisms were promoted among the resident soil microbiome by NOF and not introduced by fertilizers, as most invading microorganisms do not survive in soils for longer periods . The recovery patterns of the NOF community in this study ( ) further confirmed that ecosystem recovery is a complex and multi-step process that has been found in the secondary succession of forests and guts , . Recovery differences in soil pathogen inhibition capability between NCF and NOF were observed, supporting the third hypothesis. Fusarium oxysporum is a common soil-borne pathogen that cause Fusarium wilt in a wide range of hosts, such as tomato, banana, and pepper , . In plant-based pathogen inhibition assays, the low Fusarium abundance in the rhizosphere indicates high inhibition ability. We first observed that pathogen inhibition ability in NOF was enhanced at the late recovery stage but not in NCF. Enhancement of potential anti- Fusarium species might underlie this phenomenon and supports previous findings . Results showed that the soil respiration rate increased following prolonged time after rewetting because soil respiration was always limited by moisture when affected by several drying events . We also found that bacterial community biomass in both NCF and NOF was increased during early recovery, which is consistent with recent reports , . During the late recovery stage, lower microbial abundance with higher pathogen inhibition ability was present in NOF compared to NCF. This finding could be explained by an increase in abundant and functionally important phylotypes that drive variation in broad functions (respiration, biomass), while rarer phylotypes, such as those that are known to produce anti-microbial compounds (REFS-pseudomonas , bacillus things), drive narrower functions such as pathogen inhibition. . This result further demonstrates the unique components of complex communities that are associated with different types of ecosystem functioning. The trait-based microbial strategy and yield-resource acquisition-stress tolerator (Y-A-S) model has been modified to adapt broad situations . The NOF community was more strategy-balancing than the NCF community. The NCF community showed higher stress tolerance that was linked to lower yield and resource acquisition ability, indicating difficulties to maintain multiple functions under stress. During drought, any drought tolerance strategy involves physiological costs . Here, low physiological costs to cope with drought further revealed that the resistance in NOF was higher than in NCF, which confirms our first hypothesis. During rewetting, the high resource acquisition and yield strategies in NOF indicated a high maximum reproductive rate and carbon source utilization ability, laying the foundation for a high resilience. These results provide a new perspective to evaluate the stabilities of microbial communities, but deeper studies to link trait-based microbial strategies with ecosystem processes are needed. Our study has important implications for understanding how soil bacterial communities induced by organic fertilization respond to climate extremes. The results reveal that organic amendments could maintain high bacterial diversity during drought and increase compositional resilience under rewetting. Moreover, these features can be furtherly contributed to enhance pathogen-inhibiting function at the late recovery stage compared to chemical fertilization. We also demonstrate that a stable community tends to possess a strategy-balancing community in the yield-resource acquire-stress tolerator (Y-A-S) model. We highlight that these conclusions can provide the necessary understanding of microbiome manipulation strategies to enhance soil ecosystem stability and maintain soil functions. Compliance with Ethics Requirements This article does not contain any studies with human or animal subjects. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Gender representation in leadership & research: a 13-year review of the Annual Canadian Society of Otolaryngology Meetings	87981e28-679f-442a-8265-5c9c1e05ae07	10173511	Otolaryngology[mh]	Gender disparity in surgical disciplines, including Otolaryngology-Head and Neck Surgery (OHNS), has been highlighted in recent literature. Over the past 20 years, the proportion of female staff otolaryngologists and trainees has increased by 14.2% and 13.3% respectively, where 24.2% of staff otolaryngologists were female, and 41.9% of residents were female as of 2019 . Despite these advances, women lack proportionate representation in leadership positions in OHNS academic departments and specialty societies, though this may be improving among junior academic positions [ – ]. Termed “manels”, male-only speaking panels at major scientific conferences have been a recent focus in the literature. Women speakers were underrepresented across multiple medical and surgical specialty conferences, including in cross-sectional analyses of various American and Canadian society meetings [ – ]. In 2019, Nature Conferences and Springer Nature released a new code of conduct to formalize efforts to increase gender diversity, including no male-only organizing committees, no male-only panels, annual monitoring of progress, and sanctions when the code is not followed . Dr. Francis Collins, the National Institute of Health director, stated that women and other minorities were not equitably represented at major scientific conferences. He vowed to help “end the Manel tradition” by refusing to speak at a conference if attention to diversity was not given . Diversity in society meetings and panel-type presentations has multiple benefits. It has the potential to expand perspectives and several studies have shown that varied opinions may lead to better ideas, innovation, and an overall stronger panel . Women physicians have been shown to provide stellar patient care with excellent outcomes and have a place on these panels [ – ]. Increasing equitable representation of women and others helps perpetuate to attendees that individuals of all backgrounds are important members of the specialty society. Presentation at academic meetings and participation on scientific panels is also important for career advancement in academia. The presence of female representation helps decrease the “glass ceiling” effect noted for women in academia . Finally, this is an issue of justice and inclusivity . While there have been studies on gender diversity amongst speakers at key surgical conferences in the United States and Europe, there has not been published literature assessing this in our specialty in Canada. The Canadian Society of Otolaryngology-Head and Neck Surgery (CSO) is the major Otolaryngology society in Canada and encompasses all Otolaryngology subspecialties. Our aim was to determine the state of gender diversity amongst presenters and speakers at the annual CSO meetings. Scientific programs for the CSO Annual Meetings were obtained from their website ( www.entcanada.org ) from 2008 to 2020 by two independent groups of researchers at two Canadian institutions. Extracted information for each position included: participant name, gender, role, and subspecialty topic (General OHNS, Education, Laryngology, Pediatric, Otology, Head and Neck Surgery, Facial Plastics and Reconstructive Surgery (FPRS), Endocrinology, and Rhinology). CSO annual newsletters were also accessed to extract the name and gender of CSO executive leadership. A binary definition of gender (male or female) was chosen as a surrogate of diversity in the study population, composed of specialists trained in Otolaryngology, Otolaryngology trainees, other medical specialists, allied health members, and medical students. Gender was determined using an online search of Google Scholar, departmental websites, and public descriptions. If gender could not be determined from online information, co-authors and fellow panelists were contacted to determine this information. Leadership CSO Executive Membership was extracted from CSO annual newsletters and included members of the executive council, executive committee, regional representatives, and special interest group leaders. Each of these was defined as a leadership “opportunity spot” and the number of unique women occupying these roles was quantified. To quantify the degree of diversity, each position was counted as one “opportunity spot” as per Barinsky et al. to capture those who participate in several different roles. Invited speaking opportunities An invited speaking opportunity spot was defined as any named role in the CSO program other than paper session or poster presenter (i.e. session moderator). Of the opportunity spots occupied by a woman, the absolute number of women included was also assessed. The following roles were included: CSO president, scientific program chair, local arrangements chair (if provided), guest(s) of honour, guest speakers and special presenters, award winners, workshop presenters and panelists, and paper session chair. Composition of panels The composition of panels was separately analyzed and divided into male-only panels, female-only panels, or those with at least one female participant. The CSO meetings labelled sessions led by one or a small group of experts as “mini-workshops”, “workshops”, “courses” or “panels”. Among workshops with multiple presenters, those with two or fewer presenters were named “workshop chairs”, and those with three or more presenters were called “panelists”. Those who were designated “workshop chairs” with a separate panel were named as “workshop chairs”. All named non-otolaryngologists (including other medical specialists, allied health specialists, researchers) and non-Canadian otolaryngologists were included in the count. Descriptive statistical analysis was performed using SAS Software (Version 9.4, SAS Institute Inc., Cary, NC, USA), and consisted of counts and percentages. The two data sets produced by the two independent groups were merged into a single file to cross-check. A senior author (EG) reviewed the file and flagged any inconsistencies in the data, which were then investigated and corrected based on publicly available information. The senior author identified 48 errors (approximately 4%), which were corrected. Gender differences were analyzed using chi-square tests and logistic regression with odds ratios (OR) and 95% confidence intervals (95%CI). An alpha level of 0.05 was used to determine statistical significance. This project did not require ethics oversight as per article 2.2. of the Tri-Council Policy Statement (TCPS)-2 guidelines regarding the use of publicly available data for research purposes. CSO Executive Membership was extracted from CSO annual newsletters and included members of the executive council, executive committee, regional representatives, and special interest group leaders. Each of these was defined as a leadership “opportunity spot” and the number of unique women occupying these roles was quantified. To quantify the degree of diversity, each position was counted as one “opportunity spot” as per Barinsky et al. to capture those who participate in several different roles. An invited speaking opportunity spot was defined as any named role in the CSO program other than paper session or poster presenter (i.e. session moderator). Of the opportunity spots occupied by a woman, the absolute number of women included was also assessed. The following roles were included: CSO president, scientific program chair, local arrangements chair (if provided), guest(s) of honour, guest speakers and special presenters, award winners, workshop presenters and panelists, and paper session chair. The composition of panels was separately analyzed and divided into male-only panels, female-only panels, or those with at least one female participant. The CSO meetings labelled sessions led by one or a small group of experts as “mini-workshops”, “workshops”, “courses” or “panels”. Among workshops with multiple presenters, those with two or fewer presenters were named “workshop chairs”, and those with three or more presenters were called “panelists”. Those who were designated “workshop chairs” with a separate panel were named as “workshop chairs”. All named non-otolaryngologists (including other medical specialists, allied health specialists, researchers) and non-Canadian otolaryngologists were included in the count. Descriptive statistical analysis was performed using SAS Software (Version 9.4, SAS Institute Inc., Cary, NC, USA), and consisted of counts and percentages. The two data sets produced by the two independent groups were merged into a single file to cross-check. A senior author (EG) reviewed the file and flagged any inconsistencies in the data, which were then investigated and corrected based on publicly available information. The senior author identified 48 errors (approximately 4%), which were corrected. Gender differences were analyzed using chi-square tests and logistic regression with odds ratios (OR) and 95% confidence intervals (95%CI). An alpha level of 0.05 was used to determine statistical significance. This project did not require ethics oversight as per article 2.2. of the Tri-Council Policy Statement (TCPS)-2 guidelines regarding the use of publicly available data for research purposes. A total of 1874 opportunity spots were available during the annual CSO meetings from 2008 to 2020, of which 348 (18.6%) were filled by women (Table ). These were held by 92 unique women in total. There was an overall increase in the number and proportion of these positions held by women (Fig. ), from six leadership spots in 2008 (6.7%) to a peak of 50 spots in 2020 (23.7%). Leadership Among all CSO executive members, there were 448 men (83.0%) and 92 women (17.0%) over the studied period, encompassing 342 unique men and 63 unique females. The gender breakdown by position type, along with the number of unique individuals occupying these positions, is shown in Fig. . There was a significant difference between male and female representation in society executive members (p = 0.0009). Figure shows the change in gender representation in executive positions over the period studied, with the trend in unique individuals occupying these positions. Notably, there has been one female CSO president, and no female scientific program chairs during the period studied. Among the Guests of Honour, of which there are usually one or two per meeting, there has been only one female otolaryngologist chosen across all meetings. The CSO Awards Committee Chair, who also serves as chair of the annual Poliquin competition for resident research, has been a male surgeon until 2019 and 2020, when a female surgeon was elected to this role. From 2011 and 2014 onwards, various awards were given for lifetime achievement, recognition by Canadian region, and fellowship awards. Of the thirty-one awards, seven (22.6%) were awarded to women across all years studied. Invited speaking opportunities Overall, there were 1,136 invited speaking opportunities at CSO meetings between 2008 and 2020. Of these, 97 were part of workshops and 1,039 from panels. Females only represented 18.6% (18) of invited speakers at workshops and 18.6% (193) at panels. Across each CSO year, female representation in panels steadily increased until 2015, and then has remained constant at around 20 to 25% (Fig. ). There appears to be no discernible trend in female representation in workshops over the same period. The Scientific Program Committee consists of the CSO president, Scientific Program Chair, and Continuing Professional Development (CPD) Committee Chair. In the period studied, this committee included one woman (of 3–4 members) in 2008, 2013, and from 2015–2020. The larger Scientific Program Reviewer Committee consisted of 20–25 members representing all OHNS subspecialties, who reviewed blinded abstracts for selection of workshop/panel presenters, oral session presenters, and poster presenters. Data was available for 2018–2020 only, and there were seven female members in 2018, seven in 2019, and five in 2020. Composition of panels A total of 368 workshops (including workshops, mini-workshops, panels, courses, and CPD Corner sessions) were identified. There were 225 (61.1%) male-only panels (“manels”), while 9 (2.5%) were led by women only, and 134 (36.4%) workshops included at least one female surgeon (Fig. ). Chi-square analysis showed a significant difference between the proportion of male-only panels and those including any women (p = 0.0001). The CSO meeting in 2015 was the first year that there was a greater proportion of panels including at least one woman than those with exclusively male panelists (55.8% mixed panels), and this trend has continued for four out of six subsequent years. Female leadership was significantly underrepresented in many subspecialties (Table ). Conversely, laryngology and general OHNS workshops consistently had more female representation, even from 2008. There were only five instances, between 2008 and 2020, where females made up the majority of representatives in their discipline’s sessions compared to their male counterparts. Among all CSO executive members, there were 448 men (83.0%) and 92 women (17.0%) over the studied period, encompassing 342 unique men and 63 unique females. The gender breakdown by position type, along with the number of unique individuals occupying these positions, is shown in Fig. . There was a significant difference between male and female representation in society executive members (p = 0.0009). Figure shows the change in gender representation in executive positions over the period studied, with the trend in unique individuals occupying these positions. Notably, there has been one female CSO president, and no female scientific program chairs during the period studied. Among the Guests of Honour, of which there are usually one or two per meeting, there has been only one female otolaryngologist chosen across all meetings. The CSO Awards Committee Chair, who also serves as chair of the annual Poliquin competition for resident research, has been a male surgeon until 2019 and 2020, when a female surgeon was elected to this role. From 2011 and 2014 onwards, various awards were given for lifetime achievement, recognition by Canadian region, and fellowship awards. Of the thirty-one awards, seven (22.6%) were awarded to women across all years studied. Overall, there were 1,136 invited speaking opportunities at CSO meetings between 2008 and 2020. Of these, 97 were part of workshops and 1,039 from panels. Females only represented 18.6% (18) of invited speakers at workshops and 18.6% (193) at panels. Across each CSO year, female representation in panels steadily increased until 2015, and then has remained constant at around 20 to 25% (Fig. ). There appears to be no discernible trend in female representation in workshops over the same period. The Scientific Program Committee consists of the CSO president, Scientific Program Chair, and Continuing Professional Development (CPD) Committee Chair. In the period studied, this committee included one woman (of 3–4 members) in 2008, 2013, and from 2015–2020. The larger Scientific Program Reviewer Committee consisted of 20–25 members representing all OHNS subspecialties, who reviewed blinded abstracts for selection of workshop/panel presenters, oral session presenters, and poster presenters. Data was available for 2018–2020 only, and there were seven female members in 2018, seven in 2019, and five in 2020. A total of 368 workshops (including workshops, mini-workshops, panels, courses, and CPD Corner sessions) were identified. There were 225 (61.1%) male-only panels (“manels”), while 9 (2.5%) were led by women only, and 134 (36.4%) workshops included at least one female surgeon (Fig. ). Chi-square analysis showed a significant difference between the proportion of male-only panels and those including any women (p = 0.0001). The CSO meeting in 2015 was the first year that there was a greater proportion of panels including at least one woman than those with exclusively male panelists (55.8% mixed panels), and this trend has continued for four out of six subsequent years. Female leadership was significantly underrepresented in many subspecialties (Table ). Conversely, laryngology and general OHNS workshops consistently had more female representation, even from 2008. There were only five instances, between 2008 and 2020, where females made up the majority of representatives in their discipline’s sessions compared to their male counterparts. Our data demonstrated that female surgeons held nearly a quarter of the total speaking positions at the CSO meetings from 2008 to 2020. The most common roles held were paper session chairs and panelists (a workshop led by three or more specialists). The proportion of male-only panels and workshops (“manels”) did decrease over time, but constituted over half of all workshops in 2020. Our results align with the current literature highlighting the differential representation of women in academic conferences, particularly in medicine and in surgical subspecialties [ – , , ]. Barinsky et al. were the first and only group to publish on the gender disparity of OHNS conference speakers in the US. They showed an increase in opportunity spots occupied by women from 11.5% in 2003 to 29.5% in 2019, but that the number of unique women occupying these spots was only 24.4% of the total . Women were more likely to be oral session moderators or panelists instead of speakers, executive board members, or honoured guests. This was mirrored in our results in the Canadian population. It is promising that we have seen a trend toward increasing female representation over the past 12 years, especially amongst workshop chairs and panelists. Part of this may be attributed to an increase in the number of female otolaryngologists in Canada, from only 10% in 2000 to 24% in 2019 , which does approximate the proportion of female speakers in those years. An increase in female representation in leadership positions was seen starting in 2014. We hypothesize that there may be several contributing factors—the opening of more opportunities to present workshops, a critical mass of female staff and trainees moving through the pipeline, and the development of a formal Women in Otolaryngology section of the CSO. “Mini-workshops” and “How I Do It” workshops were first introduced at CSO meetings in 2014, though they were not always present in subsequent years. The increase in opportunities, particularly of smaller workshops, may be a way of increasing opportunities for participation from more junior staff, a pool of specialists more likely to include women . Our results also showed increased female representation in broader subspecialties starting in 2014. The proportion of Canadian and American women pursuing academic fellowships in surgical specialties has increased over the past several decades . From 2011 to 2020, the number of Canadian female otolaryngologists who have completed subspecialty fellowships has increased. Still, the gender gap was largest in head and neck surgery, rhinology, and otology, where only 28%, 29%, and 22% were female, respectively. Pediatric OHNS and laryngology were the only two fellowships with a female predominance. However, the absolute number of female graduates of otology, rhinology, and facial plastic surgery ranged from 5–10 over 2011 to 2020, whereas the numbers of female graduates of head and neck surgery and pediatric OHNS were similar at 15–20, and more than 30 new general OHNS practitioners were female . This correlates with our findings that there was less female representation in facial plastics and rhinology workshops. Increasing mentorship opportunities and visibility of women and minorities can lead to increased participation in academic activities by junior staff and trainees [ , , ]. The CSO Women in Otolaryngology (WIO) group was established in 2014, and coincides with the increased female presence at the annual meeting. The WIO hosts networking sessions with female staff and trainees from across the country and offers opportunities for society leadership, mentorship, and creates a sense of community. This may be critical for incoming and junior trainees navigating transitions and seeking career advancement opportunities. Deliberate initiatives such as this will continue to raise awareness of gender disparities in our specialty, and encourage females to pursue academic aspirations, an essential first step toward increasing representation. In 2019, 41.9% of OHNS trainees were female, and 45.3% of OHNS CaRMS applicants were female, indicating that future generations may see greater gender parity. We expect to see a similar trend in our speakers and conference leadership as more women become involved in academic endeavours. Literature shows that despite increasing proportions of female trainees and surgeons, women are still underrepresented in OHNS leadership and senior academic roles (such as assistant, associate, and full professor) compared to men [ , – ], and when compared to all specialties in medicine . However, a lag effect may be contributing to this phenomenon, in that it will take several years for the newly admitted trainees to eventually progress through their careers to leadership positions. To close the gender disparity amongst conference speakers and presenters, there must be continued efforts to close the gender gap among trainees entering the specialty, increase support for women to pursue research and academia , and develop initiatives to recruit and retain female faculty . Studies from Arora et al., Lu et al., Gerull et al., and Zaza et al. assessed the proportion of female speakers at an aggregate of over a hundred academic medical and surgical conferences across multiple specialties. They examined the correlation between the proportion of women on conference planning committees and female speakers [ , , , ]. There was a statistically significant positive correlation between the proportion of women on planning committees and society leadership and the proportion of female speakers, based on univariable analysis and still significant after controlling for regional gender balance of the specialty. For our study, the scientific planning committee information was only fully available from 2018 onwards, and while it would have been interesting to support this literature with our study, this analysis was not possible in a meaningful way. Increasing the proportion of women on conference planning committees may be a simple yet effective way to reduce the gender disparity amongst speakers [ , , , , ]. Our conference has a blinded selection process, with workshop chairs and presenters submitting blinded abstracts to be selected by the scientific planning committee. The gender disparity in workshops may not be related to gendered selection bias, but rather the number of women conducting research and their research productivity. While a 2013 study reported that women in their early career produce less research output, but at senior levels, they equal or exceed the research productivity of men , a more recent report from 2020 indicates that female otolaryngologists are maintaining research productivity in their early careers (less than 15 years into practice) to keep closer pace with men. However, women continued to lag behind men in research productivity in some subspecialties such as head and neck oncology, laryngology, and pediatrics . There are likely numerous contributing factors affecting research productivity, but the evolution of societal gender roles with more equal sharing of domestic duties and child care, greater financial and administrative support for research, and increasing mentorship opportunities will have a positive impact [ , , , ]. This study is only one component in achieving greater equity and diversity: raising awareness of disparities. Moving forward, we must consider systems-level change to improve gender parity , ]. It is critical to further assess the factors impacting speaker invitations for conferences, and women’s submissions for these opportunities. These may include personal and professional barriers, the proportion of women in the specialty, research productivity, visibility as a leader in the field, gender bias, and gender composition of the conference planning committee , , , , ]. Regular reassessment of female representation at these conferences is a crucial checkpoint ]. Ongoing analysis of equity at national society and departmental levels may be facilitated by designated diversity and inclusion leads or committees, and including these stakeholders in conference and departmental planning . With the higher proportion of women amongst younger otolaryngologists and trainees, continuing to improve the gender gap will result in a larger pool from which to select our conference leadership and presenters. Study limitations The results of the study must be interpreted within the confines of the research methodology. This study is limited in that it is a retrospective review of various publicly available databases, and thus the authors were unable to confirm the accuracy or validity of this data. Data around the proportion of abstracts submitted by female presenters versus the proportion accepted for presentation was not available. We also used a binary definition of biological sex as a surrogate for gender identity, which exists on a spectrum, and the biological sex of presenters was recorded based on public information and/or confirmation by colleagues. Lastly, the present study did not capture the many other diversity factors in the workforce. The results of the study must be interpreted within the confines of the research methodology. This study is limited in that it is a retrospective review of various publicly available databases, and thus the authors were unable to confirm the accuracy or validity of this data. Data around the proportion of abstracts submitted by female presenters versus the proportion accepted for presentation was not available. We also used a binary definition of biological sex as a surrogate for gender identity, which exists on a spectrum, and the biological sex of presenters was recorded based on public information and/or confirmation by colleagues. Lastly, the present study did not capture the many other diversity factors in the workforce. The proportion of women in speaking roles at the annual Canadian Society of Otolaryngology-Head and Neck Surgery meetings has generally increased with time, particularly among panelists. This has led to a decrease in male-only speaking panels and workshops. However, there has been a slower growth rate of unique women in leadership speaker roles. There is still room for increasing gender diversity at the major Canadian OHNS meeting. Academic mentorship, equitable allocation of opportunities and resources, and equal encouragement of research endeavours for both men and women may help contribute to this.
Effects of an online program including mindfulness, exercise therapy and patient education compared to online exercise therapy and patient education for people with Patellofemoral Pain: protocol for a randomized clinical trial	b49af4bc-ebc1-4804-821b-1702ad88ab77	10173555	Patient Education as Topic[mh]	Patellofemoral pain (PFP) is characterized by the presence of pain around or behind the patella, exacerbated by activities that increase patellofemoral joint loading . PFP is prevalent in young adults and adolescents (22.7% and 28.9%, respectively), with women being twice as likely to develop PFP than men . People with PFP typically report reduced levels of physical activity , functional capacity , and quality of life . Evidence also indicates that symptoms can be long-lasting, with 50 to 91% of people with PFP experiencing persistent pain up to 18 years after the initial diagnosis . According to the International Association for the Study of Pain (IASP) , the concept of pain encompasses more than just physical-chemical aspects of nociception. Sociocultural, emotional, and cognitive factors can also contribute to the worsening of pain and dysfunction [ – ]. This seems to apply to people with PFP as they present with altered psychological factors such as anxiety, depression, pain catastrophizing, and kinesiophobia . Moderate correlations between psychological factors with pain and disability have also been reported in people with PFP . Current recommendations on the “best management” for people with PFP are exercise therapy and patient education . Although effective in the short-term , unsatisfactory long-term prognosis remains an issue, with 57% of the people with PFP reporting unfavorable recovery at 5–8 years . This could be a reflection of the lack of consideration for psychological factors during rehabilitation. Doménech et al. have reported that patients who experience the largest decreases in pain catastrophizing, kinesiophobia, anxiety, and depression also experience greater improvement in pain and disability after a purely biomedical treatment. It has been suggested that the addition of co-interventions to address psychological factors, for example, cognitive-behavioral treatment, reassurance, and graded exposure to activity may enhance rehabilitation outcomes, such as pain and function, in individuals with PFP . The specific mechanisms by which changes in psychological factors influence physical function are not well known, however, positive coping cognitions and emotional states are thought to confer resilience to pain and resourcefulness to improve adherence to active treatments and physical activity . Therefore, further investigation is required to understand the potential additional effects of interventions that may influence psychological factors to the current best management of PFP. Mindfulness-based interventions (MBI) were developed to assist people in managing stress, anxiety and chronic pain . This evidence-based program has been increasingly used for a variety of musculoskeletal disorders . Mindfulness is defined as a form of bringing attention, friendly curiosity, and non-judgmental awareness to body sensations, thoughts, and emotions in order to reduce suffering or distress and to increase wellbeing . Previous studies have demonstrated specific brain modifications in neuroimaging evaluation in experience practitioners, such as increased grey matter volume in the frontal lobe and relatively decreased posterior cingulate cortex activity compared to novice practitioners [ – ]. This finding suggests an existence of a neural network responsible for the positive effects of MBI practices including, but not restricted to information processing, mind wondering regulation and adaptative behavior . Therefore, as part of a rehabilitation program, mindfulness may promote a better focus on rehabilitation and influence several psychological factors such as anxiety, pain catastrophizing and avoidance behaviors [ – ]. In this context, a recent study has reported that adding an MBI to exercise therapy promoted lower levels of pain during running and stepping, less functional limitations and lower pain catastrophizing as compared to exercise alone in female runners with PFP. However, this study was performed exclusively on female recreational runners with PFP, which limits the generalizability to the general population with PFP. In addition, patient education was not provided in this study, which is of utmost importance to PFP . As such, more studies are needed to investigate the effects of MBI in addition to exercise therapy and patient education in people with PFP. Internet-based interventions have been recently promoted due to their potential to overcome geographical barriers, increase access to health services, and provide alternative means to continue treating patients whenever face-to-face encounters are precluded . There is evidence supporting the use of internet-based interventions for the treatment of several conditions as they may provide similar improvements in pain and function compared to face-to-face treatments . Furthermore, internet-based interventions may allow patients to assume a more active role in their rehabilitation, encouraging strategies as self-management and self-efficacy . Online MBI has also been shown to be feasible and effective in reducing psychological factors such as stress, anxiety and depression . However, few studies have investigated the effects of internet-based interventions for PFP, especially including components targeting psychological factors such as MBI. The aims of this randomized clinical trial are: (i) to investigate the immediate (8-week) and long-term (12-month) effects of adding the MBI program to an 8-week online intervention comprised of exercise therapy and patient education on self-reported recovery, pain, function, and psychological factors in people with PFP; (ii) to investigate whether changes in psychological factors mediate changes in pain and function. We hypothesize that people in the Mindfulness group will experience greater decreases in pain, as well as higher improvements in function at 8 weeks and 12 months. We also hypothesize that psychological factors such as kinesiophobia and pain catastrophizing will mediate changes in pain and function. Protocol elaboration This protocol is reported according to the SPIRIT statement (Standard Protocol Items: Recommendations for Interventional Trials) and CONSORT Statement . Study design This is an assessor-blinded, parallel, two-arm randomized clinical trial with 12-month follow-up. All participants will receive an identical internet-based exercise therapy and patient education intervention, with one group receiving additional online MBI program. Details of participants time schedule according to the SPIRIT recommendations are available in Additional file . Participants and consent People with knee pain will be recruited through social media to participate in this study. All participants who meet the eligibility criteria will be informed about the nature of the research and receive an online consent form, prepared in accordance with the declaration of Helsinki and the 466/12 resolution of the National Health Council. Eligibility criteria The eligibility criteria were designed according to the most recent PFP consensus statement on clinical examination of PFP and will be completed through an online form. Participants’ eligibility will be confirmed by a physiotherapist with > 3 years of clinical experience managing people with PFP. All assessments, including eligibility criteria and outcomes measures (baseline, follow-up 1 and follow-up 2), will be performed through online forms. No face-to-face physical examinations will be performed. However, if further details are required to confirm the diagnosis, an online meeting between the physiotherapist and the participant will be performed. Inclusion criteria Participants will be required to meet the following criteria in order to be included in this study: (i) age between 18 and 40 years old; (ii) self-reported anterior knee pain (unilateral or bilateral) when performing at least two of the following activities: prolonged sitting, squatting, kneeling, running, ascending and descending stairs, jumping and landing ; (iii) self-reported anterior knee pain with insidious onset lasting at least 6 months ; (iv) worst self-reported pain in the previous month corresponding to at least 30 mm in a 100 mm visual analogue scale (VAS) . Exclusion criteria Participants will be excluded if they meet any of the following criteria: (i) self-reported anterior knee pain caused by trauma on the knee; (ii) self-reported history of patellar dislocation or subluxation; (iii) self-reported history of meniscal injury, ligament instability or patellar tendinopathy; (iv) history of osteoarthritis in any lower limb joint; (v) history of surgery on any lower limb joint; (vi) patient-reported rheumatic or neurologic disease; (vii) physiotherapy treatment for PFP during the preceding 6 months; (viii) answer “yes” on any questions on the PAR-Q physical activity readiness questionnaire ; (ix) history of current or past psychosis, major depressive episode, suicide attempt, post-traumatic stress disorder, bipolar disorder, manic episode, or substance dependency. Randomization and blinding The randomization list will be developed by an investigator who will not be involved in the recruitment and assessment of the participants. Randomization codes will be generated in blocks, using a custom list on the website ( https://www.sealedenvelope.com/ ), and the participants will be randomized with a 1:1 allocation to one of the two interventions. Sealed opaque envelopes, sequentially enumerated, will be used to conceal the allocation. After the baseline assessment, the investigator will open the envelope containing the participant’s random code to ensure the allocation of the participant will be concealed. Due to the nature of the interventions, participants will be informed about the type of intervention. Therefore, the study cannot be considered double-blind . The assessor will be blinded to the allocation of participants. Outcome measures Outcome measures will be assessed online at baseline, intervention endpoint (8 weeks – follow-up 1), and 12 months after intervention completion (follow-up 2). Demographic data (e.g. age, gender, duration of symptoms) will be recorded at the baseline assessment. Primary outcomes Self-reported recovery The 7-point Likert global rating of change scale (GROC) is a measure of treatment effect that has been previously used in people with PFP . The participants will be asked “How would you describe your knee pain now, compared to before you began the treatment?” The answers are marked on a 7-point Likert scale (much better, better, a little better, no change, a little worse, worse, much worse). The answers will be dichotomized in “successful” and “unsuccessful”. A successful outcome will be defined as being much better or better. Pain Participants’ self-reported pain level over the previous week will be measured with a 100 mm VAS . The VAS consists of a 0 to 100 mm horizontal line, with 0 representing “no pain” (0 mm) and 100 representing “extreme pain”. Participants will be instructed to draw a perpendicular line on the scale at the position that indicates the severity of usual and worst knee pain over the preceding week. The VAS is valid and reliable for assessing people with PFP . Secondary outcomes Self-reported function The Anterior Knee Pain Scale (AKPS) is a valid and reliable 13-item questionnaire that evaluates subjective function related to PFP . Participants will complete the AKPS based on their perceived knee condition at the prior week. The total score for the AKPS ranges from 0 (maximal disability) to 100 (no disability), with the total score being used for statistical analysis. Anxiety and depression The Hospital Anxiety and Depression Scale (HADS) is a 14-item questionnaire that evaluates the emotional state of the patient and identifies cases of mild, moderate and severe anxiety and/or depression disorders . The HADS consists of two subscales, which assess anxiety (7 items) and depression (7 items) separately. Participants will be asked to answer each item on a 4-point Likert scale (0–3), with scores ranging from 0 to 21 for anxiety and 0 to 21 for depression. Scores between 0 and 7 are classified as normal, between 8 and 10 as mild, between 11 and 14 as moderate, and between 15 and 21 as severe . Kinesiophobia The Tampa scale for kinesiophobia is a self-administered questionnaire that assesses pain-related fear associated with the avoidance behaviors, movements and physical activity . It contains 17 statements with answers in a 4-point Likert scale: Strongly disagree, Partially disagree, Partially agree and Totally agree. Participants will be instructed to choose the option according to how much they agree with each statement. The score ranges from 17 to 68 and the higher the score, the higher the fear . Pain catastrophizing The Pain Catastrophizing Scale (PCS) is a 13-item questionnaire that consists of describing thoughts and feelings that individuals experience when they have pain . Participants will be instructed to reflect on the experiences caused by pain in the past and indicate their perception on a 5-point Likert scale, where (0) represents “not at all” and “all the time”. The higher the score, the greater the pain catastrophizing . Pain self-efficacy The Chronic Pain Self-Efficacy Scale (CPSS) is a 22-item self-administered questionnaire that assesses the perception of self-efficacy and the ability to deal with the consequences of pain in patients with chronic pain . The CPSS contains 3 domains: pain control, physical function and symptom control. Participants will be asked to answer how much they agree with each of the items arranged on a Likert scale ranging from 10 to 100 points. The score ranges from 30 to 300, where the higher the score, the better the self-efficacy. Self-reported physical activity level The International Physical Activity Questionnaire short form is a 9-item questionnaire that assesses how many days and hours the participants usually spent per week doing several activities . The physical activity level will be determined by the total of vigorous and moderate exercise in the previous week and calculated according to previous studies . Interventions After the baseline assessment, participants will receive immediate access to a WEB platform developed by one of the authors available at http://www.stepslab.com.br/ where the interventions will be delivered. An individualized online meeting will be performed between participants and a physiotherapist not involved in data analysis to guide them regarding platform usage, deadlines and the importance of committing to the intervention. Participants will have access to the online interventions for 8 weeks, which will be immediately ceased at the end of the period. The exercise and education contents will be developed and pre-recorded by two physiotherapists with more than three years of clinical experience using evidence-based material. The MBI content will be developed and pre-recorded by a certified mindfulness teacher with more than 15 years of experience and revised by a psychologist to ensure psychological appropriateness. Details of the interventions according to the TIDier checklist are available in Additional file . Participants will only have access to content related to the group to which they were allocated (restricted area). Each session will be released for access on pre-defined dates relative to participants’ entry into the trial (immediately after baseline assessment). This will be performed by an investigator who will not be involved in the recruitment and assessment of the participants. The system will only allow opening of the next session if the participant had ended the previous one. At the end of each session, participants will be required to report their current level of pain on an online VAS scale, if there was any adverse event during or after the intervention, the level of satisfaction and the level of perceived exertion on a 15-point Borg scale , ranging from 6 (no exertion at all) to 20 perceiving a (maximal exertion). Participants will be instructed to not seek any other kind of knee pain treatment during the study, except in emergency cases. If necessary, participants will be able to contact the therapist through the platform’s e-mail. An outline of the study procedures is summarized in Fig. . Control group Participants allocated to this group will receive two pre-recorded video classes per week according to their availability (e.g. on Tuesdays and Thursdays) with Exercise therapy and Patient education contents (lasting 35–50 min in total). Exercise therapy The exercise therapy component will include the prescription of the exercises according to the American College of Sports Medicine Position Statement and the PFP consensus . The mean duration of the exercise videos will be approximately 30 min. There will be a one-day break between the sessions, to respect recovery time. Exercise therapy will aim to improve muscle performance, movement coordination and mobility . This intervention will target hip, knee, and ankle muscles. Exercises will be progressed in phases every two weeks (i.e., intensity, type of exercise, technique or repetitions). During the exercise, affirmative and encouraging audio messages will be displayed in order to motivate the participant to finish the session with as much effort as possible. The intensity of the exercises will be monitored through the Borg scale and must remain between 12 and 16 points. If the participant report exertion values outside this range, the exercise intensity will be modified. The full description of the exercises is available in Additional file . Patient education Educational pre-recorded video classes will cover the following topics. Week 1: Understanding my knee: anatomy and biomechanics of the knee and the relationship between pain and injury. Week 2: Understanding my knee pain: incidence and prevalence of PFP; why my knee hurts; biomechanical and psychological factors of PFP; prognosis and diagnosis of PFP. Week 3: “Too Much, too soon”: how high volume or high load intensity during daily activities or sports can lead to knee pain. Week 4: Myths and truths about my knee: Knee crepitus and movements considered harmful to the joint; fear of movement; and imaging exams. Week 5: Aspects of quality of life that influence pain: sleep quality, weight control, confidence, coping strategies, and mental health. Week 6: Taking care of my own pain: self-management of pain, motivation and responsibility for your own health. Week 7: Available treatment options: importance of adherence to active treatments, treatments that work, treatments that do not work and load management. Week 8: Take-home message: What should I do after the treatment? Motivation, habits change, behavioral change and the need to remain active (exercise/treatment). The mean duration of the patient education videos will be approximately 6 min. Mindfulness group Participants allocated to this group will receive the same intervention as the control group and an additional pre-recorded video class with MBI content according to their availability (e.g. on Mondays and Wednesdays they receive exercise therapy and Patient education contents; on Fridays they receive mindfulness contents, lasting 35–50 min). Mindfulness-based intervention The MBI component will be adapted from Mindfulness-Based Health Promotion (MBHP) model to suit patients with PFP and the online assessments of the present study. MBHP is an 8-week MBI developed in 2009 in the context of health promotion and quality of life . Inspired by Jon Kabat-Zinn’s original protocol—mindfulness-based stress reduction, MBSR (University of Massachusetts Medical Center, USA) —it also aggregates elements of other protocols, such as the MBCT (University of Toronto/Canada; University of Cambridge and Oxford University/United Kingdom) , the mindfulness programs of the Breathworks Institute (United Kingdom) , and mindfulness-based relapse prevention (MBRP, University of Washington, USA . The MBHP has been extensively used in a variety of health conditions [ – ]. Video classes will include formal and informal mindfulness practices. The description of the MBI is available in Additional file . The following themes will be covered. Week 1: Breaking the automatism. Week 2: Body awareness. Week 3: Leaving the mind and inhabiting the body. Week 4: Raising awareness. Week 5: Letting go. Week 6: Dealing with challenges and letting go of resistance. Week 7: Mindfulness and self-care. Week 8: A look to the future. In addition, participants included in this group will receive daily reminders and additional material (audios) to continue to practice formal and informal MBHP daily. The participants in this group will be encouraged to practice diaries. Adherence In order to improve the adherence to the treatments, before each session, participants of both groups will receive an automatic reminder via SMS and/or email before each session. Participants’ adherence to the interventions will be monitored through the number of accesses (date and hour), time connected to the platform, sessions visualized, number of sessions finalized, number of drop outs and others. Adverse events At the end of each video class, in a pop-up window, all participants will report the intensity of their pain and if there were any adverse events during the session. Participants will be able to contact a therapist through the platform’s email at any time. In case of a severe adverse event related to exercise therapy or MBI (e.g. strains, sprains, persistent severe pain, psychosis, mania, traumatic memories), the participant will be referred to a qualified healthcare professional for further investigation. Sample size and power The sample size calculation was performed based on the usual pain intensity data from Bagheri et al. . Considering a difference between groups of 4.2 mm and a standard deviation of 5.4 mm, with an α of 0.05 and β of 0.20, 26 participants per groups are required (52 in total). We will recruit 31 participants per trial arm to allow for up to a 20% drop-out rate at 12 months. Statistical analysis Statistical analysis will be performed by the blinded assessor using SPSS software (IBM version 23, SPSS Inc., Chicago, Il). Descriptive statistics will be computed for all variables (e.g. mean, standard deviation). Data will be tested for normal distribution by the Shapiro-Wilk test. Chi-square tests will be performed to compare self-reported recovery (successful x unsuccessful) between groups. For continuous data, the effects of group, time and their interaction will be assessed with linear mixed models. Intraclass Correlation Coefficients will be used to determine the amount of variance explained by random effects . The Bonferroni-adjusted post hoc test will be performed for multiple pairwise comparisons where appropriate. Effect sizes (95% CI) (Cohen’s d) will also be calculated and interpreted as follows: Cohen’s = 0.2 ‘small effect’; = 0.5 ‘moderate effect’; = 0.8 ‘large effect’ and = 1.3 ‘very large effect’ . Intent-to-treat analyses will be performed for all outcomes. Multiple imputation will be used to account for missing data if the proportion of missing data is > 5% . For all tests, an α level of 0.05 two-tailed will be adopted to indicate statistical significance. The mediation effects will be assessed following the 3-variable framework described by MacKinnon et al. . In this model, the intervention condition is assumed to have both direct and indirect paths to the changes in clinical outcomes. The indirect path passes through the potential mediators (anxiety, depression, kinesiophobia, pain catastrophizing and pain self-efficacy). Three multiple regressions will be performed: to test the association between the predictor (i.e., interventions) and the outcomes (i.e., pain and function); to test the association between the predictor and the potential mediators and to test the association between the potential mediators and the outcomes after controlling for the predictor. Then, it will be observed whether the association of the predictor with the outcome after controlling for potential mediators will be smaller than observed in the first regression. This protocol is reported according to the SPIRIT statement (Standard Protocol Items: Recommendations for Interventional Trials) and CONSORT Statement . This is an assessor-blinded, parallel, two-arm randomized clinical trial with 12-month follow-up. All participants will receive an identical internet-based exercise therapy and patient education intervention, with one group receiving additional online MBI program. Details of participants time schedule according to the SPIRIT recommendations are available in Additional file . People with knee pain will be recruited through social media to participate in this study. All participants who meet the eligibility criteria will be informed about the nature of the research and receive an online consent form, prepared in accordance with the declaration of Helsinki and the 466/12 resolution of the National Health Council. The eligibility criteria were designed according to the most recent PFP consensus statement on clinical examination of PFP and will be completed through an online form. Participants’ eligibility will be confirmed by a physiotherapist with > 3 years of clinical experience managing people with PFP. All assessments, including eligibility criteria and outcomes measures (baseline, follow-up 1 and follow-up 2), will be performed through online forms. No face-to-face physical examinations will be performed. However, if further details are required to confirm the diagnosis, an online meeting between the physiotherapist and the participant will be performed. Participants will be required to meet the following criteria in order to be included in this study: (i) age between 18 and 40 years old; (ii) self-reported anterior knee pain (unilateral or bilateral) when performing at least two of the following activities: prolonged sitting, squatting, kneeling, running, ascending and descending stairs, jumping and landing ; (iii) self-reported anterior knee pain with insidious onset lasting at least 6 months ; (iv) worst self-reported pain in the previous month corresponding to at least 30 mm in a 100 mm visual analogue scale (VAS) . Participants will be excluded if they meet any of the following criteria: (i) self-reported anterior knee pain caused by trauma on the knee; (ii) self-reported history of patellar dislocation or subluxation; (iii) self-reported history of meniscal injury, ligament instability or patellar tendinopathy; (iv) history of osteoarthritis in any lower limb joint; (v) history of surgery on any lower limb joint; (vi) patient-reported rheumatic or neurologic disease; (vii) physiotherapy treatment for PFP during the preceding 6 months; (viii) answer “yes” on any questions on the PAR-Q physical activity readiness questionnaire ; (ix) history of current or past psychosis, major depressive episode, suicide attempt, post-traumatic stress disorder, bipolar disorder, manic episode, or substance dependency. The randomization list will be developed by an investigator who will not be involved in the recruitment and assessment of the participants. Randomization codes will be generated in blocks, using a custom list on the website ( https://www.sealedenvelope.com/ ), and the participants will be randomized with a 1:1 allocation to one of the two interventions. Sealed opaque envelopes, sequentially enumerated, will be used to conceal the allocation. After the baseline assessment, the investigator will open the envelope containing the participant’s random code to ensure the allocation of the participant will be concealed. Due to the nature of the interventions, participants will be informed about the type of intervention. Therefore, the study cannot be considered double-blind . The assessor will be blinded to the allocation of participants. Outcome measures will be assessed online at baseline, intervention endpoint (8 weeks – follow-up 1), and 12 months after intervention completion (follow-up 2). Demographic data (e.g. age, gender, duration of symptoms) will be recorded at the baseline assessment. Primary outcomes Self-reported recovery The 7-point Likert global rating of change scale (GROC) is a measure of treatment effect that has been previously used in people with PFP . The participants will be asked “How would you describe your knee pain now, compared to before you began the treatment?” The answers are marked on a 7-point Likert scale (much better, better, a little better, no change, a little worse, worse, much worse). The answers will be dichotomized in “successful” and “unsuccessful”. A successful outcome will be defined as being much better or better. Pain Participants’ self-reported pain level over the previous week will be measured with a 100 mm VAS . The VAS consists of a 0 to 100 mm horizontal line, with 0 representing “no pain” (0 mm) and 100 representing “extreme pain”. Participants will be instructed to draw a perpendicular line on the scale at the position that indicates the severity of usual and worst knee pain over the preceding week. The VAS is valid and reliable for assessing people with PFP . Secondary outcomes Self-reported function The Anterior Knee Pain Scale (AKPS) is a valid and reliable 13-item questionnaire that evaluates subjective function related to PFP . Participants will complete the AKPS based on their perceived knee condition at the prior week. The total score for the AKPS ranges from 0 (maximal disability) to 100 (no disability), with the total score being used for statistical analysis. Anxiety and depression The Hospital Anxiety and Depression Scale (HADS) is a 14-item questionnaire that evaluates the emotional state of the patient and identifies cases of mild, moderate and severe anxiety and/or depression disorders . The HADS consists of two subscales, which assess anxiety (7 items) and depression (7 items) separately. Participants will be asked to answer each item on a 4-point Likert scale (0–3), with scores ranging from 0 to 21 for anxiety and 0 to 21 for depression. Scores between 0 and 7 are classified as normal, between 8 and 10 as mild, between 11 and 14 as moderate, and between 15 and 21 as severe . Kinesiophobia The Tampa scale for kinesiophobia is a self-administered questionnaire that assesses pain-related fear associated with the avoidance behaviors, movements and physical activity . It contains 17 statements with answers in a 4-point Likert scale: Strongly disagree, Partially disagree, Partially agree and Totally agree. Participants will be instructed to choose the option according to how much they agree with each statement. The score ranges from 17 to 68 and the higher the score, the higher the fear . Pain catastrophizing The Pain Catastrophizing Scale (PCS) is a 13-item questionnaire that consists of describing thoughts and feelings that individuals experience when they have pain . Participants will be instructed to reflect on the experiences caused by pain in the past and indicate their perception on a 5-point Likert scale, where (0) represents “not at all” and “all the time”. The higher the score, the greater the pain catastrophizing . Pain self-efficacy The Chronic Pain Self-Efficacy Scale (CPSS) is a 22-item self-administered questionnaire that assesses the perception of self-efficacy and the ability to deal with the consequences of pain in patients with chronic pain . The CPSS contains 3 domains: pain control, physical function and symptom control. Participants will be asked to answer how much they agree with each of the items arranged on a Likert scale ranging from 10 to 100 points. The score ranges from 30 to 300, where the higher the score, the better the self-efficacy. Self-reported physical activity level The International Physical Activity Questionnaire short form is a 9-item questionnaire that assesses how many days and hours the participants usually spent per week doing several activities . The physical activity level will be determined by the total of vigorous and moderate exercise in the previous week and calculated according to previous studies . Self-reported recovery The 7-point Likert global rating of change scale (GROC) is a measure of treatment effect that has been previously used in people with PFP . The participants will be asked “How would you describe your knee pain now, compared to before you began the treatment?” The answers are marked on a 7-point Likert scale (much better, better, a little better, no change, a little worse, worse, much worse). The answers will be dichotomized in “successful” and “unsuccessful”. A successful outcome will be defined as being much better or better. Pain Participants’ self-reported pain level over the previous week will be measured with a 100 mm VAS . The VAS consists of a 0 to 100 mm horizontal line, with 0 representing “no pain” (0 mm) and 100 representing “extreme pain”. Participants will be instructed to draw a perpendicular line on the scale at the position that indicates the severity of usual and worst knee pain over the preceding week. The VAS is valid and reliable for assessing people with PFP . The 7-point Likert global rating of change scale (GROC) is a measure of treatment effect that has been previously used in people with PFP . The participants will be asked “How would you describe your knee pain now, compared to before you began the treatment?” The answers are marked on a 7-point Likert scale (much better, better, a little better, no change, a little worse, worse, much worse). The answers will be dichotomized in “successful” and “unsuccessful”. A successful outcome will be defined as being much better or better. Participants’ self-reported pain level over the previous week will be measured with a 100 mm VAS . The VAS consists of a 0 to 100 mm horizontal line, with 0 representing “no pain” (0 mm) and 100 representing “extreme pain”. Participants will be instructed to draw a perpendicular line on the scale at the position that indicates the severity of usual and worst knee pain over the preceding week. The VAS is valid and reliable for assessing people with PFP . Self-reported function The Anterior Knee Pain Scale (AKPS) is a valid and reliable 13-item questionnaire that evaluates subjective function related to PFP . Participants will complete the AKPS based on their perceived knee condition at the prior week. The total score for the AKPS ranges from 0 (maximal disability) to 100 (no disability), with the total score being used for statistical analysis. Anxiety and depression The Hospital Anxiety and Depression Scale (HADS) is a 14-item questionnaire that evaluates the emotional state of the patient and identifies cases of mild, moderate and severe anxiety and/or depression disorders . The HADS consists of two subscales, which assess anxiety (7 items) and depression (7 items) separately. Participants will be asked to answer each item on a 4-point Likert scale (0–3), with scores ranging from 0 to 21 for anxiety and 0 to 21 for depression. Scores between 0 and 7 are classified as normal, between 8 and 10 as mild, between 11 and 14 as moderate, and between 15 and 21 as severe . Kinesiophobia The Tampa scale for kinesiophobia is a self-administered questionnaire that assesses pain-related fear associated with the avoidance behaviors, movements and physical activity . It contains 17 statements with answers in a 4-point Likert scale: Strongly disagree, Partially disagree, Partially agree and Totally agree. Participants will be instructed to choose the option according to how much they agree with each statement. The score ranges from 17 to 68 and the higher the score, the higher the fear . Pain catastrophizing The Pain Catastrophizing Scale (PCS) is a 13-item questionnaire that consists of describing thoughts and feelings that individuals experience when they have pain . Participants will be instructed to reflect on the experiences caused by pain in the past and indicate their perception on a 5-point Likert scale, where (0) represents “not at all” and “all the time”. The higher the score, the greater the pain catastrophizing . Pain self-efficacy The Chronic Pain Self-Efficacy Scale (CPSS) is a 22-item self-administered questionnaire that assesses the perception of self-efficacy and the ability to deal with the consequences of pain in patients with chronic pain . The CPSS contains 3 domains: pain control, physical function and symptom control. Participants will be asked to answer how much they agree with each of the items arranged on a Likert scale ranging from 10 to 100 points. The score ranges from 30 to 300, where the higher the score, the better the self-efficacy. Self-reported physical activity level The International Physical Activity Questionnaire short form is a 9-item questionnaire that assesses how many days and hours the participants usually spent per week doing several activities . The physical activity level will be determined by the total of vigorous and moderate exercise in the previous week and calculated according to previous studies . The Anterior Knee Pain Scale (AKPS) is a valid and reliable 13-item questionnaire that evaluates subjective function related to PFP . Participants will complete the AKPS based on their perceived knee condition at the prior week. The total score for the AKPS ranges from 0 (maximal disability) to 100 (no disability), with the total score being used for statistical analysis. The Hospital Anxiety and Depression Scale (HADS) is a 14-item questionnaire that evaluates the emotional state of the patient and identifies cases of mild, moderate and severe anxiety and/or depression disorders . The HADS consists of two subscales, which assess anxiety (7 items) and depression (7 items) separately. Participants will be asked to answer each item on a 4-point Likert scale (0–3), with scores ranging from 0 to 21 for anxiety and 0 to 21 for depression. Scores between 0 and 7 are classified as normal, between 8 and 10 as mild, between 11 and 14 as moderate, and between 15 and 21 as severe . The Tampa scale for kinesiophobia is a self-administered questionnaire that assesses pain-related fear associated with the avoidance behaviors, movements and physical activity . It contains 17 statements with answers in a 4-point Likert scale: Strongly disagree, Partially disagree, Partially agree and Totally agree. Participants will be instructed to choose the option according to how much they agree with each statement. The score ranges from 17 to 68 and the higher the score, the higher the fear . The Pain Catastrophizing Scale (PCS) is a 13-item questionnaire that consists of describing thoughts and feelings that individuals experience when they have pain . Participants will be instructed to reflect on the experiences caused by pain in the past and indicate their perception on a 5-point Likert scale, where (0) represents “not at all” and “all the time”. The higher the score, the greater the pain catastrophizing . The Chronic Pain Self-Efficacy Scale (CPSS) is a 22-item self-administered questionnaire that assesses the perception of self-efficacy and the ability to deal with the consequences of pain in patients with chronic pain . The CPSS contains 3 domains: pain control, physical function and symptom control. Participants will be asked to answer how much they agree with each of the items arranged on a Likert scale ranging from 10 to 100 points. The score ranges from 30 to 300, where the higher the score, the better the self-efficacy. The International Physical Activity Questionnaire short form is a 9-item questionnaire that assesses how many days and hours the participants usually spent per week doing several activities . The physical activity level will be determined by the total of vigorous and moderate exercise in the previous week and calculated according to previous studies . After the baseline assessment, participants will receive immediate access to a WEB platform developed by one of the authors available at http://www.stepslab.com.br/ where the interventions will be delivered. An individualized online meeting will be performed between participants and a physiotherapist not involved in data analysis to guide them regarding platform usage, deadlines and the importance of committing to the intervention. Participants will have access to the online interventions for 8 weeks, which will be immediately ceased at the end of the period. The exercise and education contents will be developed and pre-recorded by two physiotherapists with more than three years of clinical experience using evidence-based material. The MBI content will be developed and pre-recorded by a certified mindfulness teacher with more than 15 years of experience and revised by a psychologist to ensure psychological appropriateness. Details of the interventions according to the TIDier checklist are available in Additional file . Participants will only have access to content related to the group to which they were allocated (restricted area). Each session will be released for access on pre-defined dates relative to participants’ entry into the trial (immediately after baseline assessment). This will be performed by an investigator who will not be involved in the recruitment and assessment of the participants. The system will only allow opening of the next session if the participant had ended the previous one. At the end of each session, participants will be required to report their current level of pain on an online VAS scale, if there was any adverse event during or after the intervention, the level of satisfaction and the level of perceived exertion on a 15-point Borg scale , ranging from 6 (no exertion at all) to 20 perceiving a (maximal exertion). Participants will be instructed to not seek any other kind of knee pain treatment during the study, except in emergency cases. If necessary, participants will be able to contact the therapist through the platform’s e-mail. An outline of the study procedures is summarized in Fig. . Control group Participants allocated to this group will receive two pre-recorded video classes per week according to their availability (e.g. on Tuesdays and Thursdays) with Exercise therapy and Patient education contents (lasting 35–50 min in total). Exercise therapy The exercise therapy component will include the prescription of the exercises according to the American College of Sports Medicine Position Statement and the PFP consensus . The mean duration of the exercise videos will be approximately 30 min. There will be a one-day break between the sessions, to respect recovery time. Exercise therapy will aim to improve muscle performance, movement coordination and mobility . This intervention will target hip, knee, and ankle muscles. Exercises will be progressed in phases every two weeks (i.e., intensity, type of exercise, technique or repetitions). During the exercise, affirmative and encouraging audio messages will be displayed in order to motivate the participant to finish the session with as much effort as possible. The intensity of the exercises will be monitored through the Borg scale and must remain between 12 and 16 points. If the participant report exertion values outside this range, the exercise intensity will be modified. The full description of the exercises is available in Additional file . Patient education Educational pre-recorded video classes will cover the following topics. Week 1: Understanding my knee: anatomy and biomechanics of the knee and the relationship between pain and injury. Week 2: Understanding my knee pain: incidence and prevalence of PFP; why my knee hurts; biomechanical and psychological factors of PFP; prognosis and diagnosis of PFP. Week 3: “Too Much, too soon”: how high volume or high load intensity during daily activities or sports can lead to knee pain. Week 4: Myths and truths about my knee: Knee crepitus and movements considered harmful to the joint; fear of movement; and imaging exams. Week 5: Aspects of quality of life that influence pain: sleep quality, weight control, confidence, coping strategies, and mental health. Week 6: Taking care of my own pain: self-management of pain, motivation and responsibility for your own health. Week 7: Available treatment options: importance of adherence to active treatments, treatments that work, treatments that do not work and load management. Week 8: Take-home message: What should I do after the treatment? Motivation, habits change, behavioral change and the need to remain active (exercise/treatment). The mean duration of the patient education videos will be approximately 6 min. Mindfulness group Participants allocated to this group will receive the same intervention as the control group and an additional pre-recorded video class with MBI content according to their availability (e.g. on Mondays and Wednesdays they receive exercise therapy and Patient education contents; on Fridays they receive mindfulness contents, lasting 35–50 min). Mindfulness-based intervention The MBI component will be adapted from Mindfulness-Based Health Promotion (MBHP) model to suit patients with PFP and the online assessments of the present study. MBHP is an 8-week MBI developed in 2009 in the context of health promotion and quality of life . Inspired by Jon Kabat-Zinn’s original protocol—mindfulness-based stress reduction, MBSR (University of Massachusetts Medical Center, USA) —it also aggregates elements of other protocols, such as the MBCT (University of Toronto/Canada; University of Cambridge and Oxford University/United Kingdom) , the mindfulness programs of the Breathworks Institute (United Kingdom) , and mindfulness-based relapse prevention (MBRP, University of Washington, USA . The MBHP has been extensively used in a variety of health conditions [ – ]. Video classes will include formal and informal mindfulness practices. The description of the MBI is available in Additional file . The following themes will be covered. Week 1: Breaking the automatism. Week 2: Body awareness. Week 3: Leaving the mind and inhabiting the body. Week 4: Raising awareness. Week 5: Letting go. Week 6: Dealing with challenges and letting go of resistance. Week 7: Mindfulness and self-care. Week 8: A look to the future. In addition, participants included in this group will receive daily reminders and additional material (audios) to continue to practice formal and informal MBHP daily. The participants in this group will be encouraged to practice diaries. Participants allocated to this group will receive two pre-recorded video classes per week according to their availability (e.g. on Tuesdays and Thursdays) with Exercise therapy and Patient education contents (lasting 35–50 min in total). Exercise therapy The exercise therapy component will include the prescription of the exercises according to the American College of Sports Medicine Position Statement and the PFP consensus . The mean duration of the exercise videos will be approximately 30 min. There will be a one-day break between the sessions, to respect recovery time. Exercise therapy will aim to improve muscle performance, movement coordination and mobility . This intervention will target hip, knee, and ankle muscles. Exercises will be progressed in phases every two weeks (i.e., intensity, type of exercise, technique or repetitions). During the exercise, affirmative and encouraging audio messages will be displayed in order to motivate the participant to finish the session with as much effort as possible. The intensity of the exercises will be monitored through the Borg scale and must remain between 12 and 16 points. If the participant report exertion values outside this range, the exercise intensity will be modified. The full description of the exercises is available in Additional file . Patient education Educational pre-recorded video classes will cover the following topics. Week 1: Understanding my knee: anatomy and biomechanics of the knee and the relationship between pain and injury. Week 2: Understanding my knee pain: incidence and prevalence of PFP; why my knee hurts; biomechanical and psychological factors of PFP; prognosis and diagnosis of PFP. Week 3: “Too Much, too soon”: how high volume or high load intensity during daily activities or sports can lead to knee pain. Week 4: Myths and truths about my knee: Knee crepitus and movements considered harmful to the joint; fear of movement; and imaging exams. Week 5: Aspects of quality of life that influence pain: sleep quality, weight control, confidence, coping strategies, and mental health. Week 6: Taking care of my own pain: self-management of pain, motivation and responsibility for your own health. Week 7: Available treatment options: importance of adherence to active treatments, treatments that work, treatments that do not work and load management. Week 8: Take-home message: What should I do after the treatment? Motivation, habits change, behavioral change and the need to remain active (exercise/treatment). The mean duration of the patient education videos will be approximately 6 min. The exercise therapy component will include the prescription of the exercises according to the American College of Sports Medicine Position Statement and the PFP consensus . The mean duration of the exercise videos will be approximately 30 min. There will be a one-day break between the sessions, to respect recovery time. Exercise therapy will aim to improve muscle performance, movement coordination and mobility . This intervention will target hip, knee, and ankle muscles. Exercises will be progressed in phases every two weeks (i.e., intensity, type of exercise, technique or repetitions). During the exercise, affirmative and encouraging audio messages will be displayed in order to motivate the participant to finish the session with as much effort as possible. The intensity of the exercises will be monitored through the Borg scale and must remain between 12 and 16 points. If the participant report exertion values outside this range, the exercise intensity will be modified. The full description of the exercises is available in Additional file . Educational pre-recorded video classes will cover the following topics. Week 1: Understanding my knee: anatomy and biomechanics of the knee and the relationship between pain and injury. Week 2: Understanding my knee pain: incidence and prevalence of PFP; why my knee hurts; biomechanical and psychological factors of PFP; prognosis and diagnosis of PFP. Week 3: “Too Much, too soon”: how high volume or high load intensity during daily activities or sports can lead to knee pain. Week 4: Myths and truths about my knee: Knee crepitus and movements considered harmful to the joint; fear of movement; and imaging exams. Week 5: Aspects of quality of life that influence pain: sleep quality, weight control, confidence, coping strategies, and mental health. Week 6: Taking care of my own pain: self-management of pain, motivation and responsibility for your own health. Week 7: Available treatment options: importance of adherence to active treatments, treatments that work, treatments that do not work and load management. Week 8: Take-home message: What should I do after the treatment? Motivation, habits change, behavioral change and the need to remain active (exercise/treatment). The mean duration of the patient education videos will be approximately 6 min. Participants allocated to this group will receive the same intervention as the control group and an additional pre-recorded video class with MBI content according to their availability (e.g. on Mondays and Wednesdays they receive exercise therapy and Patient education contents; on Fridays they receive mindfulness contents, lasting 35–50 min). Mindfulness-based intervention The MBI component will be adapted from Mindfulness-Based Health Promotion (MBHP) model to suit patients with PFP and the online assessments of the present study. MBHP is an 8-week MBI developed in 2009 in the context of health promotion and quality of life . Inspired by Jon Kabat-Zinn’s original protocol—mindfulness-based stress reduction, MBSR (University of Massachusetts Medical Center, USA) —it also aggregates elements of other protocols, such as the MBCT (University of Toronto/Canada; University of Cambridge and Oxford University/United Kingdom) , the mindfulness programs of the Breathworks Institute (United Kingdom) , and mindfulness-based relapse prevention (MBRP, University of Washington, USA . The MBHP has been extensively used in a variety of health conditions [ – ]. Video classes will include formal and informal mindfulness practices. The description of the MBI is available in Additional file . The following themes will be covered. Week 1: Breaking the automatism. Week 2: Body awareness. Week 3: Leaving the mind and inhabiting the body. Week 4: Raising awareness. Week 5: Letting go. Week 6: Dealing with challenges and letting go of resistance. Week 7: Mindfulness and self-care. Week 8: A look to the future. In addition, participants included in this group will receive daily reminders and additional material (audios) to continue to practice formal and informal MBHP daily. The participants in this group will be encouraged to practice diaries. The MBI component will be adapted from Mindfulness-Based Health Promotion (MBHP) model to suit patients with PFP and the online assessments of the present study. MBHP is an 8-week MBI developed in 2009 in the context of health promotion and quality of life . Inspired by Jon Kabat-Zinn’s original protocol—mindfulness-based stress reduction, MBSR (University of Massachusetts Medical Center, USA) —it also aggregates elements of other protocols, such as the MBCT (University of Toronto/Canada; University of Cambridge and Oxford University/United Kingdom) , the mindfulness programs of the Breathworks Institute (United Kingdom) , and mindfulness-based relapse prevention (MBRP, University of Washington, USA . The MBHP has been extensively used in a variety of health conditions [ – ]. Video classes will include formal and informal mindfulness practices. The description of the MBI is available in Additional file . The following themes will be covered. Week 1: Breaking the automatism. Week 2: Body awareness. Week 3: Leaving the mind and inhabiting the body. Week 4: Raising awareness. Week 5: Letting go. Week 6: Dealing with challenges and letting go of resistance. Week 7: Mindfulness and self-care. Week 8: A look to the future. In addition, participants included in this group will receive daily reminders and additional material (audios) to continue to practice formal and informal MBHP daily. The participants in this group will be encouraged to practice diaries. In order to improve the adherence to the treatments, before each session, participants of both groups will receive an automatic reminder via SMS and/or email before each session. Participants’ adherence to the interventions will be monitored through the number of accesses (date and hour), time connected to the platform, sessions visualized, number of sessions finalized, number of drop outs and others. At the end of each video class, in a pop-up window, all participants will report the intensity of their pain and if there were any adverse events during the session. Participants will be able to contact a therapist through the platform’s email at any time. In case of a severe adverse event related to exercise therapy or MBI (e.g. strains, sprains, persistent severe pain, psychosis, mania, traumatic memories), the participant will be referred to a qualified healthcare professional for further investigation. The sample size calculation was performed based on the usual pain intensity data from Bagheri et al. . Considering a difference between groups of 4.2 mm and a standard deviation of 5.4 mm, with an α of 0.05 and β of 0.20, 26 participants per groups are required (52 in total). We will recruit 31 participants per trial arm to allow for up to a 20% drop-out rate at 12 months. Statistical analysis will be performed by the blinded assessor using SPSS software (IBM version 23, SPSS Inc., Chicago, Il). Descriptive statistics will be computed for all variables (e.g. mean, standard deviation). Data will be tested for normal distribution by the Shapiro-Wilk test. Chi-square tests will be performed to compare self-reported recovery (successful x unsuccessful) between groups. For continuous data, the effects of group, time and their interaction will be assessed with linear mixed models. Intraclass Correlation Coefficients will be used to determine the amount of variance explained by random effects . The Bonferroni-adjusted post hoc test will be performed for multiple pairwise comparisons where appropriate. Effect sizes (95% CI) (Cohen’s d) will also be calculated and interpreted as follows: Cohen’s = 0.2 ‘small effect’; = 0.5 ‘moderate effect’; = 0.8 ‘large effect’ and = 1.3 ‘very large effect’ . Intent-to-treat analyses will be performed for all outcomes. Multiple imputation will be used to account for missing data if the proportion of missing data is > 5% . For all tests, an α level of 0.05 two-tailed will be adopted to indicate statistical significance. The mediation effects will be assessed following the 3-variable framework described by MacKinnon et al. . In this model, the intervention condition is assumed to have both direct and indirect paths to the changes in clinical outcomes. The indirect path passes through the potential mediators (anxiety, depression, kinesiophobia, pain catastrophizing and pain self-efficacy). Three multiple regressions will be performed: to test the association between the predictor (i.e., interventions) and the outcomes (i.e., pain and function); to test the association between the predictor and the potential mediators and to test the association between the potential mediators and the outcomes after controlling for the predictor. Then, it will be observed whether the association of the predictor with the outcome after controlling for potential mediators will be smaller than observed in the first regression. PFP is a common and often recalcitrant knee disorder, with symptoms persisting for many years . Exercise therapy and patient education are considered the “best management” options for this population . However, unsatisfactory long-term prognosis remains an issue . It is known that people with PFP present with altered psychological factors , which should be considered during the evaluation and treatment of people with PFP . Recent studies suggest that MBI induces functional and structural brain modifications . As part of a rehabilitation program, MBI can help the patients to recognize and accept their condition, promoting a more effective focus on rehabilitation and facilitating pain relief . Therefore, adding a MBI program to the current best treatment for PFP may improve psychological outcomes, providing a better response to treatment at short and long-term. However, this hypothesis needs further investigation. The proposed trial will address this knowledge gap by evaluating the effects of adding an 8-week online MBI program to an online program of exercise therapy and patient education on self-reported recovery, pain, function and psychological factors and in people with PFP. If our hypotheses are confirmed, our findings will contribute to the discussion of a new perspective of treatment modality for people with PFP. Limitations This study investigates the additional effect of an online treatment based on mindfulness program to exercise therapy and patient education for people with PFP. Although exercise therapy and patient education are considered the cornerstones of PFP management , the additional effect of the mindfulness intervention to all physical interventions of PFP is not investigated in the present study. Future studies in this area are warranted. This study investigates the additional effect of an online treatment based on mindfulness program to exercise therapy and patient education for people with PFP. Although exercise therapy and patient education are considered the cornerstones of PFP management , the additional effect of the mindfulness intervention to all physical interventions of PFP is not investigated in the present study. Future studies in this area are warranted. Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5
Telehealth services in an outpatient nephrology clinic during the COVID-19 pandemic: a patient perspective	2b63e908-0751-450c-98a4-8b293f38a468	10173904	Internal Medicine[mh]	The COVID-19 pandemic called for a change in the way healthcare was delivered to patients in order to comply with social distancing recommendations for public safety and resulted in a dramatic shift from in-person care to a focus on virtual care across the country. The Centers for Medicare and Medicaid Services (CMS) swiftly introduced new reimbursement policy waivers that expanded telehealth coverage, eliminating geographical restrictions, and allowing both the originating and remote sites to be from the home . States passed telehealth parity laws that required commercial payers to cover telehealth services . State licensure requirements were also waived, which allowed licensed providers to provide care across state lines . The aim of this study was to assess patient experiences with telehealth services during the COVID-19 pandemic. After obtaining Institutional Review Board (#NCR213713) approval, patients who had at least one virtual visit between April 1, 2020 and March 31, 2021 in an urban university-based outpatient nephrology clinic were surveyed using a web-based survey tool to assess their telehealth experiences (see Supplement for the survey). Patients were identified using the clinic schedule from three faculty physicians who conducted telehealth encounters. Patients who were unable to read or understand English were excluded. Study data were collected and managed using Research Electronic Data Capture (REDCap) electronic data capture tools . Demographic information collected included name, medical record number, phone number, home address, gender, age, and insurance type. The patient was invited to complete the survey initially on-line by email. For individuals who did not respond to the email invitation or who did not have valid email addresses documented, they completed the survey via phone or mail. Patients rated their overall experience with telehealth encounters during the time period using a Likert scale. The Likert scale had five options: excellent, good, fair, poor, and very poor. Additionally, digital health literacy as well as barriers, such as technical issues, to successfully completing telehealth encounters were assessed. Patient satisfaction was positive if the respondent selected good or excellent. The data was de-identified prior to analysis. Patient demographic information (gender, age, race) were analyzed for number, frequency, mean, standard deviation, median and range as deemed appropriate. Chi square testing was performed to compare frequencies of categorical variables to analyze thematic patterns in survey responses. Similarly, regression analysis was performed to analyze the relationship between the independent variables of interest and patient satisfaction. Statistical significance was defined as p < 0.05. The demographic results of the survey are shown in Table . Out of 791 invitees, 166 patients completed the survey (Fig. .). A few patients did not answer all the questions, thus resulting in missing data. Most patients (138/165) reported that their overall experiences with telehealth services were positive while 20/165 reported their experiences to be fair, and seven out of 165 reported that their overall experiences were poor or very poor. Most patients (143/166) found that it was easy to make a telehealth appointment. For those who felt that it was not easy to make a telehealth appointment, the majority explained that they had difficulty with appointment scheduling that was not related to the telehealth experience. Many patients (136/163) felt it was easy to log into the virtual visit platform. A majority (107/165) responded that they were willing to use the video feature while 43/165 ranked it neutral and 15/165 did not want to use this feature. The majority of patients (133/166) wanted to have a hybrid model in the future, while 9.6% of patients wanted to have telehealth appointments only, and 8.4% wanted in-person appointments only. The remaining 1.9% of patients did not have a preference. A large number of patients (132/166) reported no technical issues during their telehealth visits. For those who had technical issues, patients reported that these issues ranged from the call being dropped to audio or video issues. Some patients reported difficulty with logging into the virtual appointments and having limited digital literacy, which led to the video visits being converted to telephone visits instead. The majority of patients (125/162) participated in video visits. Some patients (15/162) did not have camera equipment to have a video visit, while others (20/162) did not know how to use the video feature on the virtual platform. A small percentage of patients (two out of 162) did not want the physician to see them, so they opted for a telephone visit instead. Most patients (161/166) found their telehealth visits convenient. The number one reason given was not having to worry about arranging transportation. Some other popular reasons included: the doctor was on time to the visit, the visit allowed them the option of not having to take time off from work, and the visit allowed them to have someone else join in the virtual visit. For the few that responded that the telehealth visit was not convenient (five out of 166), they preferred to see the physician in-person. Most patients wanted to continue having telehealth appointments after the pandemic (153/166). Patients explained that telehealth appointments are convenient, especially for reviewing lab results. A few patients (nine out of 166) reported they were hospitalized between April 1, 2020 and March 31, 2021. Patients (eight out of nine) who were hospitalized during this time period did not think that an in-person visit instead of a virtual one could have prevented the hospitalization, while one person did not respond. Most patients (134/164) were amenable to not having an in-person physical exam. For those who wanted an in-person physical exam, patients felt that it preserved the interpersonal interactions that are important to the physician–patient relationship. The following variables had no significant impact on overall experience: age, gender, ethnicity, prior telehealth use, wanting only telehealth appointments in the future, insurance type or status (unknown insurance vs. private vs. Medicare/Medicaid), and dropped calls when using telehealth services. Table depicts the positive and negative experiences queried with regards to telehealth. This study utilizes the unique COVID-19 pandemic situation, where outpatient nephrology care shifted to telehealth-based services, to better understand patient perspectives on this form of service. Our results indicate that the majority of patients had positive experiences overall with telehealth services such that they would prefer to see a “hybrid” model in the future. The majority of patients found virtual visits convenient with few patients reporting technical issues impeding their experiences. Our research revealed benefits and challenges of this new healthcare delivery model as perceived by patients. The challenges reported by some patients included the lack of access to appropriate equipment, which sheds light into health inequities that must be addressed moving forward. These results may help the structure of future telehealth programs beyond the pandemic. Telenephrology In nephrology, there is a growing need for telehealth services to increase access to specialty care and reduce health disparities. The 2015 Workforce Report by the American Society of Nephrology found that many nephrology training programs are located in mostly urban settings, far from rural areas where kidney failure rates are high . It has been shown that patients with chronic kidney disease (CKD) in geographically isolated locations are less likely to follow up with nephrologists and receive the effective treatment they need, which ultimately leads to more hospitalizations and even higher mortality when compared to patients who have access to nephrology care . Large healthcare organizations such as the Veterans Affairs (VA) system have turned to telehealth services as a means to provide more cost-effective care and bridge the gap between underserved populations and access to care . The VA leveraged telehealth services as early as 2002 with its Clinical Video Telehealth (CVT) program that aimed to provide patients remote care through videoconferencing in various specialties . One study showed that telenephrology improved patient compliance with appointments and reduced no-show rates as well as cancellations by 50% . Moreover, they hypothesized that this may have been due to decreased travel times to the hospital. Another example of applying telehealth to increase access to specialty care employs electronic consultation (eConsult), which utilizes provider-to-provider electronic asynchronous exchanges of patient health information at a distance to improve the interface between primary care providers and specialists . Nephrologists have had opportunities to conduct telehealth encounters prior to the COVID-19 pandemic . A few patients had prior telehealth experience with one of the nephrologists under a study to improve specialty care access via telemedicine using an urban Federally Qualified Health Center as the originating site . The pandemic has pushed telehealth to the forefront of nephrology care with CMS removing barriers delineated above, licensing requirements, and reimbursement limitations based on visit type . Despite promising results on patient satisfaction and increasing access to care, the future of telenephrology largely depends on future legislation and/or regulations. Benefits and challenges for patients and practices Telehealth benefits include convenience, decrease transportation costs and time, increase accessibility to healthcare, and decrease overall opportunity costs (i.e., patients do not need to take time off from work) . There are also many challenges to this virtual healthcare model. Patients require adequate broadband services and access to equipment including a computer, tablet, or a smartphone with video and audio services. Thus, health disparities and social determinants of health pose significant barriers to having a successful telehealth encounter. Equality and social determinants of health need to be addressed. Having adequate digital health literacy ensures benefiting from telehealth services. Patients with digital illiteracy struggle to join a video visit or set up remote monitoring services, which prevents them from adequately accessing the care they need. Older CKD patients reported that the lack of a physical exam or the difficulty of breaking bad news virtually, especially with connectivity issues, detracted from their experiences and negatively affected the patient-physician relationship and fostered mistrust . Ethnicity and age were shown to negatively impact patient satisfaction, with older individuals and patients of color reporting lower patient satisfaction with using telehealth . These findings are in opposition to results from this study, which showed that ethnicity and age do not play a role in determining patient satisfaction with telehealth services. For practices, telehealth services increase the number of patients that can be seen, expand specialized care to patients who live far away, and may reduce Emergency Department visits/hospitalizations given more surveillance with remote patient monitoring . Video appointments in particular offer providers a unique opportunity to get a glimpse into the patient’s home to better inform patient care and address social determinants of health. Large practices are able to roll out telehealth programs given that they have robust electronic health record systems, adequate information technology (IT) support for troubleshooting, and adequate funds to purchase Telehealth platform licenses. Additionally, security and privacy concerns remain an issue with telehealth platforms. Purchasing licenses of telehealth platforms ensures Health Insurance Portability and Accountability Act (HIPAA) compliance and mitigates security concerns. For private practices or single provider practices, the economic means to roll out successful telehealth programs may serve as a significant challenge. Limitations The small sample size of 21% response rate, despite falling into the average range of response rates of 5 to 30%, may not represent the study population. It is possible that the remaining non-respondents may be older, have poor digital health literacy, or low satisfaction with the tool. Although the majority of patients reported a positive experience, a few patients reported a negative experience because they wanted an in-person encounter or had technical issues with audio and/or video aspects of the visit (Table ). Furthermore, the nephrology specialty relies heavily on patient symptoms, blood pressure readings and laboratory results and less on physical examination. Therefore, the findings of this study may limit its generalizability to other disciplines and specialty care. Aspirations and future challenges Various operational challenges that must be considered for widespread use of telehealth include physician licensing requirements across state lines, telehealth platforms complying with HIPAA and Health Information Technology for Economic and Clinical Health Act requirements, and Business Associates agreement with providers partaking in telehealth services . The fate of licensing requirements will depend on state policies, while the latter can be enforced with the help of an IT officer and the use of HIPAA-compliant telehealth platforms. More studies need to be conducted to determine the cost of care using CMS claims data to determine if this model of healthcare delivery endorses cost-effectiveness. Virtual visits should be used judiciously as these services are not appropriate for all patients but will depend on the physician’s judgment based on in-person visits and triage of virtual visits. In nephrology, there is a growing need for telehealth services to increase access to specialty care and reduce health disparities. The 2015 Workforce Report by the American Society of Nephrology found that many nephrology training programs are located in mostly urban settings, far from rural areas where kidney failure rates are high . It has been shown that patients with chronic kidney disease (CKD) in geographically isolated locations are less likely to follow up with nephrologists and receive the effective treatment they need, which ultimately leads to more hospitalizations and even higher mortality when compared to patients who have access to nephrology care . Large healthcare organizations such as the Veterans Affairs (VA) system have turned to telehealth services as a means to provide more cost-effective care and bridge the gap between underserved populations and access to care . The VA leveraged telehealth services as early as 2002 with its Clinical Video Telehealth (CVT) program that aimed to provide patients remote care through videoconferencing in various specialties . One study showed that telenephrology improved patient compliance with appointments and reduced no-show rates as well as cancellations by 50% . Moreover, they hypothesized that this may have been due to decreased travel times to the hospital. Another example of applying telehealth to increase access to specialty care employs electronic consultation (eConsult), which utilizes provider-to-provider electronic asynchronous exchanges of patient health information at a distance to improve the interface between primary care providers and specialists . Nephrologists have had opportunities to conduct telehealth encounters prior to the COVID-19 pandemic . A few patients had prior telehealth experience with one of the nephrologists under a study to improve specialty care access via telemedicine using an urban Federally Qualified Health Center as the originating site . The pandemic has pushed telehealth to the forefront of nephrology care with CMS removing barriers delineated above, licensing requirements, and reimbursement limitations based on visit type . Despite promising results on patient satisfaction and increasing access to care, the future of telenephrology largely depends on future legislation and/or regulations. Telehealth benefits include convenience, decrease transportation costs and time, increase accessibility to healthcare, and decrease overall opportunity costs (i.e., patients do not need to take time off from work) . There are also many challenges to this virtual healthcare model. Patients require adequate broadband services and access to equipment including a computer, tablet, or a smartphone with video and audio services. Thus, health disparities and social determinants of health pose significant barriers to having a successful telehealth encounter. Equality and social determinants of health need to be addressed. Having adequate digital health literacy ensures benefiting from telehealth services. Patients with digital illiteracy struggle to join a video visit or set up remote monitoring services, which prevents them from adequately accessing the care they need. Older CKD patients reported that the lack of a physical exam or the difficulty of breaking bad news virtually, especially with connectivity issues, detracted from their experiences and negatively affected the patient-physician relationship and fostered mistrust . Ethnicity and age were shown to negatively impact patient satisfaction, with older individuals and patients of color reporting lower patient satisfaction with using telehealth . These findings are in opposition to results from this study, which showed that ethnicity and age do not play a role in determining patient satisfaction with telehealth services. For practices, telehealth services increase the number of patients that can be seen, expand specialized care to patients who live far away, and may reduce Emergency Department visits/hospitalizations given more surveillance with remote patient monitoring . Video appointments in particular offer providers a unique opportunity to get a glimpse into the patient’s home to better inform patient care and address social determinants of health. Large practices are able to roll out telehealth programs given that they have robust electronic health record systems, adequate information technology (IT) support for troubleshooting, and adequate funds to purchase Telehealth platform licenses. Additionally, security and privacy concerns remain an issue with telehealth platforms. Purchasing licenses of telehealth platforms ensures Health Insurance Portability and Accountability Act (HIPAA) compliance and mitigates security concerns. For private practices or single provider practices, the economic means to roll out successful telehealth programs may serve as a significant challenge. The small sample size of 21% response rate, despite falling into the average range of response rates of 5 to 30%, may not represent the study population. It is possible that the remaining non-respondents may be older, have poor digital health literacy, or low satisfaction with the tool. Although the majority of patients reported a positive experience, a few patients reported a negative experience because they wanted an in-person encounter or had technical issues with audio and/or video aspects of the visit (Table ). Furthermore, the nephrology specialty relies heavily on patient symptoms, blood pressure readings and laboratory results and less on physical examination. Therefore, the findings of this study may limit its generalizability to other disciplines and specialty care. Various operational challenges that must be considered for widespread use of telehealth include physician licensing requirements across state lines, telehealth platforms complying with HIPAA and Health Information Technology for Economic and Clinical Health Act requirements, and Business Associates agreement with providers partaking in telehealth services . The fate of licensing requirements will depend on state policies, while the latter can be enforced with the help of an IT officer and the use of HIPAA-compliant telehealth platforms. More studies need to be conducted to determine the cost of care using CMS claims data to determine if this model of healthcare delivery endorses cost-effectiveness. Virtual visits should be used judiciously as these services are not appropriate for all patients but will depend on the physician’s judgment based on in-person visits and triage of virtual visits. Major findings of this study reveal that patients were not only satisfied with their telehealth experiences overall, but the majority would like to see a hybrid model of care moving forward. The patients praised the convenience of making a telehealth appointment and conducting the virtual visit, with very few patients reporting technical issues. A multitude of factors impact patient satisfaction, with the bulk of these factors directly related to navigation and use of the telehealth platform. In addition, patients enjoyed the decreased transportation costs and time, increased accessibility to healthcare, and decreased overall opportunity costs. Moreover, age, gender, and ethnicity did not impact patient satisfaction ratings as previously reported. Since a limited sample size and only patients receiving care through the nephrology specialty were studied, additional studies will need to be conducted to understand the general patient experience with telehealth. Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 20 KB)
Systematic review on the involvement and engagement of patients as advisers for the organisation of organ transplantation services	a8de239b-d434-4c6e-9194-954251bc339a	10173988	Patient-Centered Care[mh]	The deployed analytical framework can also be used as a strategic tool for the introduction of patient involvement and engagement in the organisation of transplantation services. Although a large number of studies on patient involvement and engagement were identified with the initial search strategy, only a small number of identified articles could be identified that focused on patient involvement and engagement in the organisation of organ transplantation services. Negative effects of patient involvement and engagement may be less likely reported in the literature due to publication bias. Involving and engaging patients on the organisational and management level of healthcare may help to ensure that patients’ needs and concerns are clearly understood and appropriately prioritised. Studies on primary care, acute care, oncology care and mental health services demonstrated benefits for healthcare providers and patients gained by actively involving and engaging patients in gathering information, sharing experiences and joint decision-making. Patient involvement and engagement is a concept that involves patients in decision-making about their care and giving them choice and control over the healthcare services they receive. A substantial lack of understanding in regard to the interpretation of related terms such as patient engagement, patient empowerment and patient-centred care has recently been described. As a result, confusion and misunderstandings prevail in the perception of the implications of this term. This situation provides major hurdles for the successful establishment of best practices and evidence-based approaches for patients’ involvement and engagement as advisers in the organisation of organ transplantation services and other medical services. The National Health Service in the UK has provided the following practical definition including all relevant stakeholder perspectives: ‘Patient and public engagement is active participation of patients, carers, community representatives, community groups and the public in how services and policy are planned, delivered and evaluated. It is broader and deeper than traditional consultation. It involves the ongoing process of developing and sustaining constructive relationships, building strong, active partnerships and holding a meaningful dialogue with stakeholders’. Currently, the concept of patient involvement and engagement in healthcare is not standard practice for most healthcare organisations. In this systematic review, we focus on patients as advisors for programme design and clinical process improvement within the organisation of organ transplantation services. Organ transplantation is an ethically challenging field and very cost intensive. Therefore, representative voices of patients should be included in the process of policy-making as well as in the organisation and management of organ transplantation services in a transparent and systematic way with the goal to optimise clinical processes and healthcare resource allocation. Such an approach may provide a very powerful asset for a more effective and efficient management of services mainly by the inclusion of relevant and unique insights from transplant patients. This could be realised for example by including patient representatives in relevant advisory committees and/or management boards of the transplant centre in official roles. Moreover, integration of involved and engaged patients in transplant organisations may greatly improve transparency and public understanding and could thus promote societal acceptance of and trust in national transplantation systems. Increased patient-centredness with improved service quality and transparency might improve public readiness to altruistic organ donation for transplantation, which would help to reduce the currently increasing scarcity of suitable donor organs. The importance, benefits and potentials of patient involvement and engagement in the organisation of transplantation services, as well as its effects on transplantation practice have not been investigated in a systematic manner so far. Therefore, the aim of this systematic review is to evaluate published reports and studies on this topic with the focus on goals, roles of patients, involved stages of transplantation, deployed measures and methods of implementation, barriers to implementation, and observed process benefits and outcome improvements. This review aims to derive practical lessons learnt from the published literature that may be useful for the improvement of transplantation services. This systematic review was carried out in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (see ) and has been registered in the PROSPERO database (registration code: CRD42022186467). 10.1136/bmjopen-2023-072091.supp1 Supplementary data Literature search A predefined literature search strategy was used to identify articles of potential interest. The term ‘patient involvement and engagement’ was based on the definition as published by the National Health Services in the UK and also used as a MeSH term for the search string in PubMed, while the search string for Web of Science was adapted accordingly ( ). The literature search included all publications cited in PubMed or Web of Science until 6 December 2022 deploying predefined inclusion and exclusion criteria in order to generate a highly sensitive overview of relevant publications. Furthermore, the reference lists of relevant studies identified in the literature search process were also screened in order to find additional studies. Moreover, further potential studies of interest were identified by hand search via Google Scholar to increase the sensitivity of the search. Inclusion criteria Inclusion criteria for the analysis of publications required that patients participated in transplantation practice at an organisational level as advisers or advisory committee members with the goal to participate in programme design, clinical process and outcome improvement and/or joint decision-making at the organisational level. Exclusion criteria Publications that were not published in English were eliminated. As implied by the literature search date, publications published after 6 December 2022 could not be analysed. Furthermore, surveys that collected data on patients’ attitudes and studies that focused on new methods for patient reported outcome evaluation were also excluded as well as systematic reviews, meta-analyses, abstracts-only, commentaries and letters. Data extraction After deduplication of search results, title screening was carried out based on the above mentioned predefined inclusion and exclusion criteria. The next steps were taken in a sensitive approach, which means that if one or both reviewers (ZQ and CO) found the article to be relevant, it was included for further analysis. Abstract as well as full text screening was performed to further exclude ineligible studies. All data were extracted using a standardised form that was developed prior to analysis as proposed by the PRISMA statement. Relevant information identified in finally analysed publications was systematically extracted according to a predefined structure as described by the column headings of and summarised independently by two authors (ZQ and CO). In cases of inconsistent judgements, a third reviewer (TB) was asked to evaluate and his decision on the inclusion or exclusion of any article was discussed within the research group. After reaching consensus, the results were summarised in . Development of an analytical framework for the analysis An initial literature review was conducted using a scoping rapid review process to identify existing concepts for patient involvement and engagement in healthcare. Overall, 29 conceptual publications on patient involvement and engagement in the organisation of different healthcare services and settings were identified and structured accordingly (see ). These publications were used for the development of the analytical framework ( ) which was later used for the subsequent systematic review of articles on patient involvement and engagement in the organisation of transplantation services ( ). The theoretical basis for the creation of a comparative matrix that was used to extract relevant categories and items from all systematically reviewed articles was based on these conceptual publications. Following this approach, these articles were classified and categorised for further analysis in subsequent steps ( ). Reporting quality evaluation and risk of bias assessment The updated GRIPP 2 short form (Guidance for Reporting Involvement of Patients and the Public) was used for the assessment and reporting of the quality of analysed articles in this systematic review. To assess the risk of bias, we used the Critical Appraisal Skills Programme (CASP) tool for systematic reviews of qualitative studies as suggested by Ma et al . Two authors (ZQ and HS) independently assessed the risk of bias with this tool for all included articles and subsequently reached consensus on their assessments. Patient and public involvement Neither patients nor the public were directly involved in the process of the creation of this manuscript. The plan of disseminating the current review involves direct dissemination to patient representatives and advisory committee members and the press to ensure that this research is communicated widely to all relevant stakeholders. A predefined literature search strategy was used to identify articles of potential interest. The term ‘patient involvement and engagement’ was based on the definition as published by the National Health Services in the UK and also used as a MeSH term for the search string in PubMed, while the search string for Web of Science was adapted accordingly ( ). The literature search included all publications cited in PubMed or Web of Science until 6 December 2022 deploying predefined inclusion and exclusion criteria in order to generate a highly sensitive overview of relevant publications. Furthermore, the reference lists of relevant studies identified in the literature search process were also screened in order to find additional studies. Moreover, further potential studies of interest were identified by hand search via Google Scholar to increase the sensitivity of the search. Inclusion criteria for the analysis of publications required that patients participated in transplantation practice at an organisational level as advisers or advisory committee members with the goal to participate in programme design, clinical process and outcome improvement and/or joint decision-making at the organisational level. Publications that were not published in English were eliminated. As implied by the literature search date, publications published after 6 December 2022 could not be analysed. Furthermore, surveys that collected data on patients’ attitudes and studies that focused on new methods for patient reported outcome evaluation were also excluded as well as systematic reviews, meta-analyses, abstracts-only, commentaries and letters. After deduplication of search results, title screening was carried out based on the above mentioned predefined inclusion and exclusion criteria. The next steps were taken in a sensitive approach, which means that if one or both reviewers (ZQ and CO) found the article to be relevant, it was included for further analysis. Abstract as well as full text screening was performed to further exclude ineligible studies. All data were extracted using a standardised form that was developed prior to analysis as proposed by the PRISMA statement. Relevant information identified in finally analysed publications was systematically extracted according to a predefined structure as described by the column headings of and summarised independently by two authors (ZQ and CO). In cases of inconsistent judgements, a third reviewer (TB) was asked to evaluate and his decision on the inclusion or exclusion of any article was discussed within the research group. After reaching consensus, the results were summarised in . An initial literature review was conducted using a scoping rapid review process to identify existing concepts for patient involvement and engagement in healthcare. Overall, 29 conceptual publications on patient involvement and engagement in the organisation of different healthcare services and settings were identified and structured accordingly (see ). These publications were used for the development of the analytical framework ( ) which was later used for the subsequent systematic review of articles on patient involvement and engagement in the organisation of transplantation services ( ). The theoretical basis for the creation of a comparative matrix that was used to extract relevant categories and items from all systematically reviewed articles was based on these conceptual publications. Following this approach, these articles were classified and categorised for further analysis in subsequent steps ( ). The updated GRIPP 2 short form (Guidance for Reporting Involvement of Patients and the Public) was used for the assessment and reporting of the quality of analysed articles in this systematic review. To assess the risk of bias, we used the Critical Appraisal Skills Programme (CASP) tool for systematic reviews of qualitative studies as suggested by Ma et al . Two authors (ZQ and HS) independently assessed the risk of bias with this tool for all included articles and subsequently reached consensus on their assessments. Neither patients nor the public were directly involved in the process of the creation of this manuscript. The plan of disseminating the current review involves direct dissemination to patient representatives and advisory committee members and the press to ensure that this research is communicated widely to all relevant stakeholders. Search and selection results for systematic review The electronic database search strategy yielded a total of 2263 non-duplicate records, further 2 articles were identified by hand search. Of these, 2142 were excluded after title screening based on the predefined inclusion and exclusion criteria. Eighty-six articles were further excluded after abstract screening. Of the remaining 37 articles, 11 could finally be identified as suitable for the systematic review based on full-text screening ( ). Characteristics and results of systematically reviewed articles The publication years of the 11 systematically reviewed articles ranged from 1996 to 2021. These articles originated from various countries, with most of them coming from the USA (n=5) and Norway (n=2). Among these, patient involvement and engagement in the organisation of organ transplant services was analysed in context with the support of clinical experts (n=5), nurses (n=2) and other healthcare providers (n=2). Medical students were also involved as professionals in one study ( ). The methods used to involve and engage patients included the codevelopment of a peer education programme (n=6) or a communication tool (n=1), the participation in representative or advisory groups or user panels in patient or steering committees (n=5) and a mix of different methods (n=1) ( ). Only two articles described several methods to involve and engage patients. All articles described that well-developed programmes, which involve and engage patients in the organisation of transplant services, provide benefits for patients and foster patient-centredness. However, none of these articles investigated the impact on clinical outcomes such as increased quality of life or improved coping with disease or focused on an ongoing process to involve and engage patients as required by the definition of the National Health Services. Details of deployed methods for patient involvement and engagement as well as the observed outcomes, benefits and findings are summarised in . Compared with 29 conceptual publications on patient involvement and engagement in different healthcare services and settings, the 11 articles on patient involvement and engagement in the organisation of transplantation services focused predominantly on process goals of patient involvement and engagement such as enhancing existing services or the development of new services, but rarely on outcome goals ( ). Only three reported clinical outcome goals were improvements in health status as well as increase in quality of life, while aspects such as increase in service utilisation, satisfaction with services as well as improvements in coping with the disease were not the aims. Furthermore, none of these articles involved and engaged patients at the stage of dissemination and communication ( ). Some systematically reviewed articles investigated aspects that have not been described in the conceptual publications on patient involvement and engagement in other, different healthcare services. Examples of such aspects include dynamic changes in patients’ conditions and perspectives, emotional sensitivities in regard to discussions on deceased organ donation, inadequate or inconsistent information for patients, lack of digital system interoperability and financial barriers for institutions to implement patient involvement and engagement in the organisation of transplantation services, while financial barriers for patients were not identified as an important barrier to such an implementation according to the conceptual publications ( ). Systematically reviewed articles revealed that patients and their relatives/partners were usually involved and engaged in various ways in the organisation of transplant services for patients on the waiting list for transplantation or in the organisation of transplant services for patients in the long-term follow-up period after transplantation or in both phases. All of these articles supposed that clinical transplant services would benefit from active patient involvement and engagement with different goals. The most commonly expressed goal in these articles was the improvement of existing services. In most articles, patients were assigned the following roles: Involvement in the design of focus group discussions to explore patient experiences with transplant services, or in discussions prior to and following transplantation to evaluate transplant service development. The reported approaches to patient involvement and engagement were coproduction, collaboration and consultation. Most of the articles focused on patient involvement and engagement activities in the stages of the design and planning of transplantation services. The most commonly expected mechanism was the support of the interaction between patients and their healthcare providers aiming to address patients’ needs. Notably, inadequate or inconsistent information for patients was identified as most frequent barrier. Other articles identified barriers to the involvement and engagement of patients in the organisation and management of transplant care included dynamic changes in patients’ conditions and perspectives, high emotional sensitivity to discussions on deceased organ donation, financial barriers for the institution, barriers due to lack of time or support for practitioners and lack of system interoperability (the property of systems and medical records to exchange data). All these aspects were not identified through the conceptual publications indicating the specific characteristics of the transplant field. The most frequently proposed strategies to facilitate or to improve patients’ involvement and engagement in the organisation of transplant services were better support for practitioners, adding new modes of communication between practitioners and patients, and more or clearer information for patients. For a summary of details, see . Quality assessment of systematically reviewed articles revealed that no technical report on training or on supporting information for the patients and their relatives or partners who have been involved and engaged in the organisation of transplant services was reported. None of them reported patient involvement and engagement in the plain summary or abstract according to the criteria of current recommendations in GRIPP 2. The results of the assessment of the reporting quality using the GRIPP 2 short form are summarised in . The results of the risk of bias assessment using the CASP tool are summarised in . The electronic database search strategy yielded a total of 2263 non-duplicate records, further 2 articles were identified by hand search. Of these, 2142 were excluded after title screening based on the predefined inclusion and exclusion criteria. Eighty-six articles were further excluded after abstract screening. Of the remaining 37 articles, 11 could finally be identified as suitable for the systematic review based on full-text screening ( ). The publication years of the 11 systematically reviewed articles ranged from 1996 to 2021. These articles originated from various countries, with most of them coming from the USA (n=5) and Norway (n=2). Among these, patient involvement and engagement in the organisation of organ transplant services was analysed in context with the support of clinical experts (n=5), nurses (n=2) and other healthcare providers (n=2). Medical students were also involved as professionals in one study ( ). The methods used to involve and engage patients included the codevelopment of a peer education programme (n=6) or a communication tool (n=1), the participation in representative or advisory groups or user panels in patient or steering committees (n=5) and a mix of different methods (n=1) ( ). Only two articles described several methods to involve and engage patients. All articles described that well-developed programmes, which involve and engage patients in the organisation of transplant services, provide benefits for patients and foster patient-centredness. However, none of these articles investigated the impact on clinical outcomes such as increased quality of life or improved coping with disease or focused on an ongoing process to involve and engage patients as required by the definition of the National Health Services. Details of deployed methods for patient involvement and engagement as well as the observed outcomes, benefits and findings are summarised in . Compared with 29 conceptual publications on patient involvement and engagement in different healthcare services and settings, the 11 articles on patient involvement and engagement in the organisation of transplantation services focused predominantly on process goals of patient involvement and engagement such as enhancing existing services or the development of new services, but rarely on outcome goals ( ). Only three reported clinical outcome goals were improvements in health status as well as increase in quality of life, while aspects such as increase in service utilisation, satisfaction with services as well as improvements in coping with the disease were not the aims. Furthermore, none of these articles involved and engaged patients at the stage of dissemination and communication ( ). Some systematically reviewed articles investigated aspects that have not been described in the conceptual publications on patient involvement and engagement in other, different healthcare services. Examples of such aspects include dynamic changes in patients’ conditions and perspectives, emotional sensitivities in regard to discussions on deceased organ donation, inadequate or inconsistent information for patients, lack of digital system interoperability and financial barriers for institutions to implement patient involvement and engagement in the organisation of transplantation services, while financial barriers for patients were not identified as an important barrier to such an implementation according to the conceptual publications ( ). Systematically reviewed articles revealed that patients and their relatives/partners were usually involved and engaged in various ways in the organisation of transplant services for patients on the waiting list for transplantation or in the organisation of transplant services for patients in the long-term follow-up period after transplantation or in both phases. All of these articles supposed that clinical transplant services would benefit from active patient involvement and engagement with different goals. The most commonly expressed goal in these articles was the improvement of existing services. In most articles, patients were assigned the following roles: Involvement in the design of focus group discussions to explore patient experiences with transplant services, or in discussions prior to and following transplantation to evaluate transplant service development. The reported approaches to patient involvement and engagement were coproduction, collaboration and consultation. Most of the articles focused on patient involvement and engagement activities in the stages of the design and planning of transplantation services. The most commonly expected mechanism was the support of the interaction between patients and their healthcare providers aiming to address patients’ needs. Notably, inadequate or inconsistent information for patients was identified as most frequent barrier. Other articles identified barriers to the involvement and engagement of patients in the organisation and management of transplant care included dynamic changes in patients’ conditions and perspectives, high emotional sensitivity to discussions on deceased organ donation, financial barriers for the institution, barriers due to lack of time or support for practitioners and lack of system interoperability (the property of systems and medical records to exchange data). All these aspects were not identified through the conceptual publications indicating the specific characteristics of the transplant field. The most frequently proposed strategies to facilitate or to improve patients’ involvement and engagement in the organisation of transplant services were better support for practitioners, adding new modes of communication between practitioners and patients, and more or clearer information for patients. For a summary of details, see . Quality assessment of systematically reviewed articles revealed that no technical report on training or on supporting information for the patients and their relatives or partners who have been involved and engaged in the organisation of transplant services was reported. None of them reported patient involvement and engagement in the plain summary or abstract according to the criteria of current recommendations in GRIPP 2. The results of the assessment of the reporting quality using the GRIPP 2 short form are summarised in . The results of the risk of bias assessment using the CASP tool are summarised in . This systematic review shows that the goals of active patient involvement and engagement in the organisation of transplant services mainly include the improvement of patient education and the development of new services while enhancing existing transplant services was the most common goal. We think that further management goals should also include increased transparency of processes to increase trust in transplantation and in the transplantation system with the goal to achieve more public support for transplantation in times of increasing scarcity of suitable donor organs. Appropriate approaches to patient involvement and engagement such as coproduction, collaboration and consultation can facilitate meaningful patient engagement and involvement in transplant programme design, development and enhancement as well as in process improvement projects. This could be found in the systematically reviewed articles, for example, for the codevelopment of a peer education programme, for the participation in advisory groups or in steering committees and for a mix of different methods. Patients can provide unique perspectives for the design of focus group discussions and for the exploration of patient experiences with transplant services. Patients should be involved in discussions to evaluate transplant service development. Such contributions can be valuable for the improvement of patient-centredness and for the elimination of waste in dysfunctional or suboptimal clinical service processes which are important management goals. The improved delivery of healthcare services by fostering patient-centredness with reduced waste of resources could lead to more efficient resource allocation increasingly strained healthcare systems across the globe. This may also improve work force satisfaction by reducing strain on already overburdened clinical staff. Further studies on patient involvement and engagement should therefore analyse potential quantitative evidence for process improvements, for example, improved cost-effectiveness, quantified reduced waste, shortened process time, improved process reliability, as well as improved clinical outcomes such as improved quality of life and patient survival. This may be supported by deploying Lean Six Sigma methodology for process optimisation in future studies on patient involvement and engagement. The methods summarised in this review cover a set of powerful tools for strong leadership action with high levels of transparency in the management of transplant centres and transplant institutions. These methods create and foster a deeper mutual understanding between patients and their transplant institution and thus enable better and more informed service processes. It appears that all stages of transplant care and all types of organ transplants can be improved by suitable deployment of patient involvement and engagement. The analytical framework which has been developed for the purpose of this systematic review ( ) could also be used as a concise strategic guide for patients’ involvement and engagement in the organisation of transplant services. However, such a strategic approach to transplant centre management requires strong leadership decisions and support as well as experience in change management. This becomes obvious when internal resistance to change by powerful and established decision-makers surfaces in addition to the reported and thus expected barriers to patient involvement and engagement such as dynamic changes in patients’ conditions and perspectives, high emotional sensitivity to discussions on deceased organ donation and lack of system interoperability. As this review shows, these challenges should be weighed against the above mentioned potential benefits. The systematically reviewed articles did not aim at some goals such as an increase in the usage of services or improved satisfaction with services while these goals were aimed at in the conceptual publications from other healthcare fields for example cancer screening. None of them reported the effects of patient involvement and engagement on the dissemination and communication of transplant services while conceptual publications mentioned such effects in the context of other medical fields ( ). Financial barriers for patients as a commonly mentioned barrier has not been reported, while risk of tokenism and barriers due to the lack of support for practitioners also have not been reported while these aspects have been reported as issues in other medical fields ( ). So far only conceptual publications from other medical fields did suggest the clarification of roles, the rationale and the responsibility of patients and the enhancement of access to involvement in order to improve patient involvement and engagement. We believe that future studies and projects that focus on patient involvement and engagement in the organisation of transplantation services likely benefit from considering all of these aspects. This may maximise the utility of future studies by using the lessons learnt in other medical fields. We could not identify any reason why these aspects could not be examined, introduced or discussed for the improvement of patient involvement and engagement in the organisation of transplantation services. Patient involvement and engagement navigated by nurses in cancer care could be shown to significantly improve the outcome of care as well as access to care, continuity of care and clinical efficiencies. Such an approach to a codesign together with patients may lead to better clinical processes also in transplant care while it could improve mutual learning, the sharing of knowledge and the development of new competencies. Different formats and types of patient involvement and engagement such as in temporary project-related taskforces versus longer-term membership in a management board and their effects have not been compared yet. From a management perspective the deployment of different types of patient involvement and engagement at different levels of the organisation is a strategic question that could be answered using our analytical framework depicted in . As the common stages of patient involvement and engagement are setting the priorities for patients, the intervention of patient involvement and engagement should also be prioritised and timed. For example, in the pretransplant phase the priority can be set to identify patients’ preferred ways of communication in order to communicate proactively and caringly about otherwise frequently unvoiced patients’ expectations and fears. Peer-lead workshops and experience sharing between patients, for example, in systematic cooperation with self-help organisations can be very helpful in the post-transplant phase. This likely fosters adherence to immunosuppressive medication which has relevant clinical impact on long-term prognosis. The systematically reviewed articles showed that published projects with patient involvement and engagement in the organisation of transplantation services are still localised activities and small and time-limited projects. These programmes were also described in one publication as time-consuming and energy-intensive while the impact on patient care can be marginal. In order to avoid such an unwanted situation, it is highly relevant that patient involvement and engagement activities are integrated and aligned with an overall strategic approach to improve the organisation in which it operates. We believe that this requires active and consistent leadership within the organisation and system, supportive managers to be involved and encouraged active participation of patients as coinitiators of improvement. The lack of consideration of ethical issues in three systematically reviewed articles and the lack of consideration of the role of the relationship between researchers and participants in almost all of these articles showed how the quality of studies in this area could and should be improved in the future ( ). Further improvements may be achieved by closely following the GRIPP 2 guideline ( ). It is noteworthy that one included study addressed the unique challenges in the paediatric transplant area due to limited resources, a relatively small number of cases, and the need to measure unique but significant factors, such as medication adherence, recipient health and development, and readiness for transitioning to adult care providers, which were ignored by the transplant community for a long time. Furthermore, the patients’ perspectives may be hard to integrate because many paediatric patients who require organ transplants are of very young ages and will need be represented by their parents’ voice. In addition, determining the role of paediatric centres in quality initiatives is complicated by factors such as whether they are standalone facilities, part of an adult-focused programme, or integrated within a larger hospital quality programme. However, the establishment of the Starzl Network for Excellence in Paediatric Transplantation, which aims to incorporate the patient voice at the onset of any improvement project, may significantly contribute to the continuous improvement of transplant practice with integration of emerging technologies with evidence-based patient-centred outcomes and translate evidence into practice in the future. Although studies on other organs or tissues were not identified, which is likely due to the smaller volume of publications on these topics, for example, in the contexts of intestine transplantation or vascularised composite tissue transplantation. We believe that it is important to incorporate the perspectives of patients in these contexts as well which will be more challenging due to lower case numbers of such transplantations, and the current lack of published experiences with patient involvement and engagement in the organisation of such transplantation services. There were several limitations to this review. Although a large number of studies on patient involvement and engagement were identified with the initial search strategy, only a small number of identified articles could be identified that focused on patient involvement and engagement in the organisation of organ transplantation services. Furthermore, negative effects of patient involvement and engagement may be less likely reported due to publication bias. To identify this bias, we used the GRIPP 2 checklist for reporting quality improvements in systematically reviewed articles and recommend the adherence to this checklist in future research on this topic. We derived from this systematic review the following recommendations for future projects on patient involvement and engagement in the organisation of transplant services: To empower patients, the information provided to them should be individualised to prioritise their needs. Both financial and organisational resources are important to successfully implement patient involvement and engagement at least partially. Communication should always be seen as a two-way street, because feedback to health providers is required to improve clinical workflows. Medical staff should be enabled by the provision of resources (eg, working time, training, financial resources) to improve clinical processes effectively by the involvement and engagement of patients. Active and consistent leadership within the organisation and system is required to overcome resistance to change while active participation of patients as coinitiators of improvement should be invited officially by the institution. Ideally, successful projects should analyse the impact on the improvement performance parameters of clinical processes and clinical outcomes of transplantation such as graft survival, patient survival and the improvement of patients’ quality of life. Furthermore, these projects should be focused on a longer-term period to avoid short and time-consuming projects. Reviewer comments Author's manuscript
‘It surprised me a lot that there is a link’: a qualitative study of the acceptability of periodontal treatment for individuals at risk of rheumatoid arthritis	f8078735-2dd1-49fa-8da0-ae116d16549a	10174022	Dental[mh]	The impact of poor oral health may not be well understood by individuals at risk of RA and the health professionals involved in their care. Seeking dental treatment can be hindered by dental anxiety, costs and inequalities around access to dentists. A clinical trial involving preventive periodontal treatment is potentially acceptable for individuals at risk of RA. The initiation of rheumatoid arthritis (RA) is purported to occur at mucosal sites, including the oral cavity, lung and gastrointestinal tract. Here, local inflammation may occur due to a combination of genetic and environmental risk factors. In addition, a bacterial dysbiosis may exist; it is postulated that the combination of inflammation and dysbiosis may trigger a break in immune tolerance, in particular towards citrullinated proteins. Periodontal disease (PD) is a chronic inflammatory disease that destroys the tooth supporting tissues including the alveolar bone, periodontal ligament and fibres, and the overlying gingiva. Globally, it affects 20%–50% of people, with approximately 10% suffering from its severe form. There is mounting evidence associating PD with RA. Both share common genetic and environmental risk factors, including smoking, obesity and socioeconomic status. The prevalence of PD is higher in patients with RA compared with the general population. PD is also increased in individuals at risk of RA, before the onset of clinical arthritis, suggesting that periodontal inflammation precedes joint inflammation. One bacterium of interest is Porphyromonas gingivalis , which is enriched in PD and produces a peptidyl arginine deiminase enzyme; this citrullinates c-terminus arginine residues in a-elonase and fibrinogen—two peptide targets implicated in RA. PD can be effectively treated through surgical and non-surgical interventions. Interestingly, a recent trial reported an improvement in RA disease activity following the treatment of coexistent PD. Early intervention to improve periodontal health in people with early RA or at-risk individuals may, therefore, provide a unique opportunity to delay the progression of RA or potentially prevent its onset. Many individuals at risk of RA have expressed reluctance to take preventive medications, especially when asymptomatic, while making lifestyle changes is perceived to be more acceptable. Periodontal treatment and advice could be considered a non-invasive, low-risk intervention that may provide similar systemic benefits to drug therapy, but without the risk of drug side effects, and has the additional benefit of treating a coexisting disease with its own complications such as pain and tooth loss. Lifestyle change interventions have been successfully demonstrated in patients with type II diabetes, reversing disease onset. A previous qualitative study explored the experiences and priorities concerning oral health, and barriers and facilitators for periodontal trial participation, among patients with established RA. However, to our knowledge, no previous studies have explored perceptions of oral health among individuals at risk of developing RA, nor of the healthcare professionals involved in their care. As successful periodontal treatment is dependent on both adequate service provision and patient adherence, it is necessary to explore the barriers and facilitators for accessing periodontal care and maintenance among these groups. This study aimed to explore the acceptability of periodontal treatment as a measure to potentially prevent RA among at-risk individuals and relevant healthcare professionals. This was a qualitative interview study employing a phenomenological approach to explore the meaning behind participants’ lived experiences. Our study is reported in line with the Consolidated Criteria for Reporting Qualitative Studies framework ( ). 10.1136/rmdopen-2023-003099.supp1 Supplementary data At-risk participants A purposeful sample of CCP+ at-risk individuals, with musculoskeletal (MSK) symptoms but no synovitis, were recruited from the Leeds CCP research cohort. Briefly, this is a national research cohort which recruits individuals presenting with new non-specific MSK symptoms but no clinical synovitis. Those who test positive for anti-CCP antibodies are at risk of developing RA and are followed in the Leeds CCP research clinic. At-risk participants aged 18 and above who were able to give informed consent, and able to speak and understand English, were eligible to participate. Some at-risk individuals who were invited to participate in our qualitative study had already undergone periodontal assessment delivered by a dentist, and had commenced or declined periodontal treatment as part of a separate CCP dental study (IRAS ID 213744). Participants were approached by telephone. Healthcare professional participants A wide range of healthcare professionals working in the planning and delivery of both medical and dental care services, including clinicians, commissioners and policy-makers were invited to take part in this study through purposive sampling using the authors’ professional networks. Some clinicians were involved in providing direct care to CCP+ at-risk individuals, whereas others were National Health Service (NHS) rheumatologists/nurses independent of the cohort/research team and would not be expected to have any specific knowledge of this area. Other healthcare professional participants had an indirect role through their senior leadership position in providing commissioning advice, training of health workforce, etc. Other healthcare professional participants did not have a clinical background, but had a role in policy-making and commissioning. Participants were approached by email. Data collection Individual semistructured interviews were conducted via video or telephone between February 2021 and August 2022, using topic guides ( ). Questions were open-ended and structured around the research aim. Each participant completed a single interview and provided written informed consent prior to their interview. Two female members of the research team (KV-C—a psychologist and senior qualitative researcher, and HS—a clinical academic podiatrist with experience in pre-RA research; both PhD) conducted the interviews with at-risk participants; both were previously unknown to the participants. The researchers conducted the first two interviews together to ensure consensus in the approach to questioning; the remaining interviews were undertaken by one of the two researchers. LSC observed two of the interviews. Healthcare professional participants were interviewed by one male member of the research team (SS—a specialist registrar in dental public health with experience in dental and RA research, PhD), who was known to nine of the 11 participants. All participants were briefed on the purpose of the study and the interviewing researcher’s background and personal motivation, and were given the opportunity to ask questions prior to the interview. All interviews were digitally recorded, transcribed verbatim and supplemented with field notes. The interview duration ranged from 23 to 45 min. While we did not aim for data saturation, which is arguably inappropriate for reflexive thematic analysis, we held ongoing discussions relating to recruitment to ensure our research aim was fully addressed and our final sample size was based on achieving adequate diversity of the sample and depth of data generated from participants. Patient and public involvement Patient and public involvement (PPI) contributors from local dental and rheumatology PPI groups were involved in shaping the research question, and developing the interview topic guide and participant information sheet (PIS) for at-risk participants. The wording in the topic guide and PIS changed as a result of involving PPI contributors, who suggested providing further information to participants about why the study was being conducted and what the interview data would inform. PPI contributors also informed our approach to data collection; we originally intended to conduct the interviews exclusively by telephone, but it was suggested that holding a telephone for an extended period may be difficult for participants with joint symptoms. As a result of PPI, we offered all participants the choice between a video or telephone interview. Analysis Data from interviews with at-risk participants were analysed using reflexive thematic analysis. Interviews were uploaded into NVivo V.12 (QSR International; 2018) and initially coded by one researcher (LSC), who read and reread the transcripts, generated initial codes and collated similar codes. Coding was inductive, with 10% of the transcripts second coded (KV-C), and regular coding discussions held with all other team members. Discrepancies were settled by group consensus. Codes were grouped into provisional themes through a team discussion, then reviewed against the whole data set by one other researcher (ZM). Data from interviews with healthcare professionals were then independently analysed by two researchers (LSC and ZM); coding was deductive, based on the preidentified set of constructs identified from the at-risk participant data. The two researchers discussed any discrepancies in coding until consensus was reached. The entire research team then reviewed and refined the healthcare professional content of each prespecified theme through group discussion. A purposeful sample of CCP+ at-risk individuals, with musculoskeletal (MSK) symptoms but no synovitis, were recruited from the Leeds CCP research cohort. Briefly, this is a national research cohort which recruits individuals presenting with new non-specific MSK symptoms but no clinical synovitis. Those who test positive for anti-CCP antibodies are at risk of developing RA and are followed in the Leeds CCP research clinic. At-risk participants aged 18 and above who were able to give informed consent, and able to speak and understand English, were eligible to participate. Some at-risk individuals who were invited to participate in our qualitative study had already undergone periodontal assessment delivered by a dentist, and had commenced or declined periodontal treatment as part of a separate CCP dental study (IRAS ID 213744). Participants were approached by telephone. A wide range of healthcare professionals working in the planning and delivery of both medical and dental care services, including clinicians, commissioners and policy-makers were invited to take part in this study through purposive sampling using the authors’ professional networks. Some clinicians were involved in providing direct care to CCP+ at-risk individuals, whereas others were National Health Service (NHS) rheumatologists/nurses independent of the cohort/research team and would not be expected to have any specific knowledge of this area. Other healthcare professional participants had an indirect role through their senior leadership position in providing commissioning advice, training of health workforce, etc. Other healthcare professional participants did not have a clinical background, but had a role in policy-making and commissioning. Participants were approached by email. Individual semistructured interviews were conducted via video or telephone between February 2021 and August 2022, using topic guides ( ). Questions were open-ended and structured around the research aim. Each participant completed a single interview and provided written informed consent prior to their interview. Two female members of the research team (KV-C—a psychologist and senior qualitative researcher, and HS—a clinical academic podiatrist with experience in pre-RA research; both PhD) conducted the interviews with at-risk participants; both were previously unknown to the participants. The researchers conducted the first two interviews together to ensure consensus in the approach to questioning; the remaining interviews were undertaken by one of the two researchers. LSC observed two of the interviews. Healthcare professional participants were interviewed by one male member of the research team (SS—a specialist registrar in dental public health with experience in dental and RA research, PhD), who was known to nine of the 11 participants. All participants were briefed on the purpose of the study and the interviewing researcher’s background and personal motivation, and were given the opportunity to ask questions prior to the interview. All interviews were digitally recorded, transcribed verbatim and supplemented with field notes. The interview duration ranged from 23 to 45 min. While we did not aim for data saturation, which is arguably inappropriate for reflexive thematic analysis, we held ongoing discussions relating to recruitment to ensure our research aim was fully addressed and our final sample size was based on achieving adequate diversity of the sample and depth of data generated from participants. Patient and public involvement (PPI) contributors from local dental and rheumatology PPI groups were involved in shaping the research question, and developing the interview topic guide and participant information sheet (PIS) for at-risk participants. The wording in the topic guide and PIS changed as a result of involving PPI contributors, who suggested providing further information to participants about why the study was being conducted and what the interview data would inform. PPI contributors also informed our approach to data collection; we originally intended to conduct the interviews exclusively by telephone, but it was suggested that holding a telephone for an extended period may be difficult for participants with joint symptoms. As a result of PPI, we offered all participants the choice between a video or telephone interview. Data from interviews with at-risk participants were analysed using reflexive thematic analysis. Interviews were uploaded into NVivo V.12 (QSR International; 2018) and initially coded by one researcher (LSC), who read and reread the transcripts, generated initial codes and collated similar codes. Coding was inductive, with 10% of the transcripts second coded (KV-C), and regular coding discussions held with all other team members. Discrepancies were settled by group consensus. Codes were grouped into provisional themes through a team discussion, then reviewed against the whole data set by one other researcher (ZM). Data from interviews with healthcare professionals were then independently analysed by two researchers (LSC and ZM); coding was deductive, based on the preidentified set of constructs identified from the at-risk participant data. The two researchers discussed any discrepancies in coding until consensus was reached. The entire research team then reviewed and refined the healthcare professional content of each prespecified theme through group discussion. Twenty-two individuals at risk of developing RA were approached about the study, and 19 participated. One declined participation due to ill health, while two were withdrawn from the study as they developed inflammatory arthritis prior to being interviewed. Eleven healthcare professionals were also approached about the study, all of whom participated. At-risk participant characteristics and healthcare professional characteristics are presented in , respectively. At-risk participant data were obtained from interview transcripts where possible; medical records were accessed for missing data as required. Three themes (six subthemes) were identified. A conceptual map identifying links between themes is displayed in and an example of the coding tree is provided in . Quotations supporting each theme are presented in ; quotes from at-risk participants are coded with the prefix PQ, while quotes from healthcare professionals are coded with the prefix HPQ. Knowledge of shared at-risk factors Individuals at risk of RA Participants identified various perceived risk factors for RA, including genetics, diet, being overweight, lack of exercise and smoking. However, the majority of participants were unaware of any potential link between poor oral health and RA/the risk of developing RA prior to being invited to participate in a dental research study (PQ1). Some participants recognised the negative effects of smoking on oral health (PQ2, PQ3), and half of participants were aware that smoking would increase their risk of developing RA. Among at-risk participants who were unaware of the link between smoking and the risk of developing RA, the information was unsurprising (PQ4, PQ5). In contrast, one participant commented on the lack of public awareness of the link between poor oral health and RA, including the risk of developing RA (PQ6). Some participants also perceived a potential lack of knowledge among dentists regarding their at-risk status and regarding knowledge of the association between poor oral health and RA/risk of developing RA (PQ7, PQ8, PQ9), and highlighted that from a dental perspective, medical history was focused around any medications they were taking rather than specific conditions (PQ10). Healthcare professionals Healthcare professionals highlighted a disjoin between dentistry and medicine, with some from a medical background acknowledging that it did not occur to them to send their patients to a dentist or ask about oral health, and that communication between medical and dental professionals was rare unless there was a specific problem. Inadequate holistic management of patients was identified from both a dental and rheumatology perspective (HPQ1, HPQ2), but the potential to overcome this was also recognised (HPQ3). Healthcare professionals from a dental background emphasised the difficulties of not having access to complete medical histories for patients, from a safety perspective. They felt that some patients incorrectly assumed that their healthcare records were automatically shared between healthcare professionals, whereas other patients did not recognise the importance of sharing this information and were surprised by the link between oral health and general health (HPQ4). Some healthcare professionals perceived that the disjoin between medicine and dentistry was a result of commissioning and financial barriers and inadequate training (HPQ5, HPQ6). Variations in the extent of collaboration between medicine and dentistry, due to geographical location and research activity, were also acknowledged (HPQ7, HPQ8). Information and communication Individuals at risk of RA Preference for provision of information relating to the association between oral health and RA, and to dental trial participation, varied among participants. One participant perceived that time point preferences would depend on the individual (PQ11). Some participants expressed a preference for verbal information, others preferred written information, and others felt they needed a combination of both (PQ12). Likewise, some participants felt information of this nature was best delivered by a dentist, while others felt it should come from a rheumatologist. One participant suggested a multidisciplinary approach whereby dentists and rheumatologists provide the same information, while others recognised the issues that lack of communication between dental and medical teams posed. A participant with dental phobia expressed a preference for visual aids prior to treatment. Other participants highlighted the importance of feedback and encouragement regarding the impact of dental treatment on their risk of developing RA as a motivator to continue with preventive measures (PQ13). Healthcare professionals One healthcare professional perceived that information relating to the link between oral health and the risk of developing RA should come from the rheumatology team (HPQ9). With regard to the timing of information provision, another felt the link between poor oral health and systemic conditions should be discussed at diagnosis (HPQ10). The use of guidelines and posters was suggested to aid rheumatology teams to ask patients about their oral health. Individuals at risk of RA Participants identified various perceived risk factors for RA, including genetics, diet, being overweight, lack of exercise and smoking. However, the majority of participants were unaware of any potential link between poor oral health and RA/the risk of developing RA prior to being invited to participate in a dental research study (PQ1). Some participants recognised the negative effects of smoking on oral health (PQ2, PQ3), and half of participants were aware that smoking would increase their risk of developing RA. Among at-risk participants who were unaware of the link between smoking and the risk of developing RA, the information was unsurprising (PQ4, PQ5). In contrast, one participant commented on the lack of public awareness of the link between poor oral health and RA, including the risk of developing RA (PQ6). Some participants also perceived a potential lack of knowledge among dentists regarding their at-risk status and regarding knowledge of the association between poor oral health and RA/risk of developing RA (PQ7, PQ8, PQ9), and highlighted that from a dental perspective, medical history was focused around any medications they were taking rather than specific conditions (PQ10). Healthcare professionals Healthcare professionals highlighted a disjoin between dentistry and medicine, with some from a medical background acknowledging that it did not occur to them to send their patients to a dentist or ask about oral health, and that communication between medical and dental professionals was rare unless there was a specific problem. Inadequate holistic management of patients was identified from both a dental and rheumatology perspective (HPQ1, HPQ2), but the potential to overcome this was also recognised (HPQ3). Healthcare professionals from a dental background emphasised the difficulties of not having access to complete medical histories for patients, from a safety perspective. They felt that some patients incorrectly assumed that their healthcare records were automatically shared between healthcare professionals, whereas other patients did not recognise the importance of sharing this information and were surprised by the link between oral health and general health (HPQ4). Some healthcare professionals perceived that the disjoin between medicine and dentistry was a result of commissioning and financial barriers and inadequate training (HPQ5, HPQ6). Variations in the extent of collaboration between medicine and dentistry, due to geographical location and research activity, were also acknowledged (HPQ7, HPQ8). Participants identified various perceived risk factors for RA, including genetics, diet, being overweight, lack of exercise and smoking. However, the majority of participants were unaware of any potential link between poor oral health and RA/the risk of developing RA prior to being invited to participate in a dental research study (PQ1). Some participants recognised the negative effects of smoking on oral health (PQ2, PQ3), and half of participants were aware that smoking would increase their risk of developing RA. Among at-risk participants who were unaware of the link between smoking and the risk of developing RA, the information was unsurprising (PQ4, PQ5). In contrast, one participant commented on the lack of public awareness of the link between poor oral health and RA, including the risk of developing RA (PQ6). Some participants also perceived a potential lack of knowledge among dentists regarding their at-risk status and regarding knowledge of the association between poor oral health and RA/risk of developing RA (PQ7, PQ8, PQ9), and highlighted that from a dental perspective, medical history was focused around any medications they were taking rather than specific conditions (PQ10). Healthcare professionals highlighted a disjoin between dentistry and medicine, with some from a medical background acknowledging that it did not occur to them to send their patients to a dentist or ask about oral health, and that communication between medical and dental professionals was rare unless there was a specific problem. Inadequate holistic management of patients was identified from both a dental and rheumatology perspective (HPQ1, HPQ2), but the potential to overcome this was also recognised (HPQ3). Healthcare professionals from a dental background emphasised the difficulties of not having access to complete medical histories for patients, from a safety perspective. They felt that some patients incorrectly assumed that their healthcare records were automatically shared between healthcare professionals, whereas other patients did not recognise the importance of sharing this information and were surprised by the link between oral health and general health (HPQ4). Some healthcare professionals perceived that the disjoin between medicine and dentistry was a result of commissioning and financial barriers and inadequate training (HPQ5, HPQ6). Variations in the extent of collaboration between medicine and dentistry, due to geographical location and research activity, were also acknowledged (HPQ7, HPQ8). Individuals at risk of RA Preference for provision of information relating to the association between oral health and RA, and to dental trial participation, varied among participants. One participant perceived that time point preferences would depend on the individual (PQ11). Some participants expressed a preference for verbal information, others preferred written information, and others felt they needed a combination of both (PQ12). Likewise, some participants felt information of this nature was best delivered by a dentist, while others felt it should come from a rheumatologist. One participant suggested a multidisciplinary approach whereby dentists and rheumatologists provide the same information, while others recognised the issues that lack of communication between dental and medical teams posed. A participant with dental phobia expressed a preference for visual aids prior to treatment. Other participants highlighted the importance of feedback and encouragement regarding the impact of dental treatment on their risk of developing RA as a motivator to continue with preventive measures (PQ13). Healthcare professionals One healthcare professional perceived that information relating to the link between oral health and the risk of developing RA should come from the rheumatology team (HPQ9). With regard to the timing of information provision, another felt the link between poor oral health and systemic conditions should be discussed at diagnosis (HPQ10). The use of guidelines and posters was suggested to aid rheumatology teams to ask patients about their oral health. Preference for provision of information relating to the association between oral health and RA, and to dental trial participation, varied among participants. One participant perceived that time point preferences would depend on the individual (PQ11). Some participants expressed a preference for verbal information, others preferred written information, and others felt they needed a combination of both (PQ12). Likewise, some participants felt information of this nature was best delivered by a dentist, while others felt it should come from a rheumatologist. One participant suggested a multidisciplinary approach whereby dentists and rheumatologists provide the same information, while others recognised the issues that lack of communication between dental and medical teams posed. A participant with dental phobia expressed a preference for visual aids prior to treatment. Other participants highlighted the importance of feedback and encouragement regarding the impact of dental treatment on their risk of developing RA as a motivator to continue with preventive measures (PQ13). One healthcare professional perceived that information relating to the link between oral health and the risk of developing RA should come from the rheumatology team (HPQ9). With regard to the timing of information provision, another felt the link between poor oral health and systemic conditions should be discussed at diagnosis (HPQ10). The use of guidelines and posters was suggested to aid rheumatology teams to ask patients about their oral health. Personal challenges and opportunities for dental intervention and oral health maintenance Individuals at risk of RA The majority of participants made routine visits to a dentist; however, negative perceptions of these visits were common. Participants described ‘hate’ towards the experience of visiting the dentist, ‘a little bit of fear at the general thought of it’ and ‘tensing up’. Some participants explicitly expressed a phobia of dentists and attributed their anxiety to traumatic dental experiences during childhood (PQ14). A minority of participants were comfortable visiting the dentist, with no anxiety whatsoever, but still perceived that ‘people don’t like dentists’. Comorbidities also impacted on the perceptions and priorities around oral health for some participants. One participant noted that her reflux had caused multiple fillings. Another acknowledged that his leaking heart valve meant he was supposed to look after his teeth; this participant had explicitly made his dentist aware of his heart condition, but was unsure if the dentist knew he was CCP+ at risk (PQ15). Oral health was identified as less of a priority when compared with conditions such as irritable bowel syndrome, which had a greater impact on daily life (PQ16). Another participant expressed how stress and anxiety made oral health less of a priority (PQ17). Healthcare professionals Healthcare professionals also identified dental anxiety as an issue for patients, perceiving that some patients avoided going to the dentist despite needing treatment (HPQ11, HPQ12). External barriers to dental intervention and oral health maintenance Individuals at risk of RA Many participants expressed difficulty in accessing an NHS dentist, particularly since the COVID-19 pandemic. Some participants had gone to a private dentist as a result (PQ18). One participant who did have an NHS dentist felt ‘lucky’, but emphasised the short duration of NHS dental appointments, while another expressed how her NHS dentist had not explained she had gum disease or given any advice on how to address it. This was attributed to lack of time during NHS dental appointments (PQ19, PQ20). While a minority of participants confirmed that the cost of dental treatment was not a problem for them, many participants identified that the cost of dental treatment had previously been or was still an issue. Some noted that although cost was an issue, it had not stopped them going to the dentist, whereas others explicitly stated that the cost had an impact on how frequently they were able to go. Cost also impacted on oral health maintenance; for example, one participant could not afford the upkeep of his dentures after volunteering for a limited course of free treatment at a dental school. Another participant perceived that dentistry was about making money rather than about health, while another felt that dentists had not given enough advice regarding the link between oral health and general health as a motivator to maintain good oral health habits (PQ21). Healthcare professionals Healthcare professionals identified similar barriers to dental intervention, highlighting that patients, both their own and in the wider sense, had difficulties accessing NHS dentists (HPQ13). Some attributed these access difficulties to social deprivation (HPQ14, HPQ15). This perceived lack of access to a dentist had a potential impact on healthcare professionals’ management of patients (HPQ16, HPQ17). Some healthcare professionals also identified cost as a barrier to seeking dental treatment. This included cost of both NHS treatment, and private treatment when NHS access was not possible (HPQ18, HPQ19). Individuals at risk of RA The majority of participants made routine visits to a dentist; however, negative perceptions of these visits were common. Participants described ‘hate’ towards the experience of visiting the dentist, ‘a little bit of fear at the general thought of it’ and ‘tensing up’. Some participants explicitly expressed a phobia of dentists and attributed their anxiety to traumatic dental experiences during childhood (PQ14). A minority of participants were comfortable visiting the dentist, with no anxiety whatsoever, but still perceived that ‘people don’t like dentists’. Comorbidities also impacted on the perceptions and priorities around oral health for some participants. One participant noted that her reflux had caused multiple fillings. Another acknowledged that his leaking heart valve meant he was supposed to look after his teeth; this participant had explicitly made his dentist aware of his heart condition, but was unsure if the dentist knew he was CCP+ at risk (PQ15). Oral health was identified as less of a priority when compared with conditions such as irritable bowel syndrome, which had a greater impact on daily life (PQ16). Another participant expressed how stress and anxiety made oral health less of a priority (PQ17). Healthcare professionals Healthcare professionals also identified dental anxiety as an issue for patients, perceiving that some patients avoided going to the dentist despite needing treatment (HPQ11, HPQ12). The majority of participants made routine visits to a dentist; however, negative perceptions of these visits were common. Participants described ‘hate’ towards the experience of visiting the dentist, ‘a little bit of fear at the general thought of it’ and ‘tensing up’. Some participants explicitly expressed a phobia of dentists and attributed their anxiety to traumatic dental experiences during childhood (PQ14). A minority of participants were comfortable visiting the dentist, with no anxiety whatsoever, but still perceived that ‘people don’t like dentists’. Comorbidities also impacted on the perceptions and priorities around oral health for some participants. One participant noted that her reflux had caused multiple fillings. Another acknowledged that his leaking heart valve meant he was supposed to look after his teeth; this participant had explicitly made his dentist aware of his heart condition, but was unsure if the dentist knew he was CCP+ at risk (PQ15). Oral health was identified as less of a priority when compared with conditions such as irritable bowel syndrome, which had a greater impact on daily life (PQ16). Another participant expressed how stress and anxiety made oral health less of a priority (PQ17). Healthcare professionals also identified dental anxiety as an issue for patients, perceiving that some patients avoided going to the dentist despite needing treatment (HPQ11, HPQ12). Individuals at risk of RA Many participants expressed difficulty in accessing an NHS dentist, particularly since the COVID-19 pandemic. Some participants had gone to a private dentist as a result (PQ18). One participant who did have an NHS dentist felt ‘lucky’, but emphasised the short duration of NHS dental appointments, while another expressed how her NHS dentist had not explained she had gum disease or given any advice on how to address it. This was attributed to lack of time during NHS dental appointments (PQ19, PQ20). While a minority of participants confirmed that the cost of dental treatment was not a problem for them, many participants identified that the cost of dental treatment had previously been or was still an issue. Some noted that although cost was an issue, it had not stopped them going to the dentist, whereas others explicitly stated that the cost had an impact on how frequently they were able to go. Cost also impacted on oral health maintenance; for example, one participant could not afford the upkeep of his dentures after volunteering for a limited course of free treatment at a dental school. Another participant perceived that dentistry was about making money rather than about health, while another felt that dentists had not given enough advice regarding the link between oral health and general health as a motivator to maintain good oral health habits (PQ21). Healthcare professionals Healthcare professionals identified similar barriers to dental intervention, highlighting that patients, both their own and in the wider sense, had difficulties accessing NHS dentists (HPQ13). Some attributed these access difficulties to social deprivation (HPQ14, HPQ15). This perceived lack of access to a dentist had a potential impact on healthcare professionals’ management of patients (HPQ16, HPQ17). Some healthcare professionals also identified cost as a barrier to seeking dental treatment. This included cost of both NHS treatment, and private treatment when NHS access was not possible (HPQ18, HPQ19). Many participants expressed difficulty in accessing an NHS dentist, particularly since the COVID-19 pandemic. Some participants had gone to a private dentist as a result (PQ18). One participant who did have an NHS dentist felt ‘lucky’, but emphasised the short duration of NHS dental appointments, while another expressed how her NHS dentist had not explained she had gum disease or given any advice on how to address it. This was attributed to lack of time during NHS dental appointments (PQ19, PQ20). While a minority of participants confirmed that the cost of dental treatment was not a problem for them, many participants identified that the cost of dental treatment had previously been or was still an issue. Some noted that although cost was an issue, it had not stopped them going to the dentist, whereas others explicitly stated that the cost had an impact on how frequently they were able to go. Cost also impacted on oral health maintenance; for example, one participant could not afford the upkeep of his dentures after volunteering for a limited course of free treatment at a dental school. Another participant perceived that dentistry was about making money rather than about health, while another felt that dentists had not given enough advice regarding the link between oral health and general health as a motivator to maintain good oral health habits (PQ21). Healthcare professionals identified similar barriers to dental intervention, highlighting that patients, both their own and in the wider sense, had difficulties accessing NHS dentists (HPQ13). Some attributed these access difficulties to social deprivation (HPQ14, HPQ15). This perceived lack of access to a dentist had a potential impact on healthcare professionals’ management of patients (HPQ16, HPQ17). Some healthcare professionals also identified cost as a barrier to seeking dental treatment. This included cost of both NHS treatment, and private treatment when NHS access was not possible (HPQ18, HPQ19). Making oral health changes with the aim of preventing RA Individuals at risk of RA Participants discussed oral health issues such as bleeding and sore gums, chipped and weak teeth, infections, missing teeth and self-extraction. In some cases, oral health issues were closely linked to the external barriers identified in Theme 2 (PQ22). A minority of participants stated they had no problems with their oral health. Participants described varying levels of oral health maintenance, including regular brushing, flossing, use of interdental brushes, mouthwash, and electric toothbrushes, avoidance or reduction of carbonated drinks and sugary snacks, and drinking through a straw. Although half of participants described experiencing symptoms in their hands, and discussed how joint pain had led to limitation or modification of activities, being unable to work, and relying on family members for personal care, only two reported that these symptoms had caused difficulties with oral health maintenance. Among participants who were previously aware of the link between oral health and developing RA, some had actively made changes; for example, visiting the dental hygienist more often, having a better brushing routine and quitting or reducing smoking (PQ23, PQ24). One participant stated that being at risk of developing RA resulted in being willing to pay for dental treatment. Another reported that she would only seek dental treatment due to being at risk of RA if a dentist recommended it (PQ25), whereas being told about the link between developing RA and oral health during the interview was enough for another participant to state that she would prioritise her oral health (PQ26). Healthcare professionals Some healthcare professionals identified that the importance of good oral health behaviours might be underestimated, by people in general and by rheumatology patients specifically (HPQ20, HPQ21). They concluded that patients who had previously neglected their oral health would have difficulties changing their behaviour (HPQ22). Acceptability of participation in periodontal research aiming to prevent RA Individuals at risk of RA Seventeen of the 19 participants reported that a clinical trial aiming to reduce the risk of RA through dental treatment would be acceptable to them (PQ27). In contrast, a clinical trial aiming to reduce the risk of RA through taking a medicine was less acceptable; participants identified the need to consider their risk level and side effects of the drug. Facilitators to participating in a dental trial included the personal benefits of being able to reduce their risk of developing RA (PQ28) and access free dental treatment (PQ29), and the wider societal benefit of being able to potentially help others in the future (PQ30). A participant with dental phobia felt that the acceptability of this type of trial was dependent on the clinician carrying out the treatment, and that pain or discomfort during treatment would influence his decision to participate. Other participants recognised the importance of seeing the same dentist at every visit was important and felt they would be more comfortable receiving treatment from their routine dentist rather than a new dentist (PQ31). In contrast, another participant felt that treatment as part of a research study should be done by a specialist rather than at a routine dentist appointment. Other potential barriers to dental trial participation included the location of treatment and appointment times. Some participants suggesting that a smaller, less clinical environment would be better for people with dental phobia, while others focused on ease of parking nearby and public transport routes. Participants highlighted other commitments that could affect their ability to participate, such as childcare and work (PQ32). One participant perceived that she had no oral health problems, so participation in the trial would not be a priority for her. Healthcare professionals While healthcare professionals understood the advantages of a multidisciplinary approach to managing patients with systemic diseases, some identified the cost of dentists’ time as a potential barrier. Others perceived that this could be overcome by using other members of the dental workforce (HPQ23, HPQ24). In relation to provision of preventive dental treatment for individuals at risk of RA, cost-effectiveness was also highlighted (HPQ25). The extent to which individuals at risk of developing RA would be open to receiving preventive dental treatment was perceived to depend on their current oral health behaviours (HPQ26). Individuals at risk of RA Participants discussed oral health issues such as bleeding and sore gums, chipped and weak teeth, infections, missing teeth and self-extraction. In some cases, oral health issues were closely linked to the external barriers identified in Theme 2 (PQ22). A minority of participants stated they had no problems with their oral health. Participants described varying levels of oral health maintenance, including regular brushing, flossing, use of interdental brushes, mouthwash, and electric toothbrushes, avoidance or reduction of carbonated drinks and sugary snacks, and drinking through a straw. Although half of participants described experiencing symptoms in their hands, and discussed how joint pain had led to limitation or modification of activities, being unable to work, and relying on family members for personal care, only two reported that these symptoms had caused difficulties with oral health maintenance. Among participants who were previously aware of the link between oral health and developing RA, some had actively made changes; for example, visiting the dental hygienist more often, having a better brushing routine and quitting or reducing smoking (PQ23, PQ24). One participant stated that being at risk of developing RA resulted in being willing to pay for dental treatment. Another reported that she would only seek dental treatment due to being at risk of RA if a dentist recommended it (PQ25), whereas being told about the link between developing RA and oral health during the interview was enough for another participant to state that she would prioritise her oral health (PQ26). Healthcare professionals Some healthcare professionals identified that the importance of good oral health behaviours might be underestimated, by people in general and by rheumatology patients specifically (HPQ20, HPQ21). They concluded that patients who had previously neglected their oral health would have difficulties changing their behaviour (HPQ22). Participants discussed oral health issues such as bleeding and sore gums, chipped and weak teeth, infections, missing teeth and self-extraction. In some cases, oral health issues were closely linked to the external barriers identified in Theme 2 (PQ22). A minority of participants stated they had no problems with their oral health. Participants described varying levels of oral health maintenance, including regular brushing, flossing, use of interdental brushes, mouthwash, and electric toothbrushes, avoidance or reduction of carbonated drinks and sugary snacks, and drinking through a straw. Although half of participants described experiencing symptoms in their hands, and discussed how joint pain had led to limitation or modification of activities, being unable to work, and relying on family members for personal care, only two reported that these symptoms had caused difficulties with oral health maintenance. Among participants who were previously aware of the link between oral health and developing RA, some had actively made changes; for example, visiting the dental hygienist more often, having a better brushing routine and quitting or reducing smoking (PQ23, PQ24). One participant stated that being at risk of developing RA resulted in being willing to pay for dental treatment. Another reported that she would only seek dental treatment due to being at risk of RA if a dentist recommended it (PQ25), whereas being told about the link between developing RA and oral health during the interview was enough for another participant to state that she would prioritise her oral health (PQ26). Some healthcare professionals identified that the importance of good oral health behaviours might be underestimated, by people in general and by rheumatology patients specifically (HPQ20, HPQ21). They concluded that patients who had previously neglected their oral health would have difficulties changing their behaviour (HPQ22). Individuals at risk of RA Seventeen of the 19 participants reported that a clinical trial aiming to reduce the risk of RA through dental treatment would be acceptable to them (PQ27). In contrast, a clinical trial aiming to reduce the risk of RA through taking a medicine was less acceptable; participants identified the need to consider their risk level and side effects of the drug. Facilitators to participating in a dental trial included the personal benefits of being able to reduce their risk of developing RA (PQ28) and access free dental treatment (PQ29), and the wider societal benefit of being able to potentially help others in the future (PQ30). A participant with dental phobia felt that the acceptability of this type of trial was dependent on the clinician carrying out the treatment, and that pain or discomfort during treatment would influence his decision to participate. Other participants recognised the importance of seeing the same dentist at every visit was important and felt they would be more comfortable receiving treatment from their routine dentist rather than a new dentist (PQ31). In contrast, another participant felt that treatment as part of a research study should be done by a specialist rather than at a routine dentist appointment. Other potential barriers to dental trial participation included the location of treatment and appointment times. Some participants suggesting that a smaller, less clinical environment would be better for people with dental phobia, while others focused on ease of parking nearby and public transport routes. Participants highlighted other commitments that could affect their ability to participate, such as childcare and work (PQ32). One participant perceived that she had no oral health problems, so participation in the trial would not be a priority for her. Healthcare professionals While healthcare professionals understood the advantages of a multidisciplinary approach to managing patients with systemic diseases, some identified the cost of dentists’ time as a potential barrier. Others perceived that this could be overcome by using other members of the dental workforce (HPQ23, HPQ24). In relation to provision of preventive dental treatment for individuals at risk of RA, cost-effectiveness was also highlighted (HPQ25). The extent to which individuals at risk of developing RA would be open to receiving preventive dental treatment was perceived to depend on their current oral health behaviours (HPQ26). Seventeen of the 19 participants reported that a clinical trial aiming to reduce the risk of RA through dental treatment would be acceptable to them (PQ27). In contrast, a clinical trial aiming to reduce the risk of RA through taking a medicine was less acceptable; participants identified the need to consider their risk level and side effects of the drug. Facilitators to participating in a dental trial included the personal benefits of being able to reduce their risk of developing RA (PQ28) and access free dental treatment (PQ29), and the wider societal benefit of being able to potentially help others in the future (PQ30). A participant with dental phobia felt that the acceptability of this type of trial was dependent on the clinician carrying out the treatment, and that pain or discomfort during treatment would influence his decision to participate. Other participants recognised the importance of seeing the same dentist at every visit was important and felt they would be more comfortable receiving treatment from their routine dentist rather than a new dentist (PQ31). In contrast, another participant felt that treatment as part of a research study should be done by a specialist rather than at a routine dentist appointment. Other potential barriers to dental trial participation included the location of treatment and appointment times. Some participants suggesting that a smaller, less clinical environment would be better for people with dental phobia, while others focused on ease of parking nearby and public transport routes. Participants highlighted other commitments that could affect their ability to participate, such as childcare and work (PQ32). One participant perceived that she had no oral health problems, so participation in the trial would not be a priority for her. While healthcare professionals understood the advantages of a multidisciplinary approach to managing patients with systemic diseases, some identified the cost of dentists’ time as a potential barrier. Others perceived that this could be overcome by using other members of the dental workforce (HPQ23, HPQ24). In relation to provision of preventive dental treatment for individuals at risk of RA, cost-effectiveness was also highlighted (HPQ25). The extent to which individuals at risk of developing RA would be open to receiving preventive dental treatment was perceived to depend on their current oral health behaviours (HPQ26). This study informs our understanding of the perceptions and experiences of oral health among individuals at risk of RA. Our findings indicate that dental intervention and oral health maintenance to reduce the risk of developing RA are generally perceived to be acceptable among at-risk individuals, congruent with previous studies whereby at-risk individuals were more willing to make lifestyle changes and adopt healthy behaviours than to take preventive medication. Despite growing evidence suggesting an association between PD and RA, our findings indicate that awareness of this association is limited both among patients as well as healthcare professionals. This may reflect a wider disconnect between medicine and dentistry, which was highlighted by both at-risk participants and healthcare professionals. Models to address the siloed delivery of medicine and dentistry are being explored in the UK with the view of developing pathways that facilitate access to care for high needs patients, including those with multimorbidities. Our findings emphasise that dental anxiety can impact on patients’ dental care seeking behaviours. This is congruent with a previous study focusing on attitudes towards oral health in patients with established RA, whereby previous negative experiences of dental care discouraged participation in a periodontal trial. Our study also highlighted that pursuing dental treatment can be hindered by treatment costs and further inequalities around access to an NHS dentist. This reflects inequalities throughout Europe; in 2021, 5% of the European Union (EU) had an unmet need for dental examination or treatment, due to cost, distance and waiting lists. In England, primary dental care under the NHS is not free at the point of delivery and most patients are expected to pay for their treatment, with a few exceptions such as children, pregnant women and people receiving certain state benefits. Access to NHS dental care has been highlighted as an increasingly pressing problem, especially for people living in the most socially deprived areas. Several patient organisations, trade unions and even cross-party parliamentary groups have been calling for dental system reform. The newly established Integrated Care Systems are expected to take over the commissioning responsibilities for both medical and dental care services in England, creating new opportunities for reducing the siloed delivery of these services. There have been various initiatives aimed at reducing the barriers between healthcare services, such as Making Every Contact Count (MECC). MECC aims to maximise the benefits of the interactions between various healthcare settings and patients by promoting evidence-based preventive messages and signposting between healthcare services. Dental care professionals could have an important role in providing preventive interventions and early detection of chronic conditions by capturing non-regular attendees of general healthcare services. With an expected increase in the number of people with multimorbidities, there are unique opportunities to design more person-centred, integrated medical and dental care services delivered by multidisciplinary teams. Our study has implications for clinical practice and future clinical trials. Rheumatology teams should consider oral health as part of the holistic management of RA including those at risk of developing RA, while dental care professionals should consider the implications of being CCP+ at risk when managing these patients. Whereas patients with established RA have reported difficulties in maintaining their oral hygiene due to their joint problems and a burden of numerous different hospital appointments linked to having RA, leading them to deprioritise oral health, targeting CCP+ at-risk individuals bestows an earlier opportunity to provide information and advice that may be easier to act on. When designing and conducting pre-RA clinical studies, researchers must consider the barriers and facilitators to participation reported by patients. A personalised approach that considers each participant’s level of dental anxiety, their preferred methods of information delivery, location accessibility and appointment time flexibility is recommended. Future studies in this area should focus on how at-risk individuals assess risk versus benefit in deciding on participation in preventive periodontal studies, addressing the research agenda within recent guidance for conducting clinical trials and observational studies in individuals at risk of RA. This agenda also highlights the need to understand which risk factors at-risk individuals consider to be high risk for developing RA. Our study has started to address this point, but emphasises the importance of at-risk individuals to understand all potential risk factors before considering which are high risk. Future research should also explore potential differences between CCP+ at-risk individuals who accept certain preventive measures and those who decline, to determine what factors influence different approaches to health behaviours. Our findings must be viewed in light of some limitations. First, health professionals were recruited through the authors’ professional networks and most of those interviewed were known to the interviewer, which may have introduced bias during data collection. However, in an attempt to minimise bias, the authors took care to recruit fellow health professionals who they felt did not have additional specific experience or knowledge around the impact of PD on RA, and all interviews were independently analysed by a member of the research team who is not from a dental background and had no preconceptions. Second, the rationale for this study was to explore the acceptability of periodontal treatment from the patient perspective, with input from health professionals to triangulate and explore factors that might influence patient participation. Our analysis, therefore, commenced with the patient interviews and was widened out from there, but not all data from the health professional interviews related to the patient perspective. In addition, we acknowledge that our sample of at-risk participants were already part of a CCP cohort study, and may not reflect the views of at-risk individuals who are less willing to participate in research. Nevertheless, to our knowledge, this is the first qualitative study to explore the perceptions and experiences of periodontal health in this population and provides a grounding for future research supporting the design of preventive interventional periodontal studies in individuals at risk of developing RA. The association between poor oral health and RA may not be well understood by individuals at risk of RA and the healthcare professionals involved in their care. Information relating to this association should be tailored to the individual. While PD is common in individuals at risk of RA, seeking dental treatment can be hindered by dental phobia, treatment cost and inequalities around access to an NHS dentist. A clinical trial involving preventive periodontal treatment is potentially acceptable for individuals at risk of RA.
Effects of defoliation and nitrogen on carbon dioxide (CO	0041c088-e8aa-4535-b24d-97b2136b7f79	10174058	Microbiology[mh]	Soil respiration is the primary mode of carbon dioxide (CO 2 ) exchange between the soil carbon pool and atmosphere and one of the largest fluxes in the carbon cycle of terrestrial ecosystems (78–98 Gt C per year) ( ; ). Accordingly, soil respiration can alter CO 2 in the atmosphere and carbon storage in soils, thus affecting the global carbon cycle ( ). As a vital component of land respiration in agricultural and forestry ecosystems, soil respiration is readily influenced by vegetation types, cultivation, fertilization, and other active jamming factors ( ; ; ). Reportedly, exogenous carbon inputs affect soil nutrient availability, increase microbial activities and quantity, and change the activity of soil enzymes, thus contributing to the priming effect of CO 2 emissions ( ; ; ; ). Most of the existing studies focused on the association of soil carbon dynamics and microbial communities with labile carbon inputs ( ; ; ; ; ). In addition, the effects of complex polymerized organism inputs (such as straw and litter) on soil carbon dynamics and microbial communities have been analyzed mainly in farmland and forestry ecosystems ( ; ; ; ) but rarely in orchard ecosystems ( ). Therefore, the study of defoliation inputs to orchard soils can objectively and truly reflect the characteristics of soil CO 2 emissions and microbes in the orchard industry. Nitrogen fertilizer plays a pivotal role in the management of orchards, which can increase soil nutrients and change microbial activities and communities ( ; ; ; ), and affect soil respiration ( ; ; ). Of note, some hypotheses have been proposed on the mechanism of nitrogen. For instance, the theory of “Microbial stoichiometric decomposition” ( ; ; ) demonstrates that exogenous substance inputs can contribute to high microbial activities and organic matter decomposition, which simulate high CO 2 emissions, by meeting the carbon/nitrogen demand of soil microbes, indicating that high nitrogen availability (abundant nutrients) may favor soil organic matter (SOM) decomposition. The “microbial nitrogen mining” theory shows that soil microbes decompose SOM with labile carbon as an energy source to obtain the required nitrogen and then induce carbon priming effects under low nitrogen conditions ( ; ; ; ; ), illustrating that low nitrogen availability (poor nutrients) stimulates SOM decomposition. Despite the obvious odds between the two theories, their mechanisms are closely related to organic carbon, which suggests an inherently critical association of soil carbon respiration with carbon and nitrogen inputs. In this context, there is a need to explore the impacts of exogenous defoliation inputs on CO 2 emissions and microbial mechanisms under different nitrogen levels, thus providing more data and theoretical support for understanding the soil carbon cycle. Responses of soil CO 2 emissions to different environments are variable because of different soil environments and physicochemical characteristics in soils at varying depths ( ; ; ). In top soils (≤10 cm), large amounts of SOC are produced and CO 2 emissions are increased because of litter, fertilization, soil fungi, bacteria, and animals ( ; ; ). Conversely, deep soils can sequestrate more exogenous carbon than top soils because of low SOC ( ; ). Consequently, it is urgent to investigate whether soil respiration in soils at different depths is affected by defoliation and nitrogen additions, thereby providing data to support the development of soil carbon sequestration science. The cherry industry is highly economically profitable, which has contributed to its rapid growth in northwest China. In this context, this industry has become a new source of income for local farmers to shake off poverty. Accordingly, it is necessary to explore microbial mechanisms of CO 2 emissions in dryland cherry orchards with defoliation and nitrogen addition, thus providing more data and theoretical support for sustainable development of cherry industry. In this study, soils at different depths were collected from a rain-fed cherry orchard in northwest China for an indoor incubation experiment was performed to ascertain the microbial mechanism regulating CO 2 emissions under defoliation and nitrogen addition. Then, the following three hypotheses were proposed: (1) defoliation and nitrogen addition promoted CO 2 emissions; (2) CO 2 emissions were strongly associated with microbial activities and community; (3) microbial communities varied in soils at different depths. Soil collection Soil samples were obtained at three depths (0–10 cm [top soils], 10–30 cm [middle soils], and 30–60 cm [deep soils]) of the botanical garden test site at Tianshui Normal University (Tianshui, Gansu, China; 34°34′10″N and 105°41′47″E) in 2021. The test site was built in 2002, where the cherry rootstock was Gisela 5 and the cherry variety was Provence. The basic physical and chemical properties of top, middle, and deep soils were as follows: pH: 6.7, 7.8, and 8.1; total organic carbon: 14.74 g kg −1 , 12.54 g kg −1 , and 7.92 g kg −1 ; total nitrogen: 0.74 g kg −1 , 0.69 g kg −1 , and 0.46 g kg −1 ; available phosphorus: 4.93 mg kg −1 , 5.08 mg kg −1 , and 3.07 mg kg −1 ; available potassium: 153.0 mg kg −1 , 148.0 mg kg −1 , and 100.6 mg kg −1 . Experimental design The incubation experiment was conducted as a complete factorial experiment of soils at three depths (top, middle, and deep soils) * two kinds of defoliation addition (no-defoliation [C 0 ] and defoliation [C A , 1%]) * addition of three levels of nitrogen (0 mg kg −1 [N 0 ], 90 mg kg −1 [N L ], and 135 mg kg −1 [N H ]) in three replicates with a fully randomized design. Total organic carbon was 451 g kg −1 and total nitrogen was 12.47 g kg −1 in defoliation. Incubation tanks (1 L) were respectively added with 100 g soils (dry weight) at different depths for 7 days of pre-incubation at 25 °C ( ). After that, soils were fully mixed with defoliation according to different treatments. Nitrogen and phosphate fertilizers were dissolved in distilled water and added to soils as a solution. Next, the soils were incubated at 25 °C with the moisture of 60% and the bulk density of 1.2 g cm −3 in the dark for 80 days. Gas collection and soil sampling For each treatment, three tanks were taken out at 1, 4, 13, and 80 days. The soils in the incubation tanks were stored at −20 °C for detecting microbial biomass carbon (MBC), soil enzyme activities, and microbial communities. Specifically, MBC was measured with the chloroform-fumigation extraction method ( ). The activity of alkaline phosphatase was examined with the method of Tabatabai and Bremner ( ) as described in a prior study ( ). The activity of cellulase was detected using the method of Xu and Zheng ( ). The activity of catalase was determined by titrating 0.1 mol L −1 KMnO 4 ( ). DNA extraction and high-throughput sequencing Total DNA was extracted from soils with the Fast DNA SPIN Kit for Soil and FastPrep-24 nucleic acid Extraction instrument (MP Biomedicals, Santa Ana, CA, USA) on the 4th day. The DNA was examined with 1% agarose gel electrophoresis, and its concentration was measured with NanoDrop 2000. The DNA was stored in a refrigerator at −20 °C for subsequent use. The stock solution of the DNA was diluted to about 5 mg/L as the template of polymerase chain reaction (PCR) amplification. Afterward, the 16S rRNA variable regions V3-V4 of bacteria were subjected to PCR amplification with the following primers: CCTAYGGGRBGCASCAG (341F) and GGACTACNNGGGTATCTAAT (806R). The PCR was conducted with a system of 50 µL, including 5 µL of 10 * buffer, 4 µL of dNTP, 0.5 µL of RTAQ (Takara), 1 µL of 10 µmol L −1 each front and rear primers, 36.5 µL of ddH 2 O, and 2 µL of template DNA, and the detection of each sample was repeated three times. The amplification conditions were as follows: pre-denaturation at 98 °C for 30 s, 30 cycles of chain disassembly at 98 °C for 10 s, annealing at 55 °C for 15 s, and extension at 72 °C for 1 min, and final extension at 72 °C for 10 min. The products obtained by three repeats of DNA amplification were mixed and tested with 1% agarose gel electrophoresis. The concentration of the obtained bacterial PCR products was measured with the PicoGreen kit. After the products were evenly mixed, DNA was purified and recovered with the DNA purification kit (TIANGEN Biotech, Beijing, China). Bacterial 16S rDNA was sequenced by Shenzhen Weicomeng Technology Group Co., Ltd. (Shenzhen, China) with the Novaseq 6000 PE 250 platform. According to corresponding barcodes, the samples were subjected to paired reading, and then their barcode and primer sequence were removed. The reads at both ends were combined with the FLASH (V1.2.7) software. The QIIME data analysis package was utilized to remove low-quality raw sequences (length < 250 bp, ambiguous base “N”, and the average base quality score less than 20). The chimeric sequence was discarded with the MAARJAM database ( https://maarjam.ut.ee/ ). The valid readings were assigned to operational taxon units or virtual taxa with an identity threshold of 97%, and the representative sequences were identified with SILVA (V128, http://www.arb-silva.de ) and MAARJAM databases. Calculation CO 2 emissions (mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 ) were calculated as per the titration results of hydrochloric acids. [12pt]{minimal} }{}l@{}} C{O}_{2}= _{0}-V) {C}_{HCl}}{2} 12 1000. C O 2 = V 0 − V × C H C l 2 × 12 × 1 m 1 − a × 1000 . In this formula, V 0 was the volume of titration by standard hydrochloric acid during blank calibration (mL), while V represented the volume of titration by standard hydrochloric acid during samples (mL). C HCl was the concentration of standard hydrochloric acid (1 mol L −1 ), m was soil mass (g), and a was soil water content (%). The CO 2 efflux rate (mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) was calculated with the following formula: CO 2 efflux rate = CO 2 emission/t. In this formula, t represented the day when the NaOH solution was placed in the incubation tanks. The cumulative CO 2 emission (g CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 ) was the cumulative CO 2 emission from each treatment over a given incubation time. For a given incubation time (80 days), the priming index (PI) induced by exogenous substance addition was normalized to the proportion of added non-exogenous substances to cumulative CO 2 emissions based on the following formula: Priming index (PI) = (CO 2 add − CO 2 non-add )/ CO 2 non-add . In this formula, CO 2 add represented cumulative CO 2 emissions after 80 days of defoliation and nitrogen treatment, and CO 2 non-add indicated cumulative CO 2 emissions after 80 days of treatment without defoliation and nitrogen. The PI represented the intensity of the priming effect. The PI value of 1 suggested that the amount of organic carbon mineralization was not affected by exogenous substance addition. The PI value of >1 indicated that the addition of exogenous substances caused the priming effect of organic carbon mineralization, and the larger value was associated with a stronger priming effect. The PI value of <1 represented that the addition of exogenous substances reduced the mineralization of organic carbon and produced a negative priming effect, and the smaller value illustrated the stronger negative priming effect. Statistical analysis All statistical analyses were performed with the SPSS 18.0 statistical and R programming software. The effects of the CO 2 efflux rate, cumulative CO 2 emission, PI, MBC, catalase, alkaline phosphatase, cellulase, Chao1, Shannon, and Simpson and the relative abundance of Proteobacteria and Acidobacteria were analyzed with three-way analysis of variance combined with Duncan’s multiple range test. Partial least squares discrimination analysis was conducted to analyze the distribution of soil microbial communities in soils at different depths. The Spearman correlation analysis was used to clarify the correlations of defoliation, nitrogen, and soil depth with CO 2 emissions, soil microbial activity parameters, Proteobacteria and Acidobacteria . Significance was defined at P < 0.05. Additionally, Origin18.0 and R programming software were utilized for mapping. Soil samples were obtained at three depths (0–10 cm [top soils], 10–30 cm [middle soils], and 30–60 cm [deep soils]) of the botanical garden test site at Tianshui Normal University (Tianshui, Gansu, China; 34°34′10″N and 105°41′47″E) in 2021. The test site was built in 2002, where the cherry rootstock was Gisela 5 and the cherry variety was Provence. The basic physical and chemical properties of top, middle, and deep soils were as follows: pH: 6.7, 7.8, and 8.1; total organic carbon: 14.74 g kg −1 , 12.54 g kg −1 , and 7.92 g kg −1 ; total nitrogen: 0.74 g kg −1 , 0.69 g kg −1 , and 0.46 g kg −1 ; available phosphorus: 4.93 mg kg −1 , 5.08 mg kg −1 , and 3.07 mg kg −1 ; available potassium: 153.0 mg kg −1 , 148.0 mg kg −1 , and 100.6 mg kg −1 . The incubation experiment was conducted as a complete factorial experiment of soils at three depths (top, middle, and deep soils) * two kinds of defoliation addition (no-defoliation [C 0 ] and defoliation [C A , 1%]) * addition of three levels of nitrogen (0 mg kg −1 [N 0 ], 90 mg kg −1 [N L ], and 135 mg kg −1 [N H ]) in three replicates with a fully randomized design. Total organic carbon was 451 g kg −1 and total nitrogen was 12.47 g kg −1 in defoliation. Incubation tanks (1 L) were respectively added with 100 g soils (dry weight) at different depths for 7 days of pre-incubation at 25 °C ( ). After that, soils were fully mixed with defoliation according to different treatments. Nitrogen and phosphate fertilizers were dissolved in distilled water and added to soils as a solution. Next, the soils were incubated at 25 °C with the moisture of 60% and the bulk density of 1.2 g cm −3 in the dark for 80 days. For each treatment, three tanks were taken out at 1, 4, 13, and 80 days. The soils in the incubation tanks were stored at −20 °C for detecting microbial biomass carbon (MBC), soil enzyme activities, and microbial communities. Specifically, MBC was measured with the chloroform-fumigation extraction method ( ). The activity of alkaline phosphatase was examined with the method of Tabatabai and Bremner ( ) as described in a prior study ( ). The activity of cellulase was detected using the method of Xu and Zheng ( ). The activity of catalase was determined by titrating 0.1 mol L −1 KMnO 4 ( ). Total DNA was extracted from soils with the Fast DNA SPIN Kit for Soil and FastPrep-24 nucleic acid Extraction instrument (MP Biomedicals, Santa Ana, CA, USA) on the 4th day. The DNA was examined with 1% agarose gel electrophoresis, and its concentration was measured with NanoDrop 2000. The DNA was stored in a refrigerator at −20 °C for subsequent use. The stock solution of the DNA was diluted to about 5 mg/L as the template of polymerase chain reaction (PCR) amplification. Afterward, the 16S rRNA variable regions V3-V4 of bacteria were subjected to PCR amplification with the following primers: CCTAYGGGRBGCASCAG (341F) and GGACTACNNGGGTATCTAAT (806R). The PCR was conducted with a system of 50 µL, including 5 µL of 10 * buffer, 4 µL of dNTP, 0.5 µL of RTAQ (Takara), 1 µL of 10 µmol L −1 each front and rear primers, 36.5 µL of ddH 2 O, and 2 µL of template DNA, and the detection of each sample was repeated three times. The amplification conditions were as follows: pre-denaturation at 98 °C for 30 s, 30 cycles of chain disassembly at 98 °C for 10 s, annealing at 55 °C for 15 s, and extension at 72 °C for 1 min, and final extension at 72 °C for 10 min. The products obtained by three repeats of DNA amplification were mixed and tested with 1% agarose gel electrophoresis. The concentration of the obtained bacterial PCR products was measured with the PicoGreen kit. After the products were evenly mixed, DNA was purified and recovered with the DNA purification kit (TIANGEN Biotech, Beijing, China). Bacterial 16S rDNA was sequenced by Shenzhen Weicomeng Technology Group Co., Ltd. (Shenzhen, China) with the Novaseq 6000 PE 250 platform. According to corresponding barcodes, the samples were subjected to paired reading, and then their barcode and primer sequence were removed. The reads at both ends were combined with the FLASH (V1.2.7) software. The QIIME data analysis package was utilized to remove low-quality raw sequences (length < 250 bp, ambiguous base “N”, and the average base quality score less than 20). The chimeric sequence was discarded with the MAARJAM database ( https://maarjam.ut.ee/ ). The valid readings were assigned to operational taxon units or virtual taxa with an identity threshold of 97%, and the representative sequences were identified with SILVA (V128, http://www.arb-silva.de ) and MAARJAM databases. CO 2 emissions (mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 ) were calculated as per the titration results of hydrochloric acids. [12pt]{minimal} }{}l@{}} C{O}_{2}= _{0}-V) {C}_{HCl}}{2} 12 1000. C O 2 = V 0 − V × C H C l 2 × 12 × 1 m 1 − a × 1000 . In this formula, V 0 was the volume of titration by standard hydrochloric acid during blank calibration (mL), while V represented the volume of titration by standard hydrochloric acid during samples (mL). C HCl was the concentration of standard hydrochloric acid (1 mol L −1 ), m was soil mass (g), and a was soil water content (%). The CO 2 efflux rate (mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) was calculated with the following formula: CO 2 efflux rate = CO 2 emission/t. In this formula, t represented the day when the NaOH solution was placed in the incubation tanks. The cumulative CO 2 emission (g CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 ) was the cumulative CO 2 emission from each treatment over a given incubation time. For a given incubation time (80 days), the priming index (PI) induced by exogenous substance addition was normalized to the proportion of added non-exogenous substances to cumulative CO 2 emissions based on the following formula: Priming index (PI) = (CO 2 add − CO 2 non-add )/ CO 2 non-add . In this formula, CO 2 add represented cumulative CO 2 emissions after 80 days of defoliation and nitrogen treatment, and CO 2 non-add indicated cumulative CO 2 emissions after 80 days of treatment without defoliation and nitrogen. The PI represented the intensity of the priming effect. The PI value of 1 suggested that the amount of organic carbon mineralization was not affected by exogenous substance addition. The PI value of >1 indicated that the addition of exogenous substances caused the priming effect of organic carbon mineralization, and the larger value was associated with a stronger priming effect. The PI value of <1 represented that the addition of exogenous substances reduced the mineralization of organic carbon and produced a negative priming effect, and the smaller value illustrated the stronger negative priming effect. All statistical analyses were performed with the SPSS 18.0 statistical and R programming software. The effects of the CO 2 efflux rate, cumulative CO 2 emission, PI, MBC, catalase, alkaline phosphatase, cellulase, Chao1, Shannon, and Simpson and the relative abundance of Proteobacteria and Acidobacteria were analyzed with three-way analysis of variance combined with Duncan’s multiple range test. Partial least squares discrimination analysis was conducted to analyze the distribution of soil microbial communities in soils at different depths. The Spearman correlation analysis was used to clarify the correlations of defoliation, nitrogen, and soil depth with CO 2 emissions, soil microbial activity parameters, Proteobacteria and Acidobacteria . Significance was defined at P < 0.05. Additionally, Origin18.0 and R programming software were utilized for mapping. CO 2 emissions The CO 2 efflux rate and cumulative CO 2 emissions were increased under defoliation addition, which was extremely significantly affected by the soil depth, defoliation, and nitrogen, as well as interaction among these three factors ( and , and ). The CO 2 efflux rate peaked on the 1st day and then declined with the incubation time of each treatment ( ). Among all treatments, the CO 2 efflux rate was higher in C A treatment (65.41 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) than in C 0 treatment (26.42 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) and lower in deep soils (36.1 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) than middle (51.56 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) and top (50.06 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) soils ( ). In addition, N 0 treatment (52.87 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) resulted in a higher CO 2 efflux rate than N L treatment (51.56 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) and N H (50.06 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) on the 1st day. After 80 days ( ), cumulative CO 2 emissions were averagely elevated by 2.62, 2.37, or 2.26 times under treatments of C A N 0 , C A N L , or C A N H when compared to under C 0 N 0 treatment. Cumulative CO 2 emissions were 30.28% and 28.21% higher in top and middle soils than in deep soils, respectively. Meanwhile, cumulative CO 2 emissions were markedly lower under N L and N H treatment than under N 0 treatment by 7.87% and 15.75%, respectively. PI After 80 days of incubation, defoliation addition alone or in combination with nitrogen addition resulted in positive PI in soils at the three depths ( ). PI was extremely substantially affected by defoliation, nitrogen, and soil depth ( ). PI was the highest in deep soils (2.05 [C A N 0 ], 1.83 [C A N L ], and 1.76 [C A N H ]) in defoliation and nitrogen additions. Moreover, nitrogen addition alone caused reverse PI in soils at three depths. MBC MBC was enhanced under defoliation and nitrogen addition, which was significantly or extremely significantly altered by defoliation, nitrogen and soil depth ( and ). Moreover, defoliation addition alone or combined with nitrogen addition contributed to higher MBC than C 0 N 0 treatment. On the 4th day, MBC in top soils was markedly higher under C A N 0 , C A N L , and C A N H treatment than under C 0 N 0 treatment by 66.49%, 87.82%, and 97.81%. In contrast, nitrogen addition alone diminished MBC in soils at the three depths. MBC in top and middle soils was higher than that in deep soils by 127.07% and 104.31%, respectively. Soil enzymes The activity of soil catalase, alkaline phosphatase, and cellulase was enhanced, which was markedly changed by defoliation, nitrogen and soil depth, on the 4th day ( and ). The activity of soil catalase was 32.95% and 36.94% higher, that of alkaline phosphatase was 84.64% and 83.03% higher, and that of cellulase was 174.67% and 251.12% higher under C A N L and C A N H treatment than under C 0 N 0 . Nitrogen addition alone had different influences on activity of the tested soil enzymes in soils at the three depths. The activity of catalase, alkaline phosphatase, and cellulase was higher in top soils than in deep soils. Soil bacterial diversity and community structure During the first 7 days, CO 2 emissions rapidly surged after exogenous carbon addition. Mounting studies ( ; ; ) reported that bacterial communities were changed from day 3 to day 15. Moreover, our study demonstrated that MBC peaked on the 4th day under defoliation and nitrogen addition. Accordingly, in order to evaluate the effect of defoliation, nitrogen, and soil depth on the soil microbial community structure in dryland cherry orchards, soil DNA was extracted on the 4th day of incubation to analyze the bacterial diversity. The results manifested that defoliation and nitrogen addition exerted no significant effects on the Chao1, Shannon, and Simpson index of soil bacteria ( ). However, the relative abundance of Proteobacteria markedly increased in soils at the three depths under defoliation and nitrogen addition, and that of Acidobacteria in deep soils was substantially reduced under both defoliation and nitrogen addition. In our research, 48 phyla, 133 classes, 209 orders, 260 families, and 480 genera of bacteria were obtained through high-throughput sequencing of bacterial 16S rRNA. At the level of bacteria phyla ( ), the relative abundance of Proteobacteria and Acidobacteria was greater than 10.0% in communities, which were 37.84%–60.62% and 10.70%–24.53%, respectively. The relative abundance of Proteobacteria was 15.91% higher under C A treatment than under C 0 treatment and was 12.18% and 11.85% higher under N H and N L treatment than under N 0 treatment in deep soils. The relative abundance of Acidobacteria in deep soils was lower under C A N H treatment than under C 0 N 0 treatment by 96.23%. 2 emissions The CO 2 efflux rate and cumulative CO 2 emissions were increased under defoliation addition, which was extremely significantly affected by the soil depth, defoliation, and nitrogen, as well as interaction among these three factors ( and , and ). The CO 2 efflux rate peaked on the 1st day and then declined with the incubation time of each treatment ( ). Among all treatments, the CO 2 efflux rate was higher in C A treatment (65.41 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) than in C 0 treatment (26.42 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) and lower in deep soils (36.1 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) than middle (51.56 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) and top (50.06 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) soils ( ). In addition, N 0 treatment (52.87 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) resulted in a higher CO 2 efflux rate than N L treatment (51.56 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) and N H (50.06 mg CO 2 [12pt]{minimal} }{}${}_{}^{-1}$ kg soil − 1 d −1 ) on the 1st day. After 80 days ( ), cumulative CO 2 emissions were averagely elevated by 2.62, 2.37, or 2.26 times under treatments of C A N 0 , C A N L , or C A N H when compared to under C 0 N 0 treatment. Cumulative CO 2 emissions were 30.28% and 28.21% higher in top and middle soils than in deep soils, respectively. Meanwhile, cumulative CO 2 emissions were markedly lower under N L and N H treatment than under N 0 treatment by 7.87% and 15.75%, respectively. After 80 days of incubation, defoliation addition alone or in combination with nitrogen addition resulted in positive PI in soils at the three depths ( ). PI was extremely substantially affected by defoliation, nitrogen, and soil depth ( ). PI was the highest in deep soils (2.05 [C A N 0 ], 1.83 [C A N L ], and 1.76 [C A N H ]) in defoliation and nitrogen additions. Moreover, nitrogen addition alone caused reverse PI in soils at three depths. MBC was enhanced under defoliation and nitrogen addition, which was significantly or extremely significantly altered by defoliation, nitrogen and soil depth ( and ). Moreover, defoliation addition alone or combined with nitrogen addition contributed to higher MBC than C 0 N 0 treatment. On the 4th day, MBC in top soils was markedly higher under C A N 0 , C A N L , and C A N H treatment than under C 0 N 0 treatment by 66.49%, 87.82%, and 97.81%. In contrast, nitrogen addition alone diminished MBC in soils at the three depths. MBC in top and middle soils was higher than that in deep soils by 127.07% and 104.31%, respectively. The activity of soil catalase, alkaline phosphatase, and cellulase was enhanced, which was markedly changed by defoliation, nitrogen and soil depth, on the 4th day ( and ). The activity of soil catalase was 32.95% and 36.94% higher, that of alkaline phosphatase was 84.64% and 83.03% higher, and that of cellulase was 174.67% and 251.12% higher under C A N L and C A N H treatment than under C 0 N 0 . Nitrogen addition alone had different influences on activity of the tested soil enzymes in soils at the three depths. The activity of catalase, alkaline phosphatase, and cellulase was higher in top soils than in deep soils. During the first 7 days, CO 2 emissions rapidly surged after exogenous carbon addition. Mounting studies ( ; ; ) reported that bacterial communities were changed from day 3 to day 15. Moreover, our study demonstrated that MBC peaked on the 4th day under defoliation and nitrogen addition. Accordingly, in order to evaluate the effect of defoliation, nitrogen, and soil depth on the soil microbial community structure in dryland cherry orchards, soil DNA was extracted on the 4th day of incubation to analyze the bacterial diversity. The results manifested that defoliation and nitrogen addition exerted no significant effects on the Chao1, Shannon, and Simpson index of soil bacteria ( ). However, the relative abundance of Proteobacteria markedly increased in soils at the three depths under defoliation and nitrogen addition, and that of Acidobacteria in deep soils was substantially reduced under both defoliation and nitrogen addition. In our research, 48 phyla, 133 classes, 209 orders, 260 families, and 480 genera of bacteria were obtained through high-throughput sequencing of bacterial 16S rRNA. At the level of bacteria phyla ( ), the relative abundance of Proteobacteria and Acidobacteria was greater than 10.0% in communities, which were 37.84%–60.62% and 10.70%–24.53%, respectively. The relative abundance of Proteobacteria was 15.91% higher under C A treatment than under C 0 treatment and was 12.18% and 11.85% higher under N H and N L treatment than under N 0 treatment in deep soils. The relative abundance of Acidobacteria in deep soils was lower under C A N H treatment than under C 0 N 0 treatment by 96.23%. Effects of defoliation and nitrogen addition on CO 2 emissions in soils at different depths As previously described ( ), defoliation alone or both defoliation and nitrogen was added in our study to assess organsim matter decomposition and analyze the mechanism of “microbial mining”. Our data unveiled that cumulative CO 2 emissions under defoliation addition alone or combined with nitrogen addition were 1.94−2.78 times higher than those under C 0 N 0 treatment, which may be attributable to the fact that defoliation and nitrogen addition increased the activity of soil microbes ( and ), accelerated the conversion of microbial biomass ( ), and facilitated the secretion of extracellular enzymes to decompose soil organsim matter and then deciduous residues through microbes ( ; ), thus resulting in CO 2 emissions in short-term incubation ( ; ; ). These results are consistent with hypothesis (1). Additionally, our findings also elucidated that compared to defoliation addition alone, addition of both defoliation and nitrogen decreased CO 2 emissions and PI, which is ascribed to the “mining of nitrogen” ( ). Moreover, nitrogen alone addition also reduced CO 2 emissions and caused negative PI, which is attributed to the “mining of nitrogen” ( ). Our data also revealed that in soils with relatively low total nitrogen contents (0.47−0.74 g kg −1 ), microbes were stimulated to decompose organsim matter to acquire the required nitrogen and induce CO 2 emissions under defoliation and nitrogen addition, similar to most research results ( ; ; ; ; ; ). However, our data showed no significant differences between low and high nitrogen, which indicated that the nitrogen demand of microbes could be saturated at low nitrogen levels. In addition, nitrogen addition alone reduced CO 2 emissions and then caused negative PI, which was concordant with the results of laboratory culture in the dark ( ; ) and outdoor testing ( ). Corresponding to the research with the DeNitrification-DeComposition (DNDC) model ( ), also used the DNDC model to predict the impact of 50% CO 2 emissions on nitrogen. Therefore, nitrogen addition alone exerts little effect on soil respiration, whilst combination of nitrogen and organic carbon effectively promotes the decomposition of exogenous organic substances. Soil depth is frequently associated with new increases in CO 2 emissions ( ; ). Throughout the incubation period, the CO 2 efflux rates and cumulative CO 2 emissions were higher in top and middle soils than in deep soils, which might be explained by the following factors: (i) high MBC and activities of soil catalase, alkaline phosphatase, and cellulase in top and middle soils ( and ), (ii) relatively sufficient nutrients in top and middle soils, and (iii) suitable pH and favorable soil structure and properties in top and middle soils ( ; ), which provided advantageous conditions for the decomposition of deciduous residues by microbes. Conversely, PI was higher in deep soils than in top and middle soils in our study, concurrent with the result of a prior study ( ). The proportion of residual organic carbon was calculated based on the input amount of organic carbon and emission amount of CO 2 in soils at different depths, which exhibited that the proportion of residual organic carbon was 88.88%, 88.05%, and 80.60% in top, middle, and deep soils, respectively. This result indicated that top and middle soils can retain a higher proportion of exogenous carbon than deep soils. Effects of defoliation and nitrogen addition on microbial activities and communities in different soils Microbes have been extensively recognized to play a key role in soil carbon mineralization ( ; ; ). MBC and enzyme activities are usually considered measurable proxies for microbial decomposition ( ; ) and can be optimized with the shortest incubation time to minimize microbial growth and enzyme production during the measurement ( ). Defoliation addition alone or combined with nitrogen addition elevated MBC, increased the activity of alkaline phosphatase, and cellulase, and caused positive PI in soils ( ). Defoliation shared significantly positive correlations with MBC and the activity of catalase, alkaline phosphatase, and cellulase in soils ( ). These data illustrated that microbes obtained available carbon and nitrogen from defoliation and nitrogen through various enzymes to meet their stoichiometric requirements ( ). In our study, nitrogen increased microbial biomass and the activity of the tested enzymes in defoliation and nitrogen additions ( and ) as a regulator in organic carbon mineralization, which was supported by many previous studies ( ; ; ). We also observed that at the initial stage of culture, addition of both defoliation and nitrogen (especially both defoliation and high nitrogen) markedly enhanced MBC and enzyme activities but diminished CO 2 emissions when compared with defoliation addition alone, further confirming the “microbial mining” mechanism ( ). Meanwhile, our results also unraveled that only nitrogen addition significantly decreased soil microbial biomass and changed the activity of the tested enzymes at the three depths, accompanied by reduced CO 2 emissions and negative PI. These findings illustrated that the effective combination of nitrogen and organic carbon could accelerate organism decomposition, improve soil quality and fertility, and provide guidance for production practice. In our study, Proteobacteria and Acidobacteria were dominant bacteria in soils regardless of soil depth or exogenous inputs ( ). Defoliation and nitrogen addition did not significantly change soil bacterial diversity and abundance ( ), indicating the ecological properties of Proteobacteria and Acidobacteria in soils. Defoliation addition alone or combined with nitrogen addition markedly elevated the relative abundance of Proteobacteria and reduced the relative abundance of Acidobacteria , thereby causing high CO 2 emissions. This result suggested the association of Proteobacteria and Acidobacteria with CO 2 emissions. As a copiotrophic group, Proteobacteria grows rapidly, and its abundance is increased by relying on more labile carbon sources under nutrient inputs ( ). With increasing stimulation, more Proteobacteria can participate in the synthesis of exoenzymes to mineralize substrates ( ; ), corresponding to the result that defoliation addition alone or combined with nitrogen addition substantially augmented MBC and enzyme activities, which was consistent with hypothesis (2). Our data revealed that CO 2 emissions were significantly positively correlated with Proteobacteria ( ). In addition, observed that microbial biomass increased in the first step of plant residue decomposition, which was caused by labile C rather than macromolecular compounds. Altogether, these data suggest Proteobacteria as a main participant in organic carbon mineralization in cherry orchards. However, Acidobacteria (an oligotrophic group) could be colonized on mineral surfaces under rich nutrient conditions ( ; ). Furthermore, our PLS-DA analysis ( ) demonstrated that bacterial communities were clearly separated in top and deep soils, which can be explained by the fact that nutrient composition determines the distribution of oligotrophic and hypertrophic bacteria ( ; ), consistent with hypothesis (3). Compared to top soils, deep soils had the highest relative abundance of Proteobacteria under both defoliation and nitrogen addition, which can be attributed to the highest PI in deep soils that is related to different microbial products ( ), soil pH ( ; ; ), and soil nutrient ( ). Therefore, soil microbial communities may be partially determined by soil properties. Furthermore, fungi have also been reported to directly participate in organic carbon mineralization. Accordingly, further research is warranted to determine how fungal communities respond to defoliation and nitrogen added in soils of cherry orchards in northwest China. 2 emissions in soils at different depths As previously described ( ), defoliation alone or both defoliation and nitrogen was added in our study to assess organsim matter decomposition and analyze the mechanism of “microbial mining”. Our data unveiled that cumulative CO 2 emissions under defoliation addition alone or combined with nitrogen addition were 1.94−2.78 times higher than those under C 0 N 0 treatment, which may be attributable to the fact that defoliation and nitrogen addition increased the activity of soil microbes ( and ), accelerated the conversion of microbial biomass ( ), and facilitated the secretion of extracellular enzymes to decompose soil organsim matter and then deciduous residues through microbes ( ; ), thus resulting in CO 2 emissions in short-term incubation ( ; ; ). These results are consistent with hypothesis (1). Additionally, our findings also elucidated that compared to defoliation addition alone, addition of both defoliation and nitrogen decreased CO 2 emissions and PI, which is ascribed to the “mining of nitrogen” ( ). Moreover, nitrogen alone addition also reduced CO 2 emissions and caused negative PI, which is attributed to the “mining of nitrogen” ( ). Our data also revealed that in soils with relatively low total nitrogen contents (0.47−0.74 g kg −1 ), microbes were stimulated to decompose organsim matter to acquire the required nitrogen and induce CO 2 emissions under defoliation and nitrogen addition, similar to most research results ( ; ; ; ; ; ). However, our data showed no significant differences between low and high nitrogen, which indicated that the nitrogen demand of microbes could be saturated at low nitrogen levels. In addition, nitrogen addition alone reduced CO 2 emissions and then caused negative PI, which was concordant with the results of laboratory culture in the dark ( ; ) and outdoor testing ( ). Corresponding to the research with the DeNitrification-DeComposition (DNDC) model ( ), also used the DNDC model to predict the impact of 50% CO 2 emissions on nitrogen. Therefore, nitrogen addition alone exerts little effect on soil respiration, whilst combination of nitrogen and organic carbon effectively promotes the decomposition of exogenous organic substances. Soil depth is frequently associated with new increases in CO 2 emissions ( ; ). Throughout the incubation period, the CO 2 efflux rates and cumulative CO 2 emissions were higher in top and middle soils than in deep soils, which might be explained by the following factors: (i) high MBC and activities of soil catalase, alkaline phosphatase, and cellulase in top and middle soils ( and ), (ii) relatively sufficient nutrients in top and middle soils, and (iii) suitable pH and favorable soil structure and properties in top and middle soils ( ; ), which provided advantageous conditions for the decomposition of deciduous residues by microbes. Conversely, PI was higher in deep soils than in top and middle soils in our study, concurrent with the result of a prior study ( ). The proportion of residual organic carbon was calculated based on the input amount of organic carbon and emission amount of CO 2 in soils at different depths, which exhibited that the proportion of residual organic carbon was 88.88%, 88.05%, and 80.60% in top, middle, and deep soils, respectively. This result indicated that top and middle soils can retain a higher proportion of exogenous carbon than deep soils. Microbes have been extensively recognized to play a key role in soil carbon mineralization ( ; ; ). MBC and enzyme activities are usually considered measurable proxies for microbial decomposition ( ; ) and can be optimized with the shortest incubation time to minimize microbial growth and enzyme production during the measurement ( ). Defoliation addition alone or combined with nitrogen addition elevated MBC, increased the activity of alkaline phosphatase, and cellulase, and caused positive PI in soils ( ). Defoliation shared significantly positive correlations with MBC and the activity of catalase, alkaline phosphatase, and cellulase in soils ( ). These data illustrated that microbes obtained available carbon and nitrogen from defoliation and nitrogen through various enzymes to meet their stoichiometric requirements ( ). In our study, nitrogen increased microbial biomass and the activity of the tested enzymes in defoliation and nitrogen additions ( and ) as a regulator in organic carbon mineralization, which was supported by many previous studies ( ; ; ). We also observed that at the initial stage of culture, addition of both defoliation and nitrogen (especially both defoliation and high nitrogen) markedly enhanced MBC and enzyme activities but diminished CO 2 emissions when compared with defoliation addition alone, further confirming the “microbial mining” mechanism ( ). Meanwhile, our results also unraveled that only nitrogen addition significantly decreased soil microbial biomass and changed the activity of the tested enzymes at the three depths, accompanied by reduced CO 2 emissions and negative PI. These findings illustrated that the effective combination of nitrogen and organic carbon could accelerate organism decomposition, improve soil quality and fertility, and provide guidance for production practice. In our study, Proteobacteria and Acidobacteria were dominant bacteria in soils regardless of soil depth or exogenous inputs ( ). Defoliation and nitrogen addition did not significantly change soil bacterial diversity and abundance ( ), indicating the ecological properties of Proteobacteria and Acidobacteria in soils. Defoliation addition alone or combined with nitrogen addition markedly elevated the relative abundance of Proteobacteria and reduced the relative abundance of Acidobacteria , thereby causing high CO 2 emissions. This result suggested the association of Proteobacteria and Acidobacteria with CO 2 emissions. As a copiotrophic group, Proteobacteria grows rapidly, and its abundance is increased by relying on more labile carbon sources under nutrient inputs ( ). With increasing stimulation, more Proteobacteria can participate in the synthesis of exoenzymes to mineralize substrates ( ; ), corresponding to the result that defoliation addition alone or combined with nitrogen addition substantially augmented MBC and enzyme activities, which was consistent with hypothesis (2). Our data revealed that CO 2 emissions were significantly positively correlated with Proteobacteria ( ). In addition, observed that microbial biomass increased in the first step of plant residue decomposition, which was caused by labile C rather than macromolecular compounds. Altogether, these data suggest Proteobacteria as a main participant in organic carbon mineralization in cherry orchards. However, Acidobacteria (an oligotrophic group) could be colonized on mineral surfaces under rich nutrient conditions ( ; ). Furthermore, our PLS-DA analysis ( ) demonstrated that bacterial communities were clearly separated in top and deep soils, which can be explained by the fact that nutrient composition determines the distribution of oligotrophic and hypertrophic bacteria ( ; ), consistent with hypothesis (3). Compared to top soils, deep soils had the highest relative abundance of Proteobacteria under both defoliation and nitrogen addition, which can be attributed to the highest PI in deep soils that is related to different microbial products ( ), soil pH ( ; ; ), and soil nutrient ( ). Therefore, soil microbial communities may be partially determined by soil properties. Furthermore, fungi have also been reported to directly participate in organic carbon mineralization. Accordingly, further research is warranted to determine how fungal communities respond to defoliation and nitrogen added in soils of cherry orchards in northwest China. In the early stages of the incubation period, defoliation and nitrogen addition increased MBC, the activity of catalase, alkaline phosphatase, and cellulase, and CO 2 emissions and resulted in positive PI in soils at the three depths of dryland cherry orchards. Nitrogen addition alone contributed to a significant reduction in MBC and CO 2 emissions and negative PI. Meanwhile, defoliation and nitrogen addition markedly elevated the relative abundance of Proteobacteria and reduced the relative abundance of Acidobacteria in soils at the three depths, particularly deep soils. Moreover, Proteobacteria was significantly correlated with defoliation, nitrogen, CO 2 emissions, and soil cellulase enzyme activities. In conclusion, this study unraveled the separate and interactive effects of defoliation and nitrogen addition on CO 2 emissions, soil microbial activities, and soil microbial communities composition at the soil depth scale. Likewise, this study demonstrated that both defoliation and nitrogen addition simulated exogenous carbon decomposition, affected soil microbial communities and improved soil microbial activities, ultimately enhancing soil quality in dryland cherry orchards. 10.7717/peerj.15276/supp-1 Supplemental Information 1 Data analysis Click here for additional data file. 10.7717/peerj.15276/supp-2 Supplemental Information 2 Raw data Click here for additional data file.
An eConsultant versus a hospital-based outpatient consultation for general (internal) medicine: a costing analysis	fe1f9303-f9fd-4012-be20-2dfc2439b1b0	10174616	Internal Medicine[mh]	Excessive wait times for specialist outpatient appointments are a significant problem facing health systems internationally and have increased substantially in the last decade . Lack of timely access to specialist care has been linked to inefficiencies in health care delivery, deterioration in health and dissatisfaction . In line with other countries, Australia’s ageing population, and the subsequent growth in rates of chronic, complex disease, are placing increasing demand pressures on the health system in a time of significant fiscal constraints. As a result, some patients wait longer than they should for a specialist outpatient appointment, with public hospital non-urgent cases being seen outside clinically recommended times and the number of long waits steadily increasing since 2017 . This has been exacerbated in recent years with COVID-19 placing further strain on the health system. In addition, approximately 28% of the Australian population live in rural and remote areas . These Australians face unique challenges due to their geographic location and often have poorer health outcomes than people living in metropolitan areas . People living in rural and remote areas have poorer access to health care services and accessing specialist services takes significant time and cost to attend outpatient appointments . Outpatient services are resource intense and there has been a call for realigning resources and improved processes in outpatient services . In addition, there is a recognition that patients with complex, chronic diseases require ongoing team management co-ordinated by the primary care physician/general practitioner (GP) who can provide continuity of care in partnership with other health providers . The eConsult model of care is an outpatient substitution approach which has been evaluated, and implemented extensively internationally . It provides an asynchronous, digital, clinician-to-clinician advice service, giving GPs remote access to specialist support for patient care within 3 business days . Findings from initial evaluations of Australian eConsultant services support international evidence of reduced wait times and improved access to specialist input, significant avoidance of face-to-face hospital outpatient visits and better integrated care. . Additionally, eConsultant offers workplace flexibility for the specialist, with the ability to deliver advice at the most suitable time and place. Our eConsultant service was piloted and implemented with a general (internal) medicine physician specialist, able to provide advice across all medical subspecialities bar dermatology . The specialist was employed by an urban tertiary hospital which provided hospital-based general medicine clinics. In 2019, the hospital complemented this with an eConsultant service to two Australian regions (Western Queensland, and Brisbane South), resulting in 87% of requests for advice to the eConsultant replacing a traditional outpatient referral. This paper compares the cost of delivery of a traditional hospital-based outpatient appointment with that delivered via an eConsultant service. It was hypothesised that eConsultant would be more efficient to deliver. A cost-minimisation analysis from the hospital service provider perspective was used to estimate the incremental cost per patient of an eConsult compared with the traditional referral pathway, with a year time horizon. All costs are reported in 2020/21 in Australian dollars and discounting of future costs was not applied, appropriate to the time horizon. The resources and costs to deliver the eConsultant service was compared with the cost of delivery of a traditional hospital-based general medicine outpatient appointment for the same clinical patient-population. Study setting The Mater Hospital South Brisbane (Mater) is a large urban hospital which provides seven half-day general medicine outpatient clinics per fortnight. In 2020–2021, the hospital implemented an eConsultant service with 15 general practices in Western Queensland and Brisbane South to investigate opportunities to safely substitute traditional face-to-face outpatient care. The general medicine outpatient clinics are funded by Queensland Health, and the eConsultant service is funded by the Queensland eConsultant Partnership Program (QePP) at the specialist’s current sessional appointment. The eConsultant works remote to the hospital. The traditional outpatient service GPs send a patient referral for a traditional outpatient appointment to the Mater Referral Management Centre (RMC) (Additional File ). Referrals are checked for mandatory requirements (i.e., patient details) and printed (the KPI for this step is 24 hours from receipt of referral by the RMC). The referral is then reviewed by a RMC nurse to ensure that (1) the minimum data set has been received, (2) the referral is mapped to the correct speciality, and (3) the service is available. The RMC nurse may be required to re-review the referral if more information is subsequently received. An administration staff member then updates the patient record or registers the patient, and the GP is sent a receipt of the referral and a request for more information if required. A registrar or Visiting Medical Officer then categorises the patient based on urgency (1,2 or 3) or declines the referral. In Queensland, there are three outpatient urgency categories with recommended timeframes for consultation of within 30 days of being added to the outpatient wait list for Category 1 (Urgent); within 90 days for Category 2 (Semi-urgent) and within 365 days for Category 3 patients (Non-urgent) . Once categorised, an administration officer updates the patient record with the category (the KPI for this step is 5 days from receipt of the referral by the RMC). The referral is then registered and waitlisted by the appointment management and call centre team. Face-to face, telehealth and telephone consults are conducted from the clinic. Wait time calculations starts when a referral is marked as delivered, that is, a referral is seen by the RMC and finishes once the patient is seen by the specialist. The specialist dictates a discharge letter post consultation which is typed and sent to the GP by an Administration Officer (AO) in the front desk team. The eConsultant service GPs enrolled in the eConsultant service have the option to send a Request for Advice (RFA) to the eConsultant for patients (category 1–3) who would normally be referred for a traditional outpatient appointment. GPs send a RFA, using a template, to the specialist (the eConsultant), with supporting information auto-populated from the patient’s record via the GP’s clinical information system. The RFA must include a specific question/s. The eConsultant replies within 3 business days with an answer to the problem; a request for further information; or a recommendation that the patient is referred for a traditional outpatient appointment (Fig. ). The eConsultant receives an email notification of a RFA and logs on to a web-based portal to access and reply to the RFA. This portal can be accessed from any location from a portable device or computer. An AO registers the patient (if not already registered) and records the RFA response in the patient chart. The GP receives a documented record of the eConsultant advice via secure messaging to the practice inbox. All treatment decisions are made in partnership with the patient, and on the understanding that there is the option for a usual care specialist referral. The GP has the option to send additional follow-up RFAs about the same patient. GPs use the same billing practice as they would for a regular consultation. Audit via an independent physician ensures RFA fall within appropriate categorisation for outpatient referral. Data sources A 12-month retrospective review of patient activity data for the period 2020–2021 was conducted. Costing data was collected from the outpatient service (14 clinics per month, excluding young adult and perioperative patients) and compared to data collected for the eConsultant service. Hospital-based general medicine outpatient clinic consultations included in person, telehealth provider and telephone delivered consultations. Telehealth provider consultations are via video and are face time appointments where the specialist is located at the hospital and the patient is located off site. Telephone appointments are conducted by the specialist with the patient over the phone. Peri-operative and young adult hospital-based outpatient clinic consultations were not included as these are not covered by the eConsultant. A nurse and an administration person are always assigned to a general medicine clinic (even if no other specialists are there at that time) and will be shared with other clinics running at the same time. Out-patient cost per attendance data were provided to the researchers by the hospital informatics team along with service data if the attendance was for a new or review episode, and if the attendance was face-to-face, telehealth or via telephone. Costs pertaining to the out-patient care having been attributed to each attendance using standardised managerial accounting practices by accredited hospital staff . Managerial practices automatically assign costs based on a 30 min presentation for new patients. However, an audit of the new consultations (n = 94) over this period identified the actual mean time for new face-to-face appointments was 41 (SD 14.7) minutes. As such, the cost of a face-to-face outpatient visit was scaled up using a factor of 1.36 (41/30). Costs included labour (administration and clinician including allowances), direct supplies and oncosts (Additional file ). The same cost for face-to-face outpatient attendances was used for both the traditional model as well as for those patients for whom a face-to-face attendance had been requested following an eConsultation. As digital infrastructure and secure messaging software are required by both the hospital-based outpatient and eConsultant services the costs were not included in the analysis. For the eConsultant service, the cost for new and review patients was based on the time for the administrative officer and specialist. For the administrative officer, data on time estimates was derived for both new and review patients separately, considering the longer time associated with first establishing a patient within the system. For the specialist, data on time estimates were sourced from the specialist’s logged records for each RFA response. Specialist time was derived separately for instances that subsequently resulted in a request for a face-to-face attendance and those that did not for both new and review patients. This reflects that new patients are likely to take longer than review patients and those that require face-to-face attendance may take less time in completing the initial eConsultation. The cost per staff minute were based on Mater award pay levels provided by the hospital. The administrative officer was based on a level MCA3 and assumes no shift or overtime penalties apply based on the hours of operation of the clinic. The specialist’s pay was based on a L24 M02-3 specialist and includes professional allowances. For both the administrative officer and specialist, pay levels for 2020–2021 were used and oncosts were estimated based on advice from the Mater Business and Finance of a 30% mark up on salaries. Although, general practitioners (n = 54) reported taking the same time to complete an RFA as doing a traditional referral (mean: 13.84 min, SD:8.66); this cost and the cost of follow up attendances with general practitioners was excluded as they are health care services provided in the community (and not included in a hospital health service provider perspective). Time estimates included eConsultant post consult administration and evaluation. As a hospital health service provider perspective was adopted, the cost for an outpatient attendance in which a patient does not attend was costed equal to that of one in which a patient attends. This reflects the opportunity cost of the resources (including clinician time) that has been allocated to that scheduled appointment which cannot otherwise be redirected. The probability of a patient not attending a scheduled face-to-face appointment was assumed to be 6% based on the measure for the overall outpatient clinic at the hospital. The cost of outpatient attendances subsequent to an eConsult service were assumed to be equal to that of a face-to-face outpatient attendance through the traditional referral pathway. Specifically, this is considered as: Cost of outpatient attendance subsequent to an eConsult = outpatient face-to-face attendance x ƛ where ƛ is a scale variable set to 1 in the basecase. The proportion of new patients, compared to review patients, who received an eConsult was 96.2%. The probability that an eConsult would result in a subsequent face-to-face outpatient attendance was estimated for both new and review patients based on the specialist’s logged data. For new patients, a subsequent face to face attendance was required in 15.2% of cases (19/125), and 14.3% for review patients (1/7). For the traditional model, the probability that an outpatient attendance was conducted face to face, via telehealth or telephone was estimated for both new and review patients. For new patients, face to face attendances were conducted in 77.9% of cases (106/136), and for those that did not require a face-to-face attendance, 6.7% (2/30) were conducted via telehealth and 93.3% (28/30) via telephone. For review patients, face to face attendances were conducted in 84.9% of cases (298/351), and for those that did not require a face-to-face attendance, 7.5% (4/53) were conducted via telehealth and 92.5% (49/53) via telephone. Analytical approach A decision analytic model was constructed in TreeAge Pro and graphically presented in Fig. . The expected cost per patient for both an eConsult and traditional outpatient appointment are estimated separately by multiplying the alternative specific conditional cumulative probabilities of each pathway and the cost of that pathway. Input parameters for the model are provided in Table . The primary outcome from the model is the incremental cost per patient for eConsult compared to the traditional outpatient referral pathway. Uncertainty was explored using one-way sensitivity analyses and characterised with probabilistic sensitivity analysis using 10,000 Monte Carlo simulations. For each Monte Carlo simulation an estimate for each of the models’ input parameters is drawn from a distribution. The distribution type and parameters are provided in Table . Beta distributions were used for transition probabilities, log normal distributions for count estimates (time), normal distribution for staff salary estimates and gamma distributions for outpatient clinic attendances. Distribution parameters were estimated based on (1) the data collected within the evaluation except for staff salary, for which the standard deviation was assumed to be 10% of the mean and the probability of a patient not attending a scheduled face to face attendance which assumed an alpha and beta of 6 and 94 (i.e., 6 / (6 + 94) = 6%) respectively and (2) an assumed normal distribution for the scale variable applied to the cost of an outpatient attendance subsequent to an eConsult with a standard deviation of 0.25. The Mater Hospital South Brisbane (Mater) is a large urban hospital which provides seven half-day general medicine outpatient clinics per fortnight. In 2020–2021, the hospital implemented an eConsultant service with 15 general practices in Western Queensland and Brisbane South to investigate opportunities to safely substitute traditional face-to-face outpatient care. The general medicine outpatient clinics are funded by Queensland Health, and the eConsultant service is funded by the Queensland eConsultant Partnership Program (QePP) at the specialist’s current sessional appointment. The eConsultant works remote to the hospital. GPs send a patient referral for a traditional outpatient appointment to the Mater Referral Management Centre (RMC) (Additional File ). Referrals are checked for mandatory requirements (i.e., patient details) and printed (the KPI for this step is 24 hours from receipt of referral by the RMC). The referral is then reviewed by a RMC nurse to ensure that (1) the minimum data set has been received, (2) the referral is mapped to the correct speciality, and (3) the service is available. The RMC nurse may be required to re-review the referral if more information is subsequently received. An administration staff member then updates the patient record or registers the patient, and the GP is sent a receipt of the referral and a request for more information if required. A registrar or Visiting Medical Officer then categorises the patient based on urgency (1,2 or 3) or declines the referral. In Queensland, there are three outpatient urgency categories with recommended timeframes for consultation of within 30 days of being added to the outpatient wait list for Category 1 (Urgent); within 90 days for Category 2 (Semi-urgent) and within 365 days for Category 3 patients (Non-urgent) . Once categorised, an administration officer updates the patient record with the category (the KPI for this step is 5 days from receipt of the referral by the RMC). The referral is then registered and waitlisted by the appointment management and call centre team. Face-to face, telehealth and telephone consults are conducted from the clinic. Wait time calculations starts when a referral is marked as delivered, that is, a referral is seen by the RMC and finishes once the patient is seen by the specialist. The specialist dictates a discharge letter post consultation which is typed and sent to the GP by an Administration Officer (AO) in the front desk team. GPs enrolled in the eConsultant service have the option to send a Request for Advice (RFA) to the eConsultant for patients (category 1–3) who would normally be referred for a traditional outpatient appointment. GPs send a RFA, using a template, to the specialist (the eConsultant), with supporting information auto-populated from the patient’s record via the GP’s clinical information system. The RFA must include a specific question/s. The eConsultant replies within 3 business days with an answer to the problem; a request for further information; or a recommendation that the patient is referred for a traditional outpatient appointment (Fig. ). The eConsultant receives an email notification of a RFA and logs on to a web-based portal to access and reply to the RFA. This portal can be accessed from any location from a portable device or computer. An AO registers the patient (if not already registered) and records the RFA response in the patient chart. The GP receives a documented record of the eConsultant advice via secure messaging to the practice inbox. All treatment decisions are made in partnership with the patient, and on the understanding that there is the option for a usual care specialist referral. The GP has the option to send additional follow-up RFAs about the same patient. GPs use the same billing practice as they would for a regular consultation. Audit via an independent physician ensures RFA fall within appropriate categorisation for outpatient referral. A 12-month retrospective review of patient activity data for the period 2020–2021 was conducted. Costing data was collected from the outpatient service (14 clinics per month, excluding young adult and perioperative patients) and compared to data collected for the eConsultant service. Hospital-based general medicine outpatient clinic consultations included in person, telehealth provider and telephone delivered consultations. Telehealth provider consultations are via video and are face time appointments where the specialist is located at the hospital and the patient is located off site. Telephone appointments are conducted by the specialist with the patient over the phone. Peri-operative and young adult hospital-based outpatient clinic consultations were not included as these are not covered by the eConsultant. A nurse and an administration person are always assigned to a general medicine clinic (even if no other specialists are there at that time) and will be shared with other clinics running at the same time. Out-patient cost per attendance data were provided to the researchers by the hospital informatics team along with service data if the attendance was for a new or review episode, and if the attendance was face-to-face, telehealth or via telephone. Costs pertaining to the out-patient care having been attributed to each attendance using standardised managerial accounting practices by accredited hospital staff . Managerial practices automatically assign costs based on a 30 min presentation for new patients. However, an audit of the new consultations (n = 94) over this period identified the actual mean time for new face-to-face appointments was 41 (SD 14.7) minutes. As such, the cost of a face-to-face outpatient visit was scaled up using a factor of 1.36 (41/30). Costs included labour (administration and clinician including allowances), direct supplies and oncosts (Additional file ). The same cost for face-to-face outpatient attendances was used for both the traditional model as well as for those patients for whom a face-to-face attendance had been requested following an eConsultation. As digital infrastructure and secure messaging software are required by both the hospital-based outpatient and eConsultant services the costs were not included in the analysis. For the eConsultant service, the cost for new and review patients was based on the time for the administrative officer and specialist. For the administrative officer, data on time estimates was derived for both new and review patients separately, considering the longer time associated with first establishing a patient within the system. For the specialist, data on time estimates were sourced from the specialist’s logged records for each RFA response. Specialist time was derived separately for instances that subsequently resulted in a request for a face-to-face attendance and those that did not for both new and review patients. This reflects that new patients are likely to take longer than review patients and those that require face-to-face attendance may take less time in completing the initial eConsultation. The cost per staff minute were based on Mater award pay levels provided by the hospital. The administrative officer was based on a level MCA3 and assumes no shift or overtime penalties apply based on the hours of operation of the clinic. The specialist’s pay was based on a L24 M02-3 specialist and includes professional allowances. For both the administrative officer and specialist, pay levels for 2020–2021 were used and oncosts were estimated based on advice from the Mater Business and Finance of a 30% mark up on salaries. Although, general practitioners (n = 54) reported taking the same time to complete an RFA as doing a traditional referral (mean: 13.84 min, SD:8.66); this cost and the cost of follow up attendances with general practitioners was excluded as they are health care services provided in the community (and not included in a hospital health service provider perspective). Time estimates included eConsultant post consult administration and evaluation. As a hospital health service provider perspective was adopted, the cost for an outpatient attendance in which a patient does not attend was costed equal to that of one in which a patient attends. This reflects the opportunity cost of the resources (including clinician time) that has been allocated to that scheduled appointment which cannot otherwise be redirected. The probability of a patient not attending a scheduled face-to-face appointment was assumed to be 6% based on the measure for the overall outpatient clinic at the hospital. The cost of outpatient attendances subsequent to an eConsult service were assumed to be equal to that of a face-to-face outpatient attendance through the traditional referral pathway. Specifically, this is considered as: Cost of outpatient attendance subsequent to an eConsult = outpatient face-to-face attendance x ƛ where ƛ is a scale variable set to 1 in the basecase. The proportion of new patients, compared to review patients, who received an eConsult was 96.2%. The probability that an eConsult would result in a subsequent face-to-face outpatient attendance was estimated for both new and review patients based on the specialist’s logged data. For new patients, a subsequent face to face attendance was required in 15.2% of cases (19/125), and 14.3% for review patients (1/7). For the traditional model, the probability that an outpatient attendance was conducted face to face, via telehealth or telephone was estimated for both new and review patients. For new patients, face to face attendances were conducted in 77.9% of cases (106/136), and for those that did not require a face-to-face attendance, 6.7% (2/30) were conducted via telehealth and 93.3% (28/30) via telephone. For review patients, face to face attendances were conducted in 84.9% of cases (298/351), and for those that did not require a face-to-face attendance, 7.5% (4/53) were conducted via telehealth and 92.5% (49/53) via telephone. A decision analytic model was constructed in TreeAge Pro and graphically presented in Fig. . The expected cost per patient for both an eConsult and traditional outpatient appointment are estimated separately by multiplying the alternative specific conditional cumulative probabilities of each pathway and the cost of that pathway. Input parameters for the model are provided in Table . The primary outcome from the model is the incremental cost per patient for eConsult compared to the traditional outpatient referral pathway. Uncertainty was explored using one-way sensitivity analyses and characterised with probabilistic sensitivity analysis using 10,000 Monte Carlo simulations. For each Monte Carlo simulation an estimate for each of the models’ input parameters is drawn from a distribution. The distribution type and parameters are provided in Table . Beta distributions were used for transition probabilities, log normal distributions for count estimates (time), normal distribution for staff salary estimates and gamma distributions for outpatient clinic attendances. Distribution parameters were estimated based on (1) the data collected within the evaluation except for staff salary, for which the standard deviation was assumed to be 10% of the mean and the probability of a patient not attending a scheduled face to face attendance which assumed an alpha and beta of 6 and 94 (i.e., 6 / (6 + 94) = 6%) respectively and (2) an assumed normal distribution for the scale variable applied to the cost of an outpatient attendance subsequent to an eConsult with a standard deviation of 0.25. The traditional outpatient referral pathway cost estimate was $587.20 compared to $226.13 for an eConsult, an efficiency of $361.07 per patient (a 61.5% efficiency). Results from the one-way sensitivity analyses are presented in Fig. . The incremental difference between eConsultant and traditional was most sensitive to the cost estimate of an outpatient attendance, the time for a specialist to complete an eConsult, and the probability of a patient requiring a face-to-face outpatient attendance following an eConsult. However, at the upper bounds of each of these estimates, an eConsult remained the most cost-efficient service. Increasing the cost of an outpatient attendance subsequent to an eConsult reduced the estimated difference between the eConsult and traditional referral pathways. However, the cost of a subsequent attendance would need to be 5.05 times the cost associated with an attendance via the traditional pathway for there to no longer be any difference in the expected costs between the two referral models. Results from the probabilistic sensitivity analysis are presented in Fig. . In 96.5% of the Monte Carlo simulations an eConsult was found to cost less than the traditional approach (Fig. ). Post COVID-19 health systems internationally have been energised to the opportunities of digital health to improve access and cost of care. In a period of extreme pressure on wages growth, inflation and demand management, this paper demonstrates the efficiencies to be made within the Australian health context via addition of the eConsultant service as one of these platforms. The approach preserves the desired outcome for patient and GP – specialist support for decision-making and ongoing management – but improves access, timeliness, and efficiency in clinician-to-clinician linkage. It allows maximal allocation of scarce resources to clinical rather than administrative and support costs and frees clinicians to communicate at times convenient to their work schedules. Previous work has identified the small percentage (13%) of eConsultant RFA which require subsequent face-to-face outpatient assessment. This evaluation suggests eConsultant is associated with a 61.5% efficiency to the hospital when compared with hospital-based outpatient appointments, with much of the saving related to non-clinical costs. The cost efficiency of the eConsultant service when compared to hospital-based outpatient consultations can be attributed to the cost savings due to low fixed costs and reduced clinician consulting time. Uncertainty in this analysis was principally due to the estimated cost of an outpatient attendance. The operational costs and labour for an outpatient clinic are substantial including infrastructure and information technology systems, electricity with its escalating price, and labour. Labour costs include a nurse who is allocated to a clinic regardless of the requirement for patient contact adding a substantial cost to the hospital-based models of care. This is the first costing analysis of the eConsultant service in Australia. Our findings support international research that the model is safe and less costly. Depending on the impact approach, these studies have found eConsult services provide cost savings to the healthcare system, a return of investment, and societal savings when compared with hospital-based services . Liddy’s group estimated total potential societal savings (included estimated direct costs to the payer and indirect costs to the patient) associated with their eConsult service to a remote Canadian community between August 2014 and April 2016 at $180,552.73 or $1,100.93 per eConsult . Additional potential cost savings include the prevention of deterioration in patient health due to vastly reduced wait times for specialist input and effective treatment options being provided sooner, as well as more effective future specialist consultations if needed . In addition to calculated efficiency from the hospital services perspective, eConsult services offer efficiency gains at the GP and specialist level. Previous advice options for general practitioners include phone calls to specialists, generally involving inefficient phone tag, an ad hoc advice option, and no formal documentation at the general practice or hospital service. In contrast, eConsultant is a documented advice service that operates asynchronously allowing busy clinicians to give and receive advice securely during clinical downtimes. Specialists can conduct eConsultant from home thus offering workplace flexibility and reduced travel time and costs for clinic attendance. Limitations Although international evidence, all be it limited has demonstrated that eConsults have the potential to improve health outcomes , this analysis is restricted in its comparison of hospital health service delivery costs. A full-economic evaluation, such as a cost-utlility analysis over an appropriate analytic time horizon that includes subsequent impact on health outcomes is required to address the value of eConsults. Overcoming previously identified challenges to evaluating impact on mortality and health related quality of life within a robust study design is critical . Whilst the analysis outlined has excluded digital infrastructure, software, and capital costs such as building depreciation associated with healthcare delivery models, future analyses that explicitly consider these costs in providing capacity is warranted. GP time at follow-up appointments after an eConsult was not included in the analysis as they fell within usual follow up for patients requiring care modification. Future research however could include a sensitivity analysis for these costs. Concerns regarding inappropriate outpatient categorisation have been addressed by annual audit of eConsultant referrals by an external general medicine specialist – all have met Category 1–3 criteria. The current service has delivered a 2-business day turnaround for the 15 participating practices. Scaling up without increasing wait time and preserving service integrity for hundreds of practices is the future challenge. Our partners in Ontario, have maintained this with over 100,000 consultations per year across 23 specialties . Further costing analysis will be conducted with the addition of specialties including dermatology and endocrinology. While this analysis looked at cost from a hospital system perspective, future research should include costs from a general practice and patient perspective. Additional savings to patients for an eConsult compared to face-to-face appointments can include avoidance of costs for travel, parking, and time off work for patients and their carers . Patients from rural and remote areas often have additional costs for overnight accommodation close to the hospital outpatient service. For some patients several consultations may be required following investigations if a workup is not completed prior to an initial face-to-face appointment. In 2020–21 the Queensland Government allocated $94.8 million to support rural Queenslanders to travel to their appointments and many patients still experience out of pocket expenses for travel for themselves and carers in addition to loss of income due to work leave. The eConsultant model of care meets calls for primary care system reform that improves access, and supports continuity of care and integration of primary and secondary care by embracing and building on utilisation of digital technologies post COVID-19 . By leveraging secure messaging this eConsultant service offers excellent cybersecurity. With estimates that healthcare contributes to 7% of Australia’s carbon emissions , eConsultant supports climate action goals by reducing fossil fuels required for unnecessary long-distance travel for face-to-face outpatient visits as well as hospital power consumption and cost. Although international evidence, all be it limited has demonstrated that eConsults have the potential to improve health outcomes , this analysis is restricted in its comparison of hospital health service delivery costs. A full-economic evaluation, such as a cost-utlility analysis over an appropriate analytic time horizon that includes subsequent impact on health outcomes is required to address the value of eConsults. Overcoming previously identified challenges to evaluating impact on mortality and health related quality of life within a robust study design is critical . Whilst the analysis outlined has excluded digital infrastructure, software, and capital costs such as building depreciation associated with healthcare delivery models, future analyses that explicitly consider these costs in providing capacity is warranted. GP time at follow-up appointments after an eConsult was not included in the analysis as they fell within usual follow up for patients requiring care modification. Future research however could include a sensitivity analysis for these costs. Concerns regarding inappropriate outpatient categorisation have been addressed by annual audit of eConsultant referrals by an external general medicine specialist – all have met Category 1–3 criteria. The current service has delivered a 2-business day turnaround for the 15 participating practices. Scaling up without increasing wait time and preserving service integrity for hundreds of practices is the future challenge. Our partners in Ontario, have maintained this with over 100,000 consultations per year across 23 specialties . Further costing analysis will be conducted with the addition of specialties including dermatology and endocrinology. While this analysis looked at cost from a hospital system perspective, future research should include costs from a general practice and patient perspective. Additional savings to patients for an eConsult compared to face-to-face appointments can include avoidance of costs for travel, parking, and time off work for patients and their carers . Patients from rural and remote areas often have additional costs for overnight accommodation close to the hospital outpatient service. For some patients several consultations may be required following investigations if a workup is not completed prior to an initial face-to-face appointment. In 2020–21 the Queensland Government allocated $94.8 million to support rural Queenslanders to travel to their appointments and many patients still experience out of pocket expenses for travel for themselves and carers in addition to loss of income due to work leave. The eConsultant model of care meets calls for primary care system reform that improves access, and supports continuity of care and integration of primary and secondary care by embracing and building on utilisation of digital technologies post COVID-19 . By leveraging secure messaging this eConsultant service offers excellent cybersecurity. With estimates that healthcare contributes to 7% of Australia’s carbon emissions , eConsultant supports climate action goals by reducing fossil fuels required for unnecessary long-distance travel for face-to-face outpatient visits as well as hospital power consumption and cost. This analysis of the delivery cost of Australia’s first eConsultant service compared with traditional hospital-based outpatient care suggests a 61.5% efficiency. This is achieved by reducing labour and infrastructure costs, thus allowing diversion of scarce resources to situations where face-to-face visits are essential. This is in addition to the documented benefits of reduced wait times for specialist input, reduced requirements for long distance travel, and the associated cost and loss of work time for patients and their carers. Workplace flexibility for participating specialists and reduced carbon footprint are additional gains. Further research should investigate the costing impact of scaleup to additional specialities with expanded general practice participation as the model is offered more widely across Queensland. Scale-up nationally of this service would be facilitated by an integrated national digital health infrastructure, and could contribute to a more equitable, accessible, and efficient Australian health system. Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2
Drug Shortages in Pediatrics in Europe: The Position of the European Pediatric Societies	eb69c90b-1efa-4e72-937a-fa057fe3b4df	10175075	Pediatrics[mh]	The authors declare no conflicts of interest.
Rigour and reproducibility in perinatal and paediatric epidemiologic research using big data	899c25a2-0839-4c27-89c8-73ad3f6b5b5b	10175126	Pediatrics[mh]	RIGOUR AND BIG DATA Scientific rigour requires careful application of the scientific method in the design, conduct, analysis, interpretation, and reporting of a study to ensure results are robust and minimally biased. Randomised controlled trials are thought to provide the highest-quality evidence but are difficult to conduct in perinatal research due to narrow eligibility criteria and heightened ethical concerns in pregnancy and infancy. In the absence of randomised trials, rigorous evidence can be obtained from observational studies if selection bias and confounders, both measured and unmeasured, are sufficiently accounted for. Observational studies using big data may be particularly subject to selection bias since data are typically generated for clinical or commercial purposes rather than for research. Marginalised communities may be under-represented in EHRs and claim datasets because they receive less frequent care due to structural barriers. Wearables, commercial genetic tests, and social media may be more commonly used in populations with higher income and education levels. The systematic exclusion of certain populations in epidemiologic research cannot be fixed by relying on big data, and biased sampling frames may be harder to detect in pre-existing data sources. Studies constructing cohorts from population registries, EHRs, or claims data are vulnerable to left truncation bias when excluding early pregnancy losses or delayed pregnancy detection, as is common in perinatal studies. If factors that inform if an individual is represented in data also affect the perinatal relationship being studied, the resulting estimates will undoubtedly be biased. Selection biases cannot be easily ameliorated with big data, which only underscores the importance of both classical and novel approaches to detecting and minimising bias. G-methods (e.g., inverse probability weighting and doubly robust estimation) and quasi-experimental designs (e.g., instrumental variables and Mendelian randomization) can reduce bias and mimic RCTs with big data observational epidemiology. However, the rigour of evidence produced with these designs may be highly sensitive to investigators’ assumptions. For example, in a study using Mendelian randomization, Diemer et al demonstrated how risk differences for prenatal alcohol exposure and attention deficit hyperactivity disorder are sensitive to choices in instrument selection and homogeneity assumptions. Employing causal inference methods with big data does not necessarily eliminate bias, and investigators must carefully assess causal assumptions and be transparent about potential violations. Big data may be susceptible to measurement error if they are not primarily collected for research purposes. Insurance claims and EHRs have become popular data sources for epidemiologic research, but the codes and fields they contain are not used consistently in clinical practice and may portend low sensitivity and specificity. A recent study comparing automated EHR data extraction versus manual chart abstraction for obstetrics research found that automatically extracted measurements had high reliability overall but lower accuracy for variables related to care processes (e.g., labor induction) or requiring provider interpretation (e.g., postpartum haemorrhage). Even sophisticated computational tools can be subject to classical forms of information bias. Validation studies and quantitative bias analyses are useful tools to assess the extent of, and potentially correct for, measurement error and misclassification bias in big data studies. Machine learning (ML) can be used to construct new variables from high-dimensional data and free-text fields, such as the use of clustering algorithms by Petersen et al, to learn placental features from tissue samples. However, opaque training processes can mask how information is extracted and make it difficult to assess biases in model outputs. Algorithmic bias, in which predictive performance differs across subgroups, can introduce differential misclassification if model accuracy is driven by factors that are informative to the research question at hand. Algorithmic fairness and mitigation methods are essential to minimising these biases in future big data perinatal and paediatric studies. While big data may include numerous variables that could be controlled for as potential confounders, care must be taken to avoid over-adjustment bias. For example, gestational age is commonly adjusted for as a confounder when it may in fact mediate a causal question. Similarly, whether it is appropriate to treat pregnancy history as a time-dependent confounder depends on the specific research question. Directed acyclic graphs are an essential tool for distinguishing confounders from mediators and minimising over-adjustment and collider biases. , REPRODUCIBILITY AND BIG DATA Across scientific disciplines, the “reproducibility crisis” has exposed major vulnerabilities in the practice of science that are likely to increase as big data is used more frequently. New guidelines for rigour, reproducibility, and transparency from the National Institutes of Health and other funding bodies are an important first step. Yet, assessing rigour and reproducibility during funding decisions is not enough; researchers must embrace rigour and reproducibility best practices in every stage of their research . Researchers – no matter how well-intentioned or conscientious – are prone to confirmation bias. For example, during data curation and analysis, analysts make countless seemingly innocuous judgments that could cumulatively tilt them towards observing the expected effect. In addition, selective reporting of results from multiple statistical models (i.e., “p-hacking”) can produce apparent effects that otherwise are nonexistent. Conducting data processing and analyses with a masked version of the exposure or treatment variable can reduce confirmation bias because initial estimates viewed while checking code do not reflect study findings. Documenting and registering statistical pre-analysis plans holds researchers accountable to the methods they planned before seeing data. Big data studies involve code in every step, from constructing a study population to producing statistical estimates. Adopting reproducible computational workflows can reduce the chance of errors and streamline replication efforts. Best practices for maintaining these workflows include the use of open-source statistical software, modular analysis code, bash scripts, version control, and decision logs. Statistical software, such as R and Python, can be used to manage large multi-step analyses and compare intermediate data objects during replication. Breaking an analysis into modular, self-contained scripts allows researchers to assess the results of each step and efficiently identify points of replication failure. Bash scripts can be used to easily rerun scripts in the appropriate order to replicate study findings. Version control tools, such as Github, keep a history of scripts and allow researchers to annotate changes in scripts over time. They can be used to pinpoint specific edits that cause changes in results and make coding processes transparent. Researchers can use decision logs to clearly note necessary deviations from the pre-analysis plan, which increases transparency in reporting. Ideally, a study’s computational workflow can be easily replicated from data extraction through table/figure generation, even by investigators not part of the original study team. Having multiple analysts internally replicate data processing and analysis scripts before publication can catch and resolve unintentional errors in code or identify decisions in data processing that are highly impactful in the results. Publishing data and code alongside a study manuscript allows for greater transparency and makes results findable, accessible, interoperable, and reusable (FAIR). They facilitate replication of study results by other investigators and ultimately may lead to more rapid scientific advances. However, many big data sources contain sensitive information that cannot be made publicly available. Synthetic data can be generated to allow for the publication of de-identified data that retain the statistical properties of the original, sensitive dataset. For example, Braddon et al generated synthetic data from sensitive EHRs using parametric and non-parametric methods and showed that the synthetic data could closely replicate estimates from the original data used. When generating synthetic data, context-dependent rules should be enforced to avoid unrealistic relationships between variables. With careful implementation, these developments in data sharing and security can help make perinatal epidemiologic research more transparent and reproducible. CONCLUSIONS Articles in this special “Big Data” issue of Paediatric and Perinatal Epidemiology showcase the promise of big data for perinatal and paediatric epidemiology. Yet, with big data comes big responsibility. Using novel causal inference or machine learning methods with big data does not guarantee that a study is unbiased, and foundational epidemiologic methods to maximise study rigour remain just as essential as ever. As studies using big data become increasingly computationally intensive, best practices in reproducibility must become standard tools in the epidemiology toolbox.
Predictive Potential of BCS and Pharmacokinetic Parameters on Study Outcome: Analysis of 198 In Vivo Bioequivalence Studies	ccc15824-b75f-44ca-8df7-29875b524139	10175306	Pharmacology[mh]	Conclusion of BE between a test product and a comparator product (from hereon reference product) is a critical step in the development of generic or innovative medicine, e.g., fixed-dose combination (FDC), new modified-release product, change of manufacturing site, etc. In vivo BE study is a simpler and more discriminatory surrogate for therapeutic equivalence clinical study and can evaluate equivalency of safety and efficacy profile between the generic and the reference product. BE study needs to be performed on a representative batch of the generic product, and hence is performed in late stages of the development. BE study represents a significant part of the development budget. On the other hand, affordability is inversely proportional to the product development budget and is one of the key benefits of generic medicines. The introduction of generic drugs saved the US health system nearly $1.5 trillion between 2004 and 2013 . Quality by design and related paradigms call for structured learning about product characteristics and assurance of continuous built-in quality of product throughout its lifecycle . Guidelines provided by the health authorities are an appreciated step towards standardized and optimal BE study design. Namely, protocol predefined parameters and criteria for conclusions of BE are important for making science-based decision about equivalency of generic and reference products and protecting patient’s interest and well-being. Pharmaceutical development is on the other hand interested also in the probability of BE study success. A good way to predict BE study outcome is to build in vitro–in vivo relationship . This can be done in later stages of the development, but extensive in vitro and in vivo data (BE studies) are needed and this requires a large amount of development time and budget. Probability of the success of a BE study needs to be evaluated as well in the early stages of product development where extensive data are not available. An early initial assessment of the risk related with the BE study outcome is needed for good planning and risk mitigation during product development. For such assessment, the impact of different factors on BE study outcome needs to be understood. Numerous interrelated factors impact the outcome of the BE study. Thus, it is very challenging to predict it. The BCS biowaiver approach is an example where one can, on the basis of certain API and product characteristics profiles, anticipate positive outcome of a BE study and can even completely waive a BE study in some situations. Impact of BCS on BE study outcome has been explored in real sets of data, published in conference abstracts and research articles. Two conference abstracts report work on large databases of BE studies (918 and 1200, respectively), but information available from these reports is very limited. On the other hand, two full-length research articles report the analyses of 124 and 500 BE studies, respectively, and focus mainly on the impact of BCS or Biopharmaceutical Drug Disposition Classification System (BDDCS), on acceptability of the BCS-based biowaiver approach and discriminative power of the in vitro methods. Comprehensive research for finding additional discriminatory features within each BCS Class that could help us additionally improve the risk assessment is limited. It includes research articles where the impact of area under the plasma concentration time curve to dose ratio (AUC/ D ) on BE outcome is explored on real-world datasets of BE studies with all BCS types and research articles where in silico (i.e., simulation) methodology is used to predict the impact of first-pass metabolism and intrinsic clearance variability on BE outcome and impact of T max and affinity for P-glycoprotein on BE study outcome for APIs belonging to BCS I and III . Comprehensive research on the real set of BE study data that would assess the impact of additional biopharmaceutical and pharmacokinetic parameters, such as absolute BA, affinity for P-glycoprotein, time at which maximum plasma concentration is achieved, elimination half-life ( t 1/2 ), volume of distribution, number of compartments in a compartmental model by which pharmacokinetics can be best described mathematically, in-silico permeability, plasma protein binding (PPB), etc., that would help identifying clusters of high-risk APIs within specific BCS class and that would set grounds for extended risk classification is missing. For this reason, a retrospective study was performed where a wide range of characteristics of BE study, API and product were collected, and their predictive potential of the study outcome was assessed. Database Preparation All data from in-house BE studies sponsored by Sandoz that satisfied the inclusion criteria were included in the analysis. Criteria for inclusion of the study into the database were: BE study date of completion was within the prespecified time period; Study was a pivotal BE study (i.e., pilot studies were not in the scope of this analysis, since the power of such studies is very likely insufficient); Study was completed (i.e., study phases were completed as per protocol and a report was issued); Results were not inconclusive (i.e., there was no clinical or bioanalytical deviation that would impact study results). Test product and reference product were both immediate-release products. Test product and reference product contained one API or were FDC products containing two or more APIs. In the latter case the study was treated as two or more independent BE studies for each API. BA of test and reference products were compared under the same condition (either fasting or fed). Information collected or calculated for each study is presented in Table . Database included BE studies under fasted or fed conditions. BE study was considered successful when the BE criteria defined in the study protocol were satisfied and the compared products were concluded to be BE. In case of FDC, products for each API success of BE study were evaluated separately. For example, it could happen that for API 1 of a fixed dose combination product BE was concluded and for API 2 BE was not concluded. Such product cannot be concluded as BE, but for the purpose of this analysis the outcome for the API 1 was still considered as BE. The term non-BE is used intentionally since failure to conclude BE does not lead to conclusion of bioinequivalence. In addition to the data on BE studies, we have collected biopharmaceutical and pharmacokinetic parameters of each API as summarized in Table . Data on API pharmacokinetics ( T max , peak plasma concentration ( C max ), area under the concentration time curve from zero extrapolated to infinity (AUCi), and t 1/2 were collected from the in-house studies. Parameters AUCi/ D and C max /AUCi were calculated from the dose administered, while C max and AUCi were used as observed in the study. Some parameters were used to create classes, e.g., variable T max class (short T max and long T max ) was created by splitting database through T max cut down at 1.5 h (90 min)—the upper estimated range of gastric emptying time for a capsule in fasted state . Other biopharmaceutical and pharmacokinetic characteristics, including absolute BA, presence of first-pass metabolism, apparent volume of distribution (Vd), plasma protein binding (PPB), number of pharmacokinetic compartments, substrate for P-gP transporter were collected from the literature. There were two conditions to classify drugs to have first-pass metabolism: the first was BA ≤ 80%, and the second was nonclinical or clinical literature evidence indicating first-pass metabolism. The 80% limit is arbitrary and comes from the fact that, in BE testing, ± 20% (on a normal scale) is considered a relevant BA difference. Literature claiming that API is a substrate for P-gP was the criterion for assigning API as a P-gP substrate. In the literature this claim was made on the basis of in vitro assessment, and in some cases the claim was supported by clinical data. GastroPlus v. 9.6 (Simulations Plus Inc., 42505 10th St W, Lancaster, CA 93534, USA) was used to calculate effective permeability ( P eff ) for each API using absorption, distribution, metabolism, excretion, and toxicity (ADMET) Predictor and .mol file of the API. Statistical Analysis For non-BE studies, post-hoc power was calculated considering number of subjects that completed the study, observed intrasubject variability for C max , BE criteria for C max parameter defined in the study protocol, study design, expected geometric mean ratio of C max of 95%, and type I error rate of 5%. Post-hoc power was calculated to determine whether the study failed due to inappropriate design (post-hoc power under 80%) or due to more than 5% difference in the BA of the products (post-hoc power above or equal to 80%). R version 3.5.3, RStudio version Version 1.1.456, and package PowerTOST version 1.4-9 were used to calculate post-hoc study power. Descriptive statistical analysis was performed to summarize various API parameters and BE study data within the highly and poorly soluble groups of APIs. For comparison purposes, descriptive statistics were calculated on a subset of poorly soluble APIs for the BE and non-BE group. Descriptive statistics were reported as mean and coefficient of variance (CV) for BA and permeability. For variables Vd, T max , PPB, t 1/2 , AUC/ D , C max /AUC, intra-individual coefficient of variation (intra-CV) for C max , and intra-CV for AUC, where the distribution departures from normal, descriptive statistics were reported as median and interquantile range (IQR) (Table ). To assess the association between various numerical features of poorly soluble APIs, nonparametric Spearman rank correlations tests were performed and Spearman coefficients (and associated P values) were used to assess the correlation. Correlation was assessed as weak, moderate, and strong for absolute values of Spearman coefficients below 0.4, between 0.4 and 0.7, and above 0.7, respectively. A set of univariate tests was performed to test how different variables are associated with the BE outcome. To test the association between categorical variables (Table ) and BE outcome, a chi-square test was applied. In case any cells in contingency table had five or fewer observations Fisher’s exact test was applied instead of chi-square test. Association between the numeric variables and BE outcome was assessed by analysis of variance (ANOVA) or nonparametric Kruskal–Wallis test in case of deviation from the normal distribution or outlying values. P values of < 0.1 and < 0.05 were set for conclusion of association and strong association, respectively. As this was an exploratory analysis, there was no correction for multiple comparisons (i.e., type I error rate was not controlled). To assess how well each significant parameter alone distinguishes between BE and non-BE outcome we have created receiver operating characteristic (ROC) curves and calculated area under the ROC curve (ROC AUC). BCS was the only variable tested for the association with the BE outcome on the complete set of data. Based on the information from the BCS analysis, all subsequent analyses (on categorical and numerical variables) were performed on subsets of poorly soluble APIs, which included all of the BCS II and IV APIs. Additional tests for the impact of first-pass metabolism and P-gP substrate were restricted to APIs with the absolute BA < 40%. All these analyses were repeated on subsets of studies under fasting and fed conditions to explore if the conclusions are similar when taking into account the impact of food. For the purpose of comparison with the results reported in the literature, some analyses were performed also on a subset of highly soluble APIs. These analyses included: descriptive statistics for parameters BA, first-pass metabolism, P-gP substrate, intrasubject CV, and T max . Tests used for specific parameters are presented in Table . Data were analyzed using Minitab 19.2020.3 (Minitab, Inc., 1829 Pine Hall Rd, State College, PA 16801, USA). All data from in-house BE studies sponsored by Sandoz that satisfied the inclusion criteria were included in the analysis. Criteria for inclusion of the study into the database were: BE study date of completion was within the prespecified time period; Study was a pivotal BE study (i.e., pilot studies were not in the scope of this analysis, since the power of such studies is very likely insufficient); Study was completed (i.e., study phases were completed as per protocol and a report was issued); Results were not inconclusive (i.e., there was no clinical or bioanalytical deviation that would impact study results). Test product and reference product were both immediate-release products. Test product and reference product contained one API or were FDC products containing two or more APIs. In the latter case the study was treated as two or more independent BE studies for each API. BA of test and reference products were compared under the same condition (either fasting or fed). Information collected or calculated for each study is presented in Table . Database included BE studies under fasted or fed conditions. BE study was considered successful when the BE criteria defined in the study protocol were satisfied and the compared products were concluded to be BE. In case of FDC, products for each API success of BE study were evaluated separately. For example, it could happen that for API 1 of a fixed dose combination product BE was concluded and for API 2 BE was not concluded. Such product cannot be concluded as BE, but for the purpose of this analysis the outcome for the API 1 was still considered as BE. The term non-BE is used intentionally since failure to conclude BE does not lead to conclusion of bioinequivalence. In addition to the data on BE studies, we have collected biopharmaceutical and pharmacokinetic parameters of each API as summarized in Table . Data on API pharmacokinetics ( T max , peak plasma concentration ( C max ), area under the concentration time curve from zero extrapolated to infinity (AUCi), and t 1/2 were collected from the in-house studies. Parameters AUCi/ D and C max /AUCi were calculated from the dose administered, while C max and AUCi were used as observed in the study. Some parameters were used to create classes, e.g., variable T max class (short T max and long T max ) was created by splitting database through T max cut down at 1.5 h (90 min)—the upper estimated range of gastric emptying time for a capsule in fasted state . Other biopharmaceutical and pharmacokinetic characteristics, including absolute BA, presence of first-pass metabolism, apparent volume of distribution (Vd), plasma protein binding (PPB), number of pharmacokinetic compartments, substrate for P-gP transporter were collected from the literature. There were two conditions to classify drugs to have first-pass metabolism: the first was BA ≤ 80%, and the second was nonclinical or clinical literature evidence indicating first-pass metabolism. The 80% limit is arbitrary and comes from the fact that, in BE testing, ± 20% (on a normal scale) is considered a relevant BA difference. Literature claiming that API is a substrate for P-gP was the criterion for assigning API as a P-gP substrate. In the literature this claim was made on the basis of in vitro assessment, and in some cases the claim was supported by clinical data. GastroPlus v. 9.6 (Simulations Plus Inc., 42505 10th St W, Lancaster, CA 93534, USA) was used to calculate effective permeability ( P eff ) for each API using absorption, distribution, metabolism, excretion, and toxicity (ADMET) Predictor and .mol file of the API. For non-BE studies, post-hoc power was calculated considering number of subjects that completed the study, observed intrasubject variability for C max , BE criteria for C max parameter defined in the study protocol, study design, expected geometric mean ratio of C max of 95%, and type I error rate of 5%. Post-hoc power was calculated to determine whether the study failed due to inappropriate design (post-hoc power under 80%) or due to more than 5% difference in the BA of the products (post-hoc power above or equal to 80%). R version 3.5.3, RStudio version Version 1.1.456, and package PowerTOST version 1.4-9 were used to calculate post-hoc study power. Descriptive statistical analysis was performed to summarize various API parameters and BE study data within the highly and poorly soluble groups of APIs. For comparison purposes, descriptive statistics were calculated on a subset of poorly soluble APIs for the BE and non-BE group. Descriptive statistics were reported as mean and coefficient of variance (CV) for BA and permeability. For variables Vd, T max , PPB, t 1/2 , AUC/ D , C max /AUC, intra-individual coefficient of variation (intra-CV) for C max , and intra-CV for AUC, where the distribution departures from normal, descriptive statistics were reported as median and interquantile range (IQR) (Table ). To assess the association between various numerical features of poorly soluble APIs, nonparametric Spearman rank correlations tests were performed and Spearman coefficients (and associated P values) were used to assess the correlation. Correlation was assessed as weak, moderate, and strong for absolute values of Spearman coefficients below 0.4, between 0.4 and 0.7, and above 0.7, respectively. A set of univariate tests was performed to test how different variables are associated with the BE outcome. To test the association between categorical variables (Table ) and BE outcome, a chi-square test was applied. In case any cells in contingency table had five or fewer observations Fisher’s exact test was applied instead of chi-square test. Association between the numeric variables and BE outcome was assessed by analysis of variance (ANOVA) or nonparametric Kruskal–Wallis test in case of deviation from the normal distribution or outlying values. P values of < 0.1 and < 0.05 were set for conclusion of association and strong association, respectively. As this was an exploratory analysis, there was no correction for multiple comparisons (i.e., type I error rate was not controlled). To assess how well each significant parameter alone distinguishes between BE and non-BE outcome we have created receiver operating characteristic (ROC) curves and calculated area under the ROC curve (ROC AUC). BCS was the only variable tested for the association with the BE outcome on the complete set of data. Based on the information from the BCS analysis, all subsequent analyses (on categorical and numerical variables) were performed on subsets of poorly soluble APIs, which included all of the BCS II and IV APIs. Additional tests for the impact of first-pass metabolism and P-gP substrate were restricted to APIs with the absolute BA < 40%. All these analyses were repeated on subsets of studies under fasting and fed conditions to explore if the conclusions are similar when taking into account the impact of food. For the purpose of comparison with the results reported in the literature, some analyses were performed also on a subset of highly soluble APIs. These analyses included: descriptive statistics for parameters BA, first-pass metabolism, P-gP substrate, intrasubject CV, and T max . Tests used for specific parameters are presented in Table . Data were analyzed using Minitab 19.2020.3 (Minitab, Inc., 1829 Pine Hall Rd, State College, PA 16801, USA). Full Dataset ( N = 273) The database consisted of 198 pivotal BE studies with immediate-release products containing 52 different APIs. There were no missing data for any of the observations. Among these 198 BE studies, 63 were conducted with FDC products containing two or three APIs. Each API was considered as a separate observation; thus, the database consisted of 273 observations. Before treating each API as a separate observation, FDCs were checked for pharmacokinetic drug–drug interactions. Among 17 unique API FDCs, 7 cases were found to have absence of pharmacokinetic interaction. In six cases pharmacokinetics of highly soluble API(s) in FDCs were slightly impacted by poorly soluble API (although the pharmacokinetic interaction was never clinically significant), whereas pharmacokinetics of poorly soluble API in FDC was not impacted by highly soluble API; thus, this did not impact our analysis. In four cases pharmacokinetic interactions were found also for poorly soluble APIs in FDC; however, all were reported as clinically insignificant. In two out of these four cases we confirmed these interactions do not impact our analysis by comparing pharmacokinetics when APIs were administered alone or in FDCs. Pharmacokinetic parameters in our BE studies could not be considered different. In two remaining cases only FDCs BE studies were available, so there was no confounding of single API or FDC influences. Thus, we concluded that our analysis is not impacted by pharmacokinetic drug–drug interactions. Of all 273 observations in the database, 34 were concluded as non-bioequivalent (non-BE) and 239 as BE. In all non-BE cases the results were out of limits for C max parameter (only in few of the cases also for AUC). Among non-BE studies, all except one had post-hoc study power above 80%, indicating non-BE results were not caused by insufficient study design. In 229 cases APIs were in the dosage form of tablets, in 18 cases oral suspensions, in 16 cases dispersible tablets, and in 10 cases hard-gelatin capsules. Apart from two non-BE cases with oral suspensions, other non-BE cases occurred when APIs were in the tablet formulation. Out of 52 different APIs, 26 were considered as highly soluble and 26 were considered as poorly soluble according to BCS classification. Hypothesis of existing association between a variable and BE outcome were accepted for BCS (at 5% significance) on the complete set of observations ( N = 273). ROC AUC for BCS was 0.78 (refer to supplemental data). There were no non-BE studies with BCS class I API ( N = 37), and there was only one for BCS class III API ( N = 91). BCS class II had the highest occurrence of non-BE results (25.7%), and BCS class IV was the second-riskiest class with 12.9% non-BE cases (Table and Fig. ). Highly Soluble APIs ( N =128) Highly soluble APIs had a wide range of BA, between 18% and 100%, wide range of permeability, and versatile absorption, distribution, and elimination characteristics (Table ). Regardless, only one BE study had non-BE results. When BA was above 85% all BE studies were successful. In our database, 41% of highly soluble APIs were subject to first-pass metabolism, including the API with the one non-BE result. Only 6% of cases were APIs with first-pass metabolism and high (> 30%) intrasubject variability of C max . However, non-BE results occurred for API with low intra-CV (< 30%). Of all highly soluble APIs, 17% were substrate of P-gP; however, the API with non-BE outcome was not a P-gP substrate. The study with the only non-BE result was conducted under fasted conditions. Poorly Soluble APIs ( N =145) Poorly soluble APIs had similarly versatile absorption, distribution, and elimination characteristics as highly soluble APIs, but as expected, even wider range of BA (4–100%) and volume of distribution (Table ). There were no strong associations between the numerical variables among poorly soluble APIs (refer to supplemental data). No association was found between BE outcome and APIs being part of or not part of an FDC (Table ). Strong association was shown between BE outcome and BA class, first-pass metabolism, and presence of P-gP efflux. For APIs with BA above 40%, absolute risk reduction for non-BE outcome was 21% compared with APIs with BA below 40%. For APIs without first-pass metabolism absolute risk reduction was 19% compared with APIs with first-pass metabolism. When API was not a P-gP efflux transporter substrate, absolute risk reduction for non-BE was 16% compared with studies with APIs that were substrates of P-gP (Table and Fig. ). Association with BE outcome was observed also for number of compartments in a pharmacokinetic model of API and T max class (Table and Fig. ). For APIs with short T max (≤ 1.5 h) absolute risk for non-BE outcome increased by 14% compared with APIs with longer T max (> 1.5 h) (Table ). A significant difference (at α = 5%) was observed for either mean or median difference between group of BE and non-BE study outcome for BA, T max , and permeability (Table ). When BA was above 85% all BE studies were successful. Average BA of APIs with non-BE study outcome was 15% lower (Table and Fig. A). Median T max of non-BE group was significantly lower than median of BE group (Table ). Similar trend, where T max was lower for non-BE group, was observed within fasting and fed BE studies (Fig. C). Average permeability of APIs with non-BE study outcome was 1 cm/s × 10 −4 higher (Table ). A similar trend, where permeability was higher for the non-BE group, was observed within low BA (<40%) and high BA (>40%) poorly soluble API classes (Fig. B). ROC AUC values for significant parameters mentioned in Sect. , were between 0.6 and 0.7 (please refer to supplemental data). On the other hand, no significant differences ( P > 0.05) between the groups with BE and non-BE study outcome was observed for parameters Vd, AUC/ D , C max /AUC, PPB, t 1/2 , intra-CV C max , and intra-CV AUC (Table ). None of the non-BE studies had PPB below 90% (Fig. D). Of all studies with poorly soluble APIs, 113 observations were attributable to BE studies under fasting conditions (with 24% non-BE occurrence) and 32 to BE studies under fed conditions (with 22% non-BE occurrence). The analysis on the subset of data in fasting conditions yielded the same conclusions as the analysis on the combined set of poorly soluble APIs (fasting and fed conditions) presented in Table . However, analysis on the subset of data with food revealed no significant difference between the BE and non-BE groups (Table ). This could be attributable to the lower discriminatory power of the parameters under fed conditions, or to the smaller sample size (power) of the fed subset. Poorly Soluble APIs with BA below 40% ( N = 100) Association between P-gP transport involvement and BE outcome was found for the subset of poorly soluble APIs with low BA (< 40%) (Table and Fig. ). No association between first-pass metabolism and BE outcome could be concluded on the subset of poorly soluble APIs with low BA (<40%). N = 273) The database consisted of 198 pivotal BE studies with immediate-release products containing 52 different APIs. There were no missing data for any of the observations. Among these 198 BE studies, 63 were conducted with FDC products containing two or three APIs. Each API was considered as a separate observation; thus, the database consisted of 273 observations. Before treating each API as a separate observation, FDCs were checked for pharmacokinetic drug–drug interactions. Among 17 unique API FDCs, 7 cases were found to have absence of pharmacokinetic interaction. In six cases pharmacokinetics of highly soluble API(s) in FDCs were slightly impacted by poorly soluble API (although the pharmacokinetic interaction was never clinically significant), whereas pharmacokinetics of poorly soluble API in FDC was not impacted by highly soluble API; thus, this did not impact our analysis. In four cases pharmacokinetic interactions were found also for poorly soluble APIs in FDC; however, all were reported as clinically insignificant. In two out of these four cases we confirmed these interactions do not impact our analysis by comparing pharmacokinetics when APIs were administered alone or in FDCs. Pharmacokinetic parameters in our BE studies could not be considered different. In two remaining cases only FDCs BE studies were available, so there was no confounding of single API or FDC influences. Thus, we concluded that our analysis is not impacted by pharmacokinetic drug–drug interactions. Of all 273 observations in the database, 34 were concluded as non-bioequivalent (non-BE) and 239 as BE. In all non-BE cases the results were out of limits for C max parameter (only in few of the cases also for AUC). Among non-BE studies, all except one had post-hoc study power above 80%, indicating non-BE results were not caused by insufficient study design. In 229 cases APIs were in the dosage form of tablets, in 18 cases oral suspensions, in 16 cases dispersible tablets, and in 10 cases hard-gelatin capsules. Apart from two non-BE cases with oral suspensions, other non-BE cases occurred when APIs were in the tablet formulation. Out of 52 different APIs, 26 were considered as highly soluble and 26 were considered as poorly soluble according to BCS classification. Hypothesis of existing association between a variable and BE outcome were accepted for BCS (at 5% significance) on the complete set of observations ( N = 273). ROC AUC for BCS was 0.78 (refer to supplemental data). There were no non-BE studies with BCS class I API ( N = 37), and there was only one for BCS class III API ( N = 91). BCS class II had the highest occurrence of non-BE results (25.7%), and BCS class IV was the second-riskiest class with 12.9% non-BE cases (Table and Fig. ). N =128) Highly soluble APIs had a wide range of BA, between 18% and 100%, wide range of permeability, and versatile absorption, distribution, and elimination characteristics (Table ). Regardless, only one BE study had non-BE results. When BA was above 85% all BE studies were successful. In our database, 41% of highly soluble APIs were subject to first-pass metabolism, including the API with the one non-BE result. Only 6% of cases were APIs with first-pass metabolism and high (> 30%) intrasubject variability of C max . However, non-BE results occurred for API with low intra-CV (< 30%). Of all highly soluble APIs, 17% were substrate of P-gP; however, the API with non-BE outcome was not a P-gP substrate. The study with the only non-BE result was conducted under fasted conditions. N =145) Poorly soluble APIs had similarly versatile absorption, distribution, and elimination characteristics as highly soluble APIs, but as expected, even wider range of BA (4–100%) and volume of distribution (Table ). There were no strong associations between the numerical variables among poorly soluble APIs (refer to supplemental data). No association was found between BE outcome and APIs being part of or not part of an FDC (Table ). Strong association was shown between BE outcome and BA class, first-pass metabolism, and presence of P-gP efflux. For APIs with BA above 40%, absolute risk reduction for non-BE outcome was 21% compared with APIs with BA below 40%. For APIs without first-pass metabolism absolute risk reduction was 19% compared with APIs with first-pass metabolism. When API was not a P-gP efflux transporter substrate, absolute risk reduction for non-BE was 16% compared with studies with APIs that were substrates of P-gP (Table and Fig. ). Association with BE outcome was observed also for number of compartments in a pharmacokinetic model of API and T max class (Table and Fig. ). For APIs with short T max (≤ 1.5 h) absolute risk for non-BE outcome increased by 14% compared with APIs with longer T max (> 1.5 h) (Table ). A significant difference (at α = 5%) was observed for either mean or median difference between group of BE and non-BE study outcome for BA, T max , and permeability (Table ). When BA was above 85% all BE studies were successful. Average BA of APIs with non-BE study outcome was 15% lower (Table and Fig. A). Median T max of non-BE group was significantly lower than median of BE group (Table ). Similar trend, where T max was lower for non-BE group, was observed within fasting and fed BE studies (Fig. C). Average permeability of APIs with non-BE study outcome was 1 cm/s × 10 −4 higher (Table ). A similar trend, where permeability was higher for the non-BE group, was observed within low BA (<40%) and high BA (>40%) poorly soluble API classes (Fig. B). ROC AUC values for significant parameters mentioned in Sect. , were between 0.6 and 0.7 (please refer to supplemental data). On the other hand, no significant differences ( P > 0.05) between the groups with BE and non-BE study outcome was observed for parameters Vd, AUC/ D , C max /AUC, PPB, t 1/2 , intra-CV C max , and intra-CV AUC (Table ). None of the non-BE studies had PPB below 90% (Fig. D). Of all studies with poorly soluble APIs, 113 observations were attributable to BE studies under fasting conditions (with 24% non-BE occurrence) and 32 to BE studies under fed conditions (with 22% non-BE occurrence). The analysis on the subset of data in fasting conditions yielded the same conclusions as the analysis on the combined set of poorly soluble APIs (fasting and fed conditions) presented in Table . However, analysis on the subset of data with food revealed no significant difference between the BE and non-BE groups (Table ). This could be attributable to the lower discriminatory power of the parameters under fed conditions, or to the smaller sample size (power) of the fed subset. N = 100) Association between P-gP transport involvement and BE outcome was found for the subset of poorly soluble APIs with low BA (< 40%) (Table and Fig. ). No association between first-pass metabolism and BE outcome could be concluded on the subset of poorly soluble APIs with low BA (<40%). Biopharmaceutics Classification System Significantly different percentages of non-BE studies were found across different BCS classes in our database (Table and Fig. ), indicating high association between BCS and BE study outcome. This is in line with a number of publications that supported in vivo predictive nature of BCS . Failure rate within the group of highly soluble APIs was negligible (~ 1%) and even lower than that found in the literature (10–16%), regardless of the wide range of BA (18–100%), presence of first-pass metabolism, and/or P-gP mediated efflux attributable to some highly soluble APIs. Low occurrence rate of non-BE results in our database supports BCS biowaiver approach for class I and III APIs implemented by numerous health authorities. In theory, BCS class I and III drugs were presented as less risky for non-BE outcome compared with classes with poorly soluble APIs, while the publications often reported similarly low failure rate for classes I, III, and IV, ranging from 10% to 16%. Similarity of BCS class IV APIs to BCS class I and III was usually attributed to smaller sample size of BCS IV group (i.e., insufficient power to detect differences) . Our estimated failure rate of 12.5% within BCS class IV seemed comparable to that reported in the literature; however, in our case the difference between highly soluble group of BCS and BCS class IV was obvious due to negligible failure rate in the highly soluble BCS classes. BCS class II was previously reported as the most critical for conclusion of BE, where failure rate ranged from 28% to 39% . When considering only highly variable BCS class II APIs, Lamouche reported high, flip-of-a-coin-like 54% failure rate of BE studies . In accordance with the literature data, our estimates showed that BCS class II has the highest occurrence of non-BE results (Table and Fig. ). Since the BCS classification can be done after assessment of solubility and permeability, and the assessment of permeability is sometimes challenging in the early stages of development, we decided to combine BCS class II and IV APIs into one poorly soluble group for further subanalysis. In addition, the classification available in the literature may sometimes be misleading due to the fact that the permeability assessment for BCS based biowaiver approach is many times surrogated by in vivo BA assessments. Thus, there are sometimes inconsistencies in BCS classification between BCS II and IV classes; for example, BCS II APIs with low BA but high permeability may be misclassified as BCS IV APIs. Combining both poorly soluble BCS classes we avoided any assumptions regarding permeability while assessing factors influencing BE study outcome. Bioavailability When BA was above 85%, either for BCS class I or II, all BE studies ( N = 33) were successful. Studies are usually powered at 80% or 90%, so the probability of study success, if product is indeed BE, is 80% or 90%, respectively. That further means that the scenario of observing zero non-BE studies among 33 is 0.1% (0.8^33) or 3.1% (0.9^33), which labels observation of zero non-BE studies in 33 cases, as unlikely. BCS biowaiver guidelines define BA of 85% as the limit for fraction of absorption which defines highly permeable APIs. Above this limit, less strict dissolution criteria and formulation differences requirements are set for the product being eligible for BCS biowaiver . This suggests that APIs with high extent of absorption are less risky for BE testing. This was also observed in our analysis. It seems that high permeability along with absence of presystemic API degradation/extraction processes decreases risks for BE study failure. On the other hand, regardless of wide BA range (18–100%) within highly soluble APIs, there was only one case of non-BE result. This supports acceptability of the BCS biowaiver approach for APIs with high (BCS class I) or low permeability (BCS class III). Bioavailability within the Group of Poorly Soluble APIs APIs with a wide range of BA were included in analyses. BA was a significant feature in all of the relevant analyses, i.e., lower BA (especially below 40%) was one of the key indicators for problems with BE. Low solubility might be the reason for low BA; however, it can also be caused by the low permeability, gastrointestinal instability, first-pass metabolism, and/or P-gP-mediated efflux. Some of these factors and their association with non-BE results are discussed in the Sects. , , and . Permeability within the Group of Poorly Soluble APIs As a general rule, higher permeability of poorly soluble APIs was more risky for conclusion of BE (Table and Fig. B). Namely, for APIs with high permeability, dissolution rate or solubility becomes a limiting factor for absorption, and these conditions are more challenging in terms of BE outcome where formulation performance is compared. These results correspond well with the BCS class II APIs being the most challenging group of APIs. The fact that this permeability is determined in silico by GastroPlus is at the same time an advantage, as it is easily determined early in the development, and also disadvantage, as the in silico estimate may be associated with less precision. First-Pass Metabolism First-pass metabolism can significantly affect BA and is as such of particular interest when assessing risk for concluding BE. Association between first-pass metabolism, present at 40% of highly soluble APIs, and BE outcome could not be explored within highly soluble APIs, since there was only one non-BE study. Most, i.e., 94%, of these APIs with first-pass metabolism had low variability of pharmacokinetics. The low occurrence of non-BE results is in agreement with prediction of Fernández and coworkers that highly soluble APIs with first-pass metabolism and low variability held less risk for concluding BE . On the other hand, our analyses infer that first-pass metabolism is associated with higher incidence of non-BE results in studies with poorly soluble APIs. These results are in line with the work of Cristofoletti and coworkers, where drug disposition and metabolism based classification (BDDCS) held similar predictive value for BE outcome as BCS . For the subset of APIs where BA was below 40%, presence of first-pass metabolism did not increase the absolute risk for non-BE. However, we have to take the latter conclusion with caution since there were only six observations in the group of APIs without first-pass metabolism. First-pass metabolism may be influenced by excipients , which can differ between the generic and reference product and thus impact BE outcome, and by different physiological or pathophysiological conditions, which may increase the variability of exposure. However, the impact of the latter can be excluded in our analysis since all BE studies in our database were performed with healthy volunteers. Impact of P-gP Efflux Another process that may impact BA is efflux of API from enterocytes. There are numerous transporters that perform this task. However, most often this process is governed by P-gP (MDR-1) transporter. H. Kortejärvi and coworkers have shown with simulations that, if a highly soluble API is a substrate for P-gP, this may significantly influence the outcome of BE studies . This risk cannot be observed in our database, since all studies with highly soluble APIs that were substrates of P-gP concluded BE. On the other hand, our analysis revealed that for poorly soluble APIs involvement of P-gP efflux transporter in pharmacokinetics of API seemed to significantly increase occurrence of non-BE results (Table ). Our analysis also suggests that the risk for non-BE outcome may be higher if poorly soluble API has low BA (<40%) and is a substrate for P-gP efflux. These results were not surprising, since P-gP efflux may also be influenced by excipients, which can differ between the generic and reference product and thus impact BE outcome. In addition, P-gP may impact variability of drug exposure . Time to Peak Concentration If rapid release is claimed to be clinically relevant and important for onset of action or is related to adverse events, then there should be no apparent differences in median T max and its variability between the test and reference product . In such case, T max would be a significant parameter determining the BE outcome; however, this was not the case in any of our BE trials, where primary parameters were always C max and AUC. Since T max is a composite parameter of elimination and absorption rate, the latter can be impacted by formulation (differences); thus, T max parameter could indeed hold relevant information regarding the study outcome. Furthermore, our hypothesis was that if T max is very short, gastric emptying is not playing a role in limiting the absorption rate and the absorption is rather limited by the release of API from the formulation. This was confirmed by our analysis of poorly soluble APIs where median T max of the non-BE group was 0.9 h lower than the median of the BE group (Table ). The trend was similar when the analysis was split for fasting and fed conditions (Fig. C). Similarly, the association between T max class (below or above 1.5 h) and BE outcome was observed (Table ). Splitting the dataset to fasting and fed conditions resulted in a similar conclusion under fasting conditions, but loss of association under fed conditions (even when a cutoff value higher than 1.5 h was considered). This is not surprising since under fed conditions gastric emptying limits the absorption rate. The T max could lose its predictive value for this reason. The authors acknowledge that the cutoff at 1.5 h might also be too strict to capture all cases with the fast absorption (not limited by gastric emptying), since the slow distribution/elimination can manifest in prolonged T max values. On the other hand, 1.5 h cutoff might be just what we are looking for in BE risk assessment, since the risk for non-BE is particularly high for APIs with fast absorption and fast distribution/elimination, resulting in narrow peaks in plasma concentration profile. See also Sect. where we discuss the association between number of compartments to describe distribution/elimination and BE outcome . Finally, Kortejärvi and coworkers have used simulations to show that the risk for not concluding BE is increased for highly soluble and highly permeable APIs with very short T max . However, the higher risk for non-BE results predicted by simulations has not been confirmed in our analysis of highly soluble APIs. C max /AUC within the Group of Poorly Soluble APIs Ln( C max /AUCi) of the non-BE group was lower than that of the BE group, but statistical significance could not be shown (Table ). C max /AUC was previously recommended as a less polluted measure of absorption rate as C max , but was later shown to have similar flaws as C max in lacking sensitivity in indicating changes in absorption rate constant . This might be why C max /AUC was not discriminatory in detecting non-BE results. Variability of Drug Exposure Variability of drug exposure after oral administration poses a significant challenge in modern drug development. Factors impacting variability are diverse and include human physiology variation, bioanalytical variation, and formulation technology variation . It is generally recognized that, with oral administration, distribution and elimination of API cannot be influenced by differences in formulations. For this reason, a crossover design is usually implemented for testing of BE, i.e., testing whether there are any differences in formulation performance. In such setting, “quazi” intra-individual variability (intra-CV of C max and AUC) is the variability of interest. It eliminates variability that may arise from differences related to distribution and excretion of drug between different subjects. Regardless, high intra-CV could still be one of the reasons for decreased power and was more problematic in the past when tools for handling of high intra-CV were not available or accepted by regulatory agencies. These problems were manifested in high occurrence of non-BE results, e.g., 54% of non-BE studies with BCS class II APIs with variability higher than 30% (i.e., high variability) reported by Lamouche . Examples of tools that tackle high variability are higher-order crossover design accompanied by scaling or widening of BE limits . All studies in our database of poorly soluble APIs were crossover studies (2 × 2 or higher order). Intra-CV of C max and AUC was not significantly different between non-BE and BE group in our database (Table ). This is not surprising considering the majority of non-BE studies had post-hoc power above 80%, meaning intra-CV was adequately considered in study design. It is noteworthy that higher average intra-CV of C max and AUC was observed for APIs with BA below 40% compared with APIs with BA above 40%, which suggests that processes decreasing BA increase variability of exposure. Regardless of the crossover nature of BE study, distribution and excretion properties and inter-occasion differences within one subject can still impact intra-CV and/or create more or less discriminatory environment for testing of BE. For this reason, differences between non-BE and BE group were explored for parameters that describe distribution and elimination of drug: Vd, AUC/ D (inverse of apparent clearance), PPB, t 1/2 , and number of compartments in pharmacokinetic models that best describe plasma profiles. Volume of Distribution within the Group of Poorly Soluble APIs The ln(Vd) between the BE and non-BE groups was not significantly different (Table ). Vd did not seem to correlate with intra-CV of AUC, but a moderate correlation between Vd and intra-CV of C max was found (refer to supplemental data). Apart from the impact on variability we could not conclude anything about predictable value of Vd for non-BE outcome. Inverse of Apparent Clearance (AUC/ D ) within the Group of Poorly Soluble APIs Sakuma and coworkers suggested a correlation between AUC/ D (fraction of absorption/clearance) and number of subjects in a BE study with highly soluble APIs . Yamashita and colleagues explored the correlation of AUC/ D ratio and parameters that impact BE study success. AUC/D correlated with the width of the 90% confidence interval (i.e., with variability) in BCS classes I and III . Our analysis showed that the mean ln(AUCi/ D ) was not significantly different between the BE and non-BE groups (Table ), although the mean ln(AUCi/D) of non-BE group was 0.5 h/L lower. This trend agrees with observations of Yamashita and coworkers where lower AUC/D implied higher chance of non-BE results, i.e., APIs with fast clearance, low permeability, high first-pass metabolism, and low GIT stability, were those that held the highest risk . In addition, we have also found a correlation between ln(AUCi/ D ) and intra-CV for C max . In contrast to the work by Yamashita and coworkers, the correlation was found not only for the highly soluble but also for the poorly soluble APIs (please refer to supplemental data). We could not conclude anything on the association of non-BE outcome and AUCi/D, but if the association exists, it may be confounded with the intra-CV. Plasma Protein Binding within the Group of Poorly Soluble APIs Protein binding may serve as a reservoir from which the API is slowly released as the unbound form and can prolong t 1/2 of the API. When an API is highly bound to plasma proteins, it typically has lower volume of distribution . As such, PPB may impact BA and could hypothetically create more or less discriminatory environment for testing of BE. Our analysis showed that PPB medians of BE and non-BE groups were not significantly different (Table ). Further visual analysis showed a group of poorly soluble APIs ( N = 11) with PPB below 90% where there are no non-BE results (Fig. D). Considering the study power of 90%, there was still a high 31% chance of not observing any non-BE among 11 studies (0.9^11). These APIs belonging to BCS classes II and IV, with BA in the range from 37% to 100%, had representatives within all groups in terms P-gP efflux, first-pass metabolism, and number of compartments in a pharmacokinetic model (1 and 2). Lipophilicity and acid–base properties of an API correlate significantly with PPB , and they also both essentially impact effective permeability. Our analysis shows the higher success rate when PPB was below 90% can be in all but one cases correlated with permeability being below 2 cm/s × 10 −4 . It seems that for a specific group of poorly soluble APIs lower PPB might be associated with lower risk for non-BE result, but the impact might be to certain extent confounded by other parameters that correlate with lipophilicity and acid–base characteristic of API. Elimination Half-Life within the Group of Poorly Soluble APIs For an immediate-release product, terminal half-life of an API is a hybrid measure of clearance and volume of distribution. Based on the analysis of Vd and AUC/ D , it was not expected that t 1/2 would have a direct influence on the BE study outcome. This expectation was confirmed within the group of poorly soluble APIs where no significant difference in ln( t 1/2 ) was observed between the BE and non-BE groups (Table ). Number of Compartments in a Pharmacokinetic Model within the Group of Poorly Soluble APIs Distribution of API into a peripheral compartment impacts concentration in a central compartment and can, as such, hypothetically create more or less discriminatory conditions for testing BE, i.e., the more compartments we need to describe the pharmacokinetics of API, the more complex are its distribution and elimination processes, and hypothetically, the risk for a non-BE study result is higher. We have confirmed this with our analysis, where non-BE results occurred in significantly different 11%, 22%, and 40% cases (Table ) when APIs pharmacokinetics was described by one, two, and more than two compartments, respectively. However, it should be considered that in our database we had only one API with more than two compartments. Risk Mitigation Strategy Identification of parameters that are associated with non-BE outcome calls for mitigation strategy. Noncomprehensive set of examples that may guide reader towards creation of such strategy are: (1) If bioequivalence risk assessment is early, then we can guide development to select or control excipients to minimize impact on pharmacokinetics of APIs that are subject to first-pass metabolism or P-gP transport. (2) Identification of risk parameters also directly guides selection of appropriate methodology (in vitro, ex vivo, animal in vivo, in silico) that is used in predicting human in vivo behavior of API. (3) Lastly, also appropriate BE study design is important: inclusion and exclusion criteria need to be comprehensive when API is subject to first-pass metabolism and/or P-gP transport and blood sampling schedules plan needs to be adjusted when T max is very short or when disposition is described by multicompartment models. Significantly different percentages of non-BE studies were found across different BCS classes in our database (Table and Fig. ), indicating high association between BCS and BE study outcome. This is in line with a number of publications that supported in vivo predictive nature of BCS . Failure rate within the group of highly soluble APIs was negligible (~ 1%) and even lower than that found in the literature (10–16%), regardless of the wide range of BA (18–100%), presence of first-pass metabolism, and/or P-gP mediated efflux attributable to some highly soluble APIs. Low occurrence rate of non-BE results in our database supports BCS biowaiver approach for class I and III APIs implemented by numerous health authorities. In theory, BCS class I and III drugs were presented as less risky for non-BE outcome compared with classes with poorly soluble APIs, while the publications often reported similarly low failure rate for classes I, III, and IV, ranging from 10% to 16%. Similarity of BCS class IV APIs to BCS class I and III was usually attributed to smaller sample size of BCS IV group (i.e., insufficient power to detect differences) . Our estimated failure rate of 12.5% within BCS class IV seemed comparable to that reported in the literature; however, in our case the difference between highly soluble group of BCS and BCS class IV was obvious due to negligible failure rate in the highly soluble BCS classes. BCS class II was previously reported as the most critical for conclusion of BE, where failure rate ranged from 28% to 39% . When considering only highly variable BCS class II APIs, Lamouche reported high, flip-of-a-coin-like 54% failure rate of BE studies . In accordance with the literature data, our estimates showed that BCS class II has the highest occurrence of non-BE results (Table and Fig. ). Since the BCS classification can be done after assessment of solubility and permeability, and the assessment of permeability is sometimes challenging in the early stages of development, we decided to combine BCS class II and IV APIs into one poorly soluble group for further subanalysis. In addition, the classification available in the literature may sometimes be misleading due to the fact that the permeability assessment for BCS based biowaiver approach is many times surrogated by in vivo BA assessments. Thus, there are sometimes inconsistencies in BCS classification between BCS II and IV classes; for example, BCS II APIs with low BA but high permeability may be misclassified as BCS IV APIs. Combining both poorly soluble BCS classes we avoided any assumptions regarding permeability while assessing factors influencing BE study outcome. When BA was above 85%, either for BCS class I or II, all BE studies ( N = 33) were successful. Studies are usually powered at 80% or 90%, so the probability of study success, if product is indeed BE, is 80% or 90%, respectively. That further means that the scenario of observing zero non-BE studies among 33 is 0.1% (0.8^33) or 3.1% (0.9^33), which labels observation of zero non-BE studies in 33 cases, as unlikely. BCS biowaiver guidelines define BA of 85% as the limit for fraction of absorption which defines highly permeable APIs. Above this limit, less strict dissolution criteria and formulation differences requirements are set for the product being eligible for BCS biowaiver . This suggests that APIs with high extent of absorption are less risky for BE testing. This was also observed in our analysis. It seems that high permeability along with absence of presystemic API degradation/extraction processes decreases risks for BE study failure. On the other hand, regardless of wide BA range (18–100%) within highly soluble APIs, there was only one case of non-BE result. This supports acceptability of the BCS biowaiver approach for APIs with high (BCS class I) or low permeability (BCS class III). APIs with a wide range of BA were included in analyses. BA was a significant feature in all of the relevant analyses, i.e., lower BA (especially below 40%) was one of the key indicators for problems with BE. Low solubility might be the reason for low BA; however, it can also be caused by the low permeability, gastrointestinal instability, first-pass metabolism, and/or P-gP-mediated efflux. Some of these factors and their association with non-BE results are discussed in the Sects. , , and . As a general rule, higher permeability of poorly soluble APIs was more risky for conclusion of BE (Table and Fig. B). Namely, for APIs with high permeability, dissolution rate or solubility becomes a limiting factor for absorption, and these conditions are more challenging in terms of BE outcome where formulation performance is compared. These results correspond well with the BCS class II APIs being the most challenging group of APIs. The fact that this permeability is determined in silico by GastroPlus is at the same time an advantage, as it is easily determined early in the development, and also disadvantage, as the in silico estimate may be associated with less precision. First-pass metabolism can significantly affect BA and is as such of particular interest when assessing risk for concluding BE. Association between first-pass metabolism, present at 40% of highly soluble APIs, and BE outcome could not be explored within highly soluble APIs, since there was only one non-BE study. Most, i.e., 94%, of these APIs with first-pass metabolism had low variability of pharmacokinetics. The low occurrence of non-BE results is in agreement with prediction of Fernández and coworkers that highly soluble APIs with first-pass metabolism and low variability held less risk for concluding BE . On the other hand, our analyses infer that first-pass metabolism is associated with higher incidence of non-BE results in studies with poorly soluble APIs. These results are in line with the work of Cristofoletti and coworkers, where drug disposition and metabolism based classification (BDDCS) held similar predictive value for BE outcome as BCS . For the subset of APIs where BA was below 40%, presence of first-pass metabolism did not increase the absolute risk for non-BE. However, we have to take the latter conclusion with caution since there were only six observations in the group of APIs without first-pass metabolism. First-pass metabolism may be influenced by excipients , which can differ between the generic and reference product and thus impact BE outcome, and by different physiological or pathophysiological conditions, which may increase the variability of exposure. However, the impact of the latter can be excluded in our analysis since all BE studies in our database were performed with healthy volunteers. Another process that may impact BA is efflux of API from enterocytes. There are numerous transporters that perform this task. However, most often this process is governed by P-gP (MDR-1) transporter. H. Kortejärvi and coworkers have shown with simulations that, if a highly soluble API is a substrate for P-gP, this may significantly influence the outcome of BE studies . This risk cannot be observed in our database, since all studies with highly soluble APIs that were substrates of P-gP concluded BE. On the other hand, our analysis revealed that for poorly soluble APIs involvement of P-gP efflux transporter in pharmacokinetics of API seemed to significantly increase occurrence of non-BE results (Table ). Our analysis also suggests that the risk for non-BE outcome may be higher if poorly soluble API has low BA (<40%) and is a substrate for P-gP efflux. These results were not surprising, since P-gP efflux may also be influenced by excipients, which can differ between the generic and reference product and thus impact BE outcome. In addition, P-gP may impact variability of drug exposure . If rapid release is claimed to be clinically relevant and important for onset of action or is related to adverse events, then there should be no apparent differences in median T max and its variability between the test and reference product . In such case, T max would be a significant parameter determining the BE outcome; however, this was not the case in any of our BE trials, where primary parameters were always C max and AUC. Since T max is a composite parameter of elimination and absorption rate, the latter can be impacted by formulation (differences); thus, T max parameter could indeed hold relevant information regarding the study outcome. Furthermore, our hypothesis was that if T max is very short, gastric emptying is not playing a role in limiting the absorption rate and the absorption is rather limited by the release of API from the formulation. This was confirmed by our analysis of poorly soluble APIs where median T max of the non-BE group was 0.9 h lower than the median of the BE group (Table ). The trend was similar when the analysis was split for fasting and fed conditions (Fig. C). Similarly, the association between T max class (below or above 1.5 h) and BE outcome was observed (Table ). Splitting the dataset to fasting and fed conditions resulted in a similar conclusion under fasting conditions, but loss of association under fed conditions (even when a cutoff value higher than 1.5 h was considered). This is not surprising since under fed conditions gastric emptying limits the absorption rate. The T max could lose its predictive value for this reason. The authors acknowledge that the cutoff at 1.5 h might also be too strict to capture all cases with the fast absorption (not limited by gastric emptying), since the slow distribution/elimination can manifest in prolonged T max values. On the other hand, 1.5 h cutoff might be just what we are looking for in BE risk assessment, since the risk for non-BE is particularly high for APIs with fast absorption and fast distribution/elimination, resulting in narrow peaks in plasma concentration profile. See also Sect. where we discuss the association between number of compartments to describe distribution/elimination and BE outcome . Finally, Kortejärvi and coworkers have used simulations to show that the risk for not concluding BE is increased for highly soluble and highly permeable APIs with very short T max . However, the higher risk for non-BE results predicted by simulations has not been confirmed in our analysis of highly soluble APIs. max /AUC within the Group of Poorly Soluble APIs Ln( C max /AUCi) of the non-BE group was lower than that of the BE group, but statistical significance could not be shown (Table ). C max /AUC was previously recommended as a less polluted measure of absorption rate as C max , but was later shown to have similar flaws as C max in lacking sensitivity in indicating changes in absorption rate constant . This might be why C max /AUC was not discriminatory in detecting non-BE results. Variability of drug exposure after oral administration poses a significant challenge in modern drug development. Factors impacting variability are diverse and include human physiology variation, bioanalytical variation, and formulation technology variation . It is generally recognized that, with oral administration, distribution and elimination of API cannot be influenced by differences in formulations. For this reason, a crossover design is usually implemented for testing of BE, i.e., testing whether there are any differences in formulation performance. In such setting, “quazi” intra-individual variability (intra-CV of C max and AUC) is the variability of interest. It eliminates variability that may arise from differences related to distribution and excretion of drug between different subjects. Regardless, high intra-CV could still be one of the reasons for decreased power and was more problematic in the past when tools for handling of high intra-CV were not available or accepted by regulatory agencies. These problems were manifested in high occurrence of non-BE results, e.g., 54% of non-BE studies with BCS class II APIs with variability higher than 30% (i.e., high variability) reported by Lamouche . Examples of tools that tackle high variability are higher-order crossover design accompanied by scaling or widening of BE limits . All studies in our database of poorly soluble APIs were crossover studies (2 × 2 or higher order). Intra-CV of C max and AUC was not significantly different between non-BE and BE group in our database (Table ). This is not surprising considering the majority of non-BE studies had post-hoc power above 80%, meaning intra-CV was adequately considered in study design. It is noteworthy that higher average intra-CV of C max and AUC was observed for APIs with BA below 40% compared with APIs with BA above 40%, which suggests that processes decreasing BA increase variability of exposure. Regardless of the crossover nature of BE study, distribution and excretion properties and inter-occasion differences within one subject can still impact intra-CV and/or create more or less discriminatory environment for testing of BE. For this reason, differences between non-BE and BE group were explored for parameters that describe distribution and elimination of drug: Vd, AUC/ D (inverse of apparent clearance), PPB, t 1/2 , and number of compartments in pharmacokinetic models that best describe plasma profiles. The ln(Vd) between the BE and non-BE groups was not significantly different (Table ). Vd did not seem to correlate with intra-CV of AUC, but a moderate correlation between Vd and intra-CV of C max was found (refer to supplemental data). Apart from the impact on variability we could not conclude anything about predictable value of Vd for non-BE outcome. D ) within the Group of Poorly Soluble APIs Sakuma and coworkers suggested a correlation between AUC/ D (fraction of absorption/clearance) and number of subjects in a BE study with highly soluble APIs . Yamashita and colleagues explored the correlation of AUC/ D ratio and parameters that impact BE study success. AUC/D correlated with the width of the 90% confidence interval (i.e., with variability) in BCS classes I and III . Our analysis showed that the mean ln(AUCi/ D ) was not significantly different between the BE and non-BE groups (Table ), although the mean ln(AUCi/D) of non-BE group was 0.5 h/L lower. This trend agrees with observations of Yamashita and coworkers where lower AUC/D implied higher chance of non-BE results, i.e., APIs with fast clearance, low permeability, high first-pass metabolism, and low GIT stability, were those that held the highest risk . In addition, we have also found a correlation between ln(AUCi/ D ) and intra-CV for C max . In contrast to the work by Yamashita and coworkers, the correlation was found not only for the highly soluble but also for the poorly soluble APIs (please refer to supplemental data). We could not conclude anything on the association of non-BE outcome and AUCi/D, but if the association exists, it may be confounded with the intra-CV. Protein binding may serve as a reservoir from which the API is slowly released as the unbound form and can prolong t 1/2 of the API. When an API is highly bound to plasma proteins, it typically has lower volume of distribution . As such, PPB may impact BA and could hypothetically create more or less discriminatory environment for testing of BE. Our analysis showed that PPB medians of BE and non-BE groups were not significantly different (Table ). Further visual analysis showed a group of poorly soluble APIs ( N = 11) with PPB below 90% where there are no non-BE results (Fig. D). Considering the study power of 90%, there was still a high 31% chance of not observing any non-BE among 11 studies (0.9^11). These APIs belonging to BCS classes II and IV, with BA in the range from 37% to 100%, had representatives within all groups in terms P-gP efflux, first-pass metabolism, and number of compartments in a pharmacokinetic model (1 and 2). Lipophilicity and acid–base properties of an API correlate significantly with PPB , and they also both essentially impact effective permeability. Our analysis shows the higher success rate when PPB was below 90% can be in all but one cases correlated with permeability being below 2 cm/s × 10 −4 . It seems that for a specific group of poorly soluble APIs lower PPB might be associated with lower risk for non-BE result, but the impact might be to certain extent confounded by other parameters that correlate with lipophilicity and acid–base characteristic of API. For an immediate-release product, terminal half-life of an API is a hybrid measure of clearance and volume of distribution. Based on the analysis of Vd and AUC/ D , it was not expected that t 1/2 would have a direct influence on the BE study outcome. This expectation was confirmed within the group of poorly soluble APIs where no significant difference in ln( t 1/2 ) was observed between the BE and non-BE groups (Table ). Distribution of API into a peripheral compartment impacts concentration in a central compartment and can, as such, hypothetically create more or less discriminatory conditions for testing BE, i.e., the more compartments we need to describe the pharmacokinetics of API, the more complex are its distribution and elimination processes, and hypothetically, the risk for a non-BE study result is higher. We have confirmed this with our analysis, where non-BE results occurred in significantly different 11%, 22%, and 40% cases (Table ) when APIs pharmacokinetics was described by one, two, and more than two compartments, respectively. However, it should be considered that in our database we had only one API with more than two compartments. Identification of parameters that are associated with non-BE outcome calls for mitigation strategy. Noncomprehensive set of examples that may guide reader towards creation of such strategy are: (1) If bioequivalence risk assessment is early, then we can guide development to select or control excipients to minimize impact on pharmacokinetics of APIs that are subject to first-pass metabolism or P-gP transport. (2) Identification of risk parameters also directly guides selection of appropriate methodology (in vitro, ex vivo, animal in vivo, in silico) that is used in predicting human in vivo behavior of API. (3) Lastly, also appropriate BE study design is important: inclusion and exclusion criteria need to be comprehensive when API is subject to first-pass metabolism and/or P-gP transport and blood sampling schedules plan needs to be adjusted when T max is very short or when disposition is described by multicompartment models. BCS was confirmed to be highly predictive for BE success. Only one non-BE study was determined within the group of products with a highly soluble API with wide range of BA (18–100%). This supports the BCS biowaiver approach for class I and III APIs implemented by numerous health authorities. Immediate-release products with BCS class II APIs are confirmed again to have the highest risk for non-BE results. Within groups of poorly soluble APIs (where the majority of non-BE results were observed), absolute BA was shown to be significantly lower for the group of non-BE results. This is in line with the significantly higher occurrence of non-BE results for poorly soluble API with presence of first-pass metabolism and affinity for P-gP transport (efflux). In silico estimated permeability and T max were shown as potentially relevant features for predicting BE outcome. As expected, Vd, total clearance, and t 1/2 were not associated with BE outcome. PPB between BE and non-BE group was not different; however, we have not observed any non-BE results for poorly soluble API with PPB below 90%. Our analysis also showed significantly higher occurrence of non-BE results for poorly soluble APIs with pharmacokinetics described by multicompartment model (two or more than two compartments). The conclusions for poorly soluble APIs were the same on a subset of fasting BE studies; for a subset of fed studies there were no significant differences between factors in BE and non-BE groups. One possible extension of our work could be to include additional acido-basic and specific solubility characteristics of APIs and see how these differentiate studies with regards to the BE outcome. On the other hand, it is easy to see how additional parameters showing differences in dosage form characteristics, e.g., process, composition, in vitro dissolutions, etc., could improve BE risk assessment. However, the aim of this research was to evaluate to what extent the BE risk could be predicted at the early beginning of the product development when the parameters related to the dosage form are limited or unknown for generic as well as for the innovator product. Univariate analysis or plots are simple but essential approaches to exploring the basic relationships of parameters in the dataset. There are some limitations to such approach. Firstly, the type I error is not controlled so the conclusions are to be taken with caution. Secondly, many interactions between parameters may not be found or dealt with, especially, when dealing with such interrelated parameters as presented and discussed in this paper. There are tools available to tackle these problems ranging from simple linear or logistical regression analysis to machine learning/artificial intelligence techniques with different levels of complexity. Considering the limitations, one should use findings presented in this paper as a groundwork for the further research and development of tools for early BE risk assessment. Below is the link to the electronic supplementary material. Supplementary file1 (PDF 690 KB)
A Review of the Current State of Global Surgical Oncology and the Role of Surgeons Who Treat Cancer: Our Profession’s Imperative to Act Upon a Worldwide Crisis in Evolution	e1fae806-1448-4634-843a-a07296f83dc5	10175401	Internal Medicine[mh]	The theory of ‘epidemiologic transition’ has been refined for more than 50 years within the disciplines of public health and demography. While it has its criticisms, it is instructive as a backdrop to discussing the current, pressing needs of global surgical oncology. The theory observes that as the contribution of infectious diseases to population mortality rates declines, there is also an observed increase in life expectancy and an increase in the proportion of mortality attributed to non-communicable diseases (NCDs), including cardiovascular disease and cancer, among others. This is often observed in association with a ‘demographic transition’, in which a decline in birth rate along with the above factors combine to result in an aging population. From the earliest descriptions, it has been observed that while Western Europe and other geographic areas that are now described as high-income countries (HICs) exhibited epidemiologic transition over a period of more than 150 years, more rapid changes have subsequently been observed in various parts of the world, often associated with more rapid economic, industrial, and societal changes, and tied to more organized public health interventions against infectious disease and other causes of youthful mortality. , , It is observed that as individuals live longer, they are more likely to get cancer; more people worldwide are living long enough, and the increase in the burden of neoplastic disease is dramatic. , High rates of risk factor exposure in low- and middle-income countries (LMICs), including tobacco, alcohol, and others, further exacerbate this trend. Cancer is now the second leading cause of death worldwide after cardiovascular disease and is a rising threat to population health as the global rates of cancer incidence and mortality continue to climb. , , According to data compiled in the Global Cancer Observatory, the estimated global number of new cancer cases in 2020 was almost 19.3 million, nearly doubling from 10 million in the year 2000. , The number of global cancer mortalities reported in 2020 approached 10 million deaths, an increase from approximately 6 million deaths in the year 2000. , The global number of new cancer cases is projected to increase to 30.2 million cases in 2040, associated with 16.3 million deaths. The majority of the continued increase in the cancer disease burden is expected to occur in LMICs, consistent with an understanding of the geography where epidemiologic transition is most actively occurring. , It should be noted that projections of future burden of neoplastic disease have historically been underestimations; in 2001, the projection for 2020 based on similar data sources was 15 million new cases, now thought to be an underestimation of more than 46% compared with the true increase over that interval, although the projected number of deaths appears to have been more accurate. , The underestimated prediction of future cancer incidence may be due at least in part to a low penetrance of cancer registration in LMICs, along with systematic weaknesses in cancer registration that lead to incompleteness in the data, as well as issues related to impaired access to care and low utilization of care such that the formal healthcare sector in LMICs is not encountering the true population incidence. , , , , – This is consistent with the major increase in cancer incidence occurring within the geography that is least equipped to measure it. Moreover, the increase in cancer incidence and mortality within LMICs is diverse and uneven, with rates of increase often most rapid in the lowest income settings, and heterogeneity across regions regarding the histologic makeup of the cancer burden. In some regions, a significant portion of the cancer burden continues to be secondary to infectious diseases such as hepatitis viruses, human papillomavirus, human herpesvirus 8, and human immunodeficiency virus (HIV). , , , In 1970, only 15% of reported cancer incidence was attributed to LMICs. This has now increased to an estimated 59%, while LMICs bear approximately 71% of the worldwide burden of mortality due to cancer. Individuals in LMICs with cancer are considerably more likely to die of their disease, and orders of magnitude more likely to do so without palliation of associated symptoms. Given the observed trends and causal factors relating to distribution of worldwide cancer burden, it is clear that demand for treatment of cancer will remain or become a leading concern of healthcare systems in every country on the planet for the foreseeable future. , , Although highly developed countries are now seeing a decline in mortality rates associated with cancer, it remains the leading cause of premature death (between the ages of 30 and 69 years) in those countries. Cancer currently ranks as the second, third or fourth leading cause of premature death in most LMICs, after other NCDs, but should be expected to climb in the ranks as these countries follow in the epidemiologic patterns of more developed nations. Regardless, LMICs already have a greater relative gain to be realized in overall population longevity by preventing premature deaths from cancer than HICs do. Cancer is a severe and constant threat to public health and economic health in every region of the world through its impacts on rates of disability and death; in HICs, this threat is unrelenting and in LMICs this threat is rapidly worsening. In context of these demographic and epidemiologic phenomena, experts worldwide have been progressively recognizing cause for alarm, issuing calls to action and dire warnings. , , , , , , In a 2010 paper marking the creation of the Global Task Force on Expanded Access to Cancer Care and Control in Developing Countries, Farmer et al. described ‘gaping voids in cancer care and control worldwide’ and called the need to address this crisis as ‘an urgent health and ethical priority’. Cancer in LMICs is a ‘neglected disease’ with ‘severe access limitations’ according to Eniu et al.—an ‘epidemic’ that ‘will become the leading public health issue’ in these nations. In 2022, Ngwa et al. described the ‘growing cancer crisis’ and pleaded that ‘urgent action is needed’. These discussions recognize both the current profound disparities in access to cancer care and control around the world, as well as expectation that these disparities will dramatically worsen without decisive intervention. – , , , , Increased awareness of cancer’s growing global threat has coincided with increased awareness of the inadequacy of surgical systems worldwide—recognizing surgery as an especially neglected element of healthcare systems across the developing world. , This has major implications on the ability to provide adequate treatment to the growing cancer burden. In 2015, publication of the Lancet Commission’s Global Surgery 2030 report concluded that 5 billion people worldwide lack access to surgical care, whether for benign or oncologic indications, mostly in LMICs. In the same year, Sullivan et al. published the Lancet Oncology Commission’s Global Cancer Surgery report, which added that about 80% of new cancer cases will need surgery, some of them multiple times, and estimated that in the year 2030 there would be 17.3 million patients requiring surgical procedures for diagnosis, treatment, and/or palliation of cancer worldwide. Of these individuals with cancer, fewer than 25% will be expected to have access to ‘safe, affordable, and timely surgery’. The rate in middle-income countries will be 20%, and just 5% in low-income countries. The report emphasizes the need to prioritize surgery as a core component in plans for both cancer control and universal health coverage (UHC), an objective widely supported by the global health community. Strengthening surgical systems for cancer treatment requires multiple complex and interconnected investments, including infrastructure, equipment, workforce (not only of surgeons but also of ancillary staff and complementary medical disciplines), and systems for service delivery, financing of services, and management of information. , , Advocates for surgical systems strengthening recognize that training more surgeons is a core requirement, and this includes a need to train more surgeons in their capability to treat cancer with excellent outcomes. , , , Zafar et al. analyzed the global surgical workforce for cancer and estimated that 9.5 million cancer surgeries were required in 2018, with the ratio of patients needing cancer surgery to the surgical workforce observed to be approximately 10 times higher in LICs than in HICs. Perera et al. performed a modeling study of the optimal surgical and anesthesia workforce for treating cancer, in which they estimated an increase in annual cancer surgeries needed worldwide from a little over 9 million in 2018 to over 13.8 million surgeries in 2040. Their model indicated that a workforce increase of 25% in HICs, 10% in upper MICs, 67% in lower MICs, and 383% in LICs would be needed to correct current deficiencies, and projected that the demand for cancer surgeons will be further increased by 29% in HICs, 51% in upper MICs, 67% in lower MICs, and 107% in LICs by 2040. The greatest current deficiency and projected growth in demand is in LICs, such that the current workforce of surgeons in those countries would need to multiply by a factor of 9.7 to meet anticipated 2040 demand. Both studies acknowledged limitations in the data on cancer incidence in resource-constrained settings, and other publications have given higher projections for the increase in needed cancer surgeries. , For example, Sullivan et al. suggested the number might be as high as 17.3 million cancer patients with an indication for surgery in the year 2030, with 10 million of those in LMICs. Similar workforce and access deficiencies exist across many medical specialties, including those that most actively participate in the multidisciplinary management of cancer, such as pathology, radiation oncology, and medical oncology. , , Related to this, surgeons in many LMIC settings take on the management of other oncologic treatments, such as chemotherapy. – , The crisis is not new within the history of modern oncology, but it is evolving. Roswell Park—the surgeon who in 1898 was instrumental in founding the cancer research institute that now bears his name—lobbied for public funding based on the proliferating threat of cancer and the ignorance of its causes and treatments. The second annual report for this institute for the year 1899 included a quote from the Philadelphia Medical Journal, celebrating their work to study cancer: “this pitiless enemy of civilization, which is increasing in such a startling way”. James Ewing, the pathologist who is credited as the driving force behind the early 20th century revitalization of New York City’s Memorial Hospital as a model for modern cancer hospitals, and for whom the Society of Surgical Oncology (SSO) was originally named, recognized cancer as a public health threat and the need to take a multipronged approach toward its control. – The Union for International Cancer Control was founded in 1933 on the premise that the control of cancer was an international scientific priority. Recognition of the increasing and critical public health threat represented by cancer has been a constant throughout the history of surgical oncology as a profession, and has been the driving force behind the creation of multiple other public and professional societies and institutions. The sense of nihilism and discouragement that sometimes characterizes current attitudes towards cancer in LMICs mirrors perceptions that were held in HICs just decades ago. , , Tremendous progress has transformed perspectives on cancer over the past century and more within the world’s leading cancer centers, serving to highlight the unfinished work left by past visionaries for us, their heirs. Unfortunately, our efforts contend not only against cancer and constrained resources but also the impacts of conflict and pandemic. – The work ahead is more than SWTC can accomplish alone. It will require a consistent, concerted, collaborative engagement with the many stakeholders attached to the issue, recognizing that solutions for worldwide cancer control align with crucial needs and agendas in global health and development. Strengthening systems for surgical treatment of cancer is integral to strengthening surgical systems in general, recognizing a system capable of providing (frequently more complex) cancer operations will typically also be capable of providing treatment for benign disease. Strengthening surgical systems is also integral to overall healthcare systems strengthening, as one-third of the global burden of disease is surgical, surgical treatments save lives in meaningful and cost-effective ways, and surgical and non-surgical services are complementary and interconnected in functional healthcare systems. Strengthening surgical services involves support for the entire care continuum, including perioperative care. For instance, the ASOS study showed that patients are twice as likely to die from routine surgery in many countries in sub-Saharan Africa due to deficits in supportive care. Healthcare systems strengthening is in turn integral to the global development agenda. Underscoring the importance of surgery and cancer control within this construct, the 68th World Health Assembly in 2015 approved resolution WHA 68.15, affirming the need to include surgical and anesthesia services in strengthening healthcare systems to achieve UHC. The 70th World Health Assembly in 2017 approved resolution WHA 70.12, affirming the need for national and global entities to prioritize investment in cancer prevention and control. , – In 2015, the member states of the United Nations unanimously endorsed the Sustainable Development Goals (SDGs), a group of 17 aspirational objectives with subsidiary targets and metrics to guide worldwide development plans leading up to 2030. , SDG 3 focuses on ‘Good Health and Well-Being’, encompassing multiple supporting targets. Target 3.4 aims to reduce premature deaths from NCDs such as cancer by one-third, through prevention and treatment, by 2030. Target 3.8 aims to secure UHC, ensuring access to high-quality essential health services with protection from associated financial toxicity. , Given our discussion of the rising threat of premature deaths from cancer and the role of surgical treatment in preventing them, achieving targets 3.4 and 3.8 inescapably requires efforts to strengthen systems for the surgical treatment of cancer. Improving equitable access to surgical treatment for cancer also contributes to achieving other SDGs, including SDG 1, ending poverty; SDG 4, ensuring access to education; SDG 5, ensuring gender equality; SDG 9, fostering innovation and infrastructure development; and SDG 10, reducing inequality within and among countries. Yet, while achievement of health targets will support achievement of other SDGs, these targets do not appear to enjoy indirect benefits from investments elsewhere and thus require direct investment. , , The sustainable development paradigm is complex and ambitious. Skeptics observe that as with the millennium development goals that preceded the SDGs, it is unlikely that all targets will be achieved and that there are challenges in prioritizing allocation of scarce resources among these goals. However, the SDGs provide structured acknowledgment of priorities and frontiers for development that are shared among HICs and LMICs, providing a framework for the discussion of public and private investment within which arguments can be made for investments to strengthen surgical systems for cancer. The paradigm also allows recognition that disparities and deficiencies in access to the surgical treatment of cancer are a universal challenge, and attention is needed to correct them in every country, whether an HIC or an LMIC. Barriers to equitable access to surgical treatment of cancer exist for underserved populations in HICs as well as in LMICs, as evaluated by both geographic and demographic distinctions. , , , , , Although they will not be ‘one-size-fits-all’, solutions to these disparities will be fundamentally similar whether employed within HIC or LMIC contexts, with attention needed toward context and adaptation. An understanding of how the priorities for capacity building in global surgical oncology relate to the broader priorities and agendas within global public health and worldwide sustainable development should be central to collaborative efforts between SWTC and the high-level stakeholders positioned to interact with efforts in this area, such as international agencies, national governments and their agencies including ministries of health, non-governmental organizations, and donors. Sustainable development is a paradigm for shared gain, offering prospects of improved geopolitical and economic stability and creating opportunities for enhanced international strategic partnerships built on capacity for trade rather than need for aid. Considering all this, stakeholders with an incentive to invest in strengthening surgical systems for cancer include the governments and institutions of every country on Earth. , , , Eliminating deficiencies and disparities in global surgical oncology will certainly require significant financial resources. Multiple authors have observed that both cancer control and surgical systems have been neglected by global health initiatives, seemingly under a misconception that cancer is a group of diseases that are too complex and expensive to be treated in resource-constrained settings, and that surgical treatments are likewise unaffordable in these areas. – , , , , However, many of these authors also argue that rather than seeing these challenges as needs the world cannot afford to address, they must be seen as needs the world cannot afford not to address. , , Sullivan et al. estimated that the failure to meet the needs for surgical treatment of cancer worldwide would correspond to an estimated economic loss of $6.2 trillion over the time period of 2015–2030. Alkire et al. projected a cumulative gross domestic product (GDP) loss of more than $12.1 trillion for the same time period, caused by failure to provide surgical treatment for cancer, based on a ‘value of lost output’ calculation. Worldwide economic figures from 2017 suggested the costs of the disease burden of cancer to be an estimated $1.16 trillion per year, while the resources invested on treating cancer were $300 billion. Countries around the world stand to lose at least 0.5–1.5% of GDP annually to the economic burden of surgically treatable cancers. Calculating costs associated with surgically treatable cancers on the broader ‘value of a statistical life’ methodology would suggest an annual cost of up to $7.4 trillion per year from death and disability, or a value approaching as much as 10% of annual GDP in some HICs. , Regardless of the methodology used to assess the costs, the economic impacts of the failure to surgically treat cancer are clearly very expensive in terms of lost productivity and quality of life, and the differential between the amount of resource invested in the treatment of cancer and the economic losses associated with it suggest ample opportunity to justify dramatic increases in investment. , – Resources directed to healthcare should be seen as wise allocations expected to generate meaningful yields. , , , , Investment in healthcare is a driver of broader economic growth, creating demand for jobs, higher education, industrial growth, infrastructure, and innovation; according to the McKinsey Institute, for every dollar invested in health, $2–$4 of economic benefit stands to be gained. , , , , , Even more bullish, the ‘value of additional life-years’ approach would estimate a 9- to 20-fold return on investments in health. High-level recommendations on resource allocation in global health have increased benchmarks for cancer control, suggesting that LMICs should direct more than two-thirds of their healthcare budgets toward management of NCDs such as cancer, including surgical capacity to treat these diseases. A significant number of publications have drawn attention to challenges and needs in the global practice of surgical oncology, and many have laid out broad priorities and strategies. Still, SWTC may feel that opportunities to engage with the issue are limited, intimidating, inaccessible, and/or associated with a low likelihood of professional reward. , , , , , While efforts have been made to propose the roles of surgeons in this area, there remains a need to strengthen the definition of these roles and their importance, to secure increased participation of surgeons, and to commend their contributions. , , , , , – Included below are specific categories of critical opportunities recognized across multiple assessments of the global oncology landscape. There is a critical need for SWTC to participate in advocacy . This is an opportunity to more proactively and expansively define the position of surgical oncology on the global stage. Surgeons should engage in campaigns to raise awareness among populations and their leadership regarding the concerns discussed in this paper and its cited references. SWTC should advocate for inclusion of surgical services for cancer in the design and implementation of plans for cancer control and for healthcare systems strengthening. , Within these efforts, surgeons should collaborate with other oncologic disciplines to draw attention to the importance of prevention, screening, early detection, safe and effective treatment, surveillance, and palliation, as well as the value of multidisciplinary care for cancer and the need for comprehensive systems strengthening to provide it. SWTC and their professional societies and institutions should also participate along with advocacy groups such as the Global Forum of Cancer Surgeons (GFCS) and the Global Alliance for Surgical, Obstetric, Trauma and Anaesthesia Care (G4 Alliance) to engage with high-level stakeholders to create and direct attention toward effective opportunities for investment. , , When engaged in advocacy efforts, there is an opportunity to emphasize the investment paradigm—resources directed toward these needs are not losses, but are wisely allocated, designed to prevent further loss, to improve population longevity and productivity, and to contribute to sustainable development and economic growth. , , , , Work is also needed in research . The community of SWTC has a long tradition of holding in high esteem the surgeon-scientist and recognizing the need for continued research by surgeons not only into the surgical innovations that will advance the field, but into the understanding of the molecular and biochemical structures and behaviors of cancer, and the targets these present for treatment. , , , While this work is indeed important and more of it needs to be done in diverse settings—especially including LMICs—more surgeon-scientists are also needed who will employ the sciences of sociology, economics, and public health, among others, to overcome barriers to dissemination, implementation, and access to best practices derived from what is learned by other surgeon-scientists in laboratories and leading clinical institutions. , , – Particularly in low-resource settings, surgeons find their efforts as investigators constrained between lack of funding sources and the demands of high clinical volumes. There is an opportunity to not only continue to advance the science of how to prevent and treat cancer across all settings but to exponentially increase the impact of those discoveries by advancing the science of optimizing the applicability and accessibility to those discoveries’ benefits for everyone on the planet. There is also an opportunity to recognize and clarify the academic value of all aspects of this work. , , , , Training of additional workforce for surgical treatment of cancer patients is another critical need. While there is diversity in the training pathways for SWTC around the world, work has been done to suggest core elements of knowledge and skill to be included in curricula for these surgeons. – The International Federation of Head and Neck Oncologic Surgeons (IFHNOS) has established a successful Global On-Line Fellowship (GOLF) for training of surgeons around the world in the treatment of head and neck cancers. This program creates opportunities to reach a greater number of trainees while offering them supportive learning structures to enhance their skills and knowledge within a diversity of practice settings in countries around the world; the leaders of this program offer it up as a model to replicate in other disease sites in order to meet global needs. There is an opportunity for surgeons to support and participate in the concept that training should be the ‘right size’: the optimal content and duration of training for candidates from, in, and for the populations and practice settings where they are needed, and to participate in increasing the capacity across all settings for training SWTC to meet future global workforce needs. , , , SWTC are also critically needed to participate in efforts for sustainable development and systems strengthening . There is an opportunity for surgeons to apply for and administer grants focused on development and capacity building; to create enduring programs and partnerships that enable SWTC to leverage their collective abilities and energies to address critical priorities; and to otherwise offer their own time, knowledge, and skills to enhance capacity in areas of great need. These efforts should focus on collaborative approaches to addressing needs as perceived by all relevant stakeholders, while minimizing waste and disruption and maximizing sustained benefit. The pitfalls of short-term volunteerism and of projects or programs implemented without adequate stakeholder input should be considered and avoided. , – Multiple examples exist of bilateral or multilateral international partnerships and programs that have had enduring effect; where these have achieved success and contributed sustained benefit, there is an opportunity for SWTC to emulate and expand. , , , , Within these efforts, it must be recognized that there is tremendous diversity and heterogeneity in epidemiology and resource allocation not only among countries but sometimes within countries such as India where incidence rates and treatment options vary between regions; strengthening initiatives will need to be context-specific but informed by unified guiding principles. , Within the arc of human history, the discoveries that have enabled major improvement in cancer survival are new achievements, contemporary to such fields of development as telecommunications and aviation. The area of telecommunications in particular is apt for discussion of the challenges associated with meeting the world’s need for cancer control. LMICs have proven the ability to import and adopt later, more streamlined and/or more cost-effective iterations of technologies, to develop innovative applications of these technologies, and to seize opportunities to leapfrog over earlier generations of technologies to find more accessible and scalable solutions. Both telecommunications and aviation have also combined to make the sharing of knowledge and the development of global collaboration more facile than ever before. A wealth of opportunity exists to explore innovative and collaborative approaches to the challenges discussed in this paper, including exploration and scaling up in applications of information and communication technologies to mitigate disparities in access. , , Access to surgical treatment for cancer is a public health commodity much like several that have been previously recognized and tackled, many with achievement of significant success. There is an opportunity to learn from past successes and to approach the problem of inadequate access to surgical treatment of cancer with techniques analogous to those taken to improve global access to various vaccines, mosquito nets, and antiretroviral drugs for HIV. , The history of HIV initiatives in LMICs is instructive; critics said that achieving global access to HIV therapies would be too complicated and too expensive, but programs for this have been quite effective. , Successes will be achieved by breaking the issue into component parts and taking systems-based approaches to addressing them. Successes will also be achieved by leveraging the political will and resources of multiple stakeholders. There is a wealth of opportunity to be part of these efforts and part of their successes. While the work of SWTC has achieved major advancements through the history of modern oncology, the global landscape of surgical oncology is fraught with deficiency and disparity, with critical needs for capacity building and sustainable development. The worldwide burden of neoplastic disease is increasing, and capacity for cancer control requires augmentation and attention to correcting disparity in every country on the planet. While this is true for HICs, LMICs represent the majority of both the current unmet need and the expected future increase in need. Opportunities abound for surgeons to engage in combating disparities on a global scale, as well as in their own local and national contexts. SWTC around the world are needed—both as individuals and through their collective participation in professional societies and other entities, institutions, and alliances—to engage collaboratively as leaders in advocacy, research, training, and sustainable development to create stronger systems for surgical treatment of cancer. These efforts must be bold and ambitious if we are to avoid a future in which deficiency and disparity are further exacerbated and entrenched.
Family medicine rotation in Botswana: experiences of fifth-year medical students in decentralized rural training sites	623edfa6-1156-4ad2-8a9d-293f00036437	10175651	Family Medicine[mh]	Medical students´ rural training is considered a solution to urban-rural health services inequalities in promoting rural retention regardless of a weak positive correlation between undergraduate exposure and future rural practice . The success of a rural rotation in enhancing teaching and learning depends on available communication technology, the longitudinal nature of the rotation, the focus of the school curriculum on primary care, the use of a decentralized training platform, and the ability to respond to students´ needs . A one-year longitudinal training has been suggested as a solution to geographic disparities among health professionals . There is agreement on the need of transforming medical education through the development of relevant skills and competencies required to respond to patient and community needs in the context that strengthens the health system . To support this transformation, medical schools are critical actors for improvement in the shift of training away from the tertiary institution toward rural decentralized teaching platforms (DTPs). This promotes an understanding of the context and social accountability of students . It has been also reported that medical students exposed to holistic care and continuity of care valued and supported an early exposure to family medicine away from tertiary teaching hospital . Possibly from close to a decade of exposure to family medicine, final years (5 th years) medical students have expressed the need for early rural exposure during their training . This is despite challenges with logistical and technological support, and accommodation issues requiring improvement for a successful program . However medical students were excited about potential rural clinical exposure learning, with independent working opportunities under good supervision . There are several factors including institutional and contextual ones do influence the success of a decentralized community training program and these should be considered when planning an extension of clinical training to community settings. Previous literature has reported that Family Medicine Rotations (FMR) had a positive effect on knowledge, attitude, and some skills acquired by medical students in primary care , including a positive impact on family physicians, and patients . Research in Botswana on medical students, either explored the effects of rural exposure on students´ future choice of practice location or explored fifth-year medical students´ experience of entire rural exposure during their medical training . The specific living and learning experience of fifth-year students on FMR has not yet been reported in Botswana. Considering students´ request of introducing family medicine in the curriculum as the first rotation , and based on final year medical students need for rural training , in addition to the unique context offered by decentralized sites for learning , the University of Botswana (UB), Department of Family Medicine and Public Health (DFM&PH) started training with decentralized sites away from the central campus. The early rural exposure of 32 weeks spread over the years of study was adopted. Sixteen weeks were spent under public health during their first, second, and fourth years of training. The remaining 16 weeks were spent in FMR, in their third and fifth year for eight weeks every time they were attached . Sites living experiences and learning were expected to happen through undifferentiated patients´ biopsychosocial approach. Knowing students´ experiences of FMR in Botswana would be helpful for future training adjustments. Therefore, this study intended to explore fifth-year medical students´ experiences of FMR, enablers and barriers to learning, for improvement. Design : an exploratory qualitative study using a focus FGD to collect data was adopted. Eleven students participated in the FGD. Interacting with students in a FGD was viewed as an appropriate data collection approach since students were allowed to express their views on FMR living and learning experiences and provide suggestions for improvement. Setting and study population : Maun and Mahalapye training sites were involved in the study since they hosted third- and fifth-year students. These sites operate in district hospitals which have 260 beds (Maun) and 270 beds (Mahalapye). Internal medicine, paediatrics, obstetrics-gynaecology, surgery/orthopedics, and other services like dermatology, ophthalmology, psychiatry, dental clinic, Ear-Nose-Throat (ENT), including a specialized orthopaedic service in Mahalapye, and oncology in Maun, are offered. Medical students rotated equally through the two sites. They spent eight weeks on the FMR site whenever deployed. Onsite students were attached to “four main clinical services (medicine, surgery/orthopaedics, paediatrics, obstetrics/ gynaecology)” and went to the clinic for outpatient exposure. Students were supervised and assessed by faculty supported by the district hospital staff and residents in family medicine (postgraduate students in family medicine). Expected teaching and learning were to happen during tutorials, Bedside teaching (ward round and calls), observed mini clinical examination exercise (Mini-Cex), outpatient consultations, cases presentations (three stages assessment and management plan format), directly observed clinical procedural skills (DOCPS), and participation to different weekly problem-based learning (PBL) sessions. Typically, each site received five groups of mixed (third- and fifth- years) medical students for two months starting from August of a year to May of the following year. Fifth-year medical students constituted the study population. Half of the enrolled eligible 39 fifth-year students were expected to rotate in either Maun or Mahalapye site. Recruitment and sample size : students were verbally informed throughout the year by the site´s coordinators about the upcoming study at the end of the 2016 academic year. The study was conducted in June 2016, after the last group of students completed the FMR. Two representatives of students from the two streams (Maun and Mahalapye) volunteered to facilitate the recruitment of participants. They were in touch with the main researcher (Maun site manager) who reminded them when needed, to contact and invite colleagues to participate in the study in Gaborone. Follow-up calls were done to assure that all fifth-year students were back in Gaborone and were reminded about the FGD. To reduce participants´ bias, none of the site managers attended the FGD. We purposefully selected students who rotated twice in the same site and then those who rotated once in the two different sites to allow for variability of experiences on the two sites. All fifth years were eligible and invited to participate in the study. Those who declined participation were excluded. Eleven candidates of different expected categories out of the 39 fifth years responded to the FGD in June 2016. One FGD was organized with present students. Data collection : the research assistant (MPH holder) UB employee, experienced in the qualitative study and data collection process, never involved with these students, facilitated the FGD in Gaborone. The FGD lasted for about two hours. Open-ended questions were used with the objectives of exploring fifth years´ experience of FMR, enablers, and barriers to learning, and suggestions for improvement as variables of interest. Probing, clarification, and reflective summary were done, while approval or disapproval with appropriate correction of summarized interaction content by participants to FGD for transactional credibility was done as member check. Field notes were used to complement the FGD transcript for triangulation. The discussion was done in English and the process was audiotaped followed by a verbatim transcription. Nicknamed participants as black (1), white (2), red (3), grey (4), yellow (5), blue (6), purple (7), brown (8), pink (9), orange (10) and green (11), were referred to in the result section as P1 to P11. Only clearly expressed quotes were used. Data analysis : thematic analysis was employed. Researchers familiarized themselves with the transcripts, separately coded and categorized responses which were then organized into emerging themes, according to the framework approach . Different codes, categories, and themes between researchers were harmonized for the final write-up. Atlas ti Version 7.5.18 was used to organize and capture the findings. Transferability of findings and dependability of the study were assured by setting and study processes description. Ethical considerations : permission to conduct this study was granted by the MOH / UB Ethical Review Board (HPDME: 13/18/1 Vol. X (297). To assure confidentiality, participants´ details were not included in the report. Audio records and transcripts, notes were kept safe by the principal investigator in a safe lockable cabinet. Both researchers involved in the analysis of the data had access to them. Participants signed consents before the FGD. Participants characteristics : eleven fifth-year medical students of median age 24(7) years, of which seven (63.6%) were female, participated. About half (n=6.54%,) of participants rotated once in the two training sites . Emerging themes and categories : diverse FMR experiences, inconsistency of activities and different learnings between FMR training sites, challenges, and barriers to learning during FMR, enablers to learning during FMR, and recommendations for improvement of FMR emerged . Theme 1: diverse experiences of FMR From beneficial, useful, and overwhelming to relaxing or inadequate : the individual medical student had a unique experience with FMR. It was experienced as an opportunity for revision for some. Others considered it as a break from the tension at Princess Marina Hospital (PMH) in Gaborone, contrasting with the overwhelming weekly topics experienced by others during FMR. “I was able to revise …that was the good thing about it” (P7); “We have three topics per week and one topic can take a week on itself. So, we felt overloaded, I mean three major topics per week, there is too much to cover(P11);” “You get two months away from Marina, away from Gaborone, it is just a break” (P9). Diverse experiences were summarised as: “Necessary” (P11); “it was helpful” (P10); “ amazing ” (P3); “ beneficial ”(P6); “ Crucial for skills ” (P4); “ Room for improvement ” (P1); “ It is good but could be better ” (P8); “ different from other rotation ” (P9); “ relaxing ” (P7); “ Lacking ” (P2); “ good but inadequate ” (P5). Theme 2: inconsistency of activities and learnings between sites Medical students reported inconsistency in administrative issues, supervision, teaching, and overall learning between the two sites: Different administrative standards between the two training sites : students felt that there was a double standard across a variety of administrative accountability issues, between the sites with an absence of uniformity in teaching activities resulting in a comparison of the two sites on good or bad doing: “in Mahalapye it is not compulsory for students to come and spend a night in the hospital while in Maun it is. And it is not the only thing even in terms of how long you should be in the hospital, in Maun and Mahalapye it is completely different. When I speak to my colleague in Mahalapye they are missing days, and they are leaving early. “We are not doing the same thing” (P11) . Different supervision experiences between the two sites : there was either satisfaction or dissatisfaction or else an opposing opinion regarding the type of supervision received in a training site. Family physicians´ and residents ´ involvement were different in the two sites including the inconsistency of supervision within a given site when students compared their experiences during the third and fifth year of attachment in the same site with opposing opinion on the on-site quality of supervision. “What was surprising is that in Maun the specialist goes with you guys to the clinic that was in the third year when I was there, in Mahalapye you end up transporting yourself, they through you with residents ” (P1). “ Another issue that was good about Mahalapye because I was there that their residents are very helpful even if our course coordinator is not, they were able to do with us the other activities ” (P2); “ In Maun, they [residents] are difficult to find, you only find them when you are doing PBL and maybe when you are on call a few residents were available ” (P7). “ Yes, so we had constant supervision when we were in Mahalapye; There was always somebody to guide us. In Maun once we departed from the clinic, we were in the ward with the doctor so there was very limited supervision ”(P5); “ I want to go back to what (P5) said earlier on the contact that they had with the supervisor in Mahalapye, that they had good contact with a supervisor in Mahalapye, in the third year she felt that she was learning a lot, but that was not always the case when I was doing my fifth year in Mahalapye ” (P8). Different topics and teaching methods with influence on student performance : students noted a discrepancy in teaching methods and choice of topics with potential impact on their performance during assessment which would favor the group that was taught a certain skill and surprise those who were not taught. “ Another thing to point out is that in the third year I went to Maun while in the fifth year I went to Mahalapye; well, I appreciated a bit of discrepancy in two places, I mean how they do things how they do their clinic is different. Something worth being exposed to in Mahalapye may not have happened in Maun and may be unfair because you can get a question in an exam where somebody said we were taught this in Mahalapye while someone else says ‘Nya’ [no] we were not taught that in Maun ” (P1). “ I think it will be beneficial if they teach us how we are supposed to do a blood pressure measurement, how to do a knee examination so that when we get to the exam, we do not get surprised” (P7). Different learning between Mahalapye and Maun : the group was hesitant to express their compared learning sites experiences, however, one of the students, expressed satisfaction with Maun training, while another suggested that Mahalapye was the best. “Ok I need to say that I was in Maun for family medicine rotation both third and fifth year, I enjoyed myself there ” (P11); “ Mahalapye, Mahalapye is the best” (P1). Theme 3: challenges or barriers to learning during FMR The number of onsite staff, starting family medicine as the first clinical rotation, and issues around accommodation, internet, and administrative communication were considered obstacles to learning in FMR. Insufficient onsite staff : the inadequate number of Family physicians and other specialists was considered a barrier to the learning process. “I think a very important point is the lack of lecturers. You will see that in the ward [like] internal medicine especially, the specialist only comes on Monday, the rest of the weekday you are doing round with interns and medical officers, and I feel that they cannot adequately impact some information ” (P 11). Starting clinical rotation with family medicine : starting clinical rotation with family medicine as a third-year medical student was challenging due to the broad field that is family medicine. “[…] You have never done any clinical experience, and you get there you are thrown into everything, and you do not remember most of the thing and it is very difficult if you are doing it as the first one especially if you are a third year ”(P2). Accommodation, communication issues, and poor home internet : distant accommodation from the training site and unclear communication channel with the owner of the house, electricity and water quality in Maun, and poor home internet were seen as challenging environments for learning during FMR. “In Mahalapye the place we were staying was just too far we had a problem with transport when you go there and even when you come back, it limits me in terms of if I want to do more and get more exposure, so I need to go back. The problem is that if we had a closer accommodation ” (P3). “ In Maun, it is the same thing, the place where we stayed was far away from the hospital […] it is very difficult to get transport at night and late in the evening ” (P 10). “ Any time we wanted to contact our landlord it did not happen, and we did not have the contact for the administrator ” (P6). “ We have a horrible water situation in Maun, the water is brown, and during our third year, we had to draw water from the hospital ” (P10). “ The Wi-Fi is in the house but most of the time it is very slow so in Maun we could not get to do our PBL in the room. We always did that in the hospital ” (P10). Theme 4: enablers of learning during FMR Proximity to faculty, and residents, and being part of the teamwork, mentorship, and library internet access : while residents in Mahalapye were more available compared to those in Maun, working closely with faculty and residents enabled learning. The family medicine teamwork, the mentorship initiative, and library Wi-Fi internet access promoted learning. “Mahalapye residents were available while in Maun there were not always around ” (P7). “ Working closer to a university of Botswana (UB) person, resident, and being part of the teamwork “(P5). “ I think the teamwork was appreciated [both sites], because in places like PMH the teamwork between the nurses, doctors, and students was not the same, in family medicine you do a lot of skills, you feel motivated someone doing cannula someone else doing something else and you feel encouraged ”, “ Mentorship program to help third years is a god enabler ” (P10).” The access to Wi-Fi. Library was an enabler ” (P5). Theme 5: recommendations for improvement For a better FMR experience, an individual medical student (P) and a group (G) recommended the following: 1) employment of more staff trained in family medicine: “ more staff, more family medicine specialists ” (G); 2) joint third- and fifth-year group allocation on sites to allow close guidance and student interactions. “I am saying that third years and fifth years should be always together, the third year should have fifth years, and the fifth year should have heard third years” (P11); 3)more involvement of residents in teaching and supervision activities. “ The residents should be more involved ” (P1); 4) clear details schedule of site activities: “ like a daily schedule, if I am not having a lecture, what should I be doing at a given time ” (P11). 5) consistency in teaching activities supported by the commonly used document in the two sites: “ yes I was saying consistency, also we also need that, but that can also be covered when we have clear objectives that are documented and expanded and both Mahalapye and Maun have those documents they know what they should follow then it will be easier to have consistency ” (P2); 6) closer to training site accommodation needed: “ Accommodation closer to the hospital” (G). From beneficial, useful, and overwhelming to relaxing or inadequate : the individual medical student had a unique experience with FMR. It was experienced as an opportunity for revision for some. Others considered it as a break from the tension at Princess Marina Hospital (PMH) in Gaborone, contrasting with the overwhelming weekly topics experienced by others during FMR. “I was able to revise …that was the good thing about it” (P7); “We have three topics per week and one topic can take a week on itself. So, we felt overloaded, I mean three major topics per week, there is too much to cover(P11);” “You get two months away from Marina, away from Gaborone, it is just a break” (P9). Diverse experiences were summarised as: “Necessary” (P11); “it was helpful” (P10); “ amazing ” (P3); “ beneficial ”(P6); “ Crucial for skills ” (P4); “ Room for improvement ” (P1); “ It is good but could be better ” (P8); “ different from other rotation ” (P9); “ relaxing ” (P7); “ Lacking ” (P2); “ good but inadequate ” (P5). Medical students reported inconsistency in administrative issues, supervision, teaching, and overall learning between the two sites: Different administrative standards between the two training sites : students felt that there was a double standard across a variety of administrative accountability issues, between the sites with an absence of uniformity in teaching activities resulting in a comparison of the two sites on good or bad doing: “in Mahalapye it is not compulsory for students to come and spend a night in the hospital while in Maun it is. And it is not the only thing even in terms of how long you should be in the hospital, in Maun and Mahalapye it is completely different. When I speak to my colleague in Mahalapye they are missing days, and they are leaving early. “We are not doing the same thing” (P11) . Different supervision experiences between the two sites : there was either satisfaction or dissatisfaction or else an opposing opinion regarding the type of supervision received in a training site. Family physicians´ and residents ´ involvement were different in the two sites including the inconsistency of supervision within a given site when students compared their experiences during the third and fifth year of attachment in the same site with opposing opinion on the on-site quality of supervision. “What was surprising is that in Maun the specialist goes with you guys to the clinic that was in the third year when I was there, in Mahalapye you end up transporting yourself, they through you with residents ” (P1). “ Another issue that was good about Mahalapye because I was there that their residents are very helpful even if our course coordinator is not, they were able to do with us the other activities ” (P2); “ In Maun, they [residents] are difficult to find, you only find them when you are doing PBL and maybe when you are on call a few residents were available ” (P7). “ Yes, so we had constant supervision when we were in Mahalapye; There was always somebody to guide us. In Maun once we departed from the clinic, we were in the ward with the doctor so there was very limited supervision ”(P5); “ I want to go back to what (P5) said earlier on the contact that they had with the supervisor in Mahalapye, that they had good contact with a supervisor in Mahalapye, in the third year she felt that she was learning a lot, but that was not always the case when I was doing my fifth year in Mahalapye ” (P8). Different topics and teaching methods with influence on student performance : students noted a discrepancy in teaching methods and choice of topics with potential impact on their performance during assessment which would favor the group that was taught a certain skill and surprise those who were not taught. “ Another thing to point out is that in the third year I went to Maun while in the fifth year I went to Mahalapye; well, I appreciated a bit of discrepancy in two places, I mean how they do things how they do their clinic is different. Something worth being exposed to in Mahalapye may not have happened in Maun and may be unfair because you can get a question in an exam where somebody said we were taught this in Mahalapye while someone else says ‘Nya’ [no] we were not taught that in Maun ” (P1). “ I think it will be beneficial if they teach us how we are supposed to do a blood pressure measurement, how to do a knee examination so that when we get to the exam, we do not get surprised” (P7). Different learning between Mahalapye and Maun : the group was hesitant to express their compared learning sites experiences, however, one of the students, expressed satisfaction with Maun training, while another suggested that Mahalapye was the best. “Ok I need to say that I was in Maun for family medicine rotation both third and fifth year, I enjoyed myself there ” (P11); “ Mahalapye, Mahalapye is the best” (P1). The number of onsite staff, starting family medicine as the first clinical rotation, and issues around accommodation, internet, and administrative communication were considered obstacles to learning in FMR. Insufficient onsite staff : the inadequate number of Family physicians and other specialists was considered a barrier to the learning process. “I think a very important point is the lack of lecturers. You will see that in the ward [like] internal medicine especially, the specialist only comes on Monday, the rest of the weekday you are doing round with interns and medical officers, and I feel that they cannot adequately impact some information ” (P 11). Starting clinical rotation with family medicine : starting clinical rotation with family medicine as a third-year medical student was challenging due to the broad field that is family medicine. “[…] You have never done any clinical experience, and you get there you are thrown into everything, and you do not remember most of the thing and it is very difficult if you are doing it as the first one especially if you are a third year ”(P2). Accommodation, communication issues, and poor home internet : distant accommodation from the training site and unclear communication channel with the owner of the house, electricity and water quality in Maun, and poor home internet were seen as challenging environments for learning during FMR. “In Mahalapye the place we were staying was just too far we had a problem with transport when you go there and even when you come back, it limits me in terms of if I want to do more and get more exposure, so I need to go back. The problem is that if we had a closer accommodation ” (P3). “ In Maun, it is the same thing, the place where we stayed was far away from the hospital […] it is very difficult to get transport at night and late in the evening ” (P 10). “ Any time we wanted to contact our landlord it did not happen, and we did not have the contact for the administrator ” (P6). “ We have a horrible water situation in Maun, the water is brown, and during our third year, we had to draw water from the hospital ” (P10). “ The Wi-Fi is in the house but most of the time it is very slow so in Maun we could not get to do our PBL in the room. We always did that in the hospital ” (P10). Proximity to faculty, and residents, and being part of the teamwork, mentorship, and library internet access : while residents in Mahalapye were more available compared to those in Maun, working closely with faculty and residents enabled learning. The family medicine teamwork, the mentorship initiative, and library Wi-Fi internet access promoted learning. “Mahalapye residents were available while in Maun there were not always around ” (P7). “ Working closer to a university of Botswana (UB) person, resident, and being part of the teamwork “(P5). “ I think the teamwork was appreciated [both sites], because in places like PMH the teamwork between the nurses, doctors, and students was not the same, in family medicine you do a lot of skills, you feel motivated someone doing cannula someone else doing something else and you feel encouraged ”, “ Mentorship program to help third years is a god enabler ” (P10).” The access to Wi-Fi. Library was an enabler ” (P5). For a better FMR experience, an individual medical student (P) and a group (G) recommended the following: 1) employment of more staff trained in family medicine: “ more staff, more family medicine specialists ” (G); 2) joint third- and fifth-year group allocation on sites to allow close guidance and student interactions. “I am saying that third years and fifth years should be always together, the third year should have fifth years, and the fifth year should have heard third years” (P11); 3)more involvement of residents in teaching and supervision activities. “ The residents should be more involved ” (P1); 4) clear details schedule of site activities: “ like a daily schedule, if I am not having a lecture, what should I be doing at a given time ” (P11). 5) consistency in teaching activities supported by the commonly used document in the two sites: “ yes I was saying consistency, also we also need that, but that can also be covered when we have clear objectives that are documented and expanded and both Mahalapye and Maun have those documents they know what they should follow then it will be easier to have consistency ” (P2); 6) closer to training site accommodation needed: “ Accommodation closer to the hospital” (G). This study intended to explore fifth-year medical students´ experiences of FMR. Experiences varied among participants. For some, it was different from other rotations, helpful, necessary, beneficial, crucial for skills, amazing, relaxing, and contrasting with it being inadequate requiring improvement for other participants. These diverse experiences from good to frustrating were also reported in South African settings . In a study on the rural experience of medical students conducted in Botswana two years before this one (2014), a lack of teaching during FMR was reported, implying that it was not beneficial . This negative perception is to be considered in a wide range of experiences which predominantly was positive in this study. The onsite teaching has since been introduced in the form of chosen scheduled tutorials and times of activities shared with students at each beginning of FMR for the intended eight-week periods. The positive perception expressed of FMR in this study is not an isolated perception for Botswana. FMR was reported as a good or positive experience by medical students elsewhere . It was also anticipatively considered as a positive experience by medical students based on presented expected training outcomes , hence should stand as an important component of the training of future doctors despite some negative experiences requiring improvement. Perceived inconsistency on administrative stands, training activities as well as consequent different learning outcomes between the two training sites of FMR were also reported. Although not in terms of inconsistency between the two sites, rostering of supervisors within a training site was reported in South Africa . This has been implemented by assigning a facilitator for activities as part of the eight weeks of scheduling. The consideration of the consensus key elements for decentralized training , is crucial in the development of new sites. The decentralized nature of the DFM in Botswana with two distant training sites away from the headquarter could be mainly responsible for these findings. Additionally, the discrepancy between sites in decentralized training is inherent to its nature since learning in any site depends on site dynamics, site leadership, and the type of interaction between students, lecturers, health professionals, and the community in a particular setting . These interactions and the rural setting of FMR are crucial in the transformation of medical education with the potential for the production of graduates suitable for the setting and community needs , toward which the UB is aiming. However, efforts to minimize inconsistency between the two sites were done by providing a similar schedule of activities, harmonizing onsite tutorial topics, stands on administrative arising issues, and having regular exchanges between site managers on teaching and topics to be considered. Despite the overall positive perception of FMR by participants, they reported a range of challenges due to self, logistical support, and insufficient onsite staff, including issues related to the site environment and accommodation. Starting rotation with FMR was challenging compared to doing it following other clinical rotations especially when students were in their third year of training. Poor homes Wi-Fi connectivity, including the observed insufficient onsite staff member stretched and providing limited supervision were part of rural training challenges. This limited supervision is similar to that reported in a previous study on rural experiences of medical students who happened to have spent half of their rural exposure in FMR. Inadequate staff for adequate supervision was also reported somewhere else during FMR . Onsite staff number should therefore be increased. Staff shortage has not yet been fully addressed, however alternative solution from students´ recommendations like more resident involvement in teaching and supervision has been implemented. Grouping third-and fifth-year students in interactive groups to guide the first steps of third-years, supplementing supervision deficit, and promoting onsite team build-up, peer interaction, and integration to district teamwork allowing to learn the roles of each professional actor were also implemented. This development will, however, need quality evaluation in the future for a meaningful good experience resulting in the rural choice of practice outcome , while the staffing situation in hosting training sites should be considered seriously. Distant accommodation at the training site and poor quality of water were also challenging to learning. These were also experienced in terms of logistics, technology, and accommodation issues , needing attention for optimization of student learning experiences. These challenges are however part of the complex interactions where students are to be trained and which are sites specific factors , and responsible for either individual or group different experiences of FMR. Reliable onsite internet for better connectivity has been installed, and students have been relocated to accommodation near training sites. Interaction with lecturers and residents was viewed as an enabler of learning. However, the implication of district hospital professionals in teaching should be encouraged to promote multidisciplinary learning expressed elsewhere during FMR . The role of WIFI was also acknowledged by these authors as being an enabler of learning. The important role of recommended onsite scheduling and site supervisors´ assignment was also reported elsewhere and considered crucial for the resolution of site inconsistency reported in this study. Finally, the distance between training sites and medical students´ residences was addressed since accommodation has bearing on training . Limitations: study findings are not generalizable to another cohort of students in UB or other countries. Findings are specific to the period of the study as current perceptions may have evolved by ongoing changes to address the issue reported before this write-up. Self-identified participants may have either positive or negative biases to current findings. Lastly current findings may not be exhaustive as a second FGD would have allowed the identification of potential new themes if any, in terms of saturation . Final-year medical students reported a positive perception of their FMR. The decentralized nature of FMR training sites and the reported discrepancy between sites require attention to address the factors raised by the students. The recruitment of more staff per site will certainly contribute to a better FMR experience for students at UB. What is known about this topic Rural training of medical students has the potential to promote a future choice of rural setting practice; Family medicine rural training of medical students is a skill learning, holistic approach training opportunity desirable to start early in the training; Decentralized training sites have specific sites interconnected dynamics that influence different students learning . What this study adds Medical student training in decentralized multicenter rural family medicine sites is prone to inconsistency in teaching and supervision within and between distant training sites; There is a need for manpower and synchronized teaching activities planning to minimize inconsistency between rural training sites for a good rural experience of family medicine exposure . Rural training of medical students has the potential to promote a future choice of rural setting practice; Family medicine rural training of medical students is a skill learning, holistic approach training opportunity desirable to start early in the training; Decentralized training sites have specific sites interconnected dynamics that influence different students learning . Medical student training in decentralized multicenter rural family medicine sites is prone to inconsistency in teaching and supervision within and between distant training sites; There is a need for manpower and synchronized teaching activities planning to minimize inconsistency between rural training sites for a good rural experience of family medicine exposure .
Nationwide survey of internal medicine hospitalists in Korea: motivation and sustainability of a hospitalist career	e11976ee-a9aa-4a35-a90f-eeb372cd50bd	10175863	Internal Medicine[mh]	In the United States, hospitalists were first introduced in the 1990s to address the efficiency and safety issues of inpatient care . The hospitalist system was already known to improve indicators such as duration of hospital stay, medical costs, patient satisfaction, and even mortality, and has become a larger field than any other division of internal medicine . Overall, the number of physicians caring for inpatients shrank significantly as a consequence of shortened residency durations from 4 years to 3 years in internal medicine and general surgery, as well as the legislation of resident work hour restriction. To address these issues, the need for the hospitalist system was raised, and in November 2015, a hospitalist pilot program began . Studies on the pilot program in Korea revealed that patients and nursing staff were highly satisfied with the hospitalist system . In addition to the pilot study, studies based on data of Korean hospitals showed that the hospitalist system effectively reduces the duration of stay in the emergency room and that of multimorbid patients with pneumonia or urinary tract infections . Differences in clinical results according to the working hours of hospitalists have also been revealed . The hospitalist pilot program ended in January 2021, and a management fee for hospitalist service was established. In other words, the hospitalist system officially took its first official step in Korea. By September 2021, 276 hospitalists were registered nationwide, and this number is expected to increase. However, the number of hospitalists required nationwide was estimated between 2,000 to 6,000, and the number of hospitalists required remains outmatched . To expand the hospitalist system, it was necessary to collect and share information on hospitalists’ working conditions. Thus, countrywide surveys were conducted among currently employed hospitalists on behalf of the Korean Society of Hospital Medicine (KSHM). In January 2020 and February 2022, before and after the establishment of the hospitalist fee system respectively, cross-sectional online surveys were conducted among internal medicine board-certified hospitalists based on the address book of the KSHM. KSHM is an organization established for research and educational activities in the field of hospital medicine with the support of the Korean Association of Internal Medicine. When setting up a new hospitalist model, benchmarking from an already operating hospitalist model is necessary. Most of these processes have been conducted through KSHM members. Membership in KSHM is not compulsory for hospitalists, but the number of members is increasing when fulfilling the aforementioned roles. These surveys were conducted to collect opinions on hospitalist policies among members of the KSHM and to make suggestions to the relevant ministries. At the time of the surveys, only the members working as hospitalists were sent the questionnaire. The surveys included a variety of information, such as age, sex, work experience, hospital location, working hours, the presence of night shifts, contract terms, job satisfaction, and future plans. The results of the surveys were analyzed and compared according to survey year and willingness to continue a hospitalist career. Pearson’s chi-square test or Fisher’s exact test was used to compare categorical variables, and the Mann-Whitney U test was used to compare all continuous variables that did not follow normal distribution. Backward stepwise regression analysis with a p value threshold of 0.20 was used to identify the predictive factors of willingness to continue a hospitalist career. For statistical analysis, IBM SPSS Statistics for Windows (version 27.0; IBM Corp., Armonk, NY, USA) was used, and p values less than 0.05 were deemed significant. This study was conducted with the information shared through anonymized online surveys. It did not include information that can identify research subjects, and we did not collect any sensitive information. This study was submitted to the Institutional Review Board of Seoul National University Hospital for review and exempted from approval (E-2209-022-1354). Basic information In 2020 and 2022, 59 and 64 currently working hospitalists, respectively, responded to the survey. The respondents’ age and sex ratios were similar between the two surveys. The length of hospitalist career was significantly longer in the 2022 survey (mean, 2.9 years vs. 2.2 years; p = 0.038). The percentage of people working in the Seoul metropolitan area (including Gyeonggi-do and Incheon) was approximately 75% in both surveys . Some changes were reported in terms of previous careers and motives for becoming hospitalists between the two surveys. The most common occupation, prior to working as a hospitalist, was physicians employed by non-teaching hospitals (30.5%), followed by clinical fellowship training (27.1%) and resident training (16.9%) in the 2020 survey. However, clinical fellowships were the highest previous occupations (29.7%), followed by resident training (23.4%), and specialist staff at teaching hospitals (20.3%) in the 2022 survey. In the 2020 survey, majority of the respondents cited worklife balance as the main motive for becoming a hospitalist (78.0%), followed by social needs for hospitalists (49.2%). The 2022 survey also showed that the majority of the participants cited work-life balance (75.0%), however, only 7.8% cited social needs for hospitalists as the main motive for becoming a hospitalist. The percentage of respondents citing financial benefits as a motive for becoming hospitalists was significantly higher in the 2022 survey than in the 2020 survey (34.4% vs. 10.2%; p = 0.001) . Hospitalist operating model There was a difference in the affiliation of respondents between the two surveys. The proportion of respondents belonging to independent hospitalist departments such as the division of integrated medicine or hospital medicine center increased from 47.5% in 2020 to 60.9% in 2022. As primary care physicians, 49.2% and 64.1% of the respondents had full clinical autonomy in the 2020 and 2022 surveys, respectively. Less than 20% depended on the prescriptions of the existing specialist staff as trainees. Most respondents worked solely during the daytime on weekdays, and the proportion of hospitalists practicing 24/7 was similar in both surveys. The mean working hours of the respondents were 44.2 hours per week in 2020 and 43.0 hours per week in 2022. The majority of the respondents treated 16–25 inpatients daily. The proportion of respondents responsible for more than 25 inpatients decreased from 11.9% in 2020 to 3.1% in 2022 . Contract conditions, job satisfaction, and future plans More than 60% of the respondents were 1-year contract workers in both the surveys. The mean annual salary of the respondents was 182.9 million South Korean Won (KRW) in 2022, which was significantly higher than the 163.0 million KRW in 2020 ( p = 0.006). The percentage of respondents eligible for salary increases and promotions at the time of contract renewal was significantly higher in the 2022 survey than in the 2020 survey (40.6% vs. 20.3%, p = 0.015; 21.9% vs. 5.1%; p = 0.007, respectively). There were no other significant differences in terms of compensation between the two surveys . In the 2022 survey, the percentage of respondents who played the role of clinician educators was 82.8%, which was significantly higher than 62.7% in the 2020 survey ( p = 0.012). More respondents in the 2022 survey (56.3%) than in the 2020 survey (40.7%) tended to be satisfied with the profession of a hospitalist . Regarding their future plans (their plans for the following year), 76.6% respondents of the 2022 survey said that they would continue to work at the same hospital and 4.7% said they would work as a hospitalist in a different hospital. In conclusion, 81.3% of the respondents in the 2022 survey were willing to continue their hospitalist jobs, which was higher than 64.4% in the 2020 survey ( p = 0.043) . Willingness to continue a hospitalist career Among the 90 respondents of both surveys who were willing to continue a hospitalist career, compared to the 33 who were not, the proportion of respondents in the Seoul metropolitan area tended to be higher, and the length of hospitalist careers was significantly longer. The percentage of respondents who answered that they had extra pay for night shifts (37.8% vs. 15.2%, p = 0.017) and private office spaces (70.0% vs. 45.5%, p = 0.012) were significantly higher in the group intending to pursue a hospitalist career. The percentage of respondents who answered that they were enrolled in teachers’ pensions (54.4% vs. 27.3%, p = 0.007), had the possibility of being appointed as professors (37.8% vs. 9.1%, p = 0.002), and were serving as clinician educators (80.0% vs. 54.5%, p = 0.005) also were significantly higher in the group that intended to continue their careers as hospitalists. Contrarily, there were no differences regarding age, sex, salary, or work schedule between the two groups . In the unadjusted model, the number of working years as a hospitalist, whether they worked at a hospital in the Seoul metropolitan area, if they received extra pay for night shifts, if they had private office space provision, if they were enrollment in teachers’ pension, their possibility of being appointed as a professor, and if they were clinician educators were found to be significant variables. After adjusting for age, sex, and all the variables showing significance in the unadjusted model, the possibility of being appointed as a professor (odds ratio, 4.00; 95% confidence interval, 1.09–14.75; p = 0.037) was an independent predictor of continuing a hospitalist career . In 2020 and 2022, 59 and 64 currently working hospitalists, respectively, responded to the survey. The respondents’ age and sex ratios were similar between the two surveys. The length of hospitalist career was significantly longer in the 2022 survey (mean, 2.9 years vs. 2.2 years; p = 0.038). The percentage of people working in the Seoul metropolitan area (including Gyeonggi-do and Incheon) was approximately 75% in both surveys . Some changes were reported in terms of previous careers and motives for becoming hospitalists between the two surveys. The most common occupation, prior to working as a hospitalist, was physicians employed by non-teaching hospitals (30.5%), followed by clinical fellowship training (27.1%) and resident training (16.9%) in the 2020 survey. However, clinical fellowships were the highest previous occupations (29.7%), followed by resident training (23.4%), and specialist staff at teaching hospitals (20.3%) in the 2022 survey. In the 2020 survey, majority of the respondents cited worklife balance as the main motive for becoming a hospitalist (78.0%), followed by social needs for hospitalists (49.2%). The 2022 survey also showed that the majority of the participants cited work-life balance (75.0%), however, only 7.8% cited social needs for hospitalists as the main motive for becoming a hospitalist. The percentage of respondents citing financial benefits as a motive for becoming hospitalists was significantly higher in the 2022 survey than in the 2020 survey (34.4% vs. 10.2%; p = 0.001) . There was a difference in the affiliation of respondents between the two surveys. The proportion of respondents belonging to independent hospitalist departments such as the division of integrated medicine or hospital medicine center increased from 47.5% in 2020 to 60.9% in 2022. As primary care physicians, 49.2% and 64.1% of the respondents had full clinical autonomy in the 2020 and 2022 surveys, respectively. Less than 20% depended on the prescriptions of the existing specialist staff as trainees. Most respondents worked solely during the daytime on weekdays, and the proportion of hospitalists practicing 24/7 was similar in both surveys. The mean working hours of the respondents were 44.2 hours per week in 2020 and 43.0 hours per week in 2022. The majority of the respondents treated 16–25 inpatients daily. The proportion of respondents responsible for more than 25 inpatients decreased from 11.9% in 2020 to 3.1% in 2022 . More than 60% of the respondents were 1-year contract workers in both the surveys. The mean annual salary of the respondents was 182.9 million South Korean Won (KRW) in 2022, which was significantly higher than the 163.0 million KRW in 2020 ( p = 0.006). The percentage of respondents eligible for salary increases and promotions at the time of contract renewal was significantly higher in the 2022 survey than in the 2020 survey (40.6% vs. 20.3%, p = 0.015; 21.9% vs. 5.1%; p = 0.007, respectively). There were no other significant differences in terms of compensation between the two surveys . In the 2022 survey, the percentage of respondents who played the role of clinician educators was 82.8%, which was significantly higher than 62.7% in the 2020 survey ( p = 0.012). More respondents in the 2022 survey (56.3%) than in the 2020 survey (40.7%) tended to be satisfied with the profession of a hospitalist . Regarding their future plans (their plans for the following year), 76.6% respondents of the 2022 survey said that they would continue to work at the same hospital and 4.7% said they would work as a hospitalist in a different hospital. In conclusion, 81.3% of the respondents in the 2022 survey were willing to continue their hospitalist jobs, which was higher than 64.4% in the 2020 survey ( p = 0.043) . Among the 90 respondents of both surveys who were willing to continue a hospitalist career, compared to the 33 who were not, the proportion of respondents in the Seoul metropolitan area tended to be higher, and the length of hospitalist careers was significantly longer. The percentage of respondents who answered that they had extra pay for night shifts (37.8% vs. 15.2%, p = 0.017) and private office spaces (70.0% vs. 45.5%, p = 0.012) were significantly higher in the group intending to pursue a hospitalist career. The percentage of respondents who answered that they were enrolled in teachers’ pensions (54.4% vs. 27.3%, p = 0.007), had the possibility of being appointed as professors (37.8% vs. 9.1%, p = 0.002), and were serving as clinician educators (80.0% vs. 54.5%, p = 0.005) also were significantly higher in the group that intended to continue their careers as hospitalists. Contrarily, there were no differences regarding age, sex, salary, or work schedule between the two groups . In the unadjusted model, the number of working years as a hospitalist, whether they worked at a hospital in the Seoul metropolitan area, if they received extra pay for night shifts, if they had private office space provision, if they were enrollment in teachers’ pension, their possibility of being appointed as a professor, and if they were clinician educators were found to be significant variables. After adjusting for age, sex, and all the variables showing significance in the unadjusted model, the possibility of being appointed as a professor (odds ratio, 4.00; 95% confidence interval, 1.09–14.75; p = 0.037) was an independent predictor of continuing a hospitalist career . This is the first survey study conducted on hospitalists in Korea. This study included highly realistic and detailed information about hospitalist compensation and workload, similar to previous studies . Because the Korean hospitalist model still has difficulties in recruiting and retaining hospitalists, motivation for choosing a hospitalist job and predictors of hospitalist retention were analyzed. This study identified differences in responses to various questions depending on the respondent’s condition, especially the survey year; that is, whether the hospitalist fee system was settled. In the 2020 survey, before establishing a management fee system for hospitalists, many respondents had worked at a non-teaching local hospitals before choosing a hospitalist career. However, after establishing a hospitalist fee system, there was a marked change in the respondents’ work experience. More doctors chose to have a hospitalist career immediately after residency or fellowship training. It is also worth noting that the proportion of hospitalists who transferred from an academic faculty specialist position rose to 20.3% in the 2022 survey. These changes may be because the settlement of the hospitalist fee system gives applicants a sense of security in their hospitalist careers. The motives for becoming a hospitalist also differed according to the survey year. In the 2022 survey, after the establishment of a management fee system for hospitalists, 34.4% of respondents cited financial benefits, which was higher than the 10.2% in the 2020 survey. There have been significant changes in terms of monetary compensation between the two survey years. Annual salaries were significantly higher in the 2022 survey than in the 2020 survey. Additionally, the proportion of respondents who could receive a salary increase upon renewing their contract was higher in the 2022 survey. It is known that economic factors are important in the selection of medical specialty . The establishment of a fee system for hospitalists encourages hospitals to provide more financial benefits to applicants, and this seems to be a factor that attracts new hospitalists. Also, since the fee system limits working hours and the number of patients, there were a few notable differences between the two survey years. Regarding non-monetary rewards, the opposite was observed, although it was statistically non-significant. In the 2022 survey, the proportion of respondents who had multiyear contracts, academic attendance expense support, education or research fund support, and the possibility of being appointed as professors tended to be lower than in the 2020 survey. Monetary compensation was not a significant factor for the decision of continuing hospitalist career. It might be because they had started the job after accepting the salary. This study showed that the possibility of being appointed as a professor could predict long-term work as hospitalists. The conditions for a sustainable hospitalist career identified in this study might not be applicable to all hospitalists considering that monetary compensation was cited as an important motive. However, in order for hospitalists to take on an important part of the inpatient care area and to operate stably for a long time, hospitalists that commit to developing educational roles and leadership are needed . Previous reports suggested that hospitalist clinician educators were helpful not only for clinical outcomes but also for resident education and for the maintenance of a hospitalist career . Some studies suggested that hospitalists can be better educators than traditional physicians . It is also noteworthy that the proportion of respondents who had full clinical autonomy and independent hospitalist departments tended to show higher interest in the continuing their hospitalist career. In the United States, where the hospitalist system is well established, a report showed that the proportion of hospitalists with clinical autonomy reached 97.1% . The establishment of the hospitalist fee system has forced many hospitals to hire hospitalists, meanwhile financial benefits are becoming the main motivation for choosing a hospitalist career. However, with this trend, the executives of hospitals expect hospitalists to handle the maximum workload based on their pay. This perception may adversely affect the hospitalists in establishing well-functioning positions in hospitals. Hospitalists who want to work for a long time do not earn a higher salary, but rather serve educational roles and prepare for appointments as professors. From a longterm perspective, hospitalists should not be a high-paying job that takes the place of resident doctors. Instead, they must become experts in inpatient care and clinical education. Considering that most hospitalists are currently working in tertiary general hospitals in educational roles, it can be said that this may not apply to most hospitalists in the future. However, for the hospitalist model to spread to local secondary hospitals in need, it is necessary to motivate those who were committed in the early stages of the hospitalist system introduction to become pioneers. There are some limitations to this study. First, the number of participants was small. It was difficult to secure a larger number of respondents due to the small population of hospitalists in Korea and to the voluntary participation to the surveys. Thus, these results might not fully reflect the opinions of all hospitalists in Korea. Second, because the surveys were conducted with a small pool of respondents, it is possible that the same person was included in the two surveys. However, even the same person may have responded differently in the two surveys, considering that most of respondents were 1-year contract workers. It is also meaningful to show the changes before and after the establishment of the hospitalist fee system, even for the same person. Third, only internal medicine hospitalists were included in the surveys. The authors believe that there could be considerable differences in the settings and attitudes toward the hospitalist system by department. Large-scale studies including other specialty hospitalists or in-depth qualitative researches should be undertaken in future. Since the establishment of the hospitalist fee system, monetary compensation has improved for hospitalists. The possibility of being appointed as a professor was an independent predictor of continuing a hospitalist career. We hope that this study will facilitate the recruitment and retention of hospitalists, lead to the expansion of the hospitalist system, and ultimately serve as a reference for policies regarding hospitalists. 1. Hospitalists of the recent survey tend to value financial benefits. 2. Since the official establishment of hospitalist fee system, monetary compensation has improved for hospitalists. 3. The possibility of being appointed as a professor was an independent predictor of continuing a hospitalist career.
Study design considerations to assess the impact of potential	d24a2a79-2297-42bc-942d-41619304afba	10176012	Internal Medicine[mh]	STUDIES IN ALTERNATIVE PATIENT POPULATIONS DUE TO INTERACTING CONCOMITANT THERAPIES When a new oncologic drug is not expected to be cytotoxic, genotoxic, or target a specific genetic alteration found in only the intended study population, it may be reasonable to conduct FIH studies in healthy subjects or in patients who do not routinely require the interacting concomitant therapies. These data from the FIH study can then be leveraged to select doses to be conducted in a drug‐drug interaction (DDI) study. As such, DDI studies can be conducted in a separate study to help select the recommended dosing regimen of the new oncologic drug for intended patient population when administered with interacting concomitant therapies based on exposure matching. This approach was used for glasdegib, a Hedgehog pathway inhibitor, approved for the treatment of patients with newly diagnosed AML who are on nonintensive chemotherapy. In vitro studies indicated that glasdegib is metabolized by CYP3A4. Initially, dose escalation studies were conducted in patients with hematologic malignancies and solid tumors not requiring azole antifungals. These studies demonstrated a maximum tolerated dosage of 400 mg once daily (q.d.) in patients with hematological malignancies, dose proportional pharmacokinetics (PKs), and saturable pharmacodynamic (PD) activity at 100 mg q.d. A DDI study conducted to understand the effects of a strong CYP3A inhibitor on the PKs of glasdegib following a single dose of 200 mg showed that a strong CYP3A4 inhibitor (i.e., ketoconazole) increased glasdegib exposure by twofold. Given the saturable PDs observed at 100 mg, and a twofold increase in exposure with a strong CYP3A inhibitor, 100 mg q.d. was selected for the registration trial in patients with AML to provide an adequate safety margin for the increased glasdegib exposure with concomitant use of azole antifungals. STUDIES IN THE INTENDED PATIENT POPULATION WHEN THE INTERACTING CONCOMITANT THERAPIES CANNOT BE AVOIDED FIH studies that need to be conducted in the intended patient population that require interacting concomitant therapies should be conducted with a large safety margin. Although most patients enrolled into these FIH studies will be administered an interacting concomitant therapy, a cohort of patients who will not receive the interacting concomitant therapy should be enrolled for comparative purposes, as they will provide critical PK, efficacy, and safety for the indicated patient population. This approach is illustrated by the FIH study in patients with relapsed or refractory acute leukemias for SNDX‐5613, a CYP3A4 substrate. The study included one arm with patients who required azole antifungals that are CYP3A4 inhibitors and another arm with patients that did not. Dosing for each arm were started at the same dose of 113 mg Q12h. Data from the first two dose levels indicated an increase in SDNX‐5613 exposure with strong CYP3A4 inhibitors of approximately twofold, and higher rates of adverse reactions, including QT prolongation. This approach poses several challenges, particularly the possibility of underestimating the risk for potential DDI when selecting a starting dose. It is important that the starting dose administered with the interacting medication have an adequate safety margin to support its use in FIH studies. In addition, different concomitant therapies within the same class (e.g., posaconazole vs. voriconazole) or the same CYP inhibitor in different dosage forms or dosing regimen may have different magnitudes of effect on the exposure of the new oncologic drug. For example, the exposure of ibrutinib varied with different formulations and dosing regimen of posaconazole, and with voriconazole and other strong CYP3A inhibitors. As such, early DDI studies or dose escalation studies with and without interfering drugs may not provide a complete assessment of the potential interactions, but can inform on the impact of concomitant medications on the recommended phase II dose during the course of development. FIH studies in patients who will be treated with a concomitant therapy should consider the following: Assess the potential for DDIs based on in vitro metabolic activity screens and/or in vivo DDI studies. These data will inform the selection of a safe starting/dose escalation plan, and/or the choice of medications to address the non‐disease morbidity. If proceeding without a separate DDI study, a cohort of patients without interacting concomitant therapy provides a comparison to assess the potential impact of the medications on the investigational drug. An appropriate dose escalation strategy to safely identify the recommended dosage(s) for further development of the new oncologic drug when co‐administered with the interacting concomitant therapy should be implemented. A staggered dose escalation approach may be incorporated into these FIH studies, wherein the safety, PD, and PK data from the initial cohorts are utilized to inform dosing in the interacting arm and subsequent dosing cohorts in both arms. There are instances when a new oncologic drug is also a perpetrator of the DDI and potentially affect the exposure of the commonly used concomitant therapies, thus impact the safety or effectiveness of the concomitant drugs. Physiologically‐based PK (PBPK) modeling and simulations can be leveraged to understand the impact of potential DDIs and to provide dosage recommendations to be tested during clinical development. An example which highlights this approach is ivosidenib, a kinase inhibitor approved for the treatment of adult patients with AML with a susceptible IDH1 mutation. Ivosidenib exhibits nonlinear kinetics, is metabolized by CYP3A4, and induces CYP3A4. Ivosidenib is not only prone to interactions from strong CYP3A4 inhibitors and inducers, but it may in fact also decrease exposure to antifungal drugs that are CYP3A substrates and their effectiveness. The results from an in vivo DDI study showed that concurrent use of a strong CYP3A4 inhibitor (e.g., itraconazole) increased single‐dose ivosidenib exposure by 169%. This study was used as the basis for PBPK modeling and simulations to predict the effect of ivosidenib as a CYP3A inducer: multiple doses of ivosidenib could reduce the single‐dose exposure of a CYP3A4 sensitive substrate, such as midazolam by 83%, and the steady‐state exposure of itraconazole (also a CYP3A4 substrate) by 90%. The simulations also predicted an increase in the steady‐state exposure of ivosidenib by 3.8‐fold with a strong CYP3A4 inhibitor which itself is not a substrate of CYP3A4. Further, the PBPK simulations indicated a 1.9‐fold increase in ivosidenib with a moderate CYP3A4 inhibitor (e.g., fluconazole). The PBPK modeling formed the basis for the ivosidenib labeling recommendations to healthcare providers to avoid co‐administration of azole antifungals that are CYP3A4 substrates, including itraconazole and ketoconazole, as it may result in a loss of antifungal efficacy, and to reduce the dose of ivosidenib when co‐administered with a strong CYP3A inhibitor as it may increase exposure of ivosidenib and potentially increasing the severity and incidence of adverse reactions, including QT interval prolongation. During drug development, it is critically important to identify the optimal doses of the new therapeutic for patients to assure safety and efficacy. For drugs that may be impacted by their concomitant medications, nonclinical and in vitro studies are key to understanding the potential risks for DDIs with a new oncologic drug and designing FIH studies to select doses that account for these DDI issues. Strategies, such as exploring the DDIs in healthy subjects or in patients with other cancers who do not require interacting concomitant therapies, where possible, or evaluating the drug in the intended patient population with additional strategies to safely evaluate the PKs, safety, and activity of the new oncology drug, have been successfully used to identify recommended dosage(s) for further clinical development and support labeling recommendations. These approaches help minimize exposure to toxic or ineffective dosages and help identify the recommended dosage(s) for the intended population who will be receiving interacting concomitant therapies. No funding was received for this work. The authors declared no competing interests for this work. The opinions expressed in this manuscript are those of the authors and should not be interpreted as the position of the US Food and Drug Administration.
Use of Bland-Altman Analysis to Examine the Racial and Ethnic Representativeness of Study Populations in Community-Based Pediatric Health Research	c6374a3f-c99b-4c87-874f-889767ee09c5	10176118	Pediatrics[mh]	Ensuring that marginalized racial and ethnic populations are well represented in research is essential to prevent growing health inequities. In the US, the health evidence base fails to reflect the true diversity of the population. , Black, Indigenous, and Hispanic populations are systematically excluded from health research, regardless of their awareness of and willingness to participate in studies. , , Once enrolled, these groups often exhibit lower retention rates than their White peers. , , , Disproportionate enrollment and retention compounds inequities , disproportionately conferring any potential benefits of study participation. Additionally, misrepresentation impedes the external validity of the evidence base and extends to suboptimal health policies, programs, and health care services. Investigators could improve the representativeness of study populations by consistently reporting participant race and ethnicity and sample representativeness. , , , , However, the current paradigm for quantifying representativeness lacks consistency and specificity, ultimately hindering any advances in equity. At a minimum, addressing the causes of preventable racial and ethnic health disparities begins by ensuring that study samples reflect the demographics of the US. The longstanding consensus rightfully acknowledges the overrepresentation of White individuals in study populations and the underrepresentation of marginalized racial and ethnic groups. Yet specific representation patterns may be biased by the endogeneity of race and ethnicity reporting, since public disclosure of study demographics is not mandated by journals or funders; thus, investigators often forego this reporting. For example, Rees et al examined 612 pediatric clinical trials published in top medical journals between 2011 and 2020, and found that 27.7% did not report participant race or ethnicity. Based on studies that provided data on the race of participants, the authors concluded that Black participants were enrolled in a greater proportion than their share of the US population aged younger than 19 years and were therefore overrepresented. Flores et al assessed reporting of race and ethnicity in 230 US vaccine trials, and they found that all studies reported participant age and sex but only 58.3% reported race and only 34.3% reported ethnicity. These reporting discrepancies may affect our understanding of racial and ethnic study representativeness, as findings rely on investigator transparency. When study demographics are reported in peer-reviewed literature, representativeness is traditionally quantified by comparing the overall share of each racial and ethnic group enrolled in the study against their mean share of the population without accounting for age groups or locations. , , , , This approach could obscure disparities and provides little insight into recruitment procedures and the association between factors in the environment and enrollment within and across racial and ethnic groups. Overall, the health research community could benefit from a specific and consistent approach for assessing and reporting the racial and ethnic representativeness of study populations. Here, we describe a methodology for quantifying the racial and ethnic representativeness of study populations and demonstrate its utility by compiling a convenience sample of data from 7 US community-based pediatric health studies. All studies, conducted by members of our team, collected parent- and guardian-reported child race and ethnicity and recruited at public schools. Investigators allocated extensive resources a priori to identify, partner with, and recruit from schools in communities serving predominantly marginalized populations to ensure that they were well represented . Investigators aimed to enroll samples representative of each school’s diverse population. We hypothesized that there would be no differences in mean representativeness across racial and ethnic groups. Our overall objective is to contribute to a more equitable research paradigm by presenting a standardized methodology for quantifying the representativeness of study samples. The Tufts University Institutional Review Board exempted this cross-sectional study from review because it did not constitute human participant research. For the 7 pediatric research studies described herein, informed consent was obtained from the parents or caregivers of participants. The current study followed the Strengthening the Reporting of Observational Studies in Epidemiology ( STROBE ) reporting guideline. Bland-Altman Method The Bland-Altman approach is widely accepted and applied in clinical settings to analyze the agreement between 2 quantitative measures. Bland and Altman developed a graphical technique with simple calculations for validation of measurement tools and techniques against preestablished clinical gold standards to overcome limitations of the alternative, based solely on correlation and regression analyses. The Bland-Altman method quantifies agreement between the 2 measures by assessing the mean differences of each observation and estimating the limits of agreement (ie, the range in which 95% of the differences between the first and second measures fall). Bland-Altman plots have been used previously in health disparities research. Wong et al compared racial and ethnic discrepancies in pulse oximetry and arterial oxygen saturation measures. In the current study, we visualized differences between the percentage expected for representativeness and the percentage enrolled for the following racial and ethnic subgroups: Asian, Black, Hispanic or Latino, Native Hawaiian or other Pacific Islander, White, and multiple races (eAppendix 1 in ). We standardized our findings with percentage points to assess deviations from proportionate representation. Study Population and Data Source To compute the share of each group expected for representativeness (herein referred to as percentage expected ), we used the National Center for Education Statistics (NCES) Public Elementary/Secondary School Universe Survey. These data are compiled into the Common Core of Data, which provides the total race and ethnicity enrollment counts by sex and grade for all US public schools. It is an authoritative sampling frame for pediatric research and widely relied on by investigators and in population surveillance surveys. We calculated percentage expected by dividing the total number of enrolled students in each racial and ethnic group in the target grades by the number of all enrolled in those grades. By deriving percentage expected from the schools’ true and precise population parameters, it can be conceptualized as the gold standard for proportionate representativeness. , To compute the share of each group that enrolled in studies (herein referred to as percentage observed ), we pooled data from 7 community-based health studies led by members of our research team across 104 schools in 5 states, including California, Kentucky, Massachusetts, Mississippi, and South Carolina. Individual measures from children in grades 1 to 8 who enrolled in nutritional health studies were aggregated by the school where they were recruited . Prior to baseline data collection, parents or guardians were asked to report child demographics, including grade, race, and ethnicity. Child race and ethnicity data were aggregated according to NCES reporting standards (eAppendix 1 in ). In this analysis, the Asian and Native Hawaiian or other Pacific Islander subgroups were combined. We calculated percentage observed at the school level by dividing the number of enrolled children in each racial and ethnic group by the total number of study participants from each school. We identified each study school in the NCES Common Core of Data and calculated school-level differences between percentage expected and percentage observed for each race and ethnicity. Additional details are provided in eAppendix 2 in . Interpretation of Bland-Altman Analysis to Reveal Representativeness Patterns Key components of the Bland-Altman method reveal important information regarding site-level racial and ethnic representativeness. The process for quantifying representativeness is explained further in eTable 1 in . Scatterplots and Bland-Altman plots are presented in pairs, and comparisons were made separately for each racial and ethnic group. Points on graphs represent individual schools. On the scatterplots, the x-axis presents the percentage expected and the y-axis is the percentage observed. The solid line is the line of identity, and the dashed line is the best fit. Perfect overlap of the line of identity and best fit indicates proportionate representation across schools. On the Bland-Altman plots, the x-axis denotes the mean of percentage expected and percentage observed. The y-axis indicates percentage observed minus percentage expected. The shorter-dashed line shows the mean of the absolute value of the difference between the expected and observed percentages; its distance from the line of 0 on the y-axis (herein referred to as Y,0) is the overall percentage-point amount that the racial or ethnic group was represented across studies. The longer-dashed lines indicate the SDs (2) for this range; thus, wider intervals indicate greater variability in overall representativeness. Points along Y,0 are schools that were representative. Points above Y,0 are schools where each group was overrepresented, while schools below Y,0 were underrepresented. The distance between a point and Y,0 denotes the percentage-point distance from proportionate representation. Statistical Analysis Schools were the primary unit of analyses. We visually inspected scatterplots and Bland-Altman plots to quantify representation patterns. To assess the association between percentage expected and percentage observed for each racial and ethnic group, we used Spearman rank-order correlation coefficients and a 2-sided α level of .01 to establish statistical significance. All analyses were conducted from April 1 to June 15, 2022, using Stata SE software, version 17 (StataCorp LLC). The Bland-Altman approach is widely accepted and applied in clinical settings to analyze the agreement between 2 quantitative measures. Bland and Altman developed a graphical technique with simple calculations for validation of measurement tools and techniques against preestablished clinical gold standards to overcome limitations of the alternative, based solely on correlation and regression analyses. The Bland-Altman method quantifies agreement between the 2 measures by assessing the mean differences of each observation and estimating the limits of agreement (ie, the range in which 95% of the differences between the first and second measures fall). Bland-Altman plots have been used previously in health disparities research. Wong et al compared racial and ethnic discrepancies in pulse oximetry and arterial oxygen saturation measures. In the current study, we visualized differences between the percentage expected for representativeness and the percentage enrolled for the following racial and ethnic subgroups: Asian, Black, Hispanic or Latino, Native Hawaiian or other Pacific Islander, White, and multiple races (eAppendix 1 in ). We standardized our findings with percentage points to assess deviations from proportionate representation. To compute the share of each group expected for representativeness (herein referred to as percentage expected ), we used the National Center for Education Statistics (NCES) Public Elementary/Secondary School Universe Survey. These data are compiled into the Common Core of Data, which provides the total race and ethnicity enrollment counts by sex and grade for all US public schools. It is an authoritative sampling frame for pediatric research and widely relied on by investigators and in population surveillance surveys. We calculated percentage expected by dividing the total number of enrolled students in each racial and ethnic group in the target grades by the number of all enrolled in those grades. By deriving percentage expected from the schools’ true and precise population parameters, it can be conceptualized as the gold standard for proportionate representativeness. , To compute the share of each group that enrolled in studies (herein referred to as percentage observed ), we pooled data from 7 community-based health studies led by members of our research team across 104 schools in 5 states, including California, Kentucky, Massachusetts, Mississippi, and South Carolina. Individual measures from children in grades 1 to 8 who enrolled in nutritional health studies were aggregated by the school where they were recruited . Prior to baseline data collection, parents or guardians were asked to report child demographics, including grade, race, and ethnicity. Child race and ethnicity data were aggregated according to NCES reporting standards (eAppendix 1 in ). In this analysis, the Asian and Native Hawaiian or other Pacific Islander subgroups were combined. We calculated percentage observed at the school level by dividing the number of enrolled children in each racial and ethnic group by the total number of study participants from each school. We identified each study school in the NCES Common Core of Data and calculated school-level differences between percentage expected and percentage observed for each race and ethnicity. Additional details are provided in eAppendix 2 in . Key components of the Bland-Altman method reveal important information regarding site-level racial and ethnic representativeness. The process for quantifying representativeness is explained further in eTable 1 in . Scatterplots and Bland-Altman plots are presented in pairs, and comparisons were made separately for each racial and ethnic group. Points on graphs represent individual schools. On the scatterplots, the x-axis presents the percentage expected and the y-axis is the percentage observed. The solid line is the line of identity, and the dashed line is the best fit. Perfect overlap of the line of identity and best fit indicates proportionate representation across schools. On the Bland-Altman plots, the x-axis denotes the mean of percentage expected and percentage observed. The y-axis indicates percentage observed minus percentage expected. The shorter-dashed line shows the mean of the absolute value of the difference between the expected and observed percentages; its distance from the line of 0 on the y-axis (herein referred to as Y,0) is the overall percentage-point amount that the racial or ethnic group was represented across studies. The longer-dashed lines indicate the SDs (2) for this range; thus, wider intervals indicate greater variability in overall representativeness. Points along Y,0 are schools that were representative. Points above Y,0 are schools where each group was overrepresented, while schools below Y,0 were underrepresented. The distance between a point and Y,0 denotes the percentage-point distance from proportionate representation. Schools were the primary unit of analyses. We visually inspected scatterplots and Bland-Altman plots to quantify representation patterns. To assess the association between percentage expected and percentage observed for each racial and ethnic group, we used Spearman rank-order correlation coefficients and a 2-sided α level of .01 to establish statistical significance. All analyses were conducted from April 1 to June 15, 2022, using Stata SE software, version 17 (StataCorp LLC). Informed consent was obtained for 6325 participants but 518 (8.2%) were excluded from this cross-sectional analysis because the child’s race and/or ethnicity was either missing (n = 152), reported as multiracial in studies before 2009 (n = 211), or identified with the “other” race checkbox (n = 155). In total, 104 schools (N = 5807 children) were included in this analysis. Exclusion criteria with counts are provided in eTable 2 in . presents the absolute value of the difference between expected and observed by group. shows the Bland-Altman plots and scatterplots for Asian or Native Hawaiian or other Pacific Islander children and Black children. presents the findings for Hispanic children and White children. In the eFigure in , representativeness results are presented for non-Hispanic multiracial children. Asian or Native Hawaiian or other Pacific Islander children were overrepresented by 0.45 percentage points (95% CI, −7.76 to 8.66), but most schools had small population shares. The association between percentage expected and percentage observed was significant (Spearman ρ = 0.82; P < .001; A). The scatterplot’s best-fit line fell below the line of identity as the expected percentage for the Asian or Native Hawaiian or other Pacific Islander group increased, suggesting that these children were more underrepresented when they comprised larger shares of the school population. The Bland-Altman plots indicated little variation in over- and underrepresentation across schools, since the CIs were narrow relative to the other racial and ethnic groups. Black children were overrepresented by 0.12 percentage points with variation (95% CI, −15.73 to 15.96). The positive correlation between percentage expected and percentage observed for Black children was statistically significant (Spearman ρ = 0.94; P < .001; C). The clustering of points near 0 and 100 on the scatterplot indicated that many schools had populations that were majority Black students, while others comprised only a small Black student population. The best-fit line on the scatterplot fell above the line of identity at higher values, suggesting that schools where Black children comprised larger shares of the population had a greater magnitude of overrepresentation. Notably and unlike any other predominant race and ethnicity, Black children in this study were never underrepresented in schools where they comprised more than 50% of the target grade population. Hispanic children also appeared to be proportionally represented with notable variation (0 percentage-point difference [95% CI, −17.66 to 17.66]). The positive correlation between percentage expected and percentage observed among Hispanic children was statistically significant (Spearman ρ = 0.94; P < .001; ). Extreme outliers in both directions on the Bland-Altman plots differentiated Hispanic representativeness from that of other racial and ethnic populations. White children were underrepresented by 0.72 percentage points (95% CI, −23.60 to 22.16). The positive correlation between percentage expected and percentage observed for White children was statistically significant (Spearman ρ = 0.91; P < .001; ). Although the mean difference between percentage expected and percentage observed was negative, this group appeared the least affected by bias in enrollment because the points above and below 0 were evenly distributed across the x-axis. No immediate patterns emerged between the demographic share of a school and its representation of White children. In this cross-sectional study, we provided a practical approach for quantifying the representativeness of study populations by adapting the Bland-Altman method and examining the associations between the percentage of baseline study samples relative to each group’s true population parameters by race and ethnicity. The measurement of site-level representativeness with Bland-Altman plots suggested that the mean difference between percentage expected and percentage observed—information traditionally reported in studies and shown in our analysis with the y-axis line—obscures outliers in over- and underrepresentation. By comparing this y-axis line to the spread of individual points (ie, schools), specific outlier sites were identified easily. Moreover, by examining the location of points along the x-axis of Bland-Altman plots, an association was observed between a racial and ethnic group’s share of the overall target population and their representativeness at the site. Illustrating this approach in our pooled data set underscored the importance of moving beyond reporting representativeness with group means alone. We observed that overall representativeness did not appreciably vary across racial and ethnic groups, but mean values obscured substantial within-group variation. We urge investigators to begin incorporating measures of spread into their research (1) to provide a more complete estimate of representativeness and (2) to pinpoint potential barriers and facilitators to study enrollment by race and ethnicity. Future research replicating our approach could maximize the transparency of racial and ethnic reporting, investigate whether our results in a nonrandom sample of community-based studies hold across other community-based studies in the US, and elucidate the degree to which community-based methods affected our findings. Our method of quantifying study population representativeness can be applied to a broad range of health research and policy contexts. We assessed the representativeness of enrolled children relative to their respective schools. Additionally, the method discussed supports comparisons at many other levels (eg, country, state, zip code, or census tract) or site-based locales (eg, schools, hospitals, or community centers). Our study focused on racial and ethnic representativeness, but the outcome could encompass a wide range of equity indicators (eg, income, age, nationality). For example, data from national cohort studies could be aggregated by state to examine whether the income of enrolled participants matched the mean income of their state. Future replication studies could also determine whether representativeness varies based on the study outcome or disease type. For instance, by pooling data from cancer clinical trials, the racial representativeness of study populations could be examined based on the cancer type, using the hospital as the unit of analysis. Finally, the graphical component of Bland-Altman analysis can be used to obtain additional layers of information by coding points on the graphs to convey characteristics, such as the year of study enrollment, to detect patterns in representativeness over time. Our adapted Bland-Altman method also holds promise for the more inclusive reach of federal health, nutrition, and education programs. For instance, the US government estimated that 11 million individuals were eligible to receive Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) benefits in 2019, but only 57% did. Our proposed method could quantify overall and racial and ethnic differences in WIC participation by state. Each point would indicate the zip code with individual plots for each race and ethnicity. Quantifying representativeness with this approach may provide substantial new knowledge for other federally funded programs, such as Medicare, Medicaid, Head Start, and the Supplemental Nutrition Assistance Program. However, such methodological advancement requires the cooperation of government agencies and relies on accurate and accessible sampling-frame data sets for comparison. Community-based research methods are associated with more representative enrollment outcomes than traditional clinical and population-based approaches. , , However, even the most rigorous and thoughtfully planned community-based studies, such as those in our sample, may face difficulty overcoming barriers to successfully enrolling individuals disproportionately facing adversity. For example, Asian, Black, and Hispanic children may be exposed to more discrimination at schools than their White peers, which directly impedes their health and may affect study enrollment. , , Our findings suggest that Black children, in particular, may be less likely to enroll in community-based studies when they do not belong to their school’s racial majority. Thus, efforts to include context-specific minoritized children are crucial to address rising health inequities effectively. Relatedly, we observed that the racial and ethnic composition of study settings may also incur beneficial effects. Specifically, Black children appeared more likely to be overrepresented at majority Black schools. Future qualitative research assessing inter- and intragroup social cohesion could examine how school and family racial and ethnic relations affect recruitment outcomes, which investigators should incorporate into individualized recruitment plans. Finally, each study worked in low-income and resource-limited communities. Identifying the contributing factors associated with site-level underrepresentation across racial and ethnic groups is crucial since children from low-income backgrounds remain underrepresented in the pediatric health literature. Overall, the mechanisms that limit the effectiveness of enrolling representative samples in community-based studies are not well understood but were partially revealed by our finding of school-level differences in representation based on the school majority. Limitations A main limitation of this cross-sectional study was the exclusion of enrolled children from the analysis because their race and ethnicity data could not be compared with the NCES Common Core data sets (ie, race listed as “other” in all years and multiracial respondents prior to 2010). Moreover, the magnitude of representativeness should be interpreted cautiously, particularly since some schools had small numbers of individuals of each race and ethnicity; thus, deviations from representation could indicate differences of just a few children. Race and ethnicity are fluid social constructs. Discrepancies between school and study demographic response options could have affected our findings if parents reported the child’s race and ethnicity differently on each survey. The included studies were led by investigators with expertise in community-based research. The goal of these community-based studies was to enroll eligible children from schools with certain socioeconomic compositions . These community-engaged study practices are known to include historically marginalized populations better than traditional methods. , Therefore, our representativeness findings may lack generalizability, but that extent should be confirmed with wider application of our technique. Future research could assess how additional factors, such as study design, community engagement, potential study risks, and indicators of social status (eg, parent education), are associated with racial representativeness in pediatric health studies. Notably, such data must be collected on study surveys to overcome limitations of data availability during similar retrospective assessments. A main limitation of this cross-sectional study was the exclusion of enrolled children from the analysis because their race and ethnicity data could not be compared with the NCES Common Core data sets (ie, race listed as “other” in all years and multiracial respondents prior to 2010). Moreover, the magnitude of representativeness should be interpreted cautiously, particularly since some schools had small numbers of individuals of each race and ethnicity; thus, deviations from representation could indicate differences of just a few children. Race and ethnicity are fluid social constructs. Discrepancies between school and study demographic response options could have affected our findings if parents reported the child’s race and ethnicity differently on each survey. The included studies were led by investigators with expertise in community-based research. The goal of these community-based studies was to enroll eligible children from schools with certain socioeconomic compositions . These community-engaged study practices are known to include historically marginalized populations better than traditional methods. , Therefore, our representativeness findings may lack generalizability, but that extent should be confirmed with wider application of our technique. Future research could assess how additional factors, such as study design, community engagement, potential study risks, and indicators of social status (eg, parent education), are associated with racial representativeness in pediatric health studies. Notably, such data must be collected on study surveys to overcome limitations of data availability during similar retrospective assessments. In this cross-sectional study, we confronted a critical barrier to health equity: the systematic underrepresentation of Asian, Black, and Hispanic populations in pediatric health research. Our approach for quantifying representativeness in health research is both feasible and replicable. Prevention of the racial and ethnic health disparities that manifest during childhood and persist over the life span requires a paradigm shift during recruitment and when describing study samples and results. Replication and testing of our application of Bland-Altman analysis could help improve the representation of minoritized racial and ethnic populations in the pediatric health evidence base and in health policies and programs. The proposed method may enable investigators to pinpoint the multilevel determinants of enrollment affecting racial and ethnic groups differently, and it could help alleviate preventable chronic disease disparities that are disproportionately endured by US youth from racially minoritized backgrounds.
Characteristics associated with poor COVID-19 outcomes in people with psoriasis, psoriatic arthritis and axial spondyloarthritis: data from the COVID-19 PsoProtect and Global Rheumatology Alliance physician-reported registries	2c90eb5b-137d-44e5-aee9-fc9f422a00f7	10176347	Internal Medicine[mh]	Factors associated with severe COVID-19 outcomes have been demonstrated in both registry-based and population-based studies for people with immune-mediated inflammatory diseases (IMIDs) collectively and for specific IMIDs. However, relevant risk factor data are limited for axial spondyloarthritis (axSpA) and psoriatic disease (including psoriasis without arthritis (PsO) and psoriatic arthritis (PsA)), a group of conditions that shares pathophysiological mechanisms and approved treatments, particularly targeted therapies. Older age, male sex, comorbidity burden, higher disease activity and glucocorticoid intake were associated with more severe COVID-19. Later pandemic time periods, PsO and exposure to TNFi, IL17i and IL-23i/IL-12+23i were associated with less severe COVID-19. The findings from this study will enable risk stratification for patients with PsO, PsA and axSpA. These findings will inform the development of tailored management strategies and evidence-based recommendations for patients with PsO, PsA and axSpA. The COVID-19 pandemic has significantly impacted people with immune-mediated inflammatory diseases (IMIDs), particularly those taking immunomodulatory drugs such as biological or targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs). While risk factors for severe COVID-19 outcomes have been demonstrated in both registry-based and population-based studies, for people with IMIDs collectively and for specific diseases such as rheumatoid arthritis, relevant risk factor data are limited for axial spondyloarthritis (axSpA) and psoriatic disease (including psoriasis without arthritis (PsO) and psoriatic arthritis (PsA)). The association of specific classes of b/tsDMARDs commonly used in this population, including IL-17 inhibitors (IL17i) and IL-23 or IL-12/23 inhibitors (IL-23i/IL-12+23i), with COVID-19 outcomes has not been well studied. Improved understanding of the risks associated with exposure to these medications in this population will address knowledge gaps as we continue to navigate COVID-19 risks in the postvaccination era. We used data from the COVID-19 Global Rheumatology Alliance (C19-GRA) and the Psoriasis Patient Registry for Outcomes, Therapy and Epidemiology of COVID-19 Infection (PsoProtect) physician-reported registries to evaluate the associations of baseline characteristics, including different classes of b/tsDMARDs, with COVID-19 severity in people with PsO, PsA and axSpA. Data source The C19-GRA physician-reported observational registry launched on 24 March 2020. Patients are eligible for inclusion if they have both a pre-existing rheumatic disease and SARS-CoV-2 infection. PsoProtect is a physician-reported observational registry launched on 27 March 2020. Patients are eligible for inclusion if they have both pre-existing PsO and SARS-CoV-2 infection. For both registries, data are entered voluntarily into the data entry systems by rheumatologists/dermatologists or under the supervision of rheumatologists/dermatologists. In Argentina, Brazil, France, Germany, Italy, Portugal and Sweden, C19-GRA data are transferred from national registries; in all other countries, data are entered directly into the registries’ data entry systems. Countries were categorised according to the six WHO regions ( www.who.int ); the ‘Americas’ was further divided into north and south. Further details of the registries have been described elsewhere. We used data collected on or before 25 October 2021. COVID-19 reporting and primary outcome of interest Both confirmed and presumptive cases of COVID-19 were reported. For analysis, patients were subsequently categorised into (1) confirmed or high likelihood of COVID-19 (chest imaging (CT or chest X-ray) showing bilateral infiltrates and/or symptoms after close contact with a known laboratory-confirmed COVID-19 positive patient) or (2) presumptive cases based on symptoms alone. The primary outcome of interest of this study was COVID-19 outcome, assessed by use of an ordinal COVID-19 severity scale with three mutually exclusive categories: (1) no hospitalisation and no death; (2) hospitalisation, but no death and (3) death. ‘Baseline characteristics’ refer to demographic or clinical characteristics at the time of COVID-19 symptom onset (or diagnosis if asymptomatic). IMID treatment prior to COVID-19 Medications used to treat the IMID prior to COVID-19 diagnosis were categorised into groups. Immunomodulatory drugs (conventional synthetic (cs)/biological (b)/targeted synthetic (ts) DMARDs) were distinguished from the PsO-specific non-biological systemic agent acitretin as well as from non-steroidal anti-inflammatory drugs (NSAIDs) and glucocorticoids (GC). csDMARDs included antimalarials, cyclosporine, leflunomide, methotrexate and sulfasalazine. bDMARDs included TNFi (eg, adalimumab, certolizumab, etanercept, golimumab, infliximab and TNFi biosimilars), IL-17i (eg, brodalumab, ixekizumab and secukinumab), IL-12/23i (ustekinumab) and IL-23i (eg, guselkumab, risankizumab and tildrakizumab). tsDMARDs included apremilast and JAKi (eg, baricitinib, tofacitinib and upadacitinib). IL-23i and IL-12/23i were combined in the same group for data analysis (IL-23i/IL-12+23 i). Regarding NSAIDs, we asked physicians to report if at the time of COVID-19 symptom onset (or diagnosis if asymptomatic), the patient was taking NSAIDs, without specifying a minimal duration of a continuous treatment with NSAIDs. We chose no current DMARD use as the reference group after considering the groups’ sample size and internal validity to be used as comparator for exposure to the various IMID treatments. For more details regarding the choice of DMARD reference category, refer to . 10.1136/ard-2022-223499.supp3 Supplementary data Statistical analyses Descriptive tables were produced for the whole cohort and by diagnostic group (PsO, PsA and axSpA, as defined by the reporting healthcare professional). All patients with confirmed or presumptive COVID-19 were included in the primary analysis. Independent associations between demographic and disease features and the ordinal COVID-19 outcome were estimated by multivariable ordinal logistic regression using the proportional odds model and were reported as OR and 95% CIs. In ordinal regression analysis, the effect size of a categorical predictor gives the change in log odds of being at least one level higher on the ordinal COVID-19 severity scale compared with the reference category of the predictor variable, while for a continuous predictor, it gives the change in odds of being one level higher on the ordinal COVID-19 severity scale for a unit increase in the continuous predictor. More details about assumptions of the proportional odds model are provided in . Factors potentially associated with the COVID-19 outcome considered in the models were age, sex, smoking habits (ever, unknown/missing, never), pandemic calendar period (until 15 June 2020, 16 June 2020 to 31 December 2020, 1 January 2021 and later), key comorbidities (chronic obstructive pulmonary disease (COPD) or asthma, other chronic lung disease, chronic kidney disease (CKD), hypertension, other cardiovascular disease (CVD), obesity, diabetes, cancer), IMID diagnostic category, IMID disease activity as per physician’s global assessment (remission/low vs moderate/high), DMARD treatment prior to COVID-19 diagnosis, GC use and NSAID use. For patients classified as having more than one IMID or being treated with more than one of the medications of interest, we created a hierarchy based on clinical expertise to categorise patients. This way, non-overlapping (mutually exclusive) categories are obtained, allowing a clear reference group for interpretation of the regression models, and avoiding collinearities. Patients labelled as having both PsA and axSpA were counted as PsA patients. Patients receiving multiple csDMARDs were grouped according to the following hierarchy: cyclosporine>sulfasalazine>leflunomide>methotrexate>antimalarials, where ‘A>B’ means ‘A has priority over B’. Patients receiving a b/tsDMARD and additionally a csDMARD were considered in the model solely in the b/tsDMARD group (ie, b/tsDMARD>csDMARD). We tested four two-way additive interactions in the models: hypertension and CVD; obesity and diabetes; cancer and smoking habits; and disease activity and prednisolone-equivalent GC use. provide more details regarding statistical interactions. To account for heterogeneity between participating countries regarding healthcare systems and infection dynamics, countries were considered as random effects in the regression analyses. To appropriately estimate the well-established non-linear effect of age on the outcome of SARS-CoV-2 infection, we included restricted cubic splines in the regression models. Four knots were chosen for most analyses, while three knots were chosen for the outcome mortality and the disease-specific analyses due to the limited effective sample size. Missing data were handled using multiple imputation; results of the logistic regression analyses for 10 imputed datasets were pooled by Rubin’s rules. As disease activity was missing for all patients entered from France in the C19-GRA registry, country-level life expectancy was used in the imputation model to explain potential structural differences in disease activity between countries not accounted for in the patient-level data (data from 2018, source: http://hdr.undp.org/ ). For more details regarding excluded patients and handling of missing data, refer to . IMIDs differ regarding the DMARDs approved for their treatment. To explore the impact of this heterogeneity on the associations of interest, in addition to the primary analysis with all patients, diagnostic categories were defined, and stratified secondary analyses were undertaken separately for patients with PsO, PsA and axSpA. The following sensitivity analyses were also performed to examine the robustness of our findings: (1) analysis limited to patients with confirmed or highly likely COVID-19; (2) analysis using the alternative binary outcome ‘hospitalisation’; (3) analysis using the alternative binary outcome ‘death’. In the model using death as dependent variable, comorbidities were analysed as an independent binary variable (3 or more comorbidities vs less than 3), to minimise the risk of overfitting. Data were considered statistically significant for p values<0.05. All analyses were conducted in SAS (V.9.4) and R (V.4.0.4). The C19-GRA physician-reported observational registry launched on 24 March 2020. Patients are eligible for inclusion if they have both a pre-existing rheumatic disease and SARS-CoV-2 infection. PsoProtect is a physician-reported observational registry launched on 27 March 2020. Patients are eligible for inclusion if they have both pre-existing PsO and SARS-CoV-2 infection. For both registries, data are entered voluntarily into the data entry systems by rheumatologists/dermatologists or under the supervision of rheumatologists/dermatologists. In Argentina, Brazil, France, Germany, Italy, Portugal and Sweden, C19-GRA data are transferred from national registries; in all other countries, data are entered directly into the registries’ data entry systems. Countries were categorised according to the six WHO regions ( www.who.int ); the ‘Americas’ was further divided into north and south. Further details of the registries have been described elsewhere. We used data collected on or before 25 October 2021. Both confirmed and presumptive cases of COVID-19 were reported. For analysis, patients were subsequently categorised into (1) confirmed or high likelihood of COVID-19 (chest imaging (CT or chest X-ray) showing bilateral infiltrates and/or symptoms after close contact with a known laboratory-confirmed COVID-19 positive patient) or (2) presumptive cases based on symptoms alone. The primary outcome of interest of this study was COVID-19 outcome, assessed by use of an ordinal COVID-19 severity scale with three mutually exclusive categories: (1) no hospitalisation and no death; (2) hospitalisation, but no death and (3) death. ‘Baseline characteristics’ refer to demographic or clinical characteristics at the time of COVID-19 symptom onset (or diagnosis if asymptomatic). Medications used to treat the IMID prior to COVID-19 diagnosis were categorised into groups. Immunomodulatory drugs (conventional synthetic (cs)/biological (b)/targeted synthetic (ts) DMARDs) were distinguished from the PsO-specific non-biological systemic agent acitretin as well as from non-steroidal anti-inflammatory drugs (NSAIDs) and glucocorticoids (GC). csDMARDs included antimalarials, cyclosporine, leflunomide, methotrexate and sulfasalazine. bDMARDs included TNFi (eg, adalimumab, certolizumab, etanercept, golimumab, infliximab and TNFi biosimilars), IL-17i (eg, brodalumab, ixekizumab and secukinumab), IL-12/23i (ustekinumab) and IL-23i (eg, guselkumab, risankizumab and tildrakizumab). tsDMARDs included apremilast and JAKi (eg, baricitinib, tofacitinib and upadacitinib). IL-23i and IL-12/23i were combined in the same group for data analysis (IL-23i/IL-12+23 i). Regarding NSAIDs, we asked physicians to report if at the time of COVID-19 symptom onset (or diagnosis if asymptomatic), the patient was taking NSAIDs, without specifying a minimal duration of a continuous treatment with NSAIDs. We chose no current DMARD use as the reference group after considering the groups’ sample size and internal validity to be used as comparator for exposure to the various IMID treatments. For more details regarding the choice of DMARD reference category, refer to . 10.1136/ard-2022-223499.supp3 Supplementary data Descriptive tables were produced for the whole cohort and by diagnostic group (PsO, PsA and axSpA, as defined by the reporting healthcare professional). All patients with confirmed or presumptive COVID-19 were included in the primary analysis. Independent associations between demographic and disease features and the ordinal COVID-19 outcome were estimated by multivariable ordinal logistic regression using the proportional odds model and were reported as OR and 95% CIs. In ordinal regression analysis, the effect size of a categorical predictor gives the change in log odds of being at least one level higher on the ordinal COVID-19 severity scale compared with the reference category of the predictor variable, while for a continuous predictor, it gives the change in odds of being one level higher on the ordinal COVID-19 severity scale for a unit increase in the continuous predictor. More details about assumptions of the proportional odds model are provided in . Factors potentially associated with the COVID-19 outcome considered in the models were age, sex, smoking habits (ever, unknown/missing, never), pandemic calendar period (until 15 June 2020, 16 June 2020 to 31 December 2020, 1 January 2021 and later), key comorbidities (chronic obstructive pulmonary disease (COPD) or asthma, other chronic lung disease, chronic kidney disease (CKD), hypertension, other cardiovascular disease (CVD), obesity, diabetes, cancer), IMID diagnostic category, IMID disease activity as per physician’s global assessment (remission/low vs moderate/high), DMARD treatment prior to COVID-19 diagnosis, GC use and NSAID use. For patients classified as having more than one IMID or being treated with more than one of the medications of interest, we created a hierarchy based on clinical expertise to categorise patients. This way, non-overlapping (mutually exclusive) categories are obtained, allowing a clear reference group for interpretation of the regression models, and avoiding collinearities. Patients labelled as having both PsA and axSpA were counted as PsA patients. Patients receiving multiple csDMARDs were grouped according to the following hierarchy: cyclosporine>sulfasalazine>leflunomide>methotrexate>antimalarials, where ‘A>B’ means ‘A has priority over B’. Patients receiving a b/tsDMARD and additionally a csDMARD were considered in the model solely in the b/tsDMARD group (ie, b/tsDMARD>csDMARD). We tested four two-way additive interactions in the models: hypertension and CVD; obesity and diabetes; cancer and smoking habits; and disease activity and prednisolone-equivalent GC use. provide more details regarding statistical interactions. To account for heterogeneity between participating countries regarding healthcare systems and infection dynamics, countries were considered as random effects in the regression analyses. To appropriately estimate the well-established non-linear effect of age on the outcome of SARS-CoV-2 infection, we included restricted cubic splines in the regression models. Four knots were chosen for most analyses, while three knots were chosen for the outcome mortality and the disease-specific analyses due to the limited effective sample size. Missing data were handled using multiple imputation; results of the logistic regression analyses for 10 imputed datasets were pooled by Rubin’s rules. As disease activity was missing for all patients entered from France in the C19-GRA registry, country-level life expectancy was used in the imputation model to explain potential structural differences in disease activity between countries not accounted for in the patient-level data (data from 2018, source: http://hdr.undp.org/ ). For more details regarding excluded patients and handling of missing data, refer to . IMIDs differ regarding the DMARDs approved for their treatment. To explore the impact of this heterogeneity on the associations of interest, in addition to the primary analysis with all patients, diagnostic categories were defined, and stratified secondary analyses were undertaken separately for patients with PsO, PsA and axSpA. The following sensitivity analyses were also performed to examine the robustness of our findings: (1) analysis limited to patients with confirmed or highly likely COVID-19; (2) analysis using the alternative binary outcome ‘hospitalisation’; (3) analysis using the alternative binary outcome ‘death’. In the model using death as dependent variable, comorbidities were analysed as an independent binary variable (3 or more comorbidities vs less than 3), to minimise the risk of overfitting. Data were considered statistically significant for p values<0.05. All analyses were conducted in SAS (V.9.4) and R (V.4.0.4). Study sample and baseline characteristics The study population included 5045 cases, of which 921 (18.3%) were patients with PsO, 2293 (45.5%) with PsA, and 1831 (36.3%) with axSpA. Overall, the mean age was 50 years (SD 13.5), just over half were male (51.7%) and most were from Europe (77.5%) . Cases were reported fairly equally across the three pandemic time periods. Most cases had disease (IMIDs) in remission or minimal/low disease activity (82.7%). About half had no key comorbidities reported (52.9%). Of those with comorbidities, the most reported were hypertension (26.5%) and obesity (21.1%). Any csDMARD use was reported in 30.3%, with methotrexate as the most common (23.4%). Only 5.6% reported using sulfasalazine. bDMARD use was reported in 65.7% (TNFi 45.6%, IL17i 12.1%, IL-23i/IL-12+23i 8.1%). Only 1.2% reported JAKi use. Baseline GC use was reported in only 7.3% (4.6%, 0–7.5 mg/day and 1.4%, >7.5 mg/day) and NSAID use in 24%. When stratified by condition , the main notable differences were that individuals with PsA were older (mean 53.2 years vs 46.9 years in axSpA and 48.4 years in PsO), a higher proportion of those with PsA had hypertension (32.8% vs 21.2% in axSpA and 21.3% in PsO) and a higher proportion of those with PsO were obese (29.2% vs 23% in PsA and 14.5% in axSpA). csDMARDs were most used among individuals with PsA (46.6% vs 19.3% in axSpA and 11.6% in PsO) while bDMARDs were most used among individuals with axSpA (73.6% vs 58.5% in PsA and 68.1% in PsO). Baseline GC usage was low overall but differed notably between the groups, with almost none in PsO (0.7%) vs 10.7% in PsA and 6.3% in axSpA. There was no difference across disease groups with regard to disease activity. When stratified by medication group , patients not taking DMARDs were slightly younger (mean 49.9 years) than patients taking DMARDs (range from 50 to 56.2 years, depending on the DMARD group) except for IL-17i/IL-23i/IL12+23i (mean 49.9 years) and TNFi (mean 48.3 years). Moreover, patients not taking DMARDs were slightly less often in remission/low disease activity (71.%) than patients taking DMARDs (range from 80.5% to 86.1%, depending on the DMARD group) except for JAKi (65.3% in remission/low disease activity). 10.1136/ard-2022-223499.supp2 Supplementary data COVID-19 outcomes Baseline characteristics of the study population stratified by COVID-19 outcome are shown in . Most patients (4220, 83.6%) were not hospitalised, 736 (14.6%) were hospitalised and 89 (1.8%) died. The frequency of hospitalisation (without death) and death were slightly higher in PsA (17.1% and 2.2%, respectively), compared with axSpA (12.5% and 1.4%, respectively) and PsO (12.5% and 1.3%, respectively) . Associations of baseline characteristics with COVID-19 severity The results of the primary multivariable ordinal logistic regression model are shown in and the relationship between age and probability of hospitalisation and death is shown in . Age was associated with COVID-19 severity in a non-linear way (stronger association for older age groups). Hypertension without CVD (OR 1.25, 95% CI 1.01 to 1.55), CVD without hypertension (OR 1.87, 95% CI 1.21 to 2.90), COPD or asthma (OR 1.75, 95% CI 1.33 to 2.31), other lung disease (OR 2.54, 95% CI 1.64 to 3.93), CKD (OR 2.32, 95% CI 1.50 to 3.58), cancer in patients with missing data on smoking (OR 2.89, 95% CI 1.19 to 6.97), obesity without diabetes (OR 1.35, 95% CI 1.07 to 1.70), diabetes without obesity (OR 1.84, 95% CI 1.38 to 2.45), and coexistence of obesity and diabetes (OR 1.89, 95% CI 1.34 to 2.68) were associated with greater odds of worse COVID-19 severity compared with referents without each condition. Male sex was associated with 1.54 times greater odds of worse COVID-19 severity compared with female sex (95% CI 1.30 to 1.83). Moderate/high disease activity (with or without GC use) and remission/low disease activity (with GC use) were associated with higher odds of worse COVID-19 outcomes compared with being in remission/low disease activity without GC use (OR ranging from 1.39 to 2.23). Later pandemic time periods were associated with lower odds of worse COVID-19 severity compared with the baseline period of March 2020–15 June 2020 (OR 0.42, 95% CI 0.34 to 0.51 for 16 June 2020–31 December 2020; OR 0.52, 95% CI 0.41 to 0.67 for 1 January 2021 and later). Compared with PsA, PsO was associated with less COVID-19 severity (OR 0.49, 95% CI 0.37 to 0.65). For medication classes, none were associated with higher odds of COVID-19 severity. TNFi, IL17i and IL-23i/IL-12+23i all demonstrated reduced odds of severe COVID-19 outcomes (OR 0.57, 95% CI 0.44 to 0.73; OR 0.62, 95% CI 0.45 to 0.87; OR 0.67, 95% CI 0.45 to 0.98, respectively). Finally, NSAID use compared with no use of NSAIDs was associated with lower odds of severe COVID-19 outcomes (OR 0.77, 95% CI 0.60 to 0.98). Stratified analyses When stratified by condition, results were similar to the primary model and ) with the following notable exceptions: hypertension alone and CVD alone were only significantly associated with the COVID-19 severity outcome among those with axSpA (OR 1.49, 95% CI 1.01 to 2.19; and OR 2.77, 95% CI 1.25 to 6.13; respectively) whereas COPD and asthma were associated with the COVID-19 severity outcome only among those with PsA (OR 1.95, 95% CI 1.34 to 2.82). The association of IL-23i/IL-12+23i with less severe COVID-19 outcomes was only statistically significant among those with PsO (OR 0.43, 95% CI 0.23 to 0.82); however, IL-23i/IL-12+23i were not used among patients with axSpA (not efficacious/licensed for this indication) and numbers were lower for PsA. Sensitivity analyses The results of sensitivity analyses are shown in and . When restricting the analysis to confirmed COVID-19 cases (n=4176), multivariable model results were consistent with the primary model. Results were also similar to the primary model for the binary outcome of hospitalisation. For the binary outcome of death, male sex (OR 2.00, 95% CI 1.22 to 3.26), having three or more comorbidities (OR 3.34, 95% CI 1.98 to 5.63) and baseline GC use (OR 1.91, 95% CI 1.002 to 3.64) remained associated with the outcome of interest. In this model, TNFi and IL17i continued to demonstrate reduced odds of severe COVID-19 outcomes (OR 0.50, 95% CI 0.26 to 0.98 and OR 0.11, 95% CI 0.02 to 0.51; respectively). However, sulfasalazine use (OR 2.64, 95% CI 1.13 to 6.17) and JAKi use (OR 7.49, 95% CI 2.61 to 21.47) were associated with greater odds of severe COVID-19 outcomes in this model. Discussion In this registry-based study of individuals with PsO, PsA and axSpA with SARS-CoV-2 infection, we found that known risk factors for the general population (older age, the presence of comorbidities) and for IMIDs overall (higher disease activity, higher baseline GC usage) were associated with more severe COVID-19 outcomes. In addition, a diagnosis of COVID-19 in a later time period during the pandemic was associated with lower disease severity compared with early 2020. Consistent with previous studies, baseline TNFi use was associated with lower odds for severe COVID-19 outcomes; we also found that IL17i and IL-23i/IL-12+23i use had similar associations with lower odds for severe COVID-19 outcomes. The findings of our study reiterate known risk factors in both the general population and among people with IMIDs: older age, male sex and presence of comorbidities, specifically cardiometabolic and pulmonary conditions, were associated with more severe COVID-19 outcomes. Our findings that disease activity and GC usage at baseline have an additive interaction are consistent with prior findings in the C19-GRA registry. In this study, baseline use of TNFi was associated with lower odds of severe COVID-19 outcomes. This was previously shown in the C19-GRA registry, in a combined rheumatic disease, inflammatory bowel disease (IBD) and PsO analysis, and in a US-based administrative claims database study among individuals with RA. Mechanistic plausibility for trialling TNFi therapies for COVID-19 treatment has been discussed in the literature. These therapies neutralise TNF, a major cytokine in the excess inflammatory phase of COVID-19, and several trials are ongoing. A recent preprint announced results of a large randomised, placebo-controlled clinical trial led by the National Institutes of Health showing that treating adults hospitalised with COVID-19 with infliximab (a TNFi) did not significantly shorten time to recovery but did improve 14-day clinical status and substantially reduced 28-day mortality compared with standard of care —the peer-reviewed publication is awaited. We also demonstrated that using IL17i and IL-23i/IL-12+23i was also associated with lower odds of severe COVID-19 outcomes. Prior population-level data from Israel and the UK have shown that the use of IL-17i was not associated with worse COVID-19 outcomes. At the same time, case reports and case series have also suggested that IL-17 and IL-23 inhibition may not have a negative effect on the course of COVID-19, though further inference on whether exposure to these medications might be associated with better COVID-19 outcomes is limited. IL-17 may play a pathogenic role in acute respiratory distress syndrome and lung inflammation associated with severe COVID-19. Patients with COVID-19 who experience pulmonary complications have increased and activated Th17 cell populations, and lung damage and hyperinflammation are linked to these patients’ increased Th17 cell responses. The anti-IL-17 monoclonal antibody netakimab improved survival in a small clinical trial in patients with COVID-19; it decreased lung lesion volume and the need for oxygen support. However, in another study, netakimab therapy improved some clinical parameters and decreased C reactive protein levels, but it had no effect on the need for mechanical ventilation or patient survival in COVID-19 patients. Suppressing inflammation via a variety of mechanisms has been shown to improve COVID-19 outcomes in people with severe disease (ie, GC, IL-6i, JAKi, maybe TNFi). Whether IL-17 will also have a role remains to be determined and requires further study. Importantly, in our study, we report associations and therefore we caution against interpreting our estimates causally, as the possibility of selection bias and unmeasured confounding cannot be excluded. Apremilast was not associated with the severity of COVID-19 in patients with PsO/PsA. Although the number of patients taking apremilast was low, these data are important because they add to limited previous evidence of a favourable safety profile of apremilast on COVID-19 severity in patients with these conditions. The finding that baseline NSAID use was associated with less COVID-19 severity is interesting but should be interpreted with caution. NSAID use is particularly prone to reporting bias, and inconsistencies in reporting might have resulted from the fact that we did not specify a minimal duration of a continuous treatment with NSAID and did not use a standardised questionnaire to collect NSAID data (eg, type of NSAID, dose and duration of treatment). General population studies in the UK and Denmark have not found associations between NSAID use and COVID-19-related hospitalisation or death. In our study, this association was seen particularly in individuals with axSpA and may be related to milder disease and/or well controlled of disease activity; confounding by indication cannot be excluded. Finally, the results of one sensitivity analysis indicated that use of sulfasalazine and JAKi were associated with higher odds of death (binary outcome) due to COVID-19, though there were no associations with the ordinal COVID-19 severity outcome or with hospitalisation (binary outcome). In the C-19 GRA registry, we previously found an association of sulfasalazine use with worse COVID-19 outcome, a finding which was also seen in initial analyses of the Surveillance Epidemiology of Coronavirus Under Research Exclusion (IBD) database though later analyses were null. While there are biologically plausible effects of sulfasalazine on SARS-CoV-2 viral entry, our results may be due to residual confounding. The association of JAKi usage with COVID-19 outcomes is consistent with findings from some studies focused on people with RA. However, results from this sensitivity analysis should be interpreted with caution as the proportion of patients on JAKi was low (and no patients with PsO were taking this medication) and the respective 95% CI was wide. Our study has several strengths, including the international nature of the combined registries, the large sample size and the granularity of information regarding IMID medications and disease activity. Our study also has limitations. First, the C19-GRA and PsoProtect registries were dependent on voluntary provider entry of cases, and there may be bias towards cases with more severe COVID-19 and those on DMARD therapy, as mostly secondary care clinicians were submitting cases. As such, proportions of events in our study sample should not be interpreted as incidence rates. Second, while we tried to mitigate the impacts of selection bias and confounding by indication, it is possible that our results may still be biased. However, we performed a series of sensitivity analyses to confirm the robustness of our findings, including restricting to a sample of confirmed cases of COVID-19, and our results were consistent across these additional analyses. Third, although we were able to adjust for several potential confounders in our models, there may still be residual unmeasured confounding. We did not have data available on disease duration or prior medication use, apart from what was reported at the time of COVID-19 diagnosis. Finally, vaccination status was not available for the patients in this dataset; however, the model adjustment for pandemic calendar period used in this study may act as a surrogate for vaccination status. In conclusion, more severe COVID-19 outcomes in PsO, PsA and axSpA are largely associated with age, comorbidities, active disease and GC use. None of the bDMARDs typically used in PsO, PsA and axSpA, including TNFi, IL-17i and IL-23i/IL-12+23i, were associated with severe COVID-19 outcomes, and no biologics-specific differences were found. Our findings will help clinicians, scientific societies and policy makers worldwide develop tailored management strategies for patients with PsO, PsA and axSpA during COVID-19 waves or similar future respiratory pandemics. 10.1136/ard-2022-223499.supp1 Supplementary data The study population included 5045 cases, of which 921 (18.3%) were patients with PsO, 2293 (45.5%) with PsA, and 1831 (36.3%) with axSpA. Overall, the mean age was 50 years (SD 13.5), just over half were male (51.7%) and most were from Europe (77.5%) . Cases were reported fairly equally across the three pandemic time periods. Most cases had disease (IMIDs) in remission or minimal/low disease activity (82.7%). About half had no key comorbidities reported (52.9%). Of those with comorbidities, the most reported were hypertension (26.5%) and obesity (21.1%). Any csDMARD use was reported in 30.3%, with methotrexate as the most common (23.4%). Only 5.6% reported using sulfasalazine. bDMARD use was reported in 65.7% (TNFi 45.6%, IL17i 12.1%, IL-23i/IL-12+23i 8.1%). Only 1.2% reported JAKi use. Baseline GC use was reported in only 7.3% (4.6%, 0–7.5 mg/day and 1.4%, >7.5 mg/day) and NSAID use in 24%. When stratified by condition , the main notable differences were that individuals with PsA were older (mean 53.2 years vs 46.9 years in axSpA and 48.4 years in PsO), a higher proportion of those with PsA had hypertension (32.8% vs 21.2% in axSpA and 21.3% in PsO) and a higher proportion of those with PsO were obese (29.2% vs 23% in PsA and 14.5% in axSpA). csDMARDs were most used among individuals with PsA (46.6% vs 19.3% in axSpA and 11.6% in PsO) while bDMARDs were most used among individuals with axSpA (73.6% vs 58.5% in PsA and 68.1% in PsO). Baseline GC usage was low overall but differed notably between the groups, with almost none in PsO (0.7%) vs 10.7% in PsA and 6.3% in axSpA. There was no difference across disease groups with regard to disease activity. When stratified by medication group , patients not taking DMARDs were slightly younger (mean 49.9 years) than patients taking DMARDs (range from 50 to 56.2 years, depending on the DMARD group) except for IL-17i/IL-23i/IL12+23i (mean 49.9 years) and TNFi (mean 48.3 years). Moreover, patients not taking DMARDs were slightly less often in remission/low disease activity (71.%) than patients taking DMARDs (range from 80.5% to 86.1%, depending on the DMARD group) except for JAKi (65.3% in remission/low disease activity). 10.1136/ard-2022-223499.supp2 Supplementary data Baseline characteristics of the study population stratified by COVID-19 outcome are shown in . Most patients (4220, 83.6%) were not hospitalised, 736 (14.6%) were hospitalised and 89 (1.8%) died. The frequency of hospitalisation (without death) and death were slightly higher in PsA (17.1% and 2.2%, respectively), compared with axSpA (12.5% and 1.4%, respectively) and PsO (12.5% and 1.3%, respectively) . The results of the primary multivariable ordinal logistic regression model are shown in and the relationship between age and probability of hospitalisation and death is shown in . Age was associated with COVID-19 severity in a non-linear way (stronger association for older age groups). Hypertension without CVD (OR 1.25, 95% CI 1.01 to 1.55), CVD without hypertension (OR 1.87, 95% CI 1.21 to 2.90), COPD or asthma (OR 1.75, 95% CI 1.33 to 2.31), other lung disease (OR 2.54, 95% CI 1.64 to 3.93), CKD (OR 2.32, 95% CI 1.50 to 3.58), cancer in patients with missing data on smoking (OR 2.89, 95% CI 1.19 to 6.97), obesity without diabetes (OR 1.35, 95% CI 1.07 to 1.70), diabetes without obesity (OR 1.84, 95% CI 1.38 to 2.45), and coexistence of obesity and diabetes (OR 1.89, 95% CI 1.34 to 2.68) were associated with greater odds of worse COVID-19 severity compared with referents without each condition. Male sex was associated with 1.54 times greater odds of worse COVID-19 severity compared with female sex (95% CI 1.30 to 1.83). Moderate/high disease activity (with or without GC use) and remission/low disease activity (with GC use) were associated with higher odds of worse COVID-19 outcomes compared with being in remission/low disease activity without GC use (OR ranging from 1.39 to 2.23). Later pandemic time periods were associated with lower odds of worse COVID-19 severity compared with the baseline period of March 2020–15 June 2020 (OR 0.42, 95% CI 0.34 to 0.51 for 16 June 2020–31 December 2020; OR 0.52, 95% CI 0.41 to 0.67 for 1 January 2021 and later). Compared with PsA, PsO was associated with less COVID-19 severity (OR 0.49, 95% CI 0.37 to 0.65). For medication classes, none were associated with higher odds of COVID-19 severity. TNFi, IL17i and IL-23i/IL-12+23i all demonstrated reduced odds of severe COVID-19 outcomes (OR 0.57, 95% CI 0.44 to 0.73; OR 0.62, 95% CI 0.45 to 0.87; OR 0.67, 95% CI 0.45 to 0.98, respectively). Finally, NSAID use compared with no use of NSAIDs was associated with lower odds of severe COVID-19 outcomes (OR 0.77, 95% CI 0.60 to 0.98). When stratified by condition, results were similar to the primary model and ) with the following notable exceptions: hypertension alone and CVD alone were only significantly associated with the COVID-19 severity outcome among those with axSpA (OR 1.49, 95% CI 1.01 to 2.19; and OR 2.77, 95% CI 1.25 to 6.13; respectively) whereas COPD and asthma were associated with the COVID-19 severity outcome only among those with PsA (OR 1.95, 95% CI 1.34 to 2.82). The association of IL-23i/IL-12+23i with less severe COVID-19 outcomes was only statistically significant among those with PsO (OR 0.43, 95% CI 0.23 to 0.82); however, IL-23i/IL-12+23i were not used among patients with axSpA (not efficacious/licensed for this indication) and numbers were lower for PsA. The results of sensitivity analyses are shown in and . When restricting the analysis to confirmed COVID-19 cases (n=4176), multivariable model results were consistent with the primary model. Results were also similar to the primary model for the binary outcome of hospitalisation. For the binary outcome of death, male sex (OR 2.00, 95% CI 1.22 to 3.26), having three or more comorbidities (OR 3.34, 95% CI 1.98 to 5.63) and baseline GC use (OR 1.91, 95% CI 1.002 to 3.64) remained associated with the outcome of interest. In this model, TNFi and IL17i continued to demonstrate reduced odds of severe COVID-19 outcomes (OR 0.50, 95% CI 0.26 to 0.98 and OR 0.11, 95% CI 0.02 to 0.51; respectively). However, sulfasalazine use (OR 2.64, 95% CI 1.13 to 6.17) and JAKi use (OR 7.49, 95% CI 2.61 to 21.47) were associated with greater odds of severe COVID-19 outcomes in this model. In this registry-based study of individuals with PsO, PsA and axSpA with SARS-CoV-2 infection, we found that known risk factors for the general population (older age, the presence of comorbidities) and for IMIDs overall (higher disease activity, higher baseline GC usage) were associated with more severe COVID-19 outcomes. In addition, a diagnosis of COVID-19 in a later time period during the pandemic was associated with lower disease severity compared with early 2020. Consistent with previous studies, baseline TNFi use was associated with lower odds for severe COVID-19 outcomes; we also found that IL17i and IL-23i/IL-12+23i use had similar associations with lower odds for severe COVID-19 outcomes. The findings of our study reiterate known risk factors in both the general population and among people with IMIDs: older age, male sex and presence of comorbidities, specifically cardiometabolic and pulmonary conditions, were associated with more severe COVID-19 outcomes. Our findings that disease activity and GC usage at baseline have an additive interaction are consistent with prior findings in the C19-GRA registry. In this study, baseline use of TNFi was associated with lower odds of severe COVID-19 outcomes. This was previously shown in the C19-GRA registry, in a combined rheumatic disease, inflammatory bowel disease (IBD) and PsO analysis, and in a US-based administrative claims database study among individuals with RA. Mechanistic plausibility for trialling TNFi therapies for COVID-19 treatment has been discussed in the literature. These therapies neutralise TNF, a major cytokine in the excess inflammatory phase of COVID-19, and several trials are ongoing. A recent preprint announced results of a large randomised, placebo-controlled clinical trial led by the National Institutes of Health showing that treating adults hospitalised with COVID-19 with infliximab (a TNFi) did not significantly shorten time to recovery but did improve 14-day clinical status and substantially reduced 28-day mortality compared with standard of care —the peer-reviewed publication is awaited. We also demonstrated that using IL17i and IL-23i/IL-12+23i was also associated with lower odds of severe COVID-19 outcomes. Prior population-level data from Israel and the UK have shown that the use of IL-17i was not associated with worse COVID-19 outcomes. At the same time, case reports and case series have also suggested that IL-17 and IL-23 inhibition may not have a negative effect on the course of COVID-19, though further inference on whether exposure to these medications might be associated with better COVID-19 outcomes is limited. IL-17 may play a pathogenic role in acute respiratory distress syndrome and lung inflammation associated with severe COVID-19. Patients with COVID-19 who experience pulmonary complications have increased and activated Th17 cell populations, and lung damage and hyperinflammation are linked to these patients’ increased Th17 cell responses. The anti-IL-17 monoclonal antibody netakimab improved survival in a small clinical trial in patients with COVID-19; it decreased lung lesion volume and the need for oxygen support. However, in another study, netakimab therapy improved some clinical parameters and decreased C reactive protein levels, but it had no effect on the need for mechanical ventilation or patient survival in COVID-19 patients. Suppressing inflammation via a variety of mechanisms has been shown to improve COVID-19 outcomes in people with severe disease (ie, GC, IL-6i, JAKi, maybe TNFi). Whether IL-17 will also have a role remains to be determined and requires further study. Importantly, in our study, we report associations and therefore we caution against interpreting our estimates causally, as the possibility of selection bias and unmeasured confounding cannot be excluded. Apremilast was not associated with the severity of COVID-19 in patients with PsO/PsA. Although the number of patients taking apremilast was low, these data are important because they add to limited previous evidence of a favourable safety profile of apremilast on COVID-19 severity in patients with these conditions. The finding that baseline NSAID use was associated with less COVID-19 severity is interesting but should be interpreted with caution. NSAID use is particularly prone to reporting bias, and inconsistencies in reporting might have resulted from the fact that we did not specify a minimal duration of a continuous treatment with NSAID and did not use a standardised questionnaire to collect NSAID data (eg, type of NSAID, dose and duration of treatment). General population studies in the UK and Denmark have not found associations between NSAID use and COVID-19-related hospitalisation or death. In our study, this association was seen particularly in individuals with axSpA and may be related to milder disease and/or well controlled of disease activity; confounding by indication cannot be excluded. Finally, the results of one sensitivity analysis indicated that use of sulfasalazine and JAKi were associated with higher odds of death (binary outcome) due to COVID-19, though there were no associations with the ordinal COVID-19 severity outcome or with hospitalisation (binary outcome). In the C-19 GRA registry, we previously found an association of sulfasalazine use with worse COVID-19 outcome, a finding which was also seen in initial analyses of the Surveillance Epidemiology of Coronavirus Under Research Exclusion (IBD) database though later analyses were null. While there are biologically plausible effects of sulfasalazine on SARS-CoV-2 viral entry, our results may be due to residual confounding. The association of JAKi usage with COVID-19 outcomes is consistent with findings from some studies focused on people with RA. However, results from this sensitivity analysis should be interpreted with caution as the proportion of patients on JAKi was low (and no patients with PsO were taking this medication) and the respective 95% CI was wide. Our study has several strengths, including the international nature of the combined registries, the large sample size and the granularity of information regarding IMID medications and disease activity. Our study also has limitations. First, the C19-GRA and PsoProtect registries were dependent on voluntary provider entry of cases, and there may be bias towards cases with more severe COVID-19 and those on DMARD therapy, as mostly secondary care clinicians were submitting cases. As such, proportions of events in our study sample should not be interpreted as incidence rates. Second, while we tried to mitigate the impacts of selection bias and confounding by indication, it is possible that our results may still be biased. However, we performed a series of sensitivity analyses to confirm the robustness of our findings, including restricting to a sample of confirmed cases of COVID-19, and our results were consistent across these additional analyses. Third, although we were able to adjust for several potential confounders in our models, there may still be residual unmeasured confounding. We did not have data available on disease duration or prior medication use, apart from what was reported at the time of COVID-19 diagnosis. Finally, vaccination status was not available for the patients in this dataset; however, the model adjustment for pandemic calendar period used in this study may act as a surrogate for vaccination status. In conclusion, more severe COVID-19 outcomes in PsO, PsA and axSpA are largely associated with age, comorbidities, active disease and GC use. None of the bDMARDs typically used in PsO, PsA and axSpA, including TNFi, IL-17i and IL-23i/IL-12+23i, were associated with severe COVID-19 outcomes, and no biologics-specific differences were found. Our findings will help clinicians, scientific societies and policy makers worldwide develop tailored management strategies for patients with PsO, PsA and axSpA during COVID-19 waves or similar future respiratory pandemics. 10.1136/ard-2022-223499.supp1 Supplementary data
Paediatric early warning systems: not a simple answer to a complex question	c9268738-3fc7-4a84-ae7a-33fbf158d656	10176370	Pediatrics[mh]	In adult clinical practice, the incidence of preventable inpatient mortality is significant (with potentially over 22 000 preventable deaths annually in the USA) and has precipitated the development of a variety of interventions to standardise the processes for recognition and response to evolving inpatient deterioration. These have been delivered by the use of early warning systems (EWS) to aid recognition and response to patient deterioration which have included: The development of track and trigger tools (numerical values assigned to commonly measured physiological and observation values which produce a composite score that correlates to an escalating response process dependant on the score) Rapid response teams (RRTs) (experienced staff assigned to respond to patients who trigger criteria predictive of impending need for intensive care) and Critical care inreach and outreach (deploying specialist critical care staff to non-intensive care settings to prevent admission and outreaching to prevent readmission to critical care). Determining effectiveness is challenging and data on system-level improvement in mortality are sparse. The concept of EWS in the UK was discussed as early as 1997 and the Royal College of Physicians recommended their use in 2007. Since this point, publicly available data have shown inpatient hospital mortality has actually risen from 275 000 in 2014 to 293 000 in 2019 (uncontrolled for population growth or number of hospitals). The interventions (1–3) above have different impacts and the relationship between a specific intervention and an outcome is not always clear. For example an RRT is dependent on processes which enable not only initial recognition of deterioration, but require an institution to have experienced personnel available to make time-critical decisions on patients and have the capacity to provide appropriate resources if care needs to be escalated. It has been suggested that adult RRTs were implemented to address a problem without fully understanding the context of the problem and the system in which these problems exist but the face validity that RRT should work and the impetus to improve patient safety was so strong that limited evaluation of its mechanism of action took place. A subsequent systematic review has demonstrated a variety of non-patient-based factors to also be important, such as leadership and punitive hierarchies. This is similar to the introduction of pulse oximetry, which had minimal level 1 evidence supporting its introduction. The ease of measuring oxygen levels is so great it would now be unethical to undertake a study randomising patients to a control group. Acknowledging these experiences in adult practice by examining the link between intervention and outcome with a paediatric lens, we will explore what we know about the effectiveness of EWS for use in children (paediatric early warning systems, PEWS) and what this means for the development of outcome measures. The first published Paediatric Early Warning Score in the UK was in 2005 with multiple iterations published since and nearly all hospitals in England now using some form of a score. In a recent systematic review, our group examined 30 studies to determine the effectiveness of paediatric track and trigger tools (PTTT). These studies were predominantly before and after studies using a variety of outcomes. These can be categorised into discrete outcome measures which have been grouped by frequency in . There is great variation among studies in the precise definition of a code blue, cardiac arrest, respiratory arrest, unplanned admission to paediatric intensive care unit (PICU), and the criteria for PICU admission or high-dependency admission, making comparisons of PTTTs difficult. Denominator variation for mortality and cardiac arrest figures, different types of controls used in studies or indeed no control groups, as well as study design variability, all compound that challenge. In particular, denominators are challenging to calculate as they are often recorded as patient bed days, rather than raw patient numbers which can suffer from poor data recording (eg, over weekends). It is also important to note that most studies were conducted in specialist and tertiary centres where specialist intensive care advice is available on-site. These outcomes are therefore not representative of what is measured, or can be measured, in hospitals without direct access to intensive care. In these centres, where most hospitalised children in the UK are admitted, different decisions may be made about escalation, because of the lack of immediate access to both a PICU bed and critical care trained clinicians. Furthermore, children may deteriorate in one hospital without intensive care facilities but die in another which has them. It may then not be clear which interventions in either hospital were most impactful, or lacking, making causation and the role of PTTT difficult to ascertain. Therefore, if PTTT Scores are being used in non-specialist hospitals, it cannot be assumed the outcomes from their use will be the same as in tertiary hospitals. This also applies to their use in emergency departments, and prehospital settings, by paramedical staff and in primary care. PTTTs provide a mechanism to proactively highlight the need for review of a child who is becoming clinically unstable. In prehospital and emergency department locations a high PTTT Score may be representative of a child’s initial acuity rather than evidence of deterioration. The relationship between a high PTTT Score and a specific outcome is then less certain. Furthermore, the pretest probability for serious disease (which may result in deterioration) is much lower in emergency departments and prehospital settings therefore altering a PTTT’s specificity and sensitivity. It is perhaps for these reasons a paucity of evidence exists on the performance of PTTT Scores in these environments. The use of PTTTs with low specificity and low positive predictive values may result in alarm fatigue, short cuts to bypass the system and disengagement with the system overall. Adjustment of ‘normal’ vital sign triggering thresholds may be necessary for some high-risk groups of children (eg, children with congenital heart conditions), but if done by inexperienced staff or without evidence-based data, may reduce triggering of a PTTT and consequent failure to detect deterioration. Conversely, unnecessary critical care admission reduces capacity in an already strained system, is resource-intensive and may redirect experienced staff to clinical areas where they are not needed. PEWS are complex due to the requirement for age-specific thresholds and difficulties ascertaining their effectiveness due to the lower mortality in children than in adults. All-cause mortality in children and young people continues to decline over time and a decade ago, Joffe and colleagues demonstrated that improvements in mortality previously attributed to patient safety systems, such as RRT, were also occurring in hospitals without such interventions. Childhood avoidable mortality in the UK (causes of death such as from infection or treatable conditions such as appendicitis) has fallen from 2726 in 2001, to 1902 in 2010 and then to 1473 in 2019. This trend infers that while PEWS have a role to play there are other systematic factors impacting mortality rates. Where mortality has been used as an outcome measure in evaluating the effectiveness of PTTT (and for comparison in adults there were 293 000 inpatient deaths in 2019) the studies are mixed, which has led to debate about their utility. A cluster randomised control trial comparing 10 hospitals in which the bedside PEWS (a PTTT) was implemented, with 11 control hospitals where there was no PTTT, failed to demonstrate an impact on mortality. During this study, that examined over 500 000 inpatient days in 150 000 children, there were fewer than two observed deaths per 1000 hospital discharges across both arms, and in approximately half of these deaths, ‘do not attempt resuscitation’ orders were in place. There has been one further randomised control trial published (comparing two different PTTT Scores) which concluded the results should be interpreted with caution given the low rate of clinical deterioration (only 22 unplanned transfers of care in 31 337 admissions to the recruiting paediatric hospitals). Therefore, it is reasonable to conclude the use of mortality as the primary outcome measure in these studies of PTTTs effectiveness is not robust. While mortality is infrequent, it is catastrophic for both families and staff, so it is important we understand how these interventions may impact on the variety of processes that may lead to this tragic outcome. PEWS are a complex healthcare intervention, and it is difficult to determine which components impact on the quality of clinical processes, and the mechanisms by which each part exerts its effects. National reports continue to highlight the need for quality improvement in recognising and responding to inpatient deterioration and therefore healthcare organisations and regulators need to be able to assess which components of PEWS are effective across a range of process and outcome measures. Due to the focus on mortality as the end point in the deterioration pathway, some safety mechanisms that work well before deterioration and therefore avoid intensive care, may have been missed. There is a paucity of early or intermediate outcomes. Where they have been developed, they are rarely used as primary outcome measures. It is likely directing all improvement efforts to solely preventing mortality (as has legitimately occurred in adult practice) may impact on a host of balancing measures which affect productivity and resources. Our recent evidence-based, theoretically informed, Paediatric early warning system - Utilisation and Mortality Avoidance. improvement programme (PUMA Programme) was developed and implemented in two general hospitals (no onsite PICU) and two tertiary hospitals (with onsite PICU) in the UK. We developed a composite metric (adverse events) as a primary quantitative outcome representing the number of children monthly that experienced one of the following: mortality, cardiac arrest, respiratory arrest, unplanned admission to PICU or unplanned admission to a higher dependency unit . Despite implementation challenges, all made contextually appropriate system changes with a decline in the adverse event rate at three sites. At the site in which system changes were organisationally adopted, this decline was significant. The variable impact on adverse rates highlights the dynamic qualities of PEWS. As an example, the introduction of an electronic EWS at one of the sites strengthened medical access to patient data but disrupted nursing work as there were insufficient computers available to allow nurses to enter vital signs, leading to a delay between monitoring and recording the PTTT Scores. The lack of system-wide data demonstrating a specific impact of PEWS on mortality does not mean they are not effective for other outcomes, but that these outcomes have yet to be measured or realised. Furthermore, numerous different processes might contribute to an overall benefit, but a given individual intervention might not reach significance. These interventions may be relatively simple, from the production of minimum training standards for staff who will be involved in caring for patients on PEWS pathways, to complex electronic whiteboard systems which can highlight deterioration of patients, to command control centres remote from wards. Some interventions may also be embedded in processes which already occur, such as a morning handover. These may not previously have been regarded as interventions but will probably have been recognised as being beneficial by staff and institutions. Another example of an unrecognised intervention is that of situational awareness. This encompasses processes such as bringing staff together in huddles and identifying ‘watcher’ patients through shared communication processes. Watcher patients are those identified through nursing staffs’ tacit knowledge; for example, the identification of a patient who appears well at present, but the nurse’s experience dictates they are at risk of deterioration, even though the nurse cannot completely explain exactly why. Understanding whether, and more importantly how, these interventions take place needs a system assessment requiring more of a focused evaluation than a simple ‘yes/no’ questionnaire. A clinical area may claim they don’t use watchers, but this may be what staff are doing; conversely both clinicians and operational managers may claim they have strong situational awareness, but in reality, might be dependent on traditional communication hierarchies. Finally, it is also important there are the right number of staff, with the right tools, to enable them to perform the right task, at the right time. The availability (or absence of) equipment to effectively monitor a patient (such as functional pulse oximeters and correct sized blood pressure cuffs, etc) clearly impacts on the ability to detect and plan for a child becoming unexpectedly unwell. Adequate staffing is an important safety factor in itself and demand for child health services outstrips capacity with workforce issues being ‘a significant challenge to child health service delivery and improvement’ according to the latest Royal Collage of Paediatrics and Child Health review of the medical and nursing workforce. Interventions which aid parents’, carers’ and professionals’ ability to raise concerns about imminent deterioration in a child, and processes to measure this, would also provide an organisation with knowledge of this important facet of care. Organisational cultures have a significant impact on outcomes and actively demonstrating that patients, families and carers are involved in, and feel able to share concerns around, decision making is an important outcome in itself. Discrete outcomes such as mortality or ICU admission are easy to measure but other interventions, such as improving and maintaining situational awareness at a patient, ward and organisational level are also likely to have a positive impact on deterioration. They may also improve negative hierarchical cultures (an often unexplored but significant influencer on care) and therefore are an important part of clinical practice. The absence of studies examining the link between these organisational and human factor characteristics and specific outcomes is something that needs to be addressed with further study. A scoping review of PEWS undertaken in 2021 reiterated the need to understand both the ‘technical’ nature of any score and the wider social, cultural and organisational context in which they are deployed. Our PUMA programme assessment tool evaluates a system’s ability to Detect, Plan and Act in response to the deteriorating child. It provides a detailed description of what needs to be in place for children’s deterioration to be detected and acted on. This propositional model detailed in contains interventions, tasks and skill sets that are not always possible to link to specific patient outcomes but represent a positive safety culture in a particular ward environment. Constructs such as a no blame environment, empowerment, communication, situational awareness, psychological safety, clear leadership, closed loop feedback, teaching and so on are persistently recognised as being of importance. Certainly there is evidence from our qualitative examination of these systems that PEWS aid staff in having a common language for communication about deterioration. This might explain why PEWS have continued to spread despite equipoise on their utility and is a reason why national scores exist in Scotland and Ireland. The English National Health Service has started a schedule of activity to introduce a standardised inpatient PEWS chart. This work is part of an overarching System-wide Paediatric Observation Tracking (SPOT) programme. The utilisation of standardised outcomes and end points, being developed as part of the programme, will hopefully avoid further heterogeneous results which will be difficult to interpret. An expert working group has taken a list of potential outcome measures derived from a previous systematic review and refined it to create a panel which will be measured in real time, both nationally and regionally, as the inpatient implementation rolls out. Following feasibility and usability exercises undertaken in a small group of hospitals a prototype chart has been rolled out in a number of pilot sites. This prototype chart has been developed via consensus opinion from multiple specialties with medical and nursing inputs. A community of practice has been created with weekly meetings examining enablers and barriers to effective de-implementation (all pilot sites already have a PEWS system in place and so this exercise is in changing rather than generating new practice). The pilot chart is continually being refined as a result of feedback from the pilot sites and a version 4.0 will be released in the late summer of 2022. The new version will undergo further real-world testing before a final version will be released later in 2022 for early adopter sites to implement. This programme of work has the support of National Health Service England (NHS England), The Royal College of Paediatrics and Child Health and the Royal College of Nursing with further information available via the Royal College of Paediatrics and Child Health (RCPCH) website. Box 1 Outcomes used in all the paediatric early warning systems (PEWS) ‘validation studies’ Death Hospital-wide deaths per 100 discharges Hospital-wide deaths per 1000 discharges Ward deaths only per 1000 admissions All-cause hospital-wide mortality rate Deaths on paediatric intensive care unit (PICU) PICU deaths for all ward transfers PICU mortality rates among readmitted patients in 24 hours Cardiac arrests/respiratory arrest Near cardiopulmonary arrests Cardiac arrest Actual cardiopulmonary arrests Unexpected cardiac arrests Ward respiratory arrests per 1000 patient days Preventable cardiac arrests PICU/PHDU Unplanned transfer to PICU (±within 24 hours of admission) Unplanned PICU transfers per 1000 patient days Critical deterioration events (life-sustaining interventions administered within 12 hours of PICU admission) Invasive ventilation given to emergency admissions to PICU postintervention Early intubation Postintervention rates of PICU admissions receiving mechanical ventilation Unplanned transfer to Paediatric High Dependancy Unit (PHDU) ±within 24 hours of admission) Urgent consult/review assistance/urgent review Code calls Urgent calls to inhouse paediatrician Urgent calls to respiratory therapist Rapid response team (RRT) call Outreach team calls (Phone call advice/consults to PICU team) Specific ‘Intervention’ with RRT—on ward Composite measures Combined cardiac and pulmonary arrests Critical deterioration index (non-invasive or invasive ventilation and/or inotropic support within 12 hours after admission) Critical deterioration index (non-invasive or invasive ventilation and/or inotropic support within 24 hours of admission) Hospital-wide deaths per 100 discharges Hospital-wide deaths per 1000 discharges Ward deaths only per 1000 admissions All-cause hospital-wide mortality rate Deaths on paediatric intensive care unit (PICU) PICU deaths for all ward transfers PICU mortality rates among readmitted patients in 24 hours Near cardiopulmonary arrests Cardiac arrest Actual cardiopulmonary arrests Unexpected cardiac arrests Ward respiratory arrests per 1000 patient days Preventable cardiac arrests Unplanned transfer to PICU (±within 24 hours of admission) Unplanned PICU transfers per 1000 patient days Critical deterioration events (life-sustaining interventions administered within 12 hours of PICU admission) Invasive ventilation given to emergency admissions to PICU postintervention Early intubation Postintervention rates of PICU admissions receiving mechanical ventilation Unplanned transfer to Paediatric High Dependancy Unit (PHDU) ±within 24 hours of admission) Code calls Urgent calls to inhouse paediatrician Urgent calls to respiratory therapist Rapid response team (RRT) call Outreach team calls (Phone call advice/consults to PICU team) Specific ‘Intervention’ with RRT—on ward Combined cardiac and pulmonary arrests Critical deterioration index (non-invasive or invasive ventilation and/or inotropic support within 12 hours after admission) Critical deterioration index (non-invasive or invasive ventilation and/or inotropic support within 24 hours of admission) Healthcare organisations, academic institutions and regulators recommending PEWS as an answer to the problem of inpatient deterioration without understanding all the mechanisms of their action will deliver research and implementation studies with equivocal findings. The rapid spread of PEWS over the last decade is testament to their positive face validity and undoubtedly they have contributed to an improved safety culture. Ultimately we have not yet identified the specific ingredients and mechanisms by which PEWS work as we don’t have the outcomes by which to test different recipes. To improve outcomes for children in inhospital settings we must change the emphasis from rare outcomes (mortality) and examine and measure a wide range of interventions that may aid early detection of deterioration. This work has arisen from a study/project funded by the National Institute for Health Research, Health Services and Delivery Research Programme, funder reference: 12/178/17. The views expressed are those of the author(s) and not necessarily those of the National Institute for Health Research or the Department of Health and Social Care.
Faster than light (microscopy): superiority of digital pathology over microscopy for assessment of immunohistochemistry	83f96146-699b-4a0c-9f39-c4b972a03bff	10176378	Anatomy[mh]	Over the past few years, digital pathology has been deployed for primary diagnosis in a few flagship projects around the world. Our institution now routinely digitally scans 100% of the cases coming through the department and it has become clear that these systems are able to offer numerous benefits to pathology departments such as improved workflow, reduced impact of human error and increased efficiency in diagnostic work. However, there is still insufficient research evaluating the likely advantages of digital pathology, which, in combination with poorly integrated software and high initial cost outlay, has hindered uptake particularly in smaller institutions. Of late, much of the research has been concerned with diagnostic accuracy as the introduction of a new technology must not compromise patient safety. Many single papers and a systematic review have confirmed non-inferiority of the digital microscope when compared with the light microscope. Additionally, two systems have been approved by the US Food and Drug Administration for primary diagnosis. However, accuracy is not the only aspect for concern—time taken to reach a diagnosis is also of great importance. The longer it takes to reach a diagnosis, the more effort is required by the pathologist, resulting in a reduction in their productivity. This is critical given the increasing demand for pathology services alongside a dramatic increase in retirement rate of pathologists within the UK, with only 3% of pathology departments being fully staffed. Preliminary work from our group showed that early whole slide imaging (WSI) viewers were 60% slower than the microscope, which posed a major barrier to adoption. This work led to the development and design of WSI software to focus on the need for fast viewing—the Leeds Virtual Microscope. We have shown that a digital microscope could be as quick as a light microscope for diagnostic purposes but was not faster, both using a wall-sized display and an 8-megapixel desktop setup, applied to a variety of diagnostic tasks. A more recent development of this software though allows for simultaneous viewing of multiple sections of tissue for comparison, something that is simply not possible with light microscopy. With as little as 61.9% of a pathologist’s time spent viewing an image when at the microscope, the rest of the time being consumed with manual processes such as removing slides from slide trays, adjusting slides on the microscope stage and dictating the report, we anticipate that viewing multiple slides side by side will be of significant time benefit to pathologists. It will be particularly beneficial in large resection cases, or cases where there are multiple immunohistochemical-stained slides (approximately 13% of cases within our institution). We therefore designed an experiment to compare the time with diagnosis using a digital system with the microscope. The digital system minimises the effort to manually switch between separate slides, instead offering the user a one-touch method of reviewing the case. It is hypothesised that the use of an appropriately designed digital pathology workstation can offer a reduction in time to reach a diagnosis without compromising diagnostic confidence. A purposive sampling strategy was employed to recruit 16 participants from within our institution: 8 senior trainee histopathologists and 8 consultant histopathologists from a range of subspecialist fields (not including liver pathology). Pathologists were recruited in person and those included were those who were approached and were willing to be involved. This study took place prior to the digitisation of our department resulting in a wide range of experience of using a digital microscope, from very little to many days’ cumulative experience. All participants were asked to view three cases using the digital microscope, and three different cases using the light microscope. The order of cases and interface was fully counterbalanced. All cases were liver needle core biopsies of tumours, with clinical details available in . Liver biopsy cases were chosen since many tumours are metastatic and therefore require a large panel of immunohistochemical stains to identify the location of the primary tumour. The cases were selected from archives at our institution and reviewed by a consultant histopathologist (DT). Three cases were designated as ‘set A’ and the other three as ‘set B’. Equal numbers of slides were included in each set. Details of these cases can be found in . 10.1136/jclinpath-2021-207961.supp2 Supplementary data Each case contained one or more H&E-stained slides, as well as multiple slides stained with immunohistochemical stains that were relevant to the case. An example of the slides for a case can be seen in below. Before the experiment was undertaken, each participant was given a 15-minute training session using the digital microscope. A standard training session was divided into three sections: section one in which the researcher would show the participant how to use the software, section two where the participant would use the software themselves and the researcher would evaluate their use of it, and section three where they were asked to perform a diagnostic task. All trials took place in a quiet, windowless room in the histopathology department of our institution usually used for teaching purposes. The digital microscope was placed at one end of the room with a light microscope set up on an adjacent table. The only light source was a standalone lamp placed next to the door behind the participant, to standardise the effect of ambient lighting on the display as far as possible at 10 lux. A Dell Precision T5500 with AMD W5000 graphics card was used for this experiment. A Barco Coronis Fusion (6 MP) display (Barco Limited, Kortrijk, Belgium) was used in conjunction with a Barco Nio (2 MP) display (Barco Limited, Kortrijk, Belgium). The larger 30-inch screen was a split screen setup, the left side displaying the H&E slide (which was in constant display) and the right displaying the immunohistochemical slide. The smaller screen 21-inch on the right side of the participant displayed thumbnail images of all slides and highlighted which slide was currently in view. Viewing software was the Leeds Virtual Microscope. All digital slides were scanned on an Aperio T3 scanner (Leica Biosystems UK Limited, Milton Keynes, UK) with a ×40 objective lens at 0.25 μm per pixel. Images were compressed with conventional JPEG compression. Participants were able to pan a slide using a click and drag method. The keyboard was used to zoom and change slide. A screenshot of the digital microscope display can be seen in . The light microscope was a Leica DMR microscope (Leica Biosystems UK Limited, Milton Keynes, UK) with ×2.5, ×5, ×10, ×20, ×40 and ×100 objectives and ×10 eyepiece. The microscope lens was reset by the researcher to the lowest magnification at the start of each trial (×2.5 magnification). The participants were provided with a practice slide in order to familiarise themselves with the microscope prior to the experiment. The experiment was recorded using a three-video camera setup: one captured a ‘down-the-microscope’ view from the microscope camera mount, one captured the microscope work area (placed in the corner of the room furthest from the microscope) and the other directly captured the participants’ face and body from in front while at the digital workstation (placed directly behind the digital pathology workstation). Timing began when they picked up the first glass slide and finished when the set was completed. The video data were analysed to capture each instance of the participant interacting with the slide, panning, zooming, using the microscope condenser and writing notes. A screenshot of the participant video recording with the synchronised view down the microscope can be found in . 10.1136/jclinpath-2021-207961.supp1 Supplementary data Statistical analysis was performed in Stata V.16. Data approximated a normal distribution and therefore are summarised by the mean and SD. As analyses indicated a wide variation in time to diagnosis according to the case, a normalised time to diagnosis (mean time to diagnosis for a case was calculated across both interfaces and then individual time to diagnosis on each interface expressed as a percentage of this time) is reported as the primary outcome measure. We also report actual time to diagnosis for ease of understanding. Multiple linear regression was used to estimate the time to diagnosis in minutes adjusting for the binary fixed effects of experience level (trainee vs consultant) and interface (light vs digital microscope). CIs were generated to the 95% level. A sensitivity analysis of variance (ANOVA) was also performed on normalised time with the within-subject factor being light microscope versus digital microscope and between-subject variable being experience level (trainee vs consultant). A summary of the main results can be seen in ; the normalised time results mirrored those of actual time across all outcomes. In terms of overall results, the mean time to diagnosis was 4 min 3 s using the digital microscope and 5 min 24 s using the light microscope, as shown in . This equates to a time-saving using the digital microscope of 1 min 21 s (95% CI 16 s to 2 min 26 s; p=0.02) (bootstrapped p=0.009). Overall, normalised mean time to diagnosis was 85% on the digital pathology workstation compared with 115% on the microscope; a relative reduction of 26% (95% CI 15% to 45%; p=0.0006), as can be seen in . When subcategorising the results by experience, the mean time to diagnosis for trainees using the digital microscope was 3 min 31 s, and 5 min 25 s using the light microscope. This equates to a time-saving using the digital microscope of 1 min 54 s (95% CI −3 min 11 s to −0 min 37 s; p=0.007). The mean time to diagnosis for consultants using the digital microscope was 4 min 35 s, as compared with 5 min 24 s using the light microscope. This results in a non-statistically significant time-saving using the digital microscope of 0 min 48 s (95% CI −2 min 41 s to 1 min 5 s; p=0.37). The normalised mean time to diagnosis for trainees was 74% and 116% compared with consultant times of 96% on the digital microscope and 114% on the light microscope, respectively. This equates to a reduction of 42% (95% CI 14% to 70%; p=0.006) for trainees, and again a non-statistically significant reduction of 18% (95% CI −53% to 17%; p=0.3) for consultants. When evaluating the results by set, the mean time to diagnosis for set A across all participants using the digital microscope was 3 min 13 s, and 5 min 11 s using the light microscope. This equates to a time-saving using the digital microscope of 1 min 58 s (95% CI −3 min 18 s to −0 min 36 s; p=0.008). The mean time to diagnosis for set B across all participants using the digital microscope was 4 min 53 s, and 5 min 38 s using the light microscope, a difference which was not statistically significant (95% CI −2 min 24 s to 0 min 54 s; p=0.35). The mean normalised time to diagnosis for set A was 77% compared with 123%, set B 93% compared with 107% on the digital microscope and the light microscope, respectively. This equates to a statistically significant reduction in normalised time to diagnosis per case for set A of 47% (95% CI 14% to 79%; p=0.008), but again a non-statistically significant difference of 14% (95% CI −45% to 18%; p=0.37). When combining the data across both interfaces, the data are summarised in . The mean time to diagnosis by trainees was 4 min 29 s, as compared with consultants with a mean time of 5 min 0 s. Therefore, trainees were faster across both modalities by a mean time of 31 s, but this difference was not significant (95% CI –1 min 42 s to 0 min 40 s; p=0.4). Similarly, when combining data across both interfaces, the mean time to diagnosis for set A cases was 4 min 13 s, and set B was 5 min 16 s, which represents a non-significant mean difference of 1 min 6 s (95% CI –2 min 11 s to 0 min 4 s; p=0.07). Results of multivariable linear regression adjusting for experience level were largely the same as the unadjusted results presented above. Time to diagnosis on the digital microscope was 1 min 21 s faster than on the light microscope (95% CI 0 min 16 s to 2 min 26 s; p=0.017). Consultants took 31 s longer to reach diagnoses than trainees, but this was not statistically significant (95% CI –0 min 34 s to 1 min 36 s; p=0.34). There was an adjusted reduction in normalised time to diagnosis of 30% on the digital microscope as compared with the light microscope (95% CI –52% to −8%; p=0.008). There was an adjusted reduction in normalised time to diagnosis by trainees of 10% as compared with consultants but this was, again, non-significant (95% CI −32% to 11%; p=0.34). Similarly, there were no notable differences in the results of the sensitivity ANOVA performed on actual and normalised time, the details of which can be found in . 10.1136/jclinpath-2021-207961.supp3 Supplementary data There were no major diagnostic errors made on either interface. Two participants gave a discordant diagnosis of hepatocellular carcinoma in a case of probable metastatic carcinoma—in set B. Review of the case revealed some cytological features that might support such a diagnosis and further immunohistochemistry (IHC) might be required to entirely rule it out, making the true diagnosis ambiguous. Therefore, either diagnosis was acceptable for the purposes of this study. The significant difference in time to diagnosis demonstrates that for this set of diagnostic tasks, the digital microscope was quicker with a time-saving of 1 min 21 s per case, or a 26% relative reduction in time to diagnosis. As far as we are aware, this is the first study evaluating the impact of the digital microscope on time to diagnosis for cases involving immunohistochemically stained slides. A mean reduction of 1 min 21 s per case using the digital microscope is a considerable time-saving in a health service that is struggling under ever-increasing demand for services. Previous work from our group has estimated the number of cases in our institution requiring extra stains to be approximately 5%. Given that our institution handles approximately 60 000 surgical cases per year, the use of the digital microscope for these cases alone may result in a time-saving of 67.5 histopathologist hours over the course of a year within our institution. This saving is likely to be higher in institutions where immunohistochemical or special stains are used more frequently. The effect of increased experience of many of the specialist trainee participants with digital system was possibly reflected in the results, with the trainee cohort being just over 1 min quicker than the consultants on the digital pathology workstation per case, but this difference was not statistically significant (−1 min 4 s, 95% CI −2 min 49 s to 0 min 41 s; p=0.21). Alternatively, this may reflect the fact that trainee histopathologists were more familiar with liver specimens, as no consultant liver histopathologists were included in this study. However, no major diagnostic errors were made, thus suggesting unfamiliarity had no overall impact. Moreover, the use of only liver biopsies across the two interfaces will have prevented unfamiliarity with the type of case from biasing the results. The time to diagnosis for set B was longer than the time to diagnosis for set A, although not statistically significant (mean difference of −1 min 6 s, 95% CI –2 min 11 s to 0 min and 4 s; p=0.07). This may have been due to some diagnostic difficulty surrounding one case; a diagnosis of hepatocellular carcinoma was made twice during the trial and given as a differential diagnosis once. If this was given only on the digital microscope, this may pose questions regarding image quality and the regulations regarding the use of digital slides for diagnostic work. However, these diagnoses were made on both interfaces: once on the digital microscope and the second on the light microscope. This would lead us to believe that these mistakes were due to the diagnostic difficulty surrounding of that case, as opposed to issues regarding image fidelity on the digital microscope. Participants largely reported positive experiences using the digital microscope. A large proportion of participants used the system as intended and as was shown in the training session. However, some opted for a different technique, and rather than using the digital microscope at low power to identify an area of interest and then zoom, many participants used the digital microscope at very high magnification and scrolled the whole length of the image. This technique is inevitably time-consuming and may be due to unfamiliarity of the digital microscope for some participants; time reduction per case will likely increase with continued use and increased experience. Fourteen participants reported that they found the two-screen digital microscope very useful for comparing areas of tissue side by side and aiding a diagnosis. Two participants commented on the controls and the ergonomics of the mouse and keyboard design and felt that a less cumbersome method of panning and switching between slides needed to be implemented. One participant commented that this implementation of a digital microscope to look at immunohistochemical cases was the best they had used to date. It is unsurprising that comments were not completely unanimous regarding the digital microscope workstation; it is well known that there is not a ‘one-size-fits-all’ workstation for digital radiologists. The design features of the digital microscope are a major strength to this study. The use of medical grade displays with high technical specifications enables more tissue to be viewed on the screen at one time reducing the need for interaction by the histopathologist (pan/zoom) for this low power assessment task; an 8-megapixel display shows approximately the same amount of tissue as a ×10 light microscope eyepiece. A high luminance and contrast ratio increases, the ability of the pathologist to make a confident diagnosis at low magnification. This, in combination with a viewer that facilities fast viewing, is likely responsible for the considerable time-saving afforded by the digital microscope in this study as this reduces refreshing time. Additionally, side-by-side viewing reduces cognitive load and streamlines the process of referencing the H&E WSI to check an area for relevance; light microscopy demands that pathologists remove the IHC glass slide, load the H&E glass slide and navigate to the area of interest, which is challenging and time-consuming. Previous work from our group did not demonstrate a time-saving from increased screen resolution, but the discrepancy between those results and this study is likely due to three main factors. First, that work involved a very specific search task (identifying micrometastases), as opposed to a more general low power assessment in this work. Second, the high-resolution displays were split across three screens in the previous work; the bevels were a hinderance to pathologists as they had to ensure that the micrometastasis was not being obscured by the bevels. Third, this study involved updated viewing software that was faster to respond to the user. Another study by Hanna et al found that the digital microscope was 19% slower than using the light microscope. There are again many reasons why our results were not in agreement with their findings. First, they included relatively inexperienced users; a large-scale validation study by our group demonstrated that experience of between 2 and 6 months is required for users to become proficient. Second, they employed the use of small monitors (24”) with relatively low resolution (1920×1200), equating to just 2.3 MP. Third, they used a custom Graphical User Interface which can present many difficulties in the initial phases and evolve over time, as outlined during our development of the Leeds Virtual Microscope which was initiated back in 2007. A very recent study by Borowsky et al found that the digital microscope took an average of 5.20 min as compared with 4.95 min on the light microscope. However, again, there are reasons for this discrepancy with our findings; this study included only 25% slides that were IHC or special stains and does not mention the digital microscope setup other than the use of Dell medical grade monitors. In our experience, many factors (user experience and training, details of the user interface design, task choice and technical display specifications) can all affect time to diagnosis. Objective comparison of these factors is difficult as there are complex interactions between them. Mills et al found that when including a range of surgical specimens, digital diagnosis took 4 s longer than the light microscope, but as highlighted by the authors the slowest reader got considerably quicker with digital diagnoses over the course of the study, and was similar to the light microscope by the end of the study. This demonstrates nicely that all studies of this nature are biased in favour of the light microscope due to relative inexperience with their digital counterparts. It also highlights the need to include suitable training in future longer-term studies of efficiency or time to diagnosis. The effect of working digitally does not just result in potential time-savings in time to diagnosis, but instead impacts the entire laboratory workflow. Although it is outside of the scope of this work to discuss the impact of a whole system evaluation of digitised pathology services, this has recently been addressed by the work of Baidoshvili et al . They focus on the time-savings across the entire pathology workflow when comparing analogue with digital rather than just the time to diagnosis and found that there were time-savings of approximately 19 hours within a working day across a pathology laboratory, equating to 2.63 full-time equivalent staff. Further work should be conducted prospectively on the cost:benefit as departments become fully digital. Inevitably, there were several limitations to our work. First, this was a small study with known considerable user variation. Second, the time difference to diagnosis will be affected by the participant’s familiarity with the case type. Although this should not impact the primary outcome in this study (difference between the two interfaces), having participants diagnosing case types that they are familiar with may be more reflective of the time-savings observed in routine clinical practice. It should also be noted that many routine cases do not require immunohistochemical slides to review and therefore the observed time-saving may not be applicable to these cases. Third, it would be advantageous to spend longer familiarising participants with the digital microscope; familiarity with one interface and not the other will inevitably bias the results in favour of the familiar interface. Lastly, due to the prototype nature of the digital microscope, the usability could be improved. There were some issues regarding slide registration (alignment of the H&E and immunostain images), which proved particularly problematic for participants who were less adept with the digital microscope. Further, time-saving will be likely observed as the digital microscope becomes more user-friendly. To conclude, the digital microscope reduced time per case by 1 min 21 s per case and a relative reduction of 26%, without any major diagnostic errors as compared with the light microscope. This is likely due to the ability to view multiple slides simultaneously, which is not possible using analogue systems. We anticipate that these time-savings will have a major improvement on pathologist productivity at a time where pathology services are strained, and serve as a point from which to build other user interfaces to enhance pathologist productivity. Take home messages Digital pathology may offer benefits over microscope in viewing whole slide images—one unique capability is the ability to review side by side, with synchronised pan and zoom. We designed a pragmatic study looking at evaluation of liver biopsy cases including an immunohistochemical panel, where serial comparisons are needed. We used a custom viewer that allowed side-by-side viewing, and rapid review of immunohistochemical images in the sequence. We found that mean time was 5 min 24 s on the light microscope and 4 min 3 s on digital microscope, a reduction of 1 min 21 s (95% CI 16 s to 2 min 26 s; p=0.02) and a relative reduction of 26%. These benefits were seen with relatively little training and exposure to the system, and further work is needed to evaluate the real-world impact.
ESGO/ESTRO/ESP Guidelines for the management of patients with cervical cancer – Update 2023*	ca2c709d-df87-418c-b711-090262ecf933	10176411	Internal Medicine[mh]	Cervical cancer is a major public health problem, ranking as the fourth most common cause of cancer incidence and mortality in women worldwide. There are geographical variations in cervical cancer that reflect differences particularly in the prevalence of human papillomavirus (HPV) infection and inequalities in access to adequate screening and treatment. Cervical cancer is uncommon in Europe but still remains the most frequent cause of cancer death in middle-aged women in Eastern Europe. Other epidemiologic risk factors associated with cervical cancer are notably a history of smoking, oral contraceptive use, early age of onset of coitus, number of sexual partners, history of sexually transmitted disease, certain autoimmune diseases, and chronic immunosuppression. Squamous cell carcinomas account for approximately 80% of all cervical cancers and adenocarcinoma accounts for approximately 20%. The WHO recently launched a global initiative to scale up preventive, screening, and treatment interventions relying on vaccination against HPVs, screening and treatment of detected cervical pre-invasive and invasive lesions, and offering the best possible curative care to women diagnosed with invasive cancer. As part of its mission to improve the quality of care for women with gynecological cancers across Europe, in 2018 the European Society of Gynecological Oncology (ESGO) jointly with the European Society for Radiotherapy and Oncology (ESTRO) and the European Society of Pathology (ESP) published evidence-based guidelines to improve the management of patients with cervical cancer within a multidisciplinary setting. Given the large body of new evidence addressing the management of cervical cancer, the three sister societies jointly decided to update these evidence-based guidelines and to include new topics in order to provide comprehensive guidelines on all relevant issues of diagnosis and treatment in cervical cancer. These guidelines are intended for use by gynecological oncologists, general gynecologists, surgeons, radiation oncologists, pathologists, medical and clinical oncologists, radiologists, general practitioners, palliative care teams, and allied health professionals. Even though our aim is to present the highest standard of evidence in an optimal management of patients with cervical cancer, ESGO, ESTRO, and ESP acknowledge that there will be broad variability in practices between the various centers worldwide. Moreover, there will also be significant differences in infrastructure, access to medical and surgical technology, and also training, medicolegal, financial, and cultural aspects that will affect the implementation of any guidelines. These guidelines are a statement of evidence and consensus of the multidisciplinary development group regarding their views and perspective of currently accepted approaches for the management of patients with cervical cancer. Any clinician applying or consulting these guidelines is expected to use independent medical judgment in the context of individual clinical circumstances to determine any patient’s care or treatment. These guidelines make no representations or warranties of any kind whatsoever regarding their content, use, or application and disclaim any responsibility for their application or use in any way. The guidelines were developed using a five-step process defined by the ESGO Guideline Committee (see ). The strengths of the process include creation of a multidisciplinary international development group, use of scientific evidence and international expert consensus to support the guidelines, and use of an international external review process (physicians and patients). This development process involved three meetings of the international development group, chaired by Professor David Cibula (First Faculty of Medicine, Charles University and General University Hospital, Prague, Czech Republic), Professor Jacob Christian Lindegaard (Aarhus University Hospital, Aarhus, Denmark), and Professor Maria Rosaria Raspollini (University of Florence, Florence, Italy). To serve on the expert panel, ESGO/ESTRO/ESP nominated practicing clinicians who are involved in managing patients with cervical cancer and have demonstrated leadership through their expertise in clinical care and research, national and international engagement and profile as well as dedication to the topics addressed. The objective was to assemble a multidisciplinary development group and it was therefore essential to include professionals from relevant disciplines (gynecological oncology and gynecology, medical, clinical and radiation oncology, pathology) to contribute to the validity and acceptability of the guidelines. To ensure that the statements were evidence based, the current literature was reviewed and critically appraised. A systematic, unbiased literature review of relevant studies published between January 2017 and March 2022 was carried out using the MEDLINE database (see ). The literature search was limited to publications in English. Priority was given to high-quality systematic reviews, meta-analyses, and randomized controlled trials, but studies of lower levels of evidence were also evaluated. The search strategy excluded editorials, letters, and in vitro studies. The reference list of each identified article was reviewed for other potentially relevant articles. Based on the collected evidence and clinical expertise, the international development group drafted guidelines for all the topics. The updated guidelines were retained if they were supported by a sufficiently high level of scientific evidence and/or when a large consensus among experts was obtained. An adapted version of the “Infectious Diseases Society of America–United States Public Health Service Grading System was used to define the level of evidence and grade of recommendation for each of the recommendations (see ). In the absence of any clear scientific evidence, judgment was based on the professional experience and consensus of the international development group. 10.1136/ijgc-2023-004429.supp2 Supplementary data ESGO/ESTRO/ESP established a large multidisciplinary panel of practicing clinicians who provide care to patients with cervical cancer to act as independent reviewers for the updated guidelines. These reviewers were selected according to their expertise, had to be still involved in clinical practice/research, and were from different European and non-European countries to ensure a global perspective. Patients with cervical cancer were also included. The independent reviewers were asked to evaluate each recommendation according to its relevance and feasibility in clinical practice (only physicians), so that comprehensive quantitative and qualitative evaluations of the updated guidelines were completed. Patients were asked to evaluate qualitatively each recommendation (according to their experience, personal perceptions, etc.). Evaluations of the external reviewers (n=155) were pooled and discussed by the international development group to finalize the guidelines’ updating process. The list of the 155 external reviewers is available in . The guidelines detailed in this article cover staging, management, follow-up, long-term survivorship, quality of life and palliative care. Management includes fertility sparing treatment, early and locally advanced cervical cancer, invasive cervical cancer diagnosed on a simple hysterectomy (SH) specimen, cervical cancer in pregnancy, rare tumors, recurrent and metastatic diseases. A summary of evidence supporting the guidelines is included in , available online. 10.1136/ijgc-2023-004429.supp1 Supplementary data General Recommendations Centralization of care in specialized centers and referral network is encouraged [IV, B]. Treatment planning should be made on a multidisciplinary basis (generally at a tumor board meeting as defined in the ESGO quality indicators) and based on the comprehensive and precise knowledge of prognostic and predictive factors for oncological outcome, side effects, and quality of life [IV, A]. Patients should be carefully counseled on the suggested treatment plan and potential alternatives, including risks and benefits of all options [V, A]. Treatment should be undertaken by a dedicated team of specialists in the diagnosis and management of cervical cancers [IV, A]. Enrollment of patients with cervical cancer in clinical trials is encouraged [V, B]. Staging TNM Classification and FIGO Staging Patients with cervical cancer should be staged according to the TNM classification and the International Federation of Gynaecology and Obstetrics (FIGO) staging should also be documented [IV, A]. Systematic documentation and integration of the results from clinical examination, pathology and imaging including multidisciplinary team discussions of disparate findings is recommended [IV, A]. The method used to determine tumor status (T), lymph node (LN) status (N), and systemic status (M) should be noted (clinical, imaging, pathological) [IV, A]. Lymph node (LN) metastases should be classified according to the TNM classification [IV, A]. Prognostic Factors Systematic documentation of the following major tumor-related prognostic factors is recommended [II, A]: TNM and FIGO stage, including a maximum tumor size, detailed description of extracervical tumor extension (including uterine corpus involvement) and nodal involvement (eg, total number, location, size, and metabolic activity). Pathological tumor type including HPV status (see principles of pathological evaluation). Depth of cervical stromal invasion and a minimum thickness of uninvolved cervical stroma Margin status (ectocervical, endocervical, radial/deep stromal and vaginal cuff) Presence or absence of lymphovascular space involvement (LVSI). Presence or absence of distant metastases. Local Clinical and Radiological Diagnostic Work-up Pelvic examination and biopsy±colposcopy are mandatory to diagnose cervical cancer [II, A]. Pelvic magnetic resonance imaging (MRI) is mandatory for initial assessment of pelvic tumor extent and to guide treatment options (optional for T1a tumor with free margins after conization). Endovaginal/transrectal ultrasonography is an option if performed by a properly trained sonographer [II, A]. Cystoscopy or proctoscopy are not routinely recommended [IV, D]. Nodal/Distant Diagnostic Work-up In early stages managed primarily by surgery, surgical/pathological staging of pelvic lymph node (PLN) is the standard criterion to assess the prognosis and to guide treatment (except for T1a1 and T1a2 without LVSI) [III, A]. In locally advanced cervical cancer (T1b3 and higher (except T2a1) or in early-stage disease with suspicious LN on imaging), positron emission tomography-computed tomography (PET-CT), or chest/abdomen computed tomography (CT scan) (if PET-CT is not available) is recommended for assessment of nodal and distant disease [III, B]. PET-CT is recommended before chemoradiotherapy (CTRT) with curative intent [III, B]. Para-aortic LN dissection (PALND), at least up to inferior mesenteric artery, may be considered in locally advanced cervical cancer with negative para-aortic LN on imaging for staging purposes [IV, C]. Equivocal extrauterine disease should be considered for biopsy to avoid inappropriate treatment [IV, B]. Management of T1a Disease Diagnosis of T1a Disease Diagnosis of T1a cancer should be based on a conization (or excision) specimen examined by an expert pathologist with accurate measurement of depth of invasion, margin status, coexisting pathology, and reliable assessment of LVSI [IV, B]. Loop or laser conization is preferable to cold-knife conization in women wanting to preserve fertility. Care should be taken to provide an intact (unfragmented) specimen with minimal thermal artifact. The cone specimen should be oriented for the pathologist [IV, B]. Surgical margins of the cone specimen should be clear of both invasive and preinvasive disease (except for low-grade intraepithelial lesion) [IV, B]. Management of T1a1 Disease Management of patients with T1a1 disease should be tailored to the individual depending on age, desire for fertility preservation, histological type, and the presence or absence of LVSI [III, B]. In case of positive margins (except for low-grade intraepithelial lesion in ectocervix), a repeat conization should be performed to rule out more extensive invasive disease [IV, B]. LN staging is not indicated in T1a1 LVSI-negative patients but can be considered in T1a1 LVSI-positive patients. Sentinel lymph node (SLN) biopsy (without additional PLN dissection (PLND)) is recommended in this situation [IV, B]. Conization can be considered a definitive treatment as hysterectomy does not improve the outcome [IV, C]. Radical surgical approaches such as radical hysterectomy, trachelectomy or parametrectomy represent overtreatment and should not be performed for patients with T1a1 disease [IV, D]. Patients with T1a1 adenocarcinoma who have completed childbearing should be offered SH [IV, B]. Management of T1a2 Disease Conization (with clear margins) alone or SH is an adequate treatment for patients with T1a2 disease [IV, B]. Parametrial resection is not indicated [IV, D]. SLN biopsy (without additional PLND) can be considered in LVSI-negative patients but should be performed in LVSI-positive patients [IV, B]. Patients with T1a2 adenocarcinoma who have completed childbearing should be offered SH [IV, B]. Management of T1b1, T1b2, and T2a1 Tumors General Recommendations Treatment strategy should aim to avoid combining radical surgery and radiotherapy because of the high morbidity induced by the combined treatment [IV, A]. Negative LN on Radiological Staging - Surgical Treatment Radical surgery by a gynecological oncologist is the preferred treatment modality. Laparotomy is the standard approach for all procedures which include radical parametrectomy [I, A]. Minimally invasive approach may be considered only in low risk tumors (<2 cm and free margins after conization), in high-volume centers experienced in performing radical hysterectomy with minimally invasive surgery, which meet the ESGO quality criteria for surgery, if the patient agrees after comprehensive discussion about current evidence [IV, C]. LN assessment should be performed as the first step of surgical management [IV, A]. Minimally invasive surgery is an acceptable approach for LN staging [IV, B]. SLN biopsy before pelvic lymphadenectomy should be performed. Indocyanine green is the preferred technique [III, A]. A combination of blue dye with radiocolloid is an alternative technique [IV, B]. Intra-operative assessment of LN status (evaluated by frozen section) is recommended. Sentinel nodes from both sides of the pelvis and/or any suspicious LN should be sent for intra-operative assessment [III, A]. If any LN involvement is detected intraoperatively, further PLND and radical hysterectomy should be avoided. Patients should be referred for definitive CTRT [III, A]. PALND at least up to inferior mesenteric artery may be considered for staging purposes [IV, C]. After SLN biopsy, if SLN are negative on frozen section, a systematic pelvic lymphadenectomy should be performed as the standard LN staging [III, A]. If SLN is negative bilaterally in the pelvic level I area (below iliac bifurcation) LN dissection can be limited to level I [IV, B]. If SLN is not detected on either side, LN dissection should include on that particular pelvic side the removal of lymphatic tissue from all traditional regions including obturator fossa, external iliac regions, common iliac regions, and presacral region [III, A]. After frozen section, all SLN should be processed according to pathological protocol for ultrastaging (see the principles of pathological evaluation) [III, A]. The type of radical hysterectomy (extent of parametrial resection, type A-C2) should be based on the presence of prognostic risk factors identified preoperatively such as tumor size, maximum stromal invasion, and LVSI, which are used to categorize patients at high, intermediate, and low risk of treatment failure. A complete description of the template used for radical hysterectomy should be present in the surgical report. The 2017 modification of the Querleu-Morrow classification is recommended as a tool [IV, A]. Ovarian preservation should be discussed with women in reproductive age with squamous cell carcinoma, can be considered in HPV-associated adenocarcinoma and is not recommended for HPV-independent adenocarcinomas. Opportunistic bilateral salpingectomy should be performed if ovaries are preserved. Ovarian transposition should be discussed upfront with the patient and individualized according to risk balance [IV, A]. If a combination of risk factors is known at diagnosis, which would require an adjuvant treatment, definitive CTRT and brachytherapy (BT) should be considered without previous radical pelvic surgery [IV, A]. Negative LN on Radiological Staging – Alternative Treatment Options Definitive CTRT and image-guided brachytherapy (IGBT) represent an alternative treatment option [IV, B]. Neoadjuvant chemotherapy (NACT) or CTRT followed by surgery are not recommended [IV, D]. Adjuvant Treatment After Radical Surgery Adjuvant radiotherapy should be considered in the intermediate risk group (combination of risk factors at final pathology such as tumor size, LVSI, and depth of stromal invasion) [IV, A]. When an adequate type of radical hysterectomy has been performed in intermediate risk group patients, observation is an alternative option, especially in teams experienced in this approach [IV, B]. Adjuvant CTRT is indicated in the high-risk group (see principles of radiotherapy) [IV, A]: metastatic involvement of PLN (macrometastases pN1 or micrometastases pN1(mi)) on final pathologic assessment. positive surgical margins (vagina/parametria/paracervix). parametrial involvement. Additional BT boost as part of adjuvant CTRT can be considered in cases with vaginal and/or parametrial positive disease (see principles of radiotherapy) [IV, B]. Adjuvant treatment may be considered also if only isolated tumor cells are detected in SLN, although its prognostic impact remains uncertain [IV, C]. Fertility Sparing Treatment Fertility sparing therapy is an oncologically valid alternative to radical hysterectomy for young patients with cervical cancer <2 cm (squamous cell carcinoma and HPV-related adenocarcinoma) who want to preserve the option to have children. Before initiating fertility sparing therapy, consultation at an onco-fertility center and discussion in a multidisciplinary tumor board is recommended [III, B]. Counseling of eligible patients should encompass the oncologic and obstetric risks related to this type of management as well as the risk of fertility sparing therapy abandonment if there are positive resection margins or LN involvement [III, A]. Fertility-sparing treatment should be performed exclusively in gynaecological-oncological centers with comprehensive expertise in all types of these surgical procedures [IV, A]. Fertility-sparing treatment should not be recommended for uncommon and rare histological types/subtypes of cervical cancer with aggressive behavior including neuroendocrine carcinomas, HPV-independent adenocarcinomas and carcinosarcomas [V, D]. For patients who consider fertility sparing therapy, prognostic factors, clinical staging, and preoperative work-up do not differ from those not considering fertility sparing therapy (see above). Pelvic MRI and/or expert sonography are mandatory imaging tests to measure the non-involved cervical length (upper tumor free margin) and the remaining (after cone biopsy) cervical length [III, A]. Negative PLN status is the precondition for any fertility sparing therapy. Therefore, PLN staging (SLN) should always be the first step in each fertility-sparing therapy procedure. Identification of SLN and its ultrastaging is highly recommended. Any intraoperative suspicious LN (apart from SLN) should also be removed. If SLN cannot be detected on either pelvic side, a systematic pelvic lymphadenectomy should be performed on that side. Intraoperative assessment of LN status is highly recommended. All SLN from both sides of the pelvis and any suspicious LN should be sent for frozen section. LN staging is not indicated in T1a1 LVSI negative [III, A]. In case of intraoperatively proven PLN involvement, fertility-sparing surgery should be abandoned and patients should be referred for CTRT and BT [IV, B]. PALND, at least up to inferior mesenteric artery, may be considered for staging purposes [IV, C]. Ovarian transposition cannot be recommended in N1 status [IV, D]. The specific goal of fertility-sparing surgery must be resection of invasive tumor with adequate free margins and preservation of the upper part of the cervix [IV, A]. Intraoperative frozen section is a feasible way of assessing the upper resection margin [IV, C]. LN staging follows the principles of management of early stages [III, B]. Fertility sparing procedures comprise of conization (see ), simple trachelectomy (see ), radical (vaginal) trachelectomy (see ), abdominal radical trachelectomy (see ) [III, B]. Conization and simple trachelectomy are adequate fertility sparing procedures in patients with T1a1 and T1a2 tumors, regardless of LVSI status [IV, B]. Conization or simple trachelectomy are adequate fertility sparing procedures for T1b1, LVSI negative tumors. Radical trachelectomy is still an option [IV, B]. Radical trachelectomy (type B) should be performed in patients with cervical cancer T1b1, LVSI-positive. In patients without deep stromal involvement and with a high probability of adequate endocervical tumor free margins, simple trachelectomy can be considered [III, B]. Intraoperative placement of permanent cerclage should be performed during simple or radical trachelectomy [IV, B]. Fertility sparing therapy for patients with tumors greater than 2 cm is significantly associated with a higher risk of recurrence and should not be considered as a standard treatment. The risk of recurrence must be comprehensively discussed with the patient. NACT followed by radical vaginal trachelectomy and abdominal radical trachelectomy or cone has been described for fertility sparing treatment in patients with tumors >2 cm. PLN staging should be performed before starting NACT to confirm tumor-free LN. The optimal number of chemotherapy cycles, chemotherapy regimen as well as extent of cervical resection following NACT, are still a matter of debate [IV, B]. In more advanced cases, various fertility preservation proposals such as ovarian transposition (see ), oocyte-, embryo- or ovarian tissue preservation and egg donation should be discussed with the patient. The aim of the fertility preservation should be to offer the most efficient approach in accordance with the legal country-specific regulations, while not increasing the oncological risk [IV, B]. Any pregnancy following fertility sparing therapy should be considered as a high-risk pregnancy. Following simple or radical trachelectomy with placement of a permanent cerclage, delivery can only be performed by cesarean section [IV, B]. Although evidence is limited, several antenatal management tools can be considered following fertility sparing therapy including screening and treatment of asymptomatic bacteriuria, screening for cervical incompetence and progressive cervical shortening by transvaginal ultrasonography, fetal fibronectin testing, screening (and treatment) for asymptomatic vaginal infection, vaginal progesterone application, total cervical closure according to Saling and cervical cerclage, if not placed during trachelectomy [IV, C]. Routine hysterectomy after completion of childbearing is not mandatory [V, D]. Invasive Cervical Cancer Diagnosed on a Simple Hysterectomy Specimen General Recommendations Management of disease found after SH should be based on expert pathology review and discussed in a multidisciplinary tumor board. In general, management of occult disease follows the principles of the standard management, and is based on pathologic findings, and clinical staging. Treatment strategy should aim to avoid combining further surgery and radiotherapy because of the high morbidity after combined treatment [III, B]. Before making further management decisions, optimal imaging is necessary to evaluate the local and regional (nodal) disease status. Optimal imaging follows the same recommendations as that for the standard management [III, B]. When surgical staging of nodal disease is indicated (see below for details), it can be considered either as an isolated (preferentially laparoscopic) procedure or as the first step of surgical management in radiologic node negative patients. Surgical staging of nodal disease can also be considered to assess inconclusive nodes at imaging. SLN biopsy cannot be performed in the absence of the uterus. Any suspicious LN should be sent for intraoperative assessment (frozen section) [III, B]. Para-aortic LN dissection, at least up to inferior mesenteric artery, may be considered for staging purposes in patients with positive pelvic nodes at imaging, or at frozen section [IV, C]. Management of Patients with T1a1 and T1a2 Disease In patients with T1a1 tumor regardless of LVSI status and T1a2 tumor LVSI negative with clear margins in the hysterectomy specimen, no additional treatment is recommended [III, B]. Surgical LN assessment can be considered in T1a1 tumors with LVSI and it should be performed in T1a2 LVSI positive cases [III, B]. Management of Patients with T1b1 Disease, with Clear Margins and Without Residual Tumor Surgical LN staging is recommended in patients with T1b1 tumor with clear margins and absence of residual tumor on imaging (including non-suspicious LN). In case of histological evidence of PLN involvement, definitive CTRT is recommended and PALND, at least up to inferior mesenteric artery, may be considered for staging purposes [III, B]. In pathologically node negative patients with T1b1 disease, potential disease in the parametria should be addressed. Parametrectomy and upper vaginectomy should be considered [III, B]. Radiotherapy can be considered as an alternative modality to surgical treatment, considering the risk-benefit of repeat surgery [IV, C]. Management of Patients with ≥ T1b2 Disease, Involved Surgical Margins and/or Residual Tumor (Including LN) For patients with free surgical margins and in the absence of residual tumor on imaging (including non-suspicious LN), (chemo)radiotherapy is recommended as a treatment that avoids further surgical management [IV, B]. Radical surgery (pelvic lymphadenectomy, parametrectomy and resection of the upper vagina) is an option in selected patients without expected indication for adjuvant (chemo)radiotherapy. If surgery has been performed, indications for adjuvant (chemo)radiotherapy follow the general recommendations [IV, B]. If there is residual tumor on imaging (including suspicious LN), or involved surgical margins, CTRT with or without BT is the treatment of choice (see principles of radiotherapy) [III, B]. Para-aortic LN dissection, at least up to inferior mesenteric artery, may be considered for staging purposes in patients with positive pelvic nodes and negative paraaortic LN on imaging [IV, C]. Management of Locally Advanced Cervical Cancer (T1b3-T4a) Definitive radiotherapy should include concomitant chemotherapy whenever possible [I, A]. IGBT is an essential component of definitive radiotherapy and should not be replaced with an external boost (photon or proton). If BT is not available, patients should be referred to a center where this can be done [III, B]. General recommendations for prescription of CTRT and IGBT are as follows (details given in the section on principles of radiotherapy) [III, B]: 3D imaging (preferentially both MRI and (PET-CT) with the patient in the treatment position should be used for target contouring. It is recommended to deliver external beam radiotherapy (EBRT) with a dose of 45 Gy/25 fractions or 46 Gy/23 fractions by use of intensity-modulated or volumetric arc technique. Additional dose of radiation should be applied to pathological LN on imaging, preferentially using a simultaneous integrated boost (60 Gy EQD2, combined EBRT and estimated dose from IGBT). Concomitant weekly cisplatin is standard. However, weekly carboplatin or hyperthermia can be considered as an alternative option for patients not suitable for cisplatin. Image-guided adaptive brachytherapy (IGABT) (preferentially MRI) including access to intracavitary/interstitial techniques are needed to obtain a sufficiently high dose to ensure a high rate of local control in advanced cases with poor response to initial CTRT. This is especially important for non-squamous histology. Boosting of the primary tumor and/or the parametria by use of EBRT should be avoided. The overall treatment time including both CTRT and IGBT should aim to not exceed 7 weeks. PALND (at least up to inferior mesenteric artery) may be used to assess the need for elective para-aortic EBRT in patients with negative para-aortic lymph nodes (PALN) and positive PLN on imaging [IV, C]. If PALND is not performed, risk assessment for microscopic para-aortic nodal involvement and the indication for elective para-aortic irradiation can be based on the number of level 1 positive nodes (external iliac, interiliac, internal iliac) on imaging (e.g. >2 positive nodes). However, elective para-aortic radiation should always be applied in patients who on imaging have even one positive node at level 2 (common iliac) and above. The groin should also be included in the elective target for patients with tumor involvement of the lower-third of the vagina [IV, B]. Surgical removal of large pathological pelvic and/or para-aortic nodes before definitive CTRT is not routinely recommended [IV, D]. NACT in patients who otherwise are candidates for upfront definitive CTRT and IGBT is not recommended outside of clinical trials [II, D]. Adjuvant chemotherapy following definitive CTRT and IGBT does not improve survival and enhances toxicity and should not be used outside clinical trials [IV, D]. Adjuvant/completion hysterectomy after definitive CTRT and IGBT should not be performed since it does not improve survival and is associated with both increased perioperative and late morbidities [II, E]. Patients with a persistent tumor 3–6 months after definitive CTRT and BT and without evidence of regional or metastatic disease should be referred to specialized centers for evaluating the necessity and the possibility of performing salvage surgery (see management of recurrent disease and follow-up sections) [IV, B]. Role of Surgery in T1B3 and T2a2 (LN Negative) Tumors There is limited evidence to guide the choice between surgical treatment vs CTRT with IGBT in LN negative patients with T1b3 and T2a2 tumors. Histology, tumor size, completeness of the cervical rim, uterine corpus invasion, magnitude of vaginal invasion, age, comorbidity, menopausal status, body mass index, hemoglobin and experience with type C radical hysterectomy are some of the factors to consider [IV, B]. For surgery, avoidance of the combination of radical surgery and post-operative external radiotherapy requires acceptance for modifications of the traditional selection criteria (tumor size, degree of invasion, LVSI) for adjuvant treatment [IV, B]. The patient should be discussed in a multidisciplinary team and should be counseled for the advantages and disadvantages of both treatment options (surgery vs radiotherapy) in relation to the individual presence of prognostic factors [IV, A]. Given the limited number of patients with T1b3 and T2a2 (<10%) tumors, referral to highly specialized centers for treatment is recommended [IV, A]. Type C radical hysterectomy is recommended. LN staging should follow the same principles as in T1b1-2 tumors [IV, A]. NACT followed by radical surgery should not be performed outside clinical trials [I, E]. Recurrent/Metastatic Disease General Recommendations Treatment of recurrent disease requires centralization and involvement of a broad multidisciplinary team including a gynecological oncologist, radiation oncologist, radiologist, pathologist, medical oncologist, urologist, and plastic surgeon. A structured program for multidisciplinary diagnostic work-up, treatment, and follow-up must be present in centers responsible for the treatment [IV, A]. Participation in clinical trials is encouraged [V, B]. Early involvement of a palliative care specialist is encouraged [V, B]. The patient should be carefully counseled regarding treatment options, risks and consequences [V, A]. Diagnostic Work-up The aim of the diagnostic work-up is to determine the extent of the locoregional and/or metastatic disease [V, B]. The recurrence should be confirmed by histological examination if feasible [IV, B]. Patients with multiple nodal/distant metastases (ie, not oligometastatic disease) or multifocal local disease with extensive pelvic wall involvement should not be considered as candidates for radical treatment [IV, D]. Patients with oligometastatic or oligorecurrent disease should be considered for radical and potentially curative treatment options [IV, B]. The prognostic factors should be evaluated carefully and balanced in relation to the major morbidity caused by the treatment [IV, A]. Locoregional Recurrent Disease - Central Pelvic Recurrence After Primary Surgery Definitive CTRT combined with IGABT is the treatment of choice in radiotherapy naïve patients [IV, A]. The use of boost by external beam techniques to replace IGBT is not recommended [IV, D]. Small superficial lesions (ie, <5 mm thickness) in the vagina may be treated by IGBT using a vaginal cylinder, ovoids, or mold, whereas other lesions usually require combined intracavitary-interstitial techniques [IV, C]. Locoregional Recurrent Disease - Pelvic Sidewall Recurrence After Primary Surgery Definitive CTRT is the preferred option in radiotherapy naïve patients [IV, A]. When radical radiotherapy is not feasible, extended pelvic surgery can be considered. Surgery must aim for a complete tumor resection (R=0) also with the help of special techniques (laterally extended endopelvic resection (LEER), out of box procedures), if required [IV, B]. Combined operative-radiotherapy procedures using intra-operative radiotherapy or IGBT are an option if free surgical margins are not achievable [IV, B]. Locoregional Recurrent Disease - Central Pelvic or Pelvic Sidewall Recurrence After Radiotherapy Pelvic exenteration is recommended for central pelvic recurrence where there is no involvement of the pelvic sidewall, extrapelvic nodes or peritoneal disease [IV, B]. Reirradiation with IGABT for central recurrences could be considered in selected patients taking into account volume of the disease, or time from the primary radiotherapy and total dose administered initially. This must be performed only in specialized centers [IV, C]. In patients with pelvic sidewall involvement, extended pelvic surgery can be considered in specialized centers. Surgery must aim for a complete tumor resection (R=0) also with the help of special techniqu e s (LEER, out of box procedures), if required [IV, B]. Patients who are not candidates for extensive surgery should be treated with systemic chemotherapy. Additional treatment can be considered depending of the response [IV, B]. Oligometastatic Recurrences Localized para-aortic, mediastinal, and/or peri-clavicular recurrences out of previously irradiated fields may be treated by radical EBRT with or without chemotherapy [IV, C]. The therapeutic effect of nodal resection/debulking is unclear and should, if possible, be followed by radiotherapy [IV, C]. The management of “oligo” organ metastases (lung, liver, etc.) should be discussed in a multidisciplinary setting including the team involved in the treatment of the organ-affected metastasis. Treatment options are represented by local resection, thermal ablation, interventional BT, or stereotactic ablative radiotherapy according to the size and localization [IV, B]. Distant Recurrent and Metastatic Disease Patients with recurrent/metastatic disease should have a full clinical-diagnostic evaluation to assess the extent of disease and the most appropriate treatment modality including best supportive care [V, A]. Platinum-based chemotherapy±bevacizumab is recommended for chemo-naïve, medically fit patients with recurrent/metastatic disease. Carboplatin/paclitaxel and cisplatin/paclitaxel are the preferred regimens [I, A]. The addition of bevacizumab to platinum-based chemotherapy is recommended when the risk of significant gastrointestinal/genitourinary toxicities has been carefully assessed and discussed with the patient [I, A]. The addition of pembrolizumab to platinum-based chemotherapy±bevacizumab is recommended in patients with PD-L1 positive tumors, assessed as combined positive score (CPS) of 1 or more [I, A]. Patients who progressed after first-line platinum-based chemotherapy should be offered treatment with the anti PD-1 agent, cemiplimab, regardless of PDL-1 tumor status as long as they had not previously received immunotherapy [I, A]. Patients with distant metastatic disease at diagnosis, who have responded to systemic chemotherapy, could be considered for additional radical pelvic radiotherapy (including IGBT in selected cases). Those with residual oligometastatic disease after systemic treatment could also be considered for additional regional treatment (surgery, thermal ablation, radiotherapy) to involved sites [IV, C]. Inclusion of patients with recurrent/metastatic disease in clinical trials is strongly recommended [V, A]. Follow-up During and After Treatment/Long-term Survivorship General Recommendations Patients should be informed and educated at the time of diagnosis and throughout follow-up about signs/symptoms of recurrence. They should be informed about possible side effects (by physicians, nurses, brochures, videos, etc.) [V, A]. A network of healthcare providers including all care providers should be involved in the care of survivors (eg, primary care physicians, gynecologists, psychologists, sexologists, physiotherapists, dieticians, social workers) for the follow-up [V, A]. Follow-up strategy should be individualized in terms of intensity, duration and procedures, taking into account individual risk assessment [V, A]. Available prognostic models, such as the Annual Risk Recurrence Calculator available on the ESGO website can be used to tailor surveillance strategy in an individual patient [IV, B]. Follow-up should be centralized/coordinated in a center specialized in the treatment and follow-up of gynecological cancer patients [IV, A]. Follow-up is designed to monitor disease response, to detect recurrence and to screen for subsequent primary tumors [V, B]. Regular and systematic monitoring of side effects and quality of life should be performed to improve the quality of care [V, A]. Prevention and early detection of immediate and persistent symptoms and side effects of the different cancer treatments and the individual patient supportive care needs should be identified and established at diagnosis and monitored throughout the follow-up [V, A]. All side effects should be identified and treated if possible, namely physical and psychosocial [V, A]. The development of an individual survivorship monitoring and care plan is recommended [V, B]. Recommendations for a healthy life style should include smoking cessation, regular exercise, healthy diet and weight management [V, B]. Clinical trials should address long-term cancer survivorship and should include patient related outcomes [V, B]. Quality control of care should be established [V, B]. Each visit should be composed of the following [V, A]: Patient history (including identification of relevant symptoms and side effects) Physical examination (including a speculum and bimanual pelvic examination) Imaging and laboratory tests should be performed only based on risk of recurrence, symptoms or findings suggestive of recurrence and/or side effects. Regular review of an ongoing survivorship plan that can be shared with other healthcare providers. Oncological follow-up Patients should be educated about symptoms and signs of potential recurrence [V, A]. Appropriate imaging test (MRI, ultrasound for pelvic assessment, CT scan or PET-CT for systemic assessment) should be used in symptomatic women [IV, A]. In case of suspected tumor persistence, recurrence or second primary cancer, histological verification is strongly recommended [V, A]. Vaginal vault cytology is not recommended [IV, D]. After fertility sparing treatment, follow-up should include HPV testing (at 6–12 and 24 months) [V, A]. Monitoring of quality of life and side effects Quality of life and side effects should be regularly assessed at least by the physicians/clinical care nurses and if possible by patients (using patient related outcomes). Patient self-reporting of side effects should be encouraged during and after treatment with the same frequency as medical visits [IV, B]. A checklist of potential main side effects should be included in the patient survivorship monitoring and care plan (eg, sexual dysfunction, lymphedema, menopausal symptoms and osteoporosis, genito-urinary and gastrointestinal disorders, chronic pain, fatigue) [IV, A]. After CTRT and BT, patients should be counseled about sexual rehabilitation measures including the use of vaginal dilators. Topical estrogens are indicated [IV, B]. Hormone replacement therapy is indicated to cervical cancer survivors with premature menopause and should be consistent with standard menopausal recommendation [IV, B]. Physical and lifestyle changes may also help [V, C]. Bone status should be assessed regularly in patients with early menopause [V, B]. Follow-up After Definitive CTRT and BT Follow-up should be performed/coordinated by a physician experienced with follow-up care after radiotherapy and BT including monitoring of early, and late treatment-related side effects [V, A]. The same imaging method used at the start of treatment should be used to assess tumor response [V, B]. Routine biopsy to assess complete remission should not be performed [IV, D]. Cytology is not recommended in detecting disease recurrence after radiotherapy [IV, D]. Imaging (pelvic MRI±CT scan or PET-CT) should be performed not earlier than 3 months after the end of treatment [IV, B]. In patients with uncertain complete remission at 3 months post-radiotherapy, the assessment should be repeated after an additional 2–3 months with biopsy if indicated [IV, B]. Quality of Life and Palliative Care General Recommendations Early palliative care, integrated with oncological treatments, should be offered by the clinical team to all the patients diagnosed with advanced cervical cancer for managing symptoms and improving quality of life. A multidisciplinary approach must be included in the care plan with discussion and planning for specific treatment of these symptoms [IV, A]. Pain Opioids are the main analgesics for the treatment of moderate to severe cancer-related pain; the first option is oral morphine [I, A]; but other opioids and alternative routes (transdermic, subcutaneous) can be required in specific situations (ie, intestinal obstruction, problems with swallowing, renal failure) [III, B]. If opioids alone do not provide sufficient pain relief cancer-related neuropathic pain should be treated with a combination of opioids and carefully dosed adjuvants (gabapentin, pregabalin, duloxetine, and tricyclic antidepressants) [III, B]. Severe pelvic cancer pain unresponsive to an opioid regimen can benefit from other procedures like plexus block or spinal analgesia techniques [III, B]. Palliative EBRT (if feasible) is effective for painful pelvic progression and bone metastasis [IV, B]. Renal Failure Urinary derivation by ureteral stent or percutaneous nephrostomy should be considered to treat renal failure caused by tumoral obstruction. There are no clear guidelines to predict which patients will benefit from these procedures in terms of survival and quality of life, and its indication should be discussed carefully [IV, C]. Malignant Intestinal Obstruction Medical management of malignant intestinal obstruction consists of antisecretory, corticosteroids, and antiemetic drugs. A nasogastric tube is recommended if vomiting and discomfort persist in spite of medical management. Surgical procedures can be considered in selected patients [IV, B]. Vaginal Bleeding and Discharges In the case of vaginal bleeding, vaginal packing, interventional radiology (selective embolization) or palliative radiotherapy (if feasible) are recommended. There is not enough evidence to prefer one over the other. In the case of massive refractory bleeding, palliative sedation can be considered. Malodorous vaginal discharge can be improved with vaginal washing and the use of a vaginal metronidazole tablet [IV, B]. Psychosocial Suffering In patients with cervical advanced cancer, a multidisciplinary approach of physicians, nurses, psychologists, social workers, and community health workers is needed to manage psychosocial and spiritual suffering associated with social stigma deriving from genital disease, malodorous vaginal discharge, etc [IV, A]. Cervical Cancer in Pregnancy General Recommendations Every patient diagnosed with cervical cancer in pregnancy must be counseled by a multidisciplinary team. This team should consist of experts in the fields of gynecological oncology, neonatology, obstetrics, pathology, anesthesiology, radiation oncology, medical oncology, psycho-oncology, and, spiritual and ethical counseling. National or international tumor board counseling may be considered [V, A]. Given the large spectrum of therapeutic options, the multidisciplinary team should recommend a treatment plan according to the patient’s intention, tumor stage, and gestational age of pregnancy at the time of cancer diagnosis. The primary aims of the recommended treatment plan are the oncological safety of the pregnant woman as well as the fetal survival without additional morbidity [V, A]. Treatment of patients with cervical cancer in pregnancy should be exclusively done in gynecological oncology centers associated with the highest level perinatal center with expertise in all aspects of oncologic therapy in pregnancy and intensive medical care of premature neonates [V, A]. Clinical and Imaging Diagnosis Clinical examination and histological verification of cervical cancer are mandatory [IV, A]. Pathological confirmation may be obtained by colposcopy oriented biopsy or small cone (appropriate only during the first trimester of pregnancy, endocervical curettage is contraindicated) [IV, C]. Preferred imaging modalities for clinical staging in patients with cervical cancer in pregnancy include pelvic MRI or expert ultrasound as part of the primary work-up. Gadolinium-based contrast agents should be avoided [III, A]. The use of whole-body diffusion-weighted imaging MRI (WB-DWI/MRI) can reliably obviate the need for gadolinium contrast and radiation for nodal and distant staging during pregnancy. If not available, chest CT scan with abdominal shielding is an alternative. PET-CT should be avoided during pregnancy [IV, B]. Oncological Management Tumor involvement of suspicious nodes should be histologically confirmed because of its prognostic significance and the impact on the management up to 24 weeks of gestation (fetal viability) [IV, A]. Minimally invasive approach could be considered before 14–16 weeks of gestation; however, the sentinel node biopsy concept using indocyanine green is still experimental [IV, C]. Several treatment modalities are available and should be discussed with the patient taking into account the tumor stage, gestational week of pregnancy and the patient’s preferences [IV, B]: Delay of oncological treatment until fetal maturity (if possible >34 weeks of gestation) and initiate cancer-specific treatment immediately after delivery by cesarean section. This option might be considered if the term or fetal maturity is approaching. Conization or simple trachelectomy in order to completely remove the tumor, obtain free margins and perform nodal staging if needed, with the intention to preserve the pregnancy. Radical surgery or definitive CTRT according to the disease stage as recommended outside pregnancy, if the woman decides not to preserve the pregnancy. Pregnancy termination is recommended before any treatment after the first trimester, and fetus evacuation before CTRT, if possible. Chemotherapy until term of pregnancy (37 weeks of gestation) and initiation of definitive cancer-specific treatment immediately after delivery by cesarean section. At least a 2 week interval between chemotherapy and surgery is recommended. In patients with locally advanced disease or residual tumor after surgical procedure that cannot be completely removed (risk of premature rupture of amniotic membranes and/or cervical insufficiency), chemotherapy based on cisplatin or carboplatin can be considered starting after 14 weeks of pregnancy. Combination with taxanes is an option. Bevacizumab and checkpoint inhibitors are contraindicated. Before starting each cycle of chemotherapy, an assessment of treatment response should be made by clinical examination and transvaginal or transrectal ultrasound. If no response is achieved after 2 cycles of chemotherapy during pregnancy, treatment strategy should be re-evaluated. Pregnancy Management Spontaneous delivery appears to have negative prognostic impact in patients with cervical cancer in pregnancy. Thus, cesarean section is the recommended mode of delivery [IV, B]. At the time of cesarean section, definitive cancer specific treatment should be performed corresponding to that of non-pregnant women, taking into account the treatment that has already been given during pregnancy [IV, A]. Rare Tumors Histopathological diagnosis of rare cervical tumors needs confirmation (second opinion) by an expert pathologist [IV, A]. Treatment and care of rare cervical tumors needs to be centralized at referral centers and discussed in a multidisciplinary tumor board [IV, A]. Centralization of care in specialized centers and referral network is encouraged [IV, B]. Treatment planning should be made on a multidisciplinary basis (generally at a tumor board meeting as defined in the ESGO quality indicators) and based on the comprehensive and precise knowledge of prognostic and predictive factors for oncological outcome, side effects, and quality of life [IV, A]. Patients should be carefully counseled on the suggested treatment plan and potential alternatives, including risks and benefits of all options [V, A]. Treatment should be undertaken by a dedicated team of specialists in the diagnosis and management of cervical cancers [IV, A]. Enrollment of patients with cervical cancer in clinical trials is encouraged [V, B]. TNM Classification and FIGO Staging Patients with cervical cancer should be staged according to the TNM classification and the International Federation of Gynaecology and Obstetrics (FIGO) staging should also be documented [IV, A]. Systematic documentation and integration of the results from clinical examination, pathology and imaging including multidisciplinary team discussions of disparate findings is recommended [IV, A]. The method used to determine tumor status (T), lymph node (LN) status (N), and systemic status (M) should be noted (clinical, imaging, pathological) [IV, A]. Lymph node (LN) metastases should be classified according to the TNM classification [IV, A]. Prognostic Factors Systematic documentation of the following major tumor-related prognostic factors is recommended [II, A]: TNM and FIGO stage, including a maximum tumor size, detailed description of extracervical tumor extension (including uterine corpus involvement) and nodal involvement (eg, total number, location, size, and metabolic activity). Pathological tumor type including HPV status (see principles of pathological evaluation). Depth of cervical stromal invasion and a minimum thickness of uninvolved cervical stroma Margin status (ectocervical, endocervical, radial/deep stromal and vaginal cuff) Presence or absence of lymphovascular space involvement (LVSI). Presence or absence of distant metastases. Local Clinical and Radiological Diagnostic Work-up Pelvic examination and biopsy±colposcopy are mandatory to diagnose cervical cancer [II, A]. Pelvic magnetic resonance imaging (MRI) is mandatory for initial assessment of pelvic tumor extent and to guide treatment options (optional for T1a tumor with free margins after conization). Endovaginal/transrectal ultrasonography is an option if performed by a properly trained sonographer [II, A]. Cystoscopy or proctoscopy are not routinely recommended [IV, D]. Nodal/Distant Diagnostic Work-up In early stages managed primarily by surgery, surgical/pathological staging of pelvic lymph node (PLN) is the standard criterion to assess the prognosis and to guide treatment (except for T1a1 and T1a2 without LVSI) [III, A]. In locally advanced cervical cancer (T1b3 and higher (except T2a1) or in early-stage disease with suspicious LN on imaging), positron emission tomography-computed tomography (PET-CT), or chest/abdomen computed tomography (CT scan) (if PET-CT is not available) is recommended for assessment of nodal and distant disease [III, B]. PET-CT is recommended before chemoradiotherapy (CTRT) with curative intent [III, B]. Para-aortic LN dissection (PALND), at least up to inferior mesenteric artery, may be considered in locally advanced cervical cancer with negative para-aortic LN on imaging for staging purposes [IV, C]. Equivocal extrauterine disease should be considered for biopsy to avoid inappropriate treatment [IV, B]. Patients with cervical cancer should be staged according to the TNM classification and the International Federation of Gynaecology and Obstetrics (FIGO) staging should also be documented [IV, A]. Systematic documentation and integration of the results from clinical examination, pathology and imaging including multidisciplinary team discussions of disparate findings is recommended [IV, A]. The method used to determine tumor status (T), lymph node (LN) status (N), and systemic status (M) should be noted (clinical, imaging, pathological) [IV, A]. Lymph node (LN) metastases should be classified according to the TNM classification [IV, A]. Systematic documentation of the following major tumor-related prognostic factors is recommended [II, A]: TNM and FIGO stage, including a maximum tumor size, detailed description of extracervical tumor extension (including uterine corpus involvement) and nodal involvement (eg, total number, location, size, and metabolic activity). Pathological tumor type including HPV status (see principles of pathological evaluation). Depth of cervical stromal invasion and a minimum thickness of uninvolved cervical stroma Margin status (ectocervical, endocervical, radial/deep stromal and vaginal cuff) Presence or absence of lymphovascular space involvement (LVSI). Presence or absence of distant metastases. Pelvic examination and biopsy±colposcopy are mandatory to diagnose cervical cancer [II, A]. Pelvic magnetic resonance imaging (MRI) is mandatory for initial assessment of pelvic tumor extent and to guide treatment options (optional for T1a tumor with free margins after conization). Endovaginal/transrectal ultrasonography is an option if performed by a properly trained sonographer [II, A]. Cystoscopy or proctoscopy are not routinely recommended [IV, D]. In early stages managed primarily by surgery, surgical/pathological staging of pelvic lymph node (PLN) is the standard criterion to assess the prognosis and to guide treatment (except for T1a1 and T1a2 without LVSI) [III, A]. In locally advanced cervical cancer (T1b3 and higher (except T2a1) or in early-stage disease with suspicious LN on imaging), positron emission tomography-computed tomography (PET-CT), or chest/abdomen computed tomography (CT scan) (if PET-CT is not available) is recommended for assessment of nodal and distant disease [III, B]. PET-CT is recommended before chemoradiotherapy (CTRT) with curative intent [III, B]. Para-aortic LN dissection (PALND), at least up to inferior mesenteric artery, may be considered in locally advanced cervical cancer with negative para-aortic LN on imaging for staging purposes [IV, C]. Equivocal extrauterine disease should be considered for biopsy to avoid inappropriate treatment [IV, B]. Diagnosis of T1a Disease Diagnosis of T1a cancer should be based on a conization (or excision) specimen examined by an expert pathologist with accurate measurement of depth of invasion, margin status, coexisting pathology, and reliable assessment of LVSI [IV, B]. Loop or laser conization is preferable to cold-knife conization in women wanting to preserve fertility. Care should be taken to provide an intact (unfragmented) specimen with minimal thermal artifact. The cone specimen should be oriented for the pathologist [IV, B]. Surgical margins of the cone specimen should be clear of both invasive and preinvasive disease (except for low-grade intraepithelial lesion) [IV, B]. Management of T1a1 Disease Management of patients with T1a1 disease should be tailored to the individual depending on age, desire for fertility preservation, histological type, and the presence or absence of LVSI [III, B]. In case of positive margins (except for low-grade intraepithelial lesion in ectocervix), a repeat conization should be performed to rule out more extensive invasive disease [IV, B]. LN staging is not indicated in T1a1 LVSI-negative patients but can be considered in T1a1 LVSI-positive patients. Sentinel lymph node (SLN) biopsy (without additional PLN dissection (PLND)) is recommended in this situation [IV, B]. Conization can be considered a definitive treatment as hysterectomy does not improve the outcome [IV, C]. Radical surgical approaches such as radical hysterectomy, trachelectomy or parametrectomy represent overtreatment and should not be performed for patients with T1a1 disease [IV, D]. Patients with T1a1 adenocarcinoma who have completed childbearing should be offered SH [IV, B]. Management of T1a2 Disease Conization (with clear margins) alone or SH is an adequate treatment for patients with T1a2 disease [IV, B]. Parametrial resection is not indicated [IV, D]. SLN biopsy (without additional PLND) can be considered in LVSI-negative patients but should be performed in LVSI-positive patients [IV, B]. Patients with T1a2 adenocarcinoma who have completed childbearing should be offered SH [IV, B]. Diagnosis of T1a cancer should be based on a conization (or excision) specimen examined by an expert pathologist with accurate measurement of depth of invasion, margin status, coexisting pathology, and reliable assessment of LVSI [IV, B]. Loop or laser conization is preferable to cold-knife conization in women wanting to preserve fertility. Care should be taken to provide an intact (unfragmented) specimen with minimal thermal artifact. The cone specimen should be oriented for the pathologist [IV, B]. Surgical margins of the cone specimen should be clear of both invasive and preinvasive disease (except for low-grade intraepithelial lesion) [IV, B]. Management of patients with T1a1 disease should be tailored to the individual depending on age, desire for fertility preservation, histological type, and the presence or absence of LVSI [III, B]. In case of positive margins (except for low-grade intraepithelial lesion in ectocervix), a repeat conization should be performed to rule out more extensive invasive disease [IV, B]. LN staging is not indicated in T1a1 LVSI-negative patients but can be considered in T1a1 LVSI-positive patients. Sentinel lymph node (SLN) biopsy (without additional PLN dissection (PLND)) is recommended in this situation [IV, B]. Conization can be considered a definitive treatment as hysterectomy does not improve the outcome [IV, C]. Radical surgical approaches such as radical hysterectomy, trachelectomy or parametrectomy represent overtreatment and should not be performed for patients with T1a1 disease [IV, D]. Patients with T1a1 adenocarcinoma who have completed childbearing should be offered SH [IV, B]. Conization (with clear margins) alone or SH is an adequate treatment for patients with T1a2 disease [IV, B]. Parametrial resection is not indicated [IV, D]. SLN biopsy (without additional PLND) can be considered in LVSI-negative patients but should be performed in LVSI-positive patients [IV, B]. Patients with T1a2 adenocarcinoma who have completed childbearing should be offered SH [IV, B]. General Recommendations Treatment strategy should aim to avoid combining radical surgery and radiotherapy because of the high morbidity induced by the combined treatment [IV, A]. Treatment strategy should aim to avoid combining radical surgery and radiotherapy because of the high morbidity induced by the combined treatment [IV, A]. Radical surgery by a gynecological oncologist is the preferred treatment modality. Laparotomy is the standard approach for all procedures which include radical parametrectomy [I, A]. Minimally invasive approach may be considered only in low risk tumors (<2 cm and free margins after conization), in high-volume centers experienced in performing radical hysterectomy with minimally invasive surgery, which meet the ESGO quality criteria for surgery, if the patient agrees after comprehensive discussion about current evidence [IV, C]. LN assessment should be performed as the first step of surgical management [IV, A]. Minimally invasive surgery is an acceptable approach for LN staging [IV, B]. SLN biopsy before pelvic lymphadenectomy should be performed. Indocyanine green is the preferred technique [III, A]. A combination of blue dye with radiocolloid is an alternative technique [IV, B]. Intra-operative assessment of LN status (evaluated by frozen section) is recommended. Sentinel nodes from both sides of the pelvis and/or any suspicious LN should be sent for intra-operative assessment [III, A]. If any LN involvement is detected intraoperatively, further PLND and radical hysterectomy should be avoided. Patients should be referred for definitive CTRT [III, A]. PALND at least up to inferior mesenteric artery may be considered for staging purposes [IV, C]. After SLN biopsy, if SLN are negative on frozen section, a systematic pelvic lymphadenectomy should be performed as the standard LN staging [III, A]. If SLN is negative bilaterally in the pelvic level I area (below iliac bifurcation) LN dissection can be limited to level I [IV, B]. If SLN is not detected on either side, LN dissection should include on that particular pelvic side the removal of lymphatic tissue from all traditional regions including obturator fossa, external iliac regions, common iliac regions, and presacral region [III, A]. After frozen section, all SLN should be processed according to pathological protocol for ultrastaging (see the principles of pathological evaluation) [III, A]. The type of radical hysterectomy (extent of parametrial resection, type A-C2) should be based on the presence of prognostic risk factors identified preoperatively such as tumor size, maximum stromal invasion, and LVSI, which are used to categorize patients at high, intermediate, and low risk of treatment failure. A complete description of the template used for radical hysterectomy should be present in the surgical report. The 2017 modification of the Querleu-Morrow classification is recommended as a tool [IV, A]. Ovarian preservation should be discussed with women in reproductive age with squamous cell carcinoma, can be considered in HPV-associated adenocarcinoma and is not recommended for HPV-independent adenocarcinomas. Opportunistic bilateral salpingectomy should be performed if ovaries are preserved. Ovarian transposition should be discussed upfront with the patient and individualized according to risk balance [IV, A]. If a combination of risk factors is known at diagnosis, which would require an adjuvant treatment, definitive CTRT and brachytherapy (BT) should be considered without previous radical pelvic surgery [IV, A]. Negative LN on Radiological Staging – Alternative Treatment Options Definitive CTRT and image-guided brachytherapy (IGBT) represent an alternative treatment option [IV, B]. Neoadjuvant chemotherapy (NACT) or CTRT followed by surgery are not recommended [IV, D]. Adjuvant Treatment After Radical Surgery Adjuvant radiotherapy should be considered in the intermediate risk group (combination of risk factors at final pathology such as tumor size, LVSI, and depth of stromal invasion) [IV, A]. When an adequate type of radical hysterectomy has been performed in intermediate risk group patients, observation is an alternative option, especially in teams experienced in this approach [IV, B]. Adjuvant CTRT is indicated in the high-risk group (see principles of radiotherapy) [IV, A]: metastatic involvement of PLN (macrometastases pN1 or micrometastases pN1(mi)) on final pathologic assessment. positive surgical margins (vagina/parametria/paracervix). parametrial involvement. Additional BT boost as part of adjuvant CTRT can be considered in cases with vaginal and/or parametrial positive disease (see principles of radiotherapy) [IV, B]. Adjuvant treatment may be considered also if only isolated tumor cells are detected in SLN, although its prognostic impact remains uncertain [IV, C]. Definitive CTRT and image-guided brachytherapy (IGBT) represent an alternative treatment option [IV, B]. Neoadjuvant chemotherapy (NACT) or CTRT followed by surgery are not recommended [IV, D]. Adjuvant radiotherapy should be considered in the intermediate risk group (combination of risk factors at final pathology such as tumor size, LVSI, and depth of stromal invasion) [IV, A]. When an adequate type of radical hysterectomy has been performed in intermediate risk group patients, observation is an alternative option, especially in teams experienced in this approach [IV, B]. Adjuvant CTRT is indicated in the high-risk group (see principles of radiotherapy) [IV, A]: metastatic involvement of PLN (macrometastases pN1 or micrometastases pN1(mi)) on final pathologic assessment. positive surgical margins (vagina/parametria/paracervix). parametrial involvement. Additional BT boost as part of adjuvant CTRT can be considered in cases with vaginal and/or parametrial positive disease (see principles of radiotherapy) [IV, B]. Adjuvant treatment may be considered also if only isolated tumor cells are detected in SLN, although its prognostic impact remains uncertain [IV, C]. Fertility sparing therapy is an oncologically valid alternative to radical hysterectomy for young patients with cervical cancer <2 cm (squamous cell carcinoma and HPV-related adenocarcinoma) who want to preserve the option to have children. Before initiating fertility sparing therapy, consultation at an onco-fertility center and discussion in a multidisciplinary tumor board is recommended [III, B]. Counseling of eligible patients should encompass the oncologic and obstetric risks related to this type of management as well as the risk of fertility sparing therapy abandonment if there are positive resection margins or LN involvement [III, A]. Fertility-sparing treatment should be performed exclusively in gynaecological-oncological centers with comprehensive expertise in all types of these surgical procedures [IV, A]. Fertility-sparing treatment should not be recommended for uncommon and rare histological types/subtypes of cervical cancer with aggressive behavior including neuroendocrine carcinomas, HPV-independent adenocarcinomas and carcinosarcomas [V, D]. For patients who consider fertility sparing therapy, prognostic factors, clinical staging, and preoperative work-up do not differ from those not considering fertility sparing therapy (see above). Pelvic MRI and/or expert sonography are mandatory imaging tests to measure the non-involved cervical length (upper tumor free margin) and the remaining (after cone biopsy) cervical length [III, A]. Negative PLN status is the precondition for any fertility sparing therapy. Therefore, PLN staging (SLN) should always be the first step in each fertility-sparing therapy procedure. Identification of SLN and its ultrastaging is highly recommended. Any intraoperative suspicious LN (apart from SLN) should also be removed. If SLN cannot be detected on either pelvic side, a systematic pelvic lymphadenectomy should be performed on that side. Intraoperative assessment of LN status is highly recommended. All SLN from both sides of the pelvis and any suspicious LN should be sent for frozen section. LN staging is not indicated in T1a1 LVSI negative [III, A]. In case of intraoperatively proven PLN involvement, fertility-sparing surgery should be abandoned and patients should be referred for CTRT and BT [IV, B]. PALND, at least up to inferior mesenteric artery, may be considered for staging purposes [IV, C]. Ovarian transposition cannot be recommended in N1 status [IV, D]. The specific goal of fertility-sparing surgery must be resection of invasive tumor with adequate free margins and preservation of the upper part of the cervix [IV, A]. Intraoperative frozen section is a feasible way of assessing the upper resection margin [IV, C]. LN staging follows the principles of management of early stages [III, B]. Fertility sparing procedures comprise of conization (see ), simple trachelectomy (see ), radical (vaginal) trachelectomy (see ), abdominal radical trachelectomy (see ) [III, B]. Conization and simple trachelectomy are adequate fertility sparing procedures in patients with T1a1 and T1a2 tumors, regardless of LVSI status [IV, B]. Conization or simple trachelectomy are adequate fertility sparing procedures for T1b1, LVSI negative tumors. Radical trachelectomy is still an option [IV, B]. Radical trachelectomy (type B) should be performed in patients with cervical cancer T1b1, LVSI-positive. In patients without deep stromal involvement and with a high probability of adequate endocervical tumor free margins, simple trachelectomy can be considered [III, B]. Intraoperative placement of permanent cerclage should be performed during simple or radical trachelectomy [IV, B]. Fertility sparing therapy for patients with tumors greater than 2 cm is significantly associated with a higher risk of recurrence and should not be considered as a standard treatment. The risk of recurrence must be comprehensively discussed with the patient. NACT followed by radical vaginal trachelectomy and abdominal radical trachelectomy or cone has been described for fertility sparing treatment in patients with tumors >2 cm. PLN staging should be performed before starting NACT to confirm tumor-free LN. The optimal number of chemotherapy cycles, chemotherapy regimen as well as extent of cervical resection following NACT, are still a matter of debate [IV, B]. In more advanced cases, various fertility preservation proposals such as ovarian transposition (see ), oocyte-, embryo- or ovarian tissue preservation and egg donation should be discussed with the patient. The aim of the fertility preservation should be to offer the most efficient approach in accordance with the legal country-specific regulations, while not increasing the oncological risk [IV, B]. Any pregnancy following fertility sparing therapy should be considered as a high-risk pregnancy. Following simple or radical trachelectomy with placement of a permanent cerclage, delivery can only be performed by cesarean section [IV, B]. Although evidence is limited, several antenatal management tools can be considered following fertility sparing therapy including screening and treatment of asymptomatic bacteriuria, screening for cervical incompetence and progressive cervical shortening by transvaginal ultrasonography, fetal fibronectin testing, screening (and treatment) for asymptomatic vaginal infection, vaginal progesterone application, total cervical closure according to Saling and cervical cerclage, if not placed during trachelectomy [IV, C]. Routine hysterectomy after completion of childbearing is not mandatory [V, D]. General Recommendations Management of disease found after SH should be based on expert pathology review and discussed in a multidisciplinary tumor board. In general, management of occult disease follows the principles of the standard management, and is based on pathologic findings, and clinical staging. Treatment strategy should aim to avoid combining further surgery and radiotherapy because of the high morbidity after combined treatment [III, B]. Before making further management decisions, optimal imaging is necessary to evaluate the local and regional (nodal) disease status. Optimal imaging follows the same recommendations as that for the standard management [III, B]. When surgical staging of nodal disease is indicated (see below for details), it can be considered either as an isolated (preferentially laparoscopic) procedure or as the first step of surgical management in radiologic node negative patients. Surgical staging of nodal disease can also be considered to assess inconclusive nodes at imaging. SLN biopsy cannot be performed in the absence of the uterus. Any suspicious LN should be sent for intraoperative assessment (frozen section) [III, B]. Para-aortic LN dissection, at least up to inferior mesenteric artery, may be considered for staging purposes in patients with positive pelvic nodes at imaging, or at frozen section [IV, C]. Management of Patients with T1a1 and T1a2 Disease In patients with T1a1 tumor regardless of LVSI status and T1a2 tumor LVSI negative with clear margins in the hysterectomy specimen, no additional treatment is recommended [III, B]. Surgical LN assessment can be considered in T1a1 tumors with LVSI and it should be performed in T1a2 LVSI positive cases [III, B]. Management of Patients with T1b1 Disease, with Clear Margins and Without Residual Tumor Surgical LN staging is recommended in patients with T1b1 tumor with clear margins and absence of residual tumor on imaging (including non-suspicious LN). In case of histological evidence of PLN involvement, definitive CTRT is recommended and PALND, at least up to inferior mesenteric artery, may be considered for staging purposes [III, B]. In pathologically node negative patients with T1b1 disease, potential disease in the parametria should be addressed. Parametrectomy and upper vaginectomy should be considered [III, B]. Radiotherapy can be considered as an alternative modality to surgical treatment, considering the risk-benefit of repeat surgery [IV, C]. Management of Patients with ≥ T1b2 Disease, Involved Surgical Margins and/or Residual Tumor (Including LN) For patients with free surgical margins and in the absence of residual tumor on imaging (including non-suspicious LN), (chemo)radiotherapy is recommended as a treatment that avoids further surgical management [IV, B]. Radical surgery (pelvic lymphadenectomy, parametrectomy and resection of the upper vagina) is an option in selected patients without expected indication for adjuvant (chemo)radiotherapy. If surgery has been performed, indications for adjuvant (chemo)radiotherapy follow the general recommendations [IV, B]. If there is residual tumor on imaging (including suspicious LN), or involved surgical margins, CTRT with or without BT is the treatment of choice (see principles of radiotherapy) [III, B]. Para-aortic LN dissection, at least up to inferior mesenteric artery, may be considered for staging purposes in patients with positive pelvic nodes and negative paraaortic LN on imaging [IV, C]. Management of Locally Advanced Cervical Cancer (T1b3-T4a) Definitive radiotherapy should include concomitant chemotherapy whenever possible [I, A]. IGBT is an essential component of definitive radiotherapy and should not be replaced with an external boost (photon or proton). If BT is not available, patients should be referred to a center where this can be done [III, B]. General recommendations for prescription of CTRT and IGBT are as follows (details given in the section on principles of radiotherapy) [III, B]: 3D imaging (preferentially both MRI and (PET-CT) with the patient in the treatment position should be used for target contouring. It is recommended to deliver external beam radiotherapy (EBRT) with a dose of 45 Gy/25 fractions or 46 Gy/23 fractions by use of intensity-modulated or volumetric arc technique. Additional dose of radiation should be applied to pathological LN on imaging, preferentially using a simultaneous integrated boost (60 Gy EQD2, combined EBRT and estimated dose from IGBT). Concomitant weekly cisplatin is standard. However, weekly carboplatin or hyperthermia can be considered as an alternative option for patients not suitable for cisplatin. Image-guided adaptive brachytherapy (IGABT) (preferentially MRI) including access to intracavitary/interstitial techniques are needed to obtain a sufficiently high dose to ensure a high rate of local control in advanced cases with poor response to initial CTRT. This is especially important for non-squamous histology. Boosting of the primary tumor and/or the parametria by use of EBRT should be avoided. The overall treatment time including both CTRT and IGBT should aim to not exceed 7 weeks. PALND (at least up to inferior mesenteric artery) may be used to assess the need for elective para-aortic EBRT in patients with negative para-aortic lymph nodes (PALN) and positive PLN on imaging [IV, C]. If PALND is not performed, risk assessment for microscopic para-aortic nodal involvement and the indication for elective para-aortic irradiation can be based on the number of level 1 positive nodes (external iliac, interiliac, internal iliac) on imaging (e.g. >2 positive nodes). However, elective para-aortic radiation should always be applied in patients who on imaging have even one positive node at level 2 (common iliac) and above. The groin should also be included in the elective target for patients with tumor involvement of the lower-third of the vagina [IV, B]. Surgical removal of large pathological pelvic and/or para-aortic nodes before definitive CTRT is not routinely recommended [IV, D]. NACT in patients who otherwise are candidates for upfront definitive CTRT and IGBT is not recommended outside of clinical trials [II, D]. Adjuvant chemotherapy following definitive CTRT and IGBT does not improve survival and enhances toxicity and should not be used outside clinical trials [IV, D]. Adjuvant/completion hysterectomy after definitive CTRT and IGBT should not be performed since it does not improve survival and is associated with both increased perioperative and late morbidities [II, E]. Patients with a persistent tumor 3–6 months after definitive CTRT and BT and without evidence of regional or metastatic disease should be referred to specialized centers for evaluating the necessity and the possibility of performing salvage surgery (see management of recurrent disease and follow-up sections) [IV, B]. Role of Surgery in T1B3 and T2a2 (LN Negative) Tumors There is limited evidence to guide the choice between surgical treatment vs CTRT with IGBT in LN negative patients with T1b3 and T2a2 tumors. Histology, tumor size, completeness of the cervical rim, uterine corpus invasion, magnitude of vaginal invasion, age, comorbidity, menopausal status, body mass index, hemoglobin and experience with type C radical hysterectomy are some of the factors to consider [IV, B]. For surgery, avoidance of the combination of radical surgery and post-operative external radiotherapy requires acceptance for modifications of the traditional selection criteria (tumor size, degree of invasion, LVSI) for adjuvant treatment [IV, B]. The patient should be discussed in a multidisciplinary team and should be counseled for the advantages and disadvantages of both treatment options (surgery vs radiotherapy) in relation to the individual presence of prognostic factors [IV, A]. Given the limited number of patients with T1b3 and T2a2 (<10%) tumors, referral to highly specialized centers for treatment is recommended [IV, A]. Type C radical hysterectomy is recommended. LN staging should follow the same principles as in T1b1-2 tumors [IV, A]. NACT followed by radical surgery should not be performed outside clinical trials [I, E]. Management of disease found after SH should be based on expert pathology review and discussed in a multidisciplinary tumor board. In general, management of occult disease follows the principles of the standard management, and is based on pathologic findings, and clinical staging. Treatment strategy should aim to avoid combining further surgery and radiotherapy because of the high morbidity after combined treatment [III, B]. Before making further management decisions, optimal imaging is necessary to evaluate the local and regional (nodal) disease status. Optimal imaging follows the same recommendations as that for the standard management [III, B]. When surgical staging of nodal disease is indicated (see below for details), it can be considered either as an isolated (preferentially laparoscopic) procedure or as the first step of surgical management in radiologic node negative patients. Surgical staging of nodal disease can also be considered to assess inconclusive nodes at imaging. SLN biopsy cannot be performed in the absence of the uterus. Any suspicious LN should be sent for intraoperative assessment (frozen section) [III, B]. Para-aortic LN dissection, at least up to inferior mesenteric artery, may be considered for staging purposes in patients with positive pelvic nodes at imaging, or at frozen section [IV, C]. In patients with T1a1 tumor regardless of LVSI status and T1a2 tumor LVSI negative with clear margins in the hysterectomy specimen, no additional treatment is recommended [III, B]. Surgical LN assessment can be considered in T1a1 tumors with LVSI and it should be performed in T1a2 LVSI positive cases [III, B]. Surgical LN staging is recommended in patients with T1b1 tumor with clear margins and absence of residual tumor on imaging (including non-suspicious LN). In case of histological evidence of PLN involvement, definitive CTRT is recommended and PALND, at least up to inferior mesenteric artery, may be considered for staging purposes [III, B]. In pathologically node negative patients with T1b1 disease, potential disease in the parametria should be addressed. Parametrectomy and upper vaginectomy should be considered [III, B]. Radiotherapy can be considered as an alternative modality to surgical treatment, considering the risk-benefit of repeat surgery [IV, C]. For patients with free surgical margins and in the absence of residual tumor on imaging (including non-suspicious LN), (chemo)radiotherapy is recommended as a treatment that avoids further surgical management [IV, B]. Radical surgery (pelvic lymphadenectomy, parametrectomy and resection of the upper vagina) is an option in selected patients without expected indication for adjuvant (chemo)radiotherapy. If surgery has been performed, indications for adjuvant (chemo)radiotherapy follow the general recommendations [IV, B]. If there is residual tumor on imaging (including suspicious LN), or involved surgical margins, CTRT with or without BT is the treatment of choice (see principles of radiotherapy) [III, B]. Para-aortic LN dissection, at least up to inferior mesenteric artery, may be considered for staging purposes in patients with positive pelvic nodes and negative paraaortic LN on imaging [IV, C]. Definitive radiotherapy should include concomitant chemotherapy whenever possible [I, A]. IGBT is an essential component of definitive radiotherapy and should not be replaced with an external boost (photon or proton). If BT is not available, patients should be referred to a center where this can be done [III, B]. General recommendations for prescription of CTRT and IGBT are as follows (details given in the section on principles of radiotherapy) [III, B]: 3D imaging (preferentially both MRI and (PET-CT) with the patient in the treatment position should be used for target contouring. It is recommended to deliver external beam radiotherapy (EBRT) with a dose of 45 Gy/25 fractions or 46 Gy/23 fractions by use of intensity-modulated or volumetric arc technique. Additional dose of radiation should be applied to pathological LN on imaging, preferentially using a simultaneous integrated boost (60 Gy EQD2, combined EBRT and estimated dose from IGBT). Concomitant weekly cisplatin is standard. However, weekly carboplatin or hyperthermia can be considered as an alternative option for patients not suitable for cisplatin. Image-guided adaptive brachytherapy (IGABT) (preferentially MRI) including access to intracavitary/interstitial techniques are needed to obtain a sufficiently high dose to ensure a high rate of local control in advanced cases with poor response to initial CTRT. This is especially important for non-squamous histology. Boosting of the primary tumor and/or the parametria by use of EBRT should be avoided. The overall treatment time including both CTRT and IGBT should aim to not exceed 7 weeks. PALND (at least up to inferior mesenteric artery) may be used to assess the need for elective para-aortic EBRT in patients with negative para-aortic lymph nodes (PALN) and positive PLN on imaging [IV, C]. If PALND is not performed, risk assessment for microscopic para-aortic nodal involvement and the indication for elective para-aortic irradiation can be based on the number of level 1 positive nodes (external iliac, interiliac, internal iliac) on imaging (e.g. >2 positive nodes). However, elective para-aortic radiation should always be applied in patients who on imaging have even one positive node at level 2 (common iliac) and above. The groin should also be included in the elective target for patients with tumor involvement of the lower-third of the vagina [IV, B]. Surgical removal of large pathological pelvic and/or para-aortic nodes before definitive CTRT is not routinely recommended [IV, D]. NACT in patients who otherwise are candidates for upfront definitive CTRT and IGBT is not recommended outside of clinical trials [II, D]. Adjuvant chemotherapy following definitive CTRT and IGBT does not improve survival and enhances toxicity and should not be used outside clinical trials [IV, D]. Adjuvant/completion hysterectomy after definitive CTRT and IGBT should not be performed since it does not improve survival and is associated with both increased perioperative and late morbidities [II, E]. Patients with a persistent tumor 3–6 months after definitive CTRT and BT and without evidence of regional or metastatic disease should be referred to specialized centers for evaluating the necessity and the possibility of performing salvage surgery (see management of recurrent disease and follow-up sections) [IV, B]. There is limited evidence to guide the choice between surgical treatment vs CTRT with IGBT in LN negative patients with T1b3 and T2a2 tumors. Histology, tumor size, completeness of the cervical rim, uterine corpus invasion, magnitude of vaginal invasion, age, comorbidity, menopausal status, body mass index, hemoglobin and experience with type C radical hysterectomy are some of the factors to consider [IV, B]. For surgery, avoidance of the combination of radical surgery and post-operative external radiotherapy requires acceptance for modifications of the traditional selection criteria (tumor size, degree of invasion, LVSI) for adjuvant treatment [IV, B]. The patient should be discussed in a multidisciplinary team and should be counseled for the advantages and disadvantages of both treatment options (surgery vs radiotherapy) in relation to the individual presence of prognostic factors [IV, A]. Given the limited number of patients with T1b3 and T2a2 (<10%) tumors, referral to highly specialized centers for treatment is recommended [IV, A]. Type C radical hysterectomy is recommended. LN staging should follow the same principles as in T1b1-2 tumors [IV, A]. NACT followed by radical surgery should not be performed outside clinical trials [I, E]. General Recommendations Treatment of recurrent disease requires centralization and involvement of a broad multidisciplinary team including a gynecological oncologist, radiation oncologist, radiologist, pathologist, medical oncologist, urologist, and plastic surgeon. A structured program for multidisciplinary diagnostic work-up, treatment, and follow-up must be present in centers responsible for the treatment [IV, A]. Participation in clinical trials is encouraged [V, B]. Early involvement of a palliative care specialist is encouraged [V, B]. The patient should be carefully counseled regarding treatment options, risks and consequences [V, A]. Diagnostic Work-up The aim of the diagnostic work-up is to determine the extent of the locoregional and/or metastatic disease [V, B]. The recurrence should be confirmed by histological examination if feasible [IV, B]. Patients with multiple nodal/distant metastases (ie, not oligometastatic disease) or multifocal local disease with extensive pelvic wall involvement should not be considered as candidates for radical treatment [IV, D]. Patients with oligometastatic or oligorecurrent disease should be considered for radical and potentially curative treatment options [IV, B]. The prognostic factors should be evaluated carefully and balanced in relation to the major morbidity caused by the treatment [IV, A]. Locoregional Recurrent Disease - Central Pelvic Recurrence After Primary Surgery Definitive CTRT combined with IGABT is the treatment of choice in radiotherapy naïve patients [IV, A]. The use of boost by external beam techniques to replace IGBT is not recommended [IV, D]. Small superficial lesions (ie, <5 mm thickness) in the vagina may be treated by IGBT using a vaginal cylinder, ovoids, or mold, whereas other lesions usually require combined intracavitary-interstitial techniques [IV, C]. Locoregional Recurrent Disease - Pelvic Sidewall Recurrence After Primary Surgery Definitive CTRT is the preferred option in radiotherapy naïve patients [IV, A]. When radical radiotherapy is not feasible, extended pelvic surgery can be considered. Surgery must aim for a complete tumor resection (R=0) also with the help of special techniques (laterally extended endopelvic resection (LEER), out of box procedures), if required [IV, B]. Combined operative-radiotherapy procedures using intra-operative radiotherapy or IGBT are an option if free surgical margins are not achievable [IV, B]. Locoregional Recurrent Disease - Central Pelvic or Pelvic Sidewall Recurrence After Radiotherapy Pelvic exenteration is recommended for central pelvic recurrence where there is no involvement of the pelvic sidewall, extrapelvic nodes or peritoneal disease [IV, B]. Reirradiation with IGABT for central recurrences could be considered in selected patients taking into account volume of the disease, or time from the primary radiotherapy and total dose administered initially. This must be performed only in specialized centers [IV, C]. In patients with pelvic sidewall involvement, extended pelvic surgery can be considered in specialized centers. Surgery must aim for a complete tumor resection (R=0) also with the help of special techniqu e s (LEER, out of box procedures), if required [IV, B]. Patients who are not candidates for extensive surgery should be treated with systemic chemotherapy. Additional treatment can be considered depending of the response [IV, B]. Oligometastatic Recurrences Localized para-aortic, mediastinal, and/or peri-clavicular recurrences out of previously irradiated fields may be treated by radical EBRT with or without chemotherapy [IV, C]. The therapeutic effect of nodal resection/debulking is unclear and should, if possible, be followed by radiotherapy [IV, C]. The management of “oligo” organ metastases (lung, liver, etc.) should be discussed in a multidisciplinary setting including the team involved in the treatment of the organ-affected metastasis. Treatment options are represented by local resection, thermal ablation, interventional BT, or stereotactic ablative radiotherapy according to the size and localization [IV, B]. Distant Recurrent and Metastatic Disease Patients with recurrent/metastatic disease should have a full clinical-diagnostic evaluation to assess the extent of disease and the most appropriate treatment modality including best supportive care [V, A]. Platinum-based chemotherapy±bevacizumab is recommended for chemo-naïve, medically fit patients with recurrent/metastatic disease. Carboplatin/paclitaxel and cisplatin/paclitaxel are the preferred regimens [I, A]. The addition of bevacizumab to platinum-based chemotherapy is recommended when the risk of significant gastrointestinal/genitourinary toxicities has been carefully assessed and discussed with the patient [I, A]. The addition of pembrolizumab to platinum-based chemotherapy±bevacizumab is recommended in patients with PD-L1 positive tumors, assessed as combined positive score (CPS) of 1 or more [I, A]. Patients who progressed after first-line platinum-based chemotherapy should be offered treatment with the anti PD-1 agent, cemiplimab, regardless of PDL-1 tumor status as long as they had not previously received immunotherapy [I, A]. Patients with distant metastatic disease at diagnosis, who have responded to systemic chemotherapy, could be considered for additional radical pelvic radiotherapy (including IGBT in selected cases). Those with residual oligometastatic disease after systemic treatment could also be considered for additional regional treatment (surgery, thermal ablation, radiotherapy) to involved sites [IV, C]. Inclusion of patients with recurrent/metastatic disease in clinical trials is strongly recommended [V, A]. Treatment of recurrent disease requires centralization and involvement of a broad multidisciplinary team including a gynecological oncologist, radiation oncologist, radiologist, pathologist, medical oncologist, urologist, and plastic surgeon. A structured program for multidisciplinary diagnostic work-up, treatment, and follow-up must be present in centers responsible for the treatment [IV, A]. Participation in clinical trials is encouraged [V, B]. Early involvement of a palliative care specialist is encouraged [V, B]. The patient should be carefully counseled regarding treatment options, risks and consequences [V, A]. The aim of the diagnostic work-up is to determine the extent of the locoregional and/or metastatic disease [V, B]. The recurrence should be confirmed by histological examination if feasible [IV, B]. Patients with multiple nodal/distant metastases (ie, not oligometastatic disease) or multifocal local disease with extensive pelvic wall involvement should not be considered as candidates for radical treatment [IV, D]. Patients with oligometastatic or oligorecurrent disease should be considered for radical and potentially curative treatment options [IV, B]. The prognostic factors should be evaluated carefully and balanced in relation to the major morbidity caused by the treatment [IV, A]. Definitive CTRT combined with IGABT is the treatment of choice in radiotherapy naïve patients [IV, A]. The use of boost by external beam techniques to replace IGBT is not recommended [IV, D]. Small superficial lesions (ie, <5 mm thickness) in the vagina may be treated by IGBT using a vaginal cylinder, ovoids, or mold, whereas other lesions usually require combined intracavitary-interstitial techniques [IV, C]. Definitive CTRT is the preferred option in radiotherapy naïve patients [IV, A]. When radical radiotherapy is not feasible, extended pelvic surgery can be considered. Surgery must aim for a complete tumor resection (R=0) also with the help of special techniques (laterally extended endopelvic resection (LEER), out of box procedures), if required [IV, B]. Combined operative-radiotherapy procedures using intra-operative radiotherapy or IGBT are an option if free surgical margins are not achievable [IV, B]. Pelvic exenteration is recommended for central pelvic recurrence where there is no involvement of the pelvic sidewall, extrapelvic nodes or peritoneal disease [IV, B]. Reirradiation with IGABT for central recurrences could be considered in selected patients taking into account volume of the disease, or time from the primary radiotherapy and total dose administered initially. This must be performed only in specialized centers [IV, C]. In patients with pelvic sidewall involvement, extended pelvic surgery can be considered in specialized centers. Surgery must aim for a complete tumor resection (R=0) also with the help of special techniqu e s (LEER, out of box procedures), if required [IV, B]. Patients who are not candidates for extensive surgery should be treated with systemic chemotherapy. Additional treatment can be considered depending of the response [IV, B]. Localized para-aortic, mediastinal, and/or peri-clavicular recurrences out of previously irradiated fields may be treated by radical EBRT with or without chemotherapy [IV, C]. The therapeutic effect of nodal resection/debulking is unclear and should, if possible, be followed by radiotherapy [IV, C]. The management of “oligo” organ metastases (lung, liver, etc.) should be discussed in a multidisciplinary setting including the team involved in the treatment of the organ-affected metastasis. Treatment options are represented by local resection, thermal ablation, interventional BT, or stereotactic ablative radiotherapy according to the size and localization [IV, B]. Patients with recurrent/metastatic disease should have a full clinical-diagnostic evaluation to assess the extent of disease and the most appropriate treatment modality including best supportive care [V, A]. Platinum-based chemotherapy±bevacizumab is recommended for chemo-naïve, medically fit patients with recurrent/metastatic disease. Carboplatin/paclitaxel and cisplatin/paclitaxel are the preferred regimens [I, A]. The addition of bevacizumab to platinum-based chemotherapy is recommended when the risk of significant gastrointestinal/genitourinary toxicities has been carefully assessed and discussed with the patient [I, A]. The addition of pembrolizumab to platinum-based chemotherapy±bevacizumab is recommended in patients with PD-L1 positive tumors, assessed as combined positive score (CPS) of 1 or more [I, A]. Patients who progressed after first-line platinum-based chemotherapy should be offered treatment with the anti PD-1 agent, cemiplimab, regardless of PDL-1 tumor status as long as they had not previously received immunotherapy [I, A]. Patients with distant metastatic disease at diagnosis, who have responded to systemic chemotherapy, could be considered for additional radical pelvic radiotherapy (including IGBT in selected cases). Those with residual oligometastatic disease after systemic treatment could also be considered for additional regional treatment (surgery, thermal ablation, radiotherapy) to involved sites [IV, C]. Inclusion of patients with recurrent/metastatic disease in clinical trials is strongly recommended [V, A]. General Recommendations Patients should be informed and educated at the time of diagnosis and throughout follow-up about signs/symptoms of recurrence. They should be informed about possible side effects (by physicians, nurses, brochures, videos, etc.) [V, A]. A network of healthcare providers including all care providers should be involved in the care of survivors (eg, primary care physicians, gynecologists, psychologists, sexologists, physiotherapists, dieticians, social workers) for the follow-up [V, A]. Follow-up strategy should be individualized in terms of intensity, duration and procedures, taking into account individual risk assessment [V, A]. Available prognostic models, such as the Annual Risk Recurrence Calculator available on the ESGO website can be used to tailor surveillance strategy in an individual patient [IV, B]. Follow-up should be centralized/coordinated in a center specialized in the treatment and follow-up of gynecological cancer patients [IV, A]. Follow-up is designed to monitor disease response, to detect recurrence and to screen for subsequent primary tumors [V, B]. Regular and systematic monitoring of side effects and quality of life should be performed to improve the quality of care [V, A]. Prevention and early detection of immediate and persistent symptoms and side effects of the different cancer treatments and the individual patient supportive care needs should be identified and established at diagnosis and monitored throughout the follow-up [V, A]. All side effects should be identified and treated if possible, namely physical and psychosocial [V, A]. The development of an individual survivorship monitoring and care plan is recommended [V, B]. Recommendations for a healthy life style should include smoking cessation, regular exercise, healthy diet and weight management [V, B]. Clinical trials should address long-term cancer survivorship and should include patient related outcomes [V, B]. Quality control of care should be established [V, B]. Each visit should be composed of the following [V, A]: Patient history (including identification of relevant symptoms and side effects) Physical examination (including a speculum and bimanual pelvic examination) Imaging and laboratory tests should be performed only based on risk of recurrence, symptoms or findings suggestive of recurrence and/or side effects. Regular review of an ongoing survivorship plan that can be shared with other healthcare providers. Oncological follow-up Patients should be educated about symptoms and signs of potential recurrence [V, A]. Appropriate imaging test (MRI, ultrasound for pelvic assessment, CT scan or PET-CT for systemic assessment) should be used in symptomatic women [IV, A]. In case of suspected tumor persistence, recurrence or second primary cancer, histological verification is strongly recommended [V, A]. Vaginal vault cytology is not recommended [IV, D]. After fertility sparing treatment, follow-up should include HPV testing (at 6–12 and 24 months) [V, A]. Monitoring of quality of life and side effects Quality of life and side effects should be regularly assessed at least by the physicians/clinical care nurses and if possible by patients (using patient related outcomes). Patient self-reporting of side effects should be encouraged during and after treatment with the same frequency as medical visits [IV, B]. A checklist of potential main side effects should be included in the patient survivorship monitoring and care plan (eg, sexual dysfunction, lymphedema, menopausal symptoms and osteoporosis, genito-urinary and gastrointestinal disorders, chronic pain, fatigue) [IV, A]. After CTRT and BT, patients should be counseled about sexual rehabilitation measures including the use of vaginal dilators. Topical estrogens are indicated [IV, B]. Hormone replacement therapy is indicated to cervical cancer survivors with premature menopause and should be consistent with standard menopausal recommendation [IV, B]. Physical and lifestyle changes may also help [V, C]. Bone status should be assessed regularly in patients with early menopause [V, B]. Follow-up After Definitive CTRT and BT Follow-up should be performed/coordinated by a physician experienced with follow-up care after radiotherapy and BT including monitoring of early, and late treatment-related side effects [V, A]. The same imaging method used at the start of treatment should be used to assess tumor response [V, B]. Routine biopsy to assess complete remission should not be performed [IV, D]. Cytology is not recommended in detecting disease recurrence after radiotherapy [IV, D]. Imaging (pelvic MRI±CT scan or PET-CT) should be performed not earlier than 3 months after the end of treatment [IV, B]. In patients with uncertain complete remission at 3 months post-radiotherapy, the assessment should be repeated after an additional 2–3 months with biopsy if indicated [IV, B]. Patients should be informed and educated at the time of diagnosis and throughout follow-up about signs/symptoms of recurrence. They should be informed about possible side effects (by physicians, nurses, brochures, videos, etc.) [V, A]. A network of healthcare providers including all care providers should be involved in the care of survivors (eg, primary care physicians, gynecologists, psychologists, sexologists, physiotherapists, dieticians, social workers) for the follow-up [V, A]. Follow-up strategy should be individualized in terms of intensity, duration and procedures, taking into account individual risk assessment [V, A]. Available prognostic models, such as the Annual Risk Recurrence Calculator available on the ESGO website can be used to tailor surveillance strategy in an individual patient [IV, B]. Follow-up should be centralized/coordinated in a center specialized in the treatment and follow-up of gynecological cancer patients [IV, A]. Follow-up is designed to monitor disease response, to detect recurrence and to screen for subsequent primary tumors [V, B]. Regular and systematic monitoring of side effects and quality of life should be performed to improve the quality of care [V, A]. Prevention and early detection of immediate and persistent symptoms and side effects of the different cancer treatments and the individual patient supportive care needs should be identified and established at diagnosis and monitored throughout the follow-up [V, A]. All side effects should be identified and treated if possible, namely physical and psychosocial [V, A]. The development of an individual survivorship monitoring and care plan is recommended [V, B]. Recommendations for a healthy life style should include smoking cessation, regular exercise, healthy diet and weight management [V, B]. Clinical trials should address long-term cancer survivorship and should include patient related outcomes [V, B]. Quality control of care should be established [V, B]. Each visit should be composed of the following [V, A]: Patient history (including identification of relevant symptoms and side effects) Physical examination (including a speculum and bimanual pelvic examination) Imaging and laboratory tests should be performed only based on risk of recurrence, symptoms or findings suggestive of recurrence and/or side effects. Regular review of an ongoing survivorship plan that can be shared with other healthcare providers. Oncological follow-up Patients should be educated about symptoms and signs of potential recurrence [V, A]. Appropriate imaging test (MRI, ultrasound for pelvic assessment, CT scan or PET-CT for systemic assessment) should be used in symptomatic women [IV, A]. In case of suspected tumor persistence, recurrence or second primary cancer, histological verification is strongly recommended [V, A]. Vaginal vault cytology is not recommended [IV, D]. After fertility sparing treatment, follow-up should include HPV testing (at 6–12 and 24 months) [V, A]. Monitoring of quality of life and side effects Quality of life and side effects should be regularly assessed at least by the physicians/clinical care nurses and if possible by patients (using patient related outcomes). Patient self-reporting of side effects should be encouraged during and after treatment with the same frequency as medical visits [IV, B]. A checklist of potential main side effects should be included in the patient survivorship monitoring and care plan (eg, sexual dysfunction, lymphedema, menopausal symptoms and osteoporosis, genito-urinary and gastrointestinal disorders, chronic pain, fatigue) [IV, A]. After CTRT and BT, patients should be counseled about sexual rehabilitation measures including the use of vaginal dilators. Topical estrogens are indicated [IV, B]. Hormone replacement therapy is indicated to cervical cancer survivors with premature menopause and should be consistent with standard menopausal recommendation [IV, B]. Physical and lifestyle changes may also help [V, C]. Bone status should be assessed regularly in patients with early menopause [V, B]. Follow-up should be performed/coordinated by a physician experienced with follow-up care after radiotherapy and BT including monitoring of early, and late treatment-related side effects [V, A]. The same imaging method used at the start of treatment should be used to assess tumor response [V, B]. Routine biopsy to assess complete remission should not be performed [IV, D]. Cytology is not recommended in detecting disease recurrence after radiotherapy [IV, D]. Imaging (pelvic MRI±CT scan or PET-CT) should be performed not earlier than 3 months after the end of treatment [IV, B]. In patients with uncertain complete remission at 3 months post-radiotherapy, the assessment should be repeated after an additional 2–3 months with biopsy if indicated [IV, B]. General Recommendations Early palliative care, integrated with oncological treatments, should be offered by the clinical team to all the patients diagnosed with advanced cervical cancer for managing symptoms and improving quality of life. A multidisciplinary approach must be included in the care plan with discussion and planning for specific treatment of these symptoms [IV, A]. Pain Opioids are the main analgesics for the treatment of moderate to severe cancer-related pain; the first option is oral morphine [I, A]; but other opioids and alternative routes (transdermic, subcutaneous) can be required in specific situations (ie, intestinal obstruction, problems with swallowing, renal failure) [III, B]. If opioids alone do not provide sufficient pain relief cancer-related neuropathic pain should be treated with a combination of opioids and carefully dosed adjuvants (gabapentin, pregabalin, duloxetine, and tricyclic antidepressants) [III, B]. Severe pelvic cancer pain unresponsive to an opioid regimen can benefit from other procedures like plexus block or spinal analgesia techniques [III, B]. Palliative EBRT (if feasible) is effective for painful pelvic progression and bone metastasis [IV, B]. Renal Failure Urinary derivation by ureteral stent or percutaneous nephrostomy should be considered to treat renal failure caused by tumoral obstruction. There are no clear guidelines to predict which patients will benefit from these procedures in terms of survival and quality of life, and its indication should be discussed carefully [IV, C]. Malignant Intestinal Obstruction Medical management of malignant intestinal obstruction consists of antisecretory, corticosteroids, and antiemetic drugs. A nasogastric tube is recommended if vomiting and discomfort persist in spite of medical management. Surgical procedures can be considered in selected patients [IV, B]. Vaginal Bleeding and Discharges In the case of vaginal bleeding, vaginal packing, interventional radiology (selective embolization) or palliative radiotherapy (if feasible) are recommended. There is not enough evidence to prefer one over the other. In the case of massive refractory bleeding, palliative sedation can be considered. Malodorous vaginal discharge can be improved with vaginal washing and the use of a vaginal metronidazole tablet [IV, B]. Psychosocial Suffering In patients with cervical advanced cancer, a multidisciplinary approach of physicians, nurses, psychologists, social workers, and community health workers is needed to manage psychosocial and spiritual suffering associated with social stigma deriving from genital disease, malodorous vaginal discharge, etc [IV, A]. Early palliative care, integrated with oncological treatments, should be offered by the clinical team to all the patients diagnosed with advanced cervical cancer for managing symptoms and improving quality of life. A multidisciplinary approach must be included in the care plan with discussion and planning for specific treatment of these symptoms [IV, A]. Opioids are the main analgesics for the treatment of moderate to severe cancer-related pain; the first option is oral morphine [I, A]; but other opioids and alternative routes (transdermic, subcutaneous) can be required in specific situations (ie, intestinal obstruction, problems with swallowing, renal failure) [III, B]. If opioids alone do not provide sufficient pain relief cancer-related neuropathic pain should be treated with a combination of opioids and carefully dosed adjuvants (gabapentin, pregabalin, duloxetine, and tricyclic antidepressants) [III, B]. Severe pelvic cancer pain unresponsive to an opioid regimen can benefit from other procedures like plexus block or spinal analgesia techniques [III, B]. Palliative EBRT (if feasible) is effective for painful pelvic progression and bone metastasis [IV, B]. Urinary derivation by ureteral stent or percutaneous nephrostomy should be considered to treat renal failure caused by tumoral obstruction. There are no clear guidelines to predict which patients will benefit from these procedures in terms of survival and quality of life, and its indication should be discussed carefully [IV, C]. Medical management of malignant intestinal obstruction consists of antisecretory, corticosteroids, and antiemetic drugs. A nasogastric tube is recommended if vomiting and discomfort persist in spite of medical management. Surgical procedures can be considered in selected patients [IV, B]. In the case of vaginal bleeding, vaginal packing, interventional radiology (selective embolization) or palliative radiotherapy (if feasible) are recommended. There is not enough evidence to prefer one over the other. In the case of massive refractory bleeding, palliative sedation can be considered. Malodorous vaginal discharge can be improved with vaginal washing and the use of a vaginal metronidazole tablet [IV, B]. In patients with cervical advanced cancer, a multidisciplinary approach of physicians, nurses, psychologists, social workers, and community health workers is needed to manage psychosocial and spiritual suffering associated with social stigma deriving from genital disease, malodorous vaginal discharge, etc [IV, A]. General Recommendations Every patient diagnosed with cervical cancer in pregnancy must be counseled by a multidisciplinary team. This team should consist of experts in the fields of gynecological oncology, neonatology, obstetrics, pathology, anesthesiology, radiation oncology, medical oncology, psycho-oncology, and, spiritual and ethical counseling. National or international tumor board counseling may be considered [V, A]. Given the large spectrum of therapeutic options, the multidisciplinary team should recommend a treatment plan according to the patient’s intention, tumor stage, and gestational age of pregnancy at the time of cancer diagnosis. The primary aims of the recommended treatment plan are the oncological safety of the pregnant woman as well as the fetal survival without additional morbidity [V, A]. Treatment of patients with cervical cancer in pregnancy should be exclusively done in gynecological oncology centers associated with the highest level perinatal center with expertise in all aspects of oncologic therapy in pregnancy and intensive medical care of premature neonates [V, A]. Clinical and Imaging Diagnosis Clinical examination and histological verification of cervical cancer are mandatory [IV, A]. Pathological confirmation may be obtained by colposcopy oriented biopsy or small cone (appropriate only during the first trimester of pregnancy, endocervical curettage is contraindicated) [IV, C]. Preferred imaging modalities for clinical staging in patients with cervical cancer in pregnancy include pelvic MRI or expert ultrasound as part of the primary work-up. Gadolinium-based contrast agents should be avoided [III, A]. The use of whole-body diffusion-weighted imaging MRI (WB-DWI/MRI) can reliably obviate the need for gadolinium contrast and radiation for nodal and distant staging during pregnancy. If not available, chest CT scan with abdominal shielding is an alternative. PET-CT should be avoided during pregnancy [IV, B]. Oncological Management Tumor involvement of suspicious nodes should be histologically confirmed because of its prognostic significance and the impact on the management up to 24 weeks of gestation (fetal viability) [IV, A]. Minimally invasive approach could be considered before 14–16 weeks of gestation; however, the sentinel node biopsy concept using indocyanine green is still experimental [IV, C]. Several treatment modalities are available and should be discussed with the patient taking into account the tumor stage, gestational week of pregnancy and the patient’s preferences [IV, B]: Delay of oncological treatment until fetal maturity (if possible >34 weeks of gestation) and initiate cancer-specific treatment immediately after delivery by cesarean section. This option might be considered if the term or fetal maturity is approaching. Conization or simple trachelectomy in order to completely remove the tumor, obtain free margins and perform nodal staging if needed, with the intention to preserve the pregnancy. Radical surgery or definitive CTRT according to the disease stage as recommended outside pregnancy, if the woman decides not to preserve the pregnancy. Pregnancy termination is recommended before any treatment after the first trimester, and fetus evacuation before CTRT, if possible. Chemotherapy until term of pregnancy (37 weeks of gestation) and initiation of definitive cancer-specific treatment immediately after delivery by cesarean section. At least a 2 week interval between chemotherapy and surgery is recommended. In patients with locally advanced disease or residual tumor after surgical procedure that cannot be completely removed (risk of premature rupture of amniotic membranes and/or cervical insufficiency), chemotherapy based on cisplatin or carboplatin can be considered starting after 14 weeks of pregnancy. Combination with taxanes is an option. Bevacizumab and checkpoint inhibitors are contraindicated. Before starting each cycle of chemotherapy, an assessment of treatment response should be made by clinical examination and transvaginal or transrectal ultrasound. If no response is achieved after 2 cycles of chemotherapy during pregnancy, treatment strategy should be re-evaluated. Pregnancy Management Spontaneous delivery appears to have negative prognostic impact in patients with cervical cancer in pregnancy. Thus, cesarean section is the recommended mode of delivery [IV, B]. At the time of cesarean section, definitive cancer specific treatment should be performed corresponding to that of non-pregnant women, taking into account the treatment that has already been given during pregnancy [IV, A]. Every patient diagnosed with cervical cancer in pregnancy must be counseled by a multidisciplinary team. This team should consist of experts in the fields of gynecological oncology, neonatology, obstetrics, pathology, anesthesiology, radiation oncology, medical oncology, psycho-oncology, and, spiritual and ethical counseling. National or international tumor board counseling may be considered [V, A]. Given the large spectrum of therapeutic options, the multidisciplinary team should recommend a treatment plan according to the patient’s intention, tumor stage, and gestational age of pregnancy at the time of cancer diagnosis. The primary aims of the recommended treatment plan are the oncological safety of the pregnant woman as well as the fetal survival without additional morbidity [V, A]. Treatment of patients with cervical cancer in pregnancy should be exclusively done in gynecological oncology centers associated with the highest level perinatal center with expertise in all aspects of oncologic therapy in pregnancy and intensive medical care of premature neonates [V, A]. Clinical examination and histological verification of cervical cancer are mandatory [IV, A]. Pathological confirmation may be obtained by colposcopy oriented biopsy or small cone (appropriate only during the first trimester of pregnancy, endocervical curettage is contraindicated) [IV, C]. Preferred imaging modalities for clinical staging in patients with cervical cancer in pregnancy include pelvic MRI or expert ultrasound as part of the primary work-up. Gadolinium-based contrast agents should be avoided [III, A]. The use of whole-body diffusion-weighted imaging MRI (WB-DWI/MRI) can reliably obviate the need for gadolinium contrast and radiation for nodal and distant staging during pregnancy. If not available, chest CT scan with abdominal shielding is an alternative. PET-CT should be avoided during pregnancy [IV, B]. Tumor involvement of suspicious nodes should be histologically confirmed because of its prognostic significance and the impact on the management up to 24 weeks of gestation (fetal viability) [IV, A]. Minimally invasive approach could be considered before 14–16 weeks of gestation; however, the sentinel node biopsy concept using indocyanine green is still experimental [IV, C]. Several treatment modalities are available and should be discussed with the patient taking into account the tumor stage, gestational week of pregnancy and the patient’s preferences [IV, B]: Delay of oncological treatment until fetal maturity (if possible >34 weeks of gestation) and initiate cancer-specific treatment immediately after delivery by cesarean section. This option might be considered if the term or fetal maturity is approaching. Conization or simple trachelectomy in order to completely remove the tumor, obtain free margins and perform nodal staging if needed, with the intention to preserve the pregnancy. Radical surgery or definitive CTRT according to the disease stage as recommended outside pregnancy, if the woman decides not to preserve the pregnancy. Pregnancy termination is recommended before any treatment after the first trimester, and fetus evacuation before CTRT, if possible. Chemotherapy until term of pregnancy (37 weeks of gestation) and initiation of definitive cancer-specific treatment immediately after delivery by cesarean section. At least a 2 week interval between chemotherapy and surgery is recommended. In patients with locally advanced disease or residual tumor after surgical procedure that cannot be completely removed (risk of premature rupture of amniotic membranes and/or cervical insufficiency), chemotherapy based on cisplatin or carboplatin can be considered starting after 14 weeks of pregnancy. Combination with taxanes is an option. Bevacizumab and checkpoint inhibitors are contraindicated. Before starting each cycle of chemotherapy, an assessment of treatment response should be made by clinical examination and transvaginal or transrectal ultrasound. If no response is achieved after 2 cycles of chemotherapy during pregnancy, treatment strategy should be re-evaluated. Spontaneous delivery appears to have negative prognostic impact in patients with cervical cancer in pregnancy. Thus, cesarean section is the recommended mode of delivery [IV, B]. At the time of cesarean section, definitive cancer specific treatment should be performed corresponding to that of non-pregnant women, taking into account the treatment that has already been given during pregnancy [IV, A]. Histopathological diagnosis of rare cervical tumors needs confirmation (second opinion) by an expert pathologist [IV, A]. Treatment and care of rare cervical tumors needs to be centralized at referral centers and discussed in a multidisciplinary tumor board [IV, A]. Management of T1a Disease Primary Treatment of T1b1, T1b2, and T2a1 Tumors Adjuvant Treatment of T1b1, T1b2, and T2a1 Tumors Fertility Sparing Treatment - Selection of Candidates Fertility Sparing Treatment - Management Invasive Cervical Cancer Diagnosed on a Simple Hysterectomy Specimen Management of Locally Advanced Disease Cervical Cancer in Pregnancy Recurrent Disease Distant Recurrent and Metastatic Disease Definitive CTRT and BT - General Aspects Definitive management (ie, without tumor related surgery) consists of EBRT with concomitant platinum-based chemotherapy and BT. Delay of treatment and/or treatment interruptions have to be prevented to avoid tumor progression and accelerated repopulation. The overall treatment time including both EBRT and BT should therefore not exceed 7 weeks. Definitive CTRT and BT CTRT Target contouring for EBRT should be based on 3D imaging (preferably fused MRI and PET-CT) performed in the supine treatment position. Controlled bladder filling is recommended to minimize uterus movements and to push the intestines away. The result of the gynecological examination (ie, clinical drawing and description) as well as diagnostic imaging should be available during the contouring phase. A contouring protocol including a margin strategy for handling of internal movement (ITV) should be used to minimize irradiation of organs at risk. The EMBRACE II protocol may serve as a template. The tumor related target volume for EBRT (CTV-T-LR) includes the primary cervical tumor (GTV-T), the uterus, parametria and upper vagina (or minimal 2 cm tumour-free margin below any vaginal infiltration respectively) and is optimally defined on MRI with assistance of the clinical findings. The elective target (CTV-E) includes the obturator, internal, external and common iliac and presacral regions. The inguinal nodes should be included if the primary tumor involves the distal third of the vagina. A reduced elective target volume for EBRT without the common iliac nodes may be considered in low- and intermediate-risk T1b1 patients with negative LN and no LVSI. In case of PLN involvement indicating an increased risk of PALN spread (i.e.>2 pathological LN or involvement of common iliac region) and absence of surgical para-aortic staging, the elective target for EBRT should include the para-aortic region up to the renal vessels. In case of PALN involvement, the target volume includes at a minimum the region up to the renal vessels. Pathological macroscopic LN (GTV-N) are optimally localized with PET-CT and contoured on MRI. The planning aim for EBRT is 45 Gy/25 fractions or 46 Gy/23 fractions using intensity-modulated radiotherapy/volumetric modulated arc therapy (IMRT/VMAT). A homogeneous dose from EBRT is needed in the central pelvis to ensure a safe platform for planning of BT. The use of an EBRT boost to the primary tumor and/or the parametria for complete or partial replacement of BT is not recommended. Pathological macroscopic LN (GTV-N) should receive an EBRT boost. Simultaneous integrated boosting using coverage probability planning is recommended. Depending on nodal size and the expected dose contribution from BT a total dose of approximately 60 Gy EQD2 should be the aim of treatment. An alternative treatment option is surgical removal of enlarged nodes. Image-guided radiotherapy with daily on-board 3D imaging is recommended for IMRT/VMAT to ensure safe dose application with limited PTV margins. Concomitant chemotherapy should be based on single-agent radiosensitizing chemotherapy, preferably cisplatin (weekly 40 mg/m²). If cisplatin is not applicable, alternative treatment options are weekly carboplatin (area under the curve (AUC) =2) or hyperthermia (if available). EBRT may also be applied without concomitant chemotherapy or hyperthermia according to patient selection (ie, patients unfit for any chemotherapy). Brachytherapy IGABT is recommended, preferably using MRI with applicator in place. Repeated gynaecologic examination is mandatory, and alternative imaging modalities such as CT scan and ultrasound may be used. The tumour-related targets for BT include: 1) the residual gross tumor volume (GTV-T res ) after CTRT; 2) the adaptive high-risk clinical target volume (CTV-T HR ) including the whole cervix and residual adjacent pathologic tissue; and 3) the intermediate-risk clinical target volume (CTV-T IR ) taking the initial tumor extent into consideration. The BT applicator should consist of a uterine tandem and a vaginal component (ovoids/ring/mold/combined ring/ovoid). A combined intracavitary/interstitial implant is recommended in advanced cases to achieve the dose planning aim (see below), in particular in case of residual disease in the parametrium. Ultrasound (transabdominal and/or transrectal) maybe used to intraoperatively support applicator insertion (avoidance of uterine perforation by the tandem, guidance of interstitial needles). In IGABT, the planning aim should be to deliver a BT dose of 40 to 45 Gy EQD2 to reach a total EBRT+BT dose of 85 to 95 Gy EQD2 (D90) (assuming 45 Gy through EBRT) to the CTV-T HR , equal to or greater than 60 Gy (D98) to the CTV-T IR , and equal to or greater than 90 Gy (D98) to the GTV-T res . The use of three dimensional and 2D dose volume and point constraints for rectum, bladder, vagina, sigmoid, and bowel are recommended, and they have to be based on the published clinical evidence. Even though point A dose reporting and prescription have been surpassed by the volumetric approach, a point A dose standard plan should be used as a starting point for stepwise treatment plan optimization to retain the pear shaped iso-dose pattern with a high central dose. This is especially important for the combined intracavitary/interstitial technique to avoid overloading of the interstitial needles. BT should be delivered in several fractions as high dose rate (usually 3–4) with at least 6–8 hours interval or pulse dose rate delivered in one fraction (50–60 hourly pulses) or 2–3 fractions (15–24 hourly pulses) to respect the limitations of current radiobiological models for speed and capacity of radiation damage repair. In large tumors, BT should be delivered within 1 to 2 weeks toward the end of or after CTRT. In limited-size tumors, BT may start earlier during CTRT. For the tumour-related targets (GTV-T res , CTV-T HR , CTV-T IR ), the use of external beam therapy for giving an extra dose (eg, parametrial boost, cervix boost) is not recommended, even when using advanced EBRT technology such as stereotactic radiotherapy or particle therapy. The use of a midline block for boosting the parametrium is not recommended when applying advanced image-guided radiotherapy and IGABT. Care should be taken to optimize patient comfort during (fractionated) BT. Preferably this includes a multidisciplinary approach. Intracavitary and combined intracavitary/interstitial BT implants should be performed under anesthesia. Adjuvant Radiotherapy or CTRT Adjuvant radiotherapy or CTRT follows analog principles for target contouring, dose and fractionation as outlined for definitive treatment. Different concomitant and/or sequential chemotherapy schedules have been established including cisplatin alone or combinations of cisplatin with other agents such as 5-FU or paclitaxel. Carboplatin should be considered for patients unfit for cisplatin. The application of IMRT/VMAT and image-guided radiotherapy is recommended as treatment-related morbidity is reduced. Additional BT as part of adjuvant radiotherapy or CTRT should be considered only if a well-defined limited area accessible through a BT technique is at high risk of local recurrence (eg, positive resection margins in vagina or parametrium). Such adjuvant BT should follow the major principles outlined above for IGBT. Definitive 3D Conformal EBRT or CTRT and Radiography-based BT Three-dimensional conformal radiotherapy alone or as definitive concomitant CTRT (platinum based) ± para-aortic radiotherapy and/or 2D radiography based BT is recommended, if intensity modulated radiotherapy and/or IGABT are not available. In case of 3D conformal radiotherapy and/or radiography based BT, the recommendations for EBRT and IGABT as outlined above in regard to target, dose, fractionation, and overall treatment time have to be respected as much as possible. A sequential LN boost is applied as appropriate after completion of 3D EBRT. Planning aim for BT should be based on point A. Dose to point A should be equal to or greater than 75 Gy (EQD2) in limited width adaptive CTV-T HR (≤3 cm) and should aim at higher doses in large width adaptive CTV-T HR (>4 cm). In addition, dose for the maximum width of the adaptive CTV-T HR should be reported. Radiography based dose point constraints - plus 3D dose volume constraints as available - for rectum, bladder, vagina, sigmoid, and bowel are recommended, and must be based on published clinical evidence. Definitive management (ie, without tumor related surgery) consists of EBRT with concomitant platinum-based chemotherapy and BT. Delay of treatment and/or treatment interruptions have to be prevented to avoid tumor progression and accelerated repopulation. The overall treatment time including both EBRT and BT should therefore not exceed 7 weeks. CTRT Target contouring for EBRT should be based on 3D imaging (preferably fused MRI and PET-CT) performed in the supine treatment position. Controlled bladder filling is recommended to minimize uterus movements and to push the intestines away. The result of the gynecological examination (ie, clinical drawing and description) as well as diagnostic imaging should be available during the contouring phase. A contouring protocol including a margin strategy for handling of internal movement (ITV) should be used to minimize irradiation of organs at risk. The EMBRACE II protocol may serve as a template. The tumor related target volume for EBRT (CTV-T-LR) includes the primary cervical tumor (GTV-T), the uterus, parametria and upper vagina (or minimal 2 cm tumour-free margin below any vaginal infiltration respectively) and is optimally defined on MRI with assistance of the clinical findings. The elective target (CTV-E) includes the obturator, internal, external and common iliac and presacral regions. The inguinal nodes should be included if the primary tumor involves the distal third of the vagina. A reduced elective target volume for EBRT without the common iliac nodes may be considered in low- and intermediate-risk T1b1 patients with negative LN and no LVSI. In case of PLN involvement indicating an increased risk of PALN spread (i.e.>2 pathological LN or involvement of common iliac region) and absence of surgical para-aortic staging, the elective target for EBRT should include the para-aortic region up to the renal vessels. In case of PALN involvement, the target volume includes at a minimum the region up to the renal vessels. Pathological macroscopic LN (GTV-N) are optimally localized with PET-CT and contoured on MRI. The planning aim for EBRT is 45 Gy/25 fractions or 46 Gy/23 fractions using intensity-modulated radiotherapy/volumetric modulated arc therapy (IMRT/VMAT). A homogeneous dose from EBRT is needed in the central pelvis to ensure a safe platform for planning of BT. The use of an EBRT boost to the primary tumor and/or the parametria for complete or partial replacement of BT is not recommended. Pathological macroscopic LN (GTV-N) should receive an EBRT boost. Simultaneous integrated boosting using coverage probability planning is recommended. Depending on nodal size and the expected dose contribution from BT a total dose of approximately 60 Gy EQD2 should be the aim of treatment. An alternative treatment option is surgical removal of enlarged nodes. Image-guided radiotherapy with daily on-board 3D imaging is recommended for IMRT/VMAT to ensure safe dose application with limited PTV margins. Concomitant chemotherapy should be based on single-agent radiosensitizing chemotherapy, preferably cisplatin (weekly 40 mg/m²). If cisplatin is not applicable, alternative treatment options are weekly carboplatin (area under the curve (AUC) =2) or hyperthermia (if available). EBRT may also be applied without concomitant chemotherapy or hyperthermia according to patient selection (ie, patients unfit for any chemotherapy). Brachytherapy IGABT is recommended, preferably using MRI with applicator in place. Repeated gynaecologic examination is mandatory, and alternative imaging modalities such as CT scan and ultrasound may be used. The tumour-related targets for BT include: 1) the residual gross tumor volume (GTV-T res ) after CTRT; 2) the adaptive high-risk clinical target volume (CTV-T HR ) including the whole cervix and residual adjacent pathologic tissue; and 3) the intermediate-risk clinical target volume (CTV-T IR ) taking the initial tumor extent into consideration. The BT applicator should consist of a uterine tandem and a vaginal component (ovoids/ring/mold/combined ring/ovoid). A combined intracavitary/interstitial implant is recommended in advanced cases to achieve the dose planning aim (see below), in particular in case of residual disease in the parametrium. Ultrasound (transabdominal and/or transrectal) maybe used to intraoperatively support applicator insertion (avoidance of uterine perforation by the tandem, guidance of interstitial needles). In IGABT, the planning aim should be to deliver a BT dose of 40 to 45 Gy EQD2 to reach a total EBRT+BT dose of 85 to 95 Gy EQD2 (D90) (assuming 45 Gy through EBRT) to the CTV-T HR , equal to or greater than 60 Gy (D98) to the CTV-T IR , and equal to or greater than 90 Gy (D98) to the GTV-T res . The use of three dimensional and 2D dose volume and point constraints for rectum, bladder, vagina, sigmoid, and bowel are recommended, and they have to be based on the published clinical evidence. Even though point A dose reporting and prescription have been surpassed by the volumetric approach, a point A dose standard plan should be used as a starting point for stepwise treatment plan optimization to retain the pear shaped iso-dose pattern with a high central dose. This is especially important for the combined intracavitary/interstitial technique to avoid overloading of the interstitial needles. BT should be delivered in several fractions as high dose rate (usually 3–4) with at least 6–8 hours interval or pulse dose rate delivered in one fraction (50–60 hourly pulses) or 2–3 fractions (15–24 hourly pulses) to respect the limitations of current radiobiological models for speed and capacity of radiation damage repair. In large tumors, BT should be delivered within 1 to 2 weeks toward the end of or after CTRT. In limited-size tumors, BT may start earlier during CTRT. For the tumour-related targets (GTV-T res , CTV-T HR , CTV-T IR ), the use of external beam therapy for giving an extra dose (eg, parametrial boost, cervix boost) is not recommended, even when using advanced EBRT technology such as stereotactic radiotherapy or particle therapy. The use of a midline block for boosting the parametrium is not recommended when applying advanced image-guided radiotherapy and IGABT. Care should be taken to optimize patient comfort during (fractionated) BT. Preferably this includes a multidisciplinary approach. Intracavitary and combined intracavitary/interstitial BT implants should be performed under anesthesia. Target contouring for EBRT should be based on 3D imaging (preferably fused MRI and PET-CT) performed in the supine treatment position. Controlled bladder filling is recommended to minimize uterus movements and to push the intestines away. The result of the gynecological examination (ie, clinical drawing and description) as well as diagnostic imaging should be available during the contouring phase. A contouring protocol including a margin strategy for handling of internal movement (ITV) should be used to minimize irradiation of organs at risk. The EMBRACE II protocol may serve as a template. The tumor related target volume for EBRT (CTV-T-LR) includes the primary cervical tumor (GTV-T), the uterus, parametria and upper vagina (or minimal 2 cm tumour-free margin below any vaginal infiltration respectively) and is optimally defined on MRI with assistance of the clinical findings. The elective target (CTV-E) includes the obturator, internal, external and common iliac and presacral regions. The inguinal nodes should be included if the primary tumor involves the distal third of the vagina. A reduced elective target volume for EBRT without the common iliac nodes may be considered in low- and intermediate-risk T1b1 patients with negative LN and no LVSI. In case of PLN involvement indicating an increased risk of PALN spread (i.e.>2 pathological LN or involvement of common iliac region) and absence of surgical para-aortic staging, the elective target for EBRT should include the para-aortic region up to the renal vessels. In case of PALN involvement, the target volume includes at a minimum the region up to the renal vessels. Pathological macroscopic LN (GTV-N) are optimally localized with PET-CT and contoured on MRI. The planning aim for EBRT is 45 Gy/25 fractions or 46 Gy/23 fractions using intensity-modulated radiotherapy/volumetric modulated arc therapy (IMRT/VMAT). A homogeneous dose from EBRT is needed in the central pelvis to ensure a safe platform for planning of BT. The use of an EBRT boost to the primary tumor and/or the parametria for complete or partial replacement of BT is not recommended. Pathological macroscopic LN (GTV-N) should receive an EBRT boost. Simultaneous integrated boosting using coverage probability planning is recommended. Depending on nodal size and the expected dose contribution from BT a total dose of approximately 60 Gy EQD2 should be the aim of treatment. An alternative treatment option is surgical removal of enlarged nodes. Image-guided radiotherapy with daily on-board 3D imaging is recommended for IMRT/VMAT to ensure safe dose application with limited PTV margins. Concomitant chemotherapy should be based on single-agent radiosensitizing chemotherapy, preferably cisplatin (weekly 40 mg/m²). If cisplatin is not applicable, alternative treatment options are weekly carboplatin (area under the curve (AUC) =2) or hyperthermia (if available). EBRT may also be applied without concomitant chemotherapy or hyperthermia according to patient selection (ie, patients unfit for any chemotherapy). IGABT is recommended, preferably using MRI with applicator in place. Repeated gynaecologic examination is mandatory, and alternative imaging modalities such as CT scan and ultrasound may be used. The tumour-related targets for BT include: 1) the residual gross tumor volume (GTV-T res ) after CTRT; 2) the adaptive high-risk clinical target volume (CTV-T HR ) including the whole cervix and residual adjacent pathologic tissue; and 3) the intermediate-risk clinical target volume (CTV-T IR ) taking the initial tumor extent into consideration. The BT applicator should consist of a uterine tandem and a vaginal component (ovoids/ring/mold/combined ring/ovoid). A combined intracavitary/interstitial implant is recommended in advanced cases to achieve the dose planning aim (see below), in particular in case of residual disease in the parametrium. Ultrasound (transabdominal and/or transrectal) maybe used to intraoperatively support applicator insertion (avoidance of uterine perforation by the tandem, guidance of interstitial needles). In IGABT, the planning aim should be to deliver a BT dose of 40 to 45 Gy EQD2 to reach a total EBRT+BT dose of 85 to 95 Gy EQD2 (D90) (assuming 45 Gy through EBRT) to the CTV-T HR , equal to or greater than 60 Gy (D98) to the CTV-T IR , and equal to or greater than 90 Gy (D98) to the GTV-T res . The use of three dimensional and 2D dose volume and point constraints for rectum, bladder, vagina, sigmoid, and bowel are recommended, and they have to be based on the published clinical evidence. Even though point A dose reporting and prescription have been surpassed by the volumetric approach, a point A dose standard plan should be used as a starting point for stepwise treatment plan optimization to retain the pear shaped iso-dose pattern with a high central dose. This is especially important for the combined intracavitary/interstitial technique to avoid overloading of the interstitial needles. BT should be delivered in several fractions as high dose rate (usually 3–4) with at least 6–8 hours interval or pulse dose rate delivered in one fraction (50–60 hourly pulses) or 2–3 fractions (15–24 hourly pulses) to respect the limitations of current radiobiological models for speed and capacity of radiation damage repair. In large tumors, BT should be delivered within 1 to 2 weeks toward the end of or after CTRT. In limited-size tumors, BT may start earlier during CTRT. For the tumour-related targets (GTV-T res , CTV-T HR , CTV-T IR ), the use of external beam therapy for giving an extra dose (eg, parametrial boost, cervix boost) is not recommended, even when using advanced EBRT technology such as stereotactic radiotherapy or particle therapy. The use of a midline block for boosting the parametrium is not recommended when applying advanced image-guided radiotherapy and IGABT. Care should be taken to optimize patient comfort during (fractionated) BT. Preferably this includes a multidisciplinary approach. Intracavitary and combined intracavitary/interstitial BT implants should be performed under anesthesia. Adjuvant radiotherapy or CTRT follows analog principles for target contouring, dose and fractionation as outlined for definitive treatment. Different concomitant and/or sequential chemotherapy schedules have been established including cisplatin alone or combinations of cisplatin with other agents such as 5-FU or paclitaxel. Carboplatin should be considered for patients unfit for cisplatin. The application of IMRT/VMAT and image-guided radiotherapy is recommended as treatment-related morbidity is reduced. Additional BT as part of adjuvant radiotherapy or CTRT should be considered only if a well-defined limited area accessible through a BT technique is at high risk of local recurrence (eg, positive resection margins in vagina or parametrium). Such adjuvant BT should follow the major principles outlined above for IGBT. Three-dimensional conformal radiotherapy alone or as definitive concomitant CTRT (platinum based) ± para-aortic radiotherapy and/or 2D radiography based BT is recommended, if intensity modulated radiotherapy and/or IGABT are not available. In case of 3D conformal radiotherapy and/or radiography based BT, the recommendations for EBRT and IGABT as outlined above in regard to target, dose, fractionation, and overall treatment time have to be respected as much as possible. A sequential LN boost is applied as appropriate after completion of 3D EBRT. Planning aim for BT should be based on point A. Dose to point A should be equal to or greater than 75 Gy (EQD2) in limited width adaptive CTV-T HR (≤3 cm) and should aim at higher doses in large width adaptive CTV-T HR (>4 cm). In addition, dose for the maximum width of the adaptive CTV-T HR should be reported. Radiography based dose point constraints - plus 3D dose volume constraints as available - for rectum, bladder, vagina, sigmoid, and bowel are recommended, and must be based on published clinical evidence. Requirements for Specimen Submitted for Pathological Evaluation Patient information, previous cervical cytology, histological specimens, clinical and radiological data, colposcopic findings and information on previous treatment (eg, surgery, radiotherapy) need to be included on the specimen request form. Details of cytology, biopsy, and surgical specimen (cone/loop specimen, trachelectomy, type of hysterectomy, presence of ovaries and fallopian tubes, presence of LN and designation of the LN sites, presence of vaginal cuff, and presence of parametria) need to be itemized in the specimen request form. Biopsies and surgical specimens should be sent to the pathology department in a container with liquid fixative (‘‘clamping’’ of surgical specimens on a surface may be useful). If the local situation requires biobanking of fresh tissue, surgical specimens should be submitted fresh with minimum ischemia time. Cytology specimens should be sent to the pathology department preferentially as liquid-based cytology. Smear preparations are not recommended. The former is necessary when an HPV test is requested. Immunocytochemistry is possible on LBC but of limited extent (eg, CPS score for PD-L1 cannot be assessed). Cone/loop specimen should ideally be sent intact with a suture to identify the 12-o’clock position. Specimen Grossing and Sampling Biopsy/Cone/Loop Small biopsy specimens should be enumerated. The cone/loop specimens should be measured in three dimensions according to the recent ESGO/ESP recommendations. If the cone can be oriented properly, the anterior and the posterior half should be inked with separate colors. It should further be recorded if the specimen is complete or fragmented. If more than one piece of tissue is received, every piece should be measured in three dimensions. All specimens should be entirely submitted for microscopic examination. Inking of the surgical margins of cone/loop specimens is recommended. Dissection of cone/loop specimens should be performed in a standardized procedure. All the pieces submitted should be in consecutive numerical order. This is important because if tumor is present in more than one piece, it needs to be known whether these pieces are consecutive and, thus, a single tumor is present or whether the tumor is multifocal. It is recommended to place only one piece of tissue in each cassette. There are also techniques that allow embedding of more than one piece in a cassette if they are small enough. In cases that do not comprise intact cone/loops, serial radial sectioning and placing of each slice of tissue in a single cassette should be performed. Trachelectomy The upper (proximal) surgical margin of a trachelectomy specimen should be inked. The upper margin of a trachelectomy specimen should be sampled in its entirety in a way that allows to measure the distance of the tumor to the margin. The vaginal margin should also be inked and examined totally as radial sections if no tumor is seen grossly. Hysterectomy The description of the specimen (hysterectomy, trachelectomy, presence of ovaries and fallopian tubes, presence of LN and indication of the LN sites, presence of vaginal cuff and presence of parametria) should be recorded and checked for consistency with the description given in the specimen request form. The presence of any gross abnormality in any organ should be documented. The dimensions of the uterus for a hysterectomy specimen and the cervix for a trachelectomy specimen should be documented. The minimum and maximal length of the vaginal cuff should be documented. The size of the parametria should be documented in two dimensions (vertical and horizontal). Gross tumor involvement of the parametrium, vagina, uterine corpus, or other organs should be documented. The relationship of the cervical tumor to the vaginal and parametrial margins (and upper margin in case of a trachelectomy specimen) should be measured and appropriate sections taken to demonstrate this. Radial/circumferential and vaginal margins should be inked. The gross appearance of the cervix should be documented and any gross tumor mass measured. If visible, the site of a previous cone biopsy should be documented. Gross tumors should be measured in three dimensions, namely, the horizontal extent and the depth of invasion. The tumor site within the cervix should be documented. The cervical tumor should be sampled to demonstrate the maximum depth of invasion, the relationship of the tumor with the surgical borders, and the extension to other organs. When the tumor is small (or with tumors that cannot be identified macroscopically), the cervix should be separated from the corpus, opened and processed as for a cone/ loop specimen. In the case of a large tumor, the hysterectomy or trachelectomy specimen should be opened in the sagittal plane. At least one block per centimeter of the greatest tumor dimension should be taken for large tumors. Additional blocks including the cervix adjacent to the tumor should be taken to identify precursor lesions. The whole cervix should be sampled in the case of a small tumor or where no macroscopic tumor is identified. The uterine corpus, vagina, and adnexa should be sampled according to standard protocols if not involved by tumor. If the uterine corpus and/or adnexa are grossly involved, additional blocks should be sampled. The entire vaginal margin should be blocked. The parametria should be submitted totally for histological examination to assess tumor invasion and surgical margins. The use of large sections is optional and provides good information on tumor size and marginal status. Lymph Nodes All the LN should be submitted for histological examination. If the LN are grossly involved, representative samples are sufficient. If grossly uninvolved, each node should be sliced at 2 mm interval (eg, perpendicular to its longitudinal axis) and totally embedded. From each block, hematoxylin-eosin (H&E) sections should be taken. LN should be submitted in separate cassettes according to the site recorded on the specimen request form. Pathological Analysis of SLN Intraoperative assessment of sentinel nodes is a reliable procedure but may miss micrometastases and isolated tumor cells. Intraoperative assessment should be performed on a grossly suspicious sentinel node and may be performed on a “non-suspicious” SLN because the confirmation of tumor involvement will result in abandoning a hysterectomy or trachelectomy. For intraoperative evaluation, the SLN should be sent to the pathology department in a container without liquid fixative. Intraoperative analysis requires gross dissection of the resected adipose tissue by the pathologist and selection of LN. It is important to leave some peri-nodal tissue allowing proper diagnosis of extra-nodal tumor spread. For a LN with obvious gross tumor, a single section is adequate for frozen section. Frozen section may be combined with imprint cytology. The use of one step nucleic acid amplification is not recommended particularly due to the interference with benign epithelial inclusions in PLN. Any nonsuspicious sentinel node should be bisected (if small) or sliced at (approximately) 2 mm thickness and entirely frozen. From each sample, histological sections should be cut and stained by H&E. After frozen section analysis, the tissue should be put into a cassette, fixed in liquid fixative (preferably 4% buffered formalin) and subsequently processed and embedded in paraffin. If no metastases are present in the first section, SLN should undergo ultrastaging in definitive paraffin sections, including immunohistochemistry. A minimum procedure should include five serial sections at 200 µm. At least, at two levels an additional section must be cut and stained with a broad-spectrum cytokeratin antibody (eg, AE1/AE3). If the resources of the pathology lab allow, it is recommended to cut serial sections through the whole block (eg, at 100–200 µm) and to perform about additional cytokeratin immunostainings. Cytokeratin-positive cells should always be correlated with the morphology. Müllerian inclusions (endosalpingiosis, endometriosis) and mesothelial cells may rarely be present in pelvic and PALN and are cytokeratin positive. Requirements for Pathology Report Previous pertinent histological exams of the cervical lesion/cancer, even if diagnosed in another institution, should be revised and integrated in the final report (eg, cone biopsy and hysterectomy specimen) Description of the specimen(s) submitted for histological evaluation. Macroscopic description of specimen(s) (biopsy, loop/cone, trachelectomy, hysterectomy) including specimen dimensions (three dimensions), number of tissue pieces for loop/cones, and maximum and minimum length of vaginal cuff and the parametria in two dimensions. Macroscopic tumor site(s), if the tumor is grossly visible, in trachelectomy and hysterectomy specimens. Tumor dimensions should be based on a correlation of the gross and histological features and include the depth of invasion or thickness and the horizontal dimensions. Multifocal carcinomas are separated by uninvolved cervical tissue, each should be described and measured separately, and the largest used for tumor staging. In some studies, a distance of more than 2 mm was arbitrarily used to define multifocality. Multifocal carcinomas should not be confused with the scenario in which tongues or buds of invasive carcinoma originate from more than one place in a single zone of transformed epithelium Specimens from prior conization and subsequent conization, trachelectomy, or hysterectomy should be correlated for estimation of the tumor size. This is important since different specimens may have been reported at different institutions. It should also be recognized that simply adding the maximum tumor size in separate specimens may significantly overestimate the maximum tumor dimension. Histological tumor type according to the most recent WHO classification (currently 5th edition, 2020, in its updated version). Histological tumor grade if required. It needs to be stressed that currently grading remains of uncertain value for squamous cell carcinoma and most subtypes of adenocarcinoma. For adenocarcinoma, the growth pattern (Silva Classification) is recommended. The presence or absence of lymphatic vessel invasion (LVI), which may be confirmed by immunohistochemistry. The quantification of the number of lymph vascular vessels involved by tumor cells is not mandatory but advisable for future prospective studies. The presence or absence of venous invasion (V1) and of perineural invasion (Pn1). Coexisting precursor lesions such as squamous intraepithelial lesion/cervical intraepithelial neoplasia, adenocarcinoma in situ, stratified mucin-producing intraepithelial lesion and other pathological changes of the cervix. Measurements of tumor distance to all surgical margins (including minimum distance of uninvolved cervical stroma). Margin status (invasive and preinvasive diseases). Specify all the margin(s). LN status including SLN status, the total number of nodes found, the number of positive LN, the size of the largest metastatic focus, and the presence of extra-nodal extension. In the eighth UICC TNM edition isolated tumor cell deposits are no greater than 0.2 mm (200 µm) and should be reported as pN0 (i+). Micrometastasis (200 µm to 2 mm in diameter) are reported as pN1(mi). Pathologically confirmed (if required, including immunohistochemistry/HPV DNA) distant metastases. Provisional pathological staging pretumor board/multidisciplinary team meeting (UICC TNM 9th edition; American Joint Committee on Cancer, 9th edition). Items to be Included in the Pathology Report of Carcinomas of the Cervix Ancillary Studies All invasive carcinomas and adenocarcinoma in situ require an ancillary test to show the association with HPV. The most widely available and used technique is p16 immunohistochemistry (robust surrogate marker). Alternatively, HPV DNA or mRNA E6-E7 genes, can be detected by in situ hybridization or PCR-based techniques. HPV testing of cytological specimens requires liquid based cytology and uses mostly DNA-based or less frequently RNA-based molecular techniques. PD-L1 testing for the selection of immune checkpoint therapy is performed on tumor tissue, either biopsies or surgical specimens. PD-L1 expression seems to be frequently expressed in cervical carcinomas with special emphasis on locally advanced and HPV independent tumors. Standardized testing and evaluation including regular quality assessment is required to obtain a reliable patient selection for therapy. Prospective clinical trials will provide further information on the proper use of antibodies, assays and scoring systems. Further reading is available in Patient information, previous cervical cytology, histological specimens, clinical and radiological data, colposcopic findings and information on previous treatment (eg, surgery, radiotherapy) need to be included on the specimen request form. Details of cytology, biopsy, and surgical specimen (cone/loop specimen, trachelectomy, type of hysterectomy, presence of ovaries and fallopian tubes, presence of LN and designation of the LN sites, presence of vaginal cuff, and presence of parametria) need to be itemized in the specimen request form. Biopsies and surgical specimens should be sent to the pathology department in a container with liquid fixative (‘‘clamping’’ of surgical specimens on a surface may be useful). If the local situation requires biobanking of fresh tissue, surgical specimens should be submitted fresh with minimum ischemia time. Cytology specimens should be sent to the pathology department preferentially as liquid-based cytology. Smear preparations are not recommended. The former is necessary when an HPV test is requested. Immunocytochemistry is possible on LBC but of limited extent (eg, CPS score for PD-L1 cannot be assessed). Cone/loop specimen should ideally be sent intact with a suture to identify the 12-o’clock position. Biopsy/Cone/Loop Small biopsy specimens should be enumerated. The cone/loop specimens should be measured in three dimensions according to the recent ESGO/ESP recommendations. If the cone can be oriented properly, the anterior and the posterior half should be inked with separate colors. It should further be recorded if the specimen is complete or fragmented. If more than one piece of tissue is received, every piece should be measured in three dimensions. All specimens should be entirely submitted for microscopic examination. Inking of the surgical margins of cone/loop specimens is recommended. Dissection of cone/loop specimens should be performed in a standardized procedure. All the pieces submitted should be in consecutive numerical order. This is important because if tumor is present in more than one piece, it needs to be known whether these pieces are consecutive and, thus, a single tumor is present or whether the tumor is multifocal. It is recommended to place only one piece of tissue in each cassette. There are also techniques that allow embedding of more than one piece in a cassette if they are small enough. In cases that do not comprise intact cone/loops, serial radial sectioning and placing of each slice of tissue in a single cassette should be performed. Trachelectomy The upper (proximal) surgical margin of a trachelectomy specimen should be inked. The upper margin of a trachelectomy specimen should be sampled in its entirety in a way that allows to measure the distance of the tumor to the margin. The vaginal margin should also be inked and examined totally as radial sections if no tumor is seen grossly. Hysterectomy The description of the specimen (hysterectomy, trachelectomy, presence of ovaries and fallopian tubes, presence of LN and indication of the LN sites, presence of vaginal cuff and presence of parametria) should be recorded and checked for consistency with the description given in the specimen request form. The presence of any gross abnormality in any organ should be documented. The dimensions of the uterus for a hysterectomy specimen and the cervix for a trachelectomy specimen should be documented. The minimum and maximal length of the vaginal cuff should be documented. The size of the parametria should be documented in two dimensions (vertical and horizontal). Gross tumor involvement of the parametrium, vagina, uterine corpus, or other organs should be documented. The relationship of the cervical tumor to the vaginal and parametrial margins (and upper margin in case of a trachelectomy specimen) should be measured and appropriate sections taken to demonstrate this. Radial/circumferential and vaginal margins should be inked. The gross appearance of the cervix should be documented and any gross tumor mass measured. If visible, the site of a previous cone biopsy should be documented. Gross tumors should be measured in three dimensions, namely, the horizontal extent and the depth of invasion. The tumor site within the cervix should be documented. The cervical tumor should be sampled to demonstrate the maximum depth of invasion, the relationship of the tumor with the surgical borders, and the extension to other organs. When the tumor is small (or with tumors that cannot be identified macroscopically), the cervix should be separated from the corpus, opened and processed as for a cone/ loop specimen. In the case of a large tumor, the hysterectomy or trachelectomy specimen should be opened in the sagittal plane. At least one block per centimeter of the greatest tumor dimension should be taken for large tumors. Additional blocks including the cervix adjacent to the tumor should be taken to identify precursor lesions. The whole cervix should be sampled in the case of a small tumor or where no macroscopic tumor is identified. The uterine corpus, vagina, and adnexa should be sampled according to standard protocols if not involved by tumor. If the uterine corpus and/or adnexa are grossly involved, additional blocks should be sampled. The entire vaginal margin should be blocked. The parametria should be submitted totally for histological examination to assess tumor invasion and surgical margins. The use of large sections is optional and provides good information on tumor size and marginal status. Lymph Nodes All the LN should be submitted for histological examination. If the LN are grossly involved, representative samples are sufficient. If grossly uninvolved, each node should be sliced at 2 mm interval (eg, perpendicular to its longitudinal axis) and totally embedded. From each block, hematoxylin-eosin (H&E) sections should be taken. LN should be submitted in separate cassettes according to the site recorded on the specimen request form. Small biopsy specimens should be enumerated. The cone/loop specimens should be measured in three dimensions according to the recent ESGO/ESP recommendations. If the cone can be oriented properly, the anterior and the posterior half should be inked with separate colors. It should further be recorded if the specimen is complete or fragmented. If more than one piece of tissue is received, every piece should be measured in three dimensions. All specimens should be entirely submitted for microscopic examination. Inking of the surgical margins of cone/loop specimens is recommended. Dissection of cone/loop specimens should be performed in a standardized procedure. All the pieces submitted should be in consecutive numerical order. This is important because if tumor is present in more than one piece, it needs to be known whether these pieces are consecutive and, thus, a single tumor is present or whether the tumor is multifocal. It is recommended to place only one piece of tissue in each cassette. There are also techniques that allow embedding of more than one piece in a cassette if they are small enough. In cases that do not comprise intact cone/loops, serial radial sectioning and placing of each slice of tissue in a single cassette should be performed. The upper (proximal) surgical margin of a trachelectomy specimen should be inked. The upper margin of a trachelectomy specimen should be sampled in its entirety in a way that allows to measure the distance of the tumor to the margin. The vaginal margin should also be inked and examined totally as radial sections if no tumor is seen grossly. The description of the specimen (hysterectomy, trachelectomy, presence of ovaries and fallopian tubes, presence of LN and indication of the LN sites, presence of vaginal cuff and presence of parametria) should be recorded and checked for consistency with the description given in the specimen request form. The presence of any gross abnormality in any organ should be documented. The dimensions of the uterus for a hysterectomy specimen and the cervix for a trachelectomy specimen should be documented. The minimum and maximal length of the vaginal cuff should be documented. The size of the parametria should be documented in two dimensions (vertical and horizontal). Gross tumor involvement of the parametrium, vagina, uterine corpus, or other organs should be documented. The relationship of the cervical tumor to the vaginal and parametrial margins (and upper margin in case of a trachelectomy specimen) should be measured and appropriate sections taken to demonstrate this. Radial/circumferential and vaginal margins should be inked. The gross appearance of the cervix should be documented and any gross tumor mass measured. If visible, the site of a previous cone biopsy should be documented. Gross tumors should be measured in three dimensions, namely, the horizontal extent and the depth of invasion. The tumor site within the cervix should be documented. The cervical tumor should be sampled to demonstrate the maximum depth of invasion, the relationship of the tumor with the surgical borders, and the extension to other organs. When the tumor is small (or with tumors that cannot be identified macroscopically), the cervix should be separated from the corpus, opened and processed as for a cone/ loop specimen. In the case of a large tumor, the hysterectomy or trachelectomy specimen should be opened in the sagittal plane. At least one block per centimeter of the greatest tumor dimension should be taken for large tumors. Additional blocks including the cervix adjacent to the tumor should be taken to identify precursor lesions. The whole cervix should be sampled in the case of a small tumor or where no macroscopic tumor is identified. The uterine corpus, vagina, and adnexa should be sampled according to standard protocols if not involved by tumor. If the uterine corpus and/or adnexa are grossly involved, additional blocks should be sampled. The entire vaginal margin should be blocked. The parametria should be submitted totally for histological examination to assess tumor invasion and surgical margins. The use of large sections is optional and provides good information on tumor size and marginal status. All the LN should be submitted for histological examination. If the LN are grossly involved, representative samples are sufficient. If grossly uninvolved, each node should be sliced at 2 mm interval (eg, perpendicular to its longitudinal axis) and totally embedded. From each block, hematoxylin-eosin (H&E) sections should be taken. LN should be submitted in separate cassettes according to the site recorded on the specimen request form. Intraoperative assessment of sentinel nodes is a reliable procedure but may miss micrometastases and isolated tumor cells. Intraoperative assessment should be performed on a grossly suspicious sentinel node and may be performed on a “non-suspicious” SLN because the confirmation of tumor involvement will result in abandoning a hysterectomy or trachelectomy. For intraoperative evaluation, the SLN should be sent to the pathology department in a container without liquid fixative. Intraoperative analysis requires gross dissection of the resected adipose tissue by the pathologist and selection of LN. It is important to leave some peri-nodal tissue allowing proper diagnosis of extra-nodal tumor spread. For a LN with obvious gross tumor, a single section is adequate for frozen section. Frozen section may be combined with imprint cytology. The use of one step nucleic acid amplification is not recommended particularly due to the interference with benign epithelial inclusions in PLN. Any nonsuspicious sentinel node should be bisected (if small) or sliced at (approximately) 2 mm thickness and entirely frozen. From each sample, histological sections should be cut and stained by H&E. After frozen section analysis, the tissue should be put into a cassette, fixed in liquid fixative (preferably 4% buffered formalin) and subsequently processed and embedded in paraffin. If no metastases are present in the first section, SLN should undergo ultrastaging in definitive paraffin sections, including immunohistochemistry. A minimum procedure should include five serial sections at 200 µm. At least, at two levels an additional section must be cut and stained with a broad-spectrum cytokeratin antibody (eg, AE1/AE3). If the resources of the pathology lab allow, it is recommended to cut serial sections through the whole block (eg, at 100–200 µm) and to perform about additional cytokeratin immunostainings. Cytokeratin-positive cells should always be correlated with the morphology. Müllerian inclusions (endosalpingiosis, endometriosis) and mesothelial cells may rarely be present in pelvic and PALN and are cytokeratin positive. Previous pertinent histological exams of the cervical lesion/cancer, even if diagnosed in another institution, should be revised and integrated in the final report (eg, cone biopsy and hysterectomy specimen) Description of the specimen(s) submitted for histological evaluation. Macroscopic description of specimen(s) (biopsy, loop/cone, trachelectomy, hysterectomy) including specimen dimensions (three dimensions), number of tissue pieces for loop/cones, and maximum and minimum length of vaginal cuff and the parametria in two dimensions. Macroscopic tumor site(s), if the tumor is grossly visible, in trachelectomy and hysterectomy specimens. Tumor dimensions should be based on a correlation of the gross and histological features and include the depth of invasion or thickness and the horizontal dimensions. Multifocal carcinomas are separated by uninvolved cervical tissue, each should be described and measured separately, and the largest used for tumor staging. In some studies, a distance of more than 2 mm was arbitrarily used to define multifocality. Multifocal carcinomas should not be confused with the scenario in which tongues or buds of invasive carcinoma originate from more than one place in a single zone of transformed epithelium Specimens from prior conization and subsequent conization, trachelectomy, or hysterectomy should be correlated for estimation of the tumor size. This is important since different specimens may have been reported at different institutions. It should also be recognized that simply adding the maximum tumor size in separate specimens may significantly overestimate the maximum tumor dimension. Histological tumor type according to the most recent WHO classification (currently 5th edition, 2020, in its updated version). Histological tumor grade if required. It needs to be stressed that currently grading remains of uncertain value for squamous cell carcinoma and most subtypes of adenocarcinoma. For adenocarcinoma, the growth pattern (Silva Classification) is recommended. The presence or absence of lymphatic vessel invasion (LVI), which may be confirmed by immunohistochemistry. The quantification of the number of lymph vascular vessels involved by tumor cells is not mandatory but advisable for future prospective studies. The presence or absence of venous invasion (V1) and of perineural invasion (Pn1). Coexisting precursor lesions such as squamous intraepithelial lesion/cervical intraepithelial neoplasia, adenocarcinoma in situ, stratified mucin-producing intraepithelial lesion and other pathological changes of the cervix. Measurements of tumor distance to all surgical margins (including minimum distance of uninvolved cervical stroma). Margin status (invasive and preinvasive diseases). Specify all the margin(s). LN status including SLN status, the total number of nodes found, the number of positive LN, the size of the largest metastatic focus, and the presence of extra-nodal extension. In the eighth UICC TNM edition isolated tumor cell deposits are no greater than 0.2 mm (200 µm) and should be reported as pN0 (i+). Micrometastasis (200 µm to 2 mm in diameter) are reported as pN1(mi). Pathologically confirmed (if required, including immunohistochemistry/HPV DNA) distant metastases. Provisional pathological staging pretumor board/multidisciplinary team meeting (UICC TNM 9th edition; American Joint Committee on Cancer, 9th edition). All invasive carcinomas and adenocarcinoma in situ require an ancillary test to show the association with HPV. The most widely available and used technique is p16 immunohistochemistry (robust surrogate marker). Alternatively, HPV DNA or mRNA E6-E7 genes, can be detected by in situ hybridization or PCR-based techniques. HPV testing of cytological specimens requires liquid based cytology and uses mostly DNA-based or less frequently RNA-based molecular techniques. PD-L1 testing for the selection of immune checkpoint therapy is performed on tumor tissue, either biopsies or surgical specimens. PD-L1 expression seems to be frequently expressed in cervical carcinomas with special emphasis on locally advanced and HPV independent tumors. Standardized testing and evaluation including regular quality assessment is required to obtain a reliable patient selection for therapy. Prospective clinical trials will provide further information on the proper use of antibodies, assays and scoring systems. Further reading is available in
Integrating Primary Care into the Management of Cystic Fibrosis	d166d5b3-ee82-4eb4-ac1e-0f0ecaf49a91	10176593	Patient-Centered Care[mh]	Over the last 50 years, cystic fibrosis (CF) has radically transformed from a fatal disease of infancy to a chronic disease of adulthood, primarily through advancement in targeted therapeutics. Individuals with CF (iwCF) in developed countries can expect to live well into their fifth decade. Many pursue higher education and various professions and also raise families. By 2025 it is estimated that 70% of the roughly 40 000 iwCF in the United States will be cared for by adult providers. , As clinicians with backgrounds in primary care and pulmonary medicine who care for iwCF, we anticipate that there will be an enhanced need for primary care physician (PCP) engagement, particularly regarding the current and evolving preventative care measures needed for this unique population. This commentary will provide relevant background on diagnosing and treating cystic fibrosis, current preventative screening measures, and potential models for delivering primary care to this group. CF is an autosomal recessive disease caused by mutations in the cystic fibrosis transmembrane receptor (CFTR) gene. To date, nearly 2000 mutations of the CFTR gene have been identified, the most common of which is F508del (p.Phe508del), which is present in at least one allele in roughly 85% of iwCF. mutations impact the encoded CFTR protein in various ways and are classified into 6 categories of mutations characterized by absent, reduced, or dysfunctional amounts of an epithelial transmembrane protein responsible for chloride and bicarbonate exchange. CF is fundamentally a disease of impaired mucus clearance. In childhood, the most notable symptoms are recurrent sinus and pulmonary infections and impaired exocrine pancreatic function from blocked pancreatic ducts, leading to malnutrition and steatosis. During adolescence and adulthood, a progressive decline in lung function often occurs due to recurrent pulmonary infections. In addition, individuals can develop endocrine pancreatic dysfunction leading to the destruction of islet cells, causing impaired insulin secretion and insulin resistance, referred to as CF-related diabetes. Primary care providers already play a critical role in diagnosing CF, which can occur at any stage of life. Most iwCF are diagnosed through newborn screening (NBS), performed routinely in the United States since 2010. If the newborn screen is positive, the infant should be referred to a CF specialty clinic to undergo sweat chloride skin testing. A sweat chloride value >60 mmol/L is diagnostic and considered the gold standard for diagnosis at all ages. This multistep screening algorithm has a sensitivity approaching 95%. A recent study found that roughly 65% of Canadian PCPs have reported notifying a family of a positive NBS and report they place significant importance on their involvement in disclosing these results but have moderately low confidence in doing so. This highlights the need for more effective education strategies and support structures for primary providers surrounding CF. The treatment of CF has dramatically changed over the last 2 decades and is the primary driver for improved survival. The main pillars of treatment currently focus on (1) thinning of mucus in the respiratory tract (with the use of nebulized hypertonic saline and dornase alpha), (2) clearance of mucus from the airway (via chest physiotherapy using vibratory and oscillatory devices, manual maneuvers, huff cough, and other techniques), (3) aggressive treatment of pulmonary infections especially from pseudomonas and staphylococcus aureus (including suppressive treatment with nebulized or inhaled antibiotics for pseudomonas), (4) restoration of CFTR protein function through oral therapies, commonly referred to as modulators, (5) the restoration of impaired exocrine pancreatic function through the use of supplemental pancreatic enzymes (pancrelipase or CREON), and (6) high-fat and high-calorie diets to prevent malnutrition. In 2019, the American Food and Drug Administration approved a combination medication containing 3 modulators: elexacaftor, ivacaftor, and tezacaftor (Trikafta). This medication is approved for nearly 90% of CF mutations and significantly improves CFTR function in these groups. So far, it has been shown to improve lung function, reduce the number of pulmonary exacerbations, increase body mass index, and improve quality of life. Primary care providers have reported low comfort with caring for iwCF. In the pediatric literature, there is evidence that in more rural areas or regions without a multidisciplinary CF clinic, PCPs are more likely to refer patients to pulmonologists, even for non-CF-related care. This has been shown to overburden the subspecialty clinics and make it difficult for patients to access their physicians for acute care needs. A similar phenomenon is observed among adults needing care with other chronic medical conditions, particularly those with lower incomes or limited access to healthcare. In a 2008 study of PCPs, half the internists surveyed felt primary care for individuals with child-onset chronic illness should be delegated to an adult subspecialist. This may explain, in part, our anecdotal experience of patients preferentially seeking out non-CF-related medical care needs like cancer screening and non-CF-related disease management within our CF specialty clinic, with some iwCF lacking a PCP altogether. We believe the role of a dedicated PCP will be crucial for the continued progress in health and quality-of-life outcomes, particularly related to preventative care. Preventive maintenance by a PCP spans many domains, including but not limited to vaccinations, screenings for cancer, osteoporosis, sexually transmitted infections, depression, diabetes, atherosclerotic and cardiovascular disease, Hepatitis C, and Human Immunodeficiency Virus. The CF Foundation currently requires annual or quarterly screening from accredited CF clinics for certain conditions like CF-related diabetes, osteoporosis, liver disease, mycobacterial disease, and allergic bronchopulmonary aspergillosis (ABPA); however, there are no requirements for most other age-specific preventive healthcare screening measures, which will become ever more pertinent to the longevity of iwCF. A PCP-led preventive care screening strategy can enhance compliance with guidelines while minimizing replication of effort, inconvenience, and cost. Here we outline standard screening and preventative care considerations in CF that would be ideal for shifting to the domain of a PCP. The PCP must be familiar with the need for annual diabetes screening done for non-diabetic iwCF with glucose tolerance tests rather than fasting glucose or hemoglobin A1C. Additionally, individuals with CFRD should undergo yearly screening for microvascular complications, as is done in the general population, through dilated retinal exams and urine albumin-to-creatinine ratios. There is a 5 to 10-fold increased risk of colon cancer among iwCF compared to the general population. Colon cancer screening for iwCF begins at age 40 with a colonoscopy. It is repeated every 5 years at a minimum. Those who have undergone solid organ transplantation may start screening earlier at the age of 30. Osteoporosis risk is significantly higher in iwCF. Screening for osteoporosis should be undertaken using dual-energy x-ray absorptiometry (DEXA) once by age 18. Screening should continue every 1 to 5 years based on T and Z scores. Interestingly, screening rates for osteoporosis fall short even among CF programs. Thus, it is crucial to have a comprehensive approach that is PCP driven to shore up gaps in routine care. Newer concerns unique to older iwCF that are not routinely within the scope of a CF clinic include age-related preventative care, such as screening for elevated cholesterol, abdominal aortic aneurysm, gynecologic and breast cancers, and prostate cancer, among many others. A yearly visit with a PCP also provides the opportunity for counseling on various elements that may not be factored in at a subspecialty visit, such as in-depth family planning, particularly with the increased number of pregnancies in women with the use of triple-drug modulators. Comfort with caring for iwCF has been directly linked to the volume of patients cared for, outweighing other factors like access to ancillary services and subspecialists. We believe that PCPs should be embedded into CF care teams. Several models for incorporating PCPs into the routine care of iwCF are already in use, but no universally accepted standard exists. Ideally, the PCP and pulmonologist practice together in a patient-centered medical home (PCMH) for both pediatric and adult patients. This model typically relies upon a general practitioner (pediatrician, internist, family medicine, or advanced practice provider) and pulmonologist leading multidisciplinary teams with appropriate subspecialists, providing both primary care and specialty services to patients. This model has yet to be explicitly studied in CF-related outcomes; however, some North American CF centers utilize a similar structure. There is growing evidence that these care models are associated with improved quality-of-life measures, mental health, and disease-specific outcomes like glycemic control in diabetes, cholesterol reduction in cardiovascular disease, and reduced hospitalizations in other chronic disease states. The PCMH model is not feasible for every center or patient, especially those in more rural settings. Telemedicine can address this gap by facilitating interactions between patients and providers, PCPs, and specialty clinics. Within CF, telehealth visits have been shown to enhance access to care, adherence to therapies, detection of pulmonary exacerbations, and mental health. Telemedicine may be well suited for PCPs in CF care and enhance lines of communication between practices. One example of implementing this is creating an annual visit that includes all patient care team members, including the PCP, in a virtual format. This would allow the patient and care team members to engage regardless of their distances and could be blended into the workflow of the PCP’s clinic schedule that day. This has not been explicitly studied in CF, but examples of similar and effective use of this technology to engage PCPs with specialty teams have demonstrated improved patient-centered outcomes, cost-effectiveness, and provider satisfaction in other chronic disease processes. The community of CF patients, providers, and researchers have pioneered novel management strategies that have prolonged the lives of many individuals with this disease. The models used in CF care have also had ripple effects in promoting improvements in the management strategies for other chronic childhood diseases. In our practice, we have seen growth in the burden of non-CF-related conditions like hypertension and obesity and an increased need for routine preventative screening as patients live longer. Anecdotally we find that many of our patients preferentially seek care at our specialty clinic, and some do not regularly engage with a PCP. As we enter a new era of CF management, the role of primary care medicine will become increasingly relied upon to provide multimodal care to the aging population of iwCF. To meet this moment, PCPs will need tools and practical experiences in managing this rare condition. We advocate for including and recruiting PCPs into subspecialty clinics and through engagement with community providers through readily available didactics, seminars, and telemedicine opportunities. As PCPs and CF clinicians, we feel that sharing the domain of preventative care with the expertise of a PCP will provide improved health and well-being for older iwCF. To demonstrate that these efforts are practical and effective, we must closely monitor outcomes data and patient and provider perspectives related to real-world experiences and ongoing needs assessments.
Comment on “Mutagenic damage among bronchiectasis patients attending in the pulmonology sector of a hospital in southern Brazil”	97e9dc5e-6813-4a57-a281-9b9aa5ee9ee3	10176658	Internal Medicine[mh]	The data that support the findings of this study are available from the corresponding author upon reasonable request.
European inequalities and similarities in officially recognized dental specialties	7c3608a5-dcd8-45db-b26f-1422903a1f8a	10176728	Dental[mh]	The first dental specialty, Orthodontics, was recognized in 1900 in the United States and it has been 44 years since dental specialties were established in Europe. Within the European Economic Space (EES) 25 out of the 27 countries, of which it is composed, have officially recognized dental specialties. Directives 2005/36/CE and 2013/55/EU developed the contents of three previous directives from 1978 (78/686/CEE, 78/687/CEE and 78/688/CEE) on the regulation of dental specialties. These directives were the legal support for UE professional mutual recognition and freedom of movement around the member states. Directive 2005/36/CE explicitly recognizes in annex 5.3.3. the first two professional dental specializations: Orthodontics and Oral Surgery. In order to take due account of changes in national legislation, and with a view to updating this directive, directive 2013/55/EU adds to article 35 that the Commission shall be empowered to include new dental specialties common to at least two-fifths of the Member States. Throughout the world, regulatory bodies, set up by their country’s governments, provide specialty lists of those clinicians who have undergone a recognized postgraduate training leading to a higher specialist qualification . Nowadays, the dental specialties situation remains in constant change all around Europe. Almost every country of which the EES is comprised has its own organised health system that provides conditions for the specialties regulation. While most of the member states can rely on a law regulating the dental specialties system, two countries are still omitted from these criteria: Spain and Austria, being the only ones who do not yet have this official recognition. In the case of Austria, this country is currently developing the governmental law needed to start this procedure, once they have got the parliamentary approval. Spain is currently the last country in this situation. Furthermore, the unification of names, types and number of dental specialties in the rest of these countries is non-existent, leading to a confusing situation to identify which branches of Dentistry are recognized in which country, hindering the opportunity to move to another one and keep the same job specialization . Apart from this, it seems complicated to have a professional regulated system that permits us to orientate the number of dental specialists and balance each work situation in every country, because it looks inappropriate to have a lack of or a surplus of professionals in each sector of Dentistry. Moreover, these specialties help to provide a focused treatment for suitable patients, improving both the training of dentists and the quality in clinical care received by the population, building better relationships, and thus protecting the public and the specialties themselves from unqualified clinicians. This and other observations, lead us to appreciate objective differences concerning their structure, showing health, social, economic and opportunity inequalities due to the quality of the service provided in each country, which could be partly influenced by the official recognition of dental specialties and the existence of official postgraduate studies [ – ]. Relevant information to make this comparison can be found in official documents in the national societies or associations of dentists of most of the countries. However, comparisons, even in an easy and visual way, i.e., by tables and figures, reaching the vast majority of the EES has not been reported before. Previous revised studies about this topic have analysed no more than one third of the European countries . Both, from a clinical and academic perspective, including students’ perception of dental specialties , we considered it of great interest to evidence the objective differences that make up each unique country, looking at the number and types of dental specialties. Therefore, the aim of the present study was to analyse the inequalities and similarities existing between 20 out of the 30 countries of which the EES is comprised plus the UK, highlighting the number and types of officially recognized dental specialties. Available official documents and webpages, mainly linked to regulatory bodies, official college and councils and dental institutions from 20 out of the 30 countries of which the EES plus the UK, is comprised, were collected and analysed to obtain reliable data referred to dental specialties. Additional documents like annual reports were considered to extend and corroborate the public information provided (Table ) [ , – ]. Differences in the distribution of officially recognized dental specialties in Europe were tested with the Lorentz curve and Gini test. Additionally, a Cluster analysis was performed to obtain groups of countries with a similar pattern in the number and type of dental specialties . To quantify the degree of global inequality related to the number of dental specialties of the EES analysed countries the Lorentz curve and Gini index were applied. These are two indicators commonly used to measure the unequal distribution of wealth in a society. In our context, the concept “wealth” is associated to dental wealth (number of specialties in a country), the EES acts as the society under study and each country represents an element of that society. When the relationship between the cumulative percentage of countries that make up the EES and the cumulative percentage of specialties is displayed graphically, the Lorentz curve is obtained. This curve has a very useful interpretation in order to assess inequality between countries: The closer it is to the diagonal, the more homogeneous is the number of specialties in the countries; on the contrary, if it approaches the horizontal axis, more inequality exists in the number of specialties in the different countries. One way to quantify this property is simply to measure the area between both curves and scale it, so that the result remains between 0 and 1. The result of doing this operation is known as the Gini index. The disadvantage of the Gini index is that it only captures a part of the reality, being a simplification of it. And as such, the simplification does not explore in depth the details of which countries have greater or lesser similarity depending on the number of common specialties they share. For this reason, in order to make an assessment of the inequality between EES countries and to establish groups of inequality and similarity, we extended the study of the dental reality with a Cluster analysis 32 , which is specialized in exploring this aspect. Currently we need to define the “distance between two countries” in a way that is appropriate to the available data: Each country is represented as a binary vector with 1’s where there is a certain specialty and 0’s where it does not exist. A binary distance between two countries is then defined as the proportion of 1’s and 0’s shared by the binary vectors that represent it. Based on this definition, the distances between each pair of EES countries can be calculated through Cluster analysis so that groups and associations appear according to their similarities. All the resulting information is represented by a specialized graph called a dendrogram, which is very useful for detecting associations. Figure and Table show the type, number and percentage of officially recognized dental specialties in every analysed country. Orthodontics and Oral Surgery are the most frequently recognized dental specialties of the analysed countries (90% and 81% respectively), followed by Periodontics (43%), Paediatric Dentistry (33%), Prosthodontics (29%), and Endodontics (24%). The number of recognized dental specialties was quite different as well. The most specialized country regarding the number of specialties is the UK (13 dental specialties), followed by Sweden (8), Lithuania, Norway, Poland (7), and Romania (6). Mostly of the analyzed countries recognized only 2 dental specialties. On the other hand, Austria and Spain have a total lack of officially recognized dental specialties. The 2 most frequently recognized specialties in these countries are Orthodontics and Oral Surgery, with the exception of Belgium and Netherlands. Figure shows the Lorentz curve (in red) associated with the data of our study. The grey area represents dental inequality and its assessment using the Gini index was G = 0.402. So, the total global degree of inequality related to the number of dental specialties between the analysed countries, was 40.2%. Cluster analysis differentiated three main groups of countries according to the number and type of dental specialties (Fig. ). The associated dendrogram is displayed as a forked-line diagram where each fork, groups countries at its lower level. The height of each fork indicates the maximum distance between the countries that form it. In this way, interpreting our data, if we cut, for example, at distance 2 (the vertical axis) we would obtain three groups of homogeneous countries in terms of the number and type of specialties: Two large clusters of similar countries, the first one formed by Luxembourg, Ireland, Greece, Germany, Cyprus, Denmark, Czech Republic and France and the second one by Romania, Switzerland, the UK, Poland, Norway, Sweden, Lithuania, Italy, Portugal, Belgium and The Netherlands, and finally, a small group formed by only Austria and Spain. The lowest ranges, at zero distance, in the case of Luxembourg, Ireland, Greece, Germany, Cyprus and Denmark or Austria and Spain define those countries between which there are no differences at all. The type of web-pages and documents that every state in the EES provides differ substantially depending on the way they organize their own health system, the type of annual reports that are written, the distribution of their documents or the visual aspect added to every public information provided. Nevertheless, whatever method is used, the information is accessible and clear inside the respective document or the Internet pages. A total of 21 countries having available and reliable data were included in our study [ – ]. Although the EES usually acts in a united and organized way, several main differences are found regarding the type and number of dental specialties all around this group of member states. It is important to highlight the high prevalence of two common-historical specialties, the first in being accepted and generally recognized, Orthodontics and Oral Surgery . So, nowadays they appear in every country except in Spain and Austria (without any officially recognized dental specialty) and Belgium (having recognized Orthodontics and Periodontology, but not Oral Surgery). The rest of the specialties are recognized in less than half of the countries studied in the EES. A small group of official dental recognitions are less common, finding some of them like Oral Microbiology, Oral Pathology, Public Health or Special Care Dentistry that are only regulated as specialties in the United Kingdom , the country having the highest number. The rest of specialties such as Oral Medicine, Radiology and Restorative Dentistry are locally extended, existing exceptions like Stomathognathic Physiology, that is exclusively found in Sweden ; or Maxillofacial Surgery that is only officially recognized as a dental specialty in the Netherlands and in Lithuania (in this case, it accompanies Oral Surgery as two different dental specialties). According to the number of officially recognized dental specialties quite a diverse variety was found. The number of dental specialties is really unbalanced, oscillating between 0 and 13 different specialties. Moreover, in some cases even the name and guidelines that every recognition serves is slightly different, which adds even more inequality and confusion between countries with the same type of officially recognized dental specialties . To quantify the degree of global inequality related to the number of dental specialties of the EES analysed countries plus the UK, the Lorentz curve and Gini index were applied. Globally, in the countries of the EES there is an inequality with respect to the number of specialties of 40.2%. Luxembourg, Ireland, Greece, Germany, Cyprus, and Denmark have the same situation, having recognized only the two most frequent dental specialties. Austria and Spain are in a similar situation to each other, and they are an exception inside the EES with a lack of legal regulation that prevent them having officially recognized dental specialties. Spain, curiosly, has a lot of health specialties, but not in Dentistry (Medicine, Pharmacy, Nursing, Biology, Psychology or, surprisingly, Chemistry or Physics). Influencial factors could have had an impact in Spain, as frequent governmental changes (since 1978 there have been 15 different legislatures and 26 different Health ministers) that have been delaying the process; the more prevalent number of private dental practices (who do not require so much legal regulation, as is the case in other countries in the EES); a certain historical distance from dentists to dental specialties, or an unsuccessful professional management, could have tipped the scale against national recognition until now. Additionally, in both countries, dental specialties, particularly Oral Surgery, have been fought by physicians having the specialty of Maxillofacial Surgery. The UK is the country having the highest number of dental specialties (13) and another 5 countries have more than 6 official specialties (Sweden, Lithuania, Norway, Poland and Romania), but the most of the EES countries count on just two specializations. The recognition of dental specialties of the countries integrating the EES plus the UK has meant a benchmark in terms of increasing professional development. Apart from the problems associated with inequalities regarding the number or name of dental specialties, we must consider other quality defects caused by this scenario: Lack of access to officially recognized dental specialists, lack of specialized training, many restrictions to work abroad keeping the same job specialization and guarantees, or the difficulty to control the name of specialists countrywide. These are only some deficiencies that still need to be solved to recuperate the equalities and union that the EES should provide to every member state. Our findings show the relevant dental specialties asymmetric situation existing in Europe. The results could be useful for both, dental professionals (i.e. in terms of conditions and freedom to move and work between countries across European countries, or protecting the public and the specialties themselves from unqualified clinicians) and the general population (i.e. regulating the quality in clinical care received by the population). The situation of dental specialties in the area of the EES plus the UK exhibits an unequal organization. Despite having some common directives that every country integrating the EES should follow, we can appreciate that actually it is not well implemented. Thus, the number and the name of officially recognized dental specialties across the member states is quite different and the total global degree of inequality related to the number of dental specialties of the analysed EES countries plus the UK is 40.2%, showing three different main clusters of countries with a similar pattern of dental specialties.
Effect of pharmacogenomics testing guiding on clinical outcomes in major depressive disorder: a systematic review and meta-analysis of RCT	a72feff6-6f88-480c-a955-05f850a8617e	10176803	Pharmacology[mh]	Major depressive disorder (MDD) is one of the most prevalent psychiatric diseases, affecting more than 300 million people globally . MDD leads to cognitive impairment, which results in suicidal behaviors . More than 800,000 people die from suicide each year . For decades, medications for MDD followed clinical guidelines of MDD treatment, while numerous anti-depressant drugs went on the market and were considered to be “first-line” for comparable remission and/or response rates . However, following the present anti-depressant treatment pathway, only 50% of patients achieve clinically significant responses within their first anti-depressant treatment; the number of patients achieving remission is even fewer . Thus, the current standard treatment pathway is a trial-and-error approach until a relatively effective anti-depressant or combination treatment is found that provides full remission effects . Current treatment approaches result in a prolonged duration of unremitted illness, which is related to worsened long-term prognosis, adverse changes in cognitive function and brain morphology, and increased side effects . Because of those problems with current MDD medications, better tools are urgently needed to help physicians and patients find effective treatment. Pharmacogenomic testing offers a point-of-care, personalized, and ascendable tactic to guide clinical treatment in multiple diseases pre-emptively, including MDD . A number of proprietary pharmacogenomic testing tools have been developed to guide drug selection or conversion by testing allelic gene variants of assigned genes mediating the pharmacodynamics and/or pharmacokinetics of anti-depressant medication . Numerous randomized controlled trials (RCTs) have been conducted to evaluate the efficacy and safety outcomes of pharmacogenomic testing in MDD, but the results and conclusions are still conflicting. Several large-scale RCTs depicted significant improvement in response and remission rates, while a recent study by Oslin et al. showed no significant change between the pharmacogenetic testing guided treatment group and the usual treated group . The key point of whether patients benefit from pharmacogenetic testing remains unclear. Several previous systematic review and meta-analyses had been conducted based on this issue . However, since the number of studies involved was limited, no sufficient data was provided to reach a reliable conclusion. On the other hand, cohort studies and RCTs were both included in the meta-analyses which could contribute to potential bias. Furthermore, plenty of RCTs regarding this topic have been published in recent two years , reflecting the necessity of an updated systematic review and meta-analysis. Here in our current study, we determined the effect of pharmacogenetic testing guided on clinical outcomes in MDD patients. Protocol and registration This is a protocol-driven systematic review and meta-analysis, prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO CRD42022360151). We performed this meta-analysis in accordance with the guideline of Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guideline , and the checklist is available in Supplemental Material (Method S1). Literature research We searched PubMed, Embase, and Cochrane Library of Clinical Trials from inception until August 2022. The key terms were ‘pharmacogenomic’ and ‘antidepressive’. The detail of the search terms is attached in Table . The reference lists of all included studies were also examined for relevant citations. No language restrictions were applied. Study selection Studies were considered for inclusion based on the following criteria: participants were diagnosed with major depressive disorder; the study included both pharmacogenomic testing guided group and usual care group; randomized controlled trial (RCT). The screening and scanning for eligibility were performed manually by two independent reviewers (X. W. and C. W.) through Endnote (version 9.3.2). Any disagreements were resolved by discussion with a third reviewer (Z. A.). Quality assessment Six domains (random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective reporting) were assessed by using the Cochrane Collaboration’s tool for assessing the risk of bias by two reviewers (X. W. and C. W.) independently. Disagreements on study quality assessment were discussed with another reviewer (Y. Z.) until a consensus was reached. Data extraction Data were independently extracted by two investigators (X. W. and C. W.) for eligible studies. The characteristic data obtained for each study included the first author, year of publication, study design, inclusion criteria, target genes, baseline characteristics, and industry funding. Outcomes were response (≥ 50% decrease in HAM-D17/PHQ-9 score from baseline) rate and remission (a score of ≤ 7 for HAM-D17 or ≤ 5 for PHQ-9 or ≤ 2 for CGI score) rate at week 4, 8, 12, and 24, respectively. Another outcome was medication congruence in 30 days (participants that prescribed antidepressant medication that was categorized as having no drug-gene interaction or moderate drug-gene interaction). We gave preference to data from intent-to-treat analysis over pre-protocol analysis. Data with longer follow-up duration were chosen over the shorter one when data at several time points were provided within the period. Statistical analysis Odds ratios (RR) with 95% confidence intervals (95%CIs) were calculated with R (version 4.0.5). Heterogeneity across the studies was assessed using the Cochran’s Q test; the percentage of total variability attributable to heterogeneity was quantified by the I 2 value. I 2 less than 50% indicates low or moderate heterogeneity and more than 50% indicates high heterogeneity . Random-effects model with inverse variance weighting was used for high heterogeneity, and the Mantel-Haenszel fixed-effects model was used for low or moderate heterogeneity of included studies. P values less than 0.05 were considered significant. Funnel plots were used to test the publication bias. Subgroup analysis was done according to follow-up duration for less than 12 weeks (data with the longest follow-up duration through the period was extracted). Only a few studies included provided data with longer duration, so subgroup analysis for outcomes at week 24 was not conducted. Four further subgroup analyses were carried out according to study design (single-center or multi-center), sample size (less than 200 patients or more than 200 patients), population (the majority ethnicity of the population: Caucasian or Asian), diagnostic criteria (HAM-D17 or PHQ-9), and industry funding (fully funded, partially funded, or none funded). A difference between the estimates of these subgroups was considered significant for the P-value between subgroups < 0.10 . To evaluate the stability of the results, we did a sensitivity analysis by sequentially removing every single study from the pooled effect estimates. This is a protocol-driven systematic review and meta-analysis, prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO CRD42022360151). We performed this meta-analysis in accordance with the guideline of Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guideline , and the checklist is available in Supplemental Material (Method S1). We searched PubMed, Embase, and Cochrane Library of Clinical Trials from inception until August 2022. The key terms were ‘pharmacogenomic’ and ‘antidepressive’. The detail of the search terms is attached in Table . The reference lists of all included studies were also examined for relevant citations. No language restrictions were applied. Studies were considered for inclusion based on the following criteria: participants were diagnosed with major depressive disorder; the study included both pharmacogenomic testing guided group and usual care group; randomized controlled trial (RCT). The screening and scanning for eligibility were performed manually by two independent reviewers (X. W. and C. W.) through Endnote (version 9.3.2). Any disagreements were resolved by discussion with a third reviewer (Z. A.). Six domains (random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective reporting) were assessed by using the Cochrane Collaboration’s tool for assessing the risk of bias by two reviewers (X. W. and C. W.) independently. Disagreements on study quality assessment were discussed with another reviewer (Y. Z.) until a consensus was reached. Data were independently extracted by two investigators (X. W. and C. W.) for eligible studies. The characteristic data obtained for each study included the first author, year of publication, study design, inclusion criteria, target genes, baseline characteristics, and industry funding. Outcomes were response (≥ 50% decrease in HAM-D17/PHQ-9 score from baseline) rate and remission (a score of ≤ 7 for HAM-D17 or ≤ 5 for PHQ-9 or ≤ 2 for CGI score) rate at week 4, 8, 12, and 24, respectively. Another outcome was medication congruence in 30 days (participants that prescribed antidepressant medication that was categorized as having no drug-gene interaction or moderate drug-gene interaction). We gave preference to data from intent-to-treat analysis over pre-protocol analysis. Data with longer follow-up duration were chosen over the shorter one when data at several time points were provided within the period. Odds ratios (RR) with 95% confidence intervals (95%CIs) were calculated with R (version 4.0.5). Heterogeneity across the studies was assessed using the Cochran’s Q test; the percentage of total variability attributable to heterogeneity was quantified by the I 2 value. I 2 less than 50% indicates low or moderate heterogeneity and more than 50% indicates high heterogeneity . Random-effects model with inverse variance weighting was used for high heterogeneity, and the Mantel-Haenszel fixed-effects model was used for low or moderate heterogeneity of included studies. P values less than 0.05 were considered significant. Funnel plots were used to test the publication bias. Subgroup analysis was done according to follow-up duration for less than 12 weeks (data with the longest follow-up duration through the period was extracted). Only a few studies included provided data with longer duration, so subgroup analysis for outcomes at week 24 was not conducted. Four further subgroup analyses were carried out according to study design (single-center or multi-center), sample size (less than 200 patients or more than 200 patients), population (the majority ethnicity of the population: Caucasian or Asian), diagnostic criteria (HAM-D17 or PHQ-9), and industry funding (fully funded, partially funded, or none funded). A difference between the estimates of these subgroups was considered significant for the P-value between subgroups < 0.10 . To evaluate the stability of the results, we did a sensitivity analysis by sequentially removing every single study from the pooled effect estimates. 1779 potentially relevant articles were screened. 1712 irrelevant articles were excluded based on screening of the titles and abstracts. 54 articles went through full-text review and 43 of them were excluded because they did not meet the inclusion criteria. Lastly, eleven RCT studies with a total of 5347 patients were included in the meta-analysis (Figure ). The duration of follow-up was 8 weeks in four studies, 10 weeks in one study, 12 weeks in three studies, 24 weeks in two studies, and 52 weeks in one study. The specific genes tested differed between studies (Table ). Assessment of risk The overall result of the risk assessment is summarized in Figure S2. Most of the studies were categorized as having a high bias in the other bias section since most of them were funded by industry. Only one study received no industry funding. One study was assigned high bias in the blinding of participants and personnel section since it was a single-blinded (rater-blinded) study; while others that assigned low bias were double-blind (patient- and rater-blinded) studies. Funnel plots of outcomes were presented in Figure S3. Response rate Response rates were significantly increased in the guided group compared with the usual group at week 8 (OR 1.32, 95%CI 1.15–1.53, 8 studies, 4328 participants) and week 12 (OR 1.36, 95%CI 1.15–1.62, 4 studies, 2814 participants) . However, no significant difference was observed between the guided group and the usual group at week 4 (OR 1.12, 95%CI 0.89–1.41, 2 studies, 2261 participants) or week 24 (OR 1.16, 95%CI 0.96–1.41, 2 studies, 2252 participants) (Fig. ). Remission rate Compared with the usual group, the guided group was associated with an increased remission rate at week 8 (OR 1.58, 95%CI 1.31–1.92, 8 studies, 3971 participants) and week 12 (OR 2.23, 95%CI 1.23–4.04, 5 studies, 2664 participants) . Similarly, we found no significant difference between the guided group and the usual group at week 4 (OR 1.26, 95%CI 0.93–1.72, 2 studies, 2261 participants) and week 24 (OR 1.06, 95%CI 0.83–1.34, 2 studies, 2252 participants) (Fig. ). Medication congruence in 30 days Medication congruency in 30 days was significantly reduced in the pharmacogenomic guided group compared with the usual care group (OR 2.07, 95%CI 1.69–2.54, 3 studies, 2862 participants) (Fig. ). Subgroup analysis and sensitivity analysis Table shows the subgroup analysis according to several parameters. Specifically, we found significant differences between subgroups of the target population in both response (P-value between subgroup = 0.03) and remission rate (P-value between subgroup < 0.01). In addition, significant differences between subgroup in subgroup analyses according to study design, sample size, and industry funding of remission rate were also observed, while no significant difference between subgroups was found between different diagnostic criteria groups; however, the pharmacogenomic testing guided group compared with usual group were all associated with an increased rate of remission in subgroups. No outcome changed when performing sensitivity analysis using the leave-one-out method (Figure S4). The overall result of the risk assessment is summarized in Figure S2. Most of the studies were categorized as having a high bias in the other bias section since most of them were funded by industry. Only one study received no industry funding. One study was assigned high bias in the blinding of participants and personnel section since it was a single-blinded (rater-blinded) study; while others that assigned low bias were double-blind (patient- and rater-blinded) studies. Funnel plots of outcomes were presented in Figure S3. Response rates were significantly increased in the guided group compared with the usual group at week 8 (OR 1.32, 95%CI 1.15–1.53, 8 studies, 4328 participants) and week 12 (OR 1.36, 95%CI 1.15–1.62, 4 studies, 2814 participants) . However, no significant difference was observed between the guided group and the usual group at week 4 (OR 1.12, 95%CI 0.89–1.41, 2 studies, 2261 participants) or week 24 (OR 1.16, 95%CI 0.96–1.41, 2 studies, 2252 participants) (Fig. ). Compared with the usual group, the guided group was associated with an increased remission rate at week 8 (OR 1.58, 95%CI 1.31–1.92, 8 studies, 3971 participants) and week 12 (OR 2.23, 95%CI 1.23–4.04, 5 studies, 2664 participants) . Similarly, we found no significant difference between the guided group and the usual group at week 4 (OR 1.26, 95%CI 0.93–1.72, 2 studies, 2261 participants) and week 24 (OR 1.06, 95%CI 0.83–1.34, 2 studies, 2252 participants) (Fig. ). Medication congruency in 30 days was significantly reduced in the pharmacogenomic guided group compared with the usual care group (OR 2.07, 95%CI 1.69–2.54, 3 studies, 2862 participants) (Fig. ). Table shows the subgroup analysis according to several parameters. Specifically, we found significant differences between subgroups of the target population in both response (P-value between subgroup = 0.03) and remission rate (P-value between subgroup < 0.01). In addition, significant differences between subgroup in subgroup analyses according to study design, sample size, and industry funding of remission rate were also observed, while no significant difference between subgroups was found between different diagnostic criteria groups; however, the pharmacogenomic testing guided group compared with usual group were all associated with an increased rate of remission in subgroups. No outcome changed when performing sensitivity analysis using the leave-one-out method (Figure S4). The current meta-analysis identified 11 RCTs that compared the effect of pharmacogenomic testing on target outcomes were identified. Quantitative pooled analysis showed that pharmacogenomic testing guided treatment contributed to improved response and remission rates compared to usual treatment at week 8 and 12, while no significantly higher response or remission rates were found at week 4 or 24. Pharmacogenomic testing guided treatment shortened the time to clinical remission and response to antidepressants. In addition, the pharmacogenomic testing guided treatment decreased medication congruence compared to usual care. In subgroup analysis, we found significant differences between the Asian subgroup and the Caucasian subgroup. Overall, these data showed that pharmacogenomic testing guided treatment helped MDD patients achieve clinical remission or response in a shorter time compared to usual treatment while making no difference in final response or remission rates at the end of the pharmacogenomic testing guided treatment. Notably, our results were different from previous meta-analyses, which simply showed higher response and remission rates in pharmacogenomic testing guided MDD patients than usual treated patients . When combining all the outcomes at any follow-up duration from included studies, we found similar results as the previous studies (Figure S5). However, when separating the data into week 4, 8, 12, and 24, we found significantly improved response and remission rates in week 8 and 12 rather than 4 and 24. No significant changes were found at week 4 may because of the long onset time of anti-depressants, and no significant changes were found at week 24 may because pharmacogenomic testing guidance only showed a ‘catalyst-like’ effect by accelerating the process of excluding the unsuitable anti-depressants for MDD patients rather than improving the MDD therapeutic effects. This current meta-analysis found fewer gene-drug interactions in MDD patients receiving pharmacogenomic testing guided treatment than usual in those receiving usual treatment. These results showed an effect of pharmacogenomic testing guidance on decreasing severe adverse effects induced by gene-drug interactions during MDD medications. Pharmacogenomic testing guidance provided relevant clinical information on the effect of an anti-depressant, potential gene-drug interactions, and adverse effects for physicians and patients, helping to make more accurate decisions . The genes that encode hepatic cytochrome P450 (CYP450) enzymes receive great concerns in pharmacogenomic testing, because of their roles in the regulations of psychotropic drug metabolism . Previous studies revealed the critical role of CYP450 enzymes on the pharmacokinetics of anti-depressants, especially CYP2C19 and CYP2D6 , with high-level evidence showing relevance to medications commonly used in psychiatry practice . Briefly, findings showed that variants of CYP2C19 and CYP2D6 have been associated with blood concentrations of antidepressants, adverse drug reactions, and clinical outcomes of antidepressants . Moreover, variations of other genes were also thought to be closely associated to different responses to antidepressants, such as the human gene-encoding serotonin transporter (SCL6A4) gene, catechol-o-methyltransferase (COMT) gene, and serotonin-2 A receptor (HTR2A) gene . In subgroup analysis, the current study showed a significant difference between the Asian subgroup and the Caucasian subgroup. The difference may be associated with the sub-genotype of allele frequencies of the gene variants. For example, more CYP2C19 and CYP2D6 variants were found in Asian people than in Caucasians. Almost 20% of Asians were CYP2C19 poor metabolizers while 3% of Caucasians were poor metabolizers ; only 1% of healthy Asians were attributed to poor metabolizers of CYP2D6 while the proportion of Caucasians was 5–10% . Furthermore, allele frequencies of the SLC6A4 gene variant between Caucasians and Asians are different, the S allele being present in 42% of Caucasians but in 79% of Asians . The difference in pharmacokinetics enzymes led to different responses to anti-depressants between Asian and Caucasian. Among those genes which were reported to have different responses to antidepressants, CYP450 enzyme genes, especially CYP2C19 and CYP2D6, may mask the effects of other genes because of the sensible effects on the regulation of antidepressant metabolism. But meanwhile, it’s still challenging for pharmacogenomic testing guidance utilized in clinical MDD medications. Firstly, there is not enough evidence to support the use of pharmacogenomic testing guidelines in MDD clinical practice. The US Food and Drug Administration cautioned about the content validity and potentially detrimental impact of pharmacogenomic testing panels, while Dutch and French guidelines recommended pharmacogenomic testing guided treatment as a potentially useful tool of MDD medication . Although the current study showed the effects of pharmacogenomic testing guidance on shortening the time to identify an appropriate antidepressant, there are still difficulties in developing pharmacogenomic testing guidance on MDD medications. Secondly, the cost-effectiveness of pharmacogenomic testing guidance remains unclear. The average number of genes tested in studies included in the current meta-analysis was more than 10, which may lead to huge expenditure and result in low acceptance of MDD patients. Though pharmacogenomic testing may not lead to less direct cost than usual antidepressant treatment, it’s still a very profitable tool to decrease taking inappropriate antidepressants of MDD patients and help reduce indirect costs . Moreover, several institutions provide different pharmacogenomic testing tools and services; all those tools have their unique pipelines, and results are not provided uniformly . Thus, clinicians should get support from genetic counselors, pharmacists, or companies for proper implementation . Previous case reports highlighted the key role of pharmacist assessment during drug-gene interactions and drug‐induced pheno-conversion in MDD. Further open-label RCT about pharmacist-guided pharmacogenetic testing in antidepressant therapy is in progress . Pharmacogenomic testing guidance would make a huge difference in MDD medications if those problems could be resolved. The current study included only RCTs with longer follow-up duration and a much larger sample size. We also analyzed data at multiple time points and conducted multiple comprehensive sub-group analyses. Although this study revealed the effect of pharmacogenomic testing guided treatment on MDD medications and indicated its striking value in Asian MDD patients, there are still several methodological limitations that may lead to interference and bias that emerged from this study. The reliability of the inclusion of unblinded clinicians was reduced in most of the reported pooled effect studies. In addition, patients were unblinded in the largest scale research included in this current analysis, which may also lead to detection bias of outcomes. Another significant limitation of this current study is on basis of data in per-protocol analysis rather than intention-to-treat analysis, which may lead to overestimating the outcomes of pharmacogenomic testing guided treatment on MDD medications. Nevertheless, this meta-analysis is a re-analysis based on current research so a few genotypes were involved. Previous reports focused on the effects of CYP2C19 and CYP2D6 on antidepressants, thus our research mostly mentioned the two variants . In the future, more genotypes would be involved in our prospective research. Our study demonstrated the effect of pharmacogenomic testing guidance in shortening the process of reaching clinical remission and response to anti-depressant medications in MDD. Future well-designed multi-ethnic studies are needed to confirm the benefit of pharmacogenomic testing guidance in different populations. Below is the link to the electronic supplementary material. Supplementary Material 1 Figure S1 . PRISMA Flow Diagram Supplementary Material 2 Figure S2 . Summary of Risk Assessment Using the Cochrane Collaboration’s Tool For Assessing the Risk of Bias Supplementary Material 3 Figure S3 . Funnel plot of outcomes Supplementary Material 4 Figure S4 . Sensitivity Analysis of included studies Supplementary Material 5 Figure S5 . Forest plot of response and remission rate comparing pharmacogenomic guided treatment versus usual care treatment when not separating data into different weeks Supplementary Material 6 Table S1 . Search Terms
Comparison of internal medicine applicant and resident characteristics with performance on ACGME milestones	388bfdf2-09ef-4834-aed9-d7aeb45756cd	10177668	Internal Medicine[mh]	Residency programs rely on several metrics to screen and rank applicants for the National Residency Match Program . Objective measures such as the United States Medical Licensing Examination (USMLE) Step scores are frequently used , however the correlation between Step scores and residency performance is unclear . Factors during residency may also affect performance and it is unclear which factors could contribute to residency performance. While several factors are reported to correlate with resident performance, differences in study design, tested characteristics, medical/surgical specialty, and definition of success limit the generalizability of these findings across residency programs . The lack of standardized evaluation of residents represents a notable gap, with prior studies relying on different evaluation methods . In July 2013, the Accreditation Council for Graduate Medical Education (ACGME) launched a milestone-based evaluation, which was developed as a standardized framework to assess competency across key professional domains, and to create a logical trajectory of resident performance . Recently, the relationship between internal medicine (IM) residency applicant characteristics and intern year performance on ACGME milestones reported by Golden, et al suggest applicant characteristics were not reflective of performance . Considering that resident knowledge and responsibilities typically increase over time, it is possible unique resident characteristics are more predictive of resident performance than applicant factors. To date, no studies have assessed how applicant and resident characteristics longitudinally affect residency performance. In this study, we aimed to identify applicant and resident factors associated with IM residency performance as measured by ACGME milestones during each post-graduate year. We hypothesized that a longitudinal assessment of resident performance over several years may identify characteristics predictive of performance during IM residency. Study design and setting We performed a predictive modeling study at the University of Utah IM Residency Program, an ACGME accredited site. Residents primarily rotate between three hospital-based settings: 1) University of Utah Medical Center, a 548-bed academic medical center; 2) Intermountain Medical Center, a 452-bed community hospital; and 3) George E. Wahlen Department of Veterans Affairs Medical Center, a 127-bed level 1a facility. The study period began 1 July 2015, and ended 1 July 2020. Participants Categorical IM and combined medicine/pediatric residents who completed at least three consecutive post-graduate years (PGY) of training during the study period were included. For categorical IM residents that were selected to complete an additional chief medical resident (CMR) year, the CMR year was excluded from analysis. Similarly, for combined medicine/pediatric residents, only IM clinical rotations were included. Preliminary interns were excluded as we aimed to evaluate residents longitudinally. Predictive factors and study outcomes Several factors were identified as potential predictors of resident clinical performance including applicant factors such as age when starting residency, gender identity, USMLE Step 1 and Step 2 Clinical Knowledge (CK) scores, position on rank list for residency match, and residency interview score as well as other factors that occurred during residency training such as In-Training Examination (ITE) scores, Step 3 score, and referral to the Clinical Competency Committee (CCC) for discussion of professionalism concerns. These factors were chosen as prior studies have attempted to evaluate the correlation with these factors to performance (but have not previously been studied in a longitudinal fashion over an entire Internal Medicine Residency). Interview scores are unique to our program and we wanted to study any correlation with these scores to performance. The impact of age and gender on performance is unclear but we included these factors as they had not been studied at our institution. The CCC is a forum for evaluating resident performance in any of the ACGME six core competencies. At our institution, the CCC is tasked to identify residents who are not making expected progress and design individualized learning plans. We excluded residents referred to the CCC solely for medical knowledge issues, as medical knowledge was assessed independently in our study by ITE scores. Only residents with professionalism concerns were included in the ‘referral to CCC for professionalism’ variable as these made up the majority (86%) of referrals after excluding medical knowledge deficiencies. Professionalism concerns were identified from evaluation comments and were not derived from the ACGME Professionalism core competency domain. Interview scores at our program are assigned to an applicant by each of two interviewers. Scores range from 1 to 5 with 5 being a highly desirable candidate. Interviewers have the entire Electronic Residency Application System (ERAS) application prior to the interview and the score is a subjective rating based on a holistic review of the applicant. All interviewers are given a scoring guide when rating applicants. The main outcome variables were the mean ACGME milestone score achieved at the completion of each PGY (i.e., PGY-1, PGY-2, and PGY-3). Mean milestone scores were derived from attending performance evaluations that are completed on each resident at the end of their rotation. The performance evaluations were designed to identify and rate the individual ACGME milestone core competency domains: Patient Care and Procedural Skills, Medical Knowledge, Systems-Based Practice, Practice-Based Learning and Improvement, Professionalism, Interpersonal and Communication Skills. The mean milestone score was calculated from a compilation of the core competency domain scores. Milestone scores were aggregated at the end of each residency year. This method was felt to attenuate differences in assessors and potentially differences between how different clinical rotations may be assessed. Data sources Data extracted from residency program records included age when starting residency, gender, USMLE Step 1, Step 2-CK, and Step 3 scores, residency rank list position, and ITE scores. The outcome variables were extracted from two different online databases (E-value 2015–2017 and MedHub 2017–2020) used by the residency program to track resident clinical performance evaluations during the study period. The reason for referral to the CCC was extracted from a manual review of monthly CCC meeting minutes. Statistical analysis Baseline characteristics are reported as frequency and percent for categorical variables and median with interquartile range (IQR) for continuous variables. Due to the perceived sensitivity of USMLE Step and ITE data, scores are reported as normalized values. The 2017 mean USMLE Steps 1, 2 CK, and 3 scores were used as the normalizing factor for the 25 th and 75 th percentile and median for each corresponding Step score. The median ITE score for PGY-1 (ITE 1) was used as the normalizing factor for the 25 th and 75 th percentile and median values for each ITE ( i.e ., ITE 1, ITE 2, and ITE 3). Univariable and multivariable regression models were fitted for each PGY mean milestone score from PGY-1 to PGY-3 based on prior research and content expertise. Covariates included in the final regression models were: age when starting residency; USMLE Step 1, 2 CK, and 3 scores; rank list position (as a continuous variable); average interview score; ITE scores up to current PGY level (e.g., ITE 1 and ITE 2 were included in model for PGY-2 milestones, and ITE 3 was excluded as the ITE 3 had not been completed at the end of PGY-2); and referral to CCC. Given the multiple comparisons of predictor variables with outcomes stratified by PGY, a Bonferroni correction was applied to provide a more conservative p -value threshold (<0.017) for significance . Based on the results from our primary analysis, an exploratory, post-hoc sensitivity analysis was performed to test for associations between individual ACGME core competencies and referral to the CCC. Stata/IC version 16.1 (StataCorp, College Station, TX) was used for all analyses. The local institutional review board deemed this study exempt (IRB_00127307). We performed a predictive modeling study at the University of Utah IM Residency Program, an ACGME accredited site. Residents primarily rotate between three hospital-based settings: 1) University of Utah Medical Center, a 548-bed academic medical center; 2) Intermountain Medical Center, a 452-bed community hospital; and 3) George E. Wahlen Department of Veterans Affairs Medical Center, a 127-bed level 1a facility. The study period began 1 July 2015, and ended 1 July 2020. Categorical IM and combined medicine/pediatric residents who completed at least three consecutive post-graduate years (PGY) of training during the study period were included. For categorical IM residents that were selected to complete an additional chief medical resident (CMR) year, the CMR year was excluded from analysis. Similarly, for combined medicine/pediatric residents, only IM clinical rotations were included. Preliminary interns were excluded as we aimed to evaluate residents longitudinally. Several factors were identified as potential predictors of resident clinical performance including applicant factors such as age when starting residency, gender identity, USMLE Step 1 and Step 2 Clinical Knowledge (CK) scores, position on rank list for residency match, and residency interview score as well as other factors that occurred during residency training such as In-Training Examination (ITE) scores, Step 3 score, and referral to the Clinical Competency Committee (CCC) for discussion of professionalism concerns. These factors were chosen as prior studies have attempted to evaluate the correlation with these factors to performance (but have not previously been studied in a longitudinal fashion over an entire Internal Medicine Residency). Interview scores are unique to our program and we wanted to study any correlation with these scores to performance. The impact of age and gender on performance is unclear but we included these factors as they had not been studied at our institution. The CCC is a forum for evaluating resident performance in any of the ACGME six core competencies. At our institution, the CCC is tasked to identify residents who are not making expected progress and design individualized learning plans. We excluded residents referred to the CCC solely for medical knowledge issues, as medical knowledge was assessed independently in our study by ITE scores. Only residents with professionalism concerns were included in the ‘referral to CCC for professionalism’ variable as these made up the majority (86%) of referrals after excluding medical knowledge deficiencies. Professionalism concerns were identified from evaluation comments and were not derived from the ACGME Professionalism core competency domain. Interview scores at our program are assigned to an applicant by each of two interviewers. Scores range from 1 to 5 with 5 being a highly desirable candidate. Interviewers have the entire Electronic Residency Application System (ERAS) application prior to the interview and the score is a subjective rating based on a holistic review of the applicant. All interviewers are given a scoring guide when rating applicants. The main outcome variables were the mean ACGME milestone score achieved at the completion of each PGY (i.e., PGY-1, PGY-2, and PGY-3). Mean milestone scores were derived from attending performance evaluations that are completed on each resident at the end of their rotation. The performance evaluations were designed to identify and rate the individual ACGME milestone core competency domains: Patient Care and Procedural Skills, Medical Knowledge, Systems-Based Practice, Practice-Based Learning and Improvement, Professionalism, Interpersonal and Communication Skills. The mean milestone score was calculated from a compilation of the core competency domain scores. Milestone scores were aggregated at the end of each residency year. This method was felt to attenuate differences in assessors and potentially differences between how different clinical rotations may be assessed. Data extracted from residency program records included age when starting residency, gender, USMLE Step 1, Step 2-CK, and Step 3 scores, residency rank list position, and ITE scores. The outcome variables were extracted from two different online databases (E-value 2015–2017 and MedHub 2017–2020) used by the residency program to track resident clinical performance evaluations during the study period. The reason for referral to the CCC was extracted from a manual review of monthly CCC meeting minutes. Baseline characteristics are reported as frequency and percent for categorical variables and median with interquartile range (IQR) for continuous variables. Due to the perceived sensitivity of USMLE Step and ITE data, scores are reported as normalized values. The 2017 mean USMLE Steps 1, 2 CK, and 3 scores were used as the normalizing factor for the 25 th and 75 th percentile and median for each corresponding Step score. The median ITE score for PGY-1 (ITE 1) was used as the normalizing factor for the 25 th and 75 th percentile and median values for each ITE ( i.e ., ITE 1, ITE 2, and ITE 3). Univariable and multivariable regression models were fitted for each PGY mean milestone score from PGY-1 to PGY-3 based on prior research and content expertise. Covariates included in the final regression models were: age when starting residency; USMLE Step 1, 2 CK, and 3 scores; rank list position (as a continuous variable); average interview score; ITE scores up to current PGY level (e.g., ITE 1 and ITE 2 were included in model for PGY-2 milestones, and ITE 3 was excluded as the ITE 3 had not been completed at the end of PGY-2); and referral to CCC. Given the multiple comparisons of predictor variables with outcomes stratified by PGY, a Bonferroni correction was applied to provide a more conservative p -value threshold (<0.017) for significance . Based on the results from our primary analysis, an exploratory, post-hoc sensitivity analysis was performed to test for associations between individual ACGME core competencies and referral to the CCC. Stata/IC version 16.1 (StataCorp, College Station, TX) was used for all analyses. The local institutional review board deemed this study exempt (IRB_00127307). A total of 89 individuals completed three consecutive years of IM residency training during the study period and were included. Two residents in the physician scientist training program advanced to fellowship after completion of PGY 2 and, therefore, are not included in the analysis for PGY 3. The median age at the start of residency was 28 years (IQR 27–29) and the majority were male (59.6%). The normalized median Step 1, Step 2 CK and Step 3 scores were 1.03 (IQR 0.98–1.09), 1.01 (0.98–1.05) and 1.03 (0.99–1.07) respectively. The mean ACGME milestone scores increased with each year of PGY training from 3.36 (SD 0.19) for PGY 1, to 3.80 (SD 0.15) for PGY 2, and 4.14 (SD 0.15) for PGY 3 . Normalized median ITE scores also increased for each PGY year of training, as shown in . Univariable regression analysis Results of the univariable analysis are presented in . Specifically, USMLE Step 3 scores were associated with small, significant increases in mean milestone scores during each PGY of training (beta coefficients ranged from 0.005–0.01, p = 0.001). Step 2 CK scores also appeared to be weakly associated with milestone scores during PGY-2 and PGY-3 . Similarly, ITE scores available during the current year of residency (e.g., ITE 2 during PGY-2) and preceding years of PGY training (e.g., ITE 2 during PGY-3) were associated with small, significant increases in mean milestones (beta coefficients ranged from 0.002–0.003). Significantly lower mean milestone scores during each year of residency were observed among residents referred to the CCC due to professionalism issues (beta coefficients ranged from −0.12 to −0.21). Applicant factors including age at the start of residency, gender, average interview score, and rank list position were not significantly associated with milestone scores, except during PGY-2 where a small negative association was noted between rank list and mean milestone score (beta = −0.001 [95% CI −0.001, −0.0001], p = 0.013). Multivariable regression analysis None of the resident or applicant factors included in the final multivariable regression models (age at starting residency, USMLE Step scores, interview score, rank list position, ITE scores, and referral to CCC) as predictors of resident clinical performance were significantly associated with mean ACGME milestone scores for PGY-1 and PGY-2 . For PGY-3 residents, referral to the CCC for professionalism concerns had a significant negative association with mean milestone scores (beta = −0.13 [95% CI −0.22, −0.04], p = 0.006). We performed a post-hoc sensitivity analysis to explore milestone scores for each of the ACGME core competency domains to determine if this association was largely attributable to the Professionalism domain, or if other domains were impacted. This secondary analysis suggests referral to the CCC for professionalism is negatively associated with resident milestone scores across all the core competency domains. Results of the univariable analysis are presented in . Specifically, USMLE Step 3 scores were associated with small, significant increases in mean milestone scores during each PGY of training (beta coefficients ranged from 0.005–0.01, p = 0.001). Step 2 CK scores also appeared to be weakly associated with milestone scores during PGY-2 and PGY-3 . Similarly, ITE scores available during the current year of residency (e.g., ITE 2 during PGY-2) and preceding years of PGY training (e.g., ITE 2 during PGY-3) were associated with small, significant increases in mean milestones (beta coefficients ranged from 0.002–0.003). Significantly lower mean milestone scores during each year of residency were observed among residents referred to the CCC due to professionalism issues (beta coefficients ranged from −0.12 to −0.21). Applicant factors including age at the start of residency, gender, average interview score, and rank list position were not significantly associated with milestone scores, except during PGY-2 where a small negative association was noted between rank list and mean milestone score (beta = −0.001 [95% CI −0.001, −0.0001], p = 0.013). None of the resident or applicant factors included in the final multivariable regression models (age at starting residency, USMLE Step scores, interview score, rank list position, ITE scores, and referral to CCC) as predictors of resident clinical performance were significantly associated with mean ACGME milestone scores for PGY-1 and PGY-2 . For PGY-3 residents, referral to the CCC for professionalism concerns had a significant negative association with mean milestone scores (beta = −0.13 [95% CI −0.22, −0.04], p = 0.006). We performed a post-hoc sensitivity analysis to explore milestone scores for each of the ACGME core competency domains to determine if this association was largely attributable to the Professionalism domain, or if other domains were impacted. This secondary analysis suggests referral to the CCC for professionalism is negatively associated with resident milestone scores across all the core competency domains. In this study we tested several applicant factors (e.g., Step 1 and Step 2 CK scores, rank list position, age, gender) and resident factors (e.g., Step 3 score, ITE scores, and referral to the CCC for professionalism concerns) as predictors of IM resident performance. Notably, we did not observe consistent associations between applicant or resident factors and resident performance measured by ACGME milestone scores. In our univariable analysis we did observe a negative association with referral to the CCC, as well as weak positive associations between ITE and USMLE scores, with milestone scores. However, following a multivariable analysis only the negative association with referral to the CCC for professionalism among PGY-3 residents was significant. Several studies have attempted to identify factors within residency applications and faculty or patient evaluations to predict resident performance with mixed results . The study by Fine et al . concluded there was an overemphasis on Alpha Omega Alpha (AOA) status, medical school reputation, and Step 1 scores, only modestly correlating ( r = −0.52) to resident performance evaluations . Similarly, Neely et al . examined several applicant factors to derive a weighted algorithm to predict resident performance . In contrast to the conclusions of Fine et al ., the Neely algorithm relies heavily on medical school quality and Step 1 scores. The work by Sharma et al . suggested USMLE Step 2 CK was the best predictor of residency performance when measured by a multimodal ambulatory care evaluation . These conflicting results are difficult to generalize as each of these studies relied on non-standardized assessments of resident performance prior to implementation of the ACGME milestones. Although the ACGME milestones may not be a perfect measure of performance, they stand as the most widely used metric of resident performance available . Recently, Golden et al . examined the associations between applicant factors and ACGME milestones as a reflection of resident performance limited to intern year . They concluded ‘most traditional metrics used in residency selection were not associated with early performance on ACGME milestones during internal medicine residency.’ Our study has several strengths and builds on previous efforts to predict IM resident performance and warrants further discussion. First, we utilized a widely used metric to assess resident performance, the ACGME milestones . This enhances the external validity and generalizability of our findings across IM programs compared to older studies that relied on institution-specific evaluation systems . Next, we included milestone evaluations from multiple types of practice settings (e.g., ambulatory care clinics, general medicine wards, subspecialty inpatient wards, subspecialty clinics, intensive care units, etc.) encompassing every clinical site residents rotate through in our program. This provides a more comprehensive assessment of resident performance compared to a single practice setting (i.e., ambulatory care clinic) as in the Sharma et al . study . Furthermore, we examined resident performance longitudinally across all 3 years of IM training for categorical residents entering our residency program, as opposed to restricting our analysis to the intern year alone. Longitudinal assessment could avoid missing discrepancies in performance that may arise as residents undergo shifting expectations over the course of their training. Based on these study design strengths, we conclude the applicant factors used (e.g., Step 1, Step 2 CK, rank list position, interview score) to guide the resident selection process do not predict resident performance based on ACGME milestone evaluation scores. Several associations between resident factors (e.g., Step 3 and ITE scores) and resident milestone scores were identified in our analysis, all with small effect sizes and are of uncertain significance. However, referral to the CCC for professionalism issues demonstrated the largest effect size and was negatively associated with resident milestone scores during PGY-3. We speculate the inverse relationship between referral to the CCC and resident milestone scores may be related to underlying professionalism issues negatively affecting a resident’s overall performance. Professionalism is an important characteristic to measure as previous reports suggest individuals with unprofessional behavior during medical school and residency have higher rates of disciplinary action by medical boards during their post-training careers . Furthermore, Dupras et al . reported that ‘residents in difficulty’ with professionalism concerns often had deficiencies in multiple competencies . While unprofessional behavior seems like a characteristic that would be easy to identify during the residency selection process, only one-third of program directors could retrospectively identify residents at risk for poor performance based on application materials . We do not have a solution that will assist program directors with this dilemma, though some potential tools already exist in the residency application. One of these tools is the professionalism section of the Medical Student Performance Evaluation (MSPE), which was identified as the most important section of the MSPE by program directors and selection committees across all specialties in a recent study by Bird et al . (though there was remarkable distrust of the MSPE by program directors) . Professionalism concerns may also hint to underlying factors such as burnout or mental health disorders that could broadly affect performance. In summary, we propose referral to the CCC for professionalism may be a predictor of resident performance and should alert program directors to potential professionalism issues or underlying resident factors that negatively impact resident milestone scores across all core competencies. This study is not without limitations. The single center, retrospective design is a key study limitation and may limit the generalizability of our findings. Additionally, by only including residents who matched in our IM program, we recognize that selection bias may limit our ability to detect factors that predict performance. Another limitation is that only one specialty (IM) was included so our conclusions may not apply to other specialties. The main outcome measure used in our study, mean ACGME milestone score, has been questioned and may not accurately reflect a resident’s global performance , although milestones remain the most widely studied measure to date we concede that aggregating milestone scores may not fully capture resident performance over time and assessors may vary in how they assign milestone scores. It is possible other unmeasured factors that were not studied may be more predictive of resident performance (e.g., personality traits, mental/physical illness during residency, resident wellness, medical school performance, participation in sports teams, etc.) . The use of rank list is another limitation as this is a combination of various factors both subjective and objective. Anecdotally, the generation of rank lists occupies a large portion of program directors’ time. Rank list has been shown to correlate with higher ACGME milestone scores in a univariate analysis in one study . Our results did not demonstrate a significant correlation with higher milestone scores. These conflicting findings question the utility of the rank list in predicting resident success. Lastly, the Bonferroni correction that was used is a more conservative approach to adjust for family-wise error rate when conducting multiple tests. It is possible that the type 2 error rate was increased while trying to improve the type 1 error rate . Future studies are needed to test these metrics in other specialties and verify our findings within a broader population, particularly our findings on the impact of referral to the CCC. Further studies should explore additional factors including research in medical school, volunteer work, and the professionalism section of the MSPE. Finally, while milestones represent a fair measure of assessment, continuing to evaluate their effectiveness as a metric of resident performance as compared to other measures remains important. In our study, common residency selection factors did not predict IM resident milestone evaluation scores. Referral to the CCC for professionalism was correlated with worse resident milestone scores across all domains during PGY-3, suggesting professionalism issues correlate with clinical performance.
Assessing the competitiveness of medical humanities research on psychiatry, otolaryngology, and ophthalmology residency program applications	50946f92-af8f-4b6c-af16-26ab7b50eeab	10177745	Ophthalmology[mh]	Research in the medical humanities investigates the connection between medicine and humanities fields like philosophy, history, literature, anthropology, law, music, and art. The scholarly pursuit of medical humanities can take many forms beyond the standard scientific paper, including essays, poems, visual art, music compositions, and workshops. In general, medical humanities as an academic discipline has become an increasingly important part of medical school education . In many cases, it is now a required component of the curriculum . Some have even argued that undergraduate humanities majors are the ideal candidates for medical school admissions because they may possess a more compassionate and nuanced understanding of the human condition . Early medical school admission programs, such as the Flexmed program at the Icahn School of Medicine at Mount Sinai, relieve accepted undergraduate students of premedical requirements so they may pursue humanities interests, reflecting a shift towards person-centered care and a holistic application review process. Along the same line, the Medical College Admission Test (MCAT) as of 2015 has put a stronger emphasis on humanities and social sciences . Although residency programs have begun incorporating elements of narrative medicine and storytelling into their education, it remains to be seen whether the residency admissions process embraces medical humanities on the same level as medical schools and undergraduate institutions . As residency placement becomes more competitive each year and traditional metrics, such as the United States Medical Licensing Exam (USMLE) Step 1 and pre-clinical course performance, switch to pass/fail grading systems, medical students are concerned with what residency programs want to see in their applicants. Previous studies have attempted to tease out the importance of research on residency applications with varying results. While research consistently ranks among one of the most important factors on an application , it has been shown that research quantity beyond one first author publication is not a significant factor in matching . However, the high number of students who misrepresent their research output combined with the large increase in medical student research activity in the last decade suggests that students consider it a top priority . One of the key questions students are faced with in medical school is what type of research they want to pursue. The cultural shift away from objective scores and metrics towards holistic application review raises important questions about the perception of non-clinical research, including the medical humanities. While the medical humanities are solidly embraced on the university level and increasingly so on the medical school level, residency programs have been slow to adopt them . Furthermore, researchers are still searching for a truly effective, quantitative method of measuring the impact of medical humanities on students . As a result, students may express concern that pursuing medical humanities activities detracts from their competitiveness in residency, especially if it is at the cost of not pursuing clinical research. To our knowledge, no prior study exists that directly asks residency program directors (PDs) for their perception of medical humanities research on program applications. We used a mixed methods approach to both satisfy the current lack of quantitative data existing on medical humanities and to capture the nuances that the humanities requires. We hypothesize that doing medical humanities research will not preclude students from being seriously considered for residency programs in either surgical and medicine fields, and may even help students stand out among their peers. It is our hope that this study alleviates hesitations that students may feel about pursuing medical humanities research during medical school. Our study surveyed residency PDs across New York State in both surgical and medicine fields (otolaryngology, ophthalmology, and psychiatry) about their opinion on the relative competitiveness of residency program applicants who completed medical humanities research as opposed to clinical research. Additionally, we probed what directors perceived to be the benefits, if any, of medical humanities in being a better doctor in residency and beyond. Ophthalmology and otolaryngology were selected because of their perceived competitiveness and limited residency spots. Psychiatry was chosen as a contrast to these two fields as it is known to attract more open-minded and humanities-inclined applicants . We completed this descriptive study via administration of a 5-question online survey on Google Forms (Google LLC) to residency PDs of the 2022–2023 academic year. All PDs of New York State residency programs were identified using residency program websites. A total of 64 PDs were emailed with an invitation to the survey, with two subsequent follow-up reminder emails. There was a total of 37 psychiatry PDs, 16 ophthalmology PDs, and 11 otolaryngology PDs. Compensation was not provided, and participation was voluntary. Neither patients nor the public were involved in the design, conduct, reporting, or dissemination plans of our research. The surveys were composed of four multiple choice response questions and one open-ended response question. An optional question was left for additional comments. Questions were designed to optimize ease of completion for the residency PDs and utility for potential residency program applicants in determining the positive, neutral, or negative effect of medical humanities research on residency application competitiveness. Three types of questions, dichotomous, Likert scale, and open-ended, were incorporated to maximize efficiency as well as leave space for optional expression. No names or demographic variables were recorded. All questions are included in the Results section verbatim. All questions, both quantitative and qualitative, were recorded and analyzed based on number and sentiment by the authors using Google Sheets (Google LLC). For the open-ended questions, the authors categorized the responses based on perceived sentiment (positive, neutral, or negative) on the effect of humanities research on residency applicant competitiveness. A total of 20 PDs completed the survey (31.3%). Of these, ten (27.0%) were from psychiatry, six (37.5%) from ophthalmology, and four (36.3%) from otolaryngology. The responses to the first four questions are graphically represented in . The fifth question, ‘How do you think medical humanities research and/or coursework affects an individual’s performance in residency and beyond?’ yielded a variety of responses. All are included in based on perceived positive, neutral, or negative sentiment on the effect of humanities research on residency applications. Notable positive responses included ‘Possibly that resident is more compassionate or takes in effect the psychosocial impact on healthcare’ and ‘Yes I think it can inform a better understanding of the humanistic aspects of being a doctor and psychiatrist.’ Notable neutral responses included ‘No effect’ and ‘Very little. It may make the individual more interesting, but unlikely to improve their mastery of diagnosis and treatment.’ Significantly, no responses were identified to express that medical humanities research would negatively affect a residency applicant’s competitiveness. Question 6 offered PDs a space to expound upon question 5. A total of six PDs (6/20, 30%) filled out this optional question. Of these three (3/10, 30%) psychiatry, two (2/6, 33%) ophthalmology, and one (1/4, 25%) otolaryngology PD submitted responses. One psychiatry PD noted that ‘one type of research does not weigh more strongly than another.’ One ophthalmology PD echoed this sentiment, writing that ‘on some level, research is research.’ Moreover, they added that they ‘especially enjoy talking to the students about their research in these domains.’ Conversely, a second ophthalmology PD wrote that while humanities research is a ‘nice addition,’ it was not as important as ‘basic science coursework or research.’ All responses are presented in separated based on positive, neutral, or negative sentiment. Our results support that applicants with only medical humanities research background may be seriously considered for psychiatry, ophthalmology, and otolaryngology residency programs . We conclude that students participating in medical humanities research should not be concerned about their relative competitiveness compared to their clinical research peers. In fact, the survey suggests that medical humanities research in addition to clinical research may actually give an edge to applicants . Many residency PDs expressed value in the medical humanities, saying that they encouraged more open-mindedness and empathy (‘resident is more compassionate,’ ‘increases empathy’; ). Studies about the effect of medical humanities education on medical school students have also echoed the same findings . For example, one study found that students perceived value in learning clinical communications skills after taking a medical drawing course . Other students have reported that medical humanities projects provide important self-reflection time and joy outside of their typical coursework . Some PDs were less confident in whether the medical humanities influenced performance in residency (‘No effect,’ ‘unlikely to improve their mastery of diagnosis and treatment’; ). The study of medical humanities has long carried the burden of quantitatively proving its worthiness in medical training. This has proven to be a challenge given that the emotional impact of the humanities is difficult to capture with a numerical analysis . However, the strong support (95%) of medical humanities research displayed in Question 1 of our survey indicates that it may not prevent a student from being seriously considered in both surgical and medicine residency programs . Surprisingly, the sole respondent that answered ‘No’ to Question 1 was from psychiatry, a specialty that has historically attracted humanities-minded students . Impressively, a majority of PDs (65%) said that doing medical humanities in addition to clinical research increased an applicant’s chance of being accepted . This implies that programs may be increasingly seeking well-rounded applicants with diverse interests. This sentiment was further emphasized in qualitative responses. Responses included that the medical humanities ‘helps with … understanding of diversity,’ ‘broadens the view of the individual,’ and ‘may make the individual more interesting’ . Not only may students stand out from their peers if they participate in humanities-based projects, but they could also have a broader, person-centered outlook on clinical care. Because of its inherent capability to promote self-reflection and critical thinking, the arts and humanities are considered an essential vehicle for delivering content related to diversity, equity, inclusion, and disability justice, which are topics increasingly emphasized in medical school curricula, admissions, and standardized testing . PDs were split on whether medical humanities research should pertain to their specialty of interest or not, with most saying that they did not have a preference . Rather than having projects that were in the relevant field, one psychiatry PD expressed that ‘[taking] a project from concept to publication’ was a more important marker of a good applicant . Another psychiatry PD even emphasized that these projects did not need to take the form of a published article, saying ‘we would value the work of someone who published a personal essay as much as someone who did medical humanities research’ . Different modalities of projects can be equally as valid as long as the student rigorously engages with the process. Though the survey results demonstrated that medical humanities research may be a valid and potentially advantageous experience for an applicant, PDs were still hesitant to say that the medical humanities were an important selection criterion for their program overall . However, 25% (1) of otolaryngology PDs, 17% (1) of ophthalmology PDs, and 40% (4) of psychiatry PDs agreed or strongly agreed that it was . This is a high percentage of programs (25%) given that residency programs have only sparsely adopted the medical humanities into their actual curricula. It is possible that this number foreshadows a growing importance and prevalence of the medical humanities in residency programming. Applicants with medical humanities experience may be actively sought out for their skills in the coming years as programs seek to ramp up humanities integration into the clinical space. Our study was limited by the small sample size ( n = 20) and number of specialties. The response rate of 31% could potentially be a sign of selection bias among PDs. Future studies should pursue more specialties on a larger scale. It would also be beneficial to survey PDs beyond New York State. Programs in different states and settings (e.g., rural) may have different priorities for their applicants. Though our questions were designed to be as concise and clear as possible, there is always the possibility that participants misinterpreted or misread questions. Several studies have investigated the otolaryngology and ophthalmology residency application process from the perspective of logistical and financial barriers to students . We are not aware of a previous study that has directly asked residency PDs what they are looking for in applicants. As a consequence, students are often left wondering what they need to do to be a good applicant. This results in students defaulting to what they perceive to be safe areas of research (i.e., clinical). This study was conducted in the hopes that students will read this and feel more assured in the decision to pursue medical humanities research on its own or in additional to clinical research. They should feel empowered in this choice whether they want to enter a competitive surgical field like otolaryngology and ophthalmology or a medical field like psychiatry. Doing so may even confer an advantage to individuals in the application process and in their residency and career performance.
Amino-Pyrazoles in Medicinal Chemistry: A Review	38189906-ece4-4831-9643-d83bdc711dc7	10177828	Pharmacology[mh]	The pyrazole heterocyclic ring represents an important building block in different areas of organic and medicinal chemistry as well as industrial and agricultural applications. Different pyrazole derivatives are used in supramolecular and polymer chemistry, in the food industry, as cosmetic colourings, and as UV stabilisers, while other pyrazoles have liquid crystal properties . However, pyrazole compounds are rarely found in natural products, which is probably related to the difficulty of N-N bond formation by living organisms . Pyrazole-containing molecules display a broad range of biological activities, including anti-inflammatory , anticonvulsant , anticancer , antiviral , antidepressant , analgesic , antibacterial , antifungal , and selective enzyme inhibition . Moreover, several pyrazole compounds are currently used in clinics as anti-inflammatory and analgesic (Celecoxib, Tepoxalin, and Betazole), anticancer (Crizotimib), antiobesity (Surinabant and Difenamizole), antidepressant, and tranquilliser (Fezolamine and Mepiprazole) drugs, thus confirming the pharmacological value of this heterocycle. The pharmacological properties of its nucleus are related to its particular chemical behaviour; pyrazole presents a nitrogen atom 1 (N1), named “pyrrole-like” because its unshared electrons are conjugated with the aromatic system, and a nitrogen atom 2 (N2), named “pyridine-like” since the unshared electrons are not compromised with resonance, similarly to pyridine systems. For this reason, pyrazole can react with both acids and bases . An additional structural characteristic of pyrazole is its prototrophic tautomerism; in fact, three tautomers are possible in unsubstituted pyrazoles ( A), while five tautomers can exist in mono-substituted pyrazoles ( B) . The functionalisation of the pyrazole nucleus with amino substituents in different positions has led to multifunctional pharmacologically active compounds ; in fact, aminopyrazoles (APs) represent a versatile and very useful framework in drug discovery . Some APs have the free amino group (NH 2 ), others bear a substituted amino group, and in some other derivatives, the amino function is part of other heterocycles. Several bioactive pyrazoles (e.g., Aminophenazone and Metamizole, ) present an amino group at position 4; this chemotype is shared by different drug candidates currently in clinical trials as potent inhibitors of several cyclin-dependent kinases (CDKs) AT7519 ( ; five different clinical trials) and AT9283 , known as a multitargeted kinase inhibitor with potent Aurora kinase inhibition (five clinical studies). In addition, Fipronil is a highly substituted 5-aminopyrazole with insecticidal activity. In fact, this compound is able to reversibly block pest GABA-A receptors and disturb the activity of the insect’s central nervous system, causing hyperexcitation of the nerves and muscles of the contaminated insects. Fipronil insecticide is commonly used for the protection of crops and ornamental plants against herbivorous insects and mites, urban pest control, fish farming, and veterinary applications. Due to the importance of this scaffold in medicinal chemistry research, a lot of reviews and articles have reported different summaries regarding the chemical synthesis and pharmacological properties of pyrazole derivatives, with APs being the most studied chemotype . Particular attention has also been paid to pyrazole synthesis methods . Collectively, anticancer and antimicrobial properties have emerged as the most frequent biological uses reported for the pyrazole scaffold . In view of pyrazole’s high potential behaviour, in the current review, we focused our attention on the AP scaffold, which represents one of the most studied moieties in medicinal chemistry, both in academia and in industry. This review focuses on AP compounds presenting a free amino group (hydrogen bond donor) which is responsible for the formation of numerous interactions with different targets. We, therefore, identify 3-aminopyrazoles (3APs), 4-aminopyrazoles (4APs), and 5-aminopyrazoles (5APs) which, during the last 20 years, have been extensively studied in various fields of medicinal chemistry. In addition, some examples of 3,5-diaminopyrazoles (3,5DAPs) were reported. As reported below, compounds characterised by a pyrazole scaffold substituted with an amino group in position 3 (3APs) are largely reported as anticancer and anti-inflammatory agents but are also known for their anti-infective properties. 2.1. Anti-Infective 3APs Regarding anti-infective activity, 3APs have been investigated as antibacterial and antiviral agents; in some cases, the 3AP scaffold has been incorporated into a more complex structure. In detail, Delpe-Acharige and coll. reported a series of 3APs bearing a thiourea moiety in N1 (compounds 1 , ) with sub-micromolar activity against Methicillin-sensitive Staphylococcus aureus (MSSA) and Methicillin-resistant S. aureus (MRSA) in the presence of bioavailable copper. This study clearly established the suitability of using pyrazolyl thioureas for the treatment of these types of infections . In 2020, some authors from Roche Pharma Research and Early Development identified a small library of pyrido[2,3- b ]indole 2 able to block Gram-negative strains, targeting both DNA Gyrase and Topoisomerase IV. Among all the synthesised compounds, derivative 2a , characterised by a 3AP substituent on a pyrido-indole scaffold, exhibited MIC values of 0.125 and 8 mg/mL against S. aureus and E. coli , respectively . In the same year, Fahim and coll. synthesised a series of 3APs as intermediates for the synthesis of novel pyrazolo[1,5- a ]pyrimidine derivatives. All the newly isolated compounds were screened on a Well Diffusion Assay to evaluate their antimicrobial activity against three bacteria strains ( B. subtilis , S. pneumoniae , and E. coli ) and three fungi ( A. flavus , S. racemosum , and G. candidum ). The 3APs 3a – d showed high activity against both bacteria and fungi strains (inhibition zone diameter > 15 mm), highlighting the promising properties of this class of compounds as antibacterial agents . In 2009, researchers from TIBOTEC (France), identified some 4-aryloxy-3-iodopyridin-2(1 H )-ones that have been evaluated as anti-HIV inhibitors. The authors investigated different substituents at the 5-position of this scaffold, including one compound characterised by a 3AP moiety in the 4 position (derivative 4 , ) which showed potent HIV-1 reverse transcriptase inhibitory properties against both the wild-type enzyme and simple/double mutated forms . 2.2. Anticancer and Anti-Inflammatory 3APs The most active 3AP compounds with antitumour and anti-inflammatory activities are characterised by unsubstituted scaffolds on N1 and often bear bulky aromatic rings on C4. In detail, a series of 4,5-diaryl-3APs were synthesised using Combrestatin A-4 (a well-known microtubule inhibitor) as the lead compound. The new 3APs were tested for their cytotoxic activity in vitro against five human cancer cell lines (i.e., K562, ECA-109, A549, SMMC-7721, and PC-3) and the different derivatives showed potent cytotoxicity against all the tested cell lines, with IC 50 values in the low micromolar range. Compound 5 was identified as the most interesting derivative, with IC 50 values in the 0.08–12.07 mM range. Additional biological tests indicated that Compound 5 was a potent inhibitor of tubulin polymerisation, arresting the cell cycle in the G2/M phase. In addition, in order to investigate the binding pose of this 3AP to the colchicine binding site, docking simulations were carried out using the crystal structure of the tubulin–colchicine complex. This study evidenced that: ✓ The trimethoxyphenyl moiety and 4-methoxyphenyl moiety of Compound 5 are positioned in the hydrophobic pocket between Alaβ250-Alaβ316 and Valα181-Metβ259, respectively; ✓ The trimethoxyphenyl moiety is situated in close proximity to Cysβ241; ✓ The oxygen atom of the 4-methoxy substituent forms a hydrogen bond with the thiol group of Cysβ241; ✓ The pyrazole NH group establishes one hydrogen bond with the Alaβ250 backbone NH functionality; ✓ The hydrogen atom of 3-NH 2 forms another hydrogen bond with the NH of Asnα101. Collectively, these results confirmed the experimental data and demonstrated that Compound 5 could represent an interesting chemotype for anticancer activity, suggesting that the 4,5-diaryl-3APs could be considered as Combrestatin A-4 mimetics . Simultaneously, other researchers, using different synthetic approaches, synthesised and evaluated the antitumor and antioxidant activity of some novel pyrazolo-triazines to verify their pharmacological activity . All compounds were evaluated for their in vitro anticancer effect with the standard MTT method against a panel of four human tumour cell lines, including hepatocellular carcinoma (HepG2), lung fibroblasts (WI 38), and breast cancer (MCF-7) with 5-Fluorouracil (5-FU) used as the reference compound. In addition, all derivatives were tested to evaluate their cytotoxicity against a well-known established model of Ehrlich ascites cells (EAC) in vitro and for their antioxidant activity . Within this library of pyrazolo-triazines, 3AP 6 , used as an intermediate for the synthesis of different pyrazolo-triazines, showed IC 50 values ranging from 73 to 84 mg/mL against different cancer cell lines as well as some antioxidant activity. In conclusion, the authors demonstrated that the AP moiety enhances the antioxidant properties of different heterocycle systems . Differently, other authors from deCODE Chemistry (Chicago, IL, USA), during the development of variously substituted 1,3,4-oxadiazole derivatives to obtain anti-proliferative and antimitotic agents with microtubule destabilising activities, reported that the introduction of a 1,3,4-oxadiazole scaffold on a 3AP substituent is detrimental for anticancer activity, as demonstrated by biological activity Compound 7 ( , EC 50 values > 50 mM) . In 2008, Japanese researchers reported a 3AP linked to a nucleoside analogue, with the purpose of obtaining a triplex-forming oligonucleotide (named TFOs) able to bind the major groove of the DNA duplex and therefore identify novel genomic tools. Nucleoside-3AP 8 was able to recognise the CG interrupting site, but additional studies are necessary to validate its pharmacological activity . 3APs have been also studied as inhibitors of different kinases involved in the inflammatory process, obtaining interesting results. In fact, in 2010, researchers from Novartis Institutes (Basel, Switzerland) applied a scaffold hopping strategy and identified some differently substituted 3APs 9 as MK2-inhibitors. MK2 is a direct downstream kinase substrate of p38 mitogen-activated protein kinases (MAPKs) that plays a crucial role in the signalling and synthesis of TNFa, having a central role in inflammation and auto-immune diseases. The new derivatives also were shown to inhibit the intracellular phosphorylation of HSP27 and the LPS-induced TNFa release in cells. 3AP 9a , bearing an additional indole moiety on the N1 pyrazole, emerged as the most active compound of the series. Furthermore, the compound also inhibited LPS-induced TNFa release in mice and X-ray crystallography studies of the MK2/ 9 complex evidenced an unusual binding conformation with the indole ring inserted in a new ligand-induced hydrophobic pocket behind the MK2-hinge region. In detail, the indole-NH group forms an additional hydrogen bond interaction with the backbone carbonyl of Phe90, highlighting the importance of a hydrogen donor at this position. As expected, the 3AP moiety binds to the hinge region of MK2. All these interactions seem to improve the binding to MK2 and also kinase selectivity. In a summary, the 3AP scaffold was confirmed as a new interesting portion for the identification of novel MK2-inhibitors able to treat different inflammatory disorders . Ten years later, 3APs were studied as inhibitors of GSK-3β, a serine/threonine kinase that plays critical roles in multiple cellular functions in the central nervous system and is considered a key player in Alzheimer’s disease (AD) pathophysiology. Most synthesised 3APs possess GSK-3β inhibitory activities, with IC 50 values in the micromolar ranges, and bisindole-substituted 3AP 10 displayed moderate GSK-3β inhibition (IC 50 = 1.76 ± 0.19 μM). Furthermore, the compounds influenced LPS-induced glial inflammation in BV-2 cells and glutamate-induced oxidative neurotoxicity in HT-22 cells. Additional in vivo studies confirmed the anti-inflammatory effect of 10 that proved to reduce microglial activation and astrocyte proliferation in the brain of LPS-injected mice. Overall, this study evidenced specifically that bisindole-substituted 3AP could be a useful prototype for the discovery of novel therapeutic agents to tackle AD and other GSK-3β-associated complex neurological syndromes . Finally, more recently, Lusardi and coworkers developed a new regioselective procedure to synthesise novel highly functionalised 3APs 11 . All the prepared derivatives were tested by an MTT assay on a panel of tumour cell lines and against normal fibroblast to evaluate their antiproliferative activity. Methyl 3-amino-5-[(2-nitrophenyl)amino]-1 H -pyrazole-4-carboxylate (compound 11a , ) displayed good inhibition of the proliferation on HepG2 (liver cancer cells) and HeLa (cervical cancer cells) with mean growth percentage values of 54.25% and 38.44%, respectively. On the other hand, the compound was inactive against normal fibroblasts GM-6114 (growth percentage = 80.06%), showing no toxicity to healthy cells. Interestingly, inserting different alkyl and aryl moieties at position N1 of the pyrazole led to a loss of antiproliferative activity against all tested cell lines . This evidence highlighted the importance of unsubstituted N1 nitrogen for the cytotoxic/antiproliferative activity of this class of compounds, as demonstrated by the activity of compounds 5 – 7 , and 11 . On the other hand, when a more embedded substitute was inserted in this position, as in Compound 9 , anti-inflammatory activity was reported. Regarding anti-infective activity, 3APs have been investigated as antibacterial and antiviral agents; in some cases, the 3AP scaffold has been incorporated into a more complex structure. In detail, Delpe-Acharige and coll. reported a series of 3APs bearing a thiourea moiety in N1 (compounds 1 , ) with sub-micromolar activity against Methicillin-sensitive Staphylococcus aureus (MSSA) and Methicillin-resistant S. aureus (MRSA) in the presence of bioavailable copper. This study clearly established the suitability of using pyrazolyl thioureas for the treatment of these types of infections . In 2020, some authors from Roche Pharma Research and Early Development identified a small library of pyrido[2,3- b ]indole 2 able to block Gram-negative strains, targeting both DNA Gyrase and Topoisomerase IV. Among all the synthesised compounds, derivative 2a , characterised by a 3AP substituent on a pyrido-indole scaffold, exhibited MIC values of 0.125 and 8 mg/mL against S. aureus and E. coli , respectively . In the same year, Fahim and coll. synthesised a series of 3APs as intermediates for the synthesis of novel pyrazolo[1,5- a ]pyrimidine derivatives. All the newly isolated compounds were screened on a Well Diffusion Assay to evaluate their antimicrobial activity against three bacteria strains ( B. subtilis , S. pneumoniae , and E. coli ) and three fungi ( A. flavus , S. racemosum , and G. candidum ). The 3APs 3a – d showed high activity against both bacteria and fungi strains (inhibition zone diameter > 15 mm), highlighting the promising properties of this class of compounds as antibacterial agents . In 2009, researchers from TIBOTEC (France), identified some 4-aryloxy-3-iodopyridin-2(1 H )-ones that have been evaluated as anti-HIV inhibitors. The authors investigated different substituents at the 5-position of this scaffold, including one compound characterised by a 3AP moiety in the 4 position (derivative 4 , ) which showed potent HIV-1 reverse transcriptase inhibitory properties against both the wild-type enzyme and simple/double mutated forms . The most active 3AP compounds with antitumour and anti-inflammatory activities are characterised by unsubstituted scaffolds on N1 and often bear bulky aromatic rings on C4. In detail, a series of 4,5-diaryl-3APs were synthesised using Combrestatin A-4 (a well-known microtubule inhibitor) as the lead compound. The new 3APs were tested for their cytotoxic activity in vitro against five human cancer cell lines (i.e., K562, ECA-109, A549, SMMC-7721, and PC-3) and the different derivatives showed potent cytotoxicity against all the tested cell lines, with IC 50 values in the low micromolar range. Compound 5 was identified as the most interesting derivative, with IC 50 values in the 0.08–12.07 mM range. Additional biological tests indicated that Compound 5 was a potent inhibitor of tubulin polymerisation, arresting the cell cycle in the G2/M phase. In addition, in order to investigate the binding pose of this 3AP to the colchicine binding site, docking simulations were carried out using the crystal structure of the tubulin–colchicine complex. This study evidenced that: ✓ The trimethoxyphenyl moiety and 4-methoxyphenyl moiety of Compound 5 are positioned in the hydrophobic pocket between Alaβ250-Alaβ316 and Valα181-Metβ259, respectively; ✓ The trimethoxyphenyl moiety is situated in close proximity to Cysβ241; ✓ The oxygen atom of the 4-methoxy substituent forms a hydrogen bond with the thiol group of Cysβ241; ✓ The pyrazole NH group establishes one hydrogen bond with the Alaβ250 backbone NH functionality; ✓ The hydrogen atom of 3-NH 2 forms another hydrogen bond with the NH of Asnα101. Collectively, these results confirmed the experimental data and demonstrated that Compound 5 could represent an interesting chemotype for anticancer activity, suggesting that the 4,5-diaryl-3APs could be considered as Combrestatin A-4 mimetics . Simultaneously, other researchers, using different synthetic approaches, synthesised and evaluated the antitumor and antioxidant activity of some novel pyrazolo-triazines to verify their pharmacological activity . All compounds were evaluated for their in vitro anticancer effect with the standard MTT method against a panel of four human tumour cell lines, including hepatocellular carcinoma (HepG2), lung fibroblasts (WI 38), and breast cancer (MCF-7) with 5-Fluorouracil (5-FU) used as the reference compound. In addition, all derivatives were tested to evaluate their cytotoxicity against a well-known established model of Ehrlich ascites cells (EAC) in vitro and for their antioxidant activity . Within this library of pyrazolo-triazines, 3AP 6 , used as an intermediate for the synthesis of different pyrazolo-triazines, showed IC 50 values ranging from 73 to 84 mg/mL against different cancer cell lines as well as some antioxidant activity. In conclusion, the authors demonstrated that the AP moiety enhances the antioxidant properties of different heterocycle systems . Differently, other authors from deCODE Chemistry (Chicago, IL, USA), during the development of variously substituted 1,3,4-oxadiazole derivatives to obtain anti-proliferative and antimitotic agents with microtubule destabilising activities, reported that the introduction of a 1,3,4-oxadiazole scaffold on a 3AP substituent is detrimental for anticancer activity, as demonstrated by biological activity Compound 7 ( , EC 50 values > 50 mM) . In 2008, Japanese researchers reported a 3AP linked to a nucleoside analogue, with the purpose of obtaining a triplex-forming oligonucleotide (named TFOs) able to bind the major groove of the DNA duplex and therefore identify novel genomic tools. Nucleoside-3AP 8 was able to recognise the CG interrupting site, but additional studies are necessary to validate its pharmacological activity . 3APs have been also studied as inhibitors of different kinases involved in the inflammatory process, obtaining interesting results. In fact, in 2010, researchers from Novartis Institutes (Basel, Switzerland) applied a scaffold hopping strategy and identified some differently substituted 3APs 9 as MK2-inhibitors. MK2 is a direct downstream kinase substrate of p38 mitogen-activated protein kinases (MAPKs) that plays a crucial role in the signalling and synthesis of TNFa, having a central role in inflammation and auto-immune diseases. The new derivatives also were shown to inhibit the intracellular phosphorylation of HSP27 and the LPS-induced TNFa release in cells. 3AP 9a , bearing an additional indole moiety on the N1 pyrazole, emerged as the most active compound of the series. Furthermore, the compound also inhibited LPS-induced TNFa release in mice and X-ray crystallography studies of the MK2/ 9 complex evidenced an unusual binding conformation with the indole ring inserted in a new ligand-induced hydrophobic pocket behind the MK2-hinge region. In detail, the indole-NH group forms an additional hydrogen bond interaction with the backbone carbonyl of Phe90, highlighting the importance of a hydrogen donor at this position. As expected, the 3AP moiety binds to the hinge region of MK2. All these interactions seem to improve the binding to MK2 and also kinase selectivity. In a summary, the 3AP scaffold was confirmed as a new interesting portion for the identification of novel MK2-inhibitors able to treat different inflammatory disorders . Ten years later, 3APs were studied as inhibitors of GSK-3β, a serine/threonine kinase that plays critical roles in multiple cellular functions in the central nervous system and is considered a key player in Alzheimer’s disease (AD) pathophysiology. Most synthesised 3APs possess GSK-3β inhibitory activities, with IC 50 values in the micromolar ranges, and bisindole-substituted 3AP 10 displayed moderate GSK-3β inhibition (IC 50 = 1.76 ± 0.19 μM). Furthermore, the compounds influenced LPS-induced glial inflammation in BV-2 cells and glutamate-induced oxidative neurotoxicity in HT-22 cells. Additional in vivo studies confirmed the anti-inflammatory effect of 10 that proved to reduce microglial activation and astrocyte proliferation in the brain of LPS-injected mice. Overall, this study evidenced specifically that bisindole-substituted 3AP could be a useful prototype for the discovery of novel therapeutic agents to tackle AD and other GSK-3β-associated complex neurological syndromes . Finally, more recently, Lusardi and coworkers developed a new regioselective procedure to synthesise novel highly functionalised 3APs 11 . All the prepared derivatives were tested by an MTT assay on a panel of tumour cell lines and against normal fibroblast to evaluate their antiproliferative activity. Methyl 3-amino-5-[(2-nitrophenyl)amino]-1 H -pyrazole-4-carboxylate (compound 11a , ) displayed good inhibition of the proliferation on HepG2 (liver cancer cells) and HeLa (cervical cancer cells) with mean growth percentage values of 54.25% and 38.44%, respectively. On the other hand, the compound was inactive against normal fibroblasts GM-6114 (growth percentage = 80.06%), showing no toxicity to healthy cells. Interestingly, inserting different alkyl and aryl moieties at position N1 of the pyrazole led to a loss of antiproliferative activity against all tested cell lines . This evidence highlighted the importance of unsubstituted N1 nitrogen for the cytotoxic/antiproliferative activity of this class of compounds, as demonstrated by the activity of compounds 5 – 7 , and 11 . On the other hand, when a more embedded substitute was inserted in this position, as in Compound 9 , anti-inflammatory activity was reported. According to the literature data so far available, 4AP compounds showed reduced anti-inflammatory and anticancer activity in comparison with their 3AP and 5AP isomers but attracted some attention as anticonvulsants, colouring, and antioxidant agents. Due to the variety of these biological properties, SAR considerations are difficult to define. In detail, since 1970, molecules with a pyrazole structure carrying the amino group in position 4 (4APs) have been studied for their interesting anticonvulsant properties. Compound 12 is characterised by a methyl group at position 3 and by a substituted phenyl ring at position 5; research evidenced some anticonvulsant activity, and the authors demonstrated the importance of the distance between the exocyclic NH 2 group and the endocyclic NH for this type of pharmacological activity . More recently, interesting results have been obtained for similar 4AP analogues tested as anticonvulsant agents . Different 4APs, characterised by an additional pyrazole nucleus on the N1 position ( 13 , ), were patented as keratin dyeing , evidencing the importance of this scaffold in industry as well. 4APs were also prepared as intermediates in the synthesis of a large library of thiourea and urea derivatives as anticonvulsant agents. Unfortunately, Compound 14 and related urea and thiourea derivatives were shown to be poorly active in pentylenetetrazole-induced seizure (PTZ) and maximal electroshock tests. At the same time, anti-HIV pharmacological evaluation was also carried out, obtaining negative results . However, the interest in this scaffold has grown in recent years, highlighting the different antioxidant properties of this chemotype. In 2020, a simple and efficient method for the synthesis of different pyrazole derivatives 15 , including 4APs bearing various substituents at positions 1 and 5, was developed. The prepared compounds were tested in vitro for their tuberculostatic, antibacterial, antimycotic, antioxidant, and cytotoxic activities; additionally, the analgesic and anti-inflammatory properties of the compounds were evaluated in vivo. According to the calculated ADME parameters, all obtained pyrazole derivatives evidenced favourable pharmacokinetic profiles. In detail, synthesised compounds revealed multiple pharmacological activities depending on the nature of their peripheral substituents at the pyrazole core. 4APs 15 did not show antibacterial and antimycotic activity but revealed anticancer activity against HeLa cells and moderate toxicity on human dermal fibroblasts (HDF). Moreover, the N-unsubstituted 4APs and their hydrochlorides showed good antioxidant properties whereas their N-substituted analogues were inactive or significantly less active. In addition, the introduction of thienyl moiety at position 5 significantly increased the acute toxicity of 4APs. The best analgesic activity was evidenced for 4APs having a phenyl fragment at position 5. Interestingly, among all the series of tested compounds, only 4-amino-5-phenylpyrazoles had appreciable anti-inflammatory activity. The developed SARs profile pointed to the 4-amino-3-trifluoromethyl-5-phenylpyrazoles as the lead structure for the development of new pharmacologically active compounds . More recently, the same authors reported the antioxidant activity of different Edaravone analogues characterised by the 4AP scaffold ( 16 , ). These derivatives, obtained by the reduction of 4-hydroxyiminopyrazol-5-ones, showed pronounced antioxidant activity in different assays (namely, the ABTS, FRAP, and ORAC tests), with the 4-amino-3-methyl-1-phenylpyrazol-5-ol hydrochloride ( 16a , ) derivative being the most active. Additional investigations confirmed the promising properties of 16 , which was proposed as a lead structure for developing novel therapeutic drug candidates for treating oxidative stress-related diseases. Additional chemical modifications, such as conjugation to an anticholinesterase fragment, could be performed to obtain multifunctional drugs for treating neurodegenerative diseases . The substitution of the amino group in position 5 of the pyrazole ring is largely reported in the literature due to the high versatility of this class of compounds in the medicinal chemistry field. As reported below, these derivatives have been used as kinase inhibitors (particularly p38MAPK and Bruton kinase inhibitors), anticancer, antibacterial, antimalarial, and anti-inflammatory agents. 4.1. p38 Inhibitors 5APs The MAPK family regulates a variety of cellular responses such as proliferation, differentiation, gene expression, cell survival, and apoptosis through the transduction of extracellular signals . p38, together with JNK and ERK, belongs to the MAPK family, and it is mainly involved in the regulation of pro-inflammatory cytokines such as TNF-a, IL-1, and IL-6 . For all these reasons, p38MAPK inhibitors have been largely studied as anticancer and anti-inflammatory agents. Particularly, 5-pyrazolyl ureas have been widely studied as p38MAPK inhibitors, but unfortunately, their therapeutic use is precluded by different side effects and ineffectiveness in clinical studies . For these reasons, the urea moiety on position 5 has been deleted and different 5APs have been studied. In 2006, Goldstein and coworkers designed and synthesised a new series of highly selective p38MAPK inhibitors with a 5-amino- N -phenyl-1 H -pyrazol-4-yl-3-phenylmethanones scaffold. X-ray crystallography of these new derivatives bound in the ATP binding pocket of unphosphorylated p38a was used to optimise the potency and physicochemical properties of the series. The addition of a 2,3-dihydroxypropoxy moiety on the C4 phenyl ring led to the isolation of compound RO3201195 , characterised by a higher selectivity on p38a and excellent drug-like properties, including high oral bioavailability. This 5AP molecule displayed an IC 50 of 0.7 ± 0.1 mM on p38a, good inhibition of TNFa production in a human monocytic cell line (THP1), and a reduction in IL-1b in a mononuclear cell fraction isolated from HWB. Compound RO3201195 was also tested in several acute inflammatory models in rats to evaluate its efficacy in vivo. The obtained results showed a significant dose-dependent inhibition of serum TNFa and IL-6 (IC 50 values of 0.2 and 0.3 mM, respectively) . A few years later, Bagley and coll. proposed a new synthetic method for the preparation of RO3201195 and tested the compound on Werner syndrome (WS) cells. High levels of phosphorylated p38a are expressed in proliferating WS cells, demonstrating that p38MAPK pathway inhibition could potentially interfere with the pathology. The results obtained in hTERT-immortalised HCA2 cells and primary WS cells treated with RO3201195 confirmed this hypothesis; in fact, this compound showed excellent selectivity for p38aMAPK over JNK and slowed down the accelerated aging of WS cells in culture . The same research group extended the SARs of previously synthesised 5APs and tested the newly obtained derivatives on human hTERT-immortalised HCA2 dermal cells to evaluate their ability to inhibit p38MAPK. The four compounds that displayed comparable or slightly improved potency over RO3201195 were tested on WS cells. Particularly, the 2,4-difluorophenyl substituent of compound 17a seemed to increase the activity compared to other derivatives, as well as the methyl substitution on N1 of compound 17b . These two molecules were selected as lead compounds but, despite their good p38a inhibition, they did not improve the efficacy and bioavailability of RO3201195 in in vivo evaluations . Other experiments were carried out on WS cells with RO3201195 and its analogues 17a and 17b . The results definitively confirmed that the proliferation of pathological cells is linearly correlated to p38 phosphorylation and that these 5APs significantly interfere with the protein expression and consequently with cell growth. Furthermore, the derivatives showed a better selectivity on p38 in comparison with the reference compound SB203580 which seemed to interfere in the same way with both p38 and JNK MAPK . In 2016, the screening of a DNA-encoded small molecule library allowed the identification of the highly specific and potent (IC 50 = 7 ± 0.9 nM) p38aMAPK inhibitor VPC00628 . This compound shares the N1 phenyl ring with RO3201195 and bears an amide function (rather than a keto group) on the C4 position. X-ray crystallography studies (PDB code: 5LAR) indicated that VPC00628 interacts with the ATP binding site of p38aMAPK, inducing a strong distortion of the P-loop. Interestingly, the ligand assumes an alternative binding mode as it lacks the key features of known kinase inhibitors such as a typical hinge binding motif. VPC00628 showed excellent shape complementarity and formed several specific polar interactions , assuming a canonical inactive type-II (‘DFG-out’) binding mode. This specific interaction with the inactive form of the kinase seems fundamental to increasing the potency and the selectivity of the inhibitor . Röhm and coworkers synthesised a new series of VPC00628 analogues, trying to improve the pharmacological properties of the parent compounds. Using a systematic combinatorial synthetic approach, they isolated and identified SR-318 , a 5AP characterised by a more hydrophobic amide moiety. The crystal structure of the SR-318/p38MPAK complex (PDB code: 6SFO; ) provided the structural basis for the excellent potency and selectivity for p38a/b (IC 50 a = 5 nM and IC 50 b = 32 nM) of the identified derivative. Moreover, SR-318 showed a better metabolic degradation profile in comparison with the reference drug VPC00628. To test the in vitro efficacy of SR-318, the LPS-stimulated TNF-a release in whole blood was determined. The results showed that this 5AP had an inhibition value of 97.7% at 10 mM (IC 50 = 0.283 mM) in the assay, resulting in more effectiveness than the literature-known compounds . Further studies focused on the exploration of the p38MAPK aC-out pocket showed that the hinge-binding motif of VPC00628 and SR-318 greatly enhanced the inhibitory activity compared with previously synthesised pyrazolyl-ureas. SARs extension of these reference derivatives led to the identification of Compound 18 as a selective type-II inhibitor. Despite the moderate activity of the enzyme (IC 50 = 14 nM), the crystallographic data provided valuable insights into the back-pocket interactions that were not observed in the SR-318/p38MPAK complex, thus indicating 5AP 18 as an alternative chemical tool with good cellular activity (PDB code: 6YK7; ). After demonstrating the excellent selectivity of 18 , human colon adenocarcinoma (HCT-15) cells were used to probe its efficiency on p38MPAKs. The Western blot analysis displayed the dose-dependent inhibition of p38 phosphorylation and increased phosphorylation of its downstream substrate HSP27. Furthermore, the new compound significantly inhibited TNF-a release with an IC 50 value of 0.48 mM, a better result than the one obtained with the type-I inhibitor . 4.2. Anticancer/Antiproliferative 5APs 5APs show very interesting anticancer properties; for some compounds, the intracellular targets have been identified (e.g., Bruton kinase for the recently approved Pirtobrutinib) while for other derivatives, only the antiproliferative activity on different cancer cell lines or additional intracellular mechanism have been reported (see table in ). In 2015, Ibrahim and coworkers prepared a new series of benzenesulphonamide derivatives incorporating pyrazole and isatin moieties. A biological evaluation was carried out to assess the ability of the compounds to inhibit the metalloenzyme carbonic anhydrase (CA) and more precisely the human (h) isoforms hCA I, II (cytosolic), IX, and XII (transmembrane, tumour-associated enzymes). The 15 hCA isoforms are widely distributed within different tissues and are involved in many physiological and pathological conditions. In particular, the selective inhibition of hCA IX and XII produces significant antitumour and antimetastatic effects . Derivatives 19a and 19b inhibited hCA XII with a Ki of 5.4 nM and 7.2 nM, respectively. In particular, pyrazole 19a with a 5-NO 2 substitution on the isatin ring was found to be a selective inhibitor of hCA IX and hCA XII, being more active than the acetazolamide used as the reference drug. A docking simulation confirmed that the NO 2 substituent present on 19a participates in interactions with Asp132 within the hCA IX active site and with the Lys67 and Asp130 residues in hCA XII . With the aim of identifying new chemical entities which can block the NF-kB cascade, Pippione, and coll. designed and synthesised a new series of 5APs that were biologically evaluated on four kinases involved in the NK-kB pathway (namely, IKKa, IKKb, IKKe, and NIK). In detail, 5APs 20a – c selectively inhibited NIK with an IC 50 of 8.4, 2.9, and 3.3 nM, respectively. A gene-reported assay was used to measure NK-kB activation in human multiple myeloma EJM cells, constitutively characterised by high levels of nuclear NK-kB related to NIK activation, and in breast cancer cell lines (SKBr3 and MDA-MB-231) in which the constitutive activation of nuclear NK-kB is unrelated to NIK. Compounds 20a – c seemed to inhibit the NK-kB activity in EJM cells (83.4–96.2% of inhibition at 25 mM), but not in SKBr3 and MDA-MB-231 cells, confirming their selective inhibitory activity for NIK . In 2020, Hassan and his research group designed and synthesised a novel library of twenty amino-pyrazoles, pyrazolo-pyrimidines, and their fused analogues. The compounds were tested by the National Cancer Institute (NCI, Bethesda, Maryland, USA) at a fixed concentration of 10 mM on a panel of 60 different human cancer cell lines. 5-amino-1-((4-chlorophenyl)(1-hydroxy-3,4-dihydronaphthalen-2-yl)methyl)-1 H -pyrazole-4-carbonitrile (compound 21 , ) showed the best cytotoxic activity with a cell proliferation inhibition higher than 90% on NCI-H23 (non-small cell lung cancer), HCT-15 (colon cancer), SF-295 (CNS cancer), NCI/ADR-RES (ovarian cancer), and DU-145 (prostate cancer) cells. As cyclooxygenase-2 (COX-2) inhibitors were largely reported to have antiproliferative activity against various cancer cells , docking simulations of 21 on COX-2 were carried out. Interestingly, the derivative displayed an affinity value of −7.86 kcal/mol and exhibited three hydrogen bonds with Lys137 and Gly45 side chains in the COX-2 active site . Looking for new cytotoxic derivatives, Anwer and Sayed prepared a new series of heterocyclic compounds through microwave reactions. All the synthesised molecules were tested by an MTT assay at different concentrations (1–100 mg/mL) on breast cancer (MCF-7) and colorectal carcinoma (HCT-116) cells. Among the compounds of the library, 5AP 22 (3-(4-(dimethylamino)phenyl)-1-phenyl-4-(1 H -tetrazol-5-yl)-1 H -pyrazol-5-amine, ), bearing an additional tetrazole substituent on C4, displayed the best cytotoxic activity with an IC 50 of 3.18 mM on HCT-116 and 4.63 mM on MCF-7 . In the same year and for the same purpose, Fadaly and coworkers isolated a novel series of triazole/pyrazole hybrids analogues (compounds 23 , ) of COX-2 inhibitor Celecoxib endowed with a large spectrum of activity. The reported compounds were also tested by MTT assay for their antiproliferative activity on MCF-7, HCT-116, A549 (human lung cancer cell line), PC-3 (human prostate cancer cells), and F180 normal fibroblasts. The sulphamoyl derivatives 23a and 23b were identified as the most active of the series with IC 50 of 4.22 and 6.38 μM on A549, 5.33 and 3.67 μM on MCF-7, 3.46 and 2.28 μM on HCT-116, and 1.48 and 0.33 μM on PC-3, respectively. An investigation of the mechanism of action of these molecules revealed that they were able to block the cell cycle at the G2/M phase, with a downregulation of Bcl-2 gene expression and an up-regulation of Bax expression, evidencing a pro-apoptotic mechanism. Docking simulation studies indicated the Epidermal Growth Factor Receptor (EGFR) as a potential target of 23a and 23b whose oxime functionality would form two hydrogen bonds with Thr830A and Asp831A. Enzymatic and ELISA assays confirmed the ability of the compounds to interfere with p38MAPK and VEGFR-2 signalling pathways . Very recently, Eli Lilly launched the new 5AP Pirtobrutinib (Jaypirca™, ), approved to treat mantle cell lymphoma (MCL), on the market . In detail, this (S)-5-amino-3-(4-((5-fluoro-2-methoxybenzamido)methyl)phenyl)-1-(1,1,1-trifluoropropane-2-yl)-1 H -pyrazole-4 carboxamide is a reversible inhibitor of Bruton Kinase (BTK), a nonreceptor tyrosine kinase, that represents a major therapeutic target for B-cell-driven malignancies. Different from previously approved covalent BTK inhibitors, innovative reversible BTK inhibitors, such as Pirtobrutinib, are characterised by limited off-target side effects and avoid the development of resistance mutations. These types of reversible BTK inhibitors have aroused great interest not only for the treatment of B cell malignancies but also for counteracting many autoimmune diseases . Pirtobrutinib is currently being studied in 16 clinical trials (1 completed and 5 in phase III) for chronic lymphocytic leukaemia (CLL), small lymphocytic lymphoma (SLL), and mantle cell lymphoma (MCL) . 4.3. Antibacterial 5APs In the last twenty years, a rise in the level of antimicrobial resistance among pathogens occurred due to the emergence of resistant strains of pathogenic bacteria and the increase in the number of antibiotics administered. Therefore, the development of new antibacterial compounds has become an urgent and mandatory worldwide need. For this purpose, 5APs were studied and showed more interesting results than their 3APs isomers (as compounds 2 , 3 ). As reported below, in some cases, different 5APs revealed dual activity (anticancer and antibacterial action). In 2010, Gouda and coworkers prepared a novel series of heterocyclic derivatives in which 4,5,6,7-tetrahydro-benzo[ b ]thiophene-3-carboxamide scaffold was linked to pyrazole, pyrazolo-pyridine, pyrazolo-pyrimidine, or pyrazolo-triazine core (Compound 24 and used the agar diffusion technique to evaluate the antibacterial activity of the compounds on two bacterial strains ( B. theringiensis and K. pneumoniae ) and two fungi ( B. fabe and F. oxysporum ). 5AP 24a showed a larger inhibition zone diameter on Gram-positive B. theringiensis (24 mm) and Gram-negative K. pneumoniae (22 mm) than the reference drug Ampicillin (17 mm and 20 mm, respectively) . In 2012, Behbehani and coworkers utilised 2-aminothiophenes as building blocks for the synthesis of new pyrazole, pyrimidine, quinoline, and pyridine-2-one derivatives. The isolated compounds were tested and evaluated as antimicrobial agents on a large panel of Gram-positive and Gram-negative bacteria strains and two fungi by well diffusion assay. 5AP 25 exhibited good activity against Gram-positive B. subtilis with an inhibition zone diameter of 7.3 ± 1.1 mm (Penicillin = 4.6 ± 1.1 mm). The compound also displayed moderate activity against fungi, such as C. albicans (6.6 ± 1.1 mm) and S. serevisiae (4 mm) . Another class of 5APs with antibacterial properties was synthesised by Al-Adiwish in 2013. The antibacterial activity of the newly isolated compounds was tested using the agar diffusion technique at a concentration of 1 mg/mL on Gram-positive ( Staphylococcus aureus , Bacillus subtilis , Methicillin-resistant S. aureus , Staphylococcus epidermidis , and Enterococcus faecalis ) and Gram-negative ( Escherichia coli , Pseudomonas aeruginosa , Serratia marcescens , and Salmonella typhimurium ) bacteria. 26 exhibited the best activity against S. aureus and E. coli (15 ± 0.58 mm and 15 ± 0.56 mm, respectively). The minimum inhibitory concentration (MIC) and minimum bactericidal concentration (MBC) were also determined: 26 showed a MIC of 16 mg/mL on S. aureus and 4 mg/mL on E. coli , whereas the MBC value results were 16 mg/mL on S. aureus and 8 mg/mL on E. coli . The derivative was also tested by an MTT assay on Vero cells to evaluate its in vitro cytotoxicity. Unfortunately, despite the safety of the compound (CC 50 > 20 mg/mL), the selective index was too low to assess the selectivity of the molecule on the bacteria strains . In 2018, Eldin reported the preparation, characterisation, and antimicrobial activity of new amphiphilic pyrazole-g-polyglycidyl methacrylate-based polymers. The goal of this study was to synthesise a new class of antimicrobial copolymers containing pendant amino groups in combination with hydrophobic residues to obtain new antimicrobial agents against a variety of clinically significant pathogens. In detail, 5AP 27 , characterised by an azo-substituent on the C4 position and devoid of antibacterial activity, was grafted to the epoxy rings of polyglycidyl methacrylate and poly (glycidyl methacrylate-co-methyl methacrylate) copolymers. The structure verification of the new polymers was performed using FT-IR and TGA analyses and new derivatives were tested against different Gram-positive and Gram-negative bacteria strains, showing various activity, particularly against Gram-negative pathogens. Pyrazole-g-polyglycidyl methacrylate showed the most potential antibacterial activities, followed by pyrazole-g-poly (glycidyl methacrylates-co-methyl methacrylate) copolymers. All these considerations open a new area for developing novel pyrazole-based antimicrobial agents . Selected compounds from a heterocycle library synthesised by Anwer, including previously mentioned cytotoxic derivatives 22 and 5APs 28 and 29 displayed good antibacterial properties against both Gram-negative ( E. coli and P. aeruginosa ) and Gram-positive ( B. subtilis and S. aureus ) bacteria strains. Taking Ampicillin as a standard reference antibiotic, the activity index for each species was calculated. Pyrazole 22 exhibited the best results with activity index values ≥75% in all the analysed bacteria . Recently, using a sequential ligand-based pharmacophore-based virtual screening followed by structure-based molecular docking, Indian researchers identified some 5APs endowed with anti-tubercular potential. In detail, 5-amino-3-((substitutes phenylamino)- N -(4-substituted phenylamino)-1 H -pyrazole-4-carboxamides 30a – d showed potent anti-tubercular activity (MIC between 2.23 and 4.61 mM) compared to reference standard drugs and a lower cytotoxicity when tested on Vero cells. The preliminary SAR considerations suggested that the presence of the electron-withdrawing substituents at both R and R′ positions is beneficial for antibacterial activity, while electron-donating groups caused a remarkable reduction in potency. In addition, molecular docking studies suggested InhA, an enzyme involved in fatty acid synthesis and target for the development of novel antitubercular agents, as a possible target for the most potent derivatives . 4.4. Antimalarial 5APs The increasing prevalence of drug-resistant Plasmodium falciparum strains led to an urgent priority in the development of new antimalarial drugs. In this context, Dominguenz and coworkers, utilising a previously reported synthetic methodology, prepared a new series of APs with potential antimalarial activity. The activity of the novel derivatives was tested in vitro against Plasmodium falciparum. 5APs 31a and 31b were the most active of the series with an IC 50 of 0.149 mM and 0.150 mM, respectively. From a chemical point of view, the presence of the ester functionality in position 4 of the pyrazole ring seemed to be essential for the activity; indeed, its substitution with a nitrile group led to a loss of inhibition. Interestingly, the electronic effect of the substitution in the phenyl ring seemed to be important; in fact, the substitution with the methoxy group in the meta position increased the activity of the compounds, while the same group in position ortho or para resulted in a progressive loss of potency against the plasmodium . A few years later, Verma and coworkers synthesised a series of pyrazole derivatives closely related to 31 , with an ethyl ester instead of a methyl one. All reported 5APs were tested in vitro against a chloroquine-resistant strain (FCBI) of Plasmodium falciparum . Not surprisingly, derivatives 32a and 32b displayed the best IC 50 with values of 0.149 mM and 0.150 mM, respectively, confirming the pivotal role of ester functionality and the meta substitution on the phenyl ring on C3 of the pyrazole ring. These two 5APs proved to be more effective than Pentamidine against Leishmania donovani , showing IC 50 values of 0.132 mM and 0.168 mM, respectively. Interestingly, regarding pharmacological activity, the ortho and para substitution led to a loss of activity as previously evidenced for anti-Plasmodium action . 4.5. Anti-Inflammatory 5APs In an effort to discover new selective COX-2 inhibitors without the ulcerogenic side effects of non-steroidal anti-inflammatory drugs (NSAIDs), Abdellatif and coworkers designed and prepared a new class of pyrazole derivatives possessing an amino or mehansulphonyl pharmacophore. All the compounds were preliminary tested against COX-1 and COX-2 by an in vitro colorimetric enzyme immunoassay (EIA). Furthermore, the COX-2 selectivity index (SI) was calculated [IC 50 (COX-1)/IC 50 (COX-2)] and compared to the reference drugs Indomethacin (non-selective COX inhibitor) and Celecoxib (selective COX-2 inhibitor). 5APs 33 and 34 , characterised by two linked 5AP portions, exhibited the best results with an IC 50 on COX-2 of 39 nM and 34 nM, comparable to Celecoxib (IC 50 = 45 nM), and a SI of 353.8 and 417.6, respectively. The in vivo carrageenan-induced rat paw oedema assay confirmed the in vitro data for 33 and 34 , orally administrated at a concentration of 50 mg/Kg. In addition, 33 and 34 displayed an oedema inhibition percentage of 96% and 87% after 5h, resulting in more effectiveness than the reference drug Celecoxib. In in vivo tests, the new derivatives caused a reduced number of ulcers when compared to Indomethacin and Celecoxib, with an ulcer index (UI) between 0.7 and 2 (UI Celecoxib = 2.7 and UI Indomethacin = 21.3), confirming their pharmaceutical attractiveness. A molecular docking study showed H-bond interactions between the SO 2 groups of 33 and 34 and Phe504, Arg499, Tyr341, Ser516, and Arg106 amino acids, confirming the importance of the SO 2 substituent on the 5AP scaffold to obtain COX-2 inhibition . The following year, the same research group isolated a group of Celecoxib analogues linked to oxime moiety as nitric oxide donors, previously mentioned as anticancer agents . The aim of the project was to synthesise novel anti-inflammatory NO-NSAID hybrids with selectivity for COX-2 and with a NO-donor moiety. 5APs 35a and 35b showed higher activity in an in vitro COX-2 colorimetric assay (IC 50 = 0.55 mM and 0.61 mM, respectively) compared to Celecoxib (IC 50 = 0.83 mM). These two 5APs also displayed a good selectivity index (IC 50 COX-1/IC 50 COX-2) of 9.78 and 8.57, respectively, compared to 8.68 scored by the reference Celecoxib. To evaluate their in vivo anti-inflammatory activity, the derivatives were tested at a 50 mg/kg dose with the carrageenan-induced rat paw oedema assay. Pyrazole 35a exhibited the best oedema inhibition percentage (91.11%), overcoming again the values of Celecoxib (86.66%). The ulcerogenic assay highlighted the reduced ulcerogenic effects of the newly synthesised molecules (ulcer indexes UI = 2.79–3.95) in comparison with Ibuprofen (UI = 20.25) and Celecoxib (UI = 2.93). The docking simulation of 35a in the active site of COX-2 displayed that the compound would not form any H-bonds inside the COX-2 site. Furthermore, the percent of NO released from the derivatives was determined upon incubation in phosphate-buffered-saline. In detail, 35a and 35b exhibited a NO released percentage of 3.06% and 2.15%, respectively, which indicated a slow NO release. More recently, to obtain new hybrid compounds potentially able to act on different targets involved in inflammation onset, Brullo and coworkers designed and synthesised a series of pyrazole and imidazo-pyrazole derivatives with differently decorated catechol moieties linked through an acylhydrazone chain. The inhibitory effect of the novel isolated molecules on reactive oxygen species (ROS) production on platelets and neutrophils was evaluated. 5APs 36a , 36b , and 36c displayed the most promising results, with IC 50 values on ROS production inhibition in the low micromolar range with platelets, whereas derivative 36d showed a ROS production inhibition percentage of 68% on fMLP activated-neutrophils in flow cytometric analysis. The compound was also tested by enzymatic assay on phosphodiesterase enzymes PDE4B and PDE4D, intracellular enzymes mostly involved in neuroinflammation. The assay on 36d resulted in an IC 50 of 1.05 mM on PDE4D3 and an IC 50 of 0.55 mM on PDE4B2, confirming the potential antioxidant and anti-inflammatory activity of this new chemotype . Derivatives 36a , 36b , and 36c were further investigated to evaluate their effect on several parameters indicative of oxidative status and their efficiency on aerobic metabolism. All three molecules seemed to strongly inhibit superoxide anion production, lipid peroxidation, and NADPH oxidase activity and almost restored the oxidative phosphorylation efficiency in platelets stimulated with thrombin, highlighting their potential protective effect against oxidative stress. These results were confirmed in endothelial cells in which the selected compounds showed a promising inhibition activity on H 2 O 2 -stimulated EA.hy926 cells . 4.6. 5APs with Other Pharmacological Activities In 2001, Kordik and collaborators prepared a novel series of 1,3-disubstituted-5APs with an affinity for the human neuropeptide Y (NPY) receptor Y5. NPY represents a powerful stimulant of food intake, implicated in obesity and eating; consequently, the antagonist of the NPY receptor Y5 could potentially provide new treatments for eating disorders. The synthesised 5APs, bearing a sulphonamide moiety, were tested for their binding affinity to NYP receptor Y5, using a transfected HEK293 cell line and measuring the competitive inhibition binding of 125 I-PYY. Derivative 37 displayed the best affinity in the first screening with an IC 50 value of 15 nM . Variation of the sulphonamide group led to inactive compounds, while the modification of the phenyl linker with a cyclohexyl group yielded pyrazole 38 , which showed a higher affinity for NPY Y5 than Fipronil . In 2020, Hebishy and collaborators described a new synthetic strategy to obtain novel benzamide-based 5APs as precursors of pyrazolo[1,5- a ]pyrimidine and pyrazolo-[5,1- c ][1,2,4]triazine derivatives; the derivatives were then tested as anti-influenza A agents active on subtype H5N1. The plaque reduction assay and the MTT cytotoxicity assay revealed 5AP 39 as the most promising compound, with an inhibition percentage value of 66.67% at 0.125 mmol/mL concentration and an LD 50 of 30 mmol/mL. Even though 39 resulted in less activity than the reference drug Zanamivir, it showed better potency than other reported bicycle analogues . The MAPK family regulates a variety of cellular responses such as proliferation, differentiation, gene expression, cell survival, and apoptosis through the transduction of extracellular signals . p38, together with JNK and ERK, belongs to the MAPK family, and it is mainly involved in the regulation of pro-inflammatory cytokines such as TNF-a, IL-1, and IL-6 . For all these reasons, p38MAPK inhibitors have been largely studied as anticancer and anti-inflammatory agents. Particularly, 5-pyrazolyl ureas have been widely studied as p38MAPK inhibitors, but unfortunately, their therapeutic use is precluded by different side effects and ineffectiveness in clinical studies . For these reasons, the urea moiety on position 5 has been deleted and different 5APs have been studied. In 2006, Goldstein and coworkers designed and synthesised a new series of highly selective p38MAPK inhibitors with a 5-amino- N -phenyl-1 H -pyrazol-4-yl-3-phenylmethanones scaffold. X-ray crystallography of these new derivatives bound in the ATP binding pocket of unphosphorylated p38a was used to optimise the potency and physicochemical properties of the series. The addition of a 2,3-dihydroxypropoxy moiety on the C4 phenyl ring led to the isolation of compound RO3201195 , characterised by a higher selectivity on p38a and excellent drug-like properties, including high oral bioavailability. This 5AP molecule displayed an IC 50 of 0.7 ± 0.1 mM on p38a, good inhibition of TNFa production in a human monocytic cell line (THP1), and a reduction in IL-1b in a mononuclear cell fraction isolated from HWB. Compound RO3201195 was also tested in several acute inflammatory models in rats to evaluate its efficacy in vivo. The obtained results showed a significant dose-dependent inhibition of serum TNFa and IL-6 (IC 50 values of 0.2 and 0.3 mM, respectively) . A few years later, Bagley and coll. proposed a new synthetic method for the preparation of RO3201195 and tested the compound on Werner syndrome (WS) cells. High levels of phosphorylated p38a are expressed in proliferating WS cells, demonstrating that p38MAPK pathway inhibition could potentially interfere with the pathology. The results obtained in hTERT-immortalised HCA2 cells and primary WS cells treated with RO3201195 confirmed this hypothesis; in fact, this compound showed excellent selectivity for p38aMAPK over JNK and slowed down the accelerated aging of WS cells in culture . The same research group extended the SARs of previously synthesised 5APs and tested the newly obtained derivatives on human hTERT-immortalised HCA2 dermal cells to evaluate their ability to inhibit p38MAPK. The four compounds that displayed comparable or slightly improved potency over RO3201195 were tested on WS cells. Particularly, the 2,4-difluorophenyl substituent of compound 17a seemed to increase the activity compared to other derivatives, as well as the methyl substitution on N1 of compound 17b . These two molecules were selected as lead compounds but, despite their good p38a inhibition, they did not improve the efficacy and bioavailability of RO3201195 in in vivo evaluations . Other experiments were carried out on WS cells with RO3201195 and its analogues 17a and 17b . The results definitively confirmed that the proliferation of pathological cells is linearly correlated to p38 phosphorylation and that these 5APs significantly interfere with the protein expression and consequently with cell growth. Furthermore, the derivatives showed a better selectivity on p38 in comparison with the reference compound SB203580 which seemed to interfere in the same way with both p38 and JNK MAPK . In 2016, the screening of a DNA-encoded small molecule library allowed the identification of the highly specific and potent (IC 50 = 7 ± 0.9 nM) p38aMAPK inhibitor VPC00628 . This compound shares the N1 phenyl ring with RO3201195 and bears an amide function (rather than a keto group) on the C4 position. X-ray crystallography studies (PDB code: 5LAR) indicated that VPC00628 interacts with the ATP binding site of p38aMAPK, inducing a strong distortion of the P-loop. Interestingly, the ligand assumes an alternative binding mode as it lacks the key features of known kinase inhibitors such as a typical hinge binding motif. VPC00628 showed excellent shape complementarity and formed several specific polar interactions , assuming a canonical inactive type-II (‘DFG-out’) binding mode. This specific interaction with the inactive form of the kinase seems fundamental to increasing the potency and the selectivity of the inhibitor . Röhm and coworkers synthesised a new series of VPC00628 analogues, trying to improve the pharmacological properties of the parent compounds. Using a systematic combinatorial synthetic approach, they isolated and identified SR-318 , a 5AP characterised by a more hydrophobic amide moiety. The crystal structure of the SR-318/p38MPAK complex (PDB code: 6SFO; ) provided the structural basis for the excellent potency and selectivity for p38a/b (IC 50 a = 5 nM and IC 50 b = 32 nM) of the identified derivative. Moreover, SR-318 showed a better metabolic degradation profile in comparison with the reference drug VPC00628. To test the in vitro efficacy of SR-318, the LPS-stimulated TNF-a release in whole blood was determined. The results showed that this 5AP had an inhibition value of 97.7% at 10 mM (IC 50 = 0.283 mM) in the assay, resulting in more effectiveness than the literature-known compounds . Further studies focused on the exploration of the p38MAPK aC-out pocket showed that the hinge-binding motif of VPC00628 and SR-318 greatly enhanced the inhibitory activity compared with previously synthesised pyrazolyl-ureas. SARs extension of these reference derivatives led to the identification of Compound 18 as a selective type-II inhibitor. Despite the moderate activity of the enzyme (IC 50 = 14 nM), the crystallographic data provided valuable insights into the back-pocket interactions that were not observed in the SR-318/p38MPAK complex, thus indicating 5AP 18 as an alternative chemical tool with good cellular activity (PDB code: 6YK7; ). After demonstrating the excellent selectivity of 18 , human colon adenocarcinoma (HCT-15) cells were used to probe its efficiency on p38MPAKs. The Western blot analysis displayed the dose-dependent inhibition of p38 phosphorylation and increased phosphorylation of its downstream substrate HSP27. Furthermore, the new compound significantly inhibited TNF-a release with an IC 50 value of 0.48 mM, a better result than the one obtained with the type-I inhibitor . 5APs show very interesting anticancer properties; for some compounds, the intracellular targets have been identified (e.g., Bruton kinase for the recently approved Pirtobrutinib) while for other derivatives, only the antiproliferative activity on different cancer cell lines or additional intracellular mechanism have been reported (see table in ). In 2015, Ibrahim and coworkers prepared a new series of benzenesulphonamide derivatives incorporating pyrazole and isatin moieties. A biological evaluation was carried out to assess the ability of the compounds to inhibit the metalloenzyme carbonic anhydrase (CA) and more precisely the human (h) isoforms hCA I, II (cytosolic), IX, and XII (transmembrane, tumour-associated enzymes). The 15 hCA isoforms are widely distributed within different tissues and are involved in many physiological and pathological conditions. In particular, the selective inhibition of hCA IX and XII produces significant antitumour and antimetastatic effects . Derivatives 19a and 19b inhibited hCA XII with a Ki of 5.4 nM and 7.2 nM, respectively. In particular, pyrazole 19a with a 5-NO 2 substitution on the isatin ring was found to be a selective inhibitor of hCA IX and hCA XII, being more active than the acetazolamide used as the reference drug. A docking simulation confirmed that the NO 2 substituent present on 19a participates in interactions with Asp132 within the hCA IX active site and with the Lys67 and Asp130 residues in hCA XII . With the aim of identifying new chemical entities which can block the NF-kB cascade, Pippione, and coll. designed and synthesised a new series of 5APs that were biologically evaluated on four kinases involved in the NK-kB pathway (namely, IKKa, IKKb, IKKe, and NIK). In detail, 5APs 20a – c selectively inhibited NIK with an IC 50 of 8.4, 2.9, and 3.3 nM, respectively. A gene-reported assay was used to measure NK-kB activation in human multiple myeloma EJM cells, constitutively characterised by high levels of nuclear NK-kB related to NIK activation, and in breast cancer cell lines (SKBr3 and MDA-MB-231) in which the constitutive activation of nuclear NK-kB is unrelated to NIK. Compounds 20a – c seemed to inhibit the NK-kB activity in EJM cells (83.4–96.2% of inhibition at 25 mM), but not in SKBr3 and MDA-MB-231 cells, confirming their selective inhibitory activity for NIK . In 2020, Hassan and his research group designed and synthesised a novel library of twenty amino-pyrazoles, pyrazolo-pyrimidines, and their fused analogues. The compounds were tested by the National Cancer Institute (NCI, Bethesda, Maryland, USA) at a fixed concentration of 10 mM on a panel of 60 different human cancer cell lines. 5-amino-1-((4-chlorophenyl)(1-hydroxy-3,4-dihydronaphthalen-2-yl)methyl)-1 H -pyrazole-4-carbonitrile (compound 21 , ) showed the best cytotoxic activity with a cell proliferation inhibition higher than 90% on NCI-H23 (non-small cell lung cancer), HCT-15 (colon cancer), SF-295 (CNS cancer), NCI/ADR-RES (ovarian cancer), and DU-145 (prostate cancer) cells. As cyclooxygenase-2 (COX-2) inhibitors were largely reported to have antiproliferative activity against various cancer cells , docking simulations of 21 on COX-2 were carried out. Interestingly, the derivative displayed an affinity value of −7.86 kcal/mol and exhibited three hydrogen bonds with Lys137 and Gly45 side chains in the COX-2 active site . Looking for new cytotoxic derivatives, Anwer and Sayed prepared a new series of heterocyclic compounds through microwave reactions. All the synthesised molecules were tested by an MTT assay at different concentrations (1–100 mg/mL) on breast cancer (MCF-7) and colorectal carcinoma (HCT-116) cells. Among the compounds of the library, 5AP 22 (3-(4-(dimethylamino)phenyl)-1-phenyl-4-(1 H -tetrazol-5-yl)-1 H -pyrazol-5-amine, ), bearing an additional tetrazole substituent on C4, displayed the best cytotoxic activity with an IC 50 of 3.18 mM on HCT-116 and 4.63 mM on MCF-7 . In the same year and for the same purpose, Fadaly and coworkers isolated a novel series of triazole/pyrazole hybrids analogues (compounds 23 , ) of COX-2 inhibitor Celecoxib endowed with a large spectrum of activity. The reported compounds were also tested by MTT assay for their antiproliferative activity on MCF-7, HCT-116, A549 (human lung cancer cell line), PC-3 (human prostate cancer cells), and F180 normal fibroblasts. The sulphamoyl derivatives 23a and 23b were identified as the most active of the series with IC 50 of 4.22 and 6.38 μM on A549, 5.33 and 3.67 μM on MCF-7, 3.46 and 2.28 μM on HCT-116, and 1.48 and 0.33 μM on PC-3, respectively. An investigation of the mechanism of action of these molecules revealed that they were able to block the cell cycle at the G2/M phase, with a downregulation of Bcl-2 gene expression and an up-regulation of Bax expression, evidencing a pro-apoptotic mechanism. Docking simulation studies indicated the Epidermal Growth Factor Receptor (EGFR) as a potential target of 23a and 23b whose oxime functionality would form two hydrogen bonds with Thr830A and Asp831A. Enzymatic and ELISA assays confirmed the ability of the compounds to interfere with p38MAPK and VEGFR-2 signalling pathways . Very recently, Eli Lilly launched the new 5AP Pirtobrutinib (Jaypirca™, ), approved to treat mantle cell lymphoma (MCL), on the market . In detail, this (S)-5-amino-3-(4-((5-fluoro-2-methoxybenzamido)methyl)phenyl)-1-(1,1,1-trifluoropropane-2-yl)-1 H -pyrazole-4 carboxamide is a reversible inhibitor of Bruton Kinase (BTK), a nonreceptor tyrosine kinase, that represents a major therapeutic target for B-cell-driven malignancies. Different from previously approved covalent BTK inhibitors, innovative reversible BTK inhibitors, such as Pirtobrutinib, are characterised by limited off-target side effects and avoid the development of resistance mutations. These types of reversible BTK inhibitors have aroused great interest not only for the treatment of B cell malignancies but also for counteracting many autoimmune diseases . Pirtobrutinib is currently being studied in 16 clinical trials (1 completed and 5 in phase III) for chronic lymphocytic leukaemia (CLL), small lymphocytic lymphoma (SLL), and mantle cell lymphoma (MCL) . In the last twenty years, a rise in the level of antimicrobial resistance among pathogens occurred due to the emergence of resistant strains of pathogenic bacteria and the increase in the number of antibiotics administered. Therefore, the development of new antibacterial compounds has become an urgent and mandatory worldwide need. For this purpose, 5APs were studied and showed more interesting results than their 3APs isomers (as compounds 2 , 3 ). As reported below, in some cases, different 5APs revealed dual activity (anticancer and antibacterial action). In 2010, Gouda and coworkers prepared a novel series of heterocyclic derivatives in which 4,5,6,7-tetrahydro-benzo[ b ]thiophene-3-carboxamide scaffold was linked to pyrazole, pyrazolo-pyridine, pyrazolo-pyrimidine, or pyrazolo-triazine core (Compound 24 and used the agar diffusion technique to evaluate the antibacterial activity of the compounds on two bacterial strains ( B. theringiensis and K. pneumoniae ) and two fungi ( B. fabe and F. oxysporum ). 5AP 24a showed a larger inhibition zone diameter on Gram-positive B. theringiensis (24 mm) and Gram-negative K. pneumoniae (22 mm) than the reference drug Ampicillin (17 mm and 20 mm, respectively) . In 2012, Behbehani and coworkers utilised 2-aminothiophenes as building blocks for the synthesis of new pyrazole, pyrimidine, quinoline, and pyridine-2-one derivatives. The isolated compounds were tested and evaluated as antimicrobial agents on a large panel of Gram-positive and Gram-negative bacteria strains and two fungi by well diffusion assay. 5AP 25 exhibited good activity against Gram-positive B. subtilis with an inhibition zone diameter of 7.3 ± 1.1 mm (Penicillin = 4.6 ± 1.1 mm). The compound also displayed moderate activity against fungi, such as C. albicans (6.6 ± 1.1 mm) and S. serevisiae (4 mm) . Another class of 5APs with antibacterial properties was synthesised by Al-Adiwish in 2013. The antibacterial activity of the newly isolated compounds was tested using the agar diffusion technique at a concentration of 1 mg/mL on Gram-positive ( Staphylococcus aureus , Bacillus subtilis , Methicillin-resistant S. aureus , Staphylococcus epidermidis , and Enterococcus faecalis ) and Gram-negative ( Escherichia coli , Pseudomonas aeruginosa , Serratia marcescens , and Salmonella typhimurium ) bacteria. 26 exhibited the best activity against S. aureus and E. coli (15 ± 0.58 mm and 15 ± 0.56 mm, respectively). The minimum inhibitory concentration (MIC) and minimum bactericidal concentration (MBC) were also determined: 26 showed a MIC of 16 mg/mL on S. aureus and 4 mg/mL on E. coli , whereas the MBC value results were 16 mg/mL on S. aureus and 8 mg/mL on E. coli . The derivative was also tested by an MTT assay on Vero cells to evaluate its in vitro cytotoxicity. Unfortunately, despite the safety of the compound (CC 50 > 20 mg/mL), the selective index was too low to assess the selectivity of the molecule on the bacteria strains . In 2018, Eldin reported the preparation, characterisation, and antimicrobial activity of new amphiphilic pyrazole-g-polyglycidyl methacrylate-based polymers. The goal of this study was to synthesise a new class of antimicrobial copolymers containing pendant amino groups in combination with hydrophobic residues to obtain new antimicrobial agents against a variety of clinically significant pathogens. In detail, 5AP 27 , characterised by an azo-substituent on the C4 position and devoid of antibacterial activity, was grafted to the epoxy rings of polyglycidyl methacrylate and poly (glycidyl methacrylate-co-methyl methacrylate) copolymers. The structure verification of the new polymers was performed using FT-IR and TGA analyses and new derivatives were tested against different Gram-positive and Gram-negative bacteria strains, showing various activity, particularly against Gram-negative pathogens. Pyrazole-g-polyglycidyl methacrylate showed the most potential antibacterial activities, followed by pyrazole-g-poly (glycidyl methacrylates-co-methyl methacrylate) copolymers. All these considerations open a new area for developing novel pyrazole-based antimicrobial agents . Selected compounds from a heterocycle library synthesised by Anwer, including previously mentioned cytotoxic derivatives 22 and 5APs 28 and 29 displayed good antibacterial properties against both Gram-negative ( E. coli and P. aeruginosa ) and Gram-positive ( B. subtilis and S. aureus ) bacteria strains. Taking Ampicillin as a standard reference antibiotic, the activity index for each species was calculated. Pyrazole 22 exhibited the best results with activity index values ≥75% in all the analysed bacteria . Recently, using a sequential ligand-based pharmacophore-based virtual screening followed by structure-based molecular docking, Indian researchers identified some 5APs endowed with anti-tubercular potential. In detail, 5-amino-3-((substitutes phenylamino)- N -(4-substituted phenylamino)-1 H -pyrazole-4-carboxamides 30a – d showed potent anti-tubercular activity (MIC between 2.23 and 4.61 mM) compared to reference standard drugs and a lower cytotoxicity when tested on Vero cells. The preliminary SAR considerations suggested that the presence of the electron-withdrawing substituents at both R and R′ positions is beneficial for antibacterial activity, while electron-donating groups caused a remarkable reduction in potency. In addition, molecular docking studies suggested InhA, an enzyme involved in fatty acid synthesis and target for the development of novel antitubercular agents, as a possible target for the most potent derivatives . The increasing prevalence of drug-resistant Plasmodium falciparum strains led to an urgent priority in the development of new antimalarial drugs. In this context, Dominguenz and coworkers, utilising a previously reported synthetic methodology, prepared a new series of APs with potential antimalarial activity. The activity of the novel derivatives was tested in vitro against Plasmodium falciparum. 5APs 31a and 31b were the most active of the series with an IC 50 of 0.149 mM and 0.150 mM, respectively. From a chemical point of view, the presence of the ester functionality in position 4 of the pyrazole ring seemed to be essential for the activity; indeed, its substitution with a nitrile group led to a loss of inhibition. Interestingly, the electronic effect of the substitution in the phenyl ring seemed to be important; in fact, the substitution with the methoxy group in the meta position increased the activity of the compounds, while the same group in position ortho or para resulted in a progressive loss of potency against the plasmodium . A few years later, Verma and coworkers synthesised a series of pyrazole derivatives closely related to 31 , with an ethyl ester instead of a methyl one. All reported 5APs were tested in vitro against a chloroquine-resistant strain (FCBI) of Plasmodium falciparum . Not surprisingly, derivatives 32a and 32b displayed the best IC 50 with values of 0.149 mM and 0.150 mM, respectively, confirming the pivotal role of ester functionality and the meta substitution on the phenyl ring on C3 of the pyrazole ring. These two 5APs proved to be more effective than Pentamidine against Leishmania donovani , showing IC 50 values of 0.132 mM and 0.168 mM, respectively. Interestingly, regarding pharmacological activity, the ortho and para substitution led to a loss of activity as previously evidenced for anti-Plasmodium action . In an effort to discover new selective COX-2 inhibitors without the ulcerogenic side effects of non-steroidal anti-inflammatory drugs (NSAIDs), Abdellatif and coworkers designed and prepared a new class of pyrazole derivatives possessing an amino or mehansulphonyl pharmacophore. All the compounds were preliminary tested against COX-1 and COX-2 by an in vitro colorimetric enzyme immunoassay (EIA). Furthermore, the COX-2 selectivity index (SI) was calculated [IC 50 (COX-1)/IC 50 (COX-2)] and compared to the reference drugs Indomethacin (non-selective COX inhibitor) and Celecoxib (selective COX-2 inhibitor). 5APs 33 and 34 , characterised by two linked 5AP portions, exhibited the best results with an IC 50 on COX-2 of 39 nM and 34 nM, comparable to Celecoxib (IC 50 = 45 nM), and a SI of 353.8 and 417.6, respectively. The in vivo carrageenan-induced rat paw oedema assay confirmed the in vitro data for 33 and 34 , orally administrated at a concentration of 50 mg/Kg. In addition, 33 and 34 displayed an oedema inhibition percentage of 96% and 87% after 5h, resulting in more effectiveness than the reference drug Celecoxib. In in vivo tests, the new derivatives caused a reduced number of ulcers when compared to Indomethacin and Celecoxib, with an ulcer index (UI) between 0.7 and 2 (UI Celecoxib = 2.7 and UI Indomethacin = 21.3), confirming their pharmaceutical attractiveness. A molecular docking study showed H-bond interactions between the SO 2 groups of 33 and 34 and Phe504, Arg499, Tyr341, Ser516, and Arg106 amino acids, confirming the importance of the SO 2 substituent on the 5AP scaffold to obtain COX-2 inhibition . The following year, the same research group isolated a group of Celecoxib analogues linked to oxime moiety as nitric oxide donors, previously mentioned as anticancer agents . The aim of the project was to synthesise novel anti-inflammatory NO-NSAID hybrids with selectivity for COX-2 and with a NO-donor moiety. 5APs 35a and 35b showed higher activity in an in vitro COX-2 colorimetric assay (IC 50 = 0.55 mM and 0.61 mM, respectively) compared to Celecoxib (IC 50 = 0.83 mM). These two 5APs also displayed a good selectivity index (IC 50 COX-1/IC 50 COX-2) of 9.78 and 8.57, respectively, compared to 8.68 scored by the reference Celecoxib. To evaluate their in vivo anti-inflammatory activity, the derivatives were tested at a 50 mg/kg dose with the carrageenan-induced rat paw oedema assay. Pyrazole 35a exhibited the best oedema inhibition percentage (91.11%), overcoming again the values of Celecoxib (86.66%). The ulcerogenic assay highlighted the reduced ulcerogenic effects of the newly synthesised molecules (ulcer indexes UI = 2.79–3.95) in comparison with Ibuprofen (UI = 20.25) and Celecoxib (UI = 2.93). The docking simulation of 35a in the active site of COX-2 displayed that the compound would not form any H-bonds inside the COX-2 site. Furthermore, the percent of NO released from the derivatives was determined upon incubation in phosphate-buffered-saline. In detail, 35a and 35b exhibited a NO released percentage of 3.06% and 2.15%, respectively, which indicated a slow NO release. More recently, to obtain new hybrid compounds potentially able to act on different targets involved in inflammation onset, Brullo and coworkers designed and synthesised a series of pyrazole and imidazo-pyrazole derivatives with differently decorated catechol moieties linked through an acylhydrazone chain. The inhibitory effect of the novel isolated molecules on reactive oxygen species (ROS) production on platelets and neutrophils was evaluated. 5APs 36a , 36b , and 36c displayed the most promising results, with IC 50 values on ROS production inhibition in the low micromolar range with platelets, whereas derivative 36d showed a ROS production inhibition percentage of 68% on fMLP activated-neutrophils in flow cytometric analysis. The compound was also tested by enzymatic assay on phosphodiesterase enzymes PDE4B and PDE4D, intracellular enzymes mostly involved in neuroinflammation. The assay on 36d resulted in an IC 50 of 1.05 mM on PDE4D3 and an IC 50 of 0.55 mM on PDE4B2, confirming the potential antioxidant and anti-inflammatory activity of this new chemotype . Derivatives 36a , 36b , and 36c were further investigated to evaluate their effect on several parameters indicative of oxidative status and their efficiency on aerobic metabolism. All three molecules seemed to strongly inhibit superoxide anion production, lipid peroxidation, and NADPH oxidase activity and almost restored the oxidative phosphorylation efficiency in platelets stimulated with thrombin, highlighting their potential protective effect against oxidative stress. These results were confirmed in endothelial cells in which the selected compounds showed a promising inhibition activity on H 2 O 2 -stimulated EA.hy926 cells . In 2001, Kordik and collaborators prepared a novel series of 1,3-disubstituted-5APs with an affinity for the human neuropeptide Y (NPY) receptor Y5. NPY represents a powerful stimulant of food intake, implicated in obesity and eating; consequently, the antagonist of the NPY receptor Y5 could potentially provide new treatments for eating disorders. The synthesised 5APs, bearing a sulphonamide moiety, were tested for their binding affinity to NYP receptor Y5, using a transfected HEK293 cell line and measuring the competitive inhibition binding of 125 I-PYY. Derivative 37 displayed the best affinity in the first screening with an IC 50 value of 15 nM . Variation of the sulphonamide group led to inactive compounds, while the modification of the phenyl linker with a cyclohexyl group yielded pyrazole 38 , which showed a higher affinity for NPY Y5 than Fipronil . In 2020, Hebishy and collaborators described a new synthetic strategy to obtain novel benzamide-based 5APs as precursors of pyrazolo[1,5- a ]pyrimidine and pyrazolo-[5,1- c ][1,2,4]triazine derivatives; the derivatives were then tested as anti-influenza A agents active on subtype H5N1. The plaque reduction assay and the MTT cytotoxicity assay revealed 5AP 39 as the most promising compound, with an inhibition percentage value of 66.67% at 0.125 mmol/mL concentration and an LD 50 of 30 mmol/mL. Even though 39 resulted in less activity than the reference drug Zanamivir, it showed better potency than other reported bicycle analogues . 3,5-Diaminopyrazoles (3,5-DAPs) represent the most studied diaminopyrazole class and showed different biological properties including antiproliferative, antiviral, and antibacterial activities. Interestingly, as reported below, the majority of 3,5-DAPs with pharmacological activity are unsubstituted at the N1 position. In 2006, Krystof and coworkers identified 4-arylazo-3,5-diamino-1 H -pyrazoles active as ATP antagonists with potential selectivity for CDKs. Functional kinase assays confirmed the affinity toward CDKs, with a higher selectivity for the CDK9 isoform. The most promising derivative (compound 40 , ) displayed an IC 50 of 3.5 mM on CDK2-cyclin E and additional investigations demonstrated that 40 acts as a competitive ATP inhibitor with a Ki value of 13.3 mM. Crystallographic analyses allowed the definition of the structural basis for the interaction of 3,5DAP and the CDK9 ATP binding pocket. Functional assays reported the ability of 40 to affect CDK9-related pathways including decreased phosphorylation of the retinoblastoma protein (pRb), inhibition of mRNA synthesis, and induction of the tumour suppressor p53protein. Finally, the evaluation of this new 3,5-DAP in an antiproliferation assay showed a reduced frequency of the S-phase of the cancer cell line HT-29 . Recently, Ismail and coworkers designed and synthesised a new series of azo substituent 3,5-DAPs endowed with antiproliferative activity. The cytotoxic properties were evaluated in vitro against human breast cancer cells (MCF-7) and compound 41 was seen to be the most active (IC 50 of 26.86 mM). In vitro, a kinase assay individuated CDK2/cyclin E as a potential target of the newly synthesised derivatives . Furthermore, a library of 3,5-diamino- N -aryl-1 H -pyrazole-4-carbothioamides was identified as a potential class of HIV-1 inhibitors endowed with innovative mechanisms of action. The aim of the project was to synthesise new chemical entities able to inhibit different viral functions to provide a significant advantage against drug-resistant variants. All the novel DAPs were tested against RNase H activity, and derivative 42 displayed good inhibition of viral replication and promising activity against both RNase H and RNA-dependent DNA polymerase (IC 50 on RNase H of 7 mM). A docking simulation highlighted the binding of 42 to two RT (reverse transcriptase) pockets, one close to the polymerase catalytic site (RT-pocket 1) and one close to the RNase H catalytic site (RT-pocket 2), reinforcing the hypothesis of a dual-site inhibition. Moreover, these DAPs retained good inhibition potency against viral variants resistant to three non-nucleoside RT inhibitors (NNRTI) . Finally, some 3,5-DAPs were recently reported as inhibitors of P. Aeruginosa biofilms. This type of infection was observed in several bacterial diseases such as cystic fibrosis-associated lung infections, chronic wound infections, catheter-associated urinary tract infections, and ventilator-associated pneumonia. Indeed, Jansens and coworkers synthesised sixty new derivatives and evaluated their cyclic di-GMP (c-di-GMP) reducing potency in the biofilms using the c-di-GMP monitor strain MTR235. Compound 43 , similar to 40 and 41 but bearing a fluorine group in the ortho position of the phenyl ring, was identified as the most active of the series, with a reduction in the c-di GMP level of 83%. These results suggested a stimulation of the bacteria phosphodiesterases, leading to a reduction of the cyclic nucleotide and the consequent degradation of the biofilm. The ortho-substitution of the phenyl ring as well as the primary amino groups on the pyrazole ring emerged to be essential for activity. The good activity and interesting pharmacokinetic profile of compound 43 laid a good foundation for the development of new antibiotics for the treatment of bacteria biofilms . Due to its interesting pharmacological properties, the pyrazole nucleus has been extensively studied as a pharmacophore ; in particular, this heterocycle has been used to develop herbicides, agrochemicals, anticancer, anti-inflammatory, analgesic, antioxidant, anticonvulsant, antimicrobial, antimycobacterial, antiamoebic, antidepressant, hypotensive, and ACE inhibitors . As reported in , the AP scaffold has also been deeply investigated from a medicinal chemistry point of view, particularly for its anti-inflammatory and anticancer activity . Therefore, APs have been seen to be advantageous frameworks able to provide useful ligands for receptors or enzymes such as p38MAPK, different kinases involved in cancer progression, COX, and other targets important for bacterial and virus infections . In particular, the most relevant results have been obtained for anticancer/anti-inflammatory compounds , i.e., the recent approval of 5AP Pirtobrutinib, a reversible BTK inhibitor for the treatment of MCL, demonstrates the value of this scaffold for the development of new therapeutic agents. In addition, the anti-infective (i.e., antimicrobial, antimycobacterial, antimalarial, and antiviral) properties of this chemotype are also worthy of note , being able to overcome (myco)bacteria resistance to commonly used antibiotics. Regarding SAR considerations, with the exception of the 4APs (not well investigated), it can be highlighted that: (1) 3,5DAPs are biologically active when unsubstituted at the N1 position but embedded in the C-4 position; (2) 5APs endowed with anticancer/anti-inflammatory properties are generally characterised at the N1 position by phenyl or phenethyl groups (compounds 19 – 23 , 33 – 36 ); (3) (3)5APs reported as p38 inhibitors ( RO3201195 , VPC00628 , and SR-318 ) are characterised by a phenyl group at N1 and carbonil function of C4; (4) 5APs characterised by antimalarial activity are unsubstituted at N1; (5) 3APs and 5APS with anti-infective/antiviral activity could be substituted or unsubstituted at the N1 position. In recent years, a large number of papers or reviews highlighting the design, synthesis, and biological evaluation of different classes of pyrazoles and many pyrazole-containing compounds have been reported in the literature, but an overview of APs (bearing a free amino group at the 3, 4, or 5 position) and their biological properties is still missing. With the aim to fill this gap, the present review article focuses on aminopyrazole-based compounds in different therapeutic fields, with particular attention to the design and structure-activity relationship (SAR) aspects of each class of compounds, to provide better correlation among different currently ongoing research.
Postmortem Biochemistry and Immunohistochemistry in Anaphylactic Death Due to Hymenoptera Sting: A Forensic Case Report	73a4aa05-f565-4182-af28-639aee850681	10177871	Anatomy[mh]	Anaphylaxis is commonly known as a “severe, life-threatening generalized or systemic hypersensitivity reaction” and can occur both with immunological and non-immunological mechanisms . An anaphylactic reaction is an acute IgE-mediated hypersensitivity response mediated by inflammatory mediators released in systemic circulation from mast cells and basophil; an anaphylactoid reaction has non-IgE mediated mechanisms and, usually, is clinically indistinguishable from an IgE-mediated reaction . The most common triggers of anaphylaxis are drugs, food, and insect venom, and among these, Hymenoptera stings are quite represented . The Hymenoptera order is classified into three families: bees (Apidae), wasps (Vespidae), and ants (Formicidae). These arthropods can sting humans, having the potential to cause anaphylactic and non-anaphylactic reactions. Honeybees and bumblebees have barbed stingers and generally sting only if provoked; they characteristically die after a single sting. Wasps, hornets, and most yellow jackets have no barbed stingers and can sting many times. They are usually more aggressive than bees and also sting without any provocation . Hymenoptera toxins contain various complexes of peptides, enzymes, proteins, and chemicals, and they cause cellular injury via several mechanisms . Studies on the effect of these molecules demonstrated an action similar to toxins, hormones, antibiotics, and defensins which are able to interact with different pharmacological targets, causing inflammation, pain, changes in blood pressure and heart rhythm until cardiac arrhythmia, and neurotoxicity, and are even able to lead to death . It is important to underline that in the evaluation of deaths probably related to a bee sting, it is not always possible to appreciate macroscopic signs of the sting . In fact, in some cases, the sign of the puncture can be difficult to locate, or it is absent. Moreover, if death occurs in a very short time, no local reaction can be found . Therefore, in such cases, the postmortem assessment of the cause of death needs biochemical and immunohistochemical investigations that, together with circumstantial data, clinical data, autopsy, and routine histological findings, can provide useful evidence . Here, the authors report a case of anaphylactic death due to Hymenoptera stings to highlight the contribution of several forensic investigations in assessing the cause of death. The case regards a 59-year-old Caucasian man with a history of previous sensitization to Hymenoptera sting, the result of which was that, a few years earlier, he had facial edema due to both bee and wasp stings, then confirmed by skin tests. Anamnestic data were negative for cardiovascular and respiratory diseases. On the day of his death, the man contacted an employee by phone, asking for help and reporting that he was probably stung by bees. Once arrived, the employee found the man unconscious, lying on the ground, and with a transparent liquid coming out of his mouth, and contacted an ambulance. The medical staff found the man in cardiorespiratory arrest and began resuscitation, but the man died. The autopsy was performed at 24 h after death. Body inspection showed no signs of bee puncture. The gross examination revealed mild edema of the larynx, whitish foamy liquid in the bronchial tree, and red-brownish foamy dense liquid in the lungs. The routine histology was also performed, showing subacute pulmonary emphysema, endo-alveolar edema and hemorrhage, marked congestion of the interalveolar septa, bronchospasm, and scattered bronchial obstruction due to mucus hyperproduction ( A–C). Myocardial tissue showed hypertrophic myocytes, myofiber break up, and foci of wavy fibers ( D–E); atherosclerotic plaques were observed in coronary arteries. The toxicological investigations performed were negative for alcohol, abused substances, and psychotropic drugs. Biochemical investigations have been performed on the peripheral blood (femoral vein), showing an increased level of tryptase equal to 189 µg/L, troponin I 100,000 pg/mL, and proBNP 579 pg/mL. In addition, the Immuno-CAP method was applied for the determination of total IgE antibodies, which was found to be equal to 200 kU/L. ImmunoCap (Thermo Fisher Scientific/Phadia, Uppsala, Sweden) was carried out for the specific IgE dosage against honey bee (i1), white-faced hornet (i2), common wasp (Yellow Jacket–i3), paper wasp (i4) and yellow hornet (i5); the analysis allowed the identification of honey bee IgE equal to 5.30 kUA/L and yellow jacket IgE of 3.00 kUA/L. For immunohistochemical procedures, 4-micron thick sections obtained from larynx, lung, heart, and spleen tissue blocks were deparaffinized, then washed in descending alcohol scale, treated with 3% hydrogen peroxide for 10 min, washed again in deionized water three times, and incubated with normal sheep serum to prevent unspecific adherence of serum proteins for 30 min at room temperature. After, slides were washed with deionized water and incubated for 30 min at 37 °C with primary anti-human antisera monoclonal mouse anti-tryptase antibody (Roche Diagnostics code 760-4276). Next, the sections were washed three times with PBS and incubated with a biotinylated goat anti-mouse IgG secondary antibody (1:300; Abcam, code ab7064) for 20 min at room temperature, subsequently incubated with horseradish peroxidase-labeled secondary antibody for 30 min, developed with diaminobenzidine tetrahydrochloride, and counterstained with hematoxylin using the ULTRA Staining system (Ventana Medical Systems). Negative controls were obtained by omitting the specific antisera and substituting PBS for the primary antibody. Immunohistochemical reaction revealed intense expression in the larynx and lungs, showing several immunopositive mast cells and spread immunopositivity for degranulated tryptase ( A–D); mild expression in both coronary arteries’ walls and myocardial tissue characterized by scattered positive mast cells and foci of tryptase degranulation ( E–G); and mild positivity in splenic tissue showing mast cells and spread degranulated tryptase expression ( H). The global incidence of anaphylaxis is reported between 50 and 112 episodes per 100.000 person years with a low mortality rate, estimated at 0.05–0.51 per million people/year for drugs, at 0.03–0.32 for food, and at 0.09–0.13 for venom . In Italy, Bilò et al. reported 392 cases of death from anaphylaxis, with a mortality rate of 0.51 per million people per year. Hymenoptera stings were responsible for 5.6% of these deaths, with an overall mortality rate of 0.17 per million people per year. Even if Hymenoptera stings are a frequent cause of anaphylactic reactions, there is a consistent number of related deaths that cannot be correctly identified due to the difficulty of making a postmortem diagnosis . To perform the diagnosis of death due to anaphylactic shock, it is necessary to integrate circumstantial and anamnestic data, autopsies, and histological findings . However, postmortem assessment of anaphylaxis as the cause of death is considered a challenge for forensic pathologists, because evidence emerging from autopsy and histology is often unspecific. In this context, other postmortem analyses, such as biochemistry and immunohistochemistry, can provide a useful contribution. The subject’s clinical history serves to collect information on both previous allergic reactions and sensitization phenomena to specific allergens; likewise, the circumstances of death play an important role in the forensic analysis of the case . Relevant findings can be provided from autopsy and routine histology. Sting signs and the evidence emerging from gross and microscopic analysis of the respiratory system (such as laryngeal edema, tracheo-bronchial hypersecretion, bronchoconstriction, emphysema and acute pulmonary edema, congestion, and intra-alveolar hemorrhage) support the occurrence of anaphylaxis . However, some of these findings could not be found and, even if identified, cannot be considered pathognomonic and specific. In fact, such respiratory system involvement has also been described in asthma . Many researchers have suggested the use of biochemistry and immunohistochemistry to fill the gaps related to the poor or absent autopsy and histological data in performing the postmortem diagnosis of anaphylaxis . Particularly, blood biochemical investigations to evaluate tryptase and IgE are described as useful tests to confirm deaths related to anaphylactic reactions due to bee venom, especially when there are no evident signs of stings . Serum tryptase is a neutral protease of human mast cells, mostly used as a biomarker to better define the postmortem diagnosis of anaphylaxis . Tryptase is a very stable enzyme and can be detected up to 6 days after death . Nevertheless, it must be emphasized that postmortem degradation processes can cause a reduction in the real concentration of tryptase proportionally to the increase in the postmortem interval (PMI). Therefore, if there is a suspicion of death related to anaphylaxis, it is suggested to collect a blood sample as soon as possible . Forensic literature reports variable cut-offs for serum tryptase from peripheral blood. Meyer et al. demonstrated that a level of tryptase of 10 μg/L or greater has a sensitivity of 86% and specificity of 88% for the diagnosis of postmortem anaphylaxis. Tse et al. reported a cut-off value of tryptase ≥53.8 μg/L on peripheral blood taken from the femoral vessels to make a postmortem diagnosis of anaphylaxis-related death. Edston et al. have proposed a value of 45 μg/L as a new cut-off point, especially if death is due to insect stings. The literature also offers evidence on serum tryptase measurement on blood taken from central vessels, such as the aorta, in which the suggested cut-off value is 110 µg/L . However, it was highlighted that prolonged cardiac massage or defibrillation can determine the increase in mast cell degranulation and the increase in tryptase levels, due to visceral trauma from chest compressions . In general, several studies suggest preferring peripheric blood sampling for the postmortem tryptase assay . Nevertheless, factors affecting the tryptase concentration (i.e., hemolysis, length of agonal period, the specific type of trauma, and cause of death) should always be considered in forensic practice . In fact, the increase in tryptase levels was also described in non-anaphylactic deaths, such as sudden infant death syndrome, acute deaths after heroin injection, traumatic deaths, and asphyxia . The serum concentration of tryptase found in the femoral blood of the case presented here was equal to 189 µg/L and supported the occurrence of anaphylaxis. In the presented case, other useful data were obtained from the analysis of total and specific IgE. Particularly, the analysis revealed a total IgE value of 200 kU/L and the presence of specific IgE for honey bee (5.30 kUA/L) and yellow jacket species (3.00 kUA/L), demonstrating a high (Radio-Allergo-Sorbent-Test class 4) and moderate (Radio-Allergo-Sorbent-Test class 3) level of sensibilization, respectively. Evidence in the literature suggests combining results of both mast cell tryptase and allergen-specific IgE and/or total IgE assay in postmortem serum to support the assessment of IgE-mediated fatal anaphylaxis . Even if few forensic studies investigated the behavior of postmortem serum total and specific IgE, some evidence showed relative stability of the antibodies in peripheral blood; albeit, some authors reported an increase in total IgE level proportionally with the postmortem interval . It is also important to observe that the measurement of serum IgE provides information about the atopic disposition and degree of sensitization to a particular allergen; thus, this cannot be considered a confirmation of the causal link between IgE-mediated anaphylaxis and death . Immunohistochemistry is another investigation useful for postmortem diagnosis of anaphylaxis. Many studies focused on the role of mast cell and tryptase detection in tissues, among which are bronchial, respiratory, and intestinal mucosa, and the red pulp of the spleen and connective tissue (i.e., cutaneous and perivascular) . Nevertheless, it is important to underline that the identification of mast cells in the tissues cannot be considered sufficient to make a diagnosis of certainty. These limits are related to (i) the involvement of mast cells in various biological processes (i.e., tissue remodeling, angiogenesis, fibrosis, and asphyxia), (ii) the physiological interindividual variability in the number of mast cells, and (iii) the increased detection also observed in non-anaphylactic death . Particularly, Edston et al. reported a similar number of pulmonary mast cells in both anaphylactic deaths and control cases (cardiovascular deaths), whereas a higher expression of spleen mast cells was observed in anaphylactic deaths rather than in controls. The immunohistochemical analysis performed in the presented case revealed an intense positivity of mast cells and degranulated tryptase in the larynx and lungs, together with a mild marker expression in the spleen. These expression patterns are in accordance with the evidence in the literature and support an anaphylactic death. Moreover, the immunohistochemical findings observed in the heart, together with the increased level of serum troponin and pro-BNP, could suggest a coronary hypersensitivity similar to that described in Kounis syndrome. This morbidity is associated with allergic, hypersensitivity, anaphylactic, and anaphylactoid reactions, and it is classified into three types . The type I variant, known as vasospastic allergic angina, is characterized by endothelial dysfunction or microvascular angina and occurs in subjects with normal or nearly normal coronary arteries and in the absence of predisposing factors for coronary artery disease; the releasing of inflammatory mediators due to anaphylaxis can cause coronary artery spasm until myocardial injury with impaired cardiac enzymes and troponins. The type II variant, also known as allergic myocardial infarction, has been described in subjects with quiescent pre-existing atheromatous disease. In this case, the acute release of inflammatory mediators can provoke both coronary artery spasms with normal cardiac enzymes and troponins or coronary artery spasms associated with plaque erosion or rupture. The type III variant occurs in patients with coronary artery stent in whom the inflammatory reactions cause a prothrombotic response and the stent thrombosis; eosinophils and mast cells are generally detected in thrombi and coronary wall at histological examination . Therefore, in Kounis syndrome, the myocardial damage seems related to the effect of both mast-cell degranulation and the release of inflammatory mediators that affect the cardiovascular system (i.e., coronary vasoconstriction induced from histamine) . Few cases of Kounis syndrome due to bee and wasp stings have been described in the literature . The case argumentation highlights that postmortem diagnosis of anaphylactic death is based on a combination of data about the event, medical history, gross and microscopic examination, and blood serum analyses. Particularly, even if tryptase analysis by biochemistry and immunohistochemistry and IgE dosage have limits in specificity and sensitivity, their integration with the other information is fundamental to performing a differential diagnosis and, thus, to assessing anaphylaxis . Moreover, a prompt sampling, performed as soon as possible, is crucial to prevent the effect of postmortem phenomena (i.e., cell lysis) on tryptase . In conclusion, data emerging from the forensic investigations lead to assessing the cause of death as an anaphylactic shock due to Hymenoptera stings affecting the respiratory system and the cardio-circulatory system, with possible vasospastic involvement of coronaries. The case described here supports the importance of circumstantial data in guiding postmortem investigations, especially if no external signs attributable to the insect bite and/or unspecific autopsy and histological evidence are found. Moreover, the important role of biochemistry and immunohistochemistry in demonstrating the anaphylactic reaction has been described, suggesting that these investigations should be routinely implemented in forensic practice when anaphylaxis is suspected.
Strategies for Monitoring Microbial Life in Beach Sand for Protection of Public Health	ca670362-1858-434d-ad16-9e03b91a8793	10178049	Microbiology[mh]	A day at the beach is spent mainly in the sand area where people lie down sunbathing and where children play. The wind displaces loose grains of sand, some of which end up deposited on beachgoers skin and hair or forced into the ear, nose, and mouth. At the end of that day, beachgoers return home, inevitably taking some sand and microorganisms along with them. Yet, despite the higher amount of time spent on sand than in water, the latter has well established acceptable levels of microbes based upon exposure, risk, and safety standards , while sand is lagging in such characterizations . The rationale that we spend more time on the sand than in the water was clearly recognised by the World Health Organization (WHO) in 2003, within Chapter 6 of the Guidelines for Safe Recreational Water Environments . Between 1969 and 2003, many publications addressed microbes in beach sand. Yet, it was not until 2012 that the health impacts of sand exposure were quantified. In the 2012 publication, Heaney et al. analyzed 144 wet sand samples for the presence of the fecal indicator bacteria (FIB) enterococci and conducted 4999 interviews describing contact with sand. The study found that enterococci in sand was associated with gastrointestinal (GI) illness. Beachgoers who dug in the sand or were buried in sand exhibited higher incidence rates of GI compared with those who did not. There are other organisms, such as Cryptosporidium spp., Clostridium perfringens , and Bacteroides spp., which have been used to indicate fecal contamination of bathing waters. The latter are used mainly to identify the microbiological source of fecal pollution . Fujioka and collaborators , discussed the recreational water quality criteria of 2012 of the United States of America (USA), and concluded that quality indicators would need to be adjusted in many regions to reflect emerging pathogens. For example, the regulations of the State of Hawai’i included use of C. perfringens as a FIB for decades. The rationale behind the use of C. perfringens as a fecal indicator in tropical waters is based upon the fact that C. perfringens requires anaerobic conditions to multiply outside of the human host and is therefore unable to multiply in aerobic surface soils and waters. The more mainstream FIB, such as fecal coliform, E. coli , and enterococci, multiply in the environment under warmer and more humid environmental conditions. Considering climate change, alternative indicators such as C. perfringens may become more mainstream to control for environmental regrowth as temperatures increase [ , , ]. One of the current approaches recommended in 2004 by the WHO for drinking water is the establishment of water safety plans (WSPs), to minimize threats to water supplies by systematically assessing and managing risks . This recommendation covers all possible scenarios where humans can be exposed to microbes, including sediment. The European Commission readily adopted this perspective for the most recent Bathing Water Directive 2006/7/EC . In the WHO 2020 review of the guidelines, WSPs were also recommended for recreational waters . This emphasizes the need to consider the nearshore environment as a possible source of microbial exposure, inclusive of pathogens and opportunistic organisms. 2.1. Types of Sand and Artificial Beaches Th beach is defined as the zone of unconsolidated material between the low water line and the landward limit of wave swash, often marked by either a change in landform or vegetation . Exposed sandy beaches are physically dynamic habitats, inhabited by specialized biotic assemblages that are structured mainly by physical forces . Beach sand is composed of minerals and water that can sustain entire micro-ecosystems that comprise a wide variety of biological forms. Microorganisms embed within complex biofilms that are attached to the surfaces of the grains of sand, thriving on the available nutrients . In 2016, Abreu and collaborators published a paper that compared sand grain size and composition with the microbial community . This research group found that sand granulometry and chemical composition was not significantly associated with microbial concentration. However, Valério et al. verified that different amounts of sand are needed for deoxyribonucleic acid (DNA) extraction, depending on the grain size, to achieve equivalent yields. Furthermore, Abreu et al. compared natural and artificial beaches, revealing a difference in microbial concentrations. Manmade structures, built to maintain sand in place, impeded the natural wave activity and thus limited wash-off of microorganisms from sand, resulting in microbial accumulation in the supratidal area. Coincidentally, this is the area mainly used by beachgoers for sunbathing and relaxing. Similarly, Hernandez et al. found that sand mineralogy may be related to its ability to retain microbes, with quartz sand and smoother surfaces retaining fewer microbes relative to carbonate sands with rougher surfaces and higher surface areas. Beach nourishment is a common practice for reclaiming coastlines lost to wave activity and natural erosion and for building artificial beaches . The microbial community of the sand used for renourishment should be considered since it will carry its own native microbial communities to the new location . When the artificial beach Praia da Calheta was built and nourished with sand originating in the Sahara, Morocco, the first batch of sand brought live scorpions along with it. Consequently, the project had to be halted and redesigned . Nourishing sand should thus originate as geographically close as possible to its final location, in order minimize the introduction of non-native species. In addition, to minimize losses by erosion and to maximize the chances of success of nourishment projects, the sand used to nourish beaches is recommended to have grain-sizes 1.5–2.0 times those of the original sand . 2.2. Swash Zone vs. Supratidal Zone The sand area of a beach is divided into several sections ( ). In this text, the focus shall be on the swash zone (the intertidal area limited above by the high tide reach) and on the supratidal zone (beyond wave and high tide reach). The latter tends to stay dry unless impacted by precipitation and run-off from the backshore. Backshore run-off is a well-known cause of sand contamination, as demonstrated by an episode of a destructive tropical storm hitting the island of Madeira in 2010 . The storm destroyed facilities of all kinds, including the sanitary infrastructure. Due to the heavy rain, land- and mudslides rushed down to the lower lands along the coastline. FIB from this surge could be detected within the beach sand for approximately three months after the storm event . Conversely, the swash zone is influenced by the waves and tides . A review focused on the micropsammon (bacteria, fungi, parasites, and viruses) fate and transport describes how sand microbial communities include microbial species that are both indigenous and allochthonous (i.e., non-native) to the local environment . The review also addressed the naturalisation of allochthonous micropsammon, introduced by water, air , run-off, seaweed , and by animals. For reference, FIB are considered allochthonous to beach sand . Th beach is defined as the zone of unconsolidated material between the low water line and the landward limit of wave swash, often marked by either a change in landform or vegetation . Exposed sandy beaches are physically dynamic habitats, inhabited by specialized biotic assemblages that are structured mainly by physical forces . Beach sand is composed of minerals and water that can sustain entire micro-ecosystems that comprise a wide variety of biological forms. Microorganisms embed within complex biofilms that are attached to the surfaces of the grains of sand, thriving on the available nutrients . In 2016, Abreu and collaborators published a paper that compared sand grain size and composition with the microbial community . This research group found that sand granulometry and chemical composition was not significantly associated with microbial concentration. However, Valério et al. verified that different amounts of sand are needed for deoxyribonucleic acid (DNA) extraction, depending on the grain size, to achieve equivalent yields. Furthermore, Abreu et al. compared natural and artificial beaches, revealing a difference in microbial concentrations. Manmade structures, built to maintain sand in place, impeded the natural wave activity and thus limited wash-off of microorganisms from sand, resulting in microbial accumulation in the supratidal area. Coincidentally, this is the area mainly used by beachgoers for sunbathing and relaxing. Similarly, Hernandez et al. found that sand mineralogy may be related to its ability to retain microbes, with quartz sand and smoother surfaces retaining fewer microbes relative to carbonate sands with rougher surfaces and higher surface areas. Beach nourishment is a common practice for reclaiming coastlines lost to wave activity and natural erosion and for building artificial beaches . The microbial community of the sand used for renourishment should be considered since it will carry its own native microbial communities to the new location . When the artificial beach Praia da Calheta was built and nourished with sand originating in the Sahara, Morocco, the first batch of sand brought live scorpions along with it. Consequently, the project had to be halted and redesigned . Nourishing sand should thus originate as geographically close as possible to its final location, in order minimize the introduction of non-native species. In addition, to minimize losses by erosion and to maximize the chances of success of nourishment projects, the sand used to nourish beaches is recommended to have grain-sizes 1.5–2.0 times those of the original sand . The sand area of a beach is divided into several sections ( ). In this text, the focus shall be on the swash zone (the intertidal area limited above by the high tide reach) and on the supratidal zone (beyond wave and high tide reach). The latter tends to stay dry unless impacted by precipitation and run-off from the backshore. Backshore run-off is a well-known cause of sand contamination, as demonstrated by an episode of a destructive tropical storm hitting the island of Madeira in 2010 . The storm destroyed facilities of all kinds, including the sanitary infrastructure. Due to the heavy rain, land- and mudslides rushed down to the lower lands along the coastline. FIB from this surge could be detected within the beach sand for approximately three months after the storm event . Conversely, the swash zone is influenced by the waves and tides . A review focused on the micropsammon (bacteria, fungi, parasites, and viruses) fate and transport describes how sand microbial communities include microbial species that are both indigenous and allochthonous (i.e., non-native) to the local environment . The review also addressed the naturalisation of allochthonous micropsammon, introduced by water, air , run-off, seaweed , and by animals. For reference, FIB are considered allochthonous to beach sand . Prediction and mitigation of contaminant surges at beaches requires an understanding of microbial communities in sand, their inter-species interactions in multi-communal biofilms, and their use of molecular warfare against other susceptible species, with antibiotics and statins, for example [ , , , ]. The first approach to maintaining a balanced level of microbial life in sand is to prevent contamination in the first place by implementing hygiene measures, e.g., regular trash removal, providing restroom facilities for beachgoers, removing remains of fish from the sand, etc. However, these actions may not always be possible. Brandão et al. and WHO discuss possible actions to take for remediation in case of necessity (included in ). Biofilms provide an interesting additional dynamic to controlling microbial communities in sand. In the laboratory, the biofilms may reduce the colony forming units extracted from sand due to agglomeration. However, FIB in biofilms would not be readily available as inocula to form new colonies in the natural environment. To visualize, Brandão et al. , presents pictures of initial stages of biofilm construction on the surface of sand grains, as revealed by electron microscopy. Results of the visualization show that biofilms are complex, multi-species communities that can integrate fungi with other microorganisms. For example, when black mold is a constituent of a biofilm exposed to sun radiation, its presence will protect other species from the damaging effects of ultra-violet (UV) radiation by absorbing it as a source of energy for its own growth. As a carbon source for its growth, black mold will use either decaying biofilm inhabitants or hydrocarbons present from fossil-fuel residue or shedding by more complex lifeforms visiting or inhabiting the beach . 3.1. History of Microbial Tracking in the Sand The first paper published on beach sand contaminants was in 1960 by Schönfeld, Rieth and Thianprasit, and it described a study which aimed to detect the presence of dermatophytes in a Baltic Sea resort. The study yielded only the geophilic dermatophyte Arthroderma insingulare (formerly Trichophyton terrestre ) in the superficial layers of supratidal sand . In 1973, however, Gertrud Müller in a wider study confirmed the absence of anthropophilic dermatophytes in sands. For this research, Müller’s team sampled Estoril, near Lisboa, the Adriatic Sea, at Gabicce Mare and at Grömitz, on the German Baltic coast. The team searched for Epidermophyton floccosum , an anthropophilic species, for three years, and sampled twice per week (in the same places). They also aimed to find beach usage-associated variations of this fungal species . The study revealed a clear surge in the presence of Epidermophyton floccosum in sands touched by human bare feet. It failed to isolate other species, common in other kinds of soil, showers, and pools such as Trichophyton interdigitale (formerly T. mentagrophytes var. interdigitale ) or for Nannizzia gypsea (formerly Microsporum gypseum ). The authors pointed out in their report that the inability to isolate these common agents of dermatophytosis should not rule out their presence in beach sand that had been in contact with human feet. Maria Colon Valiente, in 1990, hypothesized that the absence of dermatophytes in beach sand was likely due to the low nutrient environment or the high maximum temperatures . These observations, however, were challenged by Sousa , who reported dermatophytes in 11 out of 24 (45.83%) beaches studied in coastal areas around Lisboa. Brandão et al. sampled all regional coasts of Continental Portugal every 2 months, for 13 months, and generated 210 samples of sand from both the supratidal and the vadose zones. The three types were wild beaches, beaches with water quality problems and beaches awarded blue flag status for exceptional quality. Despite their high maintenance levels in relation to the other beach types, blue flag beaches did not exhibit the lowest fungal concentrations in sand. In this study, a higher density of beach users correlated with higher levels of dermatophytes during the summer months. Not only was the presence and concentration of fungi in beach sand highly variable, but the fungal community may have been also evolving due to land use and climate changes. Therefore, additional research and monitoring will be integral to characterizing and maintaining beach sand quality in the context of fungal contaminants. Recently, E. floccosum seems to be decreasing in its prevalence in clinical specimens worldwide, indicating that the likelihood of isolating it from beach sand may be diminished . Other dermatophytes, and even other microbial life forms, may alter their patterns of existence due to clinical intervention and climatic alterations . Oshiro and Fujioka studied the bacteriological water quality of Hanauma Bay in Hawai’i. Due to the extreme human densities associated with recreational use, water quality of Hanauma Bay was difficult to maintain at acceptable levels. The Bay has since been classified a nature preserve and is open to recreation during limited timeframes, to allow the local ecosystem to recover without the threat of constant high human densities . Because of its status as a bay full of marine wildlife, and frequent use by tourists and local beachgoers, the water quality led Oshiro and Fujioka to sample the shoreline water and sand, land runoff, and mongoose and pigeon feces, to try to find the main causes of the surges of FIB. The samples were analyzed for fecal coliforms, Escherichia coli , and enterococci, and revealed that the major sources of the periodically high levels of these bacteria in the water were contaminants founds in beach sand, namely pigeon feces . In this case, water quality was the main driver of the research, although the source came from the sand. In 2017, Argentina became the first nation to formally include sand inspection for rubbish in its water quality standards . Lithuania added monitoring of helminths in sand to their 2007 National regulation within the 2018 updates . The Lithuanian regulations are currently the only known regulation on helminths, as mentioned in WHO . During the last 20 years, several publications have identified potential microbial culture parameters for sand monitoring [ , , , ]. Others contemplate meta-genomics, raising possible alternative parameters, based on nucleic acid analyses [ , , , ]. However, until there is a regulatory document defining sand quality parameters, the choice for monitoring parameters is likely to fall on a combination of culturable fecal indicator organisms (FIO), to indicate pathogens and opportunists associated with wastewaters, and others that do not relate to fecal pollution (including most fungi). Next generation sequencing (NGS) may be an interesting screening tool to assess variations in the microbiota as shown in Taylor and Kurtz . In this study, the authors evaluated three beaches along the Grand Strand of South Carolina, USA. The authors sequenced the V4 region of the 16S rRNA gene to compute relationships between diversity and temporal or local factors. Gammaproteobacteria, Planctomycetes, Acidobacteria, and Actinobacteria were the dominating bacterial populations at these beaches. The communities were similar in overall composition and diversity, but the abundance of taxa changed over time. shows pathogens, opportunistic microorganisms, and fecal indicators in beach sand currently available in the literature. The following three sections describe the current taxa of interest for sand monitoring. WHO addresses three other biological groups of interest that are currently poorly represented in the scientific literature, namely viruses, helminths, and insects. 3.2. Fecal Indicator Bacteria (FIB) Fecal pollution remains one of the biggest problems with recreational waters and beach sand. Parts of the globe have taken action to eliminate the discharge of untreated sewage into water bodies. However, in other regions, fecal contamination of bathing and even drinking water sources remains a serious problem. One of the most notorious contamination episodes of water took place in 2010 in Haiti, due to the destruction caused by an earthquake of magnitude 7.0. This event killed an estimated 230,000 people and injured another 300,000. Haiti already had a relatively low coverage of sanitation infrastructure but after the earthquake, only 10% of the rural population and 24% of the urban population had access to improved sanitation . This low rate of access to drinking water and sanitation coupled with contamination of the water supply led to an outbreak of cholera, which resulted in 658,563 reported cases and 8111 deaths from the disease as of 2 June 2013 . The WHO guidelines for safe recreational water environments recommends the use of FIB to detect fecal pollution in recreational waters. The more recent guidelines , extend those recommendations to beach sand, precisely to avoid waterborne diseases originating from human excreta. The recommendations focus on measurements of FIB, not the actual pathogens. According to the literature, exposure to 40 CFU/mL of enterococci results in an illness probability of 1%, and up to 200 CFU/100 mL results in up to 5% probability of illness. This indicator parameter in recreational waters suggests general disease probability rather than any specific pathogen-associated illness. It also is currently in use in regulations, e.g., the European Bathing Water Directive . Fujioka et al. , Weiskerger et al. , and Teixeira et al. discuss the need to extend beyond FIB in the future, to accommodate changes arising from climatic alterations. Looking into pathogens directly instead of quantifying indicators is not only increasingly easy, but will also have to be implemented for pathogens that in the future may not be predicted through the use of FIB. 3.3. Other Bacteria As described in WHO , Sabino et al. , and Weiskerger et al. , controlling exposure to bacteria to protect human health needs to go beyond FIB, since the supratidal zone of a beach may not be greatly contaminated with fecal pollution. Instead, its main source of pollution is often skin shedding from animals, vegetable debris, and other organic matter that may serve as food for wildlife . The following bacterial taxa have been considered of relevance for the protection of human health: Staphylococcus spp. (skin infections), Vibrio spp. (cholera and necrotizing fasciitis), Clostridium perfringens (food poisoning, possible cause of bacteraemia), Campylobacter jejuni (gastroenteritis), Shigella spp. (haemorrhagic diarrhoea), and Pseudomonas aeruginosa (superficial and systemic infections) [ , , ]. Methicillin-resistant Staphylococcus aureus (MRSA) was reported in 2009 as present in beach sand and water in California and Florida, USA and more recently in South Africa, in 2015–2016 , presenting yet another emerging contaminant of concern for beachgoers. This group of sand contaminants and climate change implications is addressed further by the WHO and in Weiskerger et al. . 3.4. Fungi Most fungi are opportunistic, which means that exposure is a potential risk only for susceptible individuals. Yet, two groups of fungi are of great concern: dermatophytes and endemic fungi. Dermatophytes are keratinophilic fungi that cause dermatophytosis, superficial infections of the hair, nails, skin, and scalp. There are anthropophilic, geophilic, and zoophilic dermatophytes and they are so classified according to their transmission route: if transmission is human-to-humans, soil-to-human, or animal-to-human . The current knowledge suggests that anthropophilic dermatophytes may be susceptible to the environmental conditions in beach sand and thus may die rather quickly . This does not necessarily mean that there is no transmission at the beach, though. Shedding takes place naturally, especially from infected areas of the skin that become dry and scaly, and death due to radiation or high temperatures is not immediate. In fact, under laboratory conditions, recreational beach sands from Hawai’I were able to maintain several fungal species. In the study, Anderson tested the survival of Cutaneotrichosporon cutaneum/Trichosporon asahii, Candida albicans, Nannizzia gypsea and Trichophyton mentagrophytes (var. mentagrophytes) and all survived for at least one month in non-sterile sand inoculated with keratinized propagules. These data suggest that non-detection of anthropophilic dermatophytes may be associated more with the natural dispersion of propagules and representativeness of sampling, rather than the dermatophyte survival itself. As for the endemic fungi in sensu stricto ( Coccidioides spp., Paracoccidioides brasiliensis , Histoplasma spp., Blastomyces dermatitidis ), no known publications indicate their isolation from beach sand. However, their natural habitats suggest some degree of survival in beaches within endemic areas (inland, mainly). Coccidioides spp. inhabits dry, desert-like territories of the USA, and of Argentina. The remaining fungi are endemic to more humid habitats such as the Mississippi River Valley, USA ( Blastomyces spp.); central and eastern USA ( Histoplasma capsulatum ); Sub-Saharan Africa ( Histoplasma duboisii); and South America ( Paracoccidioides brasiliensis ) . Cryptococcus deuterogattii is also endemic to the Pacific Northwest of the North American continent but not considered dimorphic (real pathogens). There is yet to be consensus whether this species should be considered simply a genetic type of Cryptococcus gatii (type AFLP6/VGIIa). Other fungi of interest, as listed recently by WHO for exposure in natural environments, are Mucorales , allergenic fungi, and dematiaceous fungi. Mucorales is the order that includes the fungi responsible for the invasive mucormycosis. This order has raised some concern due to the opportunistic ability to start an infection in immunocompetent individuals when inoculated deeply under the skin by piercing materials. This has been well demonstrated by a cluster of cases of necrotizing cutaneous mucormycosis following a tornado in Joplin, Missouri, USA in 2011 where several individuals were injured by deep puncture wounds. Mucorales are also a relevant group of fungi in cases of near drowning, as described by Sympardi et al. . However, the distribution of fungal species has been reportedly different in different contexts. Cogliati et al. found different distributions of yeasts and molds in beaches throughout Europe depending on the latitude and heavy metal composition of soil. Yeasts seem to withstand low temperatures better during winter than molds and are therefore more prevalent in Northern European coasts. Moreover, molds are more associated with soil rich in nickel and yeasts with soils rich in cadmium. This results in a distribution of fungi mainly at the deltas of European rivers and lagoons, where these metals accumulate in river sediment. Considering that an allergy is host-dependent and that there are individual allergies to most fungi, for the purpose of this text, all fungi are considered to also cause allergies. There are, however, different types of allergies: respiratory and contact allergies. Respiratory allergies imply inhaling allergens and therefore are mainly caused by airborne-sporulating fungi. The extent to which fungal spores can travel airborne has been clearly described by Kellogg and Griffin , in an open report about the air travel of fungi from Africa to North America. Fungal spores can be airborne and can also extend their presence to an entire beach. Although information on fungal inhalation leading to infections specifically from sand is currently unavailable, Buskirk et al. showed in a murine model that the dry exposure to 10 5 spores of A. fumigatus twice per week triggers an inflammatory response in the lungs within 24 and 48 h. Additionally, an IgG elevation is observed after seven days, concomitant with spore germination. Tanaka et al. found that cytokine release in immunocompetent individuals takes place about 19 h post-inhalation. Cho et al. observed in fungal respiratory deposition models that particles of Stachybotrys chartarum might be deposited in numbers 230–250-fold higher than spores. These data suggest a delayed first response to the exposure to fungal allergens, but an existing one, nonetheless. Beach users might thus not even associate an allergic episode with a visit to the beach the day before. Lastly, exposure to volatile organic compounds (VOCs) produced by fungi like 3-Methylfuran may cause nonspecific symptoms, namely eye, nose and throat irritations, headaches, and fatigue . Aleksic et al. described the difference between the propagation of aerosolized mycotoxins (mycophenolic acid, sterigmatocystin, and macrocyclic trichothecenes) of three common fungal indoor contaminants originating in the environment: Penicillium brevicompactum, Aspergillus versicolor , and Stachybotrys chartarum . Considering the outdoor dispersion of any biochemical molecules, the toxicogenic traits of fungi should not represent a relevant threat at the beach. Dematiaceous or melanized fungi are taxa that produce melanin to harness energy from radiation to use in biochemical paths. The most common ailments associated with melanized fungi are keratitis as well as cutaneous, subcutaneous, and respiratory tract infections. Exposure to sand combined with a traumatic event may result in an invasive fungal infection (phaeohyphomycosis). The severity of the infection, though, depends on the extension of the trauma and immune response of the host . Most recently, antifungal resistant Candida auris , an emerging thermo- and halotolerant, multi resistant yeast, has been recognized as an environmental concern in beach environments [ , , ]. The first paper published on beach sand contaminants was in 1960 by Schönfeld, Rieth and Thianprasit, and it described a study which aimed to detect the presence of dermatophytes in a Baltic Sea resort. The study yielded only the geophilic dermatophyte Arthroderma insingulare (formerly Trichophyton terrestre ) in the superficial layers of supratidal sand . In 1973, however, Gertrud Müller in a wider study confirmed the absence of anthropophilic dermatophytes in sands. For this research, Müller’s team sampled Estoril, near Lisboa, the Adriatic Sea, at Gabicce Mare and at Grömitz, on the German Baltic coast. The team searched for Epidermophyton floccosum , an anthropophilic species, for three years, and sampled twice per week (in the same places). They also aimed to find beach usage-associated variations of this fungal species . The study revealed a clear surge in the presence of Epidermophyton floccosum in sands touched by human bare feet. It failed to isolate other species, common in other kinds of soil, showers, and pools such as Trichophyton interdigitale (formerly T. mentagrophytes var. interdigitale ) or for Nannizzia gypsea (formerly Microsporum gypseum ). The authors pointed out in their report that the inability to isolate these common agents of dermatophytosis should not rule out their presence in beach sand that had been in contact with human feet. Maria Colon Valiente, in 1990, hypothesized that the absence of dermatophytes in beach sand was likely due to the low nutrient environment or the high maximum temperatures . These observations, however, were challenged by Sousa , who reported dermatophytes in 11 out of 24 (45.83%) beaches studied in coastal areas around Lisboa. Brandão et al. sampled all regional coasts of Continental Portugal every 2 months, for 13 months, and generated 210 samples of sand from both the supratidal and the vadose zones. The three types were wild beaches, beaches with water quality problems and beaches awarded blue flag status for exceptional quality. Despite their high maintenance levels in relation to the other beach types, blue flag beaches did not exhibit the lowest fungal concentrations in sand. In this study, a higher density of beach users correlated with higher levels of dermatophytes during the summer months. Not only was the presence and concentration of fungi in beach sand highly variable, but the fungal community may have been also evolving due to land use and climate changes. Therefore, additional research and monitoring will be integral to characterizing and maintaining beach sand quality in the context of fungal contaminants. Recently, E. floccosum seems to be decreasing in its prevalence in clinical specimens worldwide, indicating that the likelihood of isolating it from beach sand may be diminished . Other dermatophytes, and even other microbial life forms, may alter their patterns of existence due to clinical intervention and climatic alterations . Oshiro and Fujioka studied the bacteriological water quality of Hanauma Bay in Hawai’i. Due to the extreme human densities associated with recreational use, water quality of Hanauma Bay was difficult to maintain at acceptable levels. The Bay has since been classified a nature preserve and is open to recreation during limited timeframes, to allow the local ecosystem to recover without the threat of constant high human densities . Because of its status as a bay full of marine wildlife, and frequent use by tourists and local beachgoers, the water quality led Oshiro and Fujioka to sample the shoreline water and sand, land runoff, and mongoose and pigeon feces, to try to find the main causes of the surges of FIB. The samples were analyzed for fecal coliforms, Escherichia coli , and enterococci, and revealed that the major sources of the periodically high levels of these bacteria in the water were contaminants founds in beach sand, namely pigeon feces . In this case, water quality was the main driver of the research, although the source came from the sand. In 2017, Argentina became the first nation to formally include sand inspection for rubbish in its water quality standards . Lithuania added monitoring of helminths in sand to their 2007 National regulation within the 2018 updates . The Lithuanian regulations are currently the only known regulation on helminths, as mentioned in WHO . During the last 20 years, several publications have identified potential microbial culture parameters for sand monitoring [ , , , ]. Others contemplate meta-genomics, raising possible alternative parameters, based on nucleic acid analyses [ , , , ]. However, until there is a regulatory document defining sand quality parameters, the choice for monitoring parameters is likely to fall on a combination of culturable fecal indicator organisms (FIO), to indicate pathogens and opportunists associated with wastewaters, and others that do not relate to fecal pollution (including most fungi). Next generation sequencing (NGS) may be an interesting screening tool to assess variations in the microbiota as shown in Taylor and Kurtz . In this study, the authors evaluated three beaches along the Grand Strand of South Carolina, USA. The authors sequenced the V4 region of the 16S rRNA gene to compute relationships between diversity and temporal or local factors. Gammaproteobacteria, Planctomycetes, Acidobacteria, and Actinobacteria were the dominating bacterial populations at these beaches. The communities were similar in overall composition and diversity, but the abundance of taxa changed over time. shows pathogens, opportunistic microorganisms, and fecal indicators in beach sand currently available in the literature. The following three sections describe the current taxa of interest for sand monitoring. WHO addresses three other biological groups of interest that are currently poorly represented in the scientific literature, namely viruses, helminths, and insects. Fecal pollution remains one of the biggest problems with recreational waters and beach sand. Parts of the globe have taken action to eliminate the discharge of untreated sewage into water bodies. However, in other regions, fecal contamination of bathing and even drinking water sources remains a serious problem. One of the most notorious contamination episodes of water took place in 2010 in Haiti, due to the destruction caused by an earthquake of magnitude 7.0. This event killed an estimated 230,000 people and injured another 300,000. Haiti already had a relatively low coverage of sanitation infrastructure but after the earthquake, only 10% of the rural population and 24% of the urban population had access to improved sanitation . This low rate of access to drinking water and sanitation coupled with contamination of the water supply led to an outbreak of cholera, which resulted in 658,563 reported cases and 8111 deaths from the disease as of 2 June 2013 . The WHO guidelines for safe recreational water environments recommends the use of FIB to detect fecal pollution in recreational waters. The more recent guidelines , extend those recommendations to beach sand, precisely to avoid waterborne diseases originating from human excreta. The recommendations focus on measurements of FIB, not the actual pathogens. According to the literature, exposure to 40 CFU/mL of enterococci results in an illness probability of 1%, and up to 200 CFU/100 mL results in up to 5% probability of illness. This indicator parameter in recreational waters suggests general disease probability rather than any specific pathogen-associated illness. It also is currently in use in regulations, e.g., the European Bathing Water Directive . Fujioka et al. , Weiskerger et al. , and Teixeira et al. discuss the need to extend beyond FIB in the future, to accommodate changes arising from climatic alterations. Looking into pathogens directly instead of quantifying indicators is not only increasingly easy, but will also have to be implemented for pathogens that in the future may not be predicted through the use of FIB. As described in WHO , Sabino et al. , and Weiskerger et al. , controlling exposure to bacteria to protect human health needs to go beyond FIB, since the supratidal zone of a beach may not be greatly contaminated with fecal pollution. Instead, its main source of pollution is often skin shedding from animals, vegetable debris, and other organic matter that may serve as food for wildlife . The following bacterial taxa have been considered of relevance for the protection of human health: Staphylococcus spp. (skin infections), Vibrio spp. (cholera and necrotizing fasciitis), Clostridium perfringens (food poisoning, possible cause of bacteraemia), Campylobacter jejuni (gastroenteritis), Shigella spp. (haemorrhagic diarrhoea), and Pseudomonas aeruginosa (superficial and systemic infections) [ , , ]. Methicillin-resistant Staphylococcus aureus (MRSA) was reported in 2009 as present in beach sand and water in California and Florida, USA and more recently in South Africa, in 2015–2016 , presenting yet another emerging contaminant of concern for beachgoers. This group of sand contaminants and climate change implications is addressed further by the WHO and in Weiskerger et al. . Most fungi are opportunistic, which means that exposure is a potential risk only for susceptible individuals. Yet, two groups of fungi are of great concern: dermatophytes and endemic fungi. Dermatophytes are keratinophilic fungi that cause dermatophytosis, superficial infections of the hair, nails, skin, and scalp. There are anthropophilic, geophilic, and zoophilic dermatophytes and they are so classified according to their transmission route: if transmission is human-to-humans, soil-to-human, or animal-to-human . The current knowledge suggests that anthropophilic dermatophytes may be susceptible to the environmental conditions in beach sand and thus may die rather quickly . This does not necessarily mean that there is no transmission at the beach, though. Shedding takes place naturally, especially from infected areas of the skin that become dry and scaly, and death due to radiation or high temperatures is not immediate. In fact, under laboratory conditions, recreational beach sands from Hawai’I were able to maintain several fungal species. In the study, Anderson tested the survival of Cutaneotrichosporon cutaneum/Trichosporon asahii, Candida albicans, Nannizzia gypsea and Trichophyton mentagrophytes (var. mentagrophytes) and all survived for at least one month in non-sterile sand inoculated with keratinized propagules. These data suggest that non-detection of anthropophilic dermatophytes may be associated more with the natural dispersion of propagules and representativeness of sampling, rather than the dermatophyte survival itself. As for the endemic fungi in sensu stricto ( Coccidioides spp., Paracoccidioides brasiliensis , Histoplasma spp., Blastomyces dermatitidis ), no known publications indicate their isolation from beach sand. However, their natural habitats suggest some degree of survival in beaches within endemic areas (inland, mainly). Coccidioides spp. inhabits dry, desert-like territories of the USA, and of Argentina. The remaining fungi are endemic to more humid habitats such as the Mississippi River Valley, USA ( Blastomyces spp.); central and eastern USA ( Histoplasma capsulatum ); Sub-Saharan Africa ( Histoplasma duboisii); and South America ( Paracoccidioides brasiliensis ) . Cryptococcus deuterogattii is also endemic to the Pacific Northwest of the North American continent but not considered dimorphic (real pathogens). There is yet to be consensus whether this species should be considered simply a genetic type of Cryptococcus gatii (type AFLP6/VGIIa). Other fungi of interest, as listed recently by WHO for exposure in natural environments, are Mucorales , allergenic fungi, and dematiaceous fungi. Mucorales is the order that includes the fungi responsible for the invasive mucormycosis. This order has raised some concern due to the opportunistic ability to start an infection in immunocompetent individuals when inoculated deeply under the skin by piercing materials. This has been well demonstrated by a cluster of cases of necrotizing cutaneous mucormycosis following a tornado in Joplin, Missouri, USA in 2011 where several individuals were injured by deep puncture wounds. Mucorales are also a relevant group of fungi in cases of near drowning, as described by Sympardi et al. . However, the distribution of fungal species has been reportedly different in different contexts. Cogliati et al. found different distributions of yeasts and molds in beaches throughout Europe depending on the latitude and heavy metal composition of soil. Yeasts seem to withstand low temperatures better during winter than molds and are therefore more prevalent in Northern European coasts. Moreover, molds are more associated with soil rich in nickel and yeasts with soils rich in cadmium. This results in a distribution of fungi mainly at the deltas of European rivers and lagoons, where these metals accumulate in river sediment. Considering that an allergy is host-dependent and that there are individual allergies to most fungi, for the purpose of this text, all fungi are considered to also cause allergies. There are, however, different types of allergies: respiratory and contact allergies. Respiratory allergies imply inhaling allergens and therefore are mainly caused by airborne-sporulating fungi. The extent to which fungal spores can travel airborne has been clearly described by Kellogg and Griffin , in an open report about the air travel of fungi from Africa to North America. Fungal spores can be airborne and can also extend their presence to an entire beach. Although information on fungal inhalation leading to infections specifically from sand is currently unavailable, Buskirk et al. showed in a murine model that the dry exposure to 10 5 spores of A. fumigatus twice per week triggers an inflammatory response in the lungs within 24 and 48 h. Additionally, an IgG elevation is observed after seven days, concomitant with spore germination. Tanaka et al. found that cytokine release in immunocompetent individuals takes place about 19 h post-inhalation. Cho et al. observed in fungal respiratory deposition models that particles of Stachybotrys chartarum might be deposited in numbers 230–250-fold higher than spores. These data suggest a delayed first response to the exposure to fungal allergens, but an existing one, nonetheless. Beach users might thus not even associate an allergic episode with a visit to the beach the day before. Lastly, exposure to volatile organic compounds (VOCs) produced by fungi like 3-Methylfuran may cause nonspecific symptoms, namely eye, nose and throat irritations, headaches, and fatigue . Aleksic et al. described the difference between the propagation of aerosolized mycotoxins (mycophenolic acid, sterigmatocystin, and macrocyclic trichothecenes) of three common fungal indoor contaminants originating in the environment: Penicillium brevicompactum, Aspergillus versicolor , and Stachybotrys chartarum . Considering the outdoor dispersion of any biochemical molecules, the toxicogenic traits of fungi should not represent a relevant threat at the beach. Dematiaceous or melanized fungi are taxa that produce melanin to harness energy from radiation to use in biochemical paths. The most common ailments associated with melanized fungi are keratitis as well as cutaneous, subcutaneous, and respiratory tract infections. Exposure to sand combined with a traumatic event may result in an invasive fungal infection (phaeohyphomycosis). The severity of the infection, though, depends on the extension of the trauma and immune response of the host . Most recently, antifungal resistant Candida auris , an emerging thermo- and halotolerant, multi resistant yeast, has been recognized as an environmental concern in beach environments [ , , ]. Recreational beaches are typically maintained for health and safety only during official bathing seasons. Sampling outside the bathing season serves academic purposes and can eventually help develop beach management plans to improve beach quality during the bathing season, e.g., cleaning the beach, planning the number of trash-bins, beach nourishment to increase dispersion of sunbathers, mitigation of climate change effects, and removal of animal excrements. Sampling both sand and water seems like the most reasonable plan to manage the beach, allowing a full view of simultaneous snapshots to inter-relate and assess any directionality of distinct pollution in sand and in water together. Regulations should thus recommend sampling both sand and water and leave the decision of sampling outside the bathing season to the beach managers. Water dynamics ensure that pollutants will experience high rates of diffusion facilitating the homogeneity. Sampling water at one site is thus often representative of a large volume. Sand, conversely, is patchy and thus sampling and representativeness should be considered for developing effective monitoring procedures [ , , ]. The sampling frequency and the way samples are collected and transported to a laboratory for processing should be defined with a standard approach that would render monitoring efficient and equivalent between different communities with regard to the comparability of results obtained. Brandão and WHO have addressed this subject, with a sand monitoring recommendation. Brandão et al. described how the swash zone connects water with sand of the intertidal area of a beach. The authors conclude that the microbes from the intertidal sands are represented by nearshore water, which is monitored by default, due to regulation. So, sand only needs to be monitored in the supratidal area. Brewer et al. demonstrated that single grab samples are representative of only the sample under analysis in the lab. That is the reason why an incremental approach (Incremental Sampling Methodology ) for soil sampling, although complex, is highly recommended. In this method, a decision unit that represents a significant area of volume or soil, is sampled in a regular grid pattern. Typically, at least 30 equal-mass samples are collected from the decision unit and replicates can be collected to assist in statistical analysis . For locations with no historical information to help with decision making, Brandão recommends a combination of incremental sampling and previous knowledge of hotspots of microbial contamination. This would allow for sampling of worst-case scenarios and thus render the sampling of sand relevant and representative of a beach. Should incremental sampling not be possible, a composite can be collected for beaches less than a few hundred meters in length. The composite sample would consist of three grab samples spanning the length of the beach. These three grab samples can then be combined, homogenized, and analyzed as one sample . If beaches are more than a few hundred meter in length, the beach should be divided into sub-areas and each sub-area considered a separate beach. Analytical methods are issued by the International Standards Organization (ISO), upon adoption or voted in from of a pool of possible candidate methods, and after a robust screening of all possible scenarios. The European Commission for example, issues Water Quality Directives listing a reference method and proven equivalent methods. Each individual member-state may opt for another unlisted method, if it passes a three-expert panel that contemplates reproducibility, equivalence with the reference method and representativeness. Upon acceptance, the study must be replicated in other member-states that wish to integrate that non-reference method in their own list of options . Different regions of the globe will always present specific features in public health protection from exposure to microbial life because they must address locally endemic pathogens. In the tropics, for example, the existence of percutaneous infection by nematodes, e.g., hookworm and schistosomiasis, can be a serious problem for beach users . 5.1. Two Tiers Analytical Approach Solo-Gabriele et al. recommended a two-tiered approach to sand quality monitoring, aiming to simplify the analytical approaches. Some of the methods described in the literature were extremely laborious and costly, which does not work well for routine analysis and to respond with beach management actions. Thus, the authors recommend a fast routine analysis, able to be performed in water quality laboratories and to escalate to reference laboratories only when necessary to establish the source of an outbreak or to investigate the source of a highly prevalent contaminant. 5.1.1. Analysis during Outbreak Conditions In case of an outbreak, namely the one reported by Brandão et al. , the Reference Mycology Laboratory was engaged to identify the relevant species of fungi. For this study, the fungal species isolated were typically associated with either vegetable matter (colonizers and pathogens) or with faucal contamination, which was indeed the cause of the outbreak. The local team described how a tropical winter storm delivered enormous amounts of near coastal debris into the cove, including high quantities of decaying vegetable matter. The debris was reflected in the fungal analysis, by the considerable presence of Fusarium spp., of Aspergillus section Circundati , and some Candida tropicalis , which are common plant pathogens and colonizers. Other teams have considered analyzing the total microbiome using NGS [ , , ]. This approach requires sophisticated analytical procedures and equipment, and the results are not comparable with culture-based methods. Despite analyzing the DNA present and thus detecting both viable and non-viable individuals, it is a good tool for environmental forensic analyses, since it provides microbial community composition information. One drawback is that the most successful species will be over-represented and may conceal the less represented taxa in NGS assays. Information on routine analytical approaches for viruses (mainly molecular, but also some culture based), protozoa, and helminths is extremely scarce, so they shall not be addressed further. 5.1.2. Quality Assessment Schemes Once the sample of sand arrives at the laboratory, analysis is perform to ensure reproducibility and repeatability regardless of the equipment used. The efficiency of the aqueous extraction of the sand, plating of the extract, plate reading, and registering with the necessary calculations to produce a value of CFU/g are a few examples. Boehm et al. published an article on sand bacterial analysis and performance of participating laboratories in an interlaboratory collaborative study. The authors found that there is variation in extraction efficiency between blending, shaking, and by analyst. No analysts were rejected from the study. Instead, they contributed to establish the natural variation of the results. Rinsing, decanting, and settling before using the eluent did not show any statistically significant variation ( p < 0.05). Participating in an interlaboratory assessment scheme is thus extremely relevant for sand analysis. It is the only tool that can ensure that the results obtained are independent of the analysts while serving as a training tool. Many laboratory analysts work alone and only know how consistently they perform. Distributing samples for interlaboratory assessment schemes implies processing, transporting, and handling of samples in transport to the end-user laboratory. The logistics involved have the potential to alter the sample in many ways. Non-refrigerated transport may either kill or permit multiplication of microorganisms and processing and handling might contaminate samples. Analysis of the joint results will be informative of all these possible sources of variability. The international standard used for determining a consensus value in proficiency testing by interlaboratory comparison is the ISO 13528 . 5.1.3. Outbreak Response In the absence of historical data on culturable microorganisms from beach sand, a primordial full-population-exploratory analysis may be an attractive approach. It would shortcut years of collecting scattered fragments of data to eventually generate a beach profile and be able to decide what is ordinary and what may require attention or management . However, without a history and baseline knowledge of microbial sources at a site, a relative surge of any organism is impossible to confirm. Alternatively, parametric reference values may be used as guidance for never-before tested sands, but these are yet to be published by any regulating agency. WHO provided the first regulatory document with sand microbial parameters and reference values for enterococci and fungi. It chose a total fungal count value from the work of Brandão et al. to be indicative for beach management purposes rather than a parameter value. The historical data of a beach can reveal the results associated with influencing events (e.g., windstorms, heavy rain events and beach festival parties), as described in Brandão and Brandão et al. . In the case of a beach associated outbreak by any microbial agent, an epidemiological study, such as the one described in Brandão et al. , should be conducted. In such an event, the authors recommend the analysis of many types of variables, one of them being if the beach sand might be the cause of the outbreak. To evaluate the cause of the outbreak, bathing was immediately excluded as a contributor to the outbreak because some of the infected patients did not bathe. The only common denominator to the reported macular erythematous pruritic rash outbreak was sand. As the cause was unknown during the investigation, the study was conducted, employing organic chemistry approaches, as well as inorganic chemistry, bacteriology (for FIB) and mycology. The outbreak investigation required cooperative engagement of multi-disciplinary analytical teams. Regardless of the cause of concern, mitigating actions should take place in the face of beach contamination events. In the case of human faucal pollution, there is always a concern of transmission of residual water-borne pathogens, e.g., enteric viruses, pathogenic bacteria, high concentrations of opportunistic fungi, parasites, and anti-microbial resistance genes (ARGs). 5.2. Microbial Source Tracking (MST) In the case of a surge in FIB, the inevitable question is: “Where did it come from?” Sand contaminated by water and vice versa has been extensively addressed in Whitman et al. , who described dispersion, survival, predation on, and ultimately, the fate of the FIB in sand and water. However, in addition, the ultimate origin of the FIB is important to determine. Naturally, FIB that indicate human excreta is a clear warning that human pathogens may be present and cause harm to humans. These are the most relevant FIB for human health, but not the only ones. Other FIB that indicate non-human excreta or different biological groups may indicate possible zoonoses (like bird flu), haemorrhagic E. coli , and ARGs. It is thus desirable to be able to track the origin of FIB in sand and in water during a surge, and hopefully mitigate it upstream. In 2000, Bernhard and Field developed a way to characterize fecal contamination from cows and humans, based on the 16S subunit of the ribosomal DNA, of the genus Bifidobacterium and the Bacteroides - Prevotella group. They reported PCR primers that differentiated both specific biological groups. The aim of their research was mainly to help identify and mitigate diffuse (non-point-source) fecal pollution in water, which tends to be associated mainly with run-off carrying fecal contamination from cattle in wet regions of the most developed parts of the world. Microbial Source Tracking (MST) tools have since been extensively studied and applied, especially to discriminate the origin of water pollution events. The main objective of this method is the differentiation between human and non-human fecal contamination sources. Due to inter-species barriers to transmission, pathogens of human excreta represent the main public health concern. Initially, MST was mainly based in library-dependent methods, where a database of fecal samples from known hosts were typed on an isolate-by-isolate basis and compared to the fecal sample under analysis. Currently, library-independent methods using PCR or quantitative PCR (qPCR) to target specific genes of host-associated bacteria are preferred . This latter approach is possible due to the extensive effort in developing primers sets that are specific to several biological groups, including humans , seagulls , cattle , and dogs . There are only a few sporadic reports using this methodology to unravel the contaminations source(s) in sand , however its potential has widespread support and should be further explored. Valério et al. , for instance, assessed a suspected case of fecal contamination of unknown origin that resulted from multiple contributions (horses, seagulls, and dogs). This study is a clear example of how using an MST approach for sand can help elucidate the different contamination sources present at a beach. Other methods for MST, namely culture independent technologies, are also in use given the advance of real-time qPCR and especially using next generation sequencing (NGS) based DNA analyses, but the financial and computational costs of these methods can be high . 5.3. Analyzing Microbes in Beach Sand 5.3.1. Fecal Indicator Bacteria (FIB) The FIB in sands can be analyzed by standard membrane filtration by plating filters on agar media, or by chromogenic substrate. For the chromogenic substrate, Colilert ® and Enterolert ® from IDEXX Laboratories, Inc. (Westbrook, ME, USA) represent the most common system used. The principle is to extract the FIB by shaking the sand sample immersed in distilled water, followed by processing the extract as though it is a water sample. The analysis returns the most probable number (MPN/100 mL) of coliforms, E. coli and enterococci, which must be reversed to the dilution used in the extraction. In terms of details, Sabino et al. used a 1:10 extraction (sand to distilled water) and 30 min circular shake at 50 rotations per minute (RPM). Boehm et al. recommends a faster approach by extracting 10 g of sand with 100 mL of distilled water or phosphate buffered saline (PBS), followed by membrane filtration of the eluent as if it were water to yield colony forming units (CFU) per 100 mL. If source tracking is in order, a detailed protocol for MST analyses has recently been described by Valério et al. . 5.3.2. Detection of Other Bacteria Bacteria, in general, may be difficult to isolate depending on the level of contaminants within an environmental sample. Nonetheless, selective media and specific temperatures can be useful to isolate bacteria of interest. Following the same system as for FIB, Pseudolert ® (also manufactured by IDEXX Laboratories, Inc., Westbrook, ME, USA) can be used for the detection and count of Pseudomonas aeruginosa . Other bacteria necessitate specific isolation techniques separate from those manufactured by IDEXX. The scientific literature is scarce on bacteria other than FIB. However, in 2009, Goodwin and Pobuda studied the presence of MRSA by performing a novel procedure. Two grams of sand was vigorously hand shaken in 10 mL of PBS before being vacuum filtered through a sterile 30 μm, 47 mm nylon net filter (Millipore, Beford, MA, USA). A 10 mL PBS rinse removed any residual sand from the shaking container. This process was repeated until enough sand-water solution was generated for membrane filtration. The sand–water solution was also homogenized via hand mixing prior to filtration. After filtration, the filters were incubated on SCA or C-MRSA selective and differential media for S. aureus and MRSA, respectively (BD Biosciences, San Jose, CA, USA) . Recently, a review on ESKAPE pathogens ( Enterococcus faecium , Staphylococcus aureus , Klebsiella pneumoniae , Acinetobacter baumannii , Pseudomonas aeruginosa , and Enterobacter spp.) in environmental reservoirs, such as surface water, wastewater, food, and soil, addressed these isolations from beach sand . Particularly, Akanbi et al. isolated MRSA from beach sand in the Eastern Cape Province of South Africa in 2015–2016, followed by isolation and molecular identification, and testing resistances via the Kirby-Bauer disk diffusion method on Mueller-Hinton agar . 5.3.3. Isolation of Fungi Sabino et al. describes analytical methods for fungi, whereby sand is extracted with water in a 1:1 ratio, in low energy, shaking orbitally at 100 rpm for 30 min. Sample triplicates are then plated. Malt yeast agar with chloramphenicol is used for all species and Mycosel agar (with chloramphenicol and cycloheximide), specifically for dermatophytes. Sample plates are then incubated for 5 and 15 days, respectively, at 27.5 °C. After incubation, plates undergo tentative colony identification via picking and counting one colony of each morphotype, and dividing them into yeast-like species, opportunistic and allergenic species, and dermatophytes as three parameters to assess fungi. The result is the average count of the triplicates for each parameter, per gram of sand. This method is rather laborious and time consuming but provides a complete analysis of the culturable mycobiota in a sand sample. Unlike yeasts that form mainly budding cells, molds usually occur as hyphae. If broken, any hypha will start a new colony. It is thus truly relevant how to approach fungal analyses of sand, considering that there will always be a trade-off between aqueous extraction from sand and breakage of the hyphae during the process due to vigorous agitation. In light of this trade-off, Sabino et al. opted for orbital shaking with a speed of 100 rpm. The same laboratory is still doing so today, followed by plating in malt yeast agar, supplemented with chloramphenicol (0.05%), for all fungi and Mycosel agar, supplemented with chloramphenicol and cycloheximide. The latter is used specifically for dermatophytes, given the growth speed reduction of fast-growing fungi, which allows dermatophytes to grow and be visible, instead of being overtaken by fast growing molds. In mycology, it is necessary to use a medium to recover as much of the fungi of interest as possible. Sabouraud with chloramphenicol is the traditional medium of choice as it is not selective. Yet, samples with an abundant presence of Mucorales may require the use of a medium with Dichloran Glycerol Agar combined with Rose Bengal to inhibit their excessive growth . This is frequently the case with inland beaches due to the stronger presence of vegetable matter since many species of this order are plant pathogens . Coastal beach sands may also yield some isolates of Mucorales but usually only when highly contaminated with fungi from many species. Incubation needs to extend long enough to allow as many fungal colonies as possible to become visible, but not too long to have the faster growing ones cover the slower growing ones. Again, there are trade-offs for isolating mixed cultures of unknown species from environmental samples: many dematiaceous fungi are slow growing, including dermatophytes and Exophiala spp. , which cause phaeohyphomycosis. Conversely, Mucorales and Trichoderma spp. are extremely fast growers . Moreover, yeasts grow by cell division on a growth medium, not by releasing hyphae. Hence, molds spread onto the culture media while yeast colonies have a more restricted growth. In addition, there is competition between both groups. For example, the statins , used for pharmaceutical purposes in humans to lower cholesterol levels in blood, are in fact metabolites produced by molds, intended to slow down the growth of yeasts and even of other species of molds. The statins interfere with the production of ergosterol, necessary for their growth . The inoculum from an environmental sample should thus be diluted enough to allow all species to grow without confluence, during a long enough incubation period, and at a temperature that is appropriate for the intended purpose. However, dilution of the inoculum to ensure a robust analytical disposition of all colonies will inevitably lead to the loss of species present at lower levels . Sand analysis for health protection should target a temperature that matches that of the surface of the human body. The approach used in bacteriology of selecting the pathogens from the bacteria that can grow at 37 °C does not apply to Mycology. Keratinophilic fungi infect keratinized tissue, including skin, nails, and hair. A temperature of 27.5 °C will allow all medically relevant fungi to grow . Fungi in water were addressed in Brandão et al. , but the conclusion of the study was that more data were needed before recommendations could be issued. The same team is preparing to address this data gap in the future. 5.3.4. Identification and Taxonomic Classification of Fungi Fungal taxonomy is currently undergoing a revolution, mainly due to new data arising from the use of molecular biology techniques . Still, the primary approach tends to remain identification by micro and macro characteristics of the colony. Some of the common fungi found in the beach environment are readily recognizable at first glance. Microscopic verification of the typical structures may help to distinguish, for example, some Penicillium and Aspergillus section Fumigati species. Yeasts, however, require additional biochemical testing to be differentiated. Sabino et al. described in detail how sand analysis can be performed for the identification of the fungi and bacteria. Although the same institution currently identifies many of the fungi by matrix-assisted laser desorption/ionization-time of flight (MALDI-ToF) or molecular approaches based on sequencing of the ITS1 and ITS2 regions of the ribosomal DNA. MALDI-ToF is fast and requires little handling before an acceptable identification is achieved. Molecular identification tends to be broader in the number of species that can be identified, but it is more costly and labor intensive. The process requires DNA extraction, purification, and amplification with the primers chosen for the identification intended by PCR, cleaning of the PCR product of the amplification, followed by sequencing of this product and searching for and aligning (blasting) the sequences with DNA databases . This system allows the clear distinction of many cryptic species, depending on the primers chosen to help amplify the target DNA, as shown in the study of Novak Babič et al. . 5.4. Analyzing Insects and Helminths in Beach Sand There is currently no methodology published for detecting insects in beach sand. However, WHO lists the main groups considered of interest, which are all Diptera: mosquitoes, biting midges, sand-flies, flies, and blackflies. When flies and blackflies occur in large numbers, they are a nuisance and mosquito, midge, and sand-fly bites can be painful in addition to being vectors of disease for humans and animals. Mosquito numbers typically increase during wet weather or following tide triggers and they are often most active at dawn. It is recommended that beachgoers try to avoid exposure outdoors at sunset and overnight; they should wear long sleeves and use insect repellent. As mosquitoes breed in standing water, getting rid of devices that hold water is a simple way to stop them from breeding . Biting midges are a common nuisance along certain coastal areas and are most active in intertidal zones, including canals, rivers, and estuaries. Sand-flies tend to be found on beaches where the sand is slightly earthier and are usually a problem when people lie directly on the sand for long periods of time. To help reduce nesting sites, it is advisable for authorities to remove algae and other debris brought to sands by storms. As biting midges and sand-flies are sub-optimal fliers, effective protection can be provided through the use of fans in public places to prevent bites, combined with repellents and loose clothing . Nuisance flies are attracted to dead animals, feces, and garbage, allowing them to spread a variety of disease-causing bacteria and parasites. Some flies can deliver painful bites and transmit diseases to humans and animals. Good sanitation, by eliminating flies’ breeding sites, is the main fly control recommendation at beach facilities. Thus, garbage containers must always be closed and fecal matter from dogs and other animals should be eliminated. When these dipterans become pests, public health entities and municipalities can implement chemical control programs with insecticides, such as pyrethrins and pyrethroids , while preventing the development of potential insect resistance by alternating the products. Helminths are currently monitored only in Lithuania, where the beach sand is tested once before the start of the bathing season and at least four times during the bathing season. The procedure is described in detail by the Ministry of Health of the Republic of Lithuania . Ideally, this regulation should be tested and implemented in other countries as part of beach sand monitoring activities with relevance for human health protection. Solo-Gabriele et al. recommended a two-tiered approach to sand quality monitoring, aiming to simplify the analytical approaches. Some of the methods described in the literature were extremely laborious and costly, which does not work well for routine analysis and to respond with beach management actions. Thus, the authors recommend a fast routine analysis, able to be performed in water quality laboratories and to escalate to reference laboratories only when necessary to establish the source of an outbreak or to investigate the source of a highly prevalent contaminant. 5.1.1. Analysis during Outbreak Conditions In case of an outbreak, namely the one reported by Brandão et al. , the Reference Mycology Laboratory was engaged to identify the relevant species of fungi. For this study, the fungal species isolated were typically associated with either vegetable matter (colonizers and pathogens) or with faucal contamination, which was indeed the cause of the outbreak. The local team described how a tropical winter storm delivered enormous amounts of near coastal debris into the cove, including high quantities of decaying vegetable matter. The debris was reflected in the fungal analysis, by the considerable presence of Fusarium spp., of Aspergillus section Circundati , and some Candida tropicalis , which are common plant pathogens and colonizers. Other teams have considered analyzing the total microbiome using NGS [ , , ]. This approach requires sophisticated analytical procedures and equipment, and the results are not comparable with culture-based methods. Despite analyzing the DNA present and thus detecting both viable and non-viable individuals, it is a good tool for environmental forensic analyses, since it provides microbial community composition information. One drawback is that the most successful species will be over-represented and may conceal the less represented taxa in NGS assays. Information on routine analytical approaches for viruses (mainly molecular, but also some culture based), protozoa, and helminths is extremely scarce, so they shall not be addressed further. 5.1.2. Quality Assessment Schemes Once the sample of sand arrives at the laboratory, analysis is perform to ensure reproducibility and repeatability regardless of the equipment used. The efficiency of the aqueous extraction of the sand, plating of the extract, plate reading, and registering with the necessary calculations to produce a value of CFU/g are a few examples. Boehm et al. published an article on sand bacterial analysis and performance of participating laboratories in an interlaboratory collaborative study. The authors found that there is variation in extraction efficiency between blending, shaking, and by analyst. No analysts were rejected from the study. Instead, they contributed to establish the natural variation of the results. Rinsing, decanting, and settling before using the eluent did not show any statistically significant variation ( p < 0.05). Participating in an interlaboratory assessment scheme is thus extremely relevant for sand analysis. It is the only tool that can ensure that the results obtained are independent of the analysts while serving as a training tool. Many laboratory analysts work alone and only know how consistently they perform. Distributing samples for interlaboratory assessment schemes implies processing, transporting, and handling of samples in transport to the end-user laboratory. The logistics involved have the potential to alter the sample in many ways. Non-refrigerated transport may either kill or permit multiplication of microorganisms and processing and handling might contaminate samples. Analysis of the joint results will be informative of all these possible sources of variability. The international standard used for determining a consensus value in proficiency testing by interlaboratory comparison is the ISO 13528 . 5.1.3. Outbreak Response In the absence of historical data on culturable microorganisms from beach sand, a primordial full-population-exploratory analysis may be an attractive approach. It would shortcut years of collecting scattered fragments of data to eventually generate a beach profile and be able to decide what is ordinary and what may require attention or management . However, without a history and baseline knowledge of microbial sources at a site, a relative surge of any organism is impossible to confirm. Alternatively, parametric reference values may be used as guidance for never-before tested sands, but these are yet to be published by any regulating agency. WHO provided the first regulatory document with sand microbial parameters and reference values for enterococci and fungi. It chose a total fungal count value from the work of Brandão et al. to be indicative for beach management purposes rather than a parameter value. The historical data of a beach can reveal the results associated with influencing events (e.g., windstorms, heavy rain events and beach festival parties), as described in Brandão and Brandão et al. . In the case of a beach associated outbreak by any microbial agent, an epidemiological study, such as the one described in Brandão et al. , should be conducted. In such an event, the authors recommend the analysis of many types of variables, one of them being if the beach sand might be the cause of the outbreak. To evaluate the cause of the outbreak, bathing was immediately excluded as a contributor to the outbreak because some of the infected patients did not bathe. The only common denominator to the reported macular erythematous pruritic rash outbreak was sand. As the cause was unknown during the investigation, the study was conducted, employing organic chemistry approaches, as well as inorganic chemistry, bacteriology (for FIB) and mycology. The outbreak investigation required cooperative engagement of multi-disciplinary analytical teams. Regardless of the cause of concern, mitigating actions should take place in the face of beach contamination events. In the case of human faucal pollution, there is always a concern of transmission of residual water-borne pathogens, e.g., enteric viruses, pathogenic bacteria, high concentrations of opportunistic fungi, parasites, and anti-microbial resistance genes (ARGs). In case of an outbreak, namely the one reported by Brandão et al. , the Reference Mycology Laboratory was engaged to identify the relevant species of fungi. For this study, the fungal species isolated were typically associated with either vegetable matter (colonizers and pathogens) or with faucal contamination, which was indeed the cause of the outbreak. The local team described how a tropical winter storm delivered enormous amounts of near coastal debris into the cove, including high quantities of decaying vegetable matter. The debris was reflected in the fungal analysis, by the considerable presence of Fusarium spp., of Aspergillus section Circundati , and some Candida tropicalis , which are common plant pathogens and colonizers. Other teams have considered analyzing the total microbiome using NGS [ , , ]. This approach requires sophisticated analytical procedures and equipment, and the results are not comparable with culture-based methods. Despite analyzing the DNA present and thus detecting both viable and non-viable individuals, it is a good tool for environmental forensic analyses, since it provides microbial community composition information. One drawback is that the most successful species will be over-represented and may conceal the less represented taxa in NGS assays. Information on routine analytical approaches for viruses (mainly molecular, but also some culture based), protozoa, and helminths is extremely scarce, so they shall not be addressed further. Once the sample of sand arrives at the laboratory, analysis is perform to ensure reproducibility and repeatability regardless of the equipment used. The efficiency of the aqueous extraction of the sand, plating of the extract, plate reading, and registering with the necessary calculations to produce a value of CFU/g are a few examples. Boehm et al. published an article on sand bacterial analysis and performance of participating laboratories in an interlaboratory collaborative study. The authors found that there is variation in extraction efficiency between blending, shaking, and by analyst. No analysts were rejected from the study. Instead, they contributed to establish the natural variation of the results. Rinsing, decanting, and settling before using the eluent did not show any statistically significant variation ( p < 0.05). Participating in an interlaboratory assessment scheme is thus extremely relevant for sand analysis. It is the only tool that can ensure that the results obtained are independent of the analysts while serving as a training tool. Many laboratory analysts work alone and only know how consistently they perform. Distributing samples for interlaboratory assessment schemes implies processing, transporting, and handling of samples in transport to the end-user laboratory. The logistics involved have the potential to alter the sample in many ways. Non-refrigerated transport may either kill or permit multiplication of microorganisms and processing and handling might contaminate samples. Analysis of the joint results will be informative of all these possible sources of variability. The international standard used for determining a consensus value in proficiency testing by interlaboratory comparison is the ISO 13528 . In the absence of historical data on culturable microorganisms from beach sand, a primordial full-population-exploratory analysis may be an attractive approach. It would shortcut years of collecting scattered fragments of data to eventually generate a beach profile and be able to decide what is ordinary and what may require attention or management . However, without a history and baseline knowledge of microbial sources at a site, a relative surge of any organism is impossible to confirm. Alternatively, parametric reference values may be used as guidance for never-before tested sands, but these are yet to be published by any regulating agency. WHO provided the first regulatory document with sand microbial parameters and reference values for enterococci and fungi. It chose a total fungal count value from the work of Brandão et al. to be indicative for beach management purposes rather than a parameter value. The historical data of a beach can reveal the results associated with influencing events (e.g., windstorms, heavy rain events and beach festival parties), as described in Brandão and Brandão et al. . In the case of a beach associated outbreak by any microbial agent, an epidemiological study, such as the one described in Brandão et al. , should be conducted. In such an event, the authors recommend the analysis of many types of variables, one of them being if the beach sand might be the cause of the outbreak. To evaluate the cause of the outbreak, bathing was immediately excluded as a contributor to the outbreak because some of the infected patients did not bathe. The only common denominator to the reported macular erythematous pruritic rash outbreak was sand. As the cause was unknown during the investigation, the study was conducted, employing organic chemistry approaches, as well as inorganic chemistry, bacteriology (for FIB) and mycology. The outbreak investigation required cooperative engagement of multi-disciplinary analytical teams. Regardless of the cause of concern, mitigating actions should take place in the face of beach contamination events. In the case of human faucal pollution, there is always a concern of transmission of residual water-borne pathogens, e.g., enteric viruses, pathogenic bacteria, high concentrations of opportunistic fungi, parasites, and anti-microbial resistance genes (ARGs). In the case of a surge in FIB, the inevitable question is: “Where did it come from?” Sand contaminated by water and vice versa has been extensively addressed in Whitman et al. , who described dispersion, survival, predation on, and ultimately, the fate of the FIB in sand and water. However, in addition, the ultimate origin of the FIB is important to determine. Naturally, FIB that indicate human excreta is a clear warning that human pathogens may be present and cause harm to humans. These are the most relevant FIB for human health, but not the only ones. Other FIB that indicate non-human excreta or different biological groups may indicate possible zoonoses (like bird flu), haemorrhagic E. coli , and ARGs. It is thus desirable to be able to track the origin of FIB in sand and in water during a surge, and hopefully mitigate it upstream. In 2000, Bernhard and Field developed a way to characterize fecal contamination from cows and humans, based on the 16S subunit of the ribosomal DNA, of the genus Bifidobacterium and the Bacteroides - Prevotella group. They reported PCR primers that differentiated both specific biological groups. The aim of their research was mainly to help identify and mitigate diffuse (non-point-source) fecal pollution in water, which tends to be associated mainly with run-off carrying fecal contamination from cattle in wet regions of the most developed parts of the world. Microbial Source Tracking (MST) tools have since been extensively studied and applied, especially to discriminate the origin of water pollution events. The main objective of this method is the differentiation between human and non-human fecal contamination sources. Due to inter-species barriers to transmission, pathogens of human excreta represent the main public health concern. Initially, MST was mainly based in library-dependent methods, where a database of fecal samples from known hosts were typed on an isolate-by-isolate basis and compared to the fecal sample under analysis. Currently, library-independent methods using PCR or quantitative PCR (qPCR) to target specific genes of host-associated bacteria are preferred . This latter approach is possible due to the extensive effort in developing primers sets that are specific to several biological groups, including humans , seagulls , cattle , and dogs . There are only a few sporadic reports using this methodology to unravel the contaminations source(s) in sand , however its potential has widespread support and should be further explored. Valério et al. , for instance, assessed a suspected case of fecal contamination of unknown origin that resulted from multiple contributions (horses, seagulls, and dogs). This study is a clear example of how using an MST approach for sand can help elucidate the different contamination sources present at a beach. Other methods for MST, namely culture independent technologies, are also in use given the advance of real-time qPCR and especially using next generation sequencing (NGS) based DNA analyses, but the financial and computational costs of these methods can be high . 5.3.1. Fecal Indicator Bacteria (FIB) The FIB in sands can be analyzed by standard membrane filtration by plating filters on agar media, or by chromogenic substrate. For the chromogenic substrate, Colilert ® and Enterolert ® from IDEXX Laboratories, Inc. (Westbrook, ME, USA) represent the most common system used. The principle is to extract the FIB by shaking the sand sample immersed in distilled water, followed by processing the extract as though it is a water sample. The analysis returns the most probable number (MPN/100 mL) of coliforms, E. coli and enterococci, which must be reversed to the dilution used in the extraction. In terms of details, Sabino et al. used a 1:10 extraction (sand to distilled water) and 30 min circular shake at 50 rotations per minute (RPM). Boehm et al. recommends a faster approach by extracting 10 g of sand with 100 mL of distilled water or phosphate buffered saline (PBS), followed by membrane filtration of the eluent as if it were water to yield colony forming units (CFU) per 100 mL. If source tracking is in order, a detailed protocol for MST analyses has recently been described by Valério et al. . 5.3.2. Detection of Other Bacteria Bacteria, in general, may be difficult to isolate depending on the level of contaminants within an environmental sample. Nonetheless, selective media and specific temperatures can be useful to isolate bacteria of interest. Following the same system as for FIB, Pseudolert ® (also manufactured by IDEXX Laboratories, Inc., Westbrook, ME, USA) can be used for the detection and count of Pseudomonas aeruginosa . Other bacteria necessitate specific isolation techniques separate from those manufactured by IDEXX. The scientific literature is scarce on bacteria other than FIB. However, in 2009, Goodwin and Pobuda studied the presence of MRSA by performing a novel procedure. Two grams of sand was vigorously hand shaken in 10 mL of PBS before being vacuum filtered through a sterile 30 μm, 47 mm nylon net filter (Millipore, Beford, MA, USA). A 10 mL PBS rinse removed any residual sand from the shaking container. This process was repeated until enough sand-water solution was generated for membrane filtration. The sand–water solution was also homogenized via hand mixing prior to filtration. After filtration, the filters were incubated on SCA or C-MRSA selective and differential media for S. aureus and MRSA, respectively (BD Biosciences, San Jose, CA, USA) . Recently, a review on ESKAPE pathogens ( Enterococcus faecium , Staphylococcus aureus , Klebsiella pneumoniae , Acinetobacter baumannii , Pseudomonas aeruginosa , and Enterobacter spp.) in environmental reservoirs, such as surface water, wastewater, food, and soil, addressed these isolations from beach sand . Particularly, Akanbi et al. isolated MRSA from beach sand in the Eastern Cape Province of South Africa in 2015–2016, followed by isolation and molecular identification, and testing resistances via the Kirby-Bauer disk diffusion method on Mueller-Hinton agar . 5.3.3. Isolation of Fungi Sabino et al. describes analytical methods for fungi, whereby sand is extracted with water in a 1:1 ratio, in low energy, shaking orbitally at 100 rpm for 30 min. Sample triplicates are then plated. Malt yeast agar with chloramphenicol is used for all species and Mycosel agar (with chloramphenicol and cycloheximide), specifically for dermatophytes. Sample plates are then incubated for 5 and 15 days, respectively, at 27.5 °C. After incubation, plates undergo tentative colony identification via picking and counting one colony of each morphotype, and dividing them into yeast-like species, opportunistic and allergenic species, and dermatophytes as three parameters to assess fungi. The result is the average count of the triplicates for each parameter, per gram of sand. This method is rather laborious and time consuming but provides a complete analysis of the culturable mycobiota in a sand sample. Unlike yeasts that form mainly budding cells, molds usually occur as hyphae. If broken, any hypha will start a new colony. It is thus truly relevant how to approach fungal analyses of sand, considering that there will always be a trade-off between aqueous extraction from sand and breakage of the hyphae during the process due to vigorous agitation. In light of this trade-off, Sabino et al. opted for orbital shaking with a speed of 100 rpm. The same laboratory is still doing so today, followed by plating in malt yeast agar, supplemented with chloramphenicol (0.05%), for all fungi and Mycosel agar, supplemented with chloramphenicol and cycloheximide. The latter is used specifically for dermatophytes, given the growth speed reduction of fast-growing fungi, which allows dermatophytes to grow and be visible, instead of being overtaken by fast growing molds. In mycology, it is necessary to use a medium to recover as much of the fungi of interest as possible. Sabouraud with chloramphenicol is the traditional medium of choice as it is not selective. Yet, samples with an abundant presence of Mucorales may require the use of a medium with Dichloran Glycerol Agar combined with Rose Bengal to inhibit their excessive growth . This is frequently the case with inland beaches due to the stronger presence of vegetable matter since many species of this order are plant pathogens . Coastal beach sands may also yield some isolates of Mucorales but usually only when highly contaminated with fungi from many species. Incubation needs to extend long enough to allow as many fungal colonies as possible to become visible, but not too long to have the faster growing ones cover the slower growing ones. Again, there are trade-offs for isolating mixed cultures of unknown species from environmental samples: many dematiaceous fungi are slow growing, including dermatophytes and Exophiala spp. , which cause phaeohyphomycosis. Conversely, Mucorales and Trichoderma spp. are extremely fast growers . Moreover, yeasts grow by cell division on a growth medium, not by releasing hyphae. Hence, molds spread onto the culture media while yeast colonies have a more restricted growth. In addition, there is competition between both groups. For example, the statins , used for pharmaceutical purposes in humans to lower cholesterol levels in blood, are in fact metabolites produced by molds, intended to slow down the growth of yeasts and even of other species of molds. The statins interfere with the production of ergosterol, necessary for their growth . The inoculum from an environmental sample should thus be diluted enough to allow all species to grow without confluence, during a long enough incubation period, and at a temperature that is appropriate for the intended purpose. However, dilution of the inoculum to ensure a robust analytical disposition of all colonies will inevitably lead to the loss of species present at lower levels . Sand analysis for health protection should target a temperature that matches that of the surface of the human body. The approach used in bacteriology of selecting the pathogens from the bacteria that can grow at 37 °C does not apply to Mycology. Keratinophilic fungi infect keratinized tissue, including skin, nails, and hair. A temperature of 27.5 °C will allow all medically relevant fungi to grow . Fungi in water were addressed in Brandão et al. , but the conclusion of the study was that more data were needed before recommendations could be issued. The same team is preparing to address this data gap in the future. 5.3.4. Identification and Taxonomic Classification of Fungi Fungal taxonomy is currently undergoing a revolution, mainly due to new data arising from the use of molecular biology techniques . Still, the primary approach tends to remain identification by micro and macro characteristics of the colony. Some of the common fungi found in the beach environment are readily recognizable at first glance. Microscopic verification of the typical structures may help to distinguish, for example, some Penicillium and Aspergillus section Fumigati species. Yeasts, however, require additional biochemical testing to be differentiated. Sabino et al. described in detail how sand analysis can be performed for the identification of the fungi and bacteria. Although the same institution currently identifies many of the fungi by matrix-assisted laser desorption/ionization-time of flight (MALDI-ToF) or molecular approaches based on sequencing of the ITS1 and ITS2 regions of the ribosomal DNA. MALDI-ToF is fast and requires little handling before an acceptable identification is achieved. Molecular identification tends to be broader in the number of species that can be identified, but it is more costly and labor intensive. The process requires DNA extraction, purification, and amplification with the primers chosen for the identification intended by PCR, cleaning of the PCR product of the amplification, followed by sequencing of this product and searching for and aligning (blasting) the sequences with DNA databases . This system allows the clear distinction of many cryptic species, depending on the primers chosen to help amplify the target DNA, as shown in the study of Novak Babič et al. . The FIB in sands can be analyzed by standard membrane filtration by plating filters on agar media, or by chromogenic substrate. For the chromogenic substrate, Colilert ® and Enterolert ® from IDEXX Laboratories, Inc. (Westbrook, ME, USA) represent the most common system used. The principle is to extract the FIB by shaking the sand sample immersed in distilled water, followed by processing the extract as though it is a water sample. The analysis returns the most probable number (MPN/100 mL) of coliforms, E. coli and enterococci, which must be reversed to the dilution used in the extraction. In terms of details, Sabino et al. used a 1:10 extraction (sand to distilled water) and 30 min circular shake at 50 rotations per minute (RPM). Boehm et al. recommends a faster approach by extracting 10 g of sand with 100 mL of distilled water or phosphate buffered saline (PBS), followed by membrane filtration of the eluent as if it were water to yield colony forming units (CFU) per 100 mL. If source tracking is in order, a detailed protocol for MST analyses has recently been described by Valério et al. . Bacteria, in general, may be difficult to isolate depending on the level of contaminants within an environmental sample. Nonetheless, selective media and specific temperatures can be useful to isolate bacteria of interest. Following the same system as for FIB, Pseudolert ® (also manufactured by IDEXX Laboratories, Inc., Westbrook, ME, USA) can be used for the detection and count of Pseudomonas aeruginosa . Other bacteria necessitate specific isolation techniques separate from those manufactured by IDEXX. The scientific literature is scarce on bacteria other than FIB. However, in 2009, Goodwin and Pobuda studied the presence of MRSA by performing a novel procedure. Two grams of sand was vigorously hand shaken in 10 mL of PBS before being vacuum filtered through a sterile 30 μm, 47 mm nylon net filter (Millipore, Beford, MA, USA). A 10 mL PBS rinse removed any residual sand from the shaking container. This process was repeated until enough sand-water solution was generated for membrane filtration. The sand–water solution was also homogenized via hand mixing prior to filtration. After filtration, the filters were incubated on SCA or C-MRSA selective and differential media for S. aureus and MRSA, respectively (BD Biosciences, San Jose, CA, USA) . Recently, a review on ESKAPE pathogens ( Enterococcus faecium , Staphylococcus aureus , Klebsiella pneumoniae , Acinetobacter baumannii , Pseudomonas aeruginosa , and Enterobacter spp.) in environmental reservoirs, such as surface water, wastewater, food, and soil, addressed these isolations from beach sand . Particularly, Akanbi et al. isolated MRSA from beach sand in the Eastern Cape Province of South Africa in 2015–2016, followed by isolation and molecular identification, and testing resistances via the Kirby-Bauer disk diffusion method on Mueller-Hinton agar . Sabino et al. describes analytical methods for fungi, whereby sand is extracted with water in a 1:1 ratio, in low energy, shaking orbitally at 100 rpm for 30 min. Sample triplicates are then plated. Malt yeast agar with chloramphenicol is used for all species and Mycosel agar (with chloramphenicol and cycloheximide), specifically for dermatophytes. Sample plates are then incubated for 5 and 15 days, respectively, at 27.5 °C. After incubation, plates undergo tentative colony identification via picking and counting one colony of each morphotype, and dividing them into yeast-like species, opportunistic and allergenic species, and dermatophytes as three parameters to assess fungi. The result is the average count of the triplicates for each parameter, per gram of sand. This method is rather laborious and time consuming but provides a complete analysis of the culturable mycobiota in a sand sample. Unlike yeasts that form mainly budding cells, molds usually occur as hyphae. If broken, any hypha will start a new colony. It is thus truly relevant how to approach fungal analyses of sand, considering that there will always be a trade-off between aqueous extraction from sand and breakage of the hyphae during the process due to vigorous agitation. In light of this trade-off, Sabino et al. opted for orbital shaking with a speed of 100 rpm. The same laboratory is still doing so today, followed by plating in malt yeast agar, supplemented with chloramphenicol (0.05%), for all fungi and Mycosel agar, supplemented with chloramphenicol and cycloheximide. The latter is used specifically for dermatophytes, given the growth speed reduction of fast-growing fungi, which allows dermatophytes to grow and be visible, instead of being overtaken by fast growing molds. In mycology, it is necessary to use a medium to recover as much of the fungi of interest as possible. Sabouraud with chloramphenicol is the traditional medium of choice as it is not selective. Yet, samples with an abundant presence of Mucorales may require the use of a medium with Dichloran Glycerol Agar combined with Rose Bengal to inhibit their excessive growth . This is frequently the case with inland beaches due to the stronger presence of vegetable matter since many species of this order are plant pathogens . Coastal beach sands may also yield some isolates of Mucorales but usually only when highly contaminated with fungi from many species. Incubation needs to extend long enough to allow as many fungal colonies as possible to become visible, but not too long to have the faster growing ones cover the slower growing ones. Again, there are trade-offs for isolating mixed cultures of unknown species from environmental samples: many dematiaceous fungi are slow growing, including dermatophytes and Exophiala spp. , which cause phaeohyphomycosis. Conversely, Mucorales and Trichoderma spp. are extremely fast growers . Moreover, yeasts grow by cell division on a growth medium, not by releasing hyphae. Hence, molds spread onto the culture media while yeast colonies have a more restricted growth. In addition, there is competition between both groups. For example, the statins , used for pharmaceutical purposes in humans to lower cholesterol levels in blood, are in fact metabolites produced by molds, intended to slow down the growth of yeasts and even of other species of molds. The statins interfere with the production of ergosterol, necessary for their growth . The inoculum from an environmental sample should thus be diluted enough to allow all species to grow without confluence, during a long enough incubation period, and at a temperature that is appropriate for the intended purpose. However, dilution of the inoculum to ensure a robust analytical disposition of all colonies will inevitably lead to the loss of species present at lower levels . Sand analysis for health protection should target a temperature that matches that of the surface of the human body. The approach used in bacteriology of selecting the pathogens from the bacteria that can grow at 37 °C does not apply to Mycology. Keratinophilic fungi infect keratinized tissue, including skin, nails, and hair. A temperature of 27.5 °C will allow all medically relevant fungi to grow . Fungi in water were addressed in Brandão et al. , but the conclusion of the study was that more data were needed before recommendations could be issued. The same team is preparing to address this data gap in the future. Fungal taxonomy is currently undergoing a revolution, mainly due to new data arising from the use of molecular biology techniques . Still, the primary approach tends to remain identification by micro and macro characteristics of the colony. Some of the common fungi found in the beach environment are readily recognizable at first glance. Microscopic verification of the typical structures may help to distinguish, for example, some Penicillium and Aspergillus section Fumigati species. Yeasts, however, require additional biochemical testing to be differentiated. Sabino et al. described in detail how sand analysis can be performed for the identification of the fungi and bacteria. Although the same institution currently identifies many of the fungi by matrix-assisted laser desorption/ionization-time of flight (MALDI-ToF) or molecular approaches based on sequencing of the ITS1 and ITS2 regions of the ribosomal DNA. MALDI-ToF is fast and requires little handling before an acceptable identification is achieved. Molecular identification tends to be broader in the number of species that can be identified, but it is more costly and labor intensive. The process requires DNA extraction, purification, and amplification with the primers chosen for the identification intended by PCR, cleaning of the PCR product of the amplification, followed by sequencing of this product and searching for and aligning (blasting) the sequences with DNA databases . This system allows the clear distinction of many cryptic species, depending on the primers chosen to help amplify the target DNA, as shown in the study of Novak Babič et al. . There is currently no methodology published for detecting insects in beach sand. However, WHO lists the main groups considered of interest, which are all Diptera: mosquitoes, biting midges, sand-flies, flies, and blackflies. When flies and blackflies occur in large numbers, they are a nuisance and mosquito, midge, and sand-fly bites can be painful in addition to being vectors of disease for humans and animals. Mosquito numbers typically increase during wet weather or following tide triggers and they are often most active at dawn. It is recommended that beachgoers try to avoid exposure outdoors at sunset and overnight; they should wear long sleeves and use insect repellent. As mosquitoes breed in standing water, getting rid of devices that hold water is a simple way to stop them from breeding . Biting midges are a common nuisance along certain coastal areas and are most active in intertidal zones, including canals, rivers, and estuaries. Sand-flies tend to be found on beaches where the sand is slightly earthier and are usually a problem when people lie directly on the sand for long periods of time. To help reduce nesting sites, it is advisable for authorities to remove algae and other debris brought to sands by storms. As biting midges and sand-flies are sub-optimal fliers, effective protection can be provided through the use of fans in public places to prevent bites, combined with repellents and loose clothing . Nuisance flies are attracted to dead animals, feces, and garbage, allowing them to spread a variety of disease-causing bacteria and parasites. Some flies can deliver painful bites and transmit diseases to humans and animals. Good sanitation, by eliminating flies’ breeding sites, is the main fly control recommendation at beach facilities. Thus, garbage containers must always be closed and fecal matter from dogs and other animals should be eliminated. When these dipterans become pests, public health entities and municipalities can implement chemical control programs with insecticides, such as pyrethrins and pyrethroids , while preventing the development of potential insect resistance by alternating the products. Helminths are currently monitored only in Lithuania, where the beach sand is tested once before the start of the bathing season and at least four times during the bathing season. The procedure is described in detail by the Ministry of Health of the Republic of Lithuania . Ideally, this regulation should be tested and implemented in other countries as part of beach sand monitoring activities with relevance for human health protection. One of the objectives of this review was to describe strategies to assess beach sand quality. Historically, beach sand had not been addressed although it has been long recognized since the 1960s as a fomite for human infections and a source of FIB in bathing waters. Beach sand quality did not become a regulatory concern until 2010, with the Portuguese Blue Flag program, in collaboration with the National Institute of Health Doctor Ricardo Jorge and the Portuguese Environment Agency. Between 2006 and 2011, this cooperation led to the monitoring of beach sand across Portugal . The effort found that sampling before the bathing season yields higher counts of microorganisms and yeasts, and dermatophytes, which indicate intensive human use. There is a microbial cumulative effect of beach use throughout the bathing season and there is a significant correlation between E. coli and C. albicans and between enterococci and C. albicans . The final finding confirmed the human nature of the contamination since C. albicans is nearly human specific. Since the end of the implementation in 2016, the results may have changed significantly due to the investment made by all the member-states to provide excellent water quality at bathing waters by improving the treatment of residual waters. Since this program came to an end, no other known monitoring program has taken place regarding sand. Based on the WHO recommendations of 2021 , the Blue Flag program of Portugal implemented sand monitoring as new awarding criteria for the 2022 bathing season. Knowledge in this field, however, has improved immensely with publications arising from all over the world. It is thus time to develop regulations to help guide and set a common worldwide plan of action to integrate sand quality in beach management tools. Educating the public and beach managers on how to maintain a healthy level of potential pathogens, is a crucial element in maintaining the good health of the beach and beachgoers. WHO describes a certain number of actions in this perspective ( ). These measures alone may not maintain sand microbiota at acceptable levels all the time, but they are informed by years of studies on diffuse pollution and hygiene concepts targeting human health protection in recreational water environments. In addition to research to fill the many gaps that currently exist and are discussed throughout this text, it is necessary to develop a procedure to classify beaches according to their results of sand monitoring, as currently happens with water. A future health-oriented set of safety parameters for fungi depends on epidemiological studies or quantitative microbial risk assessment (QMRA) estimates. Considering the non-normal distribution of fungal counts over time, using standard deviations and geometric means is not advised. A good alternative is to classify a beach as compliant or not compliant, allowing a certain number of results to fail, in the case of ordinary microbiota fluctuation. A 20% rejection rate, for example, is not unreasonable in the case of fungi, according to Brandão et al. . This would lead to a guidance value of 89 CFU/g of total fungi in sand and a rejection limit of the 80th percentile, which is 490 CFU/g. This means that during a sampling period, values above 490 CFU/g are acceptable in ≤20% of the samples. For enterococci, the value stated in WHO theoretically reflects the same health effect of consuming water during beach bathing. Therefore, care should be taken if samples exceed 60 CFU/g of sand. This value is considered provisional as it is the result of a QMRA calculation which does not consider the native flora of a beach. Epidemiological studies should be conducted to confirm the validity of assumptions of the calculation. In large freshwater basins, FIB may not reflect human fecal contamination, and the source can be further assessed by studying the microbiota at a genetic level, including using MST testing. With an understanding of potential sources of microbes, beach management strategies should be developed using this information to lower microbial concentrations in sands to levels below the guidelines. In the future, epidemiological studies combined with sand microbiological analysis and additional relevant parameters should be prioritized. The values used for the new WHO guidelines are currently provisional due the lack of clinical confirmation. One option would be to conduct a self-assessment study of a large population visiting defined beaches, similar to Leonard et al. . Studies should also focus on exposures to endemic fungi, bacteria other than MRSA and P. aeruginosa , and other pathogens of emerging concern in beach recreational water settings. Such studies should integrate culture-based and molecular quantification of relevant indicators of sand quality.
Setting Patient-Centered Priorities for Cardiovascular Disease in Central Appalachia: Engaging Stakeholder Experts to Develop a Research Agenda	be8342dc-2a04-4715-9a42-fdae28e24b47	10178300	Patient-Centered Care[mh]	It has been established that disparities in cardiovascular diseases (CVD) and risk factors exist across demographic and socioeconomic groups and geographic locations in the United States (U.S.) . A key goal of Healthy People 2030 is to eliminate these disparities and achieve health equity, well-being, and health literacy . Appalachian residents particularly suffer from the high burden of both diagnosed and undiagnosed CVD when compared to residents in non-Appalachian counties . Additionally, studies have indicated that the burden of CVD in the Central Appalachian Region (CAR) is disproportionately greater than in non-CAR regions in the same state and national rates . Research indicates that priorities need to be set for healthcare improvement and to develop targeted interventions for patient-centered care. However, research involving CVD and risk factors in the CAR is sparse, providing rationales for this study. Due to adverse historical experiences and sociocultural characteristics of the CAR , identifying priorities for CVD research requires engaging diverse community stakeholders . Central Appalachia is a region associated with multiple challenges that are collectively linked to social determinants of health. Some of these problems include lower per capita incomes, higher poverty rates, lower educational attainment, and healthcare-related issues, such as reduced medical care access and a higher prevalence of chronic diseases . The declining coal industry has further negatively impacted the region with problems in social and environmental conditions . These regional problems result in health disparities associated with barriers to CVD management and optimal quality of life for patients with CVD in Appalachian communities [ , , ]. The complexity of the aforementioned health disparities, both in the community and clinical settings, necessitates stakeholder engagement early on in developing research priorities, the implementation process, and later in the dissemination process to ensure the improved translation of research into practice . As guided by PCORI’s conceptualization, our concept of stakeholders refers to providers, patients, clinicians, researchers, purchasers, payers, industry, policymakers, training institutions, hospitals, and health systems . These individuals have a stake and direct interest in the health outcomes of the community being addressed . Previous research, which utilized community-oriented approaches such as principles of community-based participatory research (CBPR), has helped shift the role of the patient in the research process from the subject of research to a stakeholder in research . Community-based participatory research (CBPR) is an approach to research that utilizes community organizing principles and integrates community perspectives within every research phase . The approach is used in studies aimed at setting research priorities in other regions of the U.S. It is helpful in linking research and practice, focusing on context for creating change . The nine principles of CBPR include: (1) recognizing the community as a unit of identity; (2) building on strengths and resources within the community; (3) facilitating the collaborative, equitable involvement of all partners in all phases of the research; (4) integrating knowledge and action for the mutual benefit of all partners; (5) promoting a co-learning and empowering process that attends to social inequalities; (6) involving a cyclical and iterative process; (7) addressing health from both positive and ecological perspectives; (8) disseminating findings and knowledge gained to all partners; and (9) involving a long-term commitment by all partners . Stakeholders have been involved in research that seeks to set priorities for healthcare improvement in low-resource and rural environments. A systematic review of priority setting studies conducted in the U.S. found that most stakeholders engaged in health research priority setting were patients, caregivers, and healthcare providers . This type of collaboration highlights the much-needed place of patients as stakeholder experts and research partners having the opportunity to dialogue with other stakeholders [ , , ]. The value of patient involvement in priority setting is clear because patients have the right to be involved in their healthcare. The equitable involvement of patients in research was recommended in a study conducted on priority setting in kidney disease to ensure that the relevant priorities identified in collaboration with patients were funded . In a pilot study that engaged patients and community stakeholders, the researchers validated moving from an academic researcher-centric approach to one that embraced the knowledge and input of patients [ , , ]. A study conducted in the U.S. for chronic childhood obesity to describe the priorities of stakeholders including patients, caregivers, and health professionals indicated that 24% of stakeholders were parents and caregivers, and 5% of them were children . Thus, setting priorities for research involves collaboration across the stakeholder spectrum for health issues. Stakeholders such as patients have played an active role as partners in developing research agendas . Although studies have successfully gathered data on stakeholder preferences through consultative methods such as focus groups, this method lacks a collaborative framework for systematically engaging stakeholders for the purpose of capacity building. No studies have been conducted to set priorities for patient-centered CVD research in the CAR. Additionally, the patient experience of mistrust and skepticism of health professionals, and physician avoidance, have been identified in Appalachian communities . These serve as the impetus for this study. The purpose of this study was to engage patients, providers, and caregivers with knowledge of CVD in the CAR to identify patient-centered care research priorities for the region. With our research we seek to answer the following research questions: What are the patient-centered care research priorities peculiar to the CAR? What research priorities can we identify using the principles of CBPR? The patient-centered model has been recommended and promoted for adoption in the public health field because evidence shows improved health outcomes when implemented . Stakeholder experts, including informed and involved patients, receptive and responsive health professionals, and a collaborative, supportive healthcare environment fit well within the context of patient navigation and patient-clinician relationships where care is delivered . Thus, this study filled a gap in the literature by providing an evidence-based process of engaging stakeholder experts using a modified Delphi method, in setting patient-centered research priorities for CVD in the CAR. The study is unique because it engaged patients from the affected communities across six states as stakeholder experts, and provided new insight into consensus on managing CVD patient populations . 2.1. Study Participants Study participants comprised diverse CVD stakeholders involved in the 2018–2019 Patient-Centered Outcomes Research Institute (PCORI) engagement project. Stakeholder experts for this project included patients, family/non-professional caregivers, public health professionals, and medical/healthcare providers chosen based on their expertise, membership in a community organization or patient advocacy group, and ability and willingness to participate in the PCORI priority setting project. These criteria resulted in a group of participants representing broad interests for the region. All participants were provided with information that included the study purpose and contact details of the principal investigator and the project coordinator. In compliance with our Institutional Review Board (IRB) and Helsinki Declaration , study participants were notified that the study was voluntary and were informed of efforts in place to ensure confidentiality and privacy. Verbal consent to participate was obtained from each stakeholder. 2.2. Study Setting The study was conducted in Central Appalachia, comprising 228 contiguous and 2 non-contiguous counties in six states: Kentucky, North Carolina, Ohio, Tennessee, Virginia, and West Virginia ( ). 2.3. Study Approach This study utilized grounded theory and principles of community-based participatory research (CBPR) as approaches to incorporate the strengths and resources of multiple stakeholders involved in addressing the health issue of CVD in the CAR. Out of the nine principles of CBPR, this study utilized four principles: building on strengths and resources within the community, ensuring a cyclical and iterative process, facilitating collaborative, equitable involvement of all partners in all phases of the research, and promoting a co-learning and empowering process that attends to social inequalities . 2.4. Study Design We selected the Delphi method because it allows panelists to express their diverse perspectives on a subject matter in a structured and anonymous way . It also allows for large geographical representation, which was important for covering all the counties in the CAR. Ref. The Delphi method encompasses four principles of CBPR: use of a cyclical and iterative process, building on the strengths and resources within the community, facilitating the collaborative and equitable involvement of all partners in all stages of the research, and promoting a co-learning and empowering process that attends to social inequalities. To establish consensus, we administered surveys to address the following research questions: What are the patient-centered care research priorities peculiar to the CAR? What research priorities can we identify using the principles of CBPR? Stakeholder experts were invited to participate on the Delphi panel via email. Data were collected through open-ended questions, as well as priority rankings. We aimed to use the qualitative data gathered from stakeholders to effectively achieve consensus in establishing priority research areas in the CAR ( ). 2.5. Data Collection and Analysis Data were collected between October 2018 and July 2019 through the electronic administration of questionnaires by e-mail to stakeholders in all six states represented by the CAR ( ). The questionnaires were developed from data derived from the environmental scan, focus groups, and CVD disparity statistics for the Central Appalachian communities. The first step was conducting round 1 in an unstructured format to effectively develop a list of stakeholder priorities. Panelists were presented with an open-ended question and asked to provide their top five priorities for CVD in their community and then rank them; 1 was considered the top priority, and 5 was considered the least. Thematic analysis was used to analyze the collected data. The qualitative data were entered into Nvivo, a qualitative data management software in which the grounded theory approach was used to identify themes that support the priorities and rankings. The second step was a round of data collection conducted by means of a questionnaire. The questionnaire, which had 15 closed-ended questions, was developed using Google forms, with the content of the questionnaire derived from round one responses. The panelists were asked to rank priority items for their respective communities. A total of 42 respondents participated in round 2 (14 patients/non-licensed caregivers, 15 community stakeholders, and 13 providers). The response rate was 52%. CVD factors identified for priority ranking covered factors such as nutrition, physical activity, education, rural community outreach, health care access, quality of care, patient-provider communication, cost of medication, focus on preventive medicine, and tobacco use ( ). Panelists were asked to identify themselves as either patients, non-licensed caregivers, community stakeholders, or providers/professionals. The panelists were asked to choose their top three to four priorities from the respective questions for a total of 24 priorities each. During the response period, e-mail reminders were sent to panelists every two weeks. Data were analyzed using descriptive statistics in Microsoft Excel and presented as percentages with illustrations in bar graph format. The top 24 priorities established in round two were concretized and used to develop the questionnaire for round 3. The third step was round 3, which was the final round. A paper-based questionnaire was administered at the annual CVD Appalachia Conference in August 2019. The top priorities generated in round 2 were presented to conference stakeholders to prioritize. Each closed-ended question asked for a priority ranking, with only one priority being selected from each question. Of 40 stakeholder experts who agreed to complete the questionnaire, 31 submitted their responses (7 patients, 18 providers, and 6 community stakeholders). The response rate for round 3 was 77.5%. The modified Delphi method allows for questionnaire administration in a forum where participants are physically present . Data were analyzed using mixed methods of thematic analysis (Nvivo) and quantitative analysis with descriptive statistics for the rankings. Study participants comprised diverse CVD stakeholders involved in the 2018–2019 Patient-Centered Outcomes Research Institute (PCORI) engagement project. Stakeholder experts for this project included patients, family/non-professional caregivers, public health professionals, and medical/healthcare providers chosen based on their expertise, membership in a community organization or patient advocacy group, and ability and willingness to participate in the PCORI priority setting project. These criteria resulted in a group of participants representing broad interests for the region. All participants were provided with information that included the study purpose and contact details of the principal investigator and the project coordinator. In compliance with our Institutional Review Board (IRB) and Helsinki Declaration , study participants were notified that the study was voluntary and were informed of efforts in place to ensure confidentiality and privacy. Verbal consent to participate was obtained from each stakeholder. The study was conducted in Central Appalachia, comprising 228 contiguous and 2 non-contiguous counties in six states: Kentucky, North Carolina, Ohio, Tennessee, Virginia, and West Virginia ( ). This study utilized grounded theory and principles of community-based participatory research (CBPR) as approaches to incorporate the strengths and resources of multiple stakeholders involved in addressing the health issue of CVD in the CAR. Out of the nine principles of CBPR, this study utilized four principles: building on strengths and resources within the community, ensuring a cyclical and iterative process, facilitating collaborative, equitable involvement of all partners in all phases of the research, and promoting a co-learning and empowering process that attends to social inequalities . We selected the Delphi method because it allows panelists to express their diverse perspectives on a subject matter in a structured and anonymous way . It also allows for large geographical representation, which was important for covering all the counties in the CAR. Ref. The Delphi method encompasses four principles of CBPR: use of a cyclical and iterative process, building on the strengths and resources within the community, facilitating the collaborative and equitable involvement of all partners in all stages of the research, and promoting a co-learning and empowering process that attends to social inequalities. To establish consensus, we administered surveys to address the following research questions: What are the patient-centered care research priorities peculiar to the CAR? What research priorities can we identify using the principles of CBPR? Stakeholder experts were invited to participate on the Delphi panel via email. Data were collected through open-ended questions, as well as priority rankings. We aimed to use the qualitative data gathered from stakeholders to effectively achieve consensus in establishing priority research areas in the CAR ( ). Data were collected between October 2018 and July 2019 through the electronic administration of questionnaires by e-mail to stakeholders in all six states represented by the CAR ( ). The questionnaires were developed from data derived from the environmental scan, focus groups, and CVD disparity statistics for the Central Appalachian communities. The first step was conducting round 1 in an unstructured format to effectively develop a list of stakeholder priorities. Panelists were presented with an open-ended question and asked to provide their top five priorities for CVD in their community and then rank them; 1 was considered the top priority, and 5 was considered the least. Thematic analysis was used to analyze the collected data. The qualitative data were entered into Nvivo, a qualitative data management software in which the grounded theory approach was used to identify themes that support the priorities and rankings. The second step was a round of data collection conducted by means of a questionnaire. The questionnaire, which had 15 closed-ended questions, was developed using Google forms, with the content of the questionnaire derived from round one responses. The panelists were asked to rank priority items for their respective communities. A total of 42 respondents participated in round 2 (14 patients/non-licensed caregivers, 15 community stakeholders, and 13 providers). The response rate was 52%. CVD factors identified for priority ranking covered factors such as nutrition, physical activity, education, rural community outreach, health care access, quality of care, patient-provider communication, cost of medication, focus on preventive medicine, and tobacco use ( ). Panelists were asked to identify themselves as either patients, non-licensed caregivers, community stakeholders, or providers/professionals. The panelists were asked to choose their top three to four priorities from the respective questions for a total of 24 priorities each. During the response period, e-mail reminders were sent to panelists every two weeks. Data were analyzed using descriptive statistics in Microsoft Excel and presented as percentages with illustrations in bar graph format. The top 24 priorities established in round two were concretized and used to develop the questionnaire for round 3. The third step was round 3, which was the final round. A paper-based questionnaire was administered at the annual CVD Appalachia Conference in August 2019. The top priorities generated in round 2 were presented to conference stakeholders to prioritize. Each closed-ended question asked for a priority ranking, with only one priority being selected from each question. Of 40 stakeholder experts who agreed to complete the questionnaire, 31 submitted their responses (7 patients, 18 providers, and 6 community stakeholders). The response rate for round 3 was 77.5%. The modified Delphi method allows for questionnaire administration in a forum where participants are physically present . Data were analyzed using mixed methods of thematic analysis (Nvivo) and quantitative analysis with descriptive statistics for the rankings. CVD Patient-Centered Research Priorities Fifteen priority items for CVD care in the CAR were identified by forty-two stakeholder experts ( ). The modified Delphi outcome summary for each round, is shown in . Six of the fifteen priority items were patient-centered, and mapped to two specific areas: (1) patient-centered care; and (2) managing CVD in patient populations ( ). The consensus response rate was 77.5%. The most highly ranked priorities were divided into the following categories: overall access to quality healthcare, patient-provider communication, lifestyle modifying techniques, strategies for preventing heart disease, and medication affordability. Fifteen priority items for CVD care in the CAR were identified by forty-two stakeholder experts ( ). The modified Delphi outcome summary for each round, is shown in . Six of the fifteen priority items were patient-centered, and mapped to two specific areas: (1) patient-centered care; and (2) managing CVD in patient populations ( ). The consensus response rate was 77.5%. The most highly ranked priorities were divided into the following categories: overall access to quality healthcare, patient-provider communication, lifestyle modifying techniques, strategies for preventing heart disease, and medication affordability. To our knowledge, no other studies on setting priorities for a CVD research agenda in the CAR have been found. This gap in the literature about prioritizing research in Appalachia highlights the relevance of this study. Disparities exist in CVD risks and outcomes in the U.S., and several barriers to managing CVD have been associated with living in the Appalachian region . Moreover, individuals experiencing the greatest burden of CVD may not accept or understand the need for lifestyle changes and may not have access to medical care. The combination of acceptance and understanding requires a comprehensive approach to patient-centered care originating from stakeholder engagement. . Previous studies discuss various approaches to patient-centered care, such as patient-provider communication focusing on community health worker interventions [ , , ]. However, no studies documenting stakeholder priority setting for the CVD research agenda were found in the CAR. The results of this study converge with literature that calls for a patient-centered approach to health promotion and the reduction of health inequalities and disparities, especially in rural populations . In an environment of mistrust for medical providers, the patient-centered care approach provides an atmosphere to build trust between patients and providers . Patient reflections on patient-centered care indicate that trust remains an important part of patient-provider relationships [ , , ]. While trust may be defined in numerous ways, there is consensus in the literature that it entails believing that providers have the patient’s best interest and will act in goodwill . The trust deficit results in barriers to relationships with providers and impending patient adherence to medical recommendations and their utilization of preventive health services [ , , ]. A study in Appalachia found that distrust in the healthcare system influences patients against taking preventive health measures . The results of this study illustrate the broad range of opportunities for patient-centered, evidence-based interventions. The unique contribution of our study stems from addressing the gap in the literature. Our results confirm and enhance the patient-centered care and CVD literature by demonstrating that setting priorities for patient-centered care within disease-specific populations across large geographic areas is desired and feasible [ , , ]. This study also (1) confirms that engaging patients and community members in rural and underserved areas as stakeholders for consensus in the research process is achievable; and (2) provides a deeper understanding of patient perspectives of how CVD populations are managed in local communities . This study implies an increasing expectation from stakeholders to recognize the “patient as a person” rather than seeing the person as a patient. To see the “patient as a person”, emphasis must be on delivering patient-driven proactive and personalized care, and also understanding the patient and their illness in the context of the individual . Emerging literature has recommended a patient-centered care approach for patients with chronic diseases with the context of the whole person taken into consideration . To further exemplify the “patient as a person” in priority setting acknowledges the patient as a key stakeholder in the process, which may impact research outcomes . With this approach, the patient is not only a recipient of treatment and services, but is also an expert engaged in identifying priorities for their care in their community. The strengths of this study are the participation of a highly diverse group of stakeholder experts who represented Appalachian counties in six states, the CAR. Utilizing principles of CBPR, patients and family caregivers participated in the study by means of an existing method of priority setting, the modified Delphi method. This strategy supports the engagement of those receiving care to take an active role in their health care by becoming researchers and not merely recipients of care. Stakeholder experts were found to be highly motivated to participate in identifying health disparities in the CAR. The consensus reached after round 3 indicates that providers should come alongside patients to ensure that their needs are met, and contribute to the planning of the disease management process. Limitations One of the study’s limitations is that the panelists did not receive compensation for their time participating in the study. Logistical limitations also included a limited timeframe to complete the Delphi process. Both these factors may explain the low response rate in round 1. Nonetheless, the themes across these three data collection rounds were consistent. One of the study’s limitations is that the panelists did not receive compensation for their time participating in the study. Logistical limitations also included a limited timeframe to complete the Delphi process. Both these factors may explain the low response rate in round 1. Nonetheless, the themes across these three data collection rounds were consistent. This study aimed to fill a gap in priority setting for a CVD research agenda in the CAR. Our results suggest that the modified Delphi method successfully engaged stakeholders as experts in identifying research priorities for the region, which is rural and under-resourced. The priorities included the patient’s voice and perspective, which were identified through an iterative process based on CBPR principles. Six patient-centered priorities included research that supports: (1) access to quality healthcare providers; (2) communicating and providing education to patients on their level; (5) shorter wait times for scheduling appointments; and (6) empowering and motivating patients to take responsibility for their health. Thus, this study provides a platform for future studies to facilitate patient-centered care and patient-centered outcomes research.
Natural Killer Cells and Their Implications in Immune Response Diversification in Clinical Pathology and Neoplastic Processes	41d7f23a-5a28-4cf1-a56d-7b2cef95bf92	10178591	Pathology[mh]
FDA-Approved Fluorinated Heterocyclic Drugs from 2016 to 2022	beefae69-832e-4341-b303-1acd2ab0ec8a	10178595	Pharmacology[mh]	The presence of many different heterocyclic rings in natural products, such as alkaloids, vitamins, antibiotics, peptides, etc., prompted the introduction of structural motifs into synthetic drugs . Therefore, heterocycles are considered to be prominent scaffolds for the synthesis of biologically active compounds and prospective drugs . In fact, it is estimated that heterocyclic moieties are present in around 85% of bioactive compounds . On the other hand, in the second half of 20th century, another fundamental tool for drug design was introduced with the incorporation of fluorine atoms into drugs . Since the introduction of the first fluorocorticosteroid, fludrocortisone, in 1954 , the fluorinated drugs market has exponentially evolved, with 20% of those on the market being fluorinated drugs and around 30% of fluorinated drugs being blockbuster pharmaceuticals, such as Lipitor, Fluoxetine, Linezolid or Fluticasone . To date, more than 300 fluorinated pharmaceuticals have been approved for use as drugs . The success of the introduction of fluorine atoms is linked to the peculiar physicochemical properties of the C-F bond , which are the high bond strength, polarity and minimal steric hindrance of fluorine , combined with a general metabolic stability that, nevertheless, is an issue that is currently under exploration . The introduction of fluorine, for example, allows researchers to easily modulate the p K a of neighboring functionalities, improving the bioavailability and affinity to specific receptors . In addition, the monofluorination or trifluoromethylation of alkyl groups decreases the drug lipophilicity due to the strong electron-withdrawing capabilities of fluorine. On the other hand, fluoro-arenes are more lipophilic due to the low polarizability of the C-F bond . In addition, the presence of a fluorine atom can also enhance the membrane’s permeability . The importance of fluorinated compounds is also linked to their use as diagnostic tools within imaging techniques such as 19 F-MRI and 18 F-PET . The direct link between fluorinated moieties and heterocycles led to the formation of the sub-class of fluorinated heterocycles, which combine the strength of these two fundamental scaffolds in modern medicinal chemistry. This important class of fluorinated pharmaceuticals includes some of the selected examples of FDA-approved drugs reported in . Among these compounds, there have been several game changers over the last decades, such as fluorouracil, the class of fluoroquinolones, sitagliptin and fluorodeoxyglucose, just to mention a few . In this review, recent advances in the field of fluorinated heterocyclic drugs are presented, discussing FDA-approved molecules from 2016 to 2022. The molecules considered in this article are limited to those with a fluorinated group directly linked to the heterocyclic ring. The biological targets and the therapeutic indications are presented together with synthetic details. The fluorination strategy and influence of the fluorinated moiety on bioactivity are also discussed. The sections are reported in chronological order, starting with the most recently approved drugs; for each section, the compounds are presented in alphabetical order. Introduction of Fluorine Atoms in Organic Molecules Fluorinated starting materials used as precursors to obtain fluorinated, approved drugs can present mono-, di- and trifluoro alkyl groups; the last ones can be generally introduced via building blocks such as trifluoroacetate, trifluoroethylamine, trifluoro and ethyl triflate. Furthermore, several starting materials used for this purpose are formed by aromatic or heterocyclic rings bearing F atoms, such as (poly)fluorobenzoic acid, fluoro- or trifluoromethylpyridines, just to cite a few of them (see below). Some of the main processes used for the introduction of F atoms are summarized in . The introduction of C-F or CF 2 groups is achieved through the nucleophilic fluorination of electrophiles; some examples involving dithiane or sulphonate formation are reported in a,b . Trifluoromethylated starting materials such as trifluoroacetic acid are industrially prepared in excellent yields by the electrochemical fluorination of acetyl chloride or acetic anhydride in anhydrous hydrogen fluoride, followed by the hydrolysis of the resulting trifluoroacetyl fluoride ( c) . Fluorination methods of arenes include traditional nucleophilic substitution ( d) and transition-metal-catalyzed nucleophilic fluorination or deoxofluorination . For the introduction of trifluoromethyl groups, substrates can be trifluoromethylated by employing electrophilic trifluoromethylating reagents , such as Togni reagents , and S-(trifluoromethyl)dibenzothiophenium salts , or lower cost reagents such as CF 3 I or CF 3 H, which are favored for industrial processes, while the use of the solid and bench-top-stable reagents such as NaSO 2 CF 3 in radical trifluoromethylations for the trifluoromethylation of electron-rich arenes avoids perfluoroalkylations ( e) . Fluorinated starting materials used as precursors to obtain fluorinated, approved drugs can present mono-, di- and trifluoro alkyl groups; the last ones can be generally introduced via building blocks such as trifluoroacetate, trifluoroethylamine, trifluoro and ethyl triflate. Furthermore, several starting materials used for this purpose are formed by aromatic or heterocyclic rings bearing F atoms, such as (poly)fluorobenzoic acid, fluoro- or trifluoromethylpyridines, just to cite a few of them (see below). Some of the main processes used for the introduction of F atoms are summarized in . The introduction of C-F or CF 2 groups is achieved through the nucleophilic fluorination of electrophiles; some examples involving dithiane or sulphonate formation are reported in a,b . Trifluoromethylated starting materials such as trifluoroacetic acid are industrially prepared in excellent yields by the electrochemical fluorination of acetyl chloride or acetic anhydride in anhydrous hydrogen fluoride, followed by the hydrolysis of the resulting trifluoroacetyl fluoride ( c) . Fluorination methods of arenes include traditional nucleophilic substitution ( d) and transition-metal-catalyzed nucleophilic fluorination or deoxofluorination . For the introduction of trifluoromethyl groups, substrates can be trifluoromethylated by employing electrophilic trifluoromethylating reagents , such as Togni reagents , and S-(trifluoromethyl)dibenzothiophenium salts , or lower cost reagents such as CF 3 I or CF 3 H, which are favored for industrial processes, while the use of the solid and bench-top-stable reagents such as NaSO 2 CF 3 in radical trifluoromethylations for the trifluoromethylation of electron-rich arenes avoids perfluoroalkylations ( e) . In 2022, the FDA approved 37 new therapeutic and diagnostic products . Monoclonal antibodies (mAbs) continue to be one of the most widely licensed groups of biological therapies. Nevertheless, 22 of them are novel chemical entities (NCEs), 14 of which contain fluorine atoms and nitrogen heterocycles . Lenacapavir ( and ) and Oteseconazole , two new approved drugs released in the past year, combine these two key characteristics. 2.1. Lenacapavir Lenacapavir (SUNLENCA ® ) is a human immunodeficiency virus type 1 (HIV-1) capsid inhibitor developed by Gilead Science Inc., and it is administered in cases when conventional antiretroviral therapies are ineffective. The mechanism of action is totally different from that of other antivirals used for the treatment of HIV-1. Indeed, Lenacapavir establishes several hydrophobic and electrostatic interactions with capsid subunits (CA1 and CA2). For example, the difluorobenzyl group can be stabilized inside a hydrophobic pocket of the CA1 N -terminal domain. This entails interference with virus life cycle processes where CA is involved, such as reverse transcription, nuclear import and integration . The synthetic method to obtain Lenacapavir is divided into three steps and is reported in and . The Claisen condensation of 1 using a strong base, such as lithium bis(trimethylsilyl)amide and ethyl 2,2,2-trifluoroacetate, leads to enolate 2 , presenting a CF 3 group. At this point, pyrazole ring 3 formation occurs due to the addition of an ethyl hydrazinoacetate salt. Intermediate 4 is obtained through N -hydroxyphtalimide-catalyzed selective oxidation, followed by saponification with NaOH. To obtain building block 6 , the desulfurative fluorination of dithiolane 5 takes place, followed by supercritical fluid chromatography (SFC). The synthesis proceeds with the construction of building block 9 . To acquire substituted indazole core 8 , hydrazine is combined with fluorobenzonitrile 7 , and then a trifluoroethyl group is introduced via a substitution at position one. A cross-coupling reaction between bis(pinacolato)diboron and 8 in the presence of a palladium/triphenylphosphine catalyst and potassium propionate creates 9 . In order to produce Lenacapavir , deprotection preceded by the formation of an amide bond between carboxylic acid 6 and ammine 14 must occur. Intermediate 14 is obtained via two palladium-catalyzed coupling reactions between fluorinated compound 10 and amine 13 , followed by a protection with methanesulfonyl chloride on the amino group linked to the indazole ring. 2.2. Oteseconazole Oteseconazole (VIVJOA™) is an antifungal agent that was released by Mycovia Pharmaceuticals and is administered to reduce the incidence of recurrent vulvovaginal candidiasis (RVVC). It affects the integrity of the cell membrane of pathogenic strains of candida by interacting with cytochrome P450 (CYP51) . Oteseconazole ’s selectivity for fungal metalloenzyme CYP51 is provided by the tetrazole moiety. In turn, the heterocyclic residue is connected through a metabolically resistant difluoro methyl linker with a substituted phenyl trifluoroethyl ether. The synthesis is presented in . Starting with pyridine 15 , ethyl bromodifluoroacetate and 2,4-difluorobromobenzene are used to introduce the CF 2 linker and 2,4-difluorobenzene group, respectively, producing 16 . Through a diazomethane-mediated epoxidation reaction, intermediate 17 is obtained and reacts with 4-(trifluoromethoxy)phenylboronic acid 18 via a Pd catalyzed Suzuky–Miyaura coupling reaction, producing 19 . In the next two steps, the introduction of the triazole ring occurs via nucleophilic attack, leading to the opening of the epoxy ring. Oteseconazole is finally achieved as a single enantiomer by chiral preparative HPLC. Lenacapavir (SUNLENCA ® ) is a human immunodeficiency virus type 1 (HIV-1) capsid inhibitor developed by Gilead Science Inc., and it is administered in cases when conventional antiretroviral therapies are ineffective. The mechanism of action is totally different from that of other antivirals used for the treatment of HIV-1. Indeed, Lenacapavir establishes several hydrophobic and electrostatic interactions with capsid subunits (CA1 and CA2). For example, the difluorobenzyl group can be stabilized inside a hydrophobic pocket of the CA1 N -terminal domain. This entails interference with virus life cycle processes where CA is involved, such as reverse transcription, nuclear import and integration . The synthetic method to obtain Lenacapavir is divided into three steps and is reported in and . The Claisen condensation of 1 using a strong base, such as lithium bis(trimethylsilyl)amide and ethyl 2,2,2-trifluoroacetate, leads to enolate 2 , presenting a CF 3 group. At this point, pyrazole ring 3 formation occurs due to the addition of an ethyl hydrazinoacetate salt. Intermediate 4 is obtained through N -hydroxyphtalimide-catalyzed selective oxidation, followed by saponification with NaOH. To obtain building block 6 , the desulfurative fluorination of dithiolane 5 takes place, followed by supercritical fluid chromatography (SFC). The synthesis proceeds with the construction of building block 9 . To acquire substituted indazole core 8 , hydrazine is combined with fluorobenzonitrile 7 , and then a trifluoroethyl group is introduced via a substitution at position one. A cross-coupling reaction between bis(pinacolato)diboron and 8 in the presence of a palladium/triphenylphosphine catalyst and potassium propionate creates 9 . In order to produce Lenacapavir , deprotection preceded by the formation of an amide bond between carboxylic acid 6 and ammine 14 must occur. Intermediate 14 is obtained via two palladium-catalyzed coupling reactions between fluorinated compound 10 and amine 13 , followed by a protection with methanesulfonyl chloride on the amino group linked to the indazole ring. Oteseconazole (VIVJOA™) is an antifungal agent that was released by Mycovia Pharmaceuticals and is administered to reduce the incidence of recurrent vulvovaginal candidiasis (RVVC). It affects the integrity of the cell membrane of pathogenic strains of candida by interacting with cytochrome P450 (CYP51) . Oteseconazole ’s selectivity for fungal metalloenzyme CYP51 is provided by the tetrazole moiety. In turn, the heterocyclic residue is connected through a metabolically resistant difluoro methyl linker with a substituted phenyl trifluoroethyl ether. The synthesis is presented in . Starting with pyridine 15 , ethyl bromodifluoroacetate and 2,4-difluorobromobenzene are used to introduce the CF 2 linker and 2,4-difluorobenzene group, respectively, producing 16 . Through a diazomethane-mediated epoxidation reaction, intermediate 17 is obtained and reacts with 4-(trifluoromethoxy)phenylboronic acid 18 via a Pd catalyzed Suzuky–Miyaura coupling reaction, producing 19 . In the next two steps, the introduction of the triazole ring occurs via nucleophilic attack, leading to the opening of the epoxy ring. Oteseconazole is finally achieved as a single enantiomer by chiral preparative HPLC. In 2021, 50 new drugs were approved. Thirty-three small molecules with 10 fluorinated compounds and 28 heterocyclic compounds are included in this list, together with fluorinated heterocycles Atogepant , Piflufolastat , Sotorasib , Umbralisib , and Vericiguat ( , , , and ), as discussed below . 3.1. Atogepant Atogepant (Qulipta™) is a novel drug designed by Abb Vie for the preventive treatment of migraines in adults. Developed as a calcitonin gene-related peptide (CGRP) receptor antagonist, this neuropeptide and its receptors are located in the trigeminal nerves involved in pain sensations . The antagonist’s high affinity to the receptor is increased due to a 2,3,6 fluoro substitution on the phenyl moiety. The 2,2,2-trifluoethyl group masking the piperidinone ring improved the pharmacokinetic and pharmacodynamic characteristics more than other gepant drugs do. In addition, fluorine atoms could also be associated with a lower hepatotoxicity . The method for Atogepant synthesis is reported in . 1-(2,3,6-trifluorophenyl)propan-2-one 20 is alkylated to 21 via N-Boc-iodoserine-OMe. Piperidine intermediate 22 is a result of a reductive amination with 2,2,2-trifluoroethanamine and sodium triacetoxyborohydride as a reducing agent, followed by cyclization and optical resolution using normal-phase liquid chromatography (NPLC). In order to generate the first building block, 23 , deprotection of the amine group via hydrochloric acid takes place. The second intermediate, 26 , is achieved by performing a previously patented procedure on 24 . Azospiro bispyridine 24 undergoes diazotation–iodination process via sodium nitrite in the presence of p -toluensulfonic acid and potassium iodide. Ester 25 , obtained via palladium-catalyzed carbonylation, is subsequently saponified to obtain 26 . Finally, a coupling reaction with aminopiperidinone 23 and carboxylic acid 26 is carried out to allow the formation of an amide bond leading to the final product, Atogepant. 3.2. Piflufolastat F 18 Piflufolastat F 18 , commercially known as PYLARIFY by Progenics Pharmaceuticals Inc., is a diagnostic imaging agent radiolabeled with 18 F isotope, which detects prostate-specific membrane antigen (PSMA) via positron emission tomography (PET). It was approved on May 2021 by the FDA as a radioactive diagnostic tool, thanks to which it is possible to obtain accurate and early information on prostate cancer metastases, even in those patients with low prostate-specific antigen (PSA) levels . Piflufolastat synthesis is described in . Piflufolastat is the result of the deprotection of the carboxyl group with TFA, preceded by a nucleophilic substitution with urea derivate 27 on 6-[ 18 F]Fluoro-nicotinic acid-2,3,5,6-tetrafluoro-phenyl ester 28 , synthetized according to the previously reported procedure . 3.3. Sotorasib The brand name for the new drug marketed by Amgen is LUMAKRASTM TM . The pharmacologically active agent, Sotorasib , it is an RAS small GTPase inhibitor used to treat colorectal cancer and non-small-cell lung cancer brought on by the KRAS G12C oncogene . To improve the pharmacokinetic properties such as oral bioavailability, azaquinazolinone is designed with a fluorine atom on C6 carbon, a fluorophenol residue on C7 and a nitrogen atom instead of C8 carbon . The Sotorasib synthetic process is described in . Starting with 2,6-dichloro-5-fluoronicotinic acid 29 , acyl chloride is obtained through the use of oxalyl chloride; this is then converted to the corresponding amide, 30 . The formation of compound 31 is carried out through the reaction between nicotinamide 30 and 2-isopropyl-4-methylpyridin-3-amine. Compound 31 is then treated with potassium hexamethyldisilazane to drive cyclization and produce the duly substituted 2,4-dihydroxypyrido [2,3-d]pyrimidine ring 32. At this point, the chlorination reaction produces intermediate 33 , which, in turn, can be combined with a Boc-protected methylpiperazine to produce a selective amination. The resulting compound 34 is reacted with organotrifluoroborate salt via a Suzuki−Miyaura cross-coupling reaction to attach fluorophenol moiety 35 . Deprotection and amidation of nitrogen of the piperazine ring eventually produces Sotorasib . In 2022, the same Amgen group developed a commercial manufacturing process, in which they improved several synthetical steps, starting with compound 32 . 3.4. Umbralisib Umbralisib is sold under the brand name Ukuoniq and was developed by TG Therapeutics. It was approved in February 2021 for the treatment of marginal zone lymphoma (MZL) and follicular lymphoma (FL) . The mode of action of Umbralisib is related to the inhibition of kinase PI3K-delta and casein kinase CK1-epsilon. Umbralisib contains a 6-fluoro-chromen-4-one central heterocyclic core and two other fluorophenyl groups. The synthesis was disclosed in a patent in 2014 and is presented in . Fluorinated chromen-4-one ring 38 is constructed, starting with 3-fluorophenylacetic acid 36 , which, after conversion into chloride and subsequent acylation of 4-fluoroanisole, yields compound 37 . The treatment of phenol 37 with propionic anhydride produces 38 via acylation and subsequent cyclocondensation. The radical bromination of the methylene group with NBS yields 39 . The following steps result in the obtainment of racemic alcohol 40 after the hydrolysis of 39 , as well as the subsequent formation of two enantiomers, 42 and 44 . S enantiomer 42 could be selectively obtained via the stereoselective reduction of ketone 41 with R Alpine borane, which is obtained by means of the Swern oxidation of racemic 40 . R enantiomer 44 was obtained via a Mitsunobu reaction with 4-chlorobenzoic acid and DEAD, followed by the hydrolysis of ester 43 . Alcohol 44 is coupled with pyrazolopyrimidine 49 , again under Mitsunobu conditions, to acquire Umbralisib as a single enantiomer. Compound 49 is obtained via the Suzuki coupling of iodopyrazolopyrimidine 48 with aryl pinacolborane 47 . Compound 47 is obtained in two steps using 4-bromo-2-fluoro-phenol 45 . 3.5. Vericiguat Vericiguat is sold under the brand name Verquvo and was developed by Bayer AG and Merck & Co. It was approved in January 2021 to reduce the risks of cardiovascular death and heart failure . Vericiguat is a guanylate cyclase (sGC) stimulator with a 1H -pyrazolo [3,4-b]pyridine core bearing a fluorine atom at C-5. The presence of the fluorine atom increases the metabolic stability and induces lower clearance. The method for the synthesis of Vericiguat is reported in . Tetrafluoropropanol 50 , the starting fluorinated building block, is converted in two steps into morpholino derivative 51 , and then into morpholinium cation 52 after methylation with methyl methanesulfonate. Compound 53 is obtained after a treatment with NaOH, which induces the elimination of the first fluorine atom as HF. Other two fluorine atoms from the difluoromethyl group are lost during hydrolysis into 2-fluoroacrolein derivative 54 . α,β-unsaturated aldehyde 54 reacts with aminopyrazole 55 under acidic cyclization conditions, allowing the introduction of the 5-fluorine atom into the 1H -pyrazolo [3,4-b]pyridine core of derivative 56 . Ester 56 is then converted in three steps into amidine 59 , via amide 57 and nitrile 58 . The C-3 pyrimidine ring is then constructed with a condensation between 59 and hydrazonomalonitrile 60 . Using compound 61 , the synthesis of Vericiguat is completed in two steps via the catalytic hydrogenation of the diazo moiety of 61 to triaminopyrimidine 62 , and finally, via the formation of the carbamate group of the target compound after a treatment with methyl chloroformate. Atogepant (Qulipta™) is a novel drug designed by Abb Vie for the preventive treatment of migraines in adults. Developed as a calcitonin gene-related peptide (CGRP) receptor antagonist, this neuropeptide and its receptors are located in the trigeminal nerves involved in pain sensations . The antagonist’s high affinity to the receptor is increased due to a 2,3,6 fluoro substitution on the phenyl moiety. The 2,2,2-trifluoethyl group masking the piperidinone ring improved the pharmacokinetic and pharmacodynamic characteristics more than other gepant drugs do. In addition, fluorine atoms could also be associated with a lower hepatotoxicity . The method for Atogepant synthesis is reported in . 1-(2,3,6-trifluorophenyl)propan-2-one 20 is alkylated to 21 via N-Boc-iodoserine-OMe. Piperidine intermediate 22 is a result of a reductive amination with 2,2,2-trifluoroethanamine and sodium triacetoxyborohydride as a reducing agent, followed by cyclization and optical resolution using normal-phase liquid chromatography (NPLC). In order to generate the first building block, 23 , deprotection of the amine group via hydrochloric acid takes place. The second intermediate, 26 , is achieved by performing a previously patented procedure on 24 . Azospiro bispyridine 24 undergoes diazotation–iodination process via sodium nitrite in the presence of p -toluensulfonic acid and potassium iodide. Ester 25 , obtained via palladium-catalyzed carbonylation, is subsequently saponified to obtain 26 . Finally, a coupling reaction with aminopiperidinone 23 and carboxylic acid 26 is carried out to allow the formation of an amide bond leading to the final product, Atogepant. Piflufolastat F 18 , commercially known as PYLARIFY by Progenics Pharmaceuticals Inc., is a diagnostic imaging agent radiolabeled with 18 F isotope, which detects prostate-specific membrane antigen (PSMA) via positron emission tomography (PET). It was approved on May 2021 by the FDA as a radioactive diagnostic tool, thanks to which it is possible to obtain accurate and early information on prostate cancer metastases, even in those patients with low prostate-specific antigen (PSA) levels . Piflufolastat synthesis is described in . Piflufolastat is the result of the deprotection of the carboxyl group with TFA, preceded by a nucleophilic substitution with urea derivate 27 on 6-[ 18 F]Fluoro-nicotinic acid-2,3,5,6-tetrafluoro-phenyl ester 28 , synthetized according to the previously reported procedure . The brand name for the new drug marketed by Amgen is LUMAKRASTM TM . The pharmacologically active agent, Sotorasib , it is an RAS small GTPase inhibitor used to treat colorectal cancer and non-small-cell lung cancer brought on by the KRAS G12C oncogene . To improve the pharmacokinetic properties such as oral bioavailability, azaquinazolinone is designed with a fluorine atom on C6 carbon, a fluorophenol residue on C7 and a nitrogen atom instead of C8 carbon . The Sotorasib synthetic process is described in . Starting with 2,6-dichloro-5-fluoronicotinic acid 29 , acyl chloride is obtained through the use of oxalyl chloride; this is then converted to the corresponding amide, 30 . The formation of compound 31 is carried out through the reaction between nicotinamide 30 and 2-isopropyl-4-methylpyridin-3-amine. Compound 31 is then treated with potassium hexamethyldisilazane to drive cyclization and produce the duly substituted 2,4-dihydroxypyrido [2,3-d]pyrimidine ring 32. At this point, the chlorination reaction produces intermediate 33 , which, in turn, can be combined with a Boc-protected methylpiperazine to produce a selective amination. The resulting compound 34 is reacted with organotrifluoroborate salt via a Suzuki−Miyaura cross-coupling reaction to attach fluorophenol moiety 35 . Deprotection and amidation of nitrogen of the piperazine ring eventually produces Sotorasib . In 2022, the same Amgen group developed a commercial manufacturing process, in which they improved several synthetical steps, starting with compound 32 . Umbralisib is sold under the brand name Ukuoniq and was developed by TG Therapeutics. It was approved in February 2021 for the treatment of marginal zone lymphoma (MZL) and follicular lymphoma (FL) . The mode of action of Umbralisib is related to the inhibition of kinase PI3K-delta and casein kinase CK1-epsilon. Umbralisib contains a 6-fluoro-chromen-4-one central heterocyclic core and two other fluorophenyl groups. The synthesis was disclosed in a patent in 2014 and is presented in . Fluorinated chromen-4-one ring 38 is constructed, starting with 3-fluorophenylacetic acid 36 , which, after conversion into chloride and subsequent acylation of 4-fluoroanisole, yields compound 37 . The treatment of phenol 37 with propionic anhydride produces 38 via acylation and subsequent cyclocondensation. The radical bromination of the methylene group with NBS yields 39 . The following steps result in the obtainment of racemic alcohol 40 after the hydrolysis of 39 , as well as the subsequent formation of two enantiomers, 42 and 44 . S enantiomer 42 could be selectively obtained via the stereoselective reduction of ketone 41 with R Alpine borane, which is obtained by means of the Swern oxidation of racemic 40 . R enantiomer 44 was obtained via a Mitsunobu reaction with 4-chlorobenzoic acid and DEAD, followed by the hydrolysis of ester 43 . Alcohol 44 is coupled with pyrazolopyrimidine 49 , again under Mitsunobu conditions, to acquire Umbralisib as a single enantiomer. Compound 49 is obtained via the Suzuki coupling of iodopyrazolopyrimidine 48 with aryl pinacolborane 47 . Compound 47 is obtained in two steps using 4-bromo-2-fluoro-phenol 45 . Vericiguat is sold under the brand name Verquvo and was developed by Bayer AG and Merck & Co. It was approved in January 2021 to reduce the risks of cardiovascular death and heart failure . Vericiguat is a guanylate cyclase (sGC) stimulator with a 1H -pyrazolo [3,4-b]pyridine core bearing a fluorine atom at C-5. The presence of the fluorine atom increases the metabolic stability and induces lower clearance. The method for the synthesis of Vericiguat is reported in . Tetrafluoropropanol 50 , the starting fluorinated building block, is converted in two steps into morpholino derivative 51 , and then into morpholinium cation 52 after methylation with methyl methanesulfonate. Compound 53 is obtained after a treatment with NaOH, which induces the elimination of the first fluorine atom as HF. Other two fluorine atoms from the difluoromethyl group are lost during hydrolysis into 2-fluoroacrolein derivative 54 . α,β-unsaturated aldehyde 54 reacts with aminopyrazole 55 under acidic cyclization conditions, allowing the introduction of the 5-fluorine atom into the 1H -pyrazolo [3,4-b]pyridine core of derivative 56 . Ester 56 is then converted in three steps into amidine 59 , via amide 57 and nitrile 58 . The C-3 pyrimidine ring is then constructed with a condensation between 59 and hydrazonomalonitrile 60 . Using compound 61 , the synthesis of Vericiguat is completed in two steps via the catalytic hydrogenation of the diazo moiety of 61 to triaminopyrimidine 62 , and finally, via the formation of the carbamate group of the target compound after a treatment with methyl chloroformate. In 2020, the FDA approved 53 new molecular entities, including 34 small molecules and 4 diagnostic agents . Thirty-one out of thirty-four molecules contain at least one heterocyclic ring, and eleven out of thirty-four molecules contain at least one fluorine atom. In the following paragraph, four heterocyclic compounds bearing a fluorinated moiety directly linked to the ring are reported ( , , and ). Additionally, the approved 18 F-containing diagnostic agent, Tauvid , is presented . 4.1. Berotralstat Berotralstat is sold under the brand name Orladeyo and was developed by BioCryst Pharmaceuticals. It was approved in December 2020 to treat Hereditary Angioedema (HAE) attacks . Berotralstat is a selective inhibitor of plasma kallikrein, bearing a trifluoromethylpyrazole moiety. Another fluorine is present on the central phenyl ring. The patented synthetic approach to acquiring this drug is reported below . The trifluoromethylpyrazole portion of compound 65 is constructed through the condensation of trifluoro β-diketone 63 and arylhydrazine 64 in acetic acid. Cyanopyrazole 65 is then reduced into amine 66 , using LiAlH 4 , which is in turn protected as N -Boc derivative 67 during the successive oxidation of the furan ring to yield acid 68 . Coupling between amine 69 and amide 70 is performed using bromotris-pyrrolidino-phosphonium hexafluorophosphate (PyBrOP) as an activating agent. The formation of amine 71 occurs after the treatment of 70 with thionyl chloride, and then cyclopropanemethylamine. Berotralstat was finally obtained as a single enantiomer after acidic Boc-deprotection and chiral SFC resolution. 4.2. Cedazuridine Cedazuridine , in combination with decitabine , is sold under the brand name Inqovi and was developed by Otsuka Pharma. It was approved in July 2020 for the treatment of myelodysplastic syndromes (MDS) and chronic myelomonocytic leukemia (CMML), reducing the risk of progression of secondary acute myeloid leukemia (sAML) . Cedazuridine is a cytidine deaminase inhibitor that is able to improve the oral bioavailability of decitabine, avoiding its degradation in the gastrointestinal tract. The presence of two fluorine atoms at the ribose ring increase the level of metabolic stability under acidic conditions, improving the pharmacokinetic profile via unfluorinated analogs, retaining the same binding mode of unfluorinated tetrahydrouridines . The synthesis of Cedazuridine is performed in two steps, starting with the analogue, Gemcitabine . The Rh/C catalytic hydrogenation of Gemcitabine produces compound 76 , which is reduced into a mixture of isomers containing Cedazurine and its epimer as major products using NaBH 4 . The difluorotetrahydrofuran ring of Gemcitabine is synthesized via a Reformatzky reaction of fluorinated bromoacetate 72 with D -glyceraldehyde acetonide 73 to furnish 74 as a mixture of anti/syn diasteroisomers at a 3:1 ratio. The hydrolysis/lactonization of 74 into 75 is performed with Dowex 50 resin. Lactone 75 is then converted into Gemcitabine in four steps . 4.3. Pralsetinib Pralsetinib is sold under the brand name Gavreto and was developed by Blueprint Medicines . It was approved in September 2020 for the treatment of metastatic fusion-positive non-small-cell lung cancer . Pralsetinib is REarranged during transfection (RET) inhibitor and it is the first-in-class specific RET inhibitor with more selectivity than other kinases have. The presence of the 4-fluoropyrazolo group allows a different binding mode on the BP-II pocket, which is crucial for high-affinity binding and to avoid resistance from gatekeeper mutations . 4-Fluoropyrazole 77 gives nucleophilic displacement of bromopyridine 78 under basic conditions, yielding pyrazolylpyridine 79 . The stereoselective reductive amination to hydrochloride 82 is accomplished by means of the condensation of the acyl group of 79 with chiral sulfinamide 80 , followed by the reduction with L-Selectride and acidic hydrolysis of sulfinamide 81 . The latter fluorinated building block is coupled with acid 83 (as mixture of diastereoisomers) using PyBop as activating agents. A final chiral SFC resolution produces Pralsetinib as a single enantiomer. The method for the patented synthesis of Pralsetinib is described in . 4.4. Selumetinib Selumetinib is sold under the brand name Koselugo and was developed by AstraZeneca. It was approved in April 2020 for the treatment of neurofibromatosis type 1 (NF1) . Selumetinib , characterized by the presence of a 4-fluorobenzimidazole core, is a mitogen-activated protein kinase (MEK) inhibitor that is able to target the Raf-MEK-ERK signaling pathway. The synthesis of Selumetinib is described in . Trifluorobenzoic acid 84 was employed as starting fluorinated building block for the initial construction of the fluorobenzimidazole ring. The nitration of 84 into 85 and the nucleophilic aromatic displacement of fluoride in nitro-activated derivative 85 with ammonia led to the acquisition of compound 86 . Treatment with trimethylsilyldiazomethane (TMS-CHN 2 ) converts acid 86 into methyl ester 87 , which, in turn, reacts with aniline in xylene at 125 °C, causing the nucleophilic displacement of a second fluorine atom into 88 . The latter nitro derivative is reduced to o -diamminobenzene 89 via iron and ammonium chloride. The benzimidazole ring formation on 90 is performed with formamidine acetate (FAA) in EtOH at 80 °C. The halogenation of the anilino portion of 90 in two consecutive steps produces 91 , with the NBS-mediated introduction of bromine in the para position, and then, 92 , with ortho -chlorination performed with NCS. The methylation of 92 with methyliodide, employing K 2 CO 3 in DMF, occurs at N-1, yielding regio-isomer 93 . Basic hydrolysis into acid 94 is followed by coupling with O -(2-(vinyloxy)ethyl)hydroxylamine in the presence of EDC and HOBt. Hydroxamic acid derivative 95 is finally deprotected via the acidic hydrolysis of the vinyl ether portions, producing Selumetinib . 4.5. Tauvid Tauvid , also named Flortaucipir F18, was developed by Eli Lilly and was approved in May 2020 as a positron emission tomography (PET) imaging probe for Alzheimer’s disease (AD) . Tauvid is the first approved tracer able to bind tau protein . The [ 18 F]fluoropyridine lateral ring is the radioactive portion of this probe. Two developed synthetic procedures are described in . The first synthetic approach is based on the nucleophilic displacement of the nitro group of precursor 96 with [ 18 F]fluoride and using Kryptofix 222 (K 222 ) as a phase transfer catalyst (PTC) . This approach suffers from some drawbacks related to trace purity; therefore, a different synthetic approach was developed starting with N-Boc protected cation 97 , which undergoes nucleophilic displacement to produce radioactive fluorine, followed by acidic Boc removal . This synthetic sequence allows the obtainment of Tauvid in higher yields and purity. Interestingly, this synthesis represents the only example of late-stage fluorination among all the molecules considered in this review. Obviously, the short half-life of 18 F forces researchers to follow this peculiar synthetic approach. Berotralstat is sold under the brand name Orladeyo and was developed by BioCryst Pharmaceuticals. It was approved in December 2020 to treat Hereditary Angioedema (HAE) attacks . Berotralstat is a selective inhibitor of plasma kallikrein, bearing a trifluoromethylpyrazole moiety. Another fluorine is present on the central phenyl ring. The patented synthetic approach to acquiring this drug is reported below . The trifluoromethylpyrazole portion of compound 65 is constructed through the condensation of trifluoro β-diketone 63 and arylhydrazine 64 in acetic acid. Cyanopyrazole 65 is then reduced into amine 66 , using LiAlH 4 , which is in turn protected as N -Boc derivative 67 during the successive oxidation of the furan ring to yield acid 68 . Coupling between amine 69 and amide 70 is performed using bromotris-pyrrolidino-phosphonium hexafluorophosphate (PyBrOP) as an activating agent. The formation of amine 71 occurs after the treatment of 70 with thionyl chloride, and then cyclopropanemethylamine. Berotralstat was finally obtained as a single enantiomer after acidic Boc-deprotection and chiral SFC resolution. Cedazuridine , in combination with decitabine , is sold under the brand name Inqovi and was developed by Otsuka Pharma. It was approved in July 2020 for the treatment of myelodysplastic syndromes (MDS) and chronic myelomonocytic leukemia (CMML), reducing the risk of progression of secondary acute myeloid leukemia (sAML) . Cedazuridine is a cytidine deaminase inhibitor that is able to improve the oral bioavailability of decitabine, avoiding its degradation in the gastrointestinal tract. The presence of two fluorine atoms at the ribose ring increase the level of metabolic stability under acidic conditions, improving the pharmacokinetic profile via unfluorinated analogs, retaining the same binding mode of unfluorinated tetrahydrouridines . The synthesis of Cedazuridine is performed in two steps, starting with the analogue, Gemcitabine . The Rh/C catalytic hydrogenation of Gemcitabine produces compound 76 , which is reduced into a mixture of isomers containing Cedazurine and its epimer as major products using NaBH 4 . The difluorotetrahydrofuran ring of Gemcitabine is synthesized via a Reformatzky reaction of fluorinated bromoacetate 72 with D -glyceraldehyde acetonide 73 to furnish 74 as a mixture of anti/syn diasteroisomers at a 3:1 ratio. The hydrolysis/lactonization of 74 into 75 is performed with Dowex 50 resin. Lactone 75 is then converted into Gemcitabine in four steps . Pralsetinib is sold under the brand name Gavreto and was developed by Blueprint Medicines . It was approved in September 2020 for the treatment of metastatic fusion-positive non-small-cell lung cancer . Pralsetinib is REarranged during transfection (RET) inhibitor and it is the first-in-class specific RET inhibitor with more selectivity than other kinases have. The presence of the 4-fluoropyrazolo group allows a different binding mode on the BP-II pocket, which is crucial for high-affinity binding and to avoid resistance from gatekeeper mutations . 4-Fluoropyrazole 77 gives nucleophilic displacement of bromopyridine 78 under basic conditions, yielding pyrazolylpyridine 79 . The stereoselective reductive amination to hydrochloride 82 is accomplished by means of the condensation of the acyl group of 79 with chiral sulfinamide 80 , followed by the reduction with L-Selectride and acidic hydrolysis of sulfinamide 81 . The latter fluorinated building block is coupled with acid 83 (as mixture of diastereoisomers) using PyBop as activating agents. A final chiral SFC resolution produces Pralsetinib as a single enantiomer. The method for the patented synthesis of Pralsetinib is described in . Selumetinib is sold under the brand name Koselugo and was developed by AstraZeneca. It was approved in April 2020 for the treatment of neurofibromatosis type 1 (NF1) . Selumetinib , characterized by the presence of a 4-fluorobenzimidazole core, is a mitogen-activated protein kinase (MEK) inhibitor that is able to target the Raf-MEK-ERK signaling pathway. The synthesis of Selumetinib is described in . Trifluorobenzoic acid 84 was employed as starting fluorinated building block for the initial construction of the fluorobenzimidazole ring. The nitration of 84 into 85 and the nucleophilic aromatic displacement of fluoride in nitro-activated derivative 85 with ammonia led to the acquisition of compound 86 . Treatment with trimethylsilyldiazomethane (TMS-CHN 2 ) converts acid 86 into methyl ester 87 , which, in turn, reacts with aniline in xylene at 125 °C, causing the nucleophilic displacement of a second fluorine atom into 88 . The latter nitro derivative is reduced to o -diamminobenzene 89 via iron and ammonium chloride. The benzimidazole ring formation on 90 is performed with formamidine acetate (FAA) in EtOH at 80 °C. The halogenation of the anilino portion of 90 in two consecutive steps produces 91 , with the NBS-mediated introduction of bromine in the para position, and then, 92 , with ortho -chlorination performed with NCS. The methylation of 92 with methyliodide, employing K 2 CO 3 in DMF, occurs at N-1, yielding regio-isomer 93 . Basic hydrolysis into acid 94 is followed by coupling with O -(2-(vinyloxy)ethyl)hydroxylamine in the presence of EDC and HOBt. Hydroxamic acid derivative 95 is finally deprotected via the acidic hydrolysis of the vinyl ether portions, producing Selumetinib . Tauvid , also named Flortaucipir F18, was developed by Eli Lilly and was approved in May 2020 as a positron emission tomography (PET) imaging probe for Alzheimer’s disease (AD) . Tauvid is the first approved tracer able to bind tau protein . The [ 18 F]fluoropyridine lateral ring is the radioactive portion of this probe. Two developed synthetic procedures are described in . The first synthetic approach is based on the nucleophilic displacement of the nitro group of precursor 96 with [ 18 F]fluoride and using Kryptofix 222 (K 222 ) as a phase transfer catalyst (PTC) . This approach suffers from some drawbacks related to trace purity; therefore, a different synthetic approach was developed starting with N-Boc protected cation 97 , which undergoes nucleophilic displacement to produce radioactive fluorine, followed by acidic Boc removal . This synthetic sequence allows the obtainment of Tauvid in higher yields and purity. Interestingly, this synthesis represents the only example of late-stage fluorination among all the molecules considered in this review. Obviously, the short half-life of 18 F forces researchers to follow this peculiar synthetic approach. In 2019, the FDA approved 48 new molecular entities, including 33 small molecules and 3 diagnostic agents . Among the small molecules, 28 out of 33 contain at least one heterocyclic ring, and 11 out of 33 contain at least one fluorine atom . In the following paragraph, four heterocyclic compounds bearing a fluorinated moiety directly linked to the ring are reported ( , , and ). 5.1. Alpelisib Alpelisib is sold under the brand name Rinvoq and was developed by Novartis. It was approved in May 2019 for the treatment of advanced or metastatic breast cancer . Alpelisib is a phosphatidylinositol 3-kinase (PI3K) inhibitor possessing a trifluoro- t -butyl group at position two of the pyridine ring . The presence of the fluorinated moiety induces a higher metabolic stability and excellent oral bioavailability. Furthermore, the fluorinated group is responsible for high affinity toward the PI3K binding site due to the hydrogen bond with K802, as revealed by X-ray data . Alpelisib synthesis is performed using different approaches . One method is shown in . Fluorinated acid 98 is first converted into the corresponding chloride, 99 , via oxalyl chloride under reflux. The acylation of the methyl group of enone 100 is performed at −78 °C using LiHMDS as a strong base. The intermediate diketone is directly cyclized to pyran-4-one 101 after a treatment with trifluoroacetic acid (TFA). The reaction of 101 with ammonium hydroxide produces fluorinated pyridin-4-one 102 via an ANRORC-like reaction. The treatment with POBr 3 yields bromopyridine 103 , which is, in turn, coupled with acetylaminothiazole 104 using Pd(OAc) 2 in a CH activation process. The resulting coupled product, 105 , is hydrolyzed using HCl into corresponding amine 106 . The treatment with carbonyldiimidazole (CDI) leads to intermediate 107 , which is then converted into Alpelisib after a treatment with S -prolinamide. 5.2. Lemborexant Lemborexant is sold under the brand name Dayvigo and was developed by Purdue Pharma L. Lemborexant is a dual orexin receptor antagonist, with fluorine in the position five of the pyridine moiety . Another fluorine is also present at position three of the central phenyl ring. It was approved in December 2019 for the treatment of insomnia . The presence of each fluorine is crucial to achieve high in vitro binding affinity, good solubility and a good pharmacological profile, as evidenced during the discovery process with the screening of different fluorination patterns . The synthesis of Lemborexant starts with 3-fluorobenzyl cyanide 108 , which is converted into chiral cyclopropane derivative 109 in three steps . Dimethylpyrimidine derivative 110 is used in a reaction under Mitsunobu conditions to obtain ether 111 . The conversion of the primary alcoholic function of 111 into corresponding carboxylic acid 112 is performed in two steps with Swern oxidation, followed by Pinnick oxidation. The final production of Lemborexant requires the coupling of acid 112 with 2-amino-5-fluoropyridine 113 using HATU as an activating agent. 5.3. Pexidartinib Pexidartinib is sold under the brand name Turalio and was developed by Daiichi Sankyo Inc. It was approved in August 2019 for the treatment of symptomatic tenosynovial giant cell tumor (TGCT) . Pexidartinib is a tyrosine kinase inhibitor with selective efficacy for colony-stimulating factor (CSF) receptor; thus, it hampers the binding of CSF1 to CSF-receptor 1 (CSF1R). Three steps in the synthesis of Pexidartinib at the kilogram scale are shown in . The base-mediated reaction of azaindole 114 at position 3, over aldehyde 115 , in the presence of tetrabutylammonium hydrogen sulphate (TBAHS) yields compound 116 . The dehydroxylation of 116 with triethylsilane (TES), followed by the Boc-deprotection of the 2-aminopyiridino moiety with TFA, produce compound 117 . Pexidartinib is obtained by means of the reductive amination of 117 with 6-(trifluoromethyl)nicotinaldehyde 118 using TES as a reducing agent. 5.4. Ubrogepant Ubrogepant is sold under the brand name Ubrelvy and was developed by Allergan. It was approved in December 2019 for the treatment of migraines with or without an aura in adults . Ubrogepant is an effective calcitonin gene-related peptide (CGRP) receptor antagonist, bearing a chiral N -trifluoroethylpiperidinone ring, but its mechanism of action is still unknown. The synthesis of Ubrogepant was patented in 2012 and is reported in . The synthesis of fluorinated chiral amine 123 , starting with phenylacetone 120 , which alkylates with iodide 119 in the presence of Cs 2 CO 3 as a base, produces derivative 121 in three steps. The reductive amination of the latter substance using trifluoroethylamine in the presence of sodium triacetoxyborohydride, leads to pyridinone 122 as a single enantiomer, after chiral resolution via normal-phase liquid chromatography (NPLC). The treatment with HCl deprotects the N-Boc group, yielding 123 . The coupling of amine 123 with acidic spyro-subunit 128 (prepared as outlined in ) using BOP as an activating agent yields Ubrogepant as a single enantiomer via SFC. Alpelisib is sold under the brand name Rinvoq and was developed by Novartis. It was approved in May 2019 for the treatment of advanced or metastatic breast cancer . Alpelisib is a phosphatidylinositol 3-kinase (PI3K) inhibitor possessing a trifluoro- t -butyl group at position two of the pyridine ring . The presence of the fluorinated moiety induces a higher metabolic stability and excellent oral bioavailability. Furthermore, the fluorinated group is responsible for high affinity toward the PI3K binding site due to the hydrogen bond with K802, as revealed by X-ray data . Alpelisib synthesis is performed using different approaches . One method is shown in . Fluorinated acid 98 is first converted into the corresponding chloride, 99 , via oxalyl chloride under reflux. The acylation of the methyl group of enone 100 is performed at −78 °C using LiHMDS as a strong base. The intermediate diketone is directly cyclized to pyran-4-one 101 after a treatment with trifluoroacetic acid (TFA). The reaction of 101 with ammonium hydroxide produces fluorinated pyridin-4-one 102 via an ANRORC-like reaction. The treatment with POBr 3 yields bromopyridine 103 , which is, in turn, coupled with acetylaminothiazole 104 using Pd(OAc) 2 in a CH activation process. The resulting coupled product, 105 , is hydrolyzed using HCl into corresponding amine 106 . The treatment with carbonyldiimidazole (CDI) leads to intermediate 107 , which is then converted into Alpelisib after a treatment with S -prolinamide. Lemborexant is sold under the brand name Dayvigo and was developed by Purdue Pharma L. Lemborexant is a dual orexin receptor antagonist, with fluorine in the position five of the pyridine moiety . Another fluorine is also present at position three of the central phenyl ring. It was approved in December 2019 for the treatment of insomnia . The presence of each fluorine is crucial to achieve high in vitro binding affinity, good solubility and a good pharmacological profile, as evidenced during the discovery process with the screening of different fluorination patterns . The synthesis of Lemborexant starts with 3-fluorobenzyl cyanide 108 , which is converted into chiral cyclopropane derivative 109 in three steps . Dimethylpyrimidine derivative 110 is used in a reaction under Mitsunobu conditions to obtain ether 111 . The conversion of the primary alcoholic function of 111 into corresponding carboxylic acid 112 is performed in two steps with Swern oxidation, followed by Pinnick oxidation. The final production of Lemborexant requires the coupling of acid 112 with 2-amino-5-fluoropyridine 113 using HATU as an activating agent. Pexidartinib is sold under the brand name Turalio and was developed by Daiichi Sankyo Inc. It was approved in August 2019 for the treatment of symptomatic tenosynovial giant cell tumor (TGCT) . Pexidartinib is a tyrosine kinase inhibitor with selective efficacy for colony-stimulating factor (CSF) receptor; thus, it hampers the binding of CSF1 to CSF-receptor 1 (CSF1R). Three steps in the synthesis of Pexidartinib at the kilogram scale are shown in . The base-mediated reaction of azaindole 114 at position 3, over aldehyde 115 , in the presence of tetrabutylammonium hydrogen sulphate (TBAHS) yields compound 116 . The dehydroxylation of 116 with triethylsilane (TES), followed by the Boc-deprotection of the 2-aminopyiridino moiety with TFA, produce compound 117 . Pexidartinib is obtained by means of the reductive amination of 117 with 6-(trifluoromethyl)nicotinaldehyde 118 using TES as a reducing agent. Ubrogepant is sold under the brand name Ubrelvy and was developed by Allergan. It was approved in December 2019 for the treatment of migraines with or without an aura in adults . Ubrogepant is an effective calcitonin gene-related peptide (CGRP) receptor antagonist, bearing a chiral N -trifluoroethylpiperidinone ring, but its mechanism of action is still unknown. The synthesis of Ubrogepant was patented in 2012 and is reported in . The synthesis of fluorinated chiral amine 123 , starting with phenylacetone 120 , which alkylates with iodide 119 in the presence of Cs 2 CO 3 as a base, produces derivative 121 in three steps. The reductive amination of the latter substance using trifluoroethylamine in the presence of sodium triacetoxyborohydride, leads to pyridinone 122 as a single enantiomer, after chiral resolution via normal-phase liquid chromatography (NPLC). The treatment with HCl deprotects the N-Boc group, yielding 123 . The coupling of amine 123 with acidic spyro-subunit 128 (prepared as outlined in ) using BOP as an activating agent yields Ubrogepant as a single enantiomer via SFC. In 2018, the FDA approved a collection of 59 new molecular entities, including 39 small molecules . Thirty-two out of thirty-eight molecules contain at least one heterocyclic ring, and seventeen out of thirty-eight molecules contain at least one fluorine atom . In the following paragraphs, eight heterocyclic compounds bearing a fluorinated moiety directly linked to the ring are reported ( , , , , , , and ). 6.1. Apalutamide Apalutamide is sold under the brand name Erleada and was discovered by employees of the University of California and developed by Janssen . It was approved in February 2018 for the treatment of prostate cancer (PC) . Apalutamide is a non-steroidal oral androgen receptor inhibitor, presenting a trifluoromethylpyridine moiety linked to the central thiohydantoin core . One of the initial patented Apalutamide synthesis procedures is reported below . Chloro-trifluoromethylpyridine 129 is treated with water to induce the nucleophilic displacement of chlorine to obtain pyrimidone 130 . This compound can be easily nitrated into 131 , and again, converted into corresponding chloropyridine 132 via a treatment with PCl 5 /POCl 3 . The hydrogenation on Raney Ni into amine 133 is then followed by N-Boc protection with Boc-anhydride, yielding 134 . A Sandmeyer reaction causes cyanation to produce derivative 135 , which is subsequently deprotected into aminopyrimidine 136 using TFA. Isothiocyanate 137 is then obtained via the treatment the 136 using thiophosgene. Apalutamide is finally obtained from the construction of the thiohydantoin ring via the reaction of 137 with isocyanide 138 under microwave irradiation. The synthesis of crystalline forms of Apalutamide has recently been reviewed . 6.2. Baloxavir Marboxil Baloxavir marboxil is sold under the brand name Xofluza and was developed by Shionogi . It was approved in October 2018 for the treatment of acute uncomplicated influenza types A and B . Baloxavir marboxil is a cap-dependent endonuclease inhibitor characterized by the presence of two fluorine atoms on the 6,11-dihydrodibenzo[b,e]thiepine ring . The patented synthesis of Baloxavir marboxil is depicted in . Difluorobenzoic acid 139 is initially formylated via LDA/DMF at low temperature, yielding cyclization product 140 . A reaction with thiophenol under acidic conditions produces phenylthiolphthalide 141 . The reductive breaking of the C-O bond mediated by the action 1,1,3,3-Tetramethyldisiloxane yields thioether 142 , which undergoes intramolecular cyclization by heating at 120 °C. The obtained ketone 143 is reduced into alcohol 144 using NaBH 4 , and then coupled with chiral compound 145 under acidic conditions to produce compound 146 . The deprotection of hexyl ether into 147 , and a final reaction with chloromethyl methyl carbonate allows the obtainment of Baloxavir marboxil . 6.3. Binimetinib Binimetinib is sold under the brand name Mektovi and was developed by Array Biopharma. It was approved in June 2018 for the treatment of metastatic BRAF V600E/K-positive advanced melanoma in association with Encorafenib . Binimetinib is a potent, selective, non-ATP competitive allosteric inhibitor of MEK1 and MEK2, with a fluorobenzimidazole moiety similar to that of Selumetinib . As for the analogue, Selumetinib , the synthetic route of Binimetinib is based on ester 87 . The nucleophilic displacement of 2-fluoroaniline 148 to obtain derivative 149 is followed by hydrogenation in the presence of formic acid to directly yield benzimidazole 150 . NBS-mediated bromination and methylation at N-1 give compounds 151 and 152 , respectively. Ester 117 is, therefore, hydrolyzed using NaOH, and the resulting acid, 153 , after activation using EDC/HOBt is converted into hydroxamic acid 155 upon a reaction with O-alkyl hydroxylamine 154 . Binimetinib is finally obtained by the acidic hydrolysis of the vinyl ether group of 155 . 6.4. Doravirine Doravirine is sold under the brand name Pifeltro and was developed by Merck. It was approved in August 2018 for the treatment of human immunodeficiency virus 1 (HIV-1) infections . Doravirine is a non-nucleoside reverse transcriptase inhibitor, presenting improved ADME properties due the presence of a CF 3 -substituted pyridone central ring. In fact, the presence of this strong electron-withdrawing group is correlated with a longer elimination half-life in rats and dogs compared to that of unfluorinated analogs . The patented method of the synthesis of Doravirine is shown in . Fluorinated building block 2-chloro-3-fluoro-4-(trifluoromethyl)pyridine 156 is first used in a reaction with phenol 157 using K 2 CO 3 as a base to induce fluoride displacement and obtain ether 158 . The hydrolysis of chloropyridine 158 into pyridinone 159 is then performed via a treatment with KOH. The cyanation of the C-Br bond of the phenyl portion with CuCN yields substrate 160 , which reacts under basic condition with chloromethyltriazole 161 to produce pyridone N -alkylation product 162 . Doravirine is finally obtained by the regio-selective methylation of the 1,2,4-triazole ring at N-4 with methyliodide in DMF in the presence of K 2 CO 3 as a base. Other synthetic approaches have recently been reviewed . 6.5. Fostamatinib Fostamatinib is sold under the brand name Tavalisse and was developed by Rigel Pharmaceuticals. It was approved in April 2018 for the treatment of thrombocytopenia in adults with persistent or chronic immune thrombocytopenia (ITP) . Fostamatinib is a potent spleen tyrosine kinase (Syk) inhibitor, bearing a 5-fluoropyrimidine ring, and it is used to improve membranes’ permeability . Actually, Fostamatinib is a pro-drug of compound 168 , and its synthesis is described in . 5-Fluoropyrimidine-2,4-diol 163 is converted into dichloro-derivative 164 after a treatment with POCl 3 . Two subsequent chloride displacements with different amines then occur. The first one at C-4 with amino-pyridoxazinone 165 yields 166 ; the second one at C-2 with 3,4,5-trimethoxyaniline 167 produces compound 168 . As mentioned above, this compound is converted into the pro-drug Fostamatinib via a treatment with di-tert-butyl(chloromethyl)phosphate to produce ester 169 , which is then hydrolyzed and converted into target phosphate disodium salt. 6.6. Ivosidenib Ivosidenib is sold under the brand name Tibsovo and was developed by Servier. It was approved in July 2018 for the treatment of relapsed or refractory acute myeloid leukemia . Ivosidenib is an inhibitor of mutated cytosolic isocitrate dehydrogenase 1 (IDH1); thus, it lowers the level of oncometabolite D -2-hydroxyglutarate (2-HG) . Fluorine at the position five of the pyridine ring is crucial in order to ensure a high level of potency and metabolic stability . The synthesis of Ivosidenib at the multi-gram scale is outlined in . The synthetic process is based on a multi-component Ugi reaction of 3-amino-5-fluoropyridine 170 with 2-chlorobenzaldehyde 171 , followed by L -pyroglutamic acid and isocyanide 172 , to produce peptide derivative 173 . N-H coupling with bromopyridine 174 under Buchwald conditions produces Ivosidenib as a single enantiomer after crystallization. 6.7. Talazoparib Talazoparib is sold under the brand name Talzenna and was developed by Pfizer . It was approved in October 2018 for the treatment of locally advanced or metastatic breast cancer patients with a germline BRCA mutation . Talazoparib , a fluorine-containing tetrahydropyridophthlazinones is active as a poly(ADP-ribose) polymerase (PARP) inhibitor . To achieve inhibitory activity and metabolic stability, as well as to increase the number of interactions at the binding site via H-bonding, 5-fluoro substitution and the 4-fluorophenyl groups are crucial . The synthetic method for the preparation of Talazoparib at the 30 g scale is described in . Fluorinated dihydroquinolinone 177 is constructed with a two steps acylation/reductive amination of fluoroaniline 175 with β-ketoacid 176 . The following chiral resolution with -tartaric acid is crucial to obtain the desired enantiomer, 178 . The base-induced reaction between 178 and 5-chloro-1-methyl-1,2,4-triazole 179 leads to the desired trans stereoisomer, 180 , which is finally converted into Talazoparib via a reaction with hydrazine in EtOH under reflux. 6.8. Tezacaftor Tezacaftor is sold under the brand name Symdeko, as a co-formulation with ivacaftor and was developed by Vertex Pharms Inc. It was approved in February 2018 for the treatment of cystic fibrosis . Tezacaftor improves the processing and trafficking of cystic fibrosis transmembrane conductance regulator (CFTR) in vitro and improves CFTR’s function alone and in combination with other drugs . The synthesis of Tezacaftor has been recently reviewed , the second generation process is described in . The scheme is based on the initial formation of 6-fluoroindole’s nucleus using 3-fluoro-4-nitroaniline 181 . Bromination with elemental bromine in acetic acid yields compound 182 , and the nucleophilic ring opening by the anilino moiety on chiral epoxide 183 produces compound 184 . Nitro-group reduction with Zinc and salt formation with PTSA from amine 185 ,produce tosylate 186 . The latter one is coupled with terminal alkyne 187 under Sonogashira conditions to produce compound 188 . Indole ring formation is achieved via a Pd-catalyzed reaction using Pd(CH 3 CN) 2 Cl 2 . 6-Fluoroindole 189 , obtained as a single enantiomer, is then used in a reaction with chloride 190 to yield Bn-protected derivative 191 , which is finally converted into Tezacaftor by means of hydrogenation over Pd/C. Apalutamide is sold under the brand name Erleada and was discovered by employees of the University of California and developed by Janssen . It was approved in February 2018 for the treatment of prostate cancer (PC) . Apalutamide is a non-steroidal oral androgen receptor inhibitor, presenting a trifluoromethylpyridine moiety linked to the central thiohydantoin core . One of the initial patented Apalutamide synthesis procedures is reported below . Chloro-trifluoromethylpyridine 129 is treated with water to induce the nucleophilic displacement of chlorine to obtain pyrimidone 130 . This compound can be easily nitrated into 131 , and again, converted into corresponding chloropyridine 132 via a treatment with PCl 5 /POCl 3 . The hydrogenation on Raney Ni into amine 133 is then followed by N-Boc protection with Boc-anhydride, yielding 134 . A Sandmeyer reaction causes cyanation to produce derivative 135 , which is subsequently deprotected into aminopyrimidine 136 using TFA. Isothiocyanate 137 is then obtained via the treatment the 136 using thiophosgene. Apalutamide is finally obtained from the construction of the thiohydantoin ring via the reaction of 137 with isocyanide 138 under microwave irradiation. The synthesis of crystalline forms of Apalutamide has recently been reviewed . Baloxavir marboxil is sold under the brand name Xofluza and was developed by Shionogi . It was approved in October 2018 for the treatment of acute uncomplicated influenza types A and B . Baloxavir marboxil is a cap-dependent endonuclease inhibitor characterized by the presence of two fluorine atoms on the 6,11-dihydrodibenzo[b,e]thiepine ring . The patented synthesis of Baloxavir marboxil is depicted in . Difluorobenzoic acid 139 is initially formylated via LDA/DMF at low temperature, yielding cyclization product 140 . A reaction with thiophenol under acidic conditions produces phenylthiolphthalide 141 . The reductive breaking of the C-O bond mediated by the action 1,1,3,3-Tetramethyldisiloxane yields thioether 142 , which undergoes intramolecular cyclization by heating at 120 °C. The obtained ketone 143 is reduced into alcohol 144 using NaBH 4 , and then coupled with chiral compound 145 under acidic conditions to produce compound 146 . The deprotection of hexyl ether into 147 , and a final reaction with chloromethyl methyl carbonate allows the obtainment of Baloxavir marboxil . Binimetinib is sold under the brand name Mektovi and was developed by Array Biopharma. It was approved in June 2018 for the treatment of metastatic BRAF V600E/K-positive advanced melanoma in association with Encorafenib . Binimetinib is a potent, selective, non-ATP competitive allosteric inhibitor of MEK1 and MEK2, with a fluorobenzimidazole moiety similar to that of Selumetinib . As for the analogue, Selumetinib , the synthetic route of Binimetinib is based on ester 87 . The nucleophilic displacement of 2-fluoroaniline 148 to obtain derivative 149 is followed by hydrogenation in the presence of formic acid to directly yield benzimidazole 150 . NBS-mediated bromination and methylation at N-1 give compounds 151 and 152 , respectively. Ester 117 is, therefore, hydrolyzed using NaOH, and the resulting acid, 153 , after activation using EDC/HOBt is converted into hydroxamic acid 155 upon a reaction with O-alkyl hydroxylamine 154 . Binimetinib is finally obtained by the acidic hydrolysis of the vinyl ether group of 155 . Doravirine is sold under the brand name Pifeltro and was developed by Merck. It was approved in August 2018 for the treatment of human immunodeficiency virus 1 (HIV-1) infections . Doravirine is a non-nucleoside reverse transcriptase inhibitor, presenting improved ADME properties due the presence of a CF 3 -substituted pyridone central ring. In fact, the presence of this strong electron-withdrawing group is correlated with a longer elimination half-life in rats and dogs compared to that of unfluorinated analogs . The patented method of the synthesis of Doravirine is shown in . Fluorinated building block 2-chloro-3-fluoro-4-(trifluoromethyl)pyridine 156 is first used in a reaction with phenol 157 using K 2 CO 3 as a base to induce fluoride displacement and obtain ether 158 . The hydrolysis of chloropyridine 158 into pyridinone 159 is then performed via a treatment with KOH. The cyanation of the C-Br bond of the phenyl portion with CuCN yields substrate 160 , which reacts under basic condition with chloromethyltriazole 161 to produce pyridone N -alkylation product 162 . Doravirine is finally obtained by the regio-selective methylation of the 1,2,4-triazole ring at N-4 with methyliodide in DMF in the presence of K 2 CO 3 as a base. Other synthetic approaches have recently been reviewed . Fostamatinib is sold under the brand name Tavalisse and was developed by Rigel Pharmaceuticals. It was approved in April 2018 for the treatment of thrombocytopenia in adults with persistent or chronic immune thrombocytopenia (ITP) . Fostamatinib is a potent spleen tyrosine kinase (Syk) inhibitor, bearing a 5-fluoropyrimidine ring, and it is used to improve membranes’ permeability . Actually, Fostamatinib is a pro-drug of compound 168 , and its synthesis is described in . 5-Fluoropyrimidine-2,4-diol 163 is converted into dichloro-derivative 164 after a treatment with POCl 3 . Two subsequent chloride displacements with different amines then occur. The first one at C-4 with amino-pyridoxazinone 165 yields 166 ; the second one at C-2 with 3,4,5-trimethoxyaniline 167 produces compound 168 . As mentioned above, this compound is converted into the pro-drug Fostamatinib via a treatment with di-tert-butyl(chloromethyl)phosphate to produce ester 169 , which is then hydrolyzed and converted into target phosphate disodium salt. Ivosidenib is sold under the brand name Tibsovo and was developed by Servier. It was approved in July 2018 for the treatment of relapsed or refractory acute myeloid leukemia . Ivosidenib is an inhibitor of mutated cytosolic isocitrate dehydrogenase 1 (IDH1); thus, it lowers the level of oncometabolite D -2-hydroxyglutarate (2-HG) . Fluorine at the position five of the pyridine ring is crucial in order to ensure a high level of potency and metabolic stability . The synthesis of Ivosidenib at the multi-gram scale is outlined in . The synthetic process is based on a multi-component Ugi reaction of 3-amino-5-fluoropyridine 170 with 2-chlorobenzaldehyde 171 , followed by L -pyroglutamic acid and isocyanide 172 , to produce peptide derivative 173 . N-H coupling with bromopyridine 174 under Buchwald conditions produces Ivosidenib as a single enantiomer after crystallization. Talazoparib is sold under the brand name Talzenna and was developed by Pfizer . It was approved in October 2018 for the treatment of locally advanced or metastatic breast cancer patients with a germline BRCA mutation . Talazoparib , a fluorine-containing tetrahydropyridophthlazinones is active as a poly(ADP-ribose) polymerase (PARP) inhibitor . To achieve inhibitory activity and metabolic stability, as well as to increase the number of interactions at the binding site via H-bonding, 5-fluoro substitution and the 4-fluorophenyl groups are crucial . The synthetic method for the preparation of Talazoparib at the 30 g scale is described in . Fluorinated dihydroquinolinone 177 is constructed with a two steps acylation/reductive amination of fluoroaniline 175 with β-ketoacid 176 . The following chiral resolution with -tartaric acid is crucial to obtain the desired enantiomer, 178 . The base-induced reaction between 178 and 5-chloro-1-methyl-1,2,4-triazole 179 leads to the desired trans stereoisomer, 180 , which is finally converted into Talazoparib via a reaction with hydrazine in EtOH under reflux. Tezacaftor is sold under the brand name Symdeko, as a co-formulation with ivacaftor and was developed by Vertex Pharms Inc. It was approved in February 2018 for the treatment of cystic fibrosis . Tezacaftor improves the processing and trafficking of cystic fibrosis transmembrane conductance regulator (CFTR) in vitro and improves CFTR’s function alone and in combination with other drugs . The synthesis of Tezacaftor has been recently reviewed , the second generation process is described in . The scheme is based on the initial formation of 6-fluoroindole’s nucleus using 3-fluoro-4-nitroaniline 181 . Bromination with elemental bromine in acetic acid yields compound 182 , and the nucleophilic ring opening by the anilino moiety on chiral epoxide 183 produces compound 184 . Nitro-group reduction with Zinc and salt formation with PTSA from amine 185 ,produce tosylate 186 . The latter one is coupled with terminal alkyne 187 under Sonogashira conditions to produce compound 188 . Indole ring formation is achieved via a Pd-catalyzed reaction using Pd(CH 3 CN) 2 Cl 2 . 6-Fluoroindole 189 , obtained as a single enantiomer, is then used in a reaction with chloride 190 to yield Bn-protected derivative 191 , which is finally converted into Tezacaftor by means of hydrogenation over Pd/C. In 2017, the FDA approved 46 new drugs, including 34 small molecules . Thirty-one out of thirty-four molecules contain at least one heterocyclic ring, and ten out of thirty-four molecules contain at least one fluorine atom. In the following paragraphs, seven heterocyclic compounds bearing a fluorinated moiety directly linked to the ring are reported ( , , , , , and ). 7.1. Abemaciclib Abemaciclib is sold under the brand name Verzenio and was developed by Eli Lilly. It was approved in September 2017 for the treatment of advanced or metastatic breast cancers. Abemaciclib is a cyclin-dependent kinase (CDK) inhibitor that is selective for isoforms CDK4 and CDK6 . Abemaciclib contains two fluorinated heterocycles, namely 4-fluorobenzimidazole and 5-fluoropyrimidine, which are directly linked to form a 6-(pyrimidin-4-yl)-benzimidazole core . The synthesis described by Frederick et al. starts with the formation of this bond via a Suzuki reaction between fluorobenzimidazolyl pinacol boronate 192 and 2,4-dichloro-5-fluoropyrimidine 193 . The reaction occurs selectively with the displacement of chlorine at position 4, producing 194 , while less-reactive chlorine at position 2 is then used in a reaction with aminopyridine 195 under Buchwald–Hartwig conditions to yield intermediate 196 . The latter substance is converted into Abemaciclib through reductive amination with ethylpyperazine 197 via a Leuckart–Wallach reaction, with trimethyl orthoformate as a dehydrating agent. A further improvement has introduced a more convergent scheme, which involves the performance of flow synthesis . 7.2. Delafloxacin Delafloxacin is sold under the brand name Baxdela and was developed by Melinta. It was approved in June 2017 for the treatment of acute bacterial skin and skin structure infections . Such as other members of the fluoroquinolone family, it is a DNA gyrase topoisomerase IV inhibitor that is active against Gram-positive bacteria, including methicillin-resistant Staphylococcus Aureus (MRSA), and Gram-negative organisms, such as Escherichia Coli and Pseudomonas Aeruginosa . Additionally, some quinolone-resistant strains are susceptible to Delafloxacin . The synthetic process is in line with the classical fluoroquinolone method . Trifluorobenzoic acid 198 is initially converted into the corresponding chloride with thionylchloride, and then into β-ketoester 200 via a treatment with potassium monoethylmalonate 199 . Compound 200 is then converted into an intermediate vinylether, which is directly transformed into enamine 202 after the reaction with 2,6-diamino-3,5-difluoropyridine 201 . The cyclization of compound 202 into the corresponding quinolone by the nucleophilic displacement of ortho fluorine is induced by the addition of DBU. The second aromatic nucleophilic substitution, involving fluorine at position seven, is performed with 3-hydroxyazetidine 203 . Compound 204 is then obtained via the protection of an hydroxyl group as an ester to avoid competitive oxidation in the following chlorination step. Chlorination at position eight of the quinolone ring is selectively performed using NCS as a chlorinating agent in an acidic environment. Finally, Delafloxacin is obtained after the deprotection of the hydroxyazetine portion by means of ester hydrolysis with NaOH. 7.3. Enasidenib Enasidenib is sold under the brand name Idhifa and was developed by Celgene. It was approved in August 2017 for the treatment of relapsed or refractory acute myeloid leukemia in patients with specific mutations of the isocitrate dehydrogenase 2 (IDH2) gene . Enasidenib is a first-in-class small-molecule inhibitor of the IDH2-mutant enzyme with oral bioavailability . This drug contains two trifluoromethylpyridine rings, as demonstrated by ab initio calculations with X-ray data; one trifluoromethyl group is important for the CF 3 ···O tetrel bond with Asp312 . The same CF 3 -group is also responsible for C-H···F bonding with Asp312 and N-H···F bonding with Gln316. The synthesis of Enasidenib was patented in 2013 . Trifluoromethylpycolinate 205 is condensed using biuret 206 in refluxing EtOH in the presence of sodium metal to produce 1,3,5-triazin-2,4-dione 207 . Chlorination with PCl 5 in POCl 3 produces dichlorotriazine 208 . The nucleophilic displacement of aminoalcohol 209 produces compound 210 . The Buchwald–Hartwig Pd-catalyzed N-arylation of 4-amino-2-(trifluoromethyl)pyridine 211 with chloride 210 forms Enasidenib . 7.4. Glecaprevir Glecaprevir is sold under the brand name Mavyret, as a co-formulation with Pibrentasvir (see ), and was developed by AbbVie Inc. It was approved in August 2017 for the treatment of chronic hepatitis C virus (HCV) in adults . In 2019, the FDA expanded the use to children. Glecaprevir is a non-structural (NS) protein 3/4A protease inhibitor, presenting a macrocyclic ring with a difluoromethylene moiety directly linked to a quinoxaline ring . The enabling synthesis of Glecaprevir to produce the quantity needed for Phase I clinical trials is based on ring-closing metathesis (RCM) for the production of the 18-membered macrocycle . The synthetic route starts with the formation of fluorinated α-hydroxyester 214 via the Indium-mediated allylation of ethyl glyoxylate 213 starting with 3-bromo-3,3-difluoro-propene 212 . The Swern oxidation of propylphosphonic anhydride (T3P) into intermediate α-ketoester is followed by condensation with ortho-phenylenediamine 215 to produce gem -difluoro quinoxaline 216 . Chlorination with thionyl chloride produces derivative 217 , possessing a good leaving group for nucleophilic aromatic substitution with Boc-protected hydroxyproline 218 . Concurrent methyl ester formation and the removal of Boc protection via treating 219 with HCl in MeOH produces amine 220 , which is one of the two main building blocks for macrocycle formation. The second main component, acid 221 , is coupled, inducing amide bond formation when employing HATU as an activating agent. Diene 222 is then subjected to RCM using Zhan 1B catalyst after careful screening of the reaction conditions, thereby optimizing the formation of the desired trans macrocycle, 223 . Glecaprevir is obtained in two steps via the hydrolysis of 223 into acid 224 and its coupling with the fluorinated aminocyclopropane 225 side-chain, combining EDC and 2-hydroxypyridine- N -oxide (HOPO) as activating agents. Another approach based on ether-bond macrocyclization has also been developed for large-scale synthesis . 7.5. Letermovir Letermovir is sold under the brand name Prevymis and was developed by Merck & Co. It was approved in November 2017 for the treatment of infections caused by cytomegalovirus (CMV) after a bone marrow transplant . Letermovir’s mode of action is different from that of other antiviral agents, which act on DNA polymerase; in fact, it interferes with the activity of terminase complex of the virus . The asymmetric synthesis of Letermovir is performed in seven steps with a key part: PTC-mediated aza-Michael cyclization to obtain the chiral fluorinated dihydroquinazoline core . The formation of aminocinnamate 228 is based on a Heck reaction between fluoroaniline 226 and methyl acrylate 227 . Carbamate 229 is then obtained treating 228 with phenyl chloroformate. Urea 231 is formed via a reaction with anisidine 230 . Compound 231 is dehydrated with PCl 5 into carbodiimide 232 and directly converted into guanidine 234 via a treatment with piperazine 233 . Compound 234 is the key intermediate used for asymmetric cyclization into compound 236 using fluorinated cinchona-based derivative 235 as a PTC catalyst. Precursor 236 is converted into Letermovir via the hydrolysis of the methyl ester moiety. Other asymmetric approaches to Letermovir synthesis were developed later . 7.6. Pibrentasvir Pibrentasvir is sold under the brand name Mavyret, as a co-formulation with Glecaprevir (see ), and was developed by AbbVie Inc. It was approved in August 2017 for the treatment of chronic hepatitis C virus (HCV) in adults . Pibrentasvir is an NS5A inhibitor antiviral agent with two symmetric fluorobenzimidazole rings linked to a central trans pyrrolidine core . The method for the patented synthesis of Pibrentasvir is shown in . Fluoro-acetophenone 237 was brominated using the methyl group, producing α-bromoketone 238 . The ZnCl 2 -mediated C-C coupling of 237 with 238 produces diketone 239 . Stereoselective reduction in the presence of prolinol-derived catalyst 240 yields an intermediate diol, which is directly converted into dimesylate 241 . Double nucleophilic displacement with aniline 242 produces trans pyrrolidine 243 , which is then treated with N-Boc prolinamide under Buchwald conditions to yield N-arylation at both rings in compound 244 . The latter substance is converted into bis-benzimidazole 245 via the hydrogenation of nitro groups, AcOH-mediated cyclization and TFA-induced deprotection. Diamine 245 is finally converted into Pibrentasvir via coupling with protected O -methyl-threonine 246 with EDC and HOBt as an activating agent. 7.7. Voxilaprevir Voxilaprevir is sold under the brand name Vosevi, as a co-formulation with sofosbuvir (see ) and velpatasvir and was developed by Gilead. It was approved in July 2017 for the treatment of chronic hepatitis C virus (HCV) in adults . Voxilaprevir is an NS protein 3/4A protease inhibitor, possessing an 18-membered gem -difluoro methylene quinoxaline portion similar to that of Glecaprevir . Additionally, the synthetic pathways are quite similar . The synthetic method starts with the formation of fluorinated α-ketoester 247 via the lithium exchange and allylation of diethyl oxalate, starting with 3-bromo-3,3-difluoro-propene 212 . Condensation with methoxy ortho-phenylenediamine 248 produces gem -difluoro quinoxaline 249 , which is chlorinated with POCl 3 to produce derivative 250 , which possesses a good leaving group for nucleophilic aromatic substitution with Boc-protected ethyl hydroxyproline 251 . The Boc removal of 252 with HCl furnishes amine 253 , which is one of the two main building blocks for macrocycle formation. The second main component, acid 254 , is coupled to it by inducing amide bond formation and employing HATU as an activating agent. Diene 255 is then subjected to RCM using Zhan 1B catalyst, causing the formation of the desired trans macrocycle, 256. Glecaprevir is obtained in two steps by means of the hydrogenation of the double bond of 256 and t -Bu ester removal to produce acid 257 and its coupling with fluorinated aminocyclopropane 225 side-chain using HATU as an activating agent. Abemaciclib is sold under the brand name Verzenio and was developed by Eli Lilly. It was approved in September 2017 for the treatment of advanced or metastatic breast cancers. Abemaciclib is a cyclin-dependent kinase (CDK) inhibitor that is selective for isoforms CDK4 and CDK6 . Abemaciclib contains two fluorinated heterocycles, namely 4-fluorobenzimidazole and 5-fluoropyrimidine, which are directly linked to form a 6-(pyrimidin-4-yl)-benzimidazole core . The synthesis described by Frederick et al. starts with the formation of this bond via a Suzuki reaction between fluorobenzimidazolyl pinacol boronate 192 and 2,4-dichloro-5-fluoropyrimidine 193 . The reaction occurs selectively with the displacement of chlorine at position 4, producing 194 , while less-reactive chlorine at position 2 is then used in a reaction with aminopyridine 195 under Buchwald–Hartwig conditions to yield intermediate 196 . The latter substance is converted into Abemaciclib through reductive amination with ethylpyperazine 197 via a Leuckart–Wallach reaction, with trimethyl orthoformate as a dehydrating agent. A further improvement has introduced a more convergent scheme, which involves the performance of flow synthesis . Delafloxacin is sold under the brand name Baxdela and was developed by Melinta. It was approved in June 2017 for the treatment of acute bacterial skin and skin structure infections . Such as other members of the fluoroquinolone family, it is a DNA gyrase topoisomerase IV inhibitor that is active against Gram-positive bacteria, including methicillin-resistant Staphylococcus Aureus (MRSA), and Gram-negative organisms, such as Escherichia Coli and Pseudomonas Aeruginosa . Additionally, some quinolone-resistant strains are susceptible to Delafloxacin . The synthetic process is in line with the classical fluoroquinolone method . Trifluorobenzoic acid 198 is initially converted into the corresponding chloride with thionylchloride, and then into β-ketoester 200 via a treatment with potassium monoethylmalonate 199 . Compound 200 is then converted into an intermediate vinylether, which is directly transformed into enamine 202 after the reaction with 2,6-diamino-3,5-difluoropyridine 201 . The cyclization of compound 202 into the corresponding quinolone by the nucleophilic displacement of ortho fluorine is induced by the addition of DBU. The second aromatic nucleophilic substitution, involving fluorine at position seven, is performed with 3-hydroxyazetidine 203 . Compound 204 is then obtained via the protection of an hydroxyl group as an ester to avoid competitive oxidation in the following chlorination step. Chlorination at position eight of the quinolone ring is selectively performed using NCS as a chlorinating agent in an acidic environment. Finally, Delafloxacin is obtained after the deprotection of the hydroxyazetine portion by means of ester hydrolysis with NaOH. Enasidenib is sold under the brand name Idhifa and was developed by Celgene. It was approved in August 2017 for the treatment of relapsed or refractory acute myeloid leukemia in patients with specific mutations of the isocitrate dehydrogenase 2 (IDH2) gene . Enasidenib is a first-in-class small-molecule inhibitor of the IDH2-mutant enzyme with oral bioavailability . This drug contains two trifluoromethylpyridine rings, as demonstrated by ab initio calculations with X-ray data; one trifluoromethyl group is important for the CF 3 ···O tetrel bond with Asp312 . The same CF 3 -group is also responsible for C-H···F bonding with Asp312 and N-H···F bonding with Gln316. The synthesis of Enasidenib was patented in 2013 . Trifluoromethylpycolinate 205 is condensed using biuret 206 in refluxing EtOH in the presence of sodium metal to produce 1,3,5-triazin-2,4-dione 207 . Chlorination with PCl 5 in POCl 3 produces dichlorotriazine 208 . The nucleophilic displacement of aminoalcohol 209 produces compound 210 . The Buchwald–Hartwig Pd-catalyzed N-arylation of 4-amino-2-(trifluoromethyl)pyridine 211 with chloride 210 forms Enasidenib . Glecaprevir is sold under the brand name Mavyret, as a co-formulation with Pibrentasvir (see ), and was developed by AbbVie Inc. It was approved in August 2017 for the treatment of chronic hepatitis C virus (HCV) in adults . In 2019, the FDA expanded the use to children. Glecaprevir is a non-structural (NS) protein 3/4A protease inhibitor, presenting a macrocyclic ring with a difluoromethylene moiety directly linked to a quinoxaline ring . The enabling synthesis of Glecaprevir to produce the quantity needed for Phase I clinical trials is based on ring-closing metathesis (RCM) for the production of the 18-membered macrocycle . The synthetic route starts with the formation of fluorinated α-hydroxyester 214 via the Indium-mediated allylation of ethyl glyoxylate 213 starting with 3-bromo-3,3-difluoro-propene 212 . The Swern oxidation of propylphosphonic anhydride (T3P) into intermediate α-ketoester is followed by condensation with ortho-phenylenediamine 215 to produce gem -difluoro quinoxaline 216 . Chlorination with thionyl chloride produces derivative 217 , possessing a good leaving group for nucleophilic aromatic substitution with Boc-protected hydroxyproline 218 . Concurrent methyl ester formation and the removal of Boc protection via treating 219 with HCl in MeOH produces amine 220 , which is one of the two main building blocks for macrocycle formation. The second main component, acid 221 , is coupled, inducing amide bond formation when employing HATU as an activating agent. Diene 222 is then subjected to RCM using Zhan 1B catalyst after careful screening of the reaction conditions, thereby optimizing the formation of the desired trans macrocycle, 223 . Glecaprevir is obtained in two steps via the hydrolysis of 223 into acid 224 and its coupling with the fluorinated aminocyclopropane 225 side-chain, combining EDC and 2-hydroxypyridine- N -oxide (HOPO) as activating agents. Another approach based on ether-bond macrocyclization has also been developed for large-scale synthesis . Letermovir is sold under the brand name Prevymis and was developed by Merck & Co. It was approved in November 2017 for the treatment of infections caused by cytomegalovirus (CMV) after a bone marrow transplant . Letermovir’s mode of action is different from that of other antiviral agents, which act on DNA polymerase; in fact, it interferes with the activity of terminase complex of the virus . The asymmetric synthesis of Letermovir is performed in seven steps with a key part: PTC-mediated aza-Michael cyclization to obtain the chiral fluorinated dihydroquinazoline core . The formation of aminocinnamate 228 is based on a Heck reaction between fluoroaniline 226 and methyl acrylate 227 . Carbamate 229 is then obtained treating 228 with phenyl chloroformate. Urea 231 is formed via a reaction with anisidine 230 . Compound 231 is dehydrated with PCl 5 into carbodiimide 232 and directly converted into guanidine 234 via a treatment with piperazine 233 . Compound 234 is the key intermediate used for asymmetric cyclization into compound 236 using fluorinated cinchona-based derivative 235 as a PTC catalyst. Precursor 236 is converted into Letermovir via the hydrolysis of the methyl ester moiety. Other asymmetric approaches to Letermovir synthesis were developed later . Pibrentasvir is sold under the brand name Mavyret, as a co-formulation with Glecaprevir (see ), and was developed by AbbVie Inc. It was approved in August 2017 for the treatment of chronic hepatitis C virus (HCV) in adults . Pibrentasvir is an NS5A inhibitor antiviral agent with two symmetric fluorobenzimidazole rings linked to a central trans pyrrolidine core . The method for the patented synthesis of Pibrentasvir is shown in . Fluoro-acetophenone 237 was brominated using the methyl group, producing α-bromoketone 238 . The ZnCl 2 -mediated C-C coupling of 237 with 238 produces diketone 239 . Stereoselective reduction in the presence of prolinol-derived catalyst 240 yields an intermediate diol, which is directly converted into dimesylate 241 . Double nucleophilic displacement with aniline 242 produces trans pyrrolidine 243 , which is then treated with N-Boc prolinamide under Buchwald conditions to yield N-arylation at both rings in compound 244 . The latter substance is converted into bis-benzimidazole 245 via the hydrogenation of nitro groups, AcOH-mediated cyclization and TFA-induced deprotection. Diamine 245 is finally converted into Pibrentasvir via coupling with protected O -methyl-threonine 246 with EDC and HOBt as an activating agent. Voxilaprevir is sold under the brand name Vosevi, as a co-formulation with sofosbuvir (see ) and velpatasvir and was developed by Gilead. It was approved in July 2017 for the treatment of chronic hepatitis C virus (HCV) in adults . Voxilaprevir is an NS protein 3/4A protease inhibitor, possessing an 18-membered gem -difluoro methylene quinoxaline portion similar to that of Glecaprevir . Additionally, the synthetic pathways are quite similar . The synthetic method starts with the formation of fluorinated α-ketoester 247 via the lithium exchange and allylation of diethyl oxalate, starting with 3-bromo-3,3-difluoro-propene 212 . Condensation with methoxy ortho-phenylenediamine 248 produces gem -difluoro quinoxaline 249 , which is chlorinated with POCl 3 to produce derivative 250 , which possesses a good leaving group for nucleophilic aromatic substitution with Boc-protected ethyl hydroxyproline 251 . The Boc removal of 252 with HCl furnishes amine 253 , which is one of the two main building blocks for macrocycle formation. The second main component, acid 254 , is coupled to it by inducing amide bond formation and employing HATU as an activating agent. Diene 255 is then subjected to RCM using Zhan 1B catalyst, causing the formation of the desired trans macrocycle, 256. Glecaprevir is obtained in two steps by means of the hydrogenation of the double bond of 256 and t -Bu ester removal to produce acid 257 and its coupling with fluorinated aminocyclopropane 225 side-chain using HATU as an activating agent. In 2016, the FDA approved 22 new drugs, including 13 small molecules . Eleven out of thirteen molecules contain at least one heterocyclic ring, and four out of thirteen molecules have at least one fluorine atom. In the following paragraphs, two heterocyclic compounds bearing a fluorinated moiety directly linked to the ring are reported ( and ). 8.1. Rucaparib Rucaparib is sold under the brand name Rubraca and was developed by Clovis Oncology. It was approved in December 2016 for the treatment of ovarian cancer . Rucaparib is the first-in-class inhibitor of DNA repair enzyme poly-ADP ribose polymerase-1 (PARP-1); notably, the presence of fluorine at the indole ring enhances the in vitro potency tenfold in comparison to that of the unfluorinated analogue . Rucaparib synthesis is completed in five steps, starting with fluoroindole 258 . The alkylation of 258 with nitroacetate 259 yields compound 260 , which is reduced using Zinc in an acid medium and directly cyclized under basic conditions into azepino-indole 261 . The bromuration of indole C-2 produces 262 , which undergoes to a Suzuki reaction with formyl boronic acid 263 . The obtained aldehyde 264 is converted into Rucaparib by means of reductive amination with NaBH 3 CN. 8.2. Sofosbuvir Sofosbuvir is sold under the brand name Epclusa, as a co-formulation with velpatavsir , and was developed by Gilead Science. It was approved in June 2016 for the treatment of six major forms of HCV . Sofsbuvir has also been FDA-approved since 2014 for the treatment of HCV, alone or co-administered with other drugs . Sofosbuvir acts as an HCV NS5B polymerase inhibitor and is administered as a prodrug . The synthesis of Sofosbuvir was described by Bao et al. in 2010 . Sofosbuvir was originally obtained from cytidine derivative 270 via the hydrolysis of the amino group and Bz removal, followed by the installation of a phosphoramidate side-chain on 272 in the presence of N -methylimidazole (NMI) . Compound 270 has been previously synthesized, starting with chiral cyclic sulfate 265 . Sulfate opening using fluoride is followed by a hydrolytic step, allowing the obtainment of fluorinated chiral compound 266 . Acetonide hydrolysis, performed with HCl in EtOH, straightforwardly produces lactone 267 , which is protected as it is Bz ester 268 , and then reduced and coupled with pyrimidine derivative 269 to produce compound 270 . Rucaparib is sold under the brand name Rubraca and was developed by Clovis Oncology. It was approved in December 2016 for the treatment of ovarian cancer . Rucaparib is the first-in-class inhibitor of DNA repair enzyme poly-ADP ribose polymerase-1 (PARP-1); notably, the presence of fluorine at the indole ring enhances the in vitro potency tenfold in comparison to that of the unfluorinated analogue . Rucaparib synthesis is completed in five steps, starting with fluoroindole 258 . The alkylation of 258 with nitroacetate 259 yields compound 260 , which is reduced using Zinc in an acid medium and directly cyclized under basic conditions into azepino-indole 261 . The bromuration of indole C-2 produces 262 , which undergoes to a Suzuki reaction with formyl boronic acid 263 . The obtained aldehyde 264 is converted into Rucaparib by means of reductive amination with NaBH 3 CN. Sofosbuvir is sold under the brand name Epclusa, as a co-formulation with velpatavsir , and was developed by Gilead Science. It was approved in June 2016 for the treatment of six major forms of HCV . Sofsbuvir has also been FDA-approved since 2014 for the treatment of HCV, alone or co-administered with other drugs . Sofosbuvir acts as an HCV NS5B polymerase inhibitor and is administered as a prodrug . The synthesis of Sofosbuvir was described by Bao et al. in 2010 . Sofosbuvir was originally obtained from cytidine derivative 270 via the hydrolysis of the amino group and Bz removal, followed by the installation of a phosphoramidate side-chain on 272 in the presence of N -methylimidazole (NMI) . Compound 270 has been previously synthesized, starting with chiral cyclic sulfate 265 . Sulfate opening using fluoride is followed by a hydrolytic step, allowing the obtainment of fluorinated chiral compound 266 . Acetonide hydrolysis, performed with HCl in EtOH, straightforwardly produces lactone 267 , which is protected as it is Bz ester 268 , and then reduced and coupled with pyrimidine derivative 269 to produce compound 270 . The use and the importance of fluorinated heterocyclic compounds is continuously increasing in the field of medicinal chemistry. Drug discovery is often linked to the presence of a fluorinated moiety that is able to improve the drug’s potency or the metabolic stability of different heterocycles. These features allow the constant increase in these molecular entities, among other new drugs and among blockbuster and best-selling compounds. The only restriction, up to now, could be due to the limitation of fluorinated building blocks for synthetic purposes. In fact, all the reported synthetic schemes are based on available fluorinated heterocyclic compounds or their precursors, while late fluorination strategies are limited to 18 F probes due to the short half-life of this isotope. The installation of fluorinated moieties at the end of the synthesis could be a different approach that could characterize future research in this field.
Shotgun Proteomics Analysis, Functional Networks, and Peptide Biomarkers for Seafood-Originating Biogenic-Amine-Producing Bacteria	b0e6c3aa-6818-4963-a030-4857a5c66fa4	10178689	Microbiology[mh]	Biogenic amines (BAs) are low-molecular-weight nitrogenous compounds that are principally generated by the decarboxylation of free amino acids or by the deamination/amination or transamination of aldehydes and ketones . In food, BAs are created in the process of microbial, animal, and vegetable metabolism since the primary source of BAs is the decarboxylation of amino acids by fermentation, putrefaction, or decomposition . Fish, cheese, soy sauce, meat, wine, and beer are some products that often generate BAs . Histamine, cadaverine, tyramine, putrescine, spermidine, and spermine are the main BAs used as indicators for food spoilage ( ). Histamine is the most common BA responsible for food poisoning. Histamine was originally described by Dale in 1910 as a small molecule produced by the decarboxylation of histidine . Histamine induces a variety of biological processes, including the regulation of physiological functions in the gut, the stimulation of the nasal mucous membrane, and the release of gastric acids; additionally, the more serious processes it induces involve vasodilation and inflammation for triggering anaphylactic responses, which are similar to allergic responses and can be life-threatening . These BAs can be degraded by two enzymes, namely, diamine oxidase or histaminase and histamine-N-methyltransferase; some point mutations in the genes encoding these enzymes are associated with several disorders, such as ulcerative colitis and even autism. This suggests that rapid histamine removal is important to prevent harmful pathological events such as a bronchospasm, a dangerous symptom occurring in anaphylactic reactions. Although putrescine and cadaverine are also common BAs present in foods, these compounds were believed to only be toxic in large concentrations. However, in 2019, del Rio et al. carried out an in vitro study demonstrating that these two BAs display cytotoxic action (causing cell necrosis) at concentrations found in some foodstuffs (such as fish and fermented food). In humans, in addition to endogenously produced histamine and trace amines derived from commensal bacteria in the gut, BAs can be internalized through the ingestion of food. There are a variety of bacteria that synthetize and secrete histamine and other BAs as metabolic products, thus generating significant amounts of these compounds that can accumulate in foodstuffs (as a result of improper storage). In 1999, Ben-Gigirey et al. reported the isolation of both cadaverine- and histamine-producing bacteria from frozen or fresh albacore ( Thunnus alalunga ). BAs can accumulate in food via the metabolic processes of microorganisms that produce decarboxylases; these enzymes can exert their action on amino acid precursors, which is an absent process in the ‘normal’ metabolism of animals or plants. If a bioactive amine is produced in large quantities, the foodstuffs involved are prime candidates for food poisoning and could constitute a major threat to public health due to severe symptoms of intoxication. On the other hand, even low BA levels can lead to food intolerance among susceptible people, particularly those afflicted with low levels of diamine oxidase activity, which could be exacerbated by the intake of histamine-containing foods. An example is the so-called ‘scombroid food poisoning’, one of the main forms of seafood poisoning; this poisoning results from eating fish containing histamine (scombrotoxin), which is produced by contaminating bacteria. The symptoms appear soon after fish consumption and include headaches, flushed skin, itchy skin, or abdominal cramps and can last for 2 to 3 days. Depending on the geographical zone, different types of fish can be responsible for food poisoning, including bluefish, tuna, sardines, anchovies, and turbot ; these fish contain high levels of histidine, which is rapidly transformed into histamine by bacteria during storage . Fermented foods can also contain high levels of BAs, which are undoubtedly produced by contaminating microorganisms during fermentation that is improperly controlled . Hence, it is essential to identify the critical step in fermentation that results in bacterial contamination; this is particularly important in the dairy industry, as a variety of products are produced by microbial fermentation, such as cheeses ripened with bacterial or yeast starters, including lactic acid bacteria. It is very concerning that high amounts of BAs were not only detected in yogurt but also in both raw and processed milk, including pasteurized, UHT, and reconstituted powered milks . Contaminative biogenic-amine-producing bacteria usually belong to the group of ‘normal’ microbiota that inhabit animals or plants from which food originates, and these microorganisms include members of the family Enterobacteriaceae (i.e., Escherichia coli , Klebsiella spp., Hafnia alvei , Proteus spp., Salmonella spp., and Serratia spp.), the family Vibrionaceae (i.e., Vibrio alginoliticus ), and Pseudomonas or Pseudomonas -like species. Considering that these bacteria are usually present in the starting material and that most microorganisms can grow extremely fast, it is advisable to promptly commence the food preservation process and quickly and unambiguously identify the relevant microbial organisms present in foodstuffs. Takahashi et al. (2003) established a PCR-based strategy for the quick determination of histamine-producing Gram-negative bacteria, while Coton and Coton (2005) applied a similar method (multiplex PCR) for the discovery of bacterial histidine decarboxylase ( hdc ) genes present in Gram-positive bacteria ( Lactococcus , Enterococcus , and Streptococcus ), which have been described as more significant producers of BAs in fermented food [ , , ]. Real-time PCR was also utilized for the quantification of histamine in wine , cheese products , and fish . More recently, new techniques involving LC-ESI-MS/MS-based proteomics have provided a rapid approach to identifying the bacterial species comprising and the bacteriophages present in pathogenic bacteria [ , , , ]. This approach is also valid for studying the different antibiotic resistance mechanisms displayed by bacteria, such as the strategies used by pathogenic streptococcal species and Listeria monocytogenes . Another advantage of this novel method is that its corresponding analyses can be directly obtained from foodstuffs, as they do not require bacterial enrichment; hence, the microorganisms being studied do not have to be cultivated in a laboratory. There are currently a variety of techniques that can be applied to quantitate the levels of biogenic amines secreted by actively growing BA-producing bacteria, including HPLC-based methods or classic microbiological procedures such as the approach taken by Tao et al. (2009) , which involves bacterial growth in differential agar media. In this manuscript, the most relevant BAs in seafoods (fish) are addressed, and the relevance of these molecules for food quality and safety are reported. Fish is an extremely perishable food product and contains a vulnerable matrix that can include high levels of BAs . In this work, we used a shotgun proteomic technique to quickly and easily characterize 15 different foodborne strains of biogenic-amine-producing bacteria for the first time. The proteome repository was then subjected to some functional bioinformatics examinations, such as (i) functional pathway, gene ontology (GO), and hierarchical clustering analyses; (ii) protein network analysis; (iii) the identification of virulence factors; and (iv) the selection of putative species-specific peptide biomarkers for the distinction of foodborne biogenic-amine-producing bacteria. 2.1. Shotgun Proteomics Data Repository Fifteen different seafood-based biogenic-amine-producing bacteria were analyzed in this study ( ). Bacterial peptides were obtained via the trypsin digestion of protein mixtures and a subsequent analysis using LC-ESI-MS/MS, as presented previously [ , , , ]. A total of 10,673 peptide spectrum matches (PSMs) belonging to 4081 nonredundant peptides were determined, which belonged to 1811 annotated proteins from the Proteobacteria UniProt/TrEMBL database (August 2022) ( ). The MS/MS proteomics data were deposited in the ProteomeXchange Consortium via the PRIDE storage website with the dataset identifier PXD039320. To the best of our knowledge, the current data constitute the largest dataset of proteins and peptides of seafood-based biogenic-amine-producing bacteria identified to date. This valuable protein repository will add novel and important content to public protein databases and will hopefully be useful for novel research. 2.2. Label-Free Quantification (LFQ) of Biogenic-Amine-Producing Bacteria and Hierarchical Clustering The relative label-free quantification of each type of bacteria was also executed to define the level of protein abundance in each sample. contains these results. Comparisons of the high-abundance proteins of each species and strain were performed. a displays the distribution of the high-abundance proteins determined for each of the 15 strains. Among them, Proteus vulgaris , Stenotrophomonas maltophilia , and Morganella morganii were the three main species with the most high-abundance proteins. The distribution of the high-abundance proteins for all samples analyzed via LFQ is illustrated in a heatmap diagram in b. Euclidean hierarchical distance was used to differentiate three main clusters. Cluster A (strains H6, H2, H9, and H14: Morganella morganii , Enterobacter cloacae , Proteus vulgaris , and Stenotrophomonas maltophilia ), Cluster B (strains H12, H3, and H8: Raoutella planticola , Hafnia alvei , and Proteus penneri ), and Cluster C (strains H1, H4, and H7: Enterobacter aerogenes , Klebsiella oxytoca , and Proteus mirabilis ). As in a, the clusters of b were divided according to the number of proteins that were more upregulated (Red) (as determined via LFQ) versus those proteins that were more downregulated (Green) for the different strains. Regarding the different genera, a shows the high-abundance proteins for each genus ( Enterobacter spp., Hafnia spp., Klebsiella spp., Morganella spp., Proteus spp., Raoultella spp., and Stenotrophomonas spp.). Among them, Proteus spp. was the most represented genus with the most high-abundance proteins. The distribution of the high-abundance proteins for all samples grouped by genus and analyzed via LFQ is illustrated in a heatmap diagram in b. Finally, all strains were arranged according to Euclidean hierarchical distance. Seven principal clusters were differentiated, which corresponded to the different genus types. To obtain further insights regarding functional interpretation, the present repository was investigated using several functional in silico analyses, comprising (i) functional pathways, GO enrichment and hierarchical clustering, (ii) functional network analysis, (iii) the discovery of virulence factors, and (iv) the selection of potential species-specific peptide biomarkers. 2.3. Functional Pathways and Gene Ontology (GO) The global protein repository of foodborne strains of biogenic-amine-producing bacteria was individually examined using functional bioinformatics tools, such as functional pathway analysis and GO term enrichment. PANTHER analysis was performed using gene names (considering all nonredundant proteins), revealing the presence of 10 different molecular functions ( a), 12 different biological processes ( b), and 20 different protein classes ( c) in the complete global proteome repository. According to the molecular function classification procedure ( a), the most important molecular functions were binding (35.6%), structural molecule activity (33.1%), and catalytic activity (22.8%). Within the binding function group, ribosomal proteins, oxidorreductases, chaperones, DNA metabolism proteins, deaminases, isomerases, transferases, translation elongation factor proteins, mutases, and protein kinases were found. In the structural molecule activity group, ribosomal proteins and tubulins were detected. Regarding catalytic activity, decarboxylases, nucleotide kinases, oxidases, kinases, pyrophosphatases, isomerases, transferases, deaminases, proteases, dehydrogenases, and mutases were observed. According to the classification of biological processes ( b), the most remarkable categories were cellular processes (44.9%), metabolic processes (33.8%), biological regulation (8.3%), localization (5.9%), response to stimulus (2.4%), and signaling (0.8%). Regarding cellular processes, ribosomal proteins, decarboxylases, pyrophosphatases, vesicle coat proteins, isomerases, transferases, and translocation initiation factors were found. Concerning the metabolic process group, ribosomal proteins, decarboxylases, pyrophosphatases, translation release factor, chaperone, isomerases, transferases, and metalloproteases were detected. In the biological regulation group, ribosomal protein, chaperone, membrane traffic protein, primary active transporter, and storage proteins were observed. According to the classification of protein classes ( c), the most prominent classes were translational proteins (51.1%), metabolite interconversion enzymes (18.6%), and transporters (6.8%). Within the translational protein group, ribosomal protein and translation initiation/elongation/release factors were observed. In the metabolite interconversion enzyme group, different enzyme groups were observed, including dehydrogenases, carbohydrate kinases, aldolases, isomerases, hydrolases, glycosidases, transferases, oxidases, glucosidases, peroxidases, mutases, dehydratases, phospholiases, isomerases, and deaminases. Within the transporter category, ATP synthase, ATP-binding cassettes, amino acid transporters, and ion channels were detected. The existence of high concentrations of decarboxylases in these functional classifications influences the formation of biogenic amines by the bacteria. During the deterioration of fish, the occurrence of bacterial strains with high proteolytic enzyme activity increases the breakdown of proteins as well as the accessibility of small peptides and specific free amino acids that are decarboxylated in particular biogenic amines . In fish, the principal studied biogenic amines include histamine (derived from histidine), putrescine (derived from arginine, glutamine, methionine, and ornithine), cadaverine (derived from lysine), tyramine (derived from tyrosine), spermidine (derived from agmatine, methionine, putrescine, and spermine), and spermine (derived from agmatine, methionine, putrescine, and spermidine) . 2.4. Biogenic Amine-Related Proteins and Peptides Detected via LC-ESI-MS/MS summarizes the list of biogenic amine-related proteins and peptides detected via LC-ESI-MS/MS for the corresponding strains. Agmatine and cadaverine are aliphatic polyamine biogenic amines derived from the amino acids arginine and lysine, respectively . Two different related proteins (arginine ABC transporter substrate-binding protein and lysine–arginine–ornithine-binding periplasmic protein) and three different peptides (IDAVFGDTAVVTEWLK, CTWVGSDFDSLIPSLK, and IGTDATYAPFSSK) were detected via shotgun proteomics in the K. oxytoca strain. The metabolism of agmatine and cadaverine requires the initial presence and transport of arginine or lysine, respectively, in the periplasm of the cells. In Gram-negative bacteria, solute-binding proteins are localized in the periplasm and involved in nitrogen compound transport (GO:0071705) and amine transport (GO:0015837). Histamine is a heterocyclic biogenic amine derived from the amino acid histidine . Histamine is present in most foods but is more abundant in fish and fishery products. This biogenic amine is the major agent behind “scombroid poisoning” or “histamine poisoning” . A total of five different related proteins (histidine kinase, histidine phosphatase, histidine-binding periplasmic protein, histidine triad nucleotide-binding protein, and histidine ammonia-lyase) were detected via shotgun proteomics ( ). Nine peptides of histidine kinase (GO:0004673) were identified via LC-ESI-MS/MS analysis in different strains (IDSEDLPHVRASVAR (present in M. morganii ), LAMNLRTRLFLSISALITVALLGLLLGLVSVMQMAGSQ-EILIR ( S. maltophilia ), MIAEAANADSKQAQR ( K. oxytoca and R. planticola ), TIDQINQQKIQLEQEIADRK ( P. penneri and P. vulgaris ), GEADATLDSEVSAWRAVAR ( P. vulgaris ), LSSELWNCKIDPTQAEMAMINILANAR ( P. mirabilis ), SEASENTVDLIVEDEGSGIPK ( P. mirabilis ), NEEARDNLISELTAR ( P. vulgaris ), and RYAYSEQLGDLLQR ( S. maltophilia )). In addition, a peptide from histidine phosphatase (GO:0101006) was detected via LC-MS/MS (HAQASEYGSALFVAVGQAKQVK) in the H. alvei strain. It is well known that histidine kinase/phosphatase regulates histamine synthesis and signal transduction by activating histidine decarboxylase through phosphorylation/dephosphorylation . Moreover, a peptide of histidine-binding periplasmic protein (IGVLQGTTQETYGNEHWAPK) was detected in the K. oxytoca strain, and two peptides of histidine triad nucleotide-binding protein (EIPSDIVYQDELVTAFR, IAEQEGIAEDGYR) were detected in the E. cloacae strain. These proteins are involved in nitrogen compound transport and amine transport (GO:0071705, GO:0015837). Finally, a peptide (LAAMQQALGAQIAAVEEDR) of histidine ammonia-lyase was identified via LC-ESI-MS/MS in the M. morganii strain. This cytosolic enzyme catalyzes the first reaction in histidine catabolism: the nonoxidative deamination of histidine to trans-urocanic acid (GO:0004397). Putrescine is an aliphatic biogenic amine derived from the amino acids arginine or ornithine in one step or two steps after glutamine or methionine is transformed into ornithine and then putrescine. The ingestion of food containing high amounts of putrescine can lead to grave toxicological consequences. In fact, putrescine can react with nitrite to form N-nitrosamines, which are carcinogenic agents . Additionally, putrescine induces significant effects that enhance the toxicological effects of other BAs, particularly histamine and tyramine . In seafood such as fish, squid, and octopus, putrescines are also dominant biogenic amines . Lysine–arginine–ornithine-binding periplasmic protein and two peptides (CTWVGSDFDSLIPSLK; IGTDATYAPFSSK) were also identified via LC-ESI-MS/MS analysis in the K. oxytoca strain. The metabolism of putrescine also requires the initial presence and transport of arginine or ornithine in the periplasm of the cells (GO:0071705 and GO:0015837). In addition, three proteins responsible for glutamine and methionine transport and amino/amido transferase were identified via shotgun proteomics in K. oxytoca and H. alvei strains. These corresponded to glutamine ABC transporter periplasmic protein (peptide: AVGDSIEAQQYGIAFPK) present in K. oxytoca , glutamine-fructose-6-phosphate aminotransferase (IDAAQEAELIKALFEAPR) present in K. oxytoca , N-acetylglutaminylglutamine amidotransferase (SGANAAVDKALRLDSTVMLVDDPVK) present in H. alvei , type 1 glutamine amidotransferase domain-containing protein (IFRTLALMLLVTSATAFAASK) present in P. mirabilis , and L-glutamine-binding protein (ADAVIHDTPNILYFIK, AVGDSLEAQQYGIAFPK) present in K. oxytoca and H. alvei strains. Finally, the S-adenosylmethionine decarboxylase proenzyme (AdoMetDC) (ALSFNIYDVCYAR) was detected in the S. maltophilia strain. It is involved in the synthesis of biogenic amines in several species that use aminopropyltransferases for this pathway. AdoMetDC is involved in the production of S-adenosyl-1-(methylthio)-3-propylamine (decarboxylated S-adenosylmethionine) . In contrast to many amino acid decarboxylases that use pyridoxal 5′-phosphate as a cofactor, AdoMetDC uses a covalently bound pyruvate residue. This decarboxylase is involved in the polyamine biosynthetic pathway, as it generates the n-propylamine residue needed for the synthesis of spermidine and spermine from putrescine . Spermidine is an aliphatic polyamine derived from putrescine, agmantine, methionine, or spermine . It is a precursor to other polyamines, such as spermine and its structural isomer thermospermine. Spermidine in fish tissue can potentiate the toxic effect of histamine by inhibiting intestinal histamine-catabolic enzymes . Two spermidine-related proteins were identified via LC-MS/MS (spermidine/putrescine import ATP-binding protein and S-adenosylmethionine decarboxylase proenzyme). One peptide (VDEVHDNAEAEGLIGYIR) of spermidine/putrescine import ATP-binding protein was detected via LC-ESI-MS/MS in the P. vulgaris strain. This protein is part of the ABC transporter complex PotABCD that is involved in spermidine/putrescine import . In addition, one peptide (ALSFNIYDVCYAR) of the S-adenosylmethionine decarboxylase proenzyme was detected in the S. maltophilia strain. This enzyme is necessary for the biosynthesis of polyamines such as spermine and spermidine from the diamine putrescine . Spermine is an aliphatic polyamine derived from agmatine, methionine, or spermidine . One spermine-related protein was identified via LC-ESI-MS/MS (S-adenosylmethionine decarboxylase proenzyme). A peptide (ALSFNIYDVCYAR) of the S-adenosylmethionine decarboxylase proenzyme was detected in the S. maltophilia strain. In addition, spermine has been reported to modify the connections between polyamines and DNA. In fact, spermine has been reported to function as a free radical scavenger protecting DNA from oxidative stress . More precisely, the higher the cationic charge, the higher the degree of DNA-protein binding enhancement; thus, spermine has been characterized as more potent than spermidine and putrescine. Finally, further decarboxylases (e.g., phosphatidylserine decarboxylase and 4-carboxymuconolactone decarboxylase) and deaminases (e.g., 2-iminobutanoate/2-iminoopropanoate deaminase and glucosamine-6-phosphate deaminase) were identified via shotgun proteomics ( ), but according to the literature, they are involved in other metabolic pathways, which was also demonstrated previously via PANTHER analysis. 2.5. Network Analysis Network analysis was executed using STRING v.11.5 software ( https://string-db.org/ , accessed on 6 December 2022) , wherein all the proteins identified in this study were investigated and compared with the genome of the model organism E. coli K12 MG1655, which was the genetically closest group available in the portal ( ). Every protein—protein interaction was assigned to the network in accordance with its confidence score. To reduce the occurrence of false positives and false negatives, all expected interactions were tagged as “high-confidence” (≥0.7) in the STRING program were selected for this work. Thus, the final network for the global protein repository consisted of 260 nodes (proteins) and 1973 edges (interactions) ( ). All proteins used in the network were discovered during the proteomic experiments (see the codes of the gene column in ). This protein network is the first inclusive interactomics map for relevant seafood-based, biogenic-amine foodborne strains. Cluster networks were generated using an MCL (inflation clustering) algorithm from the STRING website, and a default value of 2 was selected for all analyses. From the cluster analysis, 42 significant clusters of interactions between nodes were obtained. highlights the most relevant clusters ( n = 15) according to the abundance of nodes involved or their biological relevance. includes information about the 42 clusters, protein names, and descriptions of the corresponding name codes. The most relevant subnetworks in terms of their number of nodes are involved in ribosomal metabolism (in red; 63 nodes), host-virus interaction/porin activity (in green; 22 nodes), transmembrane transport (in violet; 12 nodes), and glycolysis (in dark violet; 8 nodes). Other subnetworks that contain fewer nodes but have great biological importance are related to bacterial flagellum biogenesis (in red; four nodes), vancomycin (an antibiotic) resistance (in blue; three nodes), and putrescine metabolism (in pink; three nodes). Further study of the aforementioned subnetworks and protein-protein interactions will be very beneficial for the development of new therapeutic treatments for bacterial dispersion, antibiotic resistance, and food intoxication via biogenic-amine-produced putrescine. 2.6. Virulence Factors Many seafood-originating biogenic-amine-producing bacteria are pathogens with well-known virulence. It has been reported that Enterobacter bacteria are increasingly exhibiting a multidrug resistance phenotype . Moreover, K. oxytoca can acquire antimicrobial resistance and carry multiple virulence genes, such as capsular polysaccharides and fimbriae . The virulence of other species analyzed in this study, such as H. alvei , M. morganii , P. vulgaris , S. maltophilia , and R. planticola , has been previously reported. A total of 556 peptides belonging to virulence factors (nonredundant peptides) were identified in this study. They included toxins, polypeptides involved in antibiotic resistance, and proteins related to cell colonization and immune evasion. The 556 virulent peptides ( ) are displayed in groups in accordance with the principal roles in which they are involved (e.g., toxin generation/transport, colonization and immune evasion factors, antimicrobial compounds, other tolerance proteins that play a role in resistance to toxic substances, etc.). In addition, the main proteins of the identified virulence factors are displayed in . In this study, several peptides involved in antimicrobial resistance or the production/transport of toxic substances were identified ( ). Ten of the proteins characterized were associated with antibiotic resistance, and 91 peptides were related to other tolerances. Four peptides were identified as penicillin-binding proteins. Peptides associated with acriflavine and methicillin resistance and belonging to the TetR family of regulators (TFRs) were determined. TetR proteins regulate antibiotic and quorum-sensing processes as well as antibiotic resistance. In addition, two peptides of the GCN-2-related N-acetyl transferase (GNAT) family of acetyltransferases, which provide antibiotic resistance , were also identified. Peptides of proteins involved in other bacterial tolerances (e.g., thermotolerance and osmotolerance) were also identified. Accordingly, this work has identified many peptides that belong to groups of peptides of bacterial general stress response proteins, heat shock proteins, and cold-shock-like proteins (CSPs), among others . A total of 20 peptides corresponding to proteins that are involved in bacterial toxicity were identified. These peptides include ecotin, lipoprotein toxin enterocidin B, antitoxin ParD, and addition module toxin GnSA/GnsB. As an example of some of the roles these peptides play, ecotin is an inhibitor of multiple complement-dependent processes found in bacteria . In this study, 349 peptides involved in colonization and immune evasion were identified. Bacterial internalization into the host is facilitated by these proteins, resulting in subsequent infection and propagation. Transcriptional regulators involved in the control of virulence factors were also found for the analyzed strains, including two peptides identified as LysR and SlyA . LysR regulates virulence factors, such as extracellular polysaccharides, toxins, and bacteriocins. Fimbria are located on the surfaces of bacteria; they are involved in adherence to target cells and biofilm formation . Lysis proteins belonging to the LysM domain were identified; this domain was identified in enzymes involved in bacterial cell wall degradation . Additionally, several peptides of peptidases and proteases were identified. This includes members of the Lon protease family and subtilisin, among others. Different peptides of the Superoxide dismutase enzyme (SOD) were identified. SOD is a metalloenzyme that defends against reactive oxygen species produced by neutrophils and macrophages . The presence of open channels facilitates passive penetration though the outer membrane. We have identified several porins or outer-membrane proteins (OMPs), such as the porins OmpA, OmpX, and OmpC; substrate-specific porins, such as maltoporin, which is also called LamB; and TonB-dependent receptors, such as FhuA . In addition, many peptides were determined to be other virulence factors, such as VacJ family lipoprotein VacJ (virulence-associated chromosome locus J); the chaperone protein Skp, which assists in the folding and insertion of many OMPs ; and the Osmy chaperone. In this study, four peptides of antibacterial proteins were identified, including one peptide that belongs to a bacteriocin and the remaining three to a colicin-like protein. Colicins are antimicrobial proteins typically produced by E. coli that degrade internal cellular elements . ABC transporters, like many other bacterial transporters, are involved in resistance or tolerance and bacterial propagation during infection . We identified different ABC transporters related to virulence ( ). Furthermore, sixteen peptides of alternative virulent factors were identified, such as proteins related to mobile genetic elements’ transposases, recombinases, plasmids, and viral DNA fragments, which are considered the major mechanism for acquiring antibiotic resistance. Moreover, pilus conforms to a typical method of horizontal transfer between bacteria, which is another mechanism of obtaining virulence determinants . We identified 43 peptides of phage proteins, such as bacteriophage CI repressor and capsid scaffolding protein, but mainly phage shock proteins. Finally, we identified bacterial proteins determined in the UniProt database in different phage strains ( Klebsiella phage vB_KpM_FBKp24, Klebsiella phage vB_KppS-Storm, Stenotrophomonas phage BUCT608, and Stenotrophomonas phage). 2.7. Potential Species-Specific Peptide Biomarkers To select potential peptide biomarkers for the 15 different biogenic-amine-producing bacterial strains, we implemented a massive comparison of the proteomics data with respect to the proteins and peptides included in databases. The suitable peptides that were identified via LC-ESI-MS/MS in only one specific species were verified in terms of their specificity and sequence homology using the BLASTp algorithm ( ). summarizes the analysis of the 77 species-specific tryptic peptide biomarkers belonging to 64 different proteins that were suggested for the unequivocal identification of the different seafood-originating biogenic-amine-producing bacteria of 10 different species ( E. aerogenes , E. cloacae , H. alvei , K. oxytoca , M. morganii , P. mirabilis , P. penneri , P. vulgaris , R. planticola , and S. maltophilia ). All the peptides included herein have been proposed as potential biomarkers for the first time and will be very convenient for further studies using targeted proteomics approaches to identify the different seafood-originating biogenic-amine-producing bacteria in foodstuffs. Fifteen different seafood-based biogenic-amine-producing bacteria were analyzed in this study ( ). Bacterial peptides were obtained via the trypsin digestion of protein mixtures and a subsequent analysis using LC-ESI-MS/MS, as presented previously [ , , , ]. A total of 10,673 peptide spectrum matches (PSMs) belonging to 4081 nonredundant peptides were determined, which belonged to 1811 annotated proteins from the Proteobacteria UniProt/TrEMBL database (August 2022) ( ). The MS/MS proteomics data were deposited in the ProteomeXchange Consortium via the PRIDE storage website with the dataset identifier PXD039320. To the best of our knowledge, the current data constitute the largest dataset of proteins and peptides of seafood-based biogenic-amine-producing bacteria identified to date. This valuable protein repository will add novel and important content to public protein databases and will hopefully be useful for novel research. The relative label-free quantification of each type of bacteria was also executed to define the level of protein abundance in each sample. contains these results. Comparisons of the high-abundance proteins of each species and strain were performed. a displays the distribution of the high-abundance proteins determined for each of the 15 strains. Among them, Proteus vulgaris , Stenotrophomonas maltophilia , and Morganella morganii were the three main species with the most high-abundance proteins. The distribution of the high-abundance proteins for all samples analyzed via LFQ is illustrated in a heatmap diagram in b. Euclidean hierarchical distance was used to differentiate three main clusters. Cluster A (strains H6, H2, H9, and H14: Morganella morganii , Enterobacter cloacae , Proteus vulgaris , and Stenotrophomonas maltophilia ), Cluster B (strains H12, H3, and H8: Raoutella planticola , Hafnia alvei , and Proteus penneri ), and Cluster C (strains H1, H4, and H7: Enterobacter aerogenes , Klebsiella oxytoca , and Proteus mirabilis ). As in a, the clusters of b were divided according to the number of proteins that were more upregulated (Red) (as determined via LFQ) versus those proteins that were more downregulated (Green) for the different strains. Regarding the different genera, a shows the high-abundance proteins for each genus ( Enterobacter spp., Hafnia spp., Klebsiella spp., Morganella spp., Proteus spp., Raoultella spp., and Stenotrophomonas spp.). Among them, Proteus spp. was the most represented genus with the most high-abundance proteins. The distribution of the high-abundance proteins for all samples grouped by genus and analyzed via LFQ is illustrated in a heatmap diagram in b. Finally, all strains were arranged according to Euclidean hierarchical distance. Seven principal clusters were differentiated, which corresponded to the different genus types. To obtain further insights regarding functional interpretation, the present repository was investigated using several functional in silico analyses, comprising (i) functional pathways, GO enrichment and hierarchical clustering, (ii) functional network analysis, (iii) the discovery of virulence factors, and (iv) the selection of potential species-specific peptide biomarkers. The global protein repository of foodborne strains of biogenic-amine-producing bacteria was individually examined using functional bioinformatics tools, such as functional pathway analysis and GO term enrichment. PANTHER analysis was performed using gene names (considering all nonredundant proteins), revealing the presence of 10 different molecular functions ( a), 12 different biological processes ( b), and 20 different protein classes ( c) in the complete global proteome repository. According to the molecular function classification procedure ( a), the most important molecular functions were binding (35.6%), structural molecule activity (33.1%), and catalytic activity (22.8%). Within the binding function group, ribosomal proteins, oxidorreductases, chaperones, DNA metabolism proteins, deaminases, isomerases, transferases, translation elongation factor proteins, mutases, and protein kinases were found. In the structural molecule activity group, ribosomal proteins and tubulins were detected. Regarding catalytic activity, decarboxylases, nucleotide kinases, oxidases, kinases, pyrophosphatases, isomerases, transferases, deaminases, proteases, dehydrogenases, and mutases were observed. According to the classification of biological processes ( b), the most remarkable categories were cellular processes (44.9%), metabolic processes (33.8%), biological regulation (8.3%), localization (5.9%), response to stimulus (2.4%), and signaling (0.8%). Regarding cellular processes, ribosomal proteins, decarboxylases, pyrophosphatases, vesicle coat proteins, isomerases, transferases, and translocation initiation factors were found. Concerning the metabolic process group, ribosomal proteins, decarboxylases, pyrophosphatases, translation release factor, chaperone, isomerases, transferases, and metalloproteases were detected. In the biological regulation group, ribosomal protein, chaperone, membrane traffic protein, primary active transporter, and storage proteins were observed. According to the classification of protein classes ( c), the most prominent classes were translational proteins (51.1%), metabolite interconversion enzymes (18.6%), and transporters (6.8%). Within the translational protein group, ribosomal protein and translation initiation/elongation/release factors were observed. In the metabolite interconversion enzyme group, different enzyme groups were observed, including dehydrogenases, carbohydrate kinases, aldolases, isomerases, hydrolases, glycosidases, transferases, oxidases, glucosidases, peroxidases, mutases, dehydratases, phospholiases, isomerases, and deaminases. Within the transporter category, ATP synthase, ATP-binding cassettes, amino acid transporters, and ion channels were detected. The existence of high concentrations of decarboxylases in these functional classifications influences the formation of biogenic amines by the bacteria. During the deterioration of fish, the occurrence of bacterial strains with high proteolytic enzyme activity increases the breakdown of proteins as well as the accessibility of small peptides and specific free amino acids that are decarboxylated in particular biogenic amines . In fish, the principal studied biogenic amines include histamine (derived from histidine), putrescine (derived from arginine, glutamine, methionine, and ornithine), cadaverine (derived from lysine), tyramine (derived from tyrosine), spermidine (derived from agmatine, methionine, putrescine, and spermine), and spermine (derived from agmatine, methionine, putrescine, and spermidine) . summarizes the list of biogenic amine-related proteins and peptides detected via LC-ESI-MS/MS for the corresponding strains. Agmatine and cadaverine are aliphatic polyamine biogenic amines derived from the amino acids arginine and lysine, respectively . Two different related proteins (arginine ABC transporter substrate-binding protein and lysine–arginine–ornithine-binding periplasmic protein) and three different peptides (IDAVFGDTAVVTEWLK, CTWVGSDFDSLIPSLK, and IGTDATYAPFSSK) were detected via shotgun proteomics in the K. oxytoca strain. The metabolism of agmatine and cadaverine requires the initial presence and transport of arginine or lysine, respectively, in the periplasm of the cells. In Gram-negative bacteria, solute-binding proteins are localized in the periplasm and involved in nitrogen compound transport (GO:0071705) and amine transport (GO:0015837). Histamine is a heterocyclic biogenic amine derived from the amino acid histidine . Histamine is present in most foods but is more abundant in fish and fishery products. This biogenic amine is the major agent behind “scombroid poisoning” or “histamine poisoning” . A total of five different related proteins (histidine kinase, histidine phosphatase, histidine-binding periplasmic protein, histidine triad nucleotide-binding protein, and histidine ammonia-lyase) were detected via shotgun proteomics ( ). Nine peptides of histidine kinase (GO:0004673) were identified via LC-ESI-MS/MS analysis in different strains (IDSEDLPHVRASVAR (present in M. morganii ), LAMNLRTRLFLSISALITVALLGLLLGLVSVMQMAGSQ-EILIR ( S. maltophilia ), MIAEAANADSKQAQR ( K. oxytoca and R. planticola ), TIDQINQQKIQLEQEIADRK ( P. penneri and P. vulgaris ), GEADATLDSEVSAWRAVAR ( P. vulgaris ), LSSELWNCKIDPTQAEMAMINILANAR ( P. mirabilis ), SEASENTVDLIVEDEGSGIPK ( P. mirabilis ), NEEARDNLISELTAR ( P. vulgaris ), and RYAYSEQLGDLLQR ( S. maltophilia )). In addition, a peptide from histidine phosphatase (GO:0101006) was detected via LC-MS/MS (HAQASEYGSALFVAVGQAKQVK) in the H. alvei strain. It is well known that histidine kinase/phosphatase regulates histamine synthesis and signal transduction by activating histidine decarboxylase through phosphorylation/dephosphorylation . Moreover, a peptide of histidine-binding periplasmic protein (IGVLQGTTQETYGNEHWAPK) was detected in the K. oxytoca strain, and two peptides of histidine triad nucleotide-binding protein (EIPSDIVYQDELVTAFR, IAEQEGIAEDGYR) were detected in the E. cloacae strain. These proteins are involved in nitrogen compound transport and amine transport (GO:0071705, GO:0015837). Finally, a peptide (LAAMQQALGAQIAAVEEDR) of histidine ammonia-lyase was identified via LC-ESI-MS/MS in the M. morganii strain. This cytosolic enzyme catalyzes the first reaction in histidine catabolism: the nonoxidative deamination of histidine to trans-urocanic acid (GO:0004397). Putrescine is an aliphatic biogenic amine derived from the amino acids arginine or ornithine in one step or two steps after glutamine or methionine is transformed into ornithine and then putrescine. The ingestion of food containing high amounts of putrescine can lead to grave toxicological consequences. In fact, putrescine can react with nitrite to form N-nitrosamines, which are carcinogenic agents . Additionally, putrescine induces significant effects that enhance the toxicological effects of other BAs, particularly histamine and tyramine . In seafood such as fish, squid, and octopus, putrescines are also dominant biogenic amines . Lysine–arginine–ornithine-binding periplasmic protein and two peptides (CTWVGSDFDSLIPSLK; IGTDATYAPFSSK) were also identified via LC-ESI-MS/MS analysis in the K. oxytoca strain. The metabolism of putrescine also requires the initial presence and transport of arginine or ornithine in the periplasm of the cells (GO:0071705 and GO:0015837). In addition, three proteins responsible for glutamine and methionine transport and amino/amido transferase were identified via shotgun proteomics in K. oxytoca and H. alvei strains. These corresponded to glutamine ABC transporter periplasmic protein (peptide: AVGDSIEAQQYGIAFPK) present in K. oxytoca , glutamine-fructose-6-phosphate aminotransferase (IDAAQEAELIKALFEAPR) present in K. oxytoca , N-acetylglutaminylglutamine amidotransferase (SGANAAVDKALRLDSTVMLVDDPVK) present in H. alvei , type 1 glutamine amidotransferase domain-containing protein (IFRTLALMLLVTSATAFAASK) present in P. mirabilis , and L-glutamine-binding protein (ADAVIHDTPNILYFIK, AVGDSLEAQQYGIAFPK) present in K. oxytoca and H. alvei strains. Finally, the S-adenosylmethionine decarboxylase proenzyme (AdoMetDC) (ALSFNIYDVCYAR) was detected in the S. maltophilia strain. It is involved in the synthesis of biogenic amines in several species that use aminopropyltransferases for this pathway. AdoMetDC is involved in the production of S-adenosyl-1-(methylthio)-3-propylamine (decarboxylated S-adenosylmethionine) . In contrast to many amino acid decarboxylases that use pyridoxal 5′-phosphate as a cofactor, AdoMetDC uses a covalently bound pyruvate residue. This decarboxylase is involved in the polyamine biosynthetic pathway, as it generates the n-propylamine residue needed for the synthesis of spermidine and spermine from putrescine . Spermidine is an aliphatic polyamine derived from putrescine, agmantine, methionine, or spermine . It is a precursor to other polyamines, such as spermine and its structural isomer thermospermine. Spermidine in fish tissue can potentiate the toxic effect of histamine by inhibiting intestinal histamine-catabolic enzymes . Two spermidine-related proteins were identified via LC-MS/MS (spermidine/putrescine import ATP-binding protein and S-adenosylmethionine decarboxylase proenzyme). One peptide (VDEVHDNAEAEGLIGYIR) of spermidine/putrescine import ATP-binding protein was detected via LC-ESI-MS/MS in the P. vulgaris strain. This protein is part of the ABC transporter complex PotABCD that is involved in spermidine/putrescine import . In addition, one peptide (ALSFNIYDVCYAR) of the S-adenosylmethionine decarboxylase proenzyme was detected in the S. maltophilia strain. This enzyme is necessary for the biosynthesis of polyamines such as spermine and spermidine from the diamine putrescine . Spermine is an aliphatic polyamine derived from agmatine, methionine, or spermidine . One spermine-related protein was identified via LC-ESI-MS/MS (S-adenosylmethionine decarboxylase proenzyme). A peptide (ALSFNIYDVCYAR) of the S-adenosylmethionine decarboxylase proenzyme was detected in the S. maltophilia strain. In addition, spermine has been reported to modify the connections between polyamines and DNA. In fact, spermine has been reported to function as a free radical scavenger protecting DNA from oxidative stress . More precisely, the higher the cationic charge, the higher the degree of DNA-protein binding enhancement; thus, spermine has been characterized as more potent than spermidine and putrescine. Finally, further decarboxylases (e.g., phosphatidylserine decarboxylase and 4-carboxymuconolactone decarboxylase) and deaminases (e.g., 2-iminobutanoate/2-iminoopropanoate deaminase and glucosamine-6-phosphate deaminase) were identified via shotgun proteomics ( ), but according to the literature, they are involved in other metabolic pathways, which was also demonstrated previously via PANTHER analysis. Network analysis was executed using STRING v.11.5 software ( https://string-db.org/ , accessed on 6 December 2022) , wherein all the proteins identified in this study were investigated and compared with the genome of the model organism E. coli K12 MG1655, which was the genetically closest group available in the portal ( ). Every protein—protein interaction was assigned to the network in accordance with its confidence score. To reduce the occurrence of false positives and false negatives, all expected interactions were tagged as “high-confidence” (≥0.7) in the STRING program were selected for this work. Thus, the final network for the global protein repository consisted of 260 nodes (proteins) and 1973 edges (interactions) ( ). All proteins used in the network were discovered during the proteomic experiments (see the codes of the gene column in ). This protein network is the first inclusive interactomics map for relevant seafood-based, biogenic-amine foodborne strains. Cluster networks were generated using an MCL (inflation clustering) algorithm from the STRING website, and a default value of 2 was selected for all analyses. From the cluster analysis, 42 significant clusters of interactions between nodes were obtained. highlights the most relevant clusters ( n = 15) according to the abundance of nodes involved or their biological relevance. includes information about the 42 clusters, protein names, and descriptions of the corresponding name codes. The most relevant subnetworks in terms of their number of nodes are involved in ribosomal metabolism (in red; 63 nodes), host-virus interaction/porin activity (in green; 22 nodes), transmembrane transport (in violet; 12 nodes), and glycolysis (in dark violet; 8 nodes). Other subnetworks that contain fewer nodes but have great biological importance are related to bacterial flagellum biogenesis (in red; four nodes), vancomycin (an antibiotic) resistance (in blue; three nodes), and putrescine metabolism (in pink; three nodes). Further study of the aforementioned subnetworks and protein-protein interactions will be very beneficial for the development of new therapeutic treatments for bacterial dispersion, antibiotic resistance, and food intoxication via biogenic-amine-produced putrescine. Many seafood-originating biogenic-amine-producing bacteria are pathogens with well-known virulence. It has been reported that Enterobacter bacteria are increasingly exhibiting a multidrug resistance phenotype . Moreover, K. oxytoca can acquire antimicrobial resistance and carry multiple virulence genes, such as capsular polysaccharides and fimbriae . The virulence of other species analyzed in this study, such as H. alvei , M. morganii , P. vulgaris , S. maltophilia , and R. planticola , has been previously reported. A total of 556 peptides belonging to virulence factors (nonredundant peptides) were identified in this study. They included toxins, polypeptides involved in antibiotic resistance, and proteins related to cell colonization and immune evasion. The 556 virulent peptides ( ) are displayed in groups in accordance with the principal roles in which they are involved (e.g., toxin generation/transport, colonization and immune evasion factors, antimicrobial compounds, other tolerance proteins that play a role in resistance to toxic substances, etc.). In addition, the main proteins of the identified virulence factors are displayed in . In this study, several peptides involved in antimicrobial resistance or the production/transport of toxic substances were identified ( ). Ten of the proteins characterized were associated with antibiotic resistance, and 91 peptides were related to other tolerances. Four peptides were identified as penicillin-binding proteins. Peptides associated with acriflavine and methicillin resistance and belonging to the TetR family of regulators (TFRs) were determined. TetR proteins regulate antibiotic and quorum-sensing processes as well as antibiotic resistance. In addition, two peptides of the GCN-2-related N-acetyl transferase (GNAT) family of acetyltransferases, which provide antibiotic resistance , were also identified. Peptides of proteins involved in other bacterial tolerances (e.g., thermotolerance and osmotolerance) were also identified. Accordingly, this work has identified many peptides that belong to groups of peptides of bacterial general stress response proteins, heat shock proteins, and cold-shock-like proteins (CSPs), among others . A total of 20 peptides corresponding to proteins that are involved in bacterial toxicity were identified. These peptides include ecotin, lipoprotein toxin enterocidin B, antitoxin ParD, and addition module toxin GnSA/GnsB. As an example of some of the roles these peptides play, ecotin is an inhibitor of multiple complement-dependent processes found in bacteria . In this study, 349 peptides involved in colonization and immune evasion were identified. Bacterial internalization into the host is facilitated by these proteins, resulting in subsequent infection and propagation. Transcriptional regulators involved in the control of virulence factors were also found for the analyzed strains, including two peptides identified as LysR and SlyA . LysR regulates virulence factors, such as extracellular polysaccharides, toxins, and bacteriocins. Fimbria are located on the surfaces of bacteria; they are involved in adherence to target cells and biofilm formation . Lysis proteins belonging to the LysM domain were identified; this domain was identified in enzymes involved in bacterial cell wall degradation . Additionally, several peptides of peptidases and proteases were identified. This includes members of the Lon protease family and subtilisin, among others. Different peptides of the Superoxide dismutase enzyme (SOD) were identified. SOD is a metalloenzyme that defends against reactive oxygen species produced by neutrophils and macrophages . The presence of open channels facilitates passive penetration though the outer membrane. We have identified several porins or outer-membrane proteins (OMPs), such as the porins OmpA, OmpX, and OmpC; substrate-specific porins, such as maltoporin, which is also called LamB; and TonB-dependent receptors, such as FhuA . In addition, many peptides were determined to be other virulence factors, such as VacJ family lipoprotein VacJ (virulence-associated chromosome locus J); the chaperone protein Skp, which assists in the folding and insertion of many OMPs ; and the Osmy chaperone. In this study, four peptides of antibacterial proteins were identified, including one peptide that belongs to a bacteriocin and the remaining three to a colicin-like protein. Colicins are antimicrobial proteins typically produced by E. coli that degrade internal cellular elements . ABC transporters, like many other bacterial transporters, are involved in resistance or tolerance and bacterial propagation during infection . We identified different ABC transporters related to virulence ( ). Furthermore, sixteen peptides of alternative virulent factors were identified, such as proteins related to mobile genetic elements’ transposases, recombinases, plasmids, and viral DNA fragments, which are considered the major mechanism for acquiring antibiotic resistance. Moreover, pilus conforms to a typical method of horizontal transfer between bacteria, which is another mechanism of obtaining virulence determinants . We identified 43 peptides of phage proteins, such as bacteriophage CI repressor and capsid scaffolding protein, but mainly phage shock proteins. Finally, we identified bacterial proteins determined in the UniProt database in different phage strains ( Klebsiella phage vB_KpM_FBKp24, Klebsiella phage vB_KppS-Storm, Stenotrophomonas phage BUCT608, and Stenotrophomonas phage). To select potential peptide biomarkers for the 15 different biogenic-amine-producing bacterial strains, we implemented a massive comparison of the proteomics data with respect to the proteins and peptides included in databases. The suitable peptides that were identified via LC-ESI-MS/MS in only one specific species were verified in terms of their specificity and sequence homology using the BLASTp algorithm ( ). summarizes the analysis of the 77 species-specific tryptic peptide biomarkers belonging to 64 different proteins that were suggested for the unequivocal identification of the different seafood-originating biogenic-amine-producing bacteria of 10 different species ( E. aerogenes , E. cloacae , H. alvei , K. oxytoca , M. morganii , P. mirabilis , P. penneri , P. vulgaris , R. planticola , and S. maltophilia ). All the peptides included herein have been proposed as potential biomarkers for the first time and will be very convenient for further studies using targeted proteomics approaches to identify the different seafood-originating biogenic-amine-producing bacteria in foodstuffs. 3.1. Bacterial Strains A total of 15 different seafood-originating biogenic-amine-producing bacteria were included in this work ( ). Strains were previously studied via MALDI-TOF-MS and 16S rRNA sequencing . All bacterial strains were activated in brain–heart infusion (BHI) and incubated in vials at 31 °C for 24 h. Then, strain cultures were expanded on plate count agar (PCA) at 31 °C for 24 h. Samples were prepared in triplicate. 3.2. Protein Extraction Protein extracts were obtained as described by Carrera et al. (2017) [ , , ]. Concisely, the biomass of bacterial cells was mixed with a solution of 1% trifluoracetic acid/50% acetonitrile. After several extractions with glass beads conducted for 10 min at 4 °C, the supernatants were centrifuged for 10 min at 40,000× g (J221-M centrifuge, Beckman, Brea, CA, USA). The supernatant was then solubilized with lysis buffer containing 60 mM Tris-HCl pH 7.5, 1% lauryl maltoside, 5 mM phenylmethanesulfonylfluoride (PMSF), and 1% dithiothreitol (DTT). The solution was transferred to a new vial, and the quantity of protein was revealed via the bicinchoninic acid method (Sigma Chemical Co., St. Louis, MO, USA). This method was chosen because a similar procedure has been applied previously for protein extraction via MALDI-TOF MS analysis . 3.3. Peptide Sample Preparation Proteins were digested with trypsin, as described previously . A total of 100 μg of protein extracts was dried under vacuum and solubilized in 25 μL of 8 M urea in 25 mM of ammonium bicarbonate at pH 8.0. After 5 min of sonication, DTT was added at a final concentration of 10 mM and incubated at 37 °C for 1 h. Then, iodoacetamide was supplemented at a final concentration of 50 mM and incubated at room temperature in darkness for 1 h. Next, the sample was diluted four times to a final concentration of 2 M urea with 25 mM ammonium bicarbonate (pH 8.0) and subjected to digestion with trypsin (ratio 1:100) (Promega, Wisconsin, WI, USA) at 37 °C overnight. 3.4. Shotgun LC-ESI-MS/MS Analysis in a LTQ-Orbitrap Instrument Peptides were acidified with 5% formic acid (FA) until attaining pH 2, cleaned on a C18 MicroSpinTM column (The Nest Group, Southborough, MA, USA), and analyzed via LC-MS/MS using a Proxeon EASY-nLC II LC machine (Thermo Scientific, San Jose, CA, USA) coupled with an LTQ-Orbitrap XL (Thermo Fisher Scientific). Separation of peptides (2 µg) was implemented on an RP column (EASY-Spray column, 50 cm × 75 µm ID, PepMap C18, 2 µm particles, 100 Å pore size, Thermo Fisher Scientific) with a 10 mm precolumn (Accucore XL C18, Thermo Scientific) containing 0.1% FA in Milli-Q water and 98% ACN and 0.1% FA as mobile phases A and B, respectively. A 240 min linear gradient from 5 to 35% B at a flow rate of 300 nL/min was used. A capillary temperature of 230 °C and spray voltage of 1.95 kV were used for ionization. Peptides were analyzed from 400 to 1600 amu (1 µscan) in positive mode, followed by 10 data-dependent CID MS/MS scans (1 µscans) using an isolation width of 3 amu and a normalized collision energy of 35%. Fragmented masses were set in dynamic exclusion for 30 s after the second fragmentation event. Unassigned charged ions were omitted from MS/MS analysis. 3.5. LC-ESI-MS/MS Data Processing MS/MS spectra were identified using SEQUEST-HT (Proteome Discoverer 2.4 package, Thermo Fisher Scientific) and compared to the Proteobacteria UniProt/TrEMBL database (with 2,627,375 protein sequence entries dating from August 2022). MS/MS spectra were analyzed using fully tryptic cleavage constraints, and up to two missed cleavage sites were permissible. Windows for tolerance were set at 10 ppm for precursor ions and 0.06 Da for MS/MS fragment ions. The variable modifications permitted were methionine oxidation (Mox), carbamidomethylation of Cys (C), and acetylation of the N-terminus of a protein (N-Acyl). The results were subjected to statistical analysis to determine the false discovery rate (FDR) regarding peptides using a decoy database and the Percolator algorithm included in the Proteome Discoverer 2.4 program . The FDR was kept below 1% for further analysis. The MS/MS proteomics data have been deposited to the ProteomeXchange Consortium via PRIDE with the dataset identifier PXD039320. To determine relative protein abundance for each strain, a label-free quantification (LFQ) method was used by applying the Minora Feature Detector node and the ANOVA (individual proteins) method included in the Proteome Discover 2.4 software (Thermo Fisher Scientific). Peak areas of ion features from the same peptide for different charge forms were combined into one value. 3.6. Euclidean Hierarchical Clustering The function heatmap.2 of the statistical package R (version (v) 4.1.1) ( http://www.r-project.org , accessed on 25 January 2023) was used to achieve the Euclidean hierarchical clustering of the data. The Ggplots v.4.1.1 package, the Euclidean distance metric, and the complete linkage for the agglomeration method were used as constraints. 3.7. Functional Analysis: Gene Ontology (GO) and Pathways Analysis The nonredundant protein IDs (column “Gene name” in ) were submitted to the PANTHER software ( http://www.pantherdb.org/ , accessed on 30 November 2022) for grouping established based on the following main types of interpretations: molecular function, biological process, and protein class. The statistical significance was also provided as a percentage. For this procedure, all orthologous gene ID entries were included as a reference set. The pathway analysis data were clustered, thus providing an approximation of the statistical significance of over- or underrepresentation according to the GO descriptors of the proteins in the proteome. 3.8. Network Analysis Protein network was developed by incorporating the orthologous gene IDs into the STRING program (v.11.5) ( http://string-db.org/ , accessed on 6 December 2022) . STRING is an enormous database of known and predicted protein interactions. Proteins are denoted with nodes, and interactions are represented as continuous lines. All edges were reinforced by at least one reference from the literature or from canonical information deposited in the STRING dataset. The confidence score was set at ≥0.7 (high confidence). MCL algorithm included on the STRING website was used to generate cluster networks, and a default value of 2 was assigned for all analyses. 3.9. Virulence Factors Virulence Factor of Pathogenic Bacteria Database (VFDB) ( http://www.mgc.ac.cn/VFs/ , accessed on 13 December 2022) was used to characterize virulence factors. Additionally, we prolonged the analysis to include virulence factors that are contained in several scientific publications [ , , , , , , ]. 3.10. Selection of Potential Peptide Biomarkers BLASTp algorithm applied to each identified peptide by LC-MS/MS was used to determine homologies and exclusiveness with respect to protein sequences recorded in the NCBI database . A total of 15 different seafood-originating biogenic-amine-producing bacteria were included in this work ( ). Strains were previously studied via MALDI-TOF-MS and 16S rRNA sequencing . All bacterial strains were activated in brain–heart infusion (BHI) and incubated in vials at 31 °C for 24 h. Then, strain cultures were expanded on plate count agar (PCA) at 31 °C for 24 h. Samples were prepared in triplicate. Protein extracts were obtained as described by Carrera et al. (2017) [ , , ]. Concisely, the biomass of bacterial cells was mixed with a solution of 1% trifluoracetic acid/50% acetonitrile. After several extractions with glass beads conducted for 10 min at 4 °C, the supernatants were centrifuged for 10 min at 40,000× g (J221-M centrifuge, Beckman, Brea, CA, USA). The supernatant was then solubilized with lysis buffer containing 60 mM Tris-HCl pH 7.5, 1% lauryl maltoside, 5 mM phenylmethanesulfonylfluoride (PMSF), and 1% dithiothreitol (DTT). The solution was transferred to a new vial, and the quantity of protein was revealed via the bicinchoninic acid method (Sigma Chemical Co., St. Louis, MO, USA). This method was chosen because a similar procedure has been applied previously for protein extraction via MALDI-TOF MS analysis . Proteins were digested with trypsin, as described previously . A total of 100 μg of protein extracts was dried under vacuum and solubilized in 25 μL of 8 M urea in 25 mM of ammonium bicarbonate at pH 8.0. After 5 min of sonication, DTT was added at a final concentration of 10 mM and incubated at 37 °C for 1 h. Then, iodoacetamide was supplemented at a final concentration of 50 mM and incubated at room temperature in darkness for 1 h. Next, the sample was diluted four times to a final concentration of 2 M urea with 25 mM ammonium bicarbonate (pH 8.0) and subjected to digestion with trypsin (ratio 1:100) (Promega, Wisconsin, WI, USA) at 37 °C overnight. Peptides were acidified with 5% formic acid (FA) until attaining pH 2, cleaned on a C18 MicroSpinTM column (The Nest Group, Southborough, MA, USA), and analyzed via LC-MS/MS using a Proxeon EASY-nLC II LC machine (Thermo Scientific, San Jose, CA, USA) coupled with an LTQ-Orbitrap XL (Thermo Fisher Scientific). Separation of peptides (2 µg) was implemented on an RP column (EASY-Spray column, 50 cm × 75 µm ID, PepMap C18, 2 µm particles, 100 Å pore size, Thermo Fisher Scientific) with a 10 mm precolumn (Accucore XL C18, Thermo Scientific) containing 0.1% FA in Milli-Q water and 98% ACN and 0.1% FA as mobile phases A and B, respectively. A 240 min linear gradient from 5 to 35% B at a flow rate of 300 nL/min was used. A capillary temperature of 230 °C and spray voltage of 1.95 kV were used for ionization. Peptides were analyzed from 400 to 1600 amu (1 µscan) in positive mode, followed by 10 data-dependent CID MS/MS scans (1 µscans) using an isolation width of 3 amu and a normalized collision energy of 35%. Fragmented masses were set in dynamic exclusion for 30 s after the second fragmentation event. Unassigned charged ions were omitted from MS/MS analysis. MS/MS spectra were identified using SEQUEST-HT (Proteome Discoverer 2.4 package, Thermo Fisher Scientific) and compared to the Proteobacteria UniProt/TrEMBL database (with 2,627,375 protein sequence entries dating from August 2022). MS/MS spectra were analyzed using fully tryptic cleavage constraints, and up to two missed cleavage sites were permissible. Windows for tolerance were set at 10 ppm for precursor ions and 0.06 Da for MS/MS fragment ions. The variable modifications permitted were methionine oxidation (Mox), carbamidomethylation of Cys (C), and acetylation of the N-terminus of a protein (N-Acyl). The results were subjected to statistical analysis to determine the false discovery rate (FDR) regarding peptides using a decoy database and the Percolator algorithm included in the Proteome Discoverer 2.4 program . The FDR was kept below 1% for further analysis. The MS/MS proteomics data have been deposited to the ProteomeXchange Consortium via PRIDE with the dataset identifier PXD039320. To determine relative protein abundance for each strain, a label-free quantification (LFQ) method was used by applying the Minora Feature Detector node and the ANOVA (individual proteins) method included in the Proteome Discover 2.4 software (Thermo Fisher Scientific). Peak areas of ion features from the same peptide for different charge forms were combined into one value. The function heatmap.2 of the statistical package R (version (v) 4.1.1) ( http://www.r-project.org , accessed on 25 January 2023) was used to achieve the Euclidean hierarchical clustering of the data. The Ggplots v.4.1.1 package, the Euclidean distance metric, and the complete linkage for the agglomeration method were used as constraints. The nonredundant protein IDs (column “Gene name” in ) were submitted to the PANTHER software ( http://www.pantherdb.org/ , accessed on 30 November 2022) for grouping established based on the following main types of interpretations: molecular function, biological process, and protein class. The statistical significance was also provided as a percentage. For this procedure, all orthologous gene ID entries were included as a reference set. The pathway analysis data were clustered, thus providing an approximation of the statistical significance of over- or underrepresentation according to the GO descriptors of the proteins in the proteome. Protein network was developed by incorporating the orthologous gene IDs into the STRING program (v.11.5) ( http://string-db.org/ , accessed on 6 December 2022) . STRING is an enormous database of known and predicted protein interactions. Proteins are denoted with nodes, and interactions are represented as continuous lines. All edges were reinforced by at least one reference from the literature or from canonical information deposited in the STRING dataset. The confidence score was set at ≥0.7 (high confidence). MCL algorithm included on the STRING website was used to generate cluster networks, and a default value of 2 was assigned for all analyses. Virulence Factor of Pathogenic Bacteria Database (VFDB) ( http://www.mgc.ac.cn/VFs/ , accessed on 13 December 2022) was used to characterize virulence factors. Additionally, we prolonged the analysis to include virulence factors that are contained in several scientific publications [ , , , , , , ]. BLASTp algorithm applied to each identified peptide by LC-MS/MS was used to determine homologies and exclusiveness with respect to protein sequences recorded in the NCBI database . This article presents the first shotgun proteomics study of 15 different foodborne strains of biogenic-amine-producing bacteria. By means of a rapid and easy procedure for preparing proteins, the results were used to differentiate several protein datasets, which, in turn, were used to determine relevant functional pathways and differentiate strains into different Euclidean hierarchical clusters. Additionally, a predicted protein-protein interaction network for foodborne biogenic-amine-producing bacteria was created. Most proteins were classified under pathways and networks related to energy, putrescine metabolism, and host-virus interactions. Additionally, 556 different virulence factors were identified. Most of these factors corresponded to functions/roles such as toxins, antimicrobial compound production, antimicrobial resistance, additional resistances and tolerances, host colonization and immune evasion, ABC transporters, phage proteins, and alternative virulence factors and proteins involved in horizontal transfer. Finally, 77 prospective species-specific peptide biomarkers corresponding to 64 different proteins were screened to identify unique potential peptide biomarkers for 10 biogenic-amine-producing bacterial species. To date, these results constitute the largest dataset of peptides and proteins from foodborne biogenic-amine-producing bacterial species strains. This repository provides data that can be used in further studies to develop new therapeutic treatments for biogenic-amine-producing bacterial species with respect to food intoxication and for the tracking of microbial sources in foodstuffs.
Perspectives on the Person-Centered Practice of Healthcare Professionals at an Inpatient Hospital Department: A Descriptive Study	d51ba62d-9cab-4c6c-9874-bc45555fb369	10178857	Patient-Centered Care[mh]	Person-centered care is increasingly recognized as a fundamental approach to the quality, safety, and sustainability of health services, and is anchoring a new paradigm in the international development of health policies and strategies [ , , ]. Person-centered care integrates the perspectives of individuals, families, and communities as users and participants, enabling us as healthcare professionals to meet the needs of the population across the lifespan . It is defined as “…an approach to practice established by shaping and promoting healthy relationships between carers, service users, and people significant to their lives. It is underpinned by values of respect for people, individual right to self-determination, mutual respect, and understanding. It is enabled by cultures of empowerment that promote continuous approaches to practice development” . Along with the recognition of the importance of establishing integrated and person-centered health services , several global challenges for their implementation in clinical practice have been identified [ , , , ]. Contextual factors, such as the organizational culture, the characteristics of health professionals, and the practice environment, pose the greatest challenges to developing cultures that support person-centered care [ , , , ]. According to the WHO , each country should define a strategy that considers the specificities of its health system, care setting, and health professionals. The development of the person-centered practice framework (PCPF) and of the instruments to monitor its presence in clinical practice has guided its implementation in healthcare settings to date. The PCPF presents the critical domains and concepts inherent to person-centered care and provides a reference for its implementation and development . The person-centered practice inventory is an instrument that maps to the theoretical domains of the PCPF and allows for an understanding of its practice, the identification of areas of potential improvements, and the design of targeted interventions to enhance person-centeredness . How healthcare professionals think about and value person-centeredness in care can have important implications for how they construct their decisions and represent their actions in practice . Several studies have highlighted the need for health professionals to orient the focus of care towards a person’s values, thoughts, and experiences in order to be more person-centered, rather than simply adhering to pre-established norms [ , , , ]. However, few studies have focused on the perceptions of healthcare professionals in multidisciplinary teams regarding person-centered care in the hospital setting [ , , ]. This study is part of a clinical study protocol designed to provide recommendations for the development of person-centered practice in the daily care of hospitalized older adults with chronic diseases within an internal medicine department. The current practice analysis refers to how person-centered practice is perceived and identified in the context under study, considering that all principles and domains presented in the PCPF are fundamental for the implementation of this practice. Therefore, the present study aims to characterize the perceptions of person-centered practice conducted by healthcare professionals whom comprise the multidisciplinary team of an inpatient hospital unit and explore the influence of sociodemographic and professional variables on their perceptions. 2.1. Design and Population A quantitative, descriptive, cross-sectional approach was adopted in this study. It was carried out in an internal medicine inpatient unit in a secondary hospital located in an urban area in Portugal, with a direct impact area of 500,000 inhabitants. The internal medicine unit had a capacity of 55 inpatient beds. The multidisciplinary team comprises the medical staff, including 4 senior graduate physicians, 12 graduate physicians, 18 physicians, and 20 specific-training interns; the nursing staff, consisting of 50 general care nurses and 3 specialist nurses; and the physiotherapy staff, containing 2 physiotherapists. As long as they met the inclusion criteria, all physicians, nurses, and physiotherapists working in the unit were eligible for inclusion in the study sample. 2.2. Inclusion Criteria All healthcare professionals working full-time in the internal medicine unit for the last six months were eligible to participate in the study. This ensured that all healthcare professionals in the multidisciplinary team perspectives on person-centered practice in the setting were covered. 2.3. Exclusion Criteria Healthcare professionals from other departments or specialties who provided care on an occasional basis in the context of the study were excluded. 2.4. Data Collection Data were collected between December 2021 and January 2022. A questionnaire including a section for the sociodemographic characterization ( ) and the person-centered practice inventory-staff (PCPI-S) was provided to the healthcare professionals who met the defined criteria. The PCPI-S is a valid scale, psychometrically accepted and validated by an international panel of experts in person-centered practice , with proven reliability [ , , , , ]. It enabled us to understand how health professionals perceived their practice concerning person-centredness. The PCPI-S is a self-report instrument composed of 59 items on a five-point Likert-type response scale, with higher scores indicating a greater agreement . The PCPI-S measures three domains derived from the PCPF (i.e., prerequisites, the practice environment, and person-centered processes ) and comprises 17 constructs, including professionally competent, having developed interpersonal skills, knowing self, clarity of beliefs and values, commitment to the job, appropriate skill mix, shared decision-making systems, effective staff relationships, power sharing, the physical environment, supportive organizational systems, the potential for innovation and risk-taking, working with the person’s beliefs and values, sharing decision making, engaging authentically, being sympathetically present, and working holistically ( ) . The PCPI-S was translated and culturally adapted into Portuguese and showed acceptable psychometric properties and good reliability . The questionnaire was delivered to the healthcare professionals using an institutional email via Google Forms ® (Google Corp. 2018, Dublin, Ireland) to ensure the anonymity of the data collected. Healthcare professionals were encouraged to participate during the period available for data collection by sending a reminder email one week before the end and extending the response period by two weeks. 2.5. Statistical Data Analysis Quantitative data were analyzed using a statistical package for social sciences software (IBM SPSS Statistics ® for Windows, v. 27.0. IBM Corp. Released 2020, Armonk, NY, USA). A descriptive analysis (i.e., mean, standard deviation, minimum, and maximum) of the constructs comprising the PCPI-S was performed. An analysis of variance (ANOVA) was then performed to determine the effect of demographic and professional variables on the PCPI-S constructs, as described in the previous section. Specifically, dependent variables with more than two response options, for which statistically significant differences were found, were further evaluated using Tukey’s post hoc test for multiple comparisons. Model assumptions were assessed by analyzing a plot of residuals versus predicted values, a Q–Q plot of residuals, and a residual histogram. Whenever a violation of the ANOVA assumptions occurred, the dependent variable or construct was transformed using a Box–Cox transformation. A new ANOVA was then run, and the assumptions were reverified as described above. The plot of residuals versus predicted values, the Q–Q plot of residuals, and a residual histogram were generated in the R language statistical computing v.4.2.2 using the resid_auxpanel function from the ggResidpanel v.0.3.0 library after importing the SPSS data file containing the predicted and residual values of each ANOVA model using the read.spss function from the foreign v.0.8-83 library . A p -value of <0.05 was considered for statistical significance [ , , ] and values were rounded to the nearest hundredth. 2.6. Ethical Considerations The study received ethical approval from the ethics committee of the hospital where the study took place (ref. nr. 36/2021). All procedures were performed in accordance with the Declaration of Helsinki and in compliance with the General Data Protection Regulation . Permission to use the PCPI-S was requested and granted by the authors. A quantitative, descriptive, cross-sectional approach was adopted in this study. It was carried out in an internal medicine inpatient unit in a secondary hospital located in an urban area in Portugal, with a direct impact area of 500,000 inhabitants. The internal medicine unit had a capacity of 55 inpatient beds. The multidisciplinary team comprises the medical staff, including 4 senior graduate physicians, 12 graduate physicians, 18 physicians, and 20 specific-training interns; the nursing staff, consisting of 50 general care nurses and 3 specialist nurses; and the physiotherapy staff, containing 2 physiotherapists. As long as they met the inclusion criteria, all physicians, nurses, and physiotherapists working in the unit were eligible for inclusion in the study sample. All healthcare professionals working full-time in the internal medicine unit for the last six months were eligible to participate in the study. This ensured that all healthcare professionals in the multidisciplinary team perspectives on person-centered practice in the setting were covered. Healthcare professionals from other departments or specialties who provided care on an occasional basis in the context of the study were excluded. Data were collected between December 2021 and January 2022. A questionnaire including a section for the sociodemographic characterization ( ) and the person-centered practice inventory-staff (PCPI-S) was provided to the healthcare professionals who met the defined criteria. The PCPI-S is a valid scale, psychometrically accepted and validated by an international panel of experts in person-centered practice , with proven reliability [ , , , , ]. It enabled us to understand how health professionals perceived their practice concerning person-centredness. The PCPI-S is a self-report instrument composed of 59 items on a five-point Likert-type response scale, with higher scores indicating a greater agreement . The PCPI-S measures three domains derived from the PCPF (i.e., prerequisites, the practice environment, and person-centered processes ) and comprises 17 constructs, including professionally competent, having developed interpersonal skills, knowing self, clarity of beliefs and values, commitment to the job, appropriate skill mix, shared decision-making systems, effective staff relationships, power sharing, the physical environment, supportive organizational systems, the potential for innovation and risk-taking, working with the person’s beliefs and values, sharing decision making, engaging authentically, being sympathetically present, and working holistically ( ) . The PCPI-S was translated and culturally adapted into Portuguese and showed acceptable psychometric properties and good reliability . The questionnaire was delivered to the healthcare professionals using an institutional email via Google Forms ® (Google Corp. 2018, Dublin, Ireland) to ensure the anonymity of the data collected. Healthcare professionals were encouraged to participate during the period available for data collection by sending a reminder email one week before the end and extending the response period by two weeks. Quantitative data were analyzed using a statistical package for social sciences software (IBM SPSS Statistics ® for Windows, v. 27.0. IBM Corp. Released 2020, Armonk, NY, USA). A descriptive analysis (i.e., mean, standard deviation, minimum, and maximum) of the constructs comprising the PCPI-S was performed. An analysis of variance (ANOVA) was then performed to determine the effect of demographic and professional variables on the PCPI-S constructs, as described in the previous section. Specifically, dependent variables with more than two response options, for which statistically significant differences were found, were further evaluated using Tukey’s post hoc test for multiple comparisons. Model assumptions were assessed by analyzing a plot of residuals versus predicted values, a Q–Q plot of residuals, and a residual histogram. Whenever a violation of the ANOVA assumptions occurred, the dependent variable or construct was transformed using a Box–Cox transformation. A new ANOVA was then run, and the assumptions were reverified as described above. The plot of residuals versus predicted values, the Q–Q plot of residuals, and a residual histogram were generated in the R language statistical computing v.4.2.2 using the resid_auxpanel function from the ggResidpanel v.0.3.0 library after importing the SPSS data file containing the predicted and residual values of each ANOVA model using the read.spss function from the foreign v.0.8-83 library . A p -value of <0.05 was considered for statistical significance [ , , ] and values were rounded to the nearest hundredth. The study received ethical approval from the ethics committee of the hospital where the study took place (ref. nr. 36/2021). All procedures were performed in accordance with the Declaration of Helsinki and in compliance with the General Data Protection Regulation . Permission to use the PCPI-S was requested and granted by the authors. 3.1. Characteristics of the Participants The convenience sample included 109 health professionals, with a response rate of 76.15% (N = 83). The sample was predominantly composed of female professionals (79.5%), and the most representative professional group was nurses (61.5%), followed by physicians (36.1%) and physiotherapists (2.4%). Regarding the educational level, 62.7% of the health professionals had a university degree and a maximum of 24 years of education (M = 17.6; SD = 2.09). The sample had a mean of 11 years of professional experience (SD = 8.06; Min = 0; Max = 40), with 45.8% of the healthcare professionals having less than 10 years of experience, 41% having between 10 and 20 years, and 13.3% having more than 20 years of experience ( ). 3.2. Perception of Person-Centered Practice The results were analyzed using the mean score of the response scale (one to five points), accordingly to the authors’ guidelines . Constructs with a mean score greater than 2.5 were considered positive, indicating an agreement among healthcare professionals. All domains were positively rated by the different professional groups ( ). The prerequisites domain had the highest score (M = 4.12; SD = 0.36), followed by person-centered processes (M = 4.08; SD = 0.51) and the practice environment (M = 3.50; SD = 0.48). Three constructs with very high scores emerged from the prerequisites domain, namely, developed interpersonal skills (M = 4.35; SD = 0.47), which had the highest score of all the constructs analyzed, professionally competent (M = 4.26; SD = 0.42), and commitment to the job (M = 4.25; SD = 0.42). Conversely, clarity of beliefs and values (M = 3.66; SD = 0.60) and knowing self (M = 3.91; SD = 0.72) had the lowest scores. In the practice environment , the highest scoring construct was a ppropriate skill mix (M = 4.02; SD = 0.52), and the lowest scoring was supportive organization systems (M = 3.08; SD = 0.80), which represented the lowest score of all 17 constructs. In person-centered processes , working holistically (M = 4.22; SD = 0.62) and engaging authentically (M = 4.17; SD = 0.52) had the highest response scores, and sharing decision making (M = 3.91; SD = 0.72) had the lowest. 3.3. Influence of Sociodemographic and Professional Characteristics 3.3.1. Prerequisites Gender, profession, and educational level were found to have a significant effect on the constructs of the prerequisites domain. The educational level significantly influenced the health professionals’ perceptions of being professionally competent (F(1,83) = 4.98, p -value = 0.029, partial η 2 = 0.062) ( ), with a decrease in the value assigned to professionally competent of 0.218 between professionals with and without a degree. Participants’ professions also significantly influenced the perceptions of commitment to the job (F(2,83) = 5.27, p -value = 0.007, partial η 2 = 0.123) ( ). There were significant differences between the perceptions of physicians (M = 4.25; SD = 0.43) and nurses (M = 4.22; SD = 0.39) when compared to physical therapists (M = 5; SD = 0.0). In addition, the educational level significantly influenced the construct of commitment to the job (F(1,83) = 4.49, p -value = 0.037, partial η 2 = 0.056) ( ), indicating that there were significant differences between professionals who held a degree (M = 4.19; SD = 0.42) compared to those who completed postgraduate studies (M = 4.35; SD = 0.40). Participants’ genders significantly influenced knowing self ‘(F(2,83) = 3.67, p -value = 0.030, partial η 2 = 0.089) ( ), where being female increased the perception of knowing self (B = 0.527). No significant effect due to the independent variables was found in the constructs clarity of beliefs and values ( ) nor developed interpersonal skills ( ). 3.3.2. The Practice Environment The profession and gender of healthcare professionals were found to significantly influence perceptions of the practice environment domain. In addition, profession was found to significantly influence shared decision-making systems (F(2,83) = 5.38, p -value = 0.007, partial η 2 = 0.125) ( ). There were significant differences in the perceptions of physicians (M = 3.60; SD = 0.63) compared to nurses (M = 3.08; SD = 0.84) and physiotherapists (M = 2.75; SD = 1.06). Gender significantly influenced the physical environment (F(2,83) = 3.63, p -value = 0.031, partial η 2 = 0.088) ( ), with women having a higher score assigned to the perception of the physical environment (B = 0.532). No significant effect due to the independent variables was found in the appropriate skill mix ( ), effective staff relationships ( ), power sharing ( ), the potential for innovation and risk-taking ( ), and supportive organization systems ( ). 3.3.3. Person-Centered Processes No significant effect due to the independent variables was found for any of the constructs in the person-centered processes domain, namely, working with the person’s beliefs and values ( ), sharing decision making ( ), engaging authentically ( ), being sympathetically present ( ), and working holistically ( ). 3.4. Reliability of the PCPI-S All domains had an α of >0.85, and there was a significant correlation between them ( ). The internal consistency of each domain of the PCPI-S was assessed, with prerequisites and the practice environment showing good consistency (α = 0.85 and α = 0.88, respectively) and person-centered processes showing excellent consistency (α = 0.94) ( ) . Overall, an adequate internal consistency of the inventory was found when applied to the study sample. The convenience sample included 109 health professionals, with a response rate of 76.15% (N = 83). The sample was predominantly composed of female professionals (79.5%), and the most representative professional group was nurses (61.5%), followed by physicians (36.1%) and physiotherapists (2.4%). Regarding the educational level, 62.7% of the health professionals had a university degree and a maximum of 24 years of education (M = 17.6; SD = 2.09). The sample had a mean of 11 years of professional experience (SD = 8.06; Min = 0; Max = 40), with 45.8% of the healthcare professionals having less than 10 years of experience, 41% having between 10 and 20 years, and 13.3% having more than 20 years of experience ( ). The results were analyzed using the mean score of the response scale (one to five points), accordingly to the authors’ guidelines . Constructs with a mean score greater than 2.5 were considered positive, indicating an agreement among healthcare professionals. All domains were positively rated by the different professional groups ( ). The prerequisites domain had the highest score (M = 4.12; SD = 0.36), followed by person-centered processes (M = 4.08; SD = 0.51) and the practice environment (M = 3.50; SD = 0.48). Three constructs with very high scores emerged from the prerequisites domain, namely, developed interpersonal skills (M = 4.35; SD = 0.47), which had the highest score of all the constructs analyzed, professionally competent (M = 4.26; SD = 0.42), and commitment to the job (M = 4.25; SD = 0.42). Conversely, clarity of beliefs and values (M = 3.66; SD = 0.60) and knowing self (M = 3.91; SD = 0.72) had the lowest scores. In the practice environment , the highest scoring construct was a ppropriate skill mix (M = 4.02; SD = 0.52), and the lowest scoring was supportive organization systems (M = 3.08; SD = 0.80), which represented the lowest score of all 17 constructs. In person-centered processes , working holistically (M = 4.22; SD = 0.62) and engaging authentically (M = 4.17; SD = 0.52) had the highest response scores, and sharing decision making (M = 3.91; SD = 0.72) had the lowest. 3.3.1. Prerequisites Gender, profession, and educational level were found to have a significant effect on the constructs of the prerequisites domain. The educational level significantly influenced the health professionals’ perceptions of being professionally competent (F(1,83) = 4.98, p -value = 0.029, partial η 2 = 0.062) ( ), with a decrease in the value assigned to professionally competent of 0.218 between professionals with and without a degree. Participants’ professions also significantly influenced the perceptions of commitment to the job (F(2,83) = 5.27, p -value = 0.007, partial η 2 = 0.123) ( ). There were significant differences between the perceptions of physicians (M = 4.25; SD = 0.43) and nurses (M = 4.22; SD = 0.39) when compared to physical therapists (M = 5; SD = 0.0). In addition, the educational level significantly influenced the construct of commitment to the job (F(1,83) = 4.49, p -value = 0.037, partial η 2 = 0.056) ( ), indicating that there were significant differences between professionals who held a degree (M = 4.19; SD = 0.42) compared to those who completed postgraduate studies (M = 4.35; SD = 0.40). Participants’ genders significantly influenced knowing self ‘(F(2,83) = 3.67, p -value = 0.030, partial η 2 = 0.089) ( ), where being female increased the perception of knowing self (B = 0.527). No significant effect due to the independent variables was found in the constructs clarity of beliefs and values ( ) nor developed interpersonal skills ( ). 3.3.2. The Practice Environment The profession and gender of healthcare professionals were found to significantly influence perceptions of the practice environment domain. In addition, profession was found to significantly influence shared decision-making systems (F(2,83) = 5.38, p -value = 0.007, partial η 2 = 0.125) ( ). There were significant differences in the perceptions of physicians (M = 3.60; SD = 0.63) compared to nurses (M = 3.08; SD = 0.84) and physiotherapists (M = 2.75; SD = 1.06). Gender significantly influenced the physical environment (F(2,83) = 3.63, p -value = 0.031, partial η 2 = 0.088) ( ), with women having a higher score assigned to the perception of the physical environment (B = 0.532). No significant effect due to the independent variables was found in the appropriate skill mix ( ), effective staff relationships ( ), power sharing ( ), the potential for innovation and risk-taking ( ), and supportive organization systems ( ). 3.3.3. Person-Centered Processes No significant effect due to the independent variables was found for any of the constructs in the person-centered processes domain, namely, working with the person’s beliefs and values ( ), sharing decision making ( ), engaging authentically ( ), being sympathetically present ( ), and working holistically ( ). Gender, profession, and educational level were found to have a significant effect on the constructs of the prerequisites domain. The educational level significantly influenced the health professionals’ perceptions of being professionally competent (F(1,83) = 4.98, p -value = 0.029, partial η 2 = 0.062) ( ), with a decrease in the value assigned to professionally competent of 0.218 between professionals with and without a degree. Participants’ professions also significantly influenced the perceptions of commitment to the job (F(2,83) = 5.27, p -value = 0.007, partial η 2 = 0.123) ( ). There were significant differences between the perceptions of physicians (M = 4.25; SD = 0.43) and nurses (M = 4.22; SD = 0.39) when compared to physical therapists (M = 5; SD = 0.0). In addition, the educational level significantly influenced the construct of commitment to the job (F(1,83) = 4.49, p -value = 0.037, partial η 2 = 0.056) ( ), indicating that there were significant differences between professionals who held a degree (M = 4.19; SD = 0.42) compared to those who completed postgraduate studies (M = 4.35; SD = 0.40). Participants’ genders significantly influenced knowing self ‘(F(2,83) = 3.67, p -value = 0.030, partial η 2 = 0.089) ( ), where being female increased the perception of knowing self (B = 0.527). No significant effect due to the independent variables was found in the constructs clarity of beliefs and values ( ) nor developed interpersonal skills ( ). The profession and gender of healthcare professionals were found to significantly influence perceptions of the practice environment domain. In addition, profession was found to significantly influence shared decision-making systems (F(2,83) = 5.38, p -value = 0.007, partial η 2 = 0.125) ( ). There were significant differences in the perceptions of physicians (M = 3.60; SD = 0.63) compared to nurses (M = 3.08; SD = 0.84) and physiotherapists (M = 2.75; SD = 1.06). Gender significantly influenced the physical environment (F(2,83) = 3.63, p -value = 0.031, partial η 2 = 0.088) ( ), with women having a higher score assigned to the perception of the physical environment (B = 0.532). No significant effect due to the independent variables was found in the appropriate skill mix ( ), effective staff relationships ( ), power sharing ( ), the potential for innovation and risk-taking ( ), and supportive organization systems ( ). No significant effect due to the independent variables was found for any of the constructs in the person-centered processes domain, namely, working with the person’s beliefs and values ( ), sharing decision making ( ), engaging authentically ( ), being sympathetically present ( ), and working holistically ( ). All domains had an α of >0.85, and there was a significant correlation between them ( ). The internal consistency of each domain of the PCPI-S was assessed, with prerequisites and the practice environment showing good consistency (α = 0.85 and α = 0.88, respectively) and person-centered processes showing excellent consistency (α = 0.94) ( ) . Overall, an adequate internal consistency of the inventory was found when applied to the study sample. The sample reflected the Portuguese reality, where 76.5% of healthcare workers were female and the most represented professional group were nurses . According to the Organization for Economic Cooperation and Development , the nurse/physician ratio is 1.3, whereas in the sample of this study, the ratio was 1.7. Healthcare professionals’ perceptions of person-centered practice were positive, with all constructs having mean scores greater than 2.5 (min = 3.08; max = 4.35). The prerequisites related to the characteristics of the multidisciplinary team and were considered the critical foundations for the development of professionals toward a person-centered practice . Healthcare professionals valued the prerequisites of developed interpersonal skills (M = 4.26; SD = 0.42), professionally competent (M = 4.35; SD = 0.47), and commitment to the job (M = 4.25; SD = 0.42), giving relevance to communicating effectively, demonstrating commitment to finding mutual solutions, and providing holistic care that integrates knowledge, skills, and experience to negotiate care options . However, clarity of beliefs and values received the lowest score of the domain (M = 3.66; SD = 0.60). This construct related to the awareness of the impact of professionals’ beliefs and values on the care provided and the commitment to reconcile them to facilitate person-centeredness . Similarly, knowing self was related to self-awareness and to the perception of the person regarding self-knowledge, which, although showing a positive score, was below four (M = 3.91; SD = 0.72). This result may be related to the lack of critical thinking and a reflection of the practice in the study context, as both constructs depended on individual development based on reflection. McCance et al. reiterated that clarity of beliefs and values of healthcare professionals is the foundation for culture changes, which are essential for professionals to move towards person-centeredness. Participants in the study valued the interpersonal relationships and their commitment and involvement in professional practice, having the potential to develop the team’s shared professional values and demonstrate them in practice. Aligning the values adopted by the team with the behaviors experienced in practice is essential to transform the culture, context, and consistency of the care provided . When comparing the results of this study with studies using a similar methodology, in which the PCPI-S was applied to nurses in a hospital setting, such as the study by Slater et al. , similar results were obtained for each construct, where commitment to the job (M = 4.45; SD = 0.40), professionally competent (M = 4.26; SD = 0.41), and developed interpersonal skills (M = 4.37; SD = 0.38) scored highest, whilst clarity of beliefs and values (M = 3.91; SD = 0.54) and knowing self (M = 4.04; SD = 0.52) scored lowest. Tiainen et al. also obtained similar results for developed interpersonal skills (M = 4.08; SD = 0.48) and being professionally competent (M = 4.07; SD = 0.51). McCance et al. conducted a study with a multidisciplinary sample that also highlighted the commitment to the job (M = 4.39; SD = 0.47), professionally competent (M = 4.24; SD = 0.46), and developed interpersonal skills (M = 4.32; SD = 0.43) as the most valued constructs, and clarity of beliefs and values (M = 3.90; SD = 0.58) and knowing self (M = 3.96; SD = 0.58) as the least valued. The prerequisites domain is essential in triggering significant changes in the practice environment and the professionals’ involvement in person-centered processes . As this domain was the most valued by health professionals, it suggested the existence of individual conditions for the development of person-centered practice in context. The practice environment domain refers to contextual aspects and influences the operationalization of person-centered practice through its potential to facilitate or inhibit person-centered processes . Herein, the constructs that showed lower mean scores belonged to the domains of prerequisites and person-centered processes , with results that were similar to those obtained by Slater et al. , Tiainen et al. , and McCance et al. . In addition, Johnsen et al. reported that healthcare professionals working in acute inpatient hospital settings identified fewer aspects of organizational culture related to person-centered practice, reinforcing the need to emphasize the environmental aspects in this context. In the practice environment , the health professionals in the sample rated the multidisciplinary team’s knowledge and skill mix as essential to providing quality care, scoring high in appropriate skill mix (M = 4.02; SD = 0.52). The studies conducted by Slater et al. , Tiainen et al. , and McCance et al. also showed an appropriate skill mix with the highest scores (M = 4.22, SD = 0.45; M = 4.15, SD = 0.46 and M = 4.15, SD = 0.51, respectively). The lowest scoring constructs referred to supportive organization systems (M = 3.08; SD = 0.80), i.e., organizational systems that promote people’s initiative, creativity, freedom, and security, supported by a structure that privileges culture, relationships, values, communication, professional autonomy, and accountability . Of the 17 constructs analyzed, supportive organization systems received the lowest score, indicating that professionals perceived a lack of support from the organization in areas critical to practice changes. To better understand these results, it would be important to characterize the environment of care and the institution’s mission, values, and regulations to determine whether these aspects are consistent with or supportive of person-centered practice. McCance et al. and Hower et al. identified the significant impact of contextual factors on the implementation of person-centered practice. They recognized the importance of the institution in changing practice and promoting a person-centered culture. Slater et al. , Tiainen et al. , and McCance et al. also found identical results for these constructs (M = 3.43, SD = 0.66; M = 3.25; SD = 0.48 and M = 3.18, SD = 0.83, respectively). With a similarly low score, the shared decision-making systems construct (M = 3.26; sd = 0.81) refers to the organizational commitment to collaborative and participatory ways in which the healthcare team engages in decision making. McCance et al. also found this construct to be a predictor of person-centered culture, reinforcing the importance of interdisciplinarity and patient involvement in care. The low perception of shared decision-making systems in our study may indicate that healthcare professionals need to be committed to a collaborative culture that involves all participants in decision making. Otherwise, the patient’s involvement in care may be compromised. Person-centered processes describe care delivery, operationalized through a set of person-centered activities . Here, healthcare professionals scored highest on working holistically (M = 4.22, SD = 0.62), representing their value in integrating physiological, psychological, sociocultural, developmental, and spiritual dimensions into care delivery. Similarly, the scores on engaging authentically (M = 4.17, SD = 0.52) highlighted the recognition of the importance of the professional’s connection to the person being cared for and the people who matter to them, as determined by the knowledge of the person, clarity of their beliefs and values, knowing self, and professional experience . In the studies by Slater et al. , Tiainen et al. , and McCance et al. , a higher score was also found in working holistically (M = 4.40, SD = 0.44; M = 4.14, SD = 0.55 and M = 4.30, SD = 0.53, respectively). In engaging authentically , similar scores were obtained in the referred studies (M = 4.18, SD = 0.46; M = 4.01, SD = 0.47 and M = 4.20, SD = 0.47, respectively), although it was not the most valued construct within them. The lowest scores were assigned to being sympathetically present (M = 4.07, SD = 0.59), working with the person’s beliefs and values (M = 4.05; SD = 0.52), and the sharing decision making (M = 3.91; SD = 0.72) constructs. McCance et al. suggested that being sympathetically present is a core element concerning all of the other person-centered processes constructs, and is highly connected with working with the person’s beliefs and values as it depends on knowing the patients and having insight into their beliefs and values to maximize coping resources. The construct of sharing decision making (M = 3.91; SD = 0.72) showed the lowest response score within the person-centered processes , revealing that healthcare professionals should recognize their role in facilitating and reinforcing the patient’s involvement in decision making . This construct was also closely linked to working with the person’s beliefs and values as the foundation of the involvement in decision making are sustained on the person’s values, experiences, concerns, beliefs, and future aspirations. Knowing that working with the person’s beliefs and values supports and influences these structural constructs could be a key to the development of person-centered practice in this context. The low score obtained in sharing decision making was not surprising when verifying the score of shared decision-making systems . Without an organizational commitment shared among healthcare professionals, the team cannot engage with the patient in decision making . In the study of Gregório et al. , which was conducted in a representative sample of the Portuguese population, most people preferred a controlling role of the professional rather than actively participating in clinical decision making. Healthcare professionals should be alert to this fact and increasingly recognize the importance of the person’s involvement in clinical decision making for person-centeredness. Tiainen et al. had similar results (M = 3.92, SD = 0.53) in sharing decision making . However, in the studies by Slater et al. and McCance et al. , the score was higher (M = 4.21, SD = 0.52 and M = 4.09, SD = 0.58, respectively). The consistency of the results in the prerequisites and the practice environment with studies of similar characteristics may be related to the fact that all studies were conducted in Europe, namely, in England , Finland , and Ireland . The cultural similarity could have influenced the characteristics of both the practitioners and the contexts. Concerning the influence of sociodemographic and professional characteristics on the perception of person-centered practice, the female gender positively influenced the constructs of knowing self and the physical environment . In healthcare professions, women tend to be prevalent. Therefore, understanding the influence of gender on the care provided is essential. The construct of knowing self refers to how the healthcare professional gives meaning to knowledge and action, using reflection, self-awareness, and engagement with others in the search for a person-centered practice . Al-surimi et al. justified the difference in this perception, as female professionals naturally value the relational aspects of care. An aesthetically pleasant physical environment stimulates the senses and promotes healing, well-being, care, and involvement in interprofessional relationships . Female professionals value this aspect, while males may tend to focus more on the interventions and procedural aspects of care rather than on the characteristics of the physical environment . The profession significantly influenced the constructs of shared decision-making systems and commitment to the job . Gemmae et al. and Dahlke et al. reported that professional groups had different perceptions about the fundamental concepts of person-centered practice according to their intervention area. Physicians had a more positive perception of the shared decision-making systems than nurses and physiotherapists. Professional interdependence and the degree of autonomy of each professional group may explain this result. However, this construct was the least valued by the multidisciplinary team (M = 3.26; SD = 0.81), indicating the need to strengthen the commitment to participatory collaboration between team members and patients. Shared decision-making systems are a likely predictor of the development of a person-centered culture due to the importance of shared decision making among the multidisciplinary team and the person’s involvement in care . Given the results obtained in this construct, it could be expectable that the profession would exert a similar influence on the constructs of power sharing and effective staff relationships , which did not occur. These results could indicate that despite recognizing the absence of shared decision-making systems among professionals and patients, the study participants perceived the relationships in the team and power sharing as favorable to a person-centered practice. This relationship should be the focus of qualitative inquiry if the aim is to improve the quality of care toward person-centeredness. Concerning commitment to the job , physiotherapists were assigned a higher score than physicians and nurses. Commitment to persons and family through the professionals’ involvement in the relationship was valued by those who spent less time in contact with patients. However, this construct should be analyzed in a broader spectrum since commitment as a multidisciplinary team member should overlap with individual commitments . Thus, the discrepancy in perceptions between the different healthcare professionals in this study could reveal the absence of a shared commitment at the organizational level. In addition, the length of training showed an increasing influence on the commitment to the job , which was not surprising considering that engagement in the relationship is supported by a holistic view based on evidence and education . The educational level also influenced the construct of professionally competent . Professionals with higher education tend to value knowledge, skills, and attitudes for negotiating care options . The professionally competent aspect includes professional knowledge and experience. However, professional experience did not influence this construct in the sample studied. Professional experience did not influence the perception of any construct. This was in contrast to the study by Esmaeili et al. , which showed that professional experience was associated with the provision of holistic, collaborative, and comprehensive care. Similarly, the study by Tiainen et al. showed a positive influence of nurses’ professional experiences on the perception of the constructs of professionally competent and the physical environment . Johnsen et al. found that health professionals with postgraduate education showed greater involvement to patients in decision making than those with a degree. The fact that professional experience did not influence the perception of person-centeredness in this study may be related to the categorization of the variable. The categorization was determined to facilitate comparisons with previous studies that used the same methodology. However, the categorization may need to be reviewed. The statistically significant differences on the scores of the constructs between the different groups highlighted the usefulness of the PCPI-S in identifying areas of development that are appropriate for different professionals. Overall, healthcare professionals in the context studied demonstrated an understanding of person-centered practice in their work context. In summary, in prerequisites , the construct of clarity of beliefs and values revealed the need to gain an awareness of its impact on the healthcare experience and the need to develop team-aligned values to move toward a person-centered culture. The practice environment was identified as the domain requiring greatest investment with lower scores on the PCPI-S. The supportive organization systems and shared decision-making systems indicated the lack of organizational systems that promoted professional initiative, creativity, and autonomy, and ones that value communication, relationships, and participation among healthcare professionals. The low score for the shared decision-making main theme was reinforced in the person-centered processes , highlighting the need for healthcare professionals to be reinforced as facilitators of participation in the setting and to work on recognizing the person’s values, experiences, concerns, and beliefs, as their individual perspective and psychosocial role are the foundation of decision making. Therefore, these aspects should be the focus of special attention to improve person-centeredness. In order to initiate and sustain an effective change toward person-centered practice, its components must also be identified at all levels of care delivery . Therefore, in addition to aligning all levels of care with the principles of person-centered practice, it is necessary to ensure that aspects of the practice environment are sufficiently valued in the context . McCormack et al. suggested that contextual factors, such as the organizational culture, the learning environment, and the care environment itself, pose the greatest challenge to person-centeredness and the development of cultures that can support person-centered practice. This study was one of the first to systematically assess the factors influencing person-centeredness in a multidisciplinary team in a hospital setting, which limited the comparability of the results. The categorization of the sociodemographic variables was chosen in order to allow for a comparison with studies with similar methodologies. However, this may have limited us from conducting a more in-depth analysis of the impact of the variable on the healthcare professionals’ perception of person-centered care. The high scores obtained on the different constructs raised the question of whether the participants’ responses reflected their idealized practice or their real and current perception of daily care. Therefore, qualitative studies with multidisciplinary samples should be conducted to triangulate the results obtained in this study, as suggested by Vareta et al. . The fact that the research was conducted in a specific context, namely, in an internal medicine department with a small sample of healthcare professionals, limited the possibility of transferability to other care settings or populations. However, this involved a multidisciplinary team with a high response rate, which was considered a strength. The key concepts for implementing and developing person-centered practice in inpatient settings were positively identified by the professionals of the multidisciplinary team of the study context. The PCPI-S proved to be sensitive in identifying health professionals’ perceptions of the person-centeredness of their practice and in identifying significant differences in perceptions between groups, taking into account their personal and professional characteristics and, thus, contributing to the development of care practice. In addition, the influence of sociodemographic and professional characteristics on the scores obtained, considered statistically significant, allowed for the identification of groups and areas of differentiated intervention for sustainable practice developments specifically adapted to the context and person. The results of this study contributed to a growing evidence base regarding the PCPI as a psychometrically sound instrument that allows for the structural concepts of an established theory to be identified and inform changes in practice. Characterizing the culture and organizational structure of the context in future studies could allow for a deeper understanding of the relationships between the domains and the multifaceted factors that facilitate or limit them. It would be essential to analyze the influence of the values and customs of different countries on the perception of the person-centered practice.
A Longitudinal Study of Individual Radiation Responses in Pediatric Patients Treated with Proton and Photon Radiotherapy, and Interventional Cardiology: Rationale and Research Protocol of the HARMONIC Project	6ce7c350-5479-45c9-8493-32569d74bac1	10178896	Internal Medicine[mh]	The use of ionizing radiation (IR) in medicine represents significant benefits for the medical care of patients. Indeed, IR remains one of the major therapeutic options for cancer treatment . IR is also used for diagnostic and therapeutic imaging, particularly for pediatric patients with congenital or acquired heart disease, who may receive one or more cardiac catheterization procedures as part of their management . While benefits for the patients largely outweigh the risk, the potential adverse health effects of exposure to IR are particularly important to be explored in populations of young patients who are more radiosensitive and, nowadays, survive their disease for decades . IR is a well-known risk factor for cancer induction, and recent studies support the existence of an excess cancer risk, even at low doses of radiation . Children are especially vulnerable to the oncogenic effects of IR. Moreover, the oncogenic effects of IR require a long latent period (from years to decades) that varies with the type of malignancy; therefore, an infant or child has a longer lifetime risk for developing radiation-induced cancers than an adult. Radiation-induced second malignancies are one of the most serious adverse effects following radiotherapy of primary cancers in childhood cancer . Typically, radiation-induced malignancies develop in normal tissue within radiotherapy fields with a latency period of 5–10 years for hematologic malignancies and 10–60 years for solid tumors . Many clinical studies reported an increased risk for second primary cancer, histopathologically different from the first tumor, in organs inside as well as outside the primary beam . Interestingly, it was reported that leukemias and carcinomas are more often seen in organs receiving low-dose radiation (out-of-field dose), whereas sarcomas are more common in tissues or organ receiving high-dose radiation (in-field doses) . However, the exact mechanism and dose–response relationship for radiation-induced malignancy, for both in-field and out-of-field doses, are not well understood; thus, it is necessary to investigate how radiotherapy, photons, and protons impact carcinogenic risk in childhood cancer management . For instance, the therapeutic use of proton beams has the potential to provide a better depth–dose profile and remarkable reduction in the dose to the adjacent normal tissues compared with photon beams . Additionally, there is growing evidence supporting an increased risk for late adverse non-cancer conditions , including cardiac and vascular effects. Thus, late adverse effects of radiotherapy have been observed on large vessels, causing cerebrovascular and cardiovascular diseases. Nowadays, there is also evidence for a significant elevation of cancer risk in patients with acquired as well as congenital heart diseases (CHD) in response to repeated radiological exposures . However, more data are needed to better define the “malignant price of cardiac care” . The risk estimates of long-term health effects of low doses of IR (cancer and non-cancer) are still incomplete, particularly for pediatric patients. Large patient cohorts, extended follow-up, validated clinical data, and reliable dosimetry for the cohorts are needed to address this challenge. The integration of epidemiological and biological research through panels of biomarkers, together with a mechanistic understanding of the cellular responses to a particular dose and radiation quality, will provide powerful means to improve risk estimates, leading to a better quantification of the magnitude of risks associated with low-dose exposures, e.g., for out-of-field organs. The Health Effects of Cardiac Fluoroscopy and Modern Radiotherapy in Pediatrics (HARMONIC) is a five-year project funded by the European Commission to improve understanding of the long-term health risks from medical ionizing radiation exposure in children and young patients ( https://harmonicproject.eu/ , accessed on 4 March 2023). The HARMONIC project uses an integrated approach of conventional epidemiology complemented by non-invasive imaging and molecular epidemiology to assess cancer and non-cancer outcomes in pediatric patients treated with modern radiotherapy techniques (such as proton therapy) for cancer and X-ray-guided interventional catheterization procedures for CHD. The purpose of the manuscript is to describe the bioanalytical research goals of the HARMONIC project, the rationale, and the study design, including the enrolment, endpoints, and expected results of the study. The general objective of the bioanalytical studies of the HARMONIC project is to provide a mechanistic understanding of the molecular pathways and the cellular responses that are triggered by the medical applications of IR in pediatric patients. A mechanistic understanding, together with the identification of biomarkers for individuals at increased risk to develop adverse health effects, has the potential to increase the power of epidemiological studies regarding health effects caused by IR . Such biomarkers may be useful in identifying susceptible individuals who are more vulnerable to radiation damage for whom individualized treatment can be considered (by radiation sparing policy or attempts to pharmacologic or dietary radioprotection). The specific aims are to: - identify radiation-induced biochemical responses in blood and saliva from pediatric patients exposed to medical IR; - evaluate dose–response relationships for different radiation qualities and delivery techniques with regards to specific biochemical responses; - search for pre-existing biomarkers of radiation sensitivity and health effects that may be useful for molecular epidemiological studies to identify patients with a potential higher risk of radiation-induced adverse health effects. The HARMONIC biological study is a prospective observational study which aims to investigate the biological changes induced by ionizing radiation exposure at various time points before and after exposure. It will focus on specific molecular biomarkers reported as ‘early signs’ of biological damage and long-term health effects. These include biomarkers of oxidative stress (8-hydroxy-2′-deoxyguanosine) , protein markers of inflammation (PTX3, IL-6, IL-10, TNF-α, NF-kB, MCP-1, etc.) , and genetic markers (telomere shortening and mtDNA copy numbers) . To decipher significant intracellular pathways and novel potential biomarkers involved in response to the radiation regimes applied, four different approaches will be used: multiplexed protein profiling assays on blood plasma , reverse-phase protein array (RPPA) on proteins isolated from peripheral blood mononuclear cells , miRNA transcriptome on whole blood and saliva , and liquid chromatography–mass spectrometry (LC–MS) on saliva. Finally, we will develop and implement new bioinformatic models to integrate the collected biological, clinical, and dosimetry data. that may be used in epidemiological and clinical approaches to identify patients at higher risk for radiation-induced adverse health effects, not only before starting (by analyzing the sample taken before exposure), but also after finishing the exposure (by analyzing samples taken after exposure). Biomarkers will be studied in both blood and saliva to investigate whether saliva can be used as a non-invasive sample to analyze biomarkers in large-scale molecular epidemiology studies. The overall strategy of the “biology” project is presented in . 3.1. Study Population This exploratory study will include 150 patients: 50 patients treated for cancer with proton therapy, 50 patients with photon therapy, and 50 patients treated with X-ray-guided interventional catheterization procedures for CHD. Specific inclusion and exclusion criteria are listed in . Ethics approval has been already obtained in all participating centers. All eligible patients received an information brochure and are invited to participate in the study by the responsible physician. Informed consent is signed by the patient or his/her legal representative before entering the study. Detailed demographic, clinical and treatment data are retrieved by the attending physician and from the patient’s medical electronic record. Data protection officers (DPOs) from each organization will be involved to ensure compliance with General Data Protection Regulations (GDPRs). 3.2. Biological Sample Collection For radiotherapy patients, blood and saliva will be collected at three time points: before radiotherapy; three months after the last fraction (time point for the first follow-up); and one year after completion of the treatment. For X-ray-guided interventional catheterization procedures, biological samples will also be collected at three time points: before intervention; the same day after completion; and one year after completion. summarizes the protocol for the collection, preparation, and storage of biosamples. Briefly, a maximum of 12 mL blood will be collected at each time point in three different tubes, which are as follows: - one BD vacutainer ® CPT™ tube for the isolation of lymphocytes (~4 mL); - one vacutainer tube containing EDTA K2 (~4 mL); - one clot activator serum separation tube (~4 mL). The samples will be given a unique patient identification number (pseudonymization). Within two hours post collection, tubes with blood samples will be centrifuged according to standard operating procedures (SOPs) prepared by the HARMONIC consortium to obtain lymphocytes, serum, and plasma. To investigate the possible impact of pre-analytical variables, we will record and share information on the study centers, time, and calendar days when the samples are collected. We will also record and share the time that elapses between the blood draw, centrifugation, and first freezing. Regarding saliva samples, approximately 4–5 mL of saliva will be collected at each time point in a sterile 10 mL plastic tube without any additive. Saliva samples will be divided into two aliquots, 2 mL in each. Aliquots of biological samples from each donor will be immediately stored at −80 °C . The type of tubes, volume, and number of each aliquots are summarized in . At specific time points in the project, coded samples will be shipped from a respective clinic on dry ice to a centralized Biobank, where they will be stored at −80 °C until use. 3.3. Biological Measures The samples will be analyzed by state-of-the-art methods to determine the levels of selected biomarkers. Moreover, the samples will be examined by innovative high-throughput approaches, including analyses of miRNA transcriptomes and proteomics . The handling and analytical procedures will follow the respective SOP procedures for all biologic sampling, handling, shipping, and analysis. This exploratory study will include 150 patients: 50 patients treated for cancer with proton therapy, 50 patients with photon therapy, and 50 patients treated with X-ray-guided interventional catheterization procedures for CHD. Specific inclusion and exclusion criteria are listed in . Ethics approval has been already obtained in all participating centers. All eligible patients received an information brochure and are invited to participate in the study by the responsible physician. Informed consent is signed by the patient or his/her legal representative before entering the study. Detailed demographic, clinical and treatment data are retrieved by the attending physician and from the patient’s medical electronic record. Data protection officers (DPOs) from each organization will be involved to ensure compliance with General Data Protection Regulations (GDPRs). For radiotherapy patients, blood and saliva will be collected at three time points: before radiotherapy; three months after the last fraction (time point for the first follow-up); and one year after completion of the treatment. For X-ray-guided interventional catheterization procedures, biological samples will also be collected at three time points: before intervention; the same day after completion; and one year after completion. summarizes the protocol for the collection, preparation, and storage of biosamples. Briefly, a maximum of 12 mL blood will be collected at each time point in three different tubes, which are as follows: - one BD vacutainer ® CPT™ tube for the isolation of lymphocytes (~4 mL); - one vacutainer tube containing EDTA K2 (~4 mL); - one clot activator serum separation tube (~4 mL). The samples will be given a unique patient identification number (pseudonymization). Within two hours post collection, tubes with blood samples will be centrifuged according to standard operating procedures (SOPs) prepared by the HARMONIC consortium to obtain lymphocytes, serum, and plasma. To investigate the possible impact of pre-analytical variables, we will record and share information on the study centers, time, and calendar days when the samples are collected. We will also record and share the time that elapses between the blood draw, centrifugation, and first freezing. Regarding saliva samples, approximately 4–5 mL of saliva will be collected at each time point in a sterile 10 mL plastic tube without any additive. Saliva samples will be divided into two aliquots, 2 mL in each. Aliquots of biological samples from each donor will be immediately stored at −80 °C . The type of tubes, volume, and number of each aliquots are summarized in . At specific time points in the project, coded samples will be shipped from a respective clinic on dry ice to a centralized Biobank, where they will be stored at −80 °C until use. The samples will be analyzed by state-of-the-art methods to determine the levels of selected biomarkers. Moreover, the samples will be examined by innovative high-throughput approaches, including analyses of miRNA transcriptomes and proteomics . The handling and analytical procedures will follow the respective SOP procedures for all biologic sampling, handling, shipping, and analysis. 4.1. 8-Hydroxy-2′-deoxyguanosine (8-oxo-dG) and Markers of Inflammation The levels of 8-oxo-dG in serum and saliva will be determined using an ELISA method where the samples are essentially purified by Bond Elute columns, as previously described . Briefly, 800 µL blood serum or saliva will be purified using a C18 solid phase Bond Elut extraction column. The purified samples will be freeze-dried and reconstituted in PBS. Then, 300 µL of the purified sample will be mixed with 150 µL of primary antibody against 8-oxo-dG and distributed in three wells of a 96-well ELISA plate pre-coated with 8-oxo-dG and then incubated at 37 °C for 120 min. Secondary antibody body will then be added followed by the staining solution in order to quantify the yield of secondary antibodies bounded to primary antibodies in each well using a 96-well automatic ELISA plate reader. Each sample will be analyzed in triplicate. A standard curve for 8-oxo-dG (0.05–10 ng/mL) will be established for each plate and the concentration of 8-oxo-dG in each sample will be calculated based on the standard curve. 4.2. Analysis of Telomere Length (TL) and mtDNA Copy Number (mtDNA-CN) TL and mtDNA-CN will be measured on DNA extracted from 200 µL of biological samples of blood and saliva samples by real-time PCR (CFX384 Touch™ Real-Time PCR System, Bio-Rad Life Sciences) according to standardized protocols . Briefly, TL will be measured in genomic DNA by determining the ratio of a telomere repeat copy number (T) to a single-copy gene (S) and copy number (T/S ratio). The relative telomere length will be calculated using the following formula “T/S ratio = 2−ΔΔCt”, where ΔCt = Ct telomere − a Ct single-copy gene. The T/S ratio reflects the average length of the telomeres across all leukocytes. For the quantification of mtDNA-CN, the NDI1 gene in the undeleted region for the reference sequence of mtDNA will be used as an internal control (mtNDI1) and human ß-globin gene of genomic DNA (gDNA) will be amplified by PCR in both gDNA and mtDNA. ΔCt values will be calculated from the difference between the Ct for the ß-globin gene and the Ct for the NDI1 gene and used to measure mtDNA-CN relative to gDNA. mtDNA-CN will be calculated using the (2ΔCt) method (ΔCt = Ct mtNDI1 − CtgDNA). 4.3. miRNA Profiling Analysis Total RNA will be isolated from 500 µL of blood and saliva samples using a RiboPure™-Blood Kit (ThermoFisher, Waltham, MA, USA) and a miRNeasy Serum/Plasma Kit (QIAGEN, Hilden, Germany), respectively, according to the manufacturer’s protocol . The expression profiling of miRNAs will be analyzed using the Illumina MiSeq platform. For each patient, we will carry out a small RNA sequencing experiment to characterize the different miRNA expression profiles in samples for each time point. Prepared libraries will be run on Miseq, and miRNA identification and dysregulated expression analyses will be performed using latest version of iMir software ( https://www.labmedmolge.unisa.it/italiano/home/imir , accessed on 5 September 2022), a fully automated workflow for the rapid analysis of high-throughput small RNA-Seq data. Specific dysregulated miRNAs will be further validated using qRT-PCR with sequence-specific TaqMan microRNA assays and a TaqMan Universal PCR Master Mix, as opposed to AmpErase UNG (Thermo Fisher Scientific, USA), in accordance with the manufacturer’s instructions. The miRNA expression levels will be normalized to the U6 small nuclear RNA and calculated using the ΔΔCt method . The target genes from differentially expressed miRNAs will be predicted using DIANA miRPath software (microrna.gr/mirpath, accessed on 5 September 2022). Then, the targets will then be further analyzed for gene ontology (GO) function enrichment terms (geneontology.org/, accessed on 5 September 2022), Kyoto Encyclopedia of Genes Genomes (KEGG) pathway classification ( www.genome.jp/kegg/ , accessed on 5 September 2022), and Reactome pathway databases ( www.reactome.org , accessed on 5 September 2022). 4.4. Plasma Protein Profiling Briefly, from the literature, a list of 90 protein markers that have previously been identified as potential markers of diseases related to the late effects of radiation exposure, particularly vascular diseases and secondary cancer, will be established. Aliquots of 100 µL plasma will be used for plasma proteome analysis using Olink’s affinity proteomics platform. The approach is based on paired antibodies, coupled to unique and partially complementary oligonucleotides, and measured by quantitative real-time PCR. This dual-recognition DNA-coupled method provides high specificity and sensitivity for an analysis of at least 90 proteins in parallel . We plan to analyze the 98 selected proteins in all samples from the three cohorts. Different statistical and computational models for single and multivariate analysis will then be used to identify the modified pathways. 4.5. Reverse-Phase Protein Arrays (RPPAs) Reverse-phase protein arrays (RPPAs) will allow us to study protein expression levels and the activation status of cell signaling pathways. Isolated peripheral blood mononuclear cells (PBMCs) from blood collected into CPT tubes will be analyzed by customized RPPA (Proteomics Unit. IBSAL. University of Salamanca). To summarize, the cells will be lysed and the protein extract will be serially diluted with a protein lysis buffer, supplemented with proteases and phosphatase inhibitors. Five serial dilutions/sample, ranging from 2000 to 125 µg/mL, and two technical replicates per dilution will be applied on the nitrocellulose microarray membrane. In addition, a few spike-in proteins as RPPAs, such as negative and positive controls, will be included. The membranes will be incubated with primary antibodies that target proteins of interest or without primary antibodies as negative controls. All primary antibodies for RPPA screening have been previously tested by Western blotting to assess their specificity and selectivity to the targeted protein. RPPA readout is a fluorescent signal correlated to the protein expression level. Samples will be applied on membranes in three technical replicates (spots), and the membranes will then be individually incubated with antibodies targeting one protein of interest, followed by an incubation with a fluorescent dye. Fluorescent signals will be acquired by a microarray scanner at high resolutions and minimal auto-fluorescent background. The NormaCurve method will be used for data quantification and normalization . This method includes a normalization for (i) background fluorescence, (ii) variations in the total amount of spotted protein; and (iii) spatial bias on the membranes. The normalized values will be employed to compare the protein expression levels across samples. Briefly, for each spot, the raw fluorescent signal of the proteins will be corrected with the fluorescent signal of the negative control (signal obtained after incubating an array, without the targeted protein antibody). This corrected signal will be divided by the total amount of spotted protein, corresponding to the normalized signal. Finally, the normalized signals of all the proteins will be scaled according to the median for further comparisons and statistical analysis. For each sample, one value will be generated for each targeted protein, and further statistical analysis will be considered. 4.6. Saliva Protein Analysis The saliva protein concentrations will be determined using a colorimetric protein assay (BCA Protein Assay Kit, Thermo Scientific, Waltham, MA, USA). The proteomic workflow is as follows. The iST-BCT Kit (PreOmics, Martinsried, Germany) will be used to perform a fast, reliable, and reproducible sample preparation on all the patient samples. Fifty microliters of saliva will be used as the starting material (the volume will be adjusted depending on the BCA result). Saliva proteins will be precipitated with 200 μL of ethanol at −20 °C overnight. Samples will then be centrifuged (at 17,000× g for 5 min at 4 °C) and the supernatants will be removed. Salivary protein pellets will be re-suspended, lysed, reduced, and alkylated in 10 min at 95 °C. Proteins will be digested in one hour. Generated peptides will be cleaned before LC-MS injection. Purified tryptic digests will be separated with a predefined 60 SPD method (21 min gradient time and 200 ng peptides) on an Evosep One LC system (Evosep, Odense, Demmark). A fused silica 10 μm ID emitter (Bruker Daltonics, Waltham, MA, USA) is placed inside a nanoelectrospray source (CaptiveSpray source, Bruker Daltonics, Waltham, MA, USA). The emitter is connected to a 8 cm × 150 μm reverse-phase column, packed with 1.5 μm C18 beads. Mobile phases will comprise water and acetonitrile, buffered with 0.1% formic acid. The column will be heated to 40 °C in an oven compartment. LC is coupled online to a TIMS Q-TOF instrument (timsTOF Pro 2, Bruker Daltonics) with a diaPASEF acquisition method. Samples will be acquired using a diaPASEF method, consisting of 12 cycles, including a total of 34 mass width windows (25 Da width, from 350 to 1200 Da) with 2 mobility windows each, leaving a total of 68 windows that cover the ion mobility range (1/K0) from 0.64 to 1.37 V s/cm 2 . Saliva proteins will be quantified using a label-free DIA approach with DIA-NN software ( https://github.com/vdemichev/diann , accessed on 4 March 2023). DIA-NN version 1.8 will be used first to build an in silico predicted library from the human FASTA database (NextProt 2022-02-25), enabling the ‘FASTA digest for library-free search/library generation’ and ‘Deep learning-based spectra’ options, as well as RTs and IMs prediction. The predicted library will be used to analyze the diaPASEF dataset. 4.7. Radiation Doses Data For each patient, the average whole-body dose or mean/maximum dose and non-target organ (out-of-field organ) doses will be estimated in collaboration with physicists responsible for dosimetry studies in the Harmonic project ( https://harmonicproject.eu , accessed on 4 May 2023). Briefly, the strategy for dose estimation will rely on Monte Carlo simulations for CHD patients and on treatment planning systems and analytical models for cancer patients . These strategies were benchmarked against measurements on physical phantoms and reference Monte Carlo simulations . As we are analyzing plasma proteins, the total dose to the blood will also be estimated and considered. 4.8. Integrative Analysis of Biological Function and Networks Integrative data analysis of multiple sets of data types will be performed to construct an interaction network of differentially expressed features (miRNAs and proteins) to elucidate the molecular mechanisms underlying the biological response to IR and to discover new potential biomarkers. In brief, each dataset from the different independent analyses (miRNA transcriptome sequencing and proteomics) will first be analyzed in relation to the available clinical parameters, e.g., age, sex, diagnosis, and background diseases, to identify significantly different features between the baseline and post-IR exposure responses. Then, the radiation-deregulated miRNAs and proteins will be analyzed by an integrative procedure using software, such as ingenuity pathway analysis (Qiagen Bioinformatics; Redwood City, CA, USA; www.qiagen.com/ingenuity , accessed on 5 March 2023), in order to identify the most significantly affected pathways, their components, and associated signaling networks. 4.9. Sample Size and Plan for Statistical Analysis This study is exploratory as the number of available patients is limited. With the use of data from our previous study on leukocyte telomere length , priori power analysis (Spearman’s correlation test) requires a sample size of 34 patients to achieve >80% power (alpha = 0.05) and to detect an effect size of 0.5 (GPower, version 3.1.9.2). We plan to include a target population of 100 patients from each cohort, considering the feasibility aspects of the study, the estimated level of recruitment in each participating clinical site, and a drop-out rate of 50%, aiming for a minimum of 50 participating patients in each cohort. Concerning the statistical plan, a database including clinical (diagnosis, different therapies, CT and MRI images, etc.), dosimetric, and experimental data for each patient will be created in order to perform appropriate statistical analysis. Descriptive data will be presented as frequencies with proportions for categorical variables, and either as means with corresponding SDs or medians with corresponding IQRs for continuous variables depending on the distribution. Statistical tests will include Pearson’s x 2 test for frequencies, the Mann–Whitney U test for non-normally distributed continuous variables, and Student’s t -test for normally distributed variables. Spearman’s correlation test will be used to explore the association between variables and radiation doses. Exploratory analysis, including unsupervised clustering and principal component analysis, will also be performed to stratify patients according to differential protein expression or relative protein abundance. Other soft clustering, dimensionality reduction (e.g., tSNE and UMAP), or unsupervised analyses (e.g., group-based trajectory models) will be conducted to identify groups of individuals that follow similar shifts or trends on protein relative abundance or differential protein expressions over time, considering that all time points will be performed. The differential profiles will be determined based on the data analysis for baseline and each timepoint, as well as changes from baseline values (relative and absolute change). A mixed-effects model will be used to study the association between different biomarkers at different time points and the dosimetry data (dose, volume, and beam quality). Lastly, the differences in biomarker levels between a follow-up timepoint and baseline will be studied as a function of dosimetric indicators using general linear models, with adjustment for potential confounders (chemotherapy, disease history, medicines, BMI, etc.). Statistical significance for all analyses will be assessed using two-sided tests with an alpha level of 0.05 and adjustment for multiple comparisons. The levels of 8-oxo-dG in serum and saliva will be determined using an ELISA method where the samples are essentially purified by Bond Elute columns, as previously described . Briefly, 800 µL blood serum or saliva will be purified using a C18 solid phase Bond Elut extraction column. The purified samples will be freeze-dried and reconstituted in PBS. Then, 300 µL of the purified sample will be mixed with 150 µL of primary antibody against 8-oxo-dG and distributed in three wells of a 96-well ELISA plate pre-coated with 8-oxo-dG and then incubated at 37 °C for 120 min. Secondary antibody body will then be added followed by the staining solution in order to quantify the yield of secondary antibodies bounded to primary antibodies in each well using a 96-well automatic ELISA plate reader. Each sample will be analyzed in triplicate. A standard curve for 8-oxo-dG (0.05–10 ng/mL) will be established for each plate and the concentration of 8-oxo-dG in each sample will be calculated based on the standard curve. TL and mtDNA-CN will be measured on DNA extracted from 200 µL of biological samples of blood and saliva samples by real-time PCR (CFX384 Touch™ Real-Time PCR System, Bio-Rad Life Sciences) according to standardized protocols . Briefly, TL will be measured in genomic DNA by determining the ratio of a telomere repeat copy number (T) to a single-copy gene (S) and copy number (T/S ratio). The relative telomere length will be calculated using the following formula “T/S ratio = 2−ΔΔCt”, where ΔCt = Ct telomere − a Ct single-copy gene. The T/S ratio reflects the average length of the telomeres across all leukocytes. For the quantification of mtDNA-CN, the NDI1 gene in the undeleted region for the reference sequence of mtDNA will be used as an internal control (mtNDI1) and human ß-globin gene of genomic DNA (gDNA) will be amplified by PCR in both gDNA and mtDNA. ΔCt values will be calculated from the difference between the Ct for the ß-globin gene and the Ct for the NDI1 gene and used to measure mtDNA-CN relative to gDNA. mtDNA-CN will be calculated using the (2ΔCt) method (ΔCt = Ct mtNDI1 − CtgDNA). Total RNA will be isolated from 500 µL of blood and saliva samples using a RiboPure™-Blood Kit (ThermoFisher, Waltham, MA, USA) and a miRNeasy Serum/Plasma Kit (QIAGEN, Hilden, Germany), respectively, according to the manufacturer’s protocol . The expression profiling of miRNAs will be analyzed using the Illumina MiSeq platform. For each patient, we will carry out a small RNA sequencing experiment to characterize the different miRNA expression profiles in samples for each time point. Prepared libraries will be run on Miseq, and miRNA identification and dysregulated expression analyses will be performed using latest version of iMir software ( https://www.labmedmolge.unisa.it/italiano/home/imir , accessed on 5 September 2022), a fully automated workflow for the rapid analysis of high-throughput small RNA-Seq data. Specific dysregulated miRNAs will be further validated using qRT-PCR with sequence-specific TaqMan microRNA assays and a TaqMan Universal PCR Master Mix, as opposed to AmpErase UNG (Thermo Fisher Scientific, USA), in accordance with the manufacturer’s instructions. The miRNA expression levels will be normalized to the U6 small nuclear RNA and calculated using the ΔΔCt method . The target genes from differentially expressed miRNAs will be predicted using DIANA miRPath software (microrna.gr/mirpath, accessed on 5 September 2022). Then, the targets will then be further analyzed for gene ontology (GO) function enrichment terms (geneontology.org/, accessed on 5 September 2022), Kyoto Encyclopedia of Genes Genomes (KEGG) pathway classification ( www.genome.jp/kegg/ , accessed on 5 September 2022), and Reactome pathway databases ( www.reactome.org , accessed on 5 September 2022). Briefly, from the literature, a list of 90 protein markers that have previously been identified as potential markers of diseases related to the late effects of radiation exposure, particularly vascular diseases and secondary cancer, will be established. Aliquots of 100 µL plasma will be used for plasma proteome analysis using Olink’s affinity proteomics platform. The approach is based on paired antibodies, coupled to unique and partially complementary oligonucleotides, and measured by quantitative real-time PCR. This dual-recognition DNA-coupled method provides high specificity and sensitivity for an analysis of at least 90 proteins in parallel . We plan to analyze the 98 selected proteins in all samples from the three cohorts. Different statistical and computational models for single and multivariate analysis will then be used to identify the modified pathways. Reverse-phase protein arrays (RPPAs) will allow us to study protein expression levels and the activation status of cell signaling pathways. Isolated peripheral blood mononuclear cells (PBMCs) from blood collected into CPT tubes will be analyzed by customized RPPA (Proteomics Unit. IBSAL. University of Salamanca). To summarize, the cells will be lysed and the protein extract will be serially diluted with a protein lysis buffer, supplemented with proteases and phosphatase inhibitors. Five serial dilutions/sample, ranging from 2000 to 125 µg/mL, and two technical replicates per dilution will be applied on the nitrocellulose microarray membrane. In addition, a few spike-in proteins as RPPAs, such as negative and positive controls, will be included. The membranes will be incubated with primary antibodies that target proteins of interest or without primary antibodies as negative controls. All primary antibodies for RPPA screening have been previously tested by Western blotting to assess their specificity and selectivity to the targeted protein. RPPA readout is a fluorescent signal correlated to the protein expression level. Samples will be applied on membranes in three technical replicates (spots), and the membranes will then be individually incubated with antibodies targeting one protein of interest, followed by an incubation with a fluorescent dye. Fluorescent signals will be acquired by a microarray scanner at high resolutions and minimal auto-fluorescent background. The NormaCurve method will be used for data quantification and normalization . This method includes a normalization for (i) background fluorescence, (ii) variations in the total amount of spotted protein; and (iii) spatial bias on the membranes. The normalized values will be employed to compare the protein expression levels across samples. Briefly, for each spot, the raw fluorescent signal of the proteins will be corrected with the fluorescent signal of the negative control (signal obtained after incubating an array, without the targeted protein antibody). This corrected signal will be divided by the total amount of spotted protein, corresponding to the normalized signal. Finally, the normalized signals of all the proteins will be scaled according to the median for further comparisons and statistical analysis. For each sample, one value will be generated for each targeted protein, and further statistical analysis will be considered. The saliva protein concentrations will be determined using a colorimetric protein assay (BCA Protein Assay Kit, Thermo Scientific, Waltham, MA, USA). The proteomic workflow is as follows. The iST-BCT Kit (PreOmics, Martinsried, Germany) will be used to perform a fast, reliable, and reproducible sample preparation on all the patient samples. Fifty microliters of saliva will be used as the starting material (the volume will be adjusted depending on the BCA result). Saliva proteins will be precipitated with 200 μL of ethanol at −20 °C overnight. Samples will then be centrifuged (at 17,000× g for 5 min at 4 °C) and the supernatants will be removed. Salivary protein pellets will be re-suspended, lysed, reduced, and alkylated in 10 min at 95 °C. Proteins will be digested in one hour. Generated peptides will be cleaned before LC-MS injection. Purified tryptic digests will be separated with a predefined 60 SPD method (21 min gradient time and 200 ng peptides) on an Evosep One LC system (Evosep, Odense, Demmark). A fused silica 10 μm ID emitter (Bruker Daltonics, Waltham, MA, USA) is placed inside a nanoelectrospray source (CaptiveSpray source, Bruker Daltonics, Waltham, MA, USA). The emitter is connected to a 8 cm × 150 μm reverse-phase column, packed with 1.5 μm C18 beads. Mobile phases will comprise water and acetonitrile, buffered with 0.1% formic acid. The column will be heated to 40 °C in an oven compartment. LC is coupled online to a TIMS Q-TOF instrument (timsTOF Pro 2, Bruker Daltonics) with a diaPASEF acquisition method. Samples will be acquired using a diaPASEF method, consisting of 12 cycles, including a total of 34 mass width windows (25 Da width, from 350 to 1200 Da) with 2 mobility windows each, leaving a total of 68 windows that cover the ion mobility range (1/K0) from 0.64 to 1.37 V s/cm 2 . Saliva proteins will be quantified using a label-free DIA approach with DIA-NN software ( https://github.com/vdemichev/diann , accessed on 4 March 2023). DIA-NN version 1.8 will be used first to build an in silico predicted library from the human FASTA database (NextProt 2022-02-25), enabling the ‘FASTA digest for library-free search/library generation’ and ‘Deep learning-based spectra’ options, as well as RTs and IMs prediction. The predicted library will be used to analyze the diaPASEF dataset. For each patient, the average whole-body dose or mean/maximum dose and non-target organ (out-of-field organ) doses will be estimated in collaboration with physicists responsible for dosimetry studies in the Harmonic project ( https://harmonicproject.eu , accessed on 4 May 2023). Briefly, the strategy for dose estimation will rely on Monte Carlo simulations for CHD patients and on treatment planning systems and analytical models for cancer patients . These strategies were benchmarked against measurements on physical phantoms and reference Monte Carlo simulations . As we are analyzing plasma proteins, the total dose to the blood will also be estimated and considered. Integrative data analysis of multiple sets of data types will be performed to construct an interaction network of differentially expressed features (miRNAs and proteins) to elucidate the molecular mechanisms underlying the biological response to IR and to discover new potential biomarkers. In brief, each dataset from the different independent analyses (miRNA transcriptome sequencing and proteomics) will first be analyzed in relation to the available clinical parameters, e.g., age, sex, diagnosis, and background diseases, to identify significantly different features between the baseline and post-IR exposure responses. Then, the radiation-deregulated miRNAs and proteins will be analyzed by an integrative procedure using software, such as ingenuity pathway analysis (Qiagen Bioinformatics; Redwood City, CA, USA; www.qiagen.com/ingenuity , accessed on 5 March 2023), in order to identify the most significantly affected pathways, their components, and associated signaling networks. This study is exploratory as the number of available patients is limited. With the use of data from our previous study on leukocyte telomere length , priori power analysis (Spearman’s correlation test) requires a sample size of 34 patients to achieve >80% power (alpha = 0.05) and to detect an effect size of 0.5 (GPower, version 3.1.9.2). We plan to include a target population of 100 patients from each cohort, considering the feasibility aspects of the study, the estimated level of recruitment in each participating clinical site, and a drop-out rate of 50%, aiming for a minimum of 50 participating patients in each cohort. Concerning the statistical plan, a database including clinical (diagnosis, different therapies, CT and MRI images, etc.), dosimetric, and experimental data for each patient will be created in order to perform appropriate statistical analysis. Descriptive data will be presented as frequencies with proportions for categorical variables, and either as means with corresponding SDs or medians with corresponding IQRs for continuous variables depending on the distribution. Statistical tests will include Pearson’s x 2 test for frequencies, the Mann–Whitney U test for non-normally distributed continuous variables, and Student’s t -test for normally distributed variables. Spearman’s correlation test will be used to explore the association between variables and radiation doses. Exploratory analysis, including unsupervised clustering and principal component analysis, will also be performed to stratify patients according to differential protein expression or relative protein abundance. Other soft clustering, dimensionality reduction (e.g., tSNE and UMAP), or unsupervised analyses (e.g., group-based trajectory models) will be conducted to identify groups of individuals that follow similar shifts or trends on protein relative abundance or differential protein expressions over time, considering that all time points will be performed. The differential profiles will be determined based on the data analysis for baseline and each timepoint, as well as changes from baseline values (relative and absolute change). A mixed-effects model will be used to study the association between different biomarkers at different time points and the dosimetry data (dose, volume, and beam quality). Lastly, the differences in biomarker levels between a follow-up timepoint and baseline will be studied as a function of dosimetric indicators using general linear models, with adjustment for potential confounders (chemotherapy, disease history, medicines, BMI, etc.). Statistical significance for all analyses will be assessed using two-sided tests with an alpha level of 0.05 and adjustment for multiple comparisons. The HARMONIC project addresses a crucial question regarding the health risks for pediatric patients exposed to ionizing radiation from radiotherapy or interventional cardiology (UNSCEAR report 2008) . Over the recent years, there has been considerable technological advancements that improve the therapeutic gain of radiotherapy, i.e., maximizing the dose to the tumor while sparing the healthy tissue, , as well as implementing numerous dose reduction strategies in pediatric interventional cardiology . The difficulties associated with cancer and non-cancer risk assessments from pediatric radiation exposure could be partly overcome with precise dose estimation and biochemical studies to better understand the mechanisms that underly the development of disease processes and provide indicators of risk . Radiation can induce DNA damage, especially DNA double-strand breaks (DSBs), which are the most lethal type of DNA damage and can result in mutations, chromosomal abnormalities, and the further development of cancer, as well as other severe health effects. However, the full spectra of biological mechanisms underlying the adverse health effects after irradiation are only partly understood. It has been reported that several biological pathways, such as DNA repair, inflammatory response, oxidative stress induction, as well as metabolic changes, are involved in response to IR exposure . A multifactorial approach in conjunction with robust high-throughput technologies and integrated computational approaches, such as systems biology, may facilitate the discovery of new pathways and biomarkers for the prediction of long-term adverse health effects of IR exposure . Accordingly, biological research on the HARMONIC project will investigate the changes induced by medical radiation at the level of specific biomarkers which might be considered ‘early signs’ of tissue damage before the full development of adverse health effects. The project will focus on changes related to oxidative stress , inflammation , as well as nuclear and mitochondrial DNA damage , in order to identify the long-term health risks . In parallel, the study will also use high-resolution approaches of proteomics and whole miRNA transcriptomes to provide an integrated assessment of bio-molecular responses to pediatric radiation exposure by identifying biological pathways that may underlie the adverse health effects of the exposures. Of note, this study has a longitudinal design which will allow the shorter-term and longer-term biological response of IR to be compared. To account for the differences between the participants, we will anchor the data of each individual on their own baseline data. This will allow us to measure the treatment effects on an individual level immediately and up to 1 year later. An additional aspect of this study is the comparison of biomarkers in saliva with those in blood samples. Especially in vulnerable populations, such as children , saliva offers an attractive non-invasive sampling method that is relatively inexpensive, safe, and easy to use. Saliva could be a valuable alternative as a biological source for human biomonitoring in occupational and environmental medicine , but further studies are needed to explore the robustness, reproducibility, and validity of salivary biomarkers in comparison to those analyzed in blood. As the study includes a cohort of pediatric patients that will be well characterized in terms of dosimetry to “in-field” and “out-of-field” organs ( https://harmonicproject.eu , accessed on 4 March 2023), it will be possible to investigate dose–response relationships between biomarkers and doses to organs for both radiotherapy and interventional cardiology, as well as a comparison of effects of radiation quality. The HARMONIC databases will register the individual responses for a defined set of biomarkers and facilitate studies on the mechanisms that underly radiation-induced second/primary cancers, as well as cardiac and vascular damages. A mechanistic understanding, together with biomarkers for individuals at increased risk, has the potential to increase the power of epidemiological studies regarding the health effects of different radiotherapy modalities. Hence, we believe that a better understanding of the underlying biological and cellular mechanisms will complement the epidemiological approach of the HARMONIC project ( https://harmonicproject.eu/ , accessed on 4 March 2023). This will provide a unique opportunity to gain better insight into the biological effects of medical radiation doses in pediatric patients. In summary, the findings of this research project hold the potential to provide mechanistic insight in the molecular and cellular responses involved in the effects of IR of different radiation quality and doses. The HARMONIC project aims to improve the protection of patients and maximize the benefits from medical applications. The final expected output from the biological part of the project will be able to define the predictive biomarkers to be used for molecular epidemiology studies in order to identify patients at higher risk for adverse health effects.

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.