id
stringlengths 10
11
| text
stringlengths 0
185k
| title
stringlengths 0
273
| date
stringlengths 0
10
| authors
stringlengths 0
356
| language
stringclasses 2
values |
|---|---|---|---|---|---|
PMC7038347 | sensors
Article
Indoor Trajectory Reconstruction of Walking, Jogging,
and Running Activities Based on a Foot-Mounted
Inertial Pedestrian Dead-Reckoning System
Jesus D. Ceron 1
, Christine F. Martindale 2
, Diego M. López 1,*
, Felix Kluge 2
and
Bjoern M. Eskofier 2,*
1
Telematics Engineering Research Group, Telematics Department, Universidad Del Cauca (Unicauca),
Popayán 190002, Colombia; [email protected]
2
Machine Learning and Data Analytics Lab, Computer Science Department, Friedrich-Alexander University
Erlangen-Nürnberg (FAU), 91052 Erlangen, Germany; [email protected] (C.F.M.);
[email protected] (F.K.)
*
Correspondence: [email protected] (D.M.L.); bjoern.eskofi[email protected] (B.M.E.)
Received: 27 September 2019; Accepted: 1 November 2019; Published: 24 January 2020
Abstract: The evaluation of trajectory reconstruction of the human body obtained by foot-mounted
Inertial Pedestrian Dead-Reckoning (IPDR) methods has usually been carried out in controlled
environments, with very few participants and limited to walking. In this study, a pipeline for
trajectory reconstruction using a foot-mounted IPDR system is proposed and evaluated in two large
datasets containing activities that involve walking, jogging, and running, as well as movements
such as side and backward strides, sitting, and standing. First, stride segmentation is addressed using
a multi-subsequence Dynamic Time Warping method. Then, detection of Toe-Off and Mid-Stance
is performed by using two new algorithms. Finally, stride length and orientation estimation are
performed using a Zero Velocity Update algorithm empowered by a complementary Kalman filter.
As a result, the Toe-Off detection algorithm reached an F-score between 90% and 100% for activities
that do not involve stopping, and between 71% and 78% otherwise. Resulting return position
errors were in the range of 0.5% to 8.8% for non-stopping activities and 8.8% to 27.4% otherwise.
The proposed pipeline is able to reconstruct indoor trajectories of people performing activities that
involve walking, jogging, running, side and backward walking, sitting, and standing.
Keywords: trajectory reconstruction; stride segmentation; dynamic time warping; pedestrian
dead-reckoning
1. Introduction
Indoor positioning systems (IPS) enable the provision of several location-based services
such as home monitoring, rehabilitation, navigation for blind and visual impaired people, and
finding and rescuing people/firefighters in emergencies. IPSs can be divided into two approaches:
infrastructure-based and infrastructure-free [1,2]. Infrastructure-based IPS require the deployment of
devices in the indoor environment to calculate the position of the person. Among the technologies
used by this type of IPS are Wi-Fi [3], radio frequency identification (RFID) [4], Bluetooth [5],
ultra-wide band (UWB) [6], infrared [7], and video cameras [4]. Infrastructure-free IPS do not need the
deployment of devices and mainly use dead-reckoning algorithms. Those systems are called inertial
pedestrian dead-reckoning (IPDR) because they use body movement information measured by inertial
measurement units (IMU) to estimate a person’s position changes based on a previously estimated or
known position [2]. The sum of these changes of position allows the reconstruction of the person’s
trajectory [2]. An IMU usually consists of a triaxial accelerometer and gyroscope. Although some IMUs
Sensors 2020, 20, 651; doi:10.3390/s20030651
www.mdpi.com/journal/sensors
Sensors 2020, 20, 651
2 of 20
also incorporate a triaxial magnetometer, alterations of the magnetic field indoors make it unreliable
for indoor positioning [8].
The advantages of IPDR systems over infrastructure-based systems are generally lower cost, data
privacy, and ease of deployment. However, IPDR systems without correction suffer from severe drift,
as person displacement is often calculated by integrating acceleration data from the accelerometer
twice and integrating the rotational angle from the gyroscope. In consequence, intrinsic errors and IMU
noise are raised to the third power, making a person’s trajectory reconstruction by direct integration
without correction impractical [9–11].
The literature review done in this study is aimed at foot-mounted IMU IPDR systems that only use
the accelerometer and/or gyroscope. Foot-mounted IPDRs, together with a zero velocity update (ZUPT)
algorithm, have been the most widely and successful method used to mitigate the drift in trajectory
reconstruction [9]. We use only the accelerometer and gyroscope because in indoor environments,
different sources might produce alterations in the magnetic field that make the magnetometer readings
unreliable for trajectory reconstruction [8]. Most of the foot-mounted IPDR systems that only use
accelerometer and gyroscope data are based on trajectory reconstruction during normal walking.
Natural movements like avoiding obstacles, sitting, swinging legs, stopping, or performing activities
like jumping, jogging, or running have rarely been considered [9,10]. In consequence, the literature
review is focused on the foot-mounted IPDR systems that have reconstructed the trajectory of walking,
jogging, and/or running activities. Thus, only six studies met the inclusion criteria and are part of the
literature review. The foot-mounted IPDR systems are usually evaluated in closed-loop trajectories by
measuring the return position error (RPE). The RPE indicates the distance between the final position of
the person obtained by the system and the actual physical final position of the person at the end of the
trial [8].
Threshold-based and machine learning-based foot-mounted IPDR approaches have been proposed
to deal with walking and running activities [12–16]. Li et al. [12,13] proposed a threshold-based
stance-phase detector that consists of one footstep detector and two zero velocity detectors, one for
walking and another for running. The evaluation of the system was done with one pedestrian who
followed two closed-loop trajectories while walking and running. For the square-shape path (195.7 m),
the RPE was 0.24% for walking and 0.42% for running. For the eight-shape path (292.1 m), the RPE was
0.2% for walking and 1.01% for running. An adaptive zero-velocity detector that selects an optimal
threshold for zero-velocity detection depending on the movement (walking or running) of the person
was proposed by Wagstaff et al. [15]. This system was evaluated by five people who walked and ran a
distance of 130 m in an “L” shaped path. The RPE reported were 1% for walking and 3.24% for running.
Considering that zero-velocity detection using machine learning-based IPDR systems is free
of threshold-tuning, Wagstaff et al. proposed a method for zero-velocity detection by using a long
short-term memory neural network (LSTM) [16]. Five people walked and ran a 220-m “L” shaped
path. The RPE in walking was 0.49% and running 0.93%. Similarly, Ren et al. proposed a zero-velocity
detection algorithm based on HMM [14]. The system was evaluated by one person in an oval-shaped
sports field of 422 m. The RPE when walking and running was 0.6% and 1.61%, respectively.
The described works have obtained very high precision in the trajectory reconstruction of walking
and running. However, the systems were evaluated with very few participants, and the evaluated
trajectories involved continuous walking and running activities. Currently, trajectory reconstruction
methods in realistic scenarios—with several people, and considering walking, jogging, and running
strides—are still missing.
Physical activity classification and gait event detection are key components of the trajectory
reconstruction process using IPDR. Machine learning has played an important role in both topics.
In [17] it is shown how different machine learning-based algorithms are able to classify different
physical activities, including standing, sitting, walking, and running. Gait event detection has been
performed by using several machine learning algorithms such as deep learning [18], hidden Markov
models (HMM) [19,20] and neural networks [21,22].
Sensors 2020, 20, 651
3 of 20
The aim of the present work was to propose a pipeline for trajectory reconstruction using a
foot-mounted IPDR system able to reconstruct the trajectories of activities that involve walking, jogging,
and running strides as well as natural movements like stopping, standing, sitting, and side-walking.
This paper contributes to foot-mounted IPDR systems by (1) comprehensively evaluating the
trajectory reconstruction of activities that involve walking, jogging, and running strides including the
discrimination of natural activities such as stopping, sitting, and side-walking; and (2) evaluating two
algorithms for Toe-off and Mid-Stance detection during walking, jogging, and running strides adapted
from the ones proposed by Barth et al. [23].
The proposed pipeline is able to recognize walking, jogging and running strides and detect the
Toe-off and Mid-Stance events in each of them. With this information, a foot-mounted IPDR system is
able to reconstruct the person’s trajectory regardless of their gait speed. This allows the development
of new ambient assisted living applications in which indoor tracking is a ground technology as well as
the development of new applications for indoor sports.
2. Datasets
2.1. Unicauca Dataset
The objective of the Unicauca dataset was to evaluate the trajectory reconstruction of walking,
jogging, and running in similar settings as the state-of-the-art methods, which are usually evaluated in
close-loop trajectories and the activities performed by the participants include continuous walking,
jogging, or running. This dataset was collected at the University of Cauca, Popayán, Colombia.
Ten participants (mean age: 30 ± 3 years) walked, jogged, and ran a closed-loop P-shaped path of
approximately 150 m (Figure 1) with an IMU attached to the lateral side of the left shoe with a Velcro
strap (Figure 2).
Sensors 2019, 19, x FOR PEER REVIEW
3 of 20
The aim of the present work was to propose a pipeline for trajectory reconstruction using a foot-
mounted IPDR system able to reconstruct the trajectories of activities that involve walking, jogging,
and running strides as well as natural movements like stopping, standing, sitting, and side-walking.
This paper contributes to foot-mounted IPDR systems by (1) comprehensively evaluating the
trajectory reconstruction of activities that involve walking, jogging, and running strides including the
discrimination of natural activities such as stopping, sitting, and side-walking; and (2) evaluating two
algorithms for Toe-off and Mid-Stance detection during walking, jogging, and running strides
adapted from the ones proposed by Barth et al. [23].
The proposed pipeline is able to recognize walking, jogging and running strides and detect the
Toe-off and Mid-Stance events in each of them. With this information, a foot-mounted IPDR system
is able to reconstruct the person’s trajectory regardless of their gait speed. This allows the
development of new ambient assisted living applications in which indoor tracking is a ground
technology as well as the development of new applications for indoor sports.
2. Datasets
2.1. Unicauca Dataset
The objective of the Unicauca dataset was to evaluate the trajectory reconstruction of walking,
jogging, and running in similar settings as the state-of-the-art methods, which are usually evaluated
in close-loop trajectories and the activities performed by the participants include continuous walking,
jogging, or running. This dataset was collected at the University of Cauca, Popayán, Colombia. Ten
participants (mean age: 30 ± 3 years) walked, jogged, and ran a closed-loop P-shaped path of
approximately 150 m (Figure 1) with an IMU attached to the lateral side of the left shoe with a Velcro
strap (Figure 2).
Figure 1. Illustration of the path used for walking, jogging and running in the Unicauca dataset. It is
a “P” shaped path. The dotted red line represents the trajectory followed by one person, dotted black
lines show outer edges (walls) of the path, and the blue square shows the start and end point of the
trajectory.
The IMU was a Shimmer3 GSR+ (Shimmer Sensing, Dublin, Ireland). Acceleration (range: ±16 g)
and angular velocity (range: ±2000 dps) data were collected at a frequency of 200 Hz. Accelerometer
calibration consisted in leaving the sensor still for a few seconds lying on each of its 6 sides on a flat
surface. For gyroscope calibration, the sensor is rotated around the three axes. At the beginning of
each trial, the participant was asked to remain standing without moving the IMU for at least 10 s for
gyroscope bias calculation.
Figure 1. Illustration of the path used for walking, jogging and running in the Unicauca dataset. It
is a “P” shaped path. The dotted red line represents the trajectory followed by one person, dotted
black lines show outer edges (walls) of the path, and the blue square shows the start and end point of
the trajectory.
The IMU was a Shimmer3 GSR+ (Shimmer Sensing, Dublin, Ireland). Acceleration (range: ±16 g)
and angular velocity (range: ±2000 dps) data were collected at a frequency of 200 Hz. Accelerometer
calibration consisted in leaving the sensor still for a few seconds lying on each of its 6 sides on a flat
surface. For gyroscope calibration, the sensor is rotated around the three axes. At the beginning of
each trial, the participant was asked to remain standing without moving the IMU for at least 10 s for
gyroscope bias calculation.
Sensors 2020, 20, 651
4 of 20
Sensors 2019, 19, x FOR PEER REVIEW
4 of 20
(a)
(b)
Figure 2. IMU sensor placement and axis alignment. (a) Accelerometer. (b) Gyroscope.
2.2. FAU Dataset
The FAU dataset is based on a previous study evaluating a method for smart labeling of cyclic
activities [24] and is publicly available at www.activitynet.org. The dataset provides gait data in a
relatively natural setting, and its protocol consisted in the execution of 12 different task-driven
activities performed in random order for each participant. It includes data from 80 healthy
participants with a mean age of 27 ± 6 years. Data were collected from 56 participants at the Friedrich-
Alexander University Erlangen-Nürnberg (Germany) and from 24 participants at the University of
Ljubljana (Slovenia). In this study, data collected at Slovenia from 20 of the 24 participants (mean age
of 28 years) was used as training dataset [25] and data collected in Germany from the 56 participants
were used as evaluation dataset. Only the data collected from the IMU worn on the left foot was used
for trajectory reconstruction of ten activities (Table 1). Sensor placement and axis alignment are the
same used in the Unicauca dataset (Figure 2). The acceleration (range: ±8 g) and angular velocity
(range: ±2000 dps) were collected at a frequency of 200 Hz. The on-ground and off-ground phases of
each stride are labeled. The accelerometer was calibrated using six static positions and the gyroscope
was calibrated using a complete rotation about each of the three axes. Data were acquired in an indoor
environment which including chairs and tables (Figure 3). Jogging was described to the participants
as “if one would jog for exercise in the evening” and running as “if one is late for a bus”. These
instructions were the same used in the Unicauca dataset.
Figure 3. Map of the indoor environment used for collecting the FAU dataset. Blue squares represent
chairs that denote start/end positions of activities. Black rectangles represent tables, and dotted red
lines represent the possible trajectories followed by participants in each activity.
Figure 2. IMU sensor placement and axis alignment. (a) Accelerometer. (b) Gyroscope.
2.2. FAU Dataset
The FAU dataset is based on a previous study evaluating a method for smart labeling of cyclic
activities [24] and is publicly available at www.activitynet.org. The dataset provides gait data in a
relatively natural setting, and its protocol consisted in the execution of 12 different task-driven activities
performed in random order for each participant. It includes data from 80 healthy participants with
a mean age of 27 ± 6 years. Data were collected from 56 participants at the Friedrich-Alexander
University Erlangen-Nürnberg (Germany) and from 24 participants at the University of Ljubljana
(Slovenia). In this study, data collected at Slovenia from 20 of the 24 participants (mean age of 28 years)
was used as training dataset [25] and data collected in Germany from the 56 participants were used as
evaluation dataset. Only the data collected from the IMU worn on the left foot was used for trajectory
reconstruction of ten activities (Table 1). Sensor placement and axis alignment are the same used in the
Unicauca dataset (Figure 2). The acceleration (range: ±8 g) and angular velocity (range: ±2000 dps)
were collected at a frequency of 200 Hz. The on-ground and off-ground phases of each stride are
labeled. The accelerometer was calibrated using six static positions and the gyroscope was calibrated
using a complete rotation about each of the three axes. Data were acquired in an indoor environment
which including chairs and tables (Figure 3). Jogging was described to the participants as “if one
would jog for exercise in the evening” and running as “if one is late for a bus”. These instructions were
the same used in the Unicauca dataset.
Sensors 2019, 19, x FOR PEER REVIEW
4 of 20
(a)
(b)
Figure 2. IMU sensor placement and axis alignment. (a) Accelerometer. (b) Gyroscope.
2.2. FAU Dataset
The FAU dataset is based on a previous study evaluating a method for smart labeling of cyclic
activities [24] and is publicly available at www.activitynet.org. The dataset provides gait data in a
relatively natural setting, and its protocol consisted in the execution of 12 different task-driven
activities performed in random order for each participant. It includes data from 80 healthy
participants with a mean age of 27 ± 6 years. Data were collected from 56 participants at the Friedrich-
Alexander University Erlangen-Nürnberg (Germany) and from 24 participants at the University of
Ljubljana (Slovenia). In this study, data collected at Slovenia from 20 of the 24 participants (mean age
of 28 years) was used as training dataset [25] and data collected in Germany from the 56 participants
were used as evaluation dataset. Only the data collected from the IMU worn on the left foot was used
for trajectory reconstruction of ten activities (Table 1). Sensor placement and axis alignment are the
same used in the Unicauca dataset (Figure 2). The acceleration (range: ±8 g) and angular velocity
(range: ±2000 dps) were collected at a frequency of 200 Hz. The on-ground and off-ground phases of
each stride are labeled. The accelerometer was calibrated using six static positions and the gyroscope
was calibrated using a complete rotation about each of the three axes. Data were acquired in an indoor
environment which including chairs and tables (Figure 3). Jogging was described to the participants
as “if one would jog for exercise in the evening” and running as “if one is late for a bus”. These
instructions were the same used in the Unicauca dataset.
Figure 3. Map of the indoor environment used for collecting the FAU dataset. Blue squares represent
chairs that denote start/end positions of activities. Black rectangles represent tables, and dotted red
lines represent the possible trajectories followed by participants in each activity.
Figure 3. Map of the indoor environment used for collecting the FAU dataset. Blue squares represent
chairs that denote start/end positions of activities. Black rectangles represent tables, and dotted red
lines represent the possible trajectories followed by participants in each activity.
Sensors 2020, 20, 651
5 of 20
Table 1. Activity descriptions and abbreviations, shown with their relevant start and end points as
labeled in Figure 3 as well as approximated distances.
Activity
Description
Start/End Position
Approximated Distance (m)
W-Slalom
Walk slalom through 3 tables
B→B
31
W-Posters
Sign name on 5 posters on the wall
C→G
21
W-Tables
Perform task at 3 different tables while sitting
D→D
20
W-Cards
Perform task on a table while standing
E→E
6
W, J, R-20
Walk, jog, run 2 times 20 m
A→A
40
W, J, R-Circuit
Walk, jog and run half a circuit each
F, G→G,F
43
3. Methods
A trajectory reconstruction pipeline was carried out separately for each activity of both datasets
(Figure 4). This pipeline is based on previous work by Hannink et al. [26]. A type of activity classification
step was included. Toe-Off and Mid-Stance algorithms were modified in order to deal with non-walking
strides as well as a complementary filter added for stride length and orientation estimation.
Sensors 2019, 19, x FOR PEER REVIEW
5 of 20
Table 1. Activity descriptions and abbreviations, shown with their relevant start and end points as
labeled in Figure 3 as well as approximated distances.
Activity
Description
Start/End
Position
Approximated Distance
(m)
W-Slalom
Walk slalom through 3 tables
B→B
31
W-Posters
Sign name on 5 posters on the wall
C→G
21
W-Tables
Perform task at 3 different tables while
sitting
D→D
20
W-Cards
Perform task on a table while standing
E→E
6
W, J, R-20
Walk, jog, run 2 times 20m
A→A
40
W, J, R-
Circuit
Walk, jog and run half a circuit each
F, G→G,F
43
3. Methods
A trajectory reconstruction pipeline was carried out separately for each activity of both datasets
(Figure 4). This pipeline is based on previous work by Hannink et al. [26]. A type of activity
classification step was included. Toe-Off and Mid-Stance algorithms were modified in order to deal
with non-walking strides as well as a complementary filter added for stride length and orientation
estimation.
Figure 4. Pipeline for trajectory reconstruction for each activity.
3.1. Stride Segmentation
As shown by Zrenner et al., a threshold-based stride segmentation and a double integration with
the ZUPT algorithm performed better than other approaches based on stride time, foot acceleration,
and deep learning for calculating stride length in running using a foot-mounted IMU [27]. Thus,
multi-dimensional subsequence dynamic time warping (msDTW) and a double integration with
ZUPT were used as the stride segmentation and stride length and orientation estimation methods,
respectively, in this study [23].
msDTW is used to find a subsequence of continuous signal sequences similar to a given reference
pattern. In the context of stride segmentation, that pattern consists of a template of one stride. The
stride start was set to the negative peak before the swing phase and stride end to the negative peak
at the end of the stance phase (Figure 5a), according to the definition of stride given in [20]. Using
that template, msDTW looks for similarities in a movement sequence. msDTW has been shown to be
a robust method to segment strides from healthy, geriatric, and Parkinson’s patients using foot-
mounted IMUs [28].
3.1.1. Template Generation
A MatLab script was developed for template generation. It included two steps: interpolation and
averaging. Interpolation consisted of taking each stride and interpolating it to a fixed duration of 200
samples. After interpolation, the template was obtained by averaging, sample by sample, all the
strides. The templates for walking, jogging, and running were built using the 8724, 1688, and 1360
Figure 4. Pipeline for trajectory reconstruction for each activity.
3.1. Stride Segmentation
As shown by Zrenner et al., a threshold-based stride segmentation and a double integration with
the ZUPT algorithm performed better than other approaches based on stride time, foot acceleration,
and deep learning for calculating stride length in running using a foot-mounted IMU [27]. Thus,
multi-dimensional subsequence dynamic time warping (msDTW) and a double integration with
ZUPT were used as the stride segmentation and stride length and orientation estimation methods,
respectively, in this study [23].
msDTW is used to find a subsequence of continuous signal sequences similar to a given reference
pattern. In the context of stride segmentation, that pattern consists of a template of one stride. The stride
start was set to the negative peak before the swing phase and stride end to the negative peak at the end
of the stance phase (Figure 5a), according to the definition of stride given in [20]. Using that template,
msDTW looks for similarities in a movement sequence. msDTW has been shown to be a robust method
to segment strides from healthy, geriatric, and Parkinson’s patients using foot-mounted IMUs [28].
3.1.1. Template Generation
A MatLab script was developed for template generation. It included two steps: interpolation
and averaging. Interpolation consisted of taking each stride and interpolating it to a fixed duration of
200 samples. After interpolation, the template was obtained by averaging, sample by sample, all the
strides. The templates for walking, jogging, and running were built using the 8724, 1688, and 1360
walking, jogging, and running strides, respectively, of the training dataset. Unlike other studies, which
used only straight strides for building templates [23,28,29], the three templates were built with all the
strides of the activities. Thus, both straight and non-straight strides were included in the templates.
The swing-phase starts when the foot leaves the ground (Toe-Off) and ends when the heel strikes
the ground (Heel Strike). The portion of the gyroscope z-signal after Heel Strike (HS) describes the
Sensors 2020, 20, 651
6 of 20
stance-phase. A Mid-Stance (MS) event is defined as the part of the stance-phase when the signal
energy is zero [30].
walking, jogging, and running strides, respectively, of the training dataset. Unlike other studies,
which used only straight strides for building templates [23,28,29], the three templates were built with
all the strides of the activities. Thus, both straight and non-straight strides were included in the
templates.
The swing-phase starts when the foot leaves the ground (Toe-Off) and ends when the heel strikes
the ground (Heel Strike). The portion of the gyroscope z-signal after Heel Strike (HS) describes the
stance-phase. A Mid-Stance (MS) event is defined as the part of the stance-phase when the signal
energy is zero [30].
(a)
(b)
Figure 5. (a) Walking, jogging, and running templates (gyroscope z-axis). (b) Running stride example
(gyroscope z-axis).
3.1.2. Classification of Walking, Jogging, and Running Activities
In order to automatically select the walking, jogging, or running template that will be used in
the stride segmentation process, the machine learning algorithms included in the Matlab
Classification Learner app were trained using the activities of the training dataset. A window size of
200 samples (1 s of data) and an overlap of 100 samples were used for feature extraction. The features
extracted were velocity (by integrating accelerometer readings), angular velocity (by integrating
gyroscope readings) and energy of accelerometer and gyroscope axes. The most frequent value in the
result was chosen as the final classification. The evaluation was performed using ten-fold cross-
validation. As a result, the highest accuracy (98.1%) was achieved by the SVM classifier with a
polynomial kernel function of third-order.
3.1.3. Multi-Subsequence Dynamic Time Warping Implementation
The output of the stride segmentation based on msDTW is a set of segments [31]. Each segment
describes a possible stride. One issue using these resulting segments for trajectory reconstruction is
that often the end of a segment does not coincide with the start of the next segment even for
consecutive strides (Figure 6a). The solution to this issue is based on the Toe-Off (TO) detection,
which is described in the next section. Using the templates (Figure 5a), the first event detected in each
stride is TO. For this reason, TO was defined as the beginning of a stride. For consecutive strides, the
end of the stride corresponds with the beginning of the next stride (next TO), resulting in a stride
segmentation without “holes” (Figure 6b).
The precision and sensitivity of the stride segmentation using msDTW can be tuned using a
threshold. The threshold needed to detect a stride indicates the similarity between that stride and the
template used, that is, a large threshold indicates a large difference between the template and the
segmented stride [23]. Therefore, with a very small threshold, the number of false negatives strides
would increase, and a very large threshold would generate false positives strides. Thresholds from 0
to 100 in steps of 5 were tested on the training dataset. As a result, it was found that a fixed threshold
Figure 5. (a) Walking, jogging, and running templates (gyroscope z-axis). (b) Running stride example
(gyroscope z-axis).
3.1.2. Classification of Walking, Jogging, and Running Activities
In order to automatically select the walking, jogging, or running template that will be used in the
stride segmentation process, the machine learning algorithms included in the Matlab Classification
Learner app were trained using the activities of the training dataset. A window size of 200 samples (1 s
of data) and an overlap of 100 samples were used for feature extraction. The features extracted were
velocity (by integrating accelerometer readings), angular velocity (by integrating gyroscope readings)
and energy of accelerometer and gyroscope axes. The most frequent value in the result was chosen
as the final classification. The evaluation was performed using ten-fold cross-validation. As a result,
the highest accuracy (98.1%) was achieved by the SVM classifier with a polynomial kernel function
of third-order.
3.1.3. Multi-Subsequence Dynamic Time Warping Implementation
The output of the stride segmentation based on msDTW is a set of segments [31]. Each segment
describes a possible stride. One issue using these resulting segments for trajectory reconstruction is that
often the end of a segment does not coincide with the start of the next segment even for consecutive
strides (Figure 6a). The solution to this issue is based on the Toe-Off (TO) detection, which is described
in the next section. Using the templates (Figure 5a), the first event detected in each stride is TO. For this
reason, TO was defined as the beginning of a stride. For consecutive strides, the end of the stride
corresponds with the beginning of the next stride (next TO), resulting in a stride segmentation without
“holes” (Figure 6b).
The precision and sensitivity of the stride segmentation using msDTW can be tuned using a
threshold. The threshold needed to detect a stride indicates the similarity between that stride and
the template used, that is, a large threshold indicates a large difference between the template and the
segmented stride [23]. Therefore, with a very small threshold, the number of false negatives strides
would increase, and a very large threshold would generate false positives strides. Thresholds from 0 to
100 in steps of 5 were tested on the training dataset. As a result, it was found that a fixed threshold of 65
maximizes the F-score of the stride segmentation in walking, jogging, and running activities (Figure 7).
Sensors 2020, 20, 651
7 of 20
Sensors 2019, 19, x FOR PEER REVIEW
7 of 20
of 65 maximizes the F-score of the stride segmentation in walking, jogging, and running activities
(Figure 7).
(a)
(b)
Figure 6. (a) Result of stride segmentation with msDTW. (b) Final stride segmentation with TO
detection. Blue vertical lines depict TOs. Light blue rectangles are segmented strides.
Figure 6. (a) Result of stride segmentation with msDTW. (b) Final stride segmentation with TO
detection. Blue vertical lines depict TOs. Light blue rectangles are segmented strides.
Sensors 2019, 19, x FOR PEER REVIEW
7 of 20
of 65 maximizes the F-score of the stride segmentation in walking, jogging, and running activities
(Figure 7).
(a)
(b)
Figure 6. (a) Result of stride segmentation with msDTW. (b) Final stride segmentation with TO
detection. Blue vertical lines depict TOs. Light blue rectangles are segmented strides.
Figure 7.
Threshold choice for stride segmentation of walking, jogging, and running strides
using msDTW.
Sensors 2020, 20, 651
8 of 20
3.2. Toe-Off and Mid-Stance Detection
The previous algorithms for TO and MS detection [31] were modified in order to improve detection
accuracy in jogging and running. These modifications are described in this section. Both previous and
proposed algorithms use the signal of the gyroscope z-axis for TO and MS detection.
3.2.1. To Detection
At TO, the gyroscope z-axis describes a zero-crossing because of the ankle joint changes from
plantar flexion to a dorsal extension position in the sagittal plane [23]. The algorithm included in [31]
for TO detection consists of detecting the first zero-crossing in the gyroscope z-axis. Due to the abrupt
movements in jogging and running strides, in a few cases, a peak located at the beginning of the stride
causes a zero crossing. This would lead to a wrong TO detection (red circle in Figure 8). Consequently,
the adapted algorithm for TO detection (Algorithm 1) find the maximum peak of the signal and then
find the nearest zero crossing before it (blue circle in Figure 8). After the detection of all the TOs
that belong to the activity, all the portions corresponding from TO to TO are considered as strides
(Figure 6b). Considering that the stride time of walking strides is around one second [24], if one TO to
TO portion is greater than 2 s (400 samples), only the signal until 1.5 s was taken into account. This
often happens because the participant is standing still or sitting.
Algorithm 1: Toe-off (TO) detection algorithm.
1: xMP ← getMaximumPeak(stride)
2: xZC ← getZeroCrossings(stride(1 : xMP))
3: TO ← getNearestZCtoMP(xZC, xMP)
Sensors 2019, 19, x FOR PEER REVIEW
8 of 20
Figure 7. Threshold choice for stride segmentation of walking, jogging, and running strides using
msDTW.
3.2. Toe-Off and Mid-Stance Detection
The previous algorithms for TO and MS detection [31] were modified in order to improve
detection accuracy in jogging and running. These modifications are described in this section. Both
previous and proposed algorithms use the signal of the gyroscope z-axis for TO and MS detection.
3.2.1. To Detection
At TO, the gyroscope z-axis describes a zero-crossing because of the ankle joint changes from
plantar flexion to a dorsal extension position in the sagittal plane [23]. The algorithm included in [31]
for TO detection consists of detecting the first zero-crossing in the gyroscope z-axis. Due to the abrupt
movements in jogging and running strides, in a few cases, a peak located at the beginning of the
stride causes a zero crossing. This would lead to a wrong TO detection (red circle in Figure 8).
Consequently, the adapted algorithm for TO detection (Algorithm 1) find the maximum peak of the
signal and then find the nearest zero crossing before it (blue circle in Figure 8). After the detection of
all the TOs that belong to the activity, all the portions corresponding from TO to TO are considered
as strides (Figure 6b). Considering that the stride time of walking strides is around one second [24],
if one TO to TO portion is greater than 2 s (400 samples), only the signal until 1.5 s was taken into
account. This often happens because the participant is standing still or sitting.
Figure 8. Example of TO and MS detection. The red circle and square show a wrong TO and MS
detection, respectively, using the previous TO detection algorithm. The blue circle and square show
an adequate TO and MS detection, respectively, using the proposed algorithms.
Algorithm 1: Toe-off (TO) detection algorithm.
1: xMP ← getMaximumPeak(stride)
2: xZC ← getZeroCrossings(stride(1 : xMP))
3: TO ← getNearestZCtoMP(xZC, xMP)
3.2.2. Mid-Stance Detection
At Mid-Stance (MS) we define that the foot is entirely stationary on the ground [23,28] and its
velocity is zero. The gyroscope z-signal is minimal at that moment. As the speed of movement
increases from walking to running, the stance-phase time decreases (Figure 5a) making MS detection
more difficult [10]. The previous algorithm for MS detection in walking strides consists of calculating
the middle of the window with the lowest energy in the full stride’s gyroscope z-signal [23,28,31]. For
Figure 8. Example of TO and MS detection. The red circle and square show a wrong TO and MS
detection, respectively, using the previous TO detection algorithm. The blue circle and square show an
adequate TO and MS detection, respectively, using the proposed algorithms.
3.2.2. Mid-Stance Detection
At Mid-Stance (MS) we define that the foot is entirely stationary on the ground [23,28] and its
velocity is zero. The gyroscope z-signal is minimal at that moment. As the speed of movement
increases from walking to running, the stance-phase time decreases (Figure 5a) making MS detection
more difficult [10]. The previous algorithm for MS detection in walking strides consists of calculating
the middle of the window with the lowest energy in the full stride’s gyroscope z-signal [23,28,31].
For jogging and running strides, the MS is often confused with other parts of the signal like the valley
just before the HS or the peak before the next TO (red square in Figure 8).
Sensors 2020, 20, 651
9 of 20
The adaptation of the MS detection algorithm (Algorithm 2) consisted of (1) taking only the stride
portion from HS to 80% of the stride—this portion was chosen taking into account that the stance-phase
of walking strides is approximately the last 60% of the stride and for jogging and running strides it is
approximately the last 40% of the stride [25]; (2) calculating the middle of the window with the lowest
energy within that portion—to this end, a window size of 20 samples (100 ms) and a window overlap
of 10 samples (Blue square in Figure 8) are used.
Algorithm 2: Mid-Stance (MS) detection algorithm.
1: windowSize ← 20
2: overlap ← 10
3: stride ← interpolateStrideTo200Samples(stride)
4: xMP ← getMaximumPeak(stride)
5: stride ← stride(xMP : 160)
6: xHS ← getMinimumPeak(stride)
7: stride ← stride(xHS : end)
8: MS getMinimumEnergy(stride, windowSize, overlap)
3.2.3. Stride Length and Orientation Estimation
The biggest challenge to adequately estimate stride length using IMU data is the significant bias
derived from the use of IMUs, which leads to large drifts after the double-integration process. For that
reason, the ZUPT method was used. Zero-velocity detection was done by evaluating a threshold
on the magnitude of the gyroscope rate of turn of each measurement. If the measurement is less
than a threshold of 0.6 dps, that measurement is considered as a zero-velocity measurement. It has
been proved that this simple approach works properly in walking strides [11,30]. However, this
approach does not work correctly in jogging and running strides due to the abrupt signal variations.
The solution to this problem is the use of the MS detected previously. Taking into account that the
average stance-phase time in running strides is around 100 ms (20 samples), it was empirically found
that taking 5 samples to each side of the MS (which corresponds to 50 ms with the sampling frequency
used) leads to better zero-velocity detection in jogging and running strides.
After zero-velocity detection, a complementary Kalman filter (CF) was used in order to model
the error in velocity and position estimates using the ZUPTs as measurements (see Appendix A for
details). When zero-velocity is detected, but the estimated velocity is different to zero, the CF adjusts
the velocity and the corresponding displacement. The CF used in this work is based on the proposed
work by Fischer et al. [11]. Three main parameters have to be set up for CF initialization: accelerometer
and gyroscope noise (σa and σw) and the ZUPT detection noise (σv). Accelerometer and gyroscope
noise were set to equal value in both datasets (σa = 0.01 m/s2 and σw = 0.01 rad/s). ZUPT detection
noise depends on the velocity of the participant. That parameter was established by evaluating from
σv = 0.001 m/s to σv = 0.05 m/s in steps of 0.001 m/s for each trajectory performed. The σv chosen was
the one that produced the least error in the final distance evaluated. The stride length and orientation
estimation are obtained using the position increments in each MS event. Stride length, where ∇Pk is
the position increment from stride k-1 to stride k, is calculated as follows:
SLk =
q
∇Pk(x)2 + ∇Pk(y)2,
(1)
Sensors 2020, 20, 651
10 of 20
4. Results
4.1. Unicauca Dataset
4.1.1. Classification of the Type of Activity
The accuracy in the activity classification was 90%. There were only three misclassifications: two
running activities were classified as jogging activities and one jogging activity was classified as a
running activity (Figure 9).
Sensors 2019, 19, x FOR PEER REVIEW
10 of 20
4. Results
4.1. Unicauca Dataset
4.1.1. Classification of the Type of Activity
The accuracy in the activity classification was 90%. There were only three misclassifications: two
running activities were classified as jogging activities and one jogging activity was classified as a
running activity (Figure 9)
Figure 9. Confusion matrix of the classification of the type of activity in the Unicauca dataset.
4.1.2. Toe-Off and Mid-Stance Detection
In this dataset, TO and MS were manually labeled. A TO/MS is considered as a true positive (TP)
if it is located within 15% of the total number of samples of the stride to the right and left of the
TO/MS ground truth. A false positive (FP) occurs when a TO/MS is detected outside this range. A
false negative (FN) indicates that a TO/MS for a stride was not detected. Having in mind that 40%
and 60% of the stride corresponds to the stance-phase of walking and running strides, respectively
[25], the TO detection performance was evaluated in the training dataset using error ranges from 5%
to 21% of the total stride in steps of 3% (Figure 10). As a result, 15% was chosen as an acceptable error
range for TP calculation.
Results of the evaluation of the TO and MS detection using the previous and proposed
algorithms are shown in Tables 2 and 3, respectively.
Figure 10. TO performance evaluation using error ranges from 5% to 21% in steps of 3%.
Figure 9. Confusion matrix of the classification of the type of activity in the Unicauca dataset.
4.1.2. Toe-Off and Mid-Stance Detection
In this dataset, TO and MS were manually labeled. A TO/MS is considered as a true positive
(TP) if it is located within 15% of the total number of samples of the stride to the right and left of the
TO/MS ground truth. A false positive (FP) occurs when a TO/MS is detected outside this range. A false
negative (FN) indicates that a TO/MS for a stride was not detected. Having in mind that 40% and 60%
of the stride corresponds to the stance-phase of walking and running strides, respectively [25], the TO
detection performance was evaluated in the training dataset using error ranges from 5% to 21% of the
total stride in steps of 3% (Figure 10). As a result, 15% was chosen as an acceptable error range for
TP calculation.
Sensors 2019, 19, x FOR PEER REVIEW
10 of 20
4. Results
4.1. Unicauca Dataset
4.1.1. Classification of the Type of Activity
The accuracy in the activity classification was 90%. There were only three misclassifications: two
running activities were classified as jogging activities and one jogging activity was classified as a
running activity (Figure 9)
Figure 9. Confusion matrix of the classification of the type of activity in the Unicauca dataset.
4.1.2. Toe-Off and Mid-Stance Detection
In this dataset, TO and MS were manually labeled. A TO/MS is considered as a true positive (TP)
if it is located within 15% of the total number of samples of the stride to the right and left of the
TO/MS ground truth. A false positive (FP) occurs when a TO/MS is detected outside this range. A
false negative (FN) indicates that a TO/MS for a stride was not detected. Having in mind that 40%
and 60% of the stride corresponds to the stance-phase of walking and running strides, respectively
[25], the TO detection performance was evaluated in the training dataset using error ranges from 5%
to 21% of the total stride in steps of 3% (Figure 10). As a result, 15% was chosen as an acceptable error
range for TP calculation.
Results of the evaluation of the TO and MS detection using the previous and proposed
algorithms are shown in Tables 2 and 3, respectively.
Figure 10. TO performance evaluation using error ranges from 5% to 21% in steps of 3%.
Figure 10. TO performance evaluation using error ranges from 5% to 21% in steps of 3%.
Results of the evaluation of the TO and MS detection using the previous and proposed algorithms
are shown in Tables 2 and 3, respectively.
Sensors 2020, 20, 651
11 of 20
Table 2. Averaged results of TO and MS detection for the 10 participants in the Unicauca dataset using
the previous TO and MS detection algorithms.
Toe-Off
Mid-Stance
Activity
TO GT
TP
FP
FN
F-Score (%)
MS GT
TP
FP
FN
F-Score (%)
Walking
105.5
105.4
0.1
0.1
99.9
104.5
104.4
0.1
0.1
99.9
Jogging
75.4
39.4
37.1
36.2
51.5
74.4
41.2
34.2
33.3
54.9
Running
59.6
21.7
37.5
37.1
36.4
58.6
25.8
31.5
30.7
45.3
TO GT: ground truth TO rate. MS GT: ground truth MS rate. TP: true-positive rate. FP: false-positive rate. FN:
false-negative rate.
Table 3. Averaged results of TO and MS detection for the 10 participants in the Unicauca dataset using
the proposed TO and MS detection algorithms.
Toe-Off
Mid-Stance
Activity
TO GT
TP
FP
FN
F-Score (%)
MS GT
TP
FP
FN
F-Score (%)
Walking
105.5
105.5
0
0
100
104.5
104.5
0
0
100
Jogging
75.4
75.2
0.1
0.2
99.8
74.4
74.4
0
0
100
Running
59.6
59.3
0.3
0.2
99.7
58.6
58.5
0.1
0.1
99.8
TO GT: ground truth TO rate. MS GT: ground truth MS rate. TP: true-positive rate. FP: false-positive rate. FN:
false-negative rate.
A perfect F-score was obtained for TO and MS detection in walking strides. Very few mistakes
occurred for jogging and running, but the F-score remains high.
4.1.3. Trajectory Reconstruction
Two evaluation measures were used. (1) Return position error (RPE): the distance between the
coordinates of the actual final point of the activity and the coordinates of the participant’s final stride
of the corresponding activity. (2) Strides out of trajectory (SOT): All strides of the reconstructed
trajectory should be within the boundaries of the corridors represented by black dotted lines (Figure 11).
Otherwise, those strides will be counted as out of trajectory.
Higher velocity corresponds to more SOT and RPE. Although, on average, 5.7 % of the strides are
out of trajectory in the running trial, the RPE remains less than 1.0% (Table 4). Trajectories of the three
trials are mostly within the boundaries (Figure 11).
Table 4. Average results of trajectory reconstruction for each type of activity performed by the 10
participants using the previous and the proposed TO and MS detection algorithms.
Activity
SOT
RPE
[31]
New A
[31]
New A
#
%
#
%
meters
%
meters
%
Walking
1.7
1.6
1.7
1.6
0.8
0.5
0.8
0.5
Jogging
6.6
8.6
2.9
3.8
2.2
1.4
1.4
0.9
Running
5.3
9.2
3.3
5.7
2.6
1.6
1.4
0.9
SOT: strides out of trajectory, RPE: return position error, [31]: previous TO and MS detection algorithms, New A:
proposed TO and MS detection algorithms.
Sensors 2020, 20, 651
12 of 20
Sensors 2019, 19, x FOR PEER REVIEW
12 of 20
Figure 11. Trajectory reconstruction for the ten participants of the Unicauca dataset in a P shaped
path. Black dotted lines show outer edges (walls) of the possible path. Gray lines are the trajectories
reconstructed of the ten participants by using the proposed pipeline.
4.2. FAU Dataset
4.2.1. Classification of the Type of Activity
The accuracy obtained by the SVM classifier was 93%. Most of the misclassifications occurred
when classifying between running and jogging (Figure 12).
Figure 12. Confusion matrix of the classification of the type of activity classification in the FAU dataset
activities.
Figure 11. Trajectory reconstruction for the ten participants of the Unicauca dataset in a P shaped
path. Black dotted lines show outer edges (walls) of the possible path. Gray lines are the trajectories
reconstructed of the ten participants by using the proposed pipeline.
4.2. FAU Dataset
4.2.1. Classification of the Type of Activity
The accuracy obtained by the SVM classifier was 93%. Most of the misclassifications occurred
when classifying between running and jogging (Figure 12).
Sensors 2019, 19, x FOR PEER REVIEW
12 of 20
Figure 11. Trajectory reconstruction for the ten participants of the Unicauca dataset in a P shaped
path. Black dotted lines show outer edges (walls) of the possible path. Gray lines are the trajectories
reconstructed of the ten participants by using the proposed pipeline.
4.2. FAU Dataset
4.2.1. Classification of the Type of Activity
The accuracy obtained by the SVM classifier was 93%. Most of the misclassifications occurred
when classifying between running and jogging (Figure 12).
Figure 12. Confusion matrix of the classification of the type of activity classification in the FAU dataset
activities.
Figure 12. Confusion matrix of the classification of the type of activity classification in the FAU
dataset activities.
Sensors 2020, 20, 651
13 of 20
4.2.2. Toe-Off Detection
The last sample of the on-ground phase of each stride was used as ground truth for the evaluation
of the TO detection algorithm (Table 5). The same criteria used in the Unicauca dataset for TP, FP,
and FN calculations were used. The evaluation was carried out on the data collected from the 56
participants at the Friedrich-Alexander University Erlangen-Nürnberg (Germany) of FAU dataset.
Table 5. Average results of TO detection for each type of activity performed by the 56 participants in
the FAU dataset using the previous and the proposed TO detection algorithms.
Activity
TO
TP
FP
FN
F-Score (%)
[31]
New A.
[31]
New A.
[31]
New A.
[31]
New A.
W-Slalom
21.5
21.2
21.2
0.8
0.8
0.4
0.3
97
97
W-Posters
13.0
10.6
10.6
3.6
3.5
2.3
2.3
77
78
W-Tables
11.9
9.5
9.5
3.7
3.7
2.4
2.4
75
75
W-Cards
4.33
3.7
3.7
1.6
1.6
2.4
0.6
71
71
W-20
28.4
28.2
28.2
1.3
1.0
0.4
0.2
99
98
J-20
22.3
13.7
21.6
9.7
1.1
8.6
0.7
56
96
R-20
18.4
8.1
17.0
12.3
2.1
10.6
1.3
48
90
W-Circuit
28.2
27.6
27.7
0.7
0.7
0.4
0.3
98
98
J-Circuit
21.9
11.8
21.3
10.8
0.7
10
0.5
49
97
R-Circuit
17.7
7.6
17.3
10.7
0.8
10.2
0.4
40
96
TO: toe-off rate, TP: true positives rate, FP: false positives rate, FN: false negatives rate, [31]: previous TO and MS
detection algorithms, New A: proposed TO and MS detection algorithms.
4.2.3. Body Trajectory Reconstruction
For RPE estimation in FAU dataset (Table 6), it is important to note that the start/end activity
positions were defined by chairs in the indoor environment. For that reason, the actual positions
where the participants started and finished the activities were not precisely the same as the chairs’
positions since participants began each activity near the corresponding chair and did not necessarily
return to the exact point where they started the activity. Based on the videos of the data collection,
participants started and finished the activities within a radius of 1.5 m around the chairs. Light blue
and gray rectangles in Figures 13 and 14, respectively, indicate the path where all the strides related to
a certain activity should take place. If a stride is out of this path, it is considered as a Stride Out of
Trajectory (SOT). A SOT can be caused by the accumulative error of stride lengths and angle calculation
of previous strides. These zones were defined taking into account the coordinates of the chairs and
tables and the boundaries of the indoor environment.
Table 6. Averaged results of trajectory reconstruction of activities performed by the 56 participants in
the FAU dataset using the previous and the proposed TO and MS detection algorithms.
Activity
Activity
distance
SOT
RPE
[31]
New A.
[31]
New A.
meters
#
%
#
%
meters
%
meters
%
W-Slalom
31
1.1
5.2
1.1
5.2
1.7
5.5
1.7
5.5
W-Posters
21
1.0
8.0
1.0
8.0
1.9
9.0
1.8
8.8
W-Tables
20
3.1
25.9
3.1
25.9
2.8
14.1
2.8
14.1
W-Cards
6
1.3
30.5
1.3
30.5
1.6
27.4
1.6
27.4
W-20
40
3.7
13.1
3.7
13.1
1.7
4.2
1.7
4.2
J-20
40
7.5
34.2
3.9
17.9
5.5
14.2
2.0
5.1
R-20
40
4.1
22.5
3.1
17.0
5.2
13.9
2.5
6.0
W-Circuit
43
4.1
14.4
4.1
14.4
2.9
6.7
3.0
6.7
J-Circuit
43
6.4
30.2
4.5
20.4
14.5
33.7
3.6
8.8
R-Circuit
43
5.1
29.9
3.9
22.4
16.2
37.7
3.7
8.7
SOT: strides out of trajectory, RPE: return position error, [31]: previous TO and MS detection algorithms, New A:
proposed TO and MS detection algorithms.
Sensors 2020, 20, 651
14 of 20
of activities W-20, J-20, and R-20 describes two straight trajectories, joined by a 180-degree turn. The
trajectory reconstruction of W-Slalom allows sight of the area where the tables are located. The W-
Posters activity includes non-straight strides, which are well described in the trajectory obtained.
Regarding the circuit activities, although most of the strides are inside the activity zones, some
trajectories lead towards the outer part of the activity zone. Others lead towards the internal part of
the circuit (Figure 13).
Figure 13. Trajectory reconstruction of non-circuit activities for all 56 participants of the FAU dataset.
Black, blue and orange lines denote R-20, J-20, and W-20, respectively. Red, green, violet and light
green lines represent W-Cards, W-Slalom, W-Posters, and W-Tables, respectively. Gray rectangles
represent zones where all the strides related to certain activity should take place.
Figure 13. Trajectory reconstruction of non-circuit activities for all 56 participants of the FAU dataset.
Black, blue and orange lines denote R-20, J-20, and W-20, respectively. Red, green, violet and light green
lines represent W-Cards, W-Slalom, W-Posters, and W-Tables, respectively. Gray rectangles represent
zones where all the strides related to certain activity should take place.
Sensors 2019, 19, x FOR PEER REVIEW
15 of 20
Figure 14. Trajectory reconstruction of circuit activities for all 56 participants of the FAU dataset. Black
lines denote the trajectory follows by the participants. Gray zones represent the zone where all the
strides should take place.
5. Discussion
We have proposed a pipeline for indoor trajectory reconstruction of walking, jogging, and
running activities. The proposed pipeline was evaluated with two datasets. The results showed that
it is able to reconstruct a person’s trajectory regardless of their gait speed.
5.1. Classification of the Type of Activity
It was found that the classification model obtained with the SVM algorithm is able to classify the
three types of activities performed: walking, jogging, and running. The classification between jogging
and running is the one in which the classifier made more mistakes. This is possibly due to the jogging
and running speeds of some participants being similar. The use of personal models to avoid this
problem could be promising.
5.2. TO and MS Detection
Figure 14. Trajectory reconstruction of circuit activities for all 56 participants of the FAU dataset.
Black lines denote the trajectory follows by the participants. Gray zones represent the zone where all
the strides should take place.
Most of the trajectories were inside the zones (Figures 13 and 14). The trajectory reconstruction
of activities W-20, J-20, and R-20 describes two straight trajectories, joined by a 180-degree turn.
The trajectory reconstruction of W-Slalom allows sight of the area where the tables are located.
The W-Posters activity includes non-straight strides, which are well described in the trajectory obtained.
Regarding the circuit activities, although most of the strides are inside the activity zones, some
trajectories lead towards the outer part of the activity zone. Others lead towards the internal part of
the circuit (Figure 13).
Sensors 2020, 20, 651
15 of 20
5. Discussion
We have proposed a pipeline for indoor trajectory reconstruction of walking, jogging, and running
activities. The proposed pipeline was evaluated with two datasets. The results showed that it is able to
reconstruct a person’s trajectory regardless of their gait speed.
5.1. Classification of the Type of Activity
It was found that the classification model obtained with the SVM algorithm is able to classify
the three types of activities performed: walking, jogging, and running. The classification between
jogging and running is the one in which the classifier made more mistakes. This is possibly due to the
jogging and running speeds of some participants being similar. The use of personal models to avoid
this problem could be promising.
5.2. TO and MS Detection
Previous studies focused on the reconstruction of the trajectory during walking and running and
do not show results of segmentation or detection of strides [18–22]. The two datasets used in this study
allow TO evaluation. In the case of MS detection, ground truth information was not available in the
FAU dataset. Therefore, it was not possible to evaluate MS detection in that dataset. However, a high
F-score was obtained in the detection of MS in the Unicauca dataset.
While the F-score obtained for the proposed TO and MS detection algorithms is similar to that
obtained for the previous algorithms for walking activities, the F-score achieved for the proposed TO
and MS detection algorithms outperformed that achieved for the previous algorithms for all jogging
and running activities. That suggests that the proposed algorithms can detect those gait events in
walking, jogging, and running strides. The number of false positives (FP) was always higher than the
number of false negatives (FN). This could indicate that the threshold used for stride segmentation
with msDTW might have been overestimated, since stride segmentation using a large threshold implies
that there is a large difference between the template used and the segmented strides, leading to the
detection of FP strides. However, it was checked that by reducing that threshold, the number of
FN increased, causing a decrease in the F-score. Threshold-free methods based on machine learning
techniques such as those used by Ren [20] and Wagstaff [22] would make the stride segmentation
process straightforward by avoiding setting any threshold.
The lowest F-scores are obtained for three walking activities: W-Posters, W-Tables, and W-Cards,
which might be due to the fact that those activities involve non-stride movements such as stopping,
sitting, lateral and backward steps. This could be because the signal generated for those foot movements
is different from the walking/running templates. This could be accounted for by using templates
generated by those specific movements, as previously demonstrated in [29], where specific templates
were generated for each specific activity such as ascending and descending stairs. Unfortunately,
the wide range of possible natural foot movements makes this alternative hard to implement.
A hierarchical hidden Markov model (hHMM) approach has proved to be a robust method for stride
segmentation of walking activities that include non-stride movements in Parkinson’s patients [14] and
for stride segmentation of jogging activities [15]. Furthermore, hHMM is a threshold-free approach,
therefore it should be explored in order to improve the results obtained for the walking activities
that include non-stride movements such as W-Posters, W-Tables, and W-Cards, as well as for stride
segmentation of jogging and running activities.
5.3. Trajectory Reconstruction
Usually, the foot-mounted IPDR systems have been evaluated in closed-loop trajectories and by
measuring the Return Position Error (RPE) [18–22]. The purpose of the Unicauca dataset was, therefore,
to provide a starting point to allow a fair comparison with the state-of-the-art papers.
Sensors 2020, 20, 651
16 of 20
Sometimes the RPE is small, although the reconstructed trajectory does not fit the actual trajectory
performed by the person. That is why we proposed the number of strides out of the trajectory as an
additional evaluation metric. The RPEs obtained with the pipeline proposed in this paper for the three
trials collected in the Unicauca dataset are less than 1%. The results obtained by the works described
in the literature review section are also lower than 1%.
As a result of the better detection of TO and MS obtained by using the algorithms proposed in
this study, there is also a better trajectory reconstruction since there were fewer strides out of trajectory
(SOT) and shorter RPE for jogging and running activities. This demonstrates two things. The first is the
importance of performing a correct detection of TO and MS for trajectory reconstruction. The second is
that if the complementary filter does not have precise data to perform the ZVUs, it is not capable of
modeling errors in speed on its own, even if its parameters were tuned. It has also been demonstrated
that by properly detecting TO and MS, the complementary filter is capable of modeling errors in
walking, jogging, and running strides.
RPE obtained for trajectories in the FAU dataset are higher than for the Unicauca dataset. It is
important to highlight two limitations that the FAU dataset has for trajectory reconstruction. Firstly,
the position of the participants at the beginning and end of the activities is not exactly the same.
When analyzing the videos of the FAU dataset collection, it was concluded that these positions vary
approximately in a radius of one and a half meters, taking as reference the chairs that indicated the
start and end of the activities. Therefore, the RPEs calculated have an error of ±1.5 m. This fact should
be taken into account for the preparation of the protocol for the collection of a future dataset. Secondly,
it was not possible to subtract the gyroscope bias in all activities performed in the FAU dataset, because
the activities were performed continuously. A prerequisite for bias computation is that the person
stands still for a few seconds for the calculation of the mean of the gyroscope readings and then
subtracting it from the entire movement sequence.
The number of strides out of trajectory is directly related to the RPE obtained; the more strides out
of the acceptable path range, the higher the RPE. When observing the trajectory reconstruction of the
activities W-20, J-20, R-20, and W-Circuit, J-Circuit, R-Circuit, it appears that the difficulty in trajectory
reconstruction increases with stride velocity (from walking to jogging and running). This also occurred
in the five papers described in the literature review section [18–22]. In those papers, the evaluation was
performed with very few people. From our study, we can confirm that there is still a gap in trajectory
reconstruction using foot-mounted IPDR systems of jogging/running activities regarding the trajectory
reconstruction of walking activities.
The RPE of the trajectory reconstruction of W-Cards, W-Tables, and W-Posters activities are
particularly high, due to the bad detection of TOs. These activities should be treated with special care
in future works since they describe movements of daily living activities that happen frequently.
The trajectories obtained have a very well-defined shape and could be used for mapping an
indoor environment.
One important recommendation for future work in the field of trajectory reconstruction using
IPDR systems is that the datasets collected for evaluation are labeled at activity and stride/step levels,
as the FAU dataset used in this paper. Additionally, the participants of the data collection process must
start and end precisely at the indicated coordinates.
6. Conclusions
In this paper, we have proposed and evaluated a pipeline for trajectory reconstruction of walking,
jogging, and running activities using a foot-mounted inertial pedestrian dead-reckoning system.
The dynamic time warping method was adapted within this paper to segment walking, jogging, and
running strides. Stride length and orientation estimation were performed using a zero velocity update
algorithm adapted for walking, jogging, and running strides and empowered by a complementary
Kalman filter.
Sensors 2020, 20, 651
17 of 20
The presented results showed that the proposed pipeline provides good trajectory estimations
during walking, jogging, and running. TO detection algorithm reached an F-score between 92% and
100% for activities that do not involve stopping, and between 67% and 70% otherwise. Resulting return
distance errors were in the range of 0.51% to 8.67% for non-stopping activities and 8.79% to
27.36% otherwise.
To the best of the authors’ knowledge, this is the most comprehensive evaluation of a foot-mounted
IPDR system regarding the type and number of activities and quantity of people included in the
datasets and can serve as a baseline for the comparison of future systems. Future work will be focused
on using hidden Markov models in order to improve stride segmentation and fusing symbolic location
from an RSSI signal to update the indoor localization when possible.
Author Contributions: Conceptualization, J.D.C., C.F.M., D.M.L., F.K. and B.M.E.; Formal analysis, J.D.C.
and D.M.L.; Resources, C.F.M., F.K. and B.M.E.; Software, J.D.C.; Writing—original draft, J.D.C. and D.M.L.;
Writing—review & editing, C.F.M., F.K. and B.M.E. All authors have read and agreed to the published version of
the manuscript.
Funding: This research was funded by Departamento Administrativo de Ciencia, Tecnología e Innovación
(COLCIENCIAS), (Call 727-2015).
Acknowledgments: Jesus Ceron gratefully acknowledges the support of the Departamento Administrativo de
Ciencia, Tecnología e Innovación (COLCIENCIAS) within the national doctoral grants, call 727-2015. Bjoern Eskofier
gratefully acknowledges the support of the German Research Foundation (DFG) within the framework of the
Heisenberg professorship programme (grant number ES 434/8-1).
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to
publish the results.
Appendix A. Complementary Filter
The initialization of the Complementary Filter (CF) implies to establish a series of matrices. First,
the state of the CF includes the errors in orientation, position, and velocity. (A1) shows the state in an
array representation. Each array element is a 1 × 3 array containing the errors in the three-axis.
E =
h
Eo Ep Ev
i
(A1)
The error covariance matrix accumulates the error in orientation, position, and velocity produced
in each sample k:
Pk = [09x9]
(A2)
The state transition function is a matrix that is multiplied with the previous state to get the next
state, as shown in (A7). ‘S’ is the Skew-symmetric cross-product operator matrix formed from the
n-frame accelerations and is the time step equals to 0.005 s, which results from dividing 1 s between
the IMU data collection frequency (200Hz).
Fk =
I3X3
03x3
03x3
03x3
I3X3
I3X3∆t
−S∆t
03x3
I3X3
(A3)
The process noise covariance matrix is calculated for each sample by multiplying the accelerometer
and gyroscope noise by:
Qk =
h
σwx σwy σwz 0 0 0 σax σay σaz
∆t
i
(A4)
Sensors 2020, 20, 651
18 of 20
The uncertainty in velocity during each ZUPT is represented using the measurement noise
covariance matrix (A5). It is a diagonal matrix because no correlation in velocity is supposed to exist
between axes.
R =
σ2
vx
0
0
0
σ2
vy
0
0
0
σ2
vz
(A5)
The measurement function matrix is used to move from the state variables space to the measurement
variables states. In this implementation, the measurements are the ZUPTs that is when velocity is
supposed to be zero. That way, the measurement function has to contain an identity matrix in the
position of the velocity error state as follows:
Hk = [(03x3 03x3 I3x3)]
(A6)
Before running the CF, the gyroscope bias has to be removed. Gyroscope bias is obtained by
calculating the mean of the gyroscope readings while IMU is not moving just before the beginning of
the activity. The resulting value is subtracted to all gyroscope signals.
After gyroscope bias subtraction, the CF is executed. It has two phases: Prediction and update.
In the prediction phase, the error covariance matrix (Pk) is propagated using (A7):
Pk = FkPk−1FT
k + Qk
(A7)
Only when a sample k is a ZUPT, the Update phase comes into play. In this case, the Kalman gain
is calculated with (A8), and with that gain, the error is obtained using (A9).
Kk = PkHT(HPkHT + R)−1
(A8)
E =
h
Eo Ep Ev
i
= KkVk
(A9)
Finally, the velocity and position estimates are corrected as well as Pk:
Vk = Vk − Ev
(A10)
Posk = Posk − Ep
(A11)
Pk = (I9x9 − KkH)Pk
(A12)
References
1.
Zheng, L.; Zhou, W.; Tang, W.; Zheng, X.; Peng, A.; Zheng, H. A 3D indoor positioning system based on
low-cost MEMS sensors. Simul. Model. Pract. Theory 2016, 65, 45–56. [CrossRef]
2.
Susanti, R.M.; Adhinugraha, K.M.; Alamri, S.; Barolli, L.; Taniar, D. Indoor Trajectory Reconstruction Using
Mobile Devices. In Proceedings of the IEEE 32nd International Conference on Advanced Information
Networking and Applications (AINA), Krakow, Poland, 16–18 May 2018.
3.
Alarifi, A.; Al-Salman, A.; Alsaleh, M.; Alnafessah, A.; Al-Hadhrami, S.; Al-Ammar, M.A.; Al-Khalifa, H.S.
Ultra Wideband Indoor Positioning Technologies: Analysis and Recent Advances. Sensors 2016, 16, 707.
[CrossRef] [PubMed]
4.
Leong, C.Y.; Perumal, T.; Peng, K.W.; Yaakob, R. Enabling Indoor Localization with Internet of Things
(IoT). In Proceedings of the IEEE 7th Global Conference on Consumer Electronics (GCCE), Nara, Japan,
9–12 October 2018; pp. 571–573.
5.
Correa, A.; Barcelo, M.; Morell, A.; Vicario, J.L. A Review of Pedestrian Indoor Positioning Systems for Mass
Market Applications. Sensors 2017, 17, 1927. [CrossRef] [PubMed]
6.
Farid, Z.; Nordin, R.; Ismail, M. Recent Advances in Wireless Indoor Localization Techniques and System.
J. Comput. Netw. Commun. 2013, 2013, 185138. [CrossRef]
Sensors 2020, 20, 651
19 of 20
7.
Mainetti, L.; Patrono, L.; Sergi, I. A survey on indoor positioning systems. In Proceedings of the 22nd
International Conference on Software, Telecommunications and Computer Networks (SoftCOM), Split,
Croatia, 17–19 September 2014; pp. 111–120.
8.
Muhammad, M.N.; Salcic, Z.; Wang, K.I.-K. Indoor Pedestrian Tracking Using Consumer-Grade Inertial
Sensors with PZTD Heading Correction. IEEE Sens. J. 2018, 18, 5164–5172. [CrossRef]
9.
Harle, R. A Survey of Indoor Inertial Positioning Systems for Pedestrians. IEEE Commun. Surv. Tutor. 2013,
15, 1281–1293. [CrossRef]
10.
Wu, Y.; Zhu, H.B.; Du, Q.X.; Tang, S.M. A Survey of the Research Status of Pedestrian Dead Reckoning
Systems Based on Inertial Sensors. Int. J. Autom. Comput. 2019, 16, 65–83. [CrossRef]
11.
Fischer, C.; Sukumar, P.T.; Hazas, M. Tutorial: Implementing a pedestrian tracker using inertial sensors.
IEEE Pervasive Comput. 2012, 12, 17–27. [CrossRef]
12.
Li, Y.; Wang, J.J. A robust pedestrian navigation algorithm with low cost IMU. In Proceedings of
the International Conference on Indoor Positioning and Indoor Navigation (IPIN), Sydney, Australia,
13–15 November 2012; pp. 1–7.
13.
Li, Y.; Wang, J.J. A Pedestrian Navigation System Based on Low Cost IMU. J. Navig. 2014, 67, 929–949.
[CrossRef]
14.
Ren, M.; Pan, K.; Liu, Y.; Guo, H.; Zhang, X.; Wang, P. A Novel Pedestrian Navigation Algorithm for a
Foot-Mounted Inertial-Sensor-Based System. Sensors 2016, 16, 139. [CrossRef] [PubMed]
15.
Wagstaff, B.; Peretroukhin, V.; Kelly, J. Improving foot-mounted inertial navigation through real-time motion
classification. In Proceedings of the International Conference on Indoor Positioning and Indoor Navigation
(IPIN), Sapporo, Japan, 18–21 September 2017; pp. 1–8.
16.
Wagstaff, B.; Kelly, J. LSTM-Based Zero-Velocity Detection for Robust Inertial Navigation. In Proceedings
of the International Conference on Indoor Positioning and Indoor Navigation (IPIN), Nantes, France,
24–27 September 2018; pp. 1–8.
17.
Mannini, A.; Sabatini, A.M. Machine Learning Methods for Classifying Human Physical Activity from
On-Body Accelerometers. Sensors 2010, 10, 1154–1175. [CrossRef] [PubMed]
18.
Hannink, J.; Kautz, T.; Pasluosta, C.F.; Gasmann, K.G.; Klucken, J.; Eskofier, B.M. Sensor-Based Gait Parameter
Extraction with Deep Convolutional Neural Networks. IEEE J. Biomed. Health Inform. 2016, 21, 85–93.
[CrossRef] [PubMed]
19.
Ghassemi, N.H.; Hannink, J.; Martindale, C.F.; Gaßner, H.; Muller, M.; Klucken, J.; Eskofier, B.M. Segmentation
of Gait Sequences in Sensor-Based Movement Analysis: A Comparison of Methods in Parkinson’s Disease.
Sensors 2018, 18, 145. [CrossRef] [PubMed]
20.
Mannini, A.; Sabatini, A.M. Gait phase detection and discrimination between walking–jogging activities
using hidden Markov models applied to foot motion data from a gyroscope. Gait Posture 2012, 36, 657–661.
[CrossRef] [PubMed]
21.
Stetter, B.J.; Ringhof, S.; Krafft, F.C.; Sell, S.; Stein, T. Estimation of Knee Joint Forces in Sport Movements
Using Wearable Sensors and Machine Learning. Sensors 2019, 19, 3690. [CrossRef] [PubMed]
22.
Wouda, F.J.; Giuberti, M.; Bellusci, G.; Maartens, E.; Reenalda, J.; Van Beijnum, B.-J.F.; Veltink, P.H. Estimation
of Vertical Ground Reaction Forces and Sagittal Knee Kinematics During Running Using Three Inertial
Sensors. Front. Physiol. 2018, 9, 218. [CrossRef] [PubMed]
23.
Barth, J.; Oberndorfer, C.; Kugler, P.; Schuldhaus, D.; Winkler, J.; Klucken, J.; Eskofier, B.; Barth, J. Subsequence
dynamic time warping as a method for robust step segmentation using gyroscope signals of daily life
activities. In Proceedings of the 35th Annual International Conference of the IEEE Engineering in Medicine
and Biology Society (EMBC), Osaka, Japan, 3–7 July 2013; Volume 2013, pp. 6744–6747.
24.
Martindale, C.F.; Sprager, S.; Eskofier, B.M. Hidden Markov Model-Based Smart Annotation for Benchmark
Cyclic Activity Recognition Database Using Wearables. Sensors 2019, 19, 1820. [CrossRef] [PubMed]
25.
Martindale, C.F.; Roth, N.; Hannink, J.; Sprager, S.; Eskofier, B.M. Smart Annotation Tool for Multi-sensor
Gait-based Daily Activity Data. In Proceedings of the IEEE International Conference on Pervasive Computing
and Communications Workshops (PerCom Workshops), Athens, Greece, 19–23 March 2018; Institute of
Electrical and Electronics Engineers (IEEE): Piscataway, NJ, USA, 2018; pp. 549–554.
26.
Hannink, J.; Ollenschläger, M.; Kluge, F.; Roth, N.; Klucken, J.; Eskofier, B.M. Benchmarking Foot Trajectory
Estimation Methods for Mobile Gait Analysis. Sensors 2017, 17, 1940. [CrossRef] [PubMed]
Sensors 2020, 20, 651
20 of 20
27.
Zrenner, M.; Gradl, S.; Jensen, U.; Ullrich, M.; Eskofier, B.M. Comparison of Different Algorithms for
Calculating Velocity and Stride Length in Running Using Inertial Measurement Units. Sensors 2018, 18, 4194.
[CrossRef] [PubMed]
28.
Barth, J.; Oberndorfer, C.; Pasluosta, C.; Schülein, S.; Gassner, H.; Reinfelder, S.; Kugler, P.; Schuldhaus, D.;
Winkler, J.; Klucken, J.; et al. Stride Segmentation during Free Walk Movements Using Multi-Dimensional
Subsequence Dynamic Time Warping on Inertial Sensor Data. Sensors 2015, 15, 6419–6440. [CrossRef]
[PubMed]
29.
Leutheuser, H.; Doelfel, S.; Schuldhaus, D.; Reinfelder, S.; Eskofier, B.M. Performance Comparison of Two
Step Segmentation Algorithms Using Different Step Activities. In Proceedings of the 11th International
Conference on Wearable and Implantable Body Sensor Networks, Zurich, Switzerland, 16–19 June 2014;
pp. 143–148.
30.
Skog, I.; Nilsson, J.-O.; Händel, P. Evaluation of zero-velocity detectors for foot-mounted inertial navigation
systems. In Proceedings of the International Conference on Indoor Positioning and Indoor Navigation,
Zurich, Switzerland, 15–17 September 2010; pp. 1–6.
31.
Hannink, J. Mobile Gait Analysis: From Prototype towards Clinical Grade Wearable; FAU University Press:
Erlangen, Germany, 2019.
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
| Indoor Trajectory Reconstruction of Walking, Jogging, and Running Activities Based on a Foot-Mounted Inertial Pedestrian Dead-Reckoning System. | 01-24-2020 | Ceron, Jesus D,Martindale, Christine F,López, Diego M,Kluge, Felix,Eskofier, Bjoern M | eng |
PMC7538888 | Code is deposited in GitHub, see manuscript section “Code availability” for DOI.
Sample size was determined by available data during period Jul 2015
to Dec 2018 provided by Polar. This size is by far sufficient, as indicated by
smooth distribution functions.
Data exclusion criteria were pre-established. Data exclusion was applied to all available data from Polar Flow website.
Blinding was not relevant as this is not a clinical research study and differential assessment or treatment cannot occur since
individuals uploaded their data to webservice before our study started, without knowing about our analysis. We did not use any
personal information of individual besides their sex. All individuals and their activities were linked only by an anonymous ID.
| Human running performance from real-world big data. | 10-06-2020 | Emig, Thorsten,Peltonen, Jussi | eng |
PMC5919653 | S1 Data set and post‐hoc test results
Day 1
Performance time
Perceived exertion
Participant P1
P2
P3
P4
PE0 PE1 PE2 PE3 PE4 PE5 PE6 PE7
1
629
641 1727 1148
0.5
2
3
5
3
5
4
0
2
544
537 1364
935
0.5 0.5 0.5
5
3
4
3
0
3
917
615 1158 1178
1
3
3
5
5 5.5
6
5
4
720
834 1131 1185
3
5
8
8
6
7
6
4
5
806 1109 1648 1163
3 0.5
7
7
5
5
1
0
6
735
978
860
925
1
3
8
10
9
8
5
0
7 1081
467 1644 1139
0.5
2
7
8
7
7
7
0
8
693
479
884
913
0
0 0.5
10
8
7
6
0
9
440
407
679
701
0.5
1
3
6
4
3
3
2
10
736
695 1096 1587
2
3
5
7
7
5
4
2
11
465
353
778
871
0.5
1
2
8
6
5
3 0.5
12
529
354
664
667
1
1
3
10
8
8
6
1
13 1573
854 1427 1162
2
3
7
8
7
6
7 0.5
14
549
730
686
760
0 0.5
1
6
5
4
4
0
Day 2
1
596
589 1051 1079
0.5
1
3
7
5
6
5
0
2
468
466
790
894
1
2
7
6
5
4
4 0.5
3
548
592 1125 1539
3
4
6
9
6
5
4
1
4
582
614
998 1210
1
3
4
6
5
4
5
1
5
517
455
781 1075
0.5
2
3
6
4
4
3
2
6
423
511
619
770
0 0.5
2
10
9
7
5
0
7
662
608
967
966
0
1
3
9
7
4
3
0
8
613
462
849
926
0
0 0.5
8
7
7
5
0
9
376
325
595
607
0.5
1
4
8
4
4
2 0.5
10
841
546 1009 1097
2
3
6
7
6
5
5
3
11
387
409
586
659
0.5 0.5
3
7
6
5
3 0.5
12
465
327
840
601
1
1
2
9
7
6
4
1
13
561
797 1073 1115
3
4
6
8
8
7
6
1
14
404
316
743
611
0
1
1
7
5
5
4
0
1
Day 1
Cortisol
Participant C0
C1
C2
C3
C4
C5
C6
C7
1 2.738 2.760 2.965 3.206 3.207 3.107 2.939 2.906
2 2.481 2.690 2.742 2.924 3.041 2.850 2.735 2.807
3 2.446 2.814 2.891 3.053 2.963 2.850 2.857 2.598
4 2.276 2.404 2.712 2.907 2.958 2.927 2.835 2.634
5 2.300 2.649 2.998 3.126 3.051 3.057 3.025 2.840
6 2.989 3.099 3.109 3.189 3.262 3.189 3.210 3.160
7 3.086 3.021 3.085 3.154 3.160 3.132 3.168 3.108
8 3.008 3.077 3.162 3.222 3.255 3.179 3.182 3.169
9 2.958 3.005 3.005 2.947 2.991 3.144 2.992 3.159
10 3.017 3.054 3.072 3.056 3.207 3.116 3.111 3.089
11 2.496 2.763 2.902 2.923 3.078 2.945 2.933 2.855
12 2.406 2.633 2.779 2.993 3.029 2.869 2.904 2.682
13 2.504 2.635 3.067 2.985 3.034 3.086 2.922 2.884
14 2.568 2.656 2.861 2.882 2.957 2.842 2.797 2.721
Day 2
1 2.612 2.720 2.967 3.004 3.143 2.782 2.832 2.779
2 2.624 2.773 2.787 2.833 2.945 2.804 2.797 2.687
3 2.540 2.640 2.835 2.970 2.976 2.857 2.818 2.614
4 2.387 2.341 2.662 2.690 2.861 2.771 2.711 2.568
5 2.703 2.626 2.584 2.774 2.850 2.812 2.899 2.817
6 2.995 3.047 3.169 3.128 3.176 3.094 3.186 3.119
7 2.961 2.995 3.123 3.034 3.124 2.906 3.064 3.012
8 2.959 2.953 3.033 3.190 3.173 3.208 3.189 3.213
9 2.886 2.897 3.089 2.966 3.138 3.058 3.010 3.103
10 2.979 3.121 3.148 3.226 3.254 3.229 3.231 3.113
11 2.461 2.679 2.900 3.052 3.132 3.012 2.829 2.743
12 2.336 2.636 2.791 2.883 3.030 2.864 2.847 2.823
13 2.521 2.539 2.950 3.029 3.009 3.010 2.916 2.727
14 2.481 2.631 2.700 2.789 2.900 2.805 2.783 2.014
2
Day 1
Chromogranin A
Participant CgA0
CgA1
CgA2
CgA3
CgA4
CgA5
CgA6
CgA7
1
4.671
11.166
14.992 204.306
78.050
8.797
3.914
43.287
2
3.515
4.357
6.384
5.442
4.400
5.684
27.355
4.086
3
19.318
8.254
22.145 157.139 135.482 186.204 140.450
70.269
4 326.695 322.969 256.014 327.780 444.586 356.032 332.962 297.086
5
79.720 430.510 418.475 412.865 372.107 424.100 109.699 273.245
6
11.009 146.646
78.207
55.493 270.090
17.091
26.156
9.853
7
7.926
5.256
5.313
6.570
5.570
6.541
7.355
7.969
8
9.296
18.219
35.264 142.977
67.514
64.216
47.542 270.761
9 326.695 386.982 386.982 336.959 244.779
43.330
35.750 314.546
10
11.009
4.257
18.918
13.122 151.028
5.270
9.182
18.047
11
30.882
76.665 309.849 334.489 393.463 380.701 344.625
13.037
12 381.200
38.948
19.689 357.474 384.327 247.962 344.211 410.238
13
5.113 123.447
33.894
19.004 225.049 352.149
74.609
36.306
14 299.385
7.683
65.201 315.731
92.111 429.625 175.083 341.385
Day 2
1
11.024
10.624
4.143
7.041
10.353
49.398
12.094 253.659
2
3.757
5.470
4.143
4.185
7.069
4.285
4.114
7.812
3
32.823
32.138 281.211 108.472 222.951 309.192
85.373 153.298
4 170.187 284.809 311.034 325.281 352.377 334.875 220.324 200.837
5
32.095 317.872 374.533 406.769 417.547 438.390 217.697
59.762
6
81.833 308.536 359.187 390.580 281.097 149.858 320.213
6.784
7
4.014
5.770
4.185
9.667
15.963
17.862
4.228
6.912
8 107.658 214.028 152.185
97.965 118.793 129.186 140.735 189.088
9 406.455 362.770 430.710 285.037 392.022 312.704 312.090 410.152
10
7.912
22.958
76.907 305.666 300.712
17.805
5.185
23.372
11
21.559 398.874 356.831 387.724 390.209 458.819
7.840
13.351
12 227.848
21.688 194.784 367.852 381.686 240.153 331.220
45.300
13
3.929
3.543
20.803
32.266
36.107 197.368
15.178
7.026
14
88.657 265.322
29.782
29.854
89.813 197.725
75.680 381.557
3
Day 1
Memory
Visual attention
Participant M1
M2
M3
M4
VA0 VA1 VA2 VA3 VA4 VA7
1
8
10
3
4
11
18
14
22
16
25
2
7
8
4
8
13
20
18
22
18
20
3
10
10
9
8
17
26
24
21
22
26
4
7
10
7
7
11
17
22
29
22
24
5
3
4
1
2
15
18
17
19
17
27
6
6
5
4
4
11
15
21
19
20
21
7
7
6
2
4
14
18
24
22
21
20
8
6
7
6
8
11
14
14
19
15
17
9
4
7
4
7
20
17
19
20
23
20
10
3
5
7
9
17
10
13
17
14
20
11
6
5
5
4
24
22
24
23
25
20
12
6
7
7
9
16
21
26
26
26
34
13
4
8
5
3
15
12
20
17
21
22
14
5
5
6
5
17
21
17
21
16
22
Day 2
1
7
8
5
7
19
25
24
28
17
24
2
9
7
5
7
20
24
24
23
24
22
3
11
11
4
8
25
26
23
28
22
27
4
10
9
9
8
26
23
26
31
24
29
5
10
5
5
5
14
23
16
18
18
19
6
5
6
2
5
20
26
24
24
22
25
7
4
4
4
6
23
22
16
25
22
21
8
7
5
6
5
18
21
18
22
24
25
9
9
9
7
10
22
24
22
24
22
22
10
8
7
6
7
20
15
22
13
20
18
11
7
6
6
4
26
23
27
27
25
22
12
6
7
7
7
29
31
27
30
30
31
13
7
8
5
2
17
20
19
21
18
18
14
6
7
3
5
18
25
22
24
22
24
4
Day 1
Attention/mental flexibility
Functional psychobiosocial states
Participant AMF0 AMF1 AMF2 AMF3 AMF4 AMF7
FPBSS0 FPBSS1 FPBSS2 FPBSS3 FPBSS4 FPBSS5
1
61
58
50
44
55
44
1.57
1.57
1.14
0.86
1.00
2.00
2
60
55
54
56
38
41
1.00
1.14
1.57
3.57
3.14
2.71
3
65
42
38
45
42
39
2.14
2.43
2.14
2.14
2.14
2.43
4
61
61
49
36
62
58
2.57
2.71
2.71
2.57
3.14
2.43
5
70
65
65
52
65
60
1.29
2.00
1.43
1.86
2.29
1.43
6
60
70
64
54
61
47
2.43
2.43
2.29
2.29
2.71
2.29
7
63
55
60
60
49
53
1.86
1.71
1.14
1.14
1.00
1.71
8
55
59
49
44
37
64
2.00
2.14
2.14
2.43
2.71
3.00
9
60
58
50
59
58
48
2.29
2.43
2.43
2.86
2.29
3.14
10
63
60
49
52
47
62
1.14
1.43
0.86
1.00
0.86
0.86
11
63
60
48
44
42
53
2.14
1.71
1.86
1.71
2.14
2.43
12
49
43
37
32
30
30
2.29
2.57
2.43
2.14
2.14
2.14
13
67
69
63
65
61
58
3.29
3.57
3.14
3.14
2.86
3.29
14
66
64
64
66
54
73
3.00
2.14
2.57
2.14
3.29
1.86
Day 2
1
44
37
40
40
35
32
1.71
2.00
2.00
2.00
2.14
2.14
2
43
32
36
36
31
26
0.86
1.14
1.71
2.29
2.29
2.00
3
35
32
54
26
33
28
2.29
2.14
1.43
1.43
1.71
2.29
4
36
34
37
27
34
30
2.57
2.86
2.71
2.43
2.57
2.57
5
60
36
45
38
37
44
1.29
1.71
1.71
1.86
2.14
2.29
6
50
44
45
47
45
38
3.29
3.29
3.00
2.71
2.43
2.86
7
39
38
37
48
35
36
2.57
3.14
3.14
3.14
2.71
2.00
8
39
52
38
40
31
64
2.60
2.14
2.43
2.14
2.86
2.71
9
50
44
32
38
36
33
2.86
2.57
2.71
2.71
2.86
2.43
10
50
44
41
37
44
30
0.71
0.29
0.57
0.43
0.14
0.71
11
62
31
32
26
27
32
2.29
2.43
2.43
2.00
1.86
2.14
12
38
26
23
25
24
24
2.00
1.57
1.71
1.57
1.71
2.29
13
63
47
40
52
40
53
3.71
3.71
3.57
3.29
3.43
3.71
14
61
40
55
53
48
44
3.57
3.71
3.86
3.43
4.57
4.43
5
Day 1
Dysfunctional psychobiosocial states
FPBSS6 FPBSS7 Participant DPBSS0 DPBSS1 DPBSS2 DPBSS3 DPBSS4 DPBSS5 DPBSS6 DPBSS7
2.00
2.14
1
0.333
0.500
0.667
1.000
1.333
0.500
0.167
0.000
2.29
1.71
2
0.167
0.167
0.333
0.167
0.167
0.000
0.167
0.167
2.14
2.43
3
0.833
1.000
1.167
1.667
2.000
1.833
1.667
1.500
2.43
2.57
4
0.833
1.000
0.667
0.667
0.667
0.333
0.333
0.667
1.86
2.29
5
1.667
1.500
2.333
1.833
1.000
2.167
2.167
0.500
2.43
3.43
6
1.000
1.333
1.833
1.167
1.333
2.333
2.333
0.000
2.29
2.71
7
0.667
1.167
1.333
2.167
1.167
1.667
0.667
0.000
3.14
3.43
8
0.333
0.667
0.500
1.000
1.000
1.167
1.333
0.833
2.71
2.71
9
0.167
0.000
0.000
0.167
0.167
0.000
0.833
0.333
1.43
1.71
10
2.000
2.333
1.833
2.167
1.833
1.833
1.500
1.167
2.86
2.57
11
0.667
1.000
1.167
1.833
1.000
0.333
0.500
0.500
2.29
2.43
12
0.167
0.833
0.667
1.000
0.500
0.500
0.667
1.000
3.00
2.57
13
0.167
0.333
0.833
0.500
0.833
0.833
0.833
0.500
3.00
2.00
14
0.000
0.000
0.000
0.000
0.167
1.167
0.000
1.333
Day 2
2.00
2.29
1
0.500
0.500
0.333
0.500
0.500
0.500
0.500
0.167
1.71
1.14
2
0.000
0.000
0.167
0.833
0.000
0.000
0.333
0.167
2.14
2.43
3
1.667
1.667
1.500
1.500
1.333
0.500
0.833
0.833
2.71
2.71
4
0.500
1.000
0.667
0.333
0.667
0.833
0.333
0.667
2.00
2.00
5
0.500
0.333
0.167
0.667
0.167
0.000
0.500
0.333
2.71
2.14
6
0.167
0.333
0.333
0.833
0.167
0.500
0.667
1.667
2.71
3.00
7
0.167
0.167
0.667
0.500
0.833
1.167
1.000
0.500
2.43
3.29
8
0.250
0.333
0.333
1.500
0.833
0.667
0.667
0.333
2.57
2.14
9
0.167
0.167
0.167
0.167
0.167
0.167
0.167
0.000
0.71
0.86
10
2.000
2.833
2.500
3.167
3.000
2.833
2.333
2.167
2.14
2.14
11
0.167
0.333
0.167
0.500
1.000
0.667
0.500
0.667
2.29
2.29
12
0.500
0.833
1.500
1.333
1.167
0.667
0.500
0.333
3.29
3.14
13
0.167
0.167
0.333
0.500
0.667
0.667
0.500
0.167
3.57
3.43
14
0.000
0.000
0.000
0.000
0.000
0.000
0.500
0.000
6
Pairwise comparisons
Performance time Day 1
(I) Loop
(J) Loop
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
97.429
79.172
0.240
‐73.612
268.469
3
‐380.643
90.711
0.001
‐576.613
‐184.673
4
‐279.786
74.702
0.002
‐441.169
‐118.402
2
3
‐478.071
97.840
0.000
‐689.442
‐266.701
4
‐377.214
68.475
0.000
‐525.146
‐229.282
3
4
100.857
81.248
0.236
‐74.669
276.383
Performance time Day 2
(I) Loop
(J) Loop
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
30.429
33.145
0.375
‐41.178
102.035
3
‐327.357
33.702
0.000
‐400.167
‐254.548
4
‐407.571
59.712
0.000
‐536.572
‐278.571
2
3
‐357.786
32.706
0.000
‐428.442
‐287.129
4
‐438.000
52.061
0.000
‐550.471
‐325.529
3
4
‐80.214
43.384
0.087
‐173.939
13.511
1
Perceived exertion Day1
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.714
0.318
0.042
‐1.400
‐0.028
3
‐3.036
0.580
0.000
‐4.289
‐1.783
4
‐6.250
0.536
0.000
‐7.409
‐5.091
5
‐4.821
0.539
0.000
‐5.985
‐3.658
6
‐4.571
0.450
0.000
‐5.545
‐3.598
7
‐3.536
0.566
0.000
‐4.758
‐2.314
8
0.036
0.421
0.934
‐0.873
0.945
2
3
‐2.321
0.556
0.001
‐3.523
‐1.120
4
‐5.536
0.606
0.000
‐6.844
‐4.228
5
‐4.107
0.558
0.000
‐5.312
‐2.902
6
‐3.857
0.449
0.000
‐4.827
‐2.887
7
‐2.821
0.447
0.000
‐3.786
‐1.857
8
0.750
0.358
0.057
‐0.024
1.524
3
4
‐3.214
0.752
0.001
‐4.838
‐1.590
5
‐1.786
0.714
0.027
‐3.329
‐0.243
6
‐1.536
0.666
0.038
‐2.975
‐0.097
7
‐0.500
0.784
0.535
‐2.195
1.195
8
3.071
0.808
0.002
1.325
4.818
4
5
1.429
0.202
0.000
0.992
1.865
6
1.679
0.285
0.000
1.062
2.295
7
2.714
0.518
0.000
1.595
3.834
8
6.286
0.728
0.000
4.714
7.858
5
6
0.250
0.291
0.405
‐0.378
0.878
7
1.286
0.462
0.016
0.287
2.284
8
4.857
0.688
0.000
3.370
6.344
6
7
1.036
0.365
0.014
0.247
1.824
8
4.607
0.600
0.000
3.310
5.904
7
8
3.571
0.559
0.000
2.363
4.780
2
Perceived exertion Day2
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.786
0.155
0.000
‐1.120
‐0.451
3
‐2.679
0.369
0.000
‐3.476
‐1.881
4
‐6.714
0.431
0.000
‐7.646
‐5.782
5
‐5.071
0.450
0.000
‐6.045
‐4.098
6
‐4.286
0.398
0.000
‐5.146
‐3.425
7
‐3.214
0.318
0.000
‐3.900
‐2.528
8
0.179
0.249
0.486
‐0.360
0.717
2
3
‐1.893
0.336
0.000
‐2.619
‐1.166
4
‐5.929
0.532
0.000
‐7.077
‐4.780
5
‐4.286
0.541
0.000
‐5.455
‐3.116
6
‐3.500
0.508
0.000
‐4.598
‐2.402
7
‐2.429
0.370
0.000
‐3.228
‐1.629
8
0.964
0.285
0.005
0.350
1.579
3
4
‐4.036
0.700
0.000
‐5.548
‐2.523
5
‐2.393
0.707
0.005
‐3.921
‐0.864
6
‐1.607
0.704
0.040
‐3.127
‐0.087
7
‐0.536
0.580
0.373
‐1.789
0.717
8
2.857
0.467
0.000
1.848
3.866
4
5
1.643
0.269
0.000
1.061
2.225
6
2.429
0.327
0.000
1.723
3.134
7
3.500
0.442
0.000
2.546
4.454
8
6.893
0.472
0.000
5.873
7.913
5
6
0.786
0.261
0.010
0.223
1.349
7
1.857
0.345
0.000
1.111
2.603
8
5.250
0.497
0.000
4.177
6.323
6
7
1.071
0.245
0.001
0.542
1.601
8
4.464
0.437
0.000
3.521
5.408
7
8
3.393
0.360
0.000
2.615
4.171
3
Cortisol Day 1
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.142
0.034
0.001
‐0.215
‐0.069
3
‐0.291
0.055
0.000
‐0.411
‐0.172
4
‐0.378
0.067
0.000
‐0.522
‐0.234
5
‐0.423
0.060
0.000
‐0.553
‐0.292
6
‐0.359
0.056
0.000
‐0.480
‐0.237
7
‐0.310
0.054
0.000
‐0.427
‐0.193
8
‐0.238
0.037
0.000
‐0.319
‐0.157
2
3
‐0.149
0.036
0.001
‐0.227
‐0.072
4
‐0.236
0.046
0.000
‐0.336
‐0.136
5
‐0.281
0.042
0.000
‐0.371
‐0.191
6
‐0.217
0.041
0.000
‐0.306
‐0.128
7
‐0.168
0.035
0.000
‐0.244
‐0.093
8
‐0.097
0.030
0.006
‐0.161
‐0.032
3
4
‐0.087
0.028
0.008
‐0.146
‐0.027
5
‐0.132
0.027
0.000
‐0.190
‐0.073
6
‐0.068
0.018
0.003
‐0.107
‐0.028
7
‐0.019
0.020
0.368
‐0.062
0.025
8
0.053
0.031
0.115
‐0.015
0.120
4
5
‐0.045
0.019
0.035
‐0.086
‐0.004
6
0.019
0.027
0.482
‐0.038
0.077
7
0.068
0.026
0.021
0.012
0.124
8
0.140
0.046
0.009
0.040
0.239
5
6
0.064
0.024
0.019
0.012
0.116
7
0.113
0.024
0.000
0.061
0.165
8
0.184
0.038
0.000
0.102
0.267
6
7
0.049
0.020
0.030
0.005
0.092
8
0.120
0.028
0.001
0.061
0.180
7
8
0.072
0.031
0.038
0.005
0.138
4
Cortisol Day 2
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.082
0.028
0.011
‐0.142
‐0.022
3
‐0.235
0.042
0.000
‐0.326
‐0.145
4
‐0.294
0.047
0.000
‐0.397
‐0.192
5
‐0.376
0.049
0.000
‐0.481
‐0.271
6
‐0.269
0.047
0.000
‐0.371
‐0.168
7
‐0.262
0.030
0.000
‐0.326
‐0.198
8
‐0.135
0.055
0.029
‐0.254
‐0.016
2
3
‐0.153
0.033
0.000
‐0.224
‐0.082
4
‐0.212
0.037
0.000
‐0.293
‐0.132
5
‐0.294
0.037
0.000
‐0.373
‐0.215
6
‐0.187
0.041
0.001
‐0.276
‐0.098
7
‐0.180
0.028
0.000
‐0.240
‐0.119
8
‐0.053
0.059
0.387
‐0.180
0.075
3
4
‐0.059
0.025
0.032
‐0.112
‐0.006
5
‐0.141
0.023
0.000
‐0.191
‐0.091
6
‐0.034
0.034
0.334
‐0.107
0.039
7
‐0.027
0.030
0.397
‐0.092
0.039
8
0.100
0.058
0.106
‐0.025
0.226
4
5
‐0.082
0.017
0.000
‐0.119
‐0.044
6
0.025
0.022
0.277
‐0.023
0.074
7
0.033
0.026
0.237
‐0.024
0.089
8
0.160
0.061
0.022
0.027
0.292
5
6
0.107
0.026
0.001
0.050
0.164
7
0.114
0.029
0.002
0.051
0.177
8
0.241
0.062
0.002
0.107
0.375
6
7
0.007
0.023
0.756
‐0.042
0.056
8
0.134
0.061
0.045
0.003
0.265
7
8
0.127
0.054
0.034
0.011
0.243
5
Chromogranin A Day 1
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.404
1.962
0.840
‐4.642
3.835
3
‐0.997
1.875
0.604
‐5.048
3.055
4
‐4.300
1.383
0.008
‐7.289
‐1.312
5
‐5.091
1.756
0.012
‐8.886
‐1.297
6
‐3.296
1.993
0.122
‐7.601
1.010
7
‐1.417
1.542
0.375
‐4.749
1.915
8
‐2.347
1.093
0.051
‐4.708
0.014
2
3
‐0.593
0.935
0.537
‐2.613
1.428
4
‐3.897
1.828
0.053
‐7.845
0.052
5
‐4.688
1.345
0.004
‐7.594
‐1.781
6
‐2.892
2.143
0.200
‐7.522
1.739
7
‐1.014
2.085
0.635
‐5.519
3.491
8
‐1.944
2.016
0.353
‐6.298
2.411
3
4
‐3.304
1.422
0.037
‐6.376
‐0.232
5
‐4.095
1.345
0.009
‐7.000
‐1.190
6
‐2.299
1.928
0.254
‐6.464
1.865
7
‐0.421
1.820
0.821
‐4.354
3.512
8
‐1.351
1.980
0.507
‐5.629
2.928
4
5
‐0.791
1.468
0.599
‐3.961
2.379
6
1.005
1.675
0.559
‐2.615
4.624
7
2.883
1.416
0.063
‐0.176
5.942
8
1.953
1.281
0.151
‐0.815
4.721
5
6
1.796
1.635
0.292
‐1.736
5.328
7
3.674
1.325
0.016
0.812
6.535
8
2.744
1.918
0.176
‐1.400
6.888
6
7
1.878
1.140
0.123
‐0.584
4.340
8
0.948
1.945
0.634
‐3.254
5.150
7
8
‐0.930
1.762
0.607
‐4.736
2.876
6
Chromogranin A Day 2
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐2.998
1.697
0.101
‐6.664
0.668
3
‐4.166
1.578
0.020
‐7.576
‐0.756
4
‐4.660
1.767
0.021
‐8.477
‐0.842
5
‐5.597
1.549
0.003
‐8.944
‐2.250
6
‐5.389
1.627
0.006
‐8.904
‐1.875
7
‐1.747
0.937
0.085
‐3.771
0.276
8
‐1.744
1.479
0.259
‐4.938
1.451
2
3
‐1.167
1.380
0.413
‐4.150
1.815
4
‐1.662
1.702
0.347
‐5.338
2.015
5
‐2.599
1.593
0.127
‐6.040
0.842
6
‐2.391
1.492
0.133
‐5.614
0.831
7
1.251
1.755
0.489
‐2.541
5.042
8
1.254
2.098
0.560
‐3.277
5.786
3
4
‐0.494
0.955
0.614
‐2.557
1.569
5
‐1.431
0.792
0.094
‐3.142
0.279
6
‐1.224
1.198
0.326
‐3.812
1.365
7
2.418
1.380
0.103
‐0.564
5.400
8
2.422
2.401
0.332
‐2.765
7.609
4
5
‐0.937
0.498
0.082
‐2.012
0.138
6
‐0.730
1.591
0.654
‐4.167
2.708
7
2.912
1.595
0.091
‐0.533
6.358
8
2.916
2.705
0.301
‐2.927
8.759
5
6
0.208
1.334
0.879
‐2.674
3.090
7
3.850
1.491
0.023
0.628
7.072
8
3.853
2.397
0.132
‐1.325
9.031
6
7
3.642
1.613
0.042
0.158
7.126
8
3.646
2.060
0.100
‐0.806
8.097
7
8
0.004
1.985
0.999
‐4.285
4.293
7
Memory Day 1
(I) Loop
(J) Loop
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐1.071
0.425
0.026
‐1.990
‐0.153
3
0.857
0.653
0.212
‐0.555
2.269
4
0.000
0.734
1.000
‐1.585
1.585
2
3
1.929
0.633
0.009
0.561
3.296
4
1.071
0.699
0.149
‐0.439
2.582
3
4
‐0.857
0.467
0.089
‐1.866
0.152
Memory Day 2
(I) Loop
(J) Loop
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
0.500
0.454
0.291
‐0.481
1.481
3
2.286
0.549
0.001
1.100
3.472
4
1.429
0.571
0.027
0.194
2.663
2
3
1.786
0.595
0.010
0.501
3.070
4
0.929
0.518
0.097
‐0.192
2.049
3
4
‐0.857
0.553
0.145
‐2.052
0.338
Visual attention Day 1
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐2.643
1.239
0.052
‐5.319
0.033
3
‐4.357
1.269
0.004
‐7.099
‐1.615
4
‐6.071
1.396
0.001
‐9.088
‐3.055
5
‐4.571
1.073
0.001
‐6.889
‐2.254
6
‐7.571
1.540
0.000
‐10.898
‐4.245
2
3
‐1.714
1.056
0.128
‐3.995
0.567
4
‐3.429
1.026
0.005
‐5.644
‐1.213
5
‐1.929
1.112
0.106
‐4.330
0.473
6
‐4.929
1.220
0.001
‐7.564
‐2.293
3
4
‐1.714
0.986
0.106
‐3.844
0.415
5
‐0.214
0.459
0.648
‐1.206
0.777
6
‐3.214
1.223
0.021
‐5.857
‐0.572
4
5
1.500
0.924
0.129
‐0.497
3.497
6
‐1.500
1.088
0.191
‐3.851
0.851
5
6
‐3.000
1.186
0.025
‐5.562
‐0.438
8
Visual attention Day 2
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐2.214
1.100
0.065
‐4.591
0.163
3
‐0.929
0.848
0.293
‐2.761
0.904
4
‐2.929
0.946
0.009
‐4.973
‐0.885
5
‐0.929
0.730
0.226
‐2.505
0.648
6
‐2.143
0.882
0.030
‐4.049
‐0.237
2
3
1.286
1.019
0.229
‐0.915
3.486
4
‐0.714
0.854
0.418
‐2.560
1.132
5
1.286
0.922
0.187
‐0.707
3.278
6
0.071
0.722
0.923
‐1.489
1.632
3
4
‐2.000
1.084
0.088
‐4.343
0.343
5
0.000
0.914
1.000
‐1.974
1.974
6
‐1.214
0.921
0.210
‐3.204
0.775
4
5
2.000
1.144
0.104
‐0.470
4.470
6
0.786
0.757
0.318
‐0.850
2.422
5
6
‐1.214
0.786
0.146
‐2.912
0.483
Attention/mental flexibility Day 1
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
3.143
1.955
0.132
‐1.081
7.367
3
8.786
1.989
0.001
4.488
13.084
4
11.000
2.226
0.000
6.191
15.809
5
11.571
2.301
0.000
6.601
16.542
6
9.500
2.629
0.003
3.821
15.179
2
3
5.643
1.369
0.001
2.685
8.601
4
7.857
2.463
0.007
2.537
13.177
5
8.429
2.005
0.001
4.098
12.759
6
6.357
2.286
0.016
1.418
11.296
3
4
2.214
1.849
0.253
‐1.781
6.209
5
2.786
2.209
0.230
‐1.987
7.559
6
0.714
2.590
0.787
‐4.881
6.310
4
5
0.571
3.063
0.855
‐6.046
7.188
6
‐1.500
3.038
0.630
‐8.063
5.063
5
6
‐2.071
3.211
0.530
‐9.009
4.866
9
Attention/mental flexibility Day 2
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
9.500
2.942
0.007
3.144
15.856
3
8.214
3.173
0.022
1.359
15.069
4
9.786
2.798
0.004
3.742
15.829
5
12.143
2.518
0.000
6.704
17.582
6
11.143
3.305
0.005
4.003
18.282
2
3
‐1.286
2.579
0.626
‐6.858
4.287
4
0.286
1.920
0.884
‐3.862
4.433
5
2.643
1.750
0.155
‐1.137
6.423
6
1.643
1.929
0.410
‐2.524
5.810
3
4
1.571
2.639
0.562
‐4.130
7.272
5
3.929
1.659
0.034
0.344
7.513
6
2.929
3.239
0.382
‐4.068
9.925
4
5
2.357
1.737
0.198
‐1.396
6.110
6
1.357
2.548
0.603
‐4.147
6.861
5
6
‐1.000
3.024
0.746
‐7.532
5.532
10
Functional psychobiosocial states Day 1
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.071
0.099
0.484
‐0.285
0.143
3
0.082
0.088
0.372
‐0.109
0.272
4
‐0.061
0.227
0.792
‐0.553
0.430
5
‐0.194
0.202
0.354
‐0.630
0.242
6
‐0.194
0.181
0.304
‐0.586
0.198
7
‐0.347
0.125
0.016
‐0.617
‐0.077
8
‐0.408
0.175
0.037
‐0.787
‐0.030
2
3
0.153
0.092
0.119
‐0.045
0.351
4
0.010
0.206
0.961
‐0.434
0.454
5
‐0.122
0.209
0.567
‐0.573
0.328
6
‐0.122
0.171
0.486
‐0.491
0.246
7
‐0.276
0.157
0.102
‐0.614
0.063
8
‐0.337
0.162
0.058
‐0.686
0.013
3
4
‐0.143
0.158
0.384
‐0.485
0.200
5
‐0.276
0.142
0.074
‐0.582
0.030
6
‐0.276
0.142
0.074
‐0.582
0.030
7
‐0.429
0.126
0.005
‐0.701
‐0.156
8
‐0.490
0.180
0.018
‐0.879
‐0.101
4
5
‐0.133
0.121
0.293
‐0.394
0.129
6
‐0.133
0.138
0.353
‐0.430
0.165
7
‐0.286
0.181
0.138
‐0.677
0.105
8
‐0.347
0.236
0.166
‐0.857
0.163
5
6
0.000
0.186
1.000
‐0.403
0.403
7
‐0.153
0.168
0.380
‐0.517
0.211
8
‐0.214
0.234
0.377
‐0.720
0.292
6
7
‐0.153
0.119
0.222
‐0.411
0.105
8
‐0.214
0.169
0.226
‐0.579
0.150
7
8
‐0.061
0.133
0.653
‐0.348
0.226
11
Functional psychobiosocial states Day 2
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.029
0.090
0.755
‐0.222
0.165
3
‐0.049
0.115
0.677
‐0.297
0.199
4
0.063
0.159
0.697
‐0.279
0.406
5
‐0.080
0.177
0.660
‐0.462
0.302
6
‐0.161
0.142
0.277
‐0.468
0.146
7
‐0.049
0.107
0.654
‐0.280
0.182
8
‐0.049
0.146
0.742
‐0.364
0.266
2
3
‐0.020
0.080
0.804
‐0.194
0.153
4
0.092
0.122
0.466
‐0.172
0.356
5
‐0.051
0.156
0.749
‐0.388
0.286
6
‐0.133
0.149
0.391
‐0.455
0.190
7
‐0.020
0.105
0.848
‐0.247
0.206
8
‐0.020
0.156
0.898
‐0.356
0.316
3
4
0.112
0.071
0.136
‐0.040
0.265
5
‐0.031
0.114
0.793
‐0.277
0.216
6
‐0.112
0.134
0.418
‐0.402
0.177
7
0.000
0.090
1.000
‐0.194
0.194
8
0.000
0.151
1.000
‐0.327
0.327
4
5
‐0.143
0.109
0.213
‐0.378
0.093
6
‐0.224
0.146
0.147
‐0.539
0.090
7
‐0.112
0.096
0.264
‐0.320
0.095
8
‐0.112
0.166
0.511
‐0.471
0.247
5
6
‐0.082
0.107
0.458
‐0.312
0.149
7
0.031
0.121
0.804
‐0.231
0.292
8
0.031
0.166
0.857
‐0.328
0.390
6
7
0.112
0.094
0.253
‐0.090
0.315
8
0.112
0.148
0.461
‐0.207
0.431
7
8
0.000
0.102
1.000
‐0.219
0.219
12
Dysfunctional psychobiosocial states Day 1
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.104
0.054
0.076
‐0.221
0.013
3
‐0.160
0.064
0.026
‐0.298
‐0.023
4
‐0.244
0.068
0.003
‐0.391
‐0.096
5
‐0.205
0.071
0.013
‐0.358
‐0.052
6
‐0.192
0.117
0.125
‐0.445
0.061
7
‐0.163
0.084
0.075
‐0.344
0.019
8
0.047
0.156
0.769
‐0.291
0.385
2
3
‐0.056
0.043
0.218
‐0.150
0.038
4
‐0.140
0.050
0.016
‐0.248
‐0.031
5
‐0.101
0.069
0.168
‐0.250
0.048
6
‐0.088
0.110
0.439
‐0.326
0.150
7
‐0.059
0.100
0.568
‐0.275
0.158
8
0.151
0.166
0.380
‐0.208
0.510
3
4
‐0.083
0.059
0.181
‐0.211
0.044
5
‐0.044
0.074
0.557
‐0.204
0.115
6
‐0.032
0.107
0.772
‐0.263
0.200
7
‐0.002
0.095
0.981
‐0.207
0.202
8
0.207
0.181
0.273
‐0.183
0.598
4
5
0.039
0.064
0.552
‐0.099
0.176
6
0.052
0.120
0.673
‐0.207
0.310
7
0.081
0.099
0.429
‐0.133
0.295
8
0.291
0.174
0.119
‐0.086
0.667
5
6
0.013
0.095
0.895
‐0.192
0.218
7
0.042
0.094
0.660
‐0.160
0.244
8
0.252
0.147
0.111
‐0.067
0.570
6
7
0.029
0.118
0.807
‐0.225
0.284
8
0.239
0.168
0.178
‐0.123
0.601
7
8
0.210
0.162
0.219
‐0.141
0.560
13
Dysfunctional psychobiosocial states Day 2
(I) Phase
(J) Phase
Mean difference (I‐J) Std. Error
p level
95% CI
Lower limit Higher limit
1
2
‐0.075
0.033
0.041
‐0.147
‐0.003
3
‐0.109
0.060
0.090
‐0.238
0.019
4
‐0.269
0.082
0.006
‐0.447
‐0.091
5
‐0.163
0.071
0.040
‐0.317
‐0.009
6
‐0.100
0.100
0.334
‐0.315
0.115
7
‐0.200
0.083
0.031
‐0.379
‐0.022
8
‐0.064
0.094
0.506
‐0.267
0.138
2
3
‐0.034
0.056
0.551
‐0.155
0.086
4
‐0.194
0.086
0.042
‐0.379
‐0.008
5
‐0.088
0.064
0.193
‐0.226
0.050
6
‐0.025
0.089
0.786
‐0.216
0.167
7
‐0.125
0.093
0.204
‐0.327
0.077
8
0.011
0.088
0.901
‐0.179
0.201
3
4
‐0.160
0.068
0.035
‐0.306
‐0.013
5
‐0.054
0.064
0.418
‐0.192
0.085
6
0.010
0.080
0.907
‐0.164
0.183
7
‐0.091
0.081
0.283
‐0.266
0.084
8
0.045
0.089
0.620
‐0.147
0.237
4
5
0.106
0.088
0.250
‐0.084
0.296
6
0.169
0.104
0.128
‐0.056
0.394
7
0.069
0.081
0.410
‐0.105
0.243
8
0.205
0.082
0.027
0.027
0.382
5
6
0.063
0.055
0.272
‐0.056
0.182
7
‐0.037
0.090
0.686
‐0.232
0.158
8
0.099
0.100
0.342
‐0.118
0.315
6
7
‐0.101
0.088
0.277
‐0.292
0.091
8
0.036
0.094
0.710
‐0.167
0.238
7
8
0.136
0.077
0.102
‐0.031
0.303
14
| Psychophysiological responses of junior orienteers under competitive pressure. | 04-26-2018 | Robazza, Claudio,Izzicupo, Pascal,D'Amico, Maria Angela,Ghinassi, Barbara,Crippa, Maria Chiara,Di Cecco, Vincenzo,Ruiz, Montse C,Bortoli, Laura,Di Baldassarre, Angela | eng |
PMC9794057 |
1
S2 Table. Qualifications and responsibilities of the expert panel.
Qualifications
-
Possess extensive experience with and knowledge of endurance performance
-
Interest in the topic
Responsibilities
-
Participate anonymously in three rounds of questionnaires (15 minutes each)
| Factors associated with high-level endurance performance: An expert consensus derived via the Delphi technique. | 12-27-2022 | Konopka, Magdalena J,Zeegers, Maurice P,Solberg, Paul A,Delhaije, Louis,Meeusen, Romain,Ruigrok, Geert,Rietjens, Gerard,Sperlich, Billy | eng |
PMC9784187 | Citation: Sawada, T.; Okawara, H.;
Nakashima, D.; Ikeda, K.; Nagahara,
J.; Fujitsuka, H.; Hoshino, S.; Maeda,
Y.; Katsumata, Y.; Nakamura, M.;
et al. Constant Load Pedaling
Exercise Combined with Electrical
Muscle Stimulation Leads to an Early
Increase in Sweat Lactate Levels.
Sensors 2022, 22, 9585. https://
doi.org/10.3390/s22249585
Academic Editors: Gian Marco Revel
and Sara Casaccia
Received: 9 November 2022
Accepted: 6 December 2022
Published: 7 December 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright:
© 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed
under
the
terms
and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
sensors
Article
Constant Load Pedaling Exercise Combined with Electrical
Muscle Stimulation Leads to an Early Increase in Sweat
Lactate Levels
Tomonori Sawada 1
, Hiroki Okawara 1
, Daisuke Nakashima 1,*
, Kaito Ikeda 1, Joji Nagahara 1,
Haruki Fujitsuka 1, Sosuke Hoshino 1, Yuta Maeda 1, Yoshinori Katsumata 2,3, Masaya Nakamura 1
and Takeo Nagura 1,4
1
Department of Orthopaedic Surgery, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku,
Tokyo 160-8582, Japan
2
Institute for Integrated Sports Medicine, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku,
Tokyo 160-8582, Japan
3
Department of Cardiology, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku,
Tokyo 160-8582, Japan
4
Department of Clinical Biomechanics, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku,
Tokyo 160-8582, Japan
*
Correspondence: [email protected]; Tel.: +81-3-5363-3812
Abstract: A novel exercise modality combined with electrical muscle stimulation (EMS) has been
reported to increase cardiovascular and metabolic responses, such as blood lactate concentration. We
aimed to examine the effect of constant load pedaling exercise, combined with EMS, by non-invasively
and continuously measuring sweat lactate levels. A total of 22 healthy young men (20.7 ± 0.8 years)
performed a constant load pedaling exercise for 20 min at 125% of the pre-measured ventilatory
work threshold with (EMS condition) and without (control condition) EMS stimulation. Blood lactate
concentration was measured by blood samples obtained from the earlobe every minute. Sweat lactate
was monitored in real time using a sensor placed on the forearm. The sweat lactate threshold (sLT)
was defined as the point of increase in sweat lactate. sLT occurred significantly earlier in the EMS
condition than in the control condition. In the single regression analysis, the difference in sLT between
the two conditions, as the independent variable, was a significant predictor of the difference in blood
lactate concentrations at the end of the exercise (p < 0.05, r = −0.52). Sweat lactate measurement may
be a noninvasive and simple alternative to blood lactate measurement to determine the effectiveness
of exercise combined with EMS.
Keywords: sweat lactate; blood lactate; electrical muscle stimulation; exercise
1. Introduction
Electrical muscle stimulation (EMS) has been widely used in rehabilitation and sports
to assist in exercise therapy or to increase the exercise load, in combination with regular
exercise [1–6]. Adding EMS has a significant impact on muscle metabolism and induces
substantial physiological adaptations [7,8]. Stimulated contractions caused by adding
EMS intensively activates anaerobic glycolysis for energy production by phosphocreatine
and glycogen degradation, leading to an increase in blood lactate concentration [9–11].
Lactate, which is produced in the glycolytic pathway, is widely recognized as an efficient
energy source used by systemic organs [12]. Additionally, it has been reported that lactate
increases the peroxisome proliferator-activated receptor-γ coactivator- (PGC-) 1α mRNA
expression and mitochondrial biogenesis [13–17]. PGC-1α is the master controller of mi-
tochondrial biogenesis and promotes mitochondrial biogenesis through the activation of
various transcription factors [18]. Since mitochondrial content is generally considered an
indicator of endurance performance [19], increasing mitochondrial biogenesis induced by
Sensors 2022, 22, 9585. https://doi.org/10.3390/s22249585
https://www.mdpi.com/journal/sensors
Sensors 2022, 22, 9585
2 of 11
exercise training is believed to be an important adaptative event that improves exercise
tolerance capacity [20,21]. Further, an exercise-induced increase in PGC-1α mRNA was
observed above the first lactate threshold, but not below it [22]. Therefore, it is important
to determine the EMS intensity that increases the blood lactate levels without fatigue, as
compared to exercise without EMS, to establish an efficient program of exercise training
combined with EMS. However, measuring the blood lactate levels frequently to determine
the appropriate EMS output is not feasible, since it requires invasive blood sampling. Addi-
tionally, it is unclear whether the optimal measurement time should be at the moment of
puncture, or when the blood is discharged after a time lag of 5–10 s following the puncture.
Thus, establishing a noninvasive method of monitoring the dynamics of metabolism during
exercise combined with EMS is necessary.
To date, noninvasive biosensors of lactate using sweat from the body surface have
been reported [23] and have shown that sweat lactate levels increase in conjunction with
exercise intensity [24–26]. In a recent study, the point of sweat lactate level elevation
during incremental exercise was well correlated with the lactate threshold determined
from the blood lactate concentration and the ventilatory threshold (VT), determined using
a respiratory gas analyzer in patients with cardiovascular disease, as well as healthy
individuals [27]. When this technology is applied to exercise combined with EMS, the
noninvasive and continuous monitoring of lactate behavior during exercise could be
possible, and increased blood lactate levels may be more easily detected by measuring
sweat lactate levels. In particular, it would be suitable for the remote rehabilitation of
patients and athletes outside of medical institutions and may enable effective and safe
exercise based on daily physical conditions.
Thus, the current study aimed to examine whether sweat lactate concentrations could
be used to detect increased blood lactate levels during exercise combined with EMS. We
hypothesized that the increase in blood lactate concentration during exercise combined
with EMS would be predicted, to some extent, by the increase in sweat lactate levels.
2. Materials and Methods
2.1. Participants
After recruiting participants from one university starting in May 2021, 22 healthy
recreationally trained men (average age, 20.3 years) participated in this study conducted
between June and September 2021 (Table 1). The inclusion criteria were (1) age ≥18 years
and (2) no medical history of illness or injury, and not currently taking any medication.
The exclusion criteria were (1) lower extremity injury or disorder that hinders complete
participation in exercise, (2) metabolic, cardiac, respiratory, and psychiatric diseases, and
(3) severe skin disease. The study protocol was conducted in compliance with the ethical
guidelines for medical and health research involving human subjects and was approved
by the Institutional Review Board of our institution (approval number: 20190229). Written
informed consent was obtained from the individuals for study participation and publication
of the findings before enrollment.
Table 1. Participant characteristics (n = 22).
Mean (SD)
Range
Age (years)
20.7 (0.8)
19–22
Height (cm)
174.0 (5.4)
161.0–184.0
Weight (kg)
66.8 (9.0)
46.8–91.5
BMI (kg/m2)
22.0 (2.2)
18.1–27.0
Body fat ratio (%)
15.4 (4.3)
8.6–23.8
Fat mass (kg)
10.5 (4.0)
4.0–20.1
Lean body mass (kg)
56.4 (6.2)
42.8–71.4
Muscle mass (kg)
53.4 (5.9)
40.5–67.7
Total body water (kg)
40.5 (6.5)
30.3–57.6
Body water (%)
59.7 (4.7)
49.6–65.9
BMI, body mass index.
Sensors 2022, 22, 9585
3 of 11
2.2. Experimental Protocol
All participants were asked to visit our laboratory three times. For each of the three vis-
its, an interval of at least 3–14 days was allowed. In all three sessions, the room temperature
was set to the same level and pedaling exercises were performed using an electromagneti-
cally braked ergometer (POWER MAX V3 Pro; Konami Sports Co., Ltd., Tokyo, Japan) with
a target of 70 rpm. At the first visit, prior to an incremental load test, body composition
was measured using a multi-frequency body composition meter (MC780A-N; TANITA Cor-
poration, Tokyo, Japan). Then, the exercise test was performed, and the VT was determined
for each individual with a respiratory gas analyzer using the ventilatory equivalent, excess
carbon dioxide, and modified V-slope methods [28]. Specifically, following a 2 min rest
to stabilize the heart rate and respiration, the participants performed a 4 min warm-up,
pedaling at 20 W, and then exercised at increasing intensity until they could no longer
maintain the pedaling rate (volitional exhaustion). The resistance was increased in 25 W
increments from 50 W at 1 min intervals. Once the exercise tests were terminated, the
participants were instructed to stop pedaling and remain on the ergometer for 3 min. The
expired gas flow was measured using a breath-by-breath automated system (Aeromonitor®;
Minato Medical Science Co., Ltd., Osaka, Japan). Prior to the second and third visits, the
participants were instructed to keep a fast for 3 h prior to the measurements and to refrain
from caffeine and alcohol intake and engaging in intensive exercises within 12 h. Partici-
pants were also asked to drink 500 mL of water before the exercise. On the second visit,
constant load pedaling exercise, without EMS stimulation (control condition, hereinafter
called “CR condition”), was performed. To perform the exercise, the sweat lactate level
was monitored with a sweat lactate sensor (Grace Imaging Inc., Tokyo, Japan) attached to
the left forearm, and a Fitbit Inspire HR (Fitbit Inc., San Francisco, CA, USA) was attached
to the left wrist, two-finger widths above the ulnar styloid process, to measure the heart
rate. After an initial 2 min of rest and measurement of sweat and blood lactate levels, a
20 min pedaling exercise was performed at a constant load of 125% of the pre-measured
ventilatory work threshold. The loading of VT125% was determined based on a previous
study [11] reporting that the difference in blood lactate concentrations between conditions
with and without EMS was significantly greater in the VT125% exercise loading than in the
VT50% and VT75% exercise loading. During the pedaling exercise, the sweat lactate level
was measured at 1 Hz, and the blood lactate level was measured (Lactate Pro 2; ARKRAY,
Inc., Kyoto, Japan) by drawing blood from the earlobe every minute. The exercise was
stopped under the following conditions, even if the duration was <20 min: when the heart
rate exceeded 190 bpm; when the examiner judged the exercise to be dangerous to the
patient; when the participant requested to stop; when the participant became exhausted;
or when it became difficult to maintain the target speed of 70 rpm. In such cases, blood
lactate levels were measured at the end of the exercise. At the third visit, the exercise was
performed in the same way as the second visit, but with EMS stimulation. In the EMS
condition, participants wore a commercially available EMS suit separated into a top and a
bottom section (Powersuit; MTG Ltd., Nagoya, Japan) and performed pedaling exercises
while seven muscles on each side of the upper and lower trunk were stimulated (biceps
brachii, triceps brachii, rectus abdominis, oblique abdominis, gluteus medius, quadriceps
femoris, and hamstrings). In accordance with a previous study [11], all electrode pairs were
synchronized and biphasic square current pulses with a 100 µs duration were constantly
applied at a stimulation frequency of 4 Hz. The EMS intensity was set to a maximum
intensity at which each participant did not feel pain or discomfort. The maximal electrical
potential and current of this device were 50 V and 4.85 mA, respectively.
Sensors 2022, 22, 9585
4 of 11
2.3. Sweat Lactate Threshold Measurements
The sweat lactate level was measured using a wearable sensor, which quantifies lactate
concentration as a current value because it reacts with sweat lactate and generates an
electric current [27]. The current value can be obtained as continuous data within 0.1–80 µA
in 0.1-µA increments. After calibration using saline for approximately 3 min, the sensor
chip connected to the sensor device was attached to the participants’ dorsal left forearm,
which was cleaned with an alcohol-free cloth. Additionally, the data were recorded at a
sampling frequency of 1 Hz for mobile applications using a Bluetooth connection. The
recorded data were converted to moving average values over 13 s intervals and individually
underwent zero correction using the baseline value. Sweat lactate threshold (sLT) was
defined as the first significant increase in the sweat lactate level above baseline, based on
graphical plots [27] by three researchers in consultation.
2.4. Statistical Analysis
All data are presented as the mean plus standard deviation. For the heart rate and
blood lactate concentration data, in a population who performed 20 min of exercise in
both conditions, two-way analysis of variance with repeated measures was used to test
the main effects under two conditions (CR and EMS) and five time points (baseline, 5, 10,
15, and 20 min), as well as the interaction effect between the time point and the condition.
Bonferroni correction was performed for post hoc pairwise comparison. For sLTs, paired
t-tests were conducted on participants for whom sLTs could be defined in order to compare
values between the two conditions. Additionally, to predict the increase in blood lactate
concentration in the EMS condition relative to the CR condition, a single regression analysis
was performed, with the blood lactate concentration change between the two conditions as
the dependent variable, and the change in the time to reach sLT between the two conditions
as the independent variable. In the regression analysis, the blood lactate concentration
was used as the value after 20 min, or at the end of exercise, because the analysis included
patients who could not exercise for 20 min. All statistical analyses were performed using
SPSS Statistics version 27.0 (IBM Corp., Armonk, NY, USA), with statistical significance set
at 0.05.
3. Results
Representative data of sweat and blood lactate concentrations during a constant load
pedaling exercise are shown in Figure 1. Of the 22 participants, 15 completed 20 min of
constant load exercise in both conditions, and the remaining 7 had difficulty maintaining
the target speed of 70 rpm due to fatigue and stopped the exercise before 20 min. Four of the
seven participants had difficulty completing 20 min in both conditions, two had difficulty
completing 20 min only in the EMS condition, and one had difficulty completing 20 min
only in the CR condition. Additionally, detecting sLT was difficult in 2 out of 22 participants,
as the sensor did not respond, due to lack of sweating in one participant, and the sensor
chip was poorly connected in the other participant. Therefore, only 20 participants were
included in the analysis of sLT.
Sensors 2022, 22, 9585
5 of 11
Figure 1. Representative data of sweat lactate and blood lactate concentrations during constant load
pedaling exercise. EMS, electrical muscle stimulation; CR, without EMS stimulation; bLA, blood
lactate; sLA, sweat lactate; sLT, sweat lactate threshold.
Figure 2 shows the comparison of the heart rate under the two conditions. As a result,
there was a main effect of time point observed [F (4, 56) = 167.831, p < 0.01]. Post hoc test
results showed significant differences between BL and other measurement time points (5,
10, 15, and 20 min), between 5 and 10, 5 and 15, and 5 and 20 min. Contrastingly, there
was no main effect for condition, and no interaction effect between condition and meas-
urement point. Figure 3 also depicts the comparison of blood lactate concentration be-
tween the two conditions. For blood lactate concentration, there was a main effect of time
point [F (4, 56) = 19.703, p < 0.01] and condition [F (1, 14) = 11.050, p < 0.01] and an interac-
tion effect between time point and condition [F (4, 56) = 4.069, p < 0.01]. Post hoc test results
showed significant differences between BL and other measurement points (5, 10, 15, and
20 min). Additionally, significant differences were observed between the conditions at 10,
15, and 20 min, with the EMS condition showing an increase in blood lactate concentration
compared to the CR condition.
-1
0
0
120
240
360
480
600
720
840
960
1080
1200
1320
Time [s]
Figure 1. Representative data of sweat lactate and blood lactate concentrations during constant load
pedaling exercise. EMS, electrical muscle stimulation; CR, without EMS stimulation; bLA, blood
lactate; sLA, sweat lactate; sLT, sweat lactate threshold.
Figure 2 shows the comparison of the heart rate under the two conditions. As a result,
there was a main effect of time point observed [F (4, 56) = 167.831, p < 0.01]. Post hoc test
results showed significant differences between BL and other measurement time points (5,
10, 15, and 20 min), between 5 and 10, 5 and 15, and 5 and 20 min. Contrastingly, there was
no main effect for condition, and no interaction effect between condition and measurement
point. Figure 3 also depicts the comparison of blood lactate concentration between the
two conditions. For blood lactate concentration, there was a main effect of time point
[F (4, 56) = 19.703, p < 0.01] and condition [F (1, 14) = 11.050, p < 0.01] and an interaction
effect between time point and condition [F (4, 56) = 4.069, p < 0.01]. Post hoc test results
showed significant differences between BL and other measurement points (5, 10, 15, and
20 min). Additionally, significant differences were observed between the conditions at 10,
15, and 20 min, with the EMS condition showing an increase in blood lactate concentration
compared to the CR condition.
Sensors 2022, 22, x FOR PEER REVIEW
5 of 11
Figure 1. Representative data of sweat lactate and blood lactate concentrations during constant load
pedaling exercise. EMS, electrical muscle stimulation; CR, without EMS stimulation; bLA, blood
lactate; sLA, sweat lactate; sLT, sweat lactate threshold.
Figure 2 shows the comparison of the heart rate under the two conditions. As a result,
there was a main effect of time point observed [F (4, 56) = 167.831, p < 0.01]. Post hoc test
results showed significant differences between BL and other measurement time points (5,
10, 15, and 20 min), between 5 and 10, 5 and 15, and 5 and 20 min. Contrastingly, there
was no main effect for condition, and no interaction effect between condition and meas-
urement point. Figure 3 also depicts the comparison of blood lactate concentration be-
tween the two conditions. For blood lactate concentration, there was a main effect of time
point [F (4, 56) = 19.703, p < 0.01] and condition [F (1, 14) = 11.050, p < 0.01] and an interac-
tion effect between time point and condition [F (4, 56) = 4.069, p < 0.01]. Post hoc test results
showed significant differences between BL and other measurement points (5, 10, 15, and
20 min). Additionally, significant differences were observed between the conditions at 10,
15, and 20 min, with the EMS condition showing an increase in blood lactate concentration
compared to the CR condition.
Figure 2. Comparison of the heart rate between the EMS and CR conditions (n = 15). †: p < 0.01 signif-
icant main effect at time point. EMS, electrical muscle stimulation; CR, without EMS stimulation.
-1
0
1
2
3
4
5
6
7
0
0.5
1
1.5
2
2.5
3
3.5
4
0
120
240
360
480
600
720
840
960
1080
1200
1320
Sweat lactate [µA]
Blood lactate [mmol/l]
Time [s]
bLA_CR
bLA_EMS
sLA_CR
sLA_EMS
20 minutes of constant load pedaling exercise
sLT
369 [sec]
sLT
296 [sec]
Figure 2. Comparison of the heart rate between the EMS and CR conditions (n = 15). †: p < 0.01
significant main effect at time point. EMS, electrical muscle stimulation; CR, without EMS stimulation.
Sensors 2022, 22, 9585
6 of 11
Sensors 2022, 22, x FOR PEER REVIEW
6 of 11
Figure 3. Comparison of blood lactate concentrations between the EMS and CR conditions (n = 15).
†: p < 0.01 significant main effect at time point. * p < 0.05, ** p < 0.01 significant difference between
conditions. EMS, electrical muscle stimulation; CR, without EMS stimulation.
Regarding sLT, it occurred significantly earlier in the EMS condition than in the CR
condition (Figure 4, 215.2 ± 74.5 s and 271.8 ± 104.3 s, respectively; p < 0.05). Additionally,
in the single regression analysis, the difference in sLT between the two conditions as an
independent variable was a significant predictor of the difference in blood lactate concen-
trations at the end of the exercise (Figure 5, p < 0.05; r = −0.52).
Figure 4. Comparison of sweat lactate threshold between the EMS and CR conditions (n = 20). * p <
0.05. EMS, electrical muscle stimulation; CR, without EMS stimulation.
Figure 3. Comparison of blood lactate concentrations between the EMS and CR conditions (n = 15).
†: p < 0.01 significant main effect at time point. * p < 0.05, ** p < 0.01 significant difference between
conditions. EMS, electrical muscle stimulation; CR, without EMS stimulation.
Regarding sLT, it occurred significantly earlier in the EMS condition than in the CR
condition (Figure 4, 215.2 ± 74.5 s and 271.8 ± 104.3 s, respectively; p < 0.05). Additionally,
in the single regression analysis, the difference in sLT between the two conditions as
an independent variable was a significant predictor of the difference in blood lactate
concentrations at the end of the exercise (Figure 5, p < 0.05; r = −0.52).
Sensors 2022, 22, x FOR PEER REVIEW
6 of 11
Figure 3. Comparison of blood lactate concentrations between the EMS and CR conditions (n = 15).
†: p < 0.01 significant main effect at time point. * p < 0.05, ** p < 0.01 significant difference between
conditions. EMS, electrical muscle stimulation; CR, without EMS stimulation.
Regarding sLT, it occurred significantly earlier in the EMS condition than in the CR
condition (Figure 4, 215.2 ± 74.5 s and 271.8 ± 104.3 s, respectively; p < 0.05). Additionally,
in the single regression analysis, the difference in sLT between the two conditions as an
independent variable was a significant predictor of the difference in blood lactate concen-
trations at the end of the exercise (Figure 5, p < 0.05; r = −0.52).
Figure 4. Comparison of sweat lactate threshold between the EMS and CR conditions (n = 20). * p <
0.05. EMS, electrical muscle stimulation; CR, without EMS stimulation.
Figure 4. Comparison of sweat lactate threshold between the EMS and CR conditions (n = 20).
* p < 0.05. EMS, electrical muscle stimulation; CR, without EMS stimulation.
Sensors 2022, 22, 9585
7 of 11
Sensors 2022, 22, x FOR PEER REVIEW
7 of 11
Figure 5. Relationship between the change in blood lactate concentration at the end of exercise and
the change in sweat lactate threshold (n = 20). The amount of change is calculated as the value of the
EMS condition minus the value of the CR condition. EMS, electrical muscle stimulation; CR, without
EMS stimulation.
4. Discussion
The primary result of this study was that the increase in blood lactate concentration
due to exercise combined with EMS could be explained, to some extent, by the sLT
changes, which can be measured noninvasively. Previous studies have reported that ex-
ercise combined with EMS leads to an increase in blood lactate concentration [9–11]. The
mechanism of increased blood lactate concentration is considered to be due to the recruit-
ment of high-threshold motor units and muscle fibers by the additional use of EMS [29–
32]. Therefore, to accurately assess the load imposed on the skeletal muscle by exercise
combined with EMS, it was necessary to frequently measure lactate concentration, which
is considered to directly reflect an increase in metabolic rate and glycolytic carbon flow in
the skeletal muscles [33]. Additionally, as lactate is also considered a signal molecule that
induces mitochondrial neogenesis in skeletal muscle cells [14], it is beneficial to determine
the EMS loads that increases blood lactate concentration in each individual. However,
measuring the blood lactate concentration was not feasible because it required blood sam-
pling. Thus, we focused on the lactate contained in sweat. Sweat lactate measurement has
the potential to compensate for the disadvantages of conventional evaluation methods, as
it can be measured noninvasively and easily, and this could be demonstrated through
exercise combined with EMS. The sweat lactate device that we used is capable of measur-
ing sweat lactate continuously and over a long period of time by adjusting the thickness
and composition of the topcoat applied to the upper layer of lactate oxidase on the sensor
chip [27]. Therefore, it is able to monitor lactate level changes without deactivation of the
enzyme in a single measurement. Our results suggest that this device may be applied in
the future for setting EMS loads and/or determining the effectiveness of exercise in vari-
ous training environments, such as gyms, outdoors, and at home.
As for sweat lactate, it reportedly does not reflect the blood lactate levels during ex-
ercise [34,35]. While lactate is produced in sweat reflecting exercise intensity, it is influ-
enced by the body’s production of lactate, the rate of sweating, and metabolic kinetics in
the sweat glands [34,36]. On the other hand, treating sweat lactate as an elevated point
during incremental load exercise has been verified in a previous report to be consistent
with LT obtained from blood [27]. One possible reason for this measurement consistency
is that increased lactate production from the muscle cells reflecting LT may induce a sim-
ultaneous increase in sweat lactate values through changes in autonomic balance,
Figure 5. Relationship between the change in blood lactate concentration at the end of exercise and
the change in sweat lactate threshold (n = 20). The amount of change is calculated as the value of the
EMS condition minus the value of the CR condition. EMS, electrical muscle stimulation; CR, without
EMS stimulation.
4. Discussion
The primary result of this study was that the increase in blood lactate concentration
due to exercise combined with EMS could be explained, to some extent, by the sLT changes,
which can be measured noninvasively. Previous studies have reported that exercise com-
bined with EMS leads to an increase in blood lactate concentration [9–11]. The mechanism
of increased blood lactate concentration is considered to be due to the recruitment of high-
threshold motor units and muscle fibers by the additional use of EMS [29–32]. Therefore,
to accurately assess the load imposed on the skeletal muscle by exercise combined with
EMS, it was necessary to frequently measure lactate concentration, which is considered
to directly reflect an increase in metabolic rate and glycolytic carbon flow in the skeletal
muscles [33]. Additionally, as lactate is also considered a signal molecule that induces
mitochondrial neogenesis in skeletal muscle cells [14], it is beneficial to determine the EMS
loads that increases blood lactate concentration in each individual. However, measuring
the blood lactate concentration was not feasible because it required blood sampling. Thus,
we focused on the lactate contained in sweat. Sweat lactate measurement has the poten-
tial to compensate for the disadvantages of conventional evaluation methods, as it can
be measured noninvasively and easily, and this could be demonstrated through exercise
combined with EMS. The sweat lactate device that we used is capable of measuring sweat
lactate continuously and over a long period of time by adjusting the thickness and compo-
sition of the topcoat applied to the upper layer of lactate oxidase on the sensor chip [27].
Therefore, it is able to monitor lactate level changes without deactivation of the enzyme in
a single measurement. Our results suggest that this device may be applied in the future
for setting EMS loads and/or determining the effectiveness of exercise in various training
environments, such as gyms, outdoors, and at home.
As for sweat lactate, it reportedly does not reflect the blood lactate levels during exer-
cise [34,35]. While lactate is produced in sweat reflecting exercise intensity, it is influenced
by the body’s production of lactate, the rate of sweating, and metabolic kinetics in the
sweat glands [34,36]. On the other hand, treating sweat lactate as an elevated point during
incremental load exercise has been verified in a previous report to be consistent with LT
obtained from blood [27]. One possible reason for this measurement consistency is that
increased lactate production from the muscle cells reflecting LT may induce a simultaneous
increase in sweat lactate values through changes in autonomic balance, hormones, acid-base
equilibrium, and metabolic dynamics [37,38]. Although the exercise protocol used in the
Sensors 2022, 22, 9585
8 of 11
current study was a constant load rather than an incremental load, because of the load
above AT (VT125%), the lactate discharged from the sweat glands is thought to strongly
reflect anaerobic metabolism by the skeletal muscle, in addition to sweat gland metabolism.
In such a load setting, blood and sweat lactate concentrations increase in the CR condi-
tion itself, which would likely increase further when combined with EMS. In our results
for the single regression analysis, R2 was 0.2726, and in terms of correlation coefficient,
r = −0.52. Therefore, our results showed a moderate association between the increase in
blood lactate at 20 min (or the end of exercise) due to the addition of EMS and the time
change to the point of sLT, and the amount of change in the blood lactate concentration
between conditions could be explained by the amount of change in the point of sLT. More
recently, there have been reports on the development of lactate biosensors that are not
affected by sweat secretion rate [39] and on the development of systems that combine
colorimetric analysis with deep learning to detect sweat lactate [40]. These techniques may
contribute to more accurate sweat lactate measurement during exercise. Moreover, the
results of this study suggest that noninvasive skeletal muscle metabolic assessment, which
previously required blood sampling, may be possible with sweat lactate.
Our results showed that there was no difference in the heart rate between the two con-
ditions. A previous study reported a difference between the conditions with and without
EMS at 80% VT loading [9], which was inconsistent with the results of this study. One pos-
sible reason for the inconsistency in the results could be the difference in the exercise load
intensity. Watanabe et al. examined the effects of different voluntary exercise intensities
(50%, 75%, 100%, and 125% of VT) on metabolic responses to exercise combined with EMS
and reported that the increment in oxygen consumption at 125% of VT was significantly
lower than those at lower exercise intensities [11], suggesting that the additional recruit-
ments of motor units associated with EMS would be attenuated during high-intensity
voluntary exercise, especially over the AT. Contrarily, EMS can induce a greater reliance on
anaerobic glycolysis for energy production, along with phosphocreatine degradation and
lactate formation [41–44]. In this study, the blood lactate concentration was also increased
during exercise combined with EMS. Therefore, the constant load exercise combined with
EMS at 125% of VT may have further increased the glycolytic metabolism, through large
and fatigable fast-twitch motor units with glycolytic fibers. Although future investigations
should be conducted to include exercise intensities of 80% VT and 100% VT, as in previous
studies, there may be a limited increase in heart rate with the addition of EMS under
exercise intensities above AT, in which lactic can accumulate.
This study has several limitations. First, all 22 study participants were university
students and male, suggesting a population bias. Therefore, future evaluations including
participants with different characteristics, such as age, sex, and exercise capacity, would
allow for a wider interpretation and application of the results. Second, it was difficult to
measure the heart rate using an electrocardiogram at the chest due to wearing a pair of
upper and lower full-body suits. Therefore, measurements were taken from the peripheral
wrist with a Fitbit instead. Thus, it may be necessary to consider the influence of mea-
surement uncertainty [45,46] as a reason why there was no difference between the two
conditions with respect to the heart rate. Third, the EMS load intensity was set based on the
subjectivity of everyone; thus, the stimulus intensity differed among participants. Fourth,
we cannot eliminate the possibility that the participants were not given any control, such as
fasting or prohibiting exercise on the previous day, which may have affected the pattern of
blood lactate level changes. Therefore, it is considered that such controls are also necessary
to conduct more precise experiments.
5. Conclusions
Constant load pedaling exercise combined with EMS resulted in an early increase
in sweat lactate levels, which could explain, to some extent, the increase in blood lactate
concentrations. Since sweat lactate measurement is noninvasive, continuous, and easy to
perform, it may be expected to be used as an alternative to blood lactate measurement
Sensors 2022, 22, 9585
9 of 11
for monitoring metabolism during exercise combined with EMS and for determining its
effectiveness in the future.
Author Contributions: T.S., H.O., D.N. and Y.K. conceived and designed the research; T.S., H.O.,
K.I., J.N., H.F., S.H. and Y.M. conducted the experiments and analyzed the data; T.S. drafted the
manuscript and prepared the tables/figures; all the authors edited and revised the manuscript; M.N.
contributed to the supervision of the research project; D.N. and T.N. fulfilled the role of project
administration and funding acquisition. All authors have read and agreed to the published version
of the manuscript.
Funding: This research was funded by KGRI/loT Healthcare Research Consortium (Grant number
02-066-0008).
Institutional Review Board Statement: The study was conducted in accordance with the Declaration
of Helsinki and ethical guidelines for medical and health research involving human participants and
was approved by the Institutional Review Board of our institution (approval no. 20190357).
Informed Consent Statement: Informed consent was obtained from all subjects involved in the
study. Written informed consent has been obtained from the subjects to publish this paper.
Data Availability Statement: The data that support the findings of this study are available from the
corresponding author upon reasonable request.
Acknowledgments: We thank Yoshikazu Kikuchi for his assistance with data collection.
Conflicts of Interest: Daisuke Nakashima is the president of Grace Imaging Inc. and holds shares in
this company. This company has no involvement in this research. The other 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.
References
1.
Malone, J.K.; Blake, C.; Caulfield, B.M. Neuromuscular electrical stimulation during recovery from exercise: A systematic review.
J. Strength Cond. Res. 2014, 28, 2478–2506. [CrossRef] [PubMed]
2.
Paillard, T.; Noe, F.; Passelergue, P.; Dupui, P. Electrical Stimulation Superimposed onto Voluntary Muscular Contraction. Sports
Med. 2005, 35, 951–966. [CrossRef] [PubMed]
3.
Maffiuletti, N.A. Physiological and methodological considerations for the use of neuromuscular electrical stimulation. Eur. J.
Appl. Physiol. 2010, 110, 223–234. [CrossRef] [PubMed]
4.
Fuentes, J.P.; Olivo, S.A.; Magee, D.J.; Gross, D.P. Effectiveness of Interferential Current Therapy in the Management of Muscu-
loskeletal Pain: A Systematic Review and Meta-Analysis. Phys. Ther. 2010, 90, 1219–1238. [CrossRef]
5.
Kristensen, M.G.H.; Busk, H.; Wienecke, T. Neuromuscular Electrical Stimulation Improves Activities of Daily Living Post Stroke:
A Systematic Review and Meta-analysis. Arch. Rehabilitation Res. Clin. Transl. 2021, 4, 100167. [CrossRef]
6.
Wang, H.-Y.; Chen, Y.-H.; Kuan, Y.-C.; Huang, S.-W.; Lin, L.-F.; Chen, H.-C. The effectiveness of functional electrical stimulation of
the legs in patients with heart failure: A systematic review and meta-analysis of randomized controlled trials. Clin. Rehabilitation
2021, 36, 303–316. [CrossRef]
7.
Sánchez, B.R.; Puche, P.P.; González-Badillo, J.J. Percutaneous Electrical Stimulation in Strength Training: An Update. J. Strength
Cond. Res. 2005, 19, 438–448. [CrossRef]
8.
Paillard, T. Combined Application of Neuromuscular Electrical Stimulation and Voluntary Muscular Contractions. Sports Med.
2008, 38, 161–177. [CrossRef]
9.
Watanabe, K.; Taniguchi, Y.; Moritani, T. Metabolic and cardiovascular responses during voluntary pedaling exercise with
electrical muscle stimulation. Eur. J. App. Physiol. 2014, 114, 1801–1807. [CrossRef]
10.
Watanabe, K.; Yoshida, T.; Ishikawa, T.; Kawade, S.; Moritani, T. Effect of the Combination of Whole-Body Neuromuscular
Electrical Stimulation and Voluntary Exercise on Metabolic Responses in Human. Front. Physiol. 2019, 10, 291. [CrossRef]
11.
Watanabe, K.; Takada, T.; Kawade, S.; Moritani, T. Effect of exercise intensity on metabolic responses on combined application of
electrical stimulation and voluntary exercise. Physiol. Rep. 2021, 9, e14758. [CrossRef]
12.
Hui, S.; Ghergurovich, J.M.; Morscher, R.J.; Jang, C.; Teng, X.; Lu, W.; Esparza, L.A.; Reya, T.; Zhan, L.; Guo, J.Y.; et al. Glucose
feeds the TCA cycle via circulating lactate. Nature 2017, 551, 115–118. [CrossRef]
13.
Pilegaard, H.; Saltin, B.; Neufer, P.D. Exercise induces transient transcriptional activation of the PGC-1α gene in human skeletal
muscle. J. Physiol. 2003, 546 Pt 3, 851–858. [CrossRef]
14.
Hashimoto, T.; Hussien, R.; Oommen, S.; Gohil, K.; Brooks, G.A. Lactate sensitive transcription factor network in L6 cells:
Activation of MCT1 and mitochondrial biogenesis. FASEB J. 2007, 21, 2602–2612. [CrossRef]
Sensors 2022, 22, 9585
10 of 11
15.
Perry, C.G.R.; Lally, J.; Holloway, G.P.; Heigenhauser, G.J.F.; Bonen, A.; Spriet, L.L. Repeated transient mRNA bursts precede
increases in transcriptional and mitochondrial proteins during training in human skeletal muscle. J. Physiol. 2010, 588 Pt 23,
4795–4810. [CrossRef] [PubMed]
16.
Kitaoka, Y.; Takeda, K.; Tamura, Y.; Hatta, H. Lactate administration increases mRNA expression of PGC-1α and UCP3 in mouse
skeletal muscle. Appl. Physiol. Nutr. Metab. 2016, 41, 695–698. [CrossRef] [PubMed]
17.
Fiorenza, M.; Gunnarsson, T.P.; Hostrup, M.; Iaia, F.M.; Schena, F.; Pilegaard, H.; Bangsbo, J. Metabolic stress-dependent regulation
of the mitochondrial biogenic molecular response to high-intensity exercise in human skeletal muscle. J. Physiol. 2018, 596,
2823–2840. [CrossRef] [PubMed]
18.
Chodari, L.; Aytemir, M.D.; Vahedi, P.; Alipour, M.; Vahed, S.Z.; Khatibi, S.M.H.; Ahmadian, E.; Ardalan, M.; Eftekhari, A.
Targeting Mitochondrial Biogenesis with Polyphenol Compounds. Oxidative Med. Cell Longev. 2021, 2021, 4946711. [CrossRef]
19.
Oliveira, A.N.; Hood, D.A. Exercise is mitochondrial medicine for muscle. Sports Med. Health Sci. 2019, 1, 11–18. [CrossRef]
20.
Yan, Z.; Lira, V.A.; Greene, N.P. Exercise Training-Induced Regulation of Mitochondrial Quality. Exerc. Sport Sci. Rev. 2012, 40,
159–164. [CrossRef]
21.
Lundby, C.; Jacobs, R.A. Adaptations of skeletal muscle mitochondria to exercise training. Exp. Physiol. 2016, 101, 17–22.
[CrossRef]
22.
Tobina, T.; Yoshioka, K.; Hirata, A.; Mori, S.; Kiyonaga, A.; Tanaka, H. Peroxisomal proliferator-activated receptor gamma
co-activator-1 alpha gene expression increases above the lactate threshold in human skeletal muscle. J. Sports Med. Phys. Fit. 2011,
51, 683–688.
23.
Van Hoovels, K.; Xuan, X.; Cuartero, M.; Gijssel, M.; Swarén, M.; Crespo, G.A. Can Wearable Sweat Lactate Sensors Contribute to
Sports Physiology? ACS Sens. 2021, 6, 3496–3508. [CrossRef] [PubMed]
24.
Jia, W.; Bandodkar, A.J.; Valdés-Ramírez, G.; Windmiller, J.R.; Yang, Z.; Ramírez, J.; Chan, G.; Wang, J. Electrochemical Tattoo
Biosensors for Real-Time Noninvasive Lactate Monitoring in Human Perspiration. Anal. Chem. 2013, 85, 6553–6560. [CrossRef]
[PubMed]
25.
Gao, W.; Emaminejad, S.; Nyein, H.Y.Y.; Challa, S.; Chen, K.; Peck, A.; Fahad, H.M.; Ota, H.; Shiraki, H.; Kiriya, D.; et al. Fully
integrated wearable sensor arrays for multiplexed in situ perspiration analysis. Nature 2016, 529, 509–514. [CrossRef] [PubMed]
26.
Imani, S.; Bandodkar, A.J.; Mohan, A.M.V.; Kumar, R.; Yu, S.; Wang, J.; Mercier, P.P. A wearable chemical–electrophysiological
hybrid biosensing system for real-time health and fitness monitoring. Nat. Commun. 2016, 7, 11650. [CrossRef]
27.
Seki, Y.; Nakashima, D.; Shiraishi, Y.; Ryuzaki, T.; Ikura, H.; Miura, K.; Suzuki, M.; Watanabe, T.; Nagura, T.; Matsumato, M.; et al.
A novel device for detecting anaerobic threshold using sweat lactate during exercise. Sci. Rep. 2021, 11, 4929. [CrossRef]
28.
Gaskill, S.E.; Ruby, B.C.; Walker, A.J.; Sanchez, O.A.; Serfass, R.C.; Leon, A.S. Validity and reliability of combining three methods
to determine ventilatory threshold. Med. Sci. Sports Exerc. 2001, 33, 1841–1848. [CrossRef]
29.
Hamada, T.; Kimura, T.; Moritani, T. Selective fatigue of fast motor units after electrically elicited muscle contractions.
J. Electromyogr. Kinesiol. 2004, 14, 531–538. [CrossRef]
30.
Gregory, C.M.; Bickel, C.S. Recruitment Patterns in Human Skeletal Muscle During Electrical Stimulation. Phys. Ther. 2005, 85,
358–364. [CrossRef]
31.
Jubeau, M.; Gondin, J.; Martin, A.; Sartorio, A.; Maffiuletti, N.A. Random Motor Unit Activation by Electrostimulation. Endoscopy
2007, 28, 901–904. [CrossRef]
32.
Bickel, C.S.; Gregory, C.M.; Dean, J. Motor unit recruitment during neuromuscular electrical stimulation: A critical appraisal. Eur.
J. Appl. Physiol. 2011, 111, 2399–2407. [CrossRef]
33.
Brooks, G.A. Anaerobic threshold: Review of the concept and directions for future research. Med. Sci. Sports Exerc. 1985, 17, 22–34.
[CrossRef]
34.
Lamont, L.S. Sweat lactate secretion during exercise in relation to women’s aerobic capacity. J. Appl. Physiol. 1987, 62, 194–198.
[CrossRef]
35.
Green, J.M.; Bishop, P.A.; Muir, I.H.; McLester, J.R., Jr.; Heath, H.E. Effects of High and Low Blood Lactate Concentrations on
Sweat Lactate Response. Int. J. Sports Med. 2000, 21, 556–560. [CrossRef]
36.
Buono, M.J.; Lee, N.V.L.; Miller, P.W. The relationship between exercise intensity and the sweat lactate excretion rate. J. Physiol.
Sci. 2009, 60, 103–107. [CrossRef] [PubMed]
37.
Alvear-Ordenes, I.; García-López, D.; De Paz, J.A.; González-Gallego, J. Sweat Lactate, Ammonia, and Urea in Rugby Players.
Endoscopy 2005, 26, 632–637. [CrossRef]
38.
Shiraishi, Y.; Katsumata, Y.; Sadahiro, T.; Azuma, K.; Akita, K.; Isobe, S.; Yashima, F.; Miyamoto, K.; Nishiyama, T.; Tamura, Y.;
et al. Real-Time Analysis of the Heart Rate Variability During Incremental Exercise for the Detection of the Ventilatory Threshold.
J. Am. Heart Assoc. 2018, 7, e006612. [CrossRef] [PubMed]
39.
Komkova, M.A.; Eliseev, A.A.; Poyarkov, A.A.; Daboss, E.V.; Evdokimov, P.V.; Eliseev, A.A.; Karyakin, A.A. Simultaneous
monitoring of sweat lactate content and sweat secretion rate by wearable remote biosensors. Biosens. Bioelectron. 2022, 202, 113970.
[CrossRef] [PubMed]
40.
Yüzer, E.; Do˘gan, V.; Kılıç, V.; ¸Sen, M. Smartphone embedded deep learning approach for highly accurate and automated
colorimetric lactate analysis in sweat. Sensors Actuators B Chem. 2022, 371, 132489. [CrossRef]
41.
Spriet, L.L.; Söderlund, K.; Bergström, M.; Hultman, E. Anaerobic energy release in skeletal muscle during electrical stimulation
in men. J. Appl. Physiol. 1987, 62, 611–615. [CrossRef] [PubMed]
Sensors 2022, 22, 9585
11 of 11
42.
Bergstrom, M.; Hultman, E. Energy cost and fatigue during intermittent electrical stimulation of human skeletal muscle. J. Appl.
Physiol. 1988, 65, 1500–1505. [CrossRef] [PubMed]
43.
Greenhaff, P.; Söderlund, K.; Ren, J.M.; Hultman, E. Energy metabolism in single human muscle fibres during intermittent
contraction with occluded circulation. J. Physiol. 1993, 460, 443–453. [CrossRef] [PubMed]
44.
Vanderthommen, M.; Duteil, S.; Wary, C.; Raynaud, J.S.; Leroy-Willig, A.; Crielaard, J.M.; Carlier, P.G. A comparison of voluntary
and electrically induced contractions by interleaved 1H- and 31P-NMRS in humans. J. Appl. Physiol. 2003, 94, 1012–1024.
[CrossRef] [PubMed]
45.
Morresi, N.; Casaccia, S.; Sorcinelli, M.; Arnesano, M.; Revel, G.M. Analysing performances of Heart Rate Variability measurement
through a smartwatch. In Proceedings of the 2020 IEEE International Symposium on Medical Measurements and Applications
(MeMeA), Bari, Italy, 1 June–1 July 2020; IEEE: Piscataway, NJ, USA, 2020; pp. 1–6.
46.
Morresi, N.; Casaccia, S.; Revel, G.M. Metrological characterization and signal processing of a wearable sensor for the measurement
of heart rate variability. In Proceedings of the 2021 IEEE International Symposium on Medical Measurements and Applications
(MeMeA), Lausanne, Switzerland, 23–25 June 2021; IEEE: Piscataway, NJ, USA, 2021; pp. 1–6.
| Constant Load Pedaling Exercise Combined with Electrical Muscle Stimulation Leads to an Early Increase in Sweat Lactate Levels. | 12-07-2022 | Sawada, Tomonori,Okawara, Hiroki,Nakashima, Daisuke,Ikeda, Kaito,Nagahara, Joji,Fujitsuka, Haruki,Hoshino, Sosuke,Maeda, Yuta,Katsumata, Yoshinori,Nakamura, Masaya,Nagura, Takeo | eng |
PMC7379642 | Supplement Table 9. Standardized proportions with a low VO2max (ml·min-1·kg-1) using different cut-offs, in the total population and by sex.
Year
<32 ml
<28.5 ml
<25 ml
<21.5 ml
<32 ml
<28.5 ml
<25 ml
<21.5 ml
<32 ml
<28.5 ml
<25 ml
<21.5 ml
95-97
28,3%
16,5%
7,9%
3,6%
26,3%
14,5%
6,2%
2,5%
27,3%
15,6%
7,1%
3,1%
98-99
37,9%
26,9%
8,7%
2,8%
35,1%
20,3%
11,6%
2,3%
36,5%
23,6%
10,2%
2,5%
00-01
36,7%
20,6%
11,3%
3,2%
38,0%
18,3%
9,7%
5,0%
37,4%
19,4%
10,5%
4,1%
02-03
43,7%
27,3%
14,1%
5,5%
39,7%
27,2%
16,3%
4,0%
41,7%
27,2%
15,2%
4,8%
04-05
43,3%
29,1%
15,6%
6,2%
37,5%
23,3%
9,2%
3,7%
40,4%
26,2%
12,4%
5,0%
06-07
41,0%
26,8%
14,4%
4,7%
38,4%
24,3%
12,1%
4,6%
39,7%
25,5%
13,3%
4,7%
08-09
40,3%
25,5%
14,0%
5,8%
40,5%
26,6%
13,7%
4,7%
40,4%
26,1%
13,9%
5,2%
10-11
43,6%
29,1%
16,8%
6,9%
39,4%
26,1%
13,9%
5,1%
41,5%
27,6%
15,3%
6,0%
12-13
43,2%
29,8%
16,7%
7,3%
43,2%
28,5%
15,6%
6,8%
43,2%
29,1%
16,1%
7,0%
14-15
44,9%
31,4%
18,7%
8,5%
45,8%
30,7%
16,4%
7,0%
45,4%
31,0%
17,5%
7,8%
16-17
45,6%
31,8%
17,2%
6,7%
45,5%
31,4%
18,6%
7,6%
45,6%
31,6%
17,9%
7,2%
Women
Men
Total
| Decline in cardiorespiratory fitness in the Swedish working force between 1995 and 2017. | 11-15-2018 | Ekblom-Bak, Elin,Ekblom, Örjan,Andersson, Gunnar,Wallin, Peter,Söderling, Jonas,Hemmingsson, Erik,Ekblom, Björn | eng |
PMC6647242 | Vol.:(0123456789)
1 3
European Journal of Applied Physiology (2019) 119:1865–1874
https://doi.org/10.1007/s00421-019-04175-w
ORIGINAL ARTICLE
Quantification of aerobic determinants of performance
in post‑pubertal adolescent middle‑distance runners
Richard C. Blagrove1 · Glyn Howatson2,3 · Charles R. Pedlar4,5,6 · Philip R. Hayes2
Received: 26 March 2019 / Accepted: 7 June 2019 / Published online: 17 June 2019
© The Author(s) 2019
Abstract
Purpose The use of oxygen cost ( ̇Oaero) parameters to predict endurance performance has recently been criticized. Instead, it
is suggested that aerobic energy cost ( ̇Eaero ) provides greater validity; however, a comparison of these quantification methods
has not previously been made.
Methods Fifty-six male (n = 34) and female (n = 22) competitive adolescent (17 ± 1 years) middle-distance runners par-
ticipated in a sub-maximal and maximal incremental treadmill test. Running economy (RE) was measured at the speed
corresponding to lactate turnpoint, and the three speeds prior. Maximal oxygen uptake ( ̇VO2max), speed at ̇VO2max and
fraction of ̇VO2max utilized across a range of intensities, and speeds from 0.8, 1.5 and 3 km races were also quantified. RE
and fractional utilization were calculated in units of ̇Oaero and ̇Eaero.
Results Multiple linear regression models demonstrated no discernible difference in the predictive capability of RE, frac-
tional utilization and ̇VO2max when expressed as ̇Oaero or ̇Eaero in both sexes. When plotted as a function of running speed, ̇O
aero displayed a stepwise decrease (F = 11.59, p < 0.001) whereas ̇Eaero exhibited a curvilinear response (F = 4.74, p = 0.015).
Differences were also evident in the slopes plotted for % ̇VO2max and % ̇Eaeromax against running speed (F = 5.38, p = 0.021).
Conclusions Quantifying aerobic determinants of performance in units of ̇Eaero provides no greater validity compared to ̇O
aero-based measurement. Although both ̇Eaero and ̇Oaero are sensitive to changes in speed, ̇Eaero provides the more valid reflec-
tion of the underlying metabolic cost of running. Physiologists should also be aware of the potential differences between
expression of aerobic running intensity based upon % ̇VO2max compared to % ̇Eaeromax.
Keywords Running economy · Maximal oxygen uptake · Fractional utilization · Youth
Abbreviations
ANCOVA Analysis of co-variance
ANOVA
Analysis of variance
CI
Confidence interval
̇Eaero
Aerobic energy cost
̇Eaeromax
Maximal aerobic energy expenditure
LTP
Lactate turnpoint
MDC
Minimal detectable change
̇Oaero
Oxygen cost
RE
Running economy
RER
Respiratory exchange ratio
SD
Standard deviation
sLT
Speed corresponding to lactate threshold
sLTP
Speed corresponding to lactate turnpoint
s ̇VO2max
Speed at ̇VO2max
̇VO2
Oxygen uptake
̇VO2max
Maximal oxygen uptake
Communicated by Jean-René Lacour.
This article is submitted as part of the Topical Collection on
‘Energetics of Human Locomotion’.
* Richard C. Blagrove
[email protected]
1
School of Sport, Exercise and Health Sciences,
Loughborough University, Epinal Way, Loughborough,
Leicestershire, UK
2
Department of Sport, Exercise and Rehabilitation,
Northumbria University, Newcastle-upon-Tyne, UK
3
Water Research Group, Northwest University, Potchefstroom,
South Africa
4
School of Sport, Health and Applied Science, St Mary’s
University, Twickenham, UK
5
Orreco Ltd, National University of Ireland Business
Innovation Centre, Galway, Ireland
6
Division of Surgery and Interventional Science, University
College London, London, UK
1866
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
Introduction
Distance running performance is largely dependent upon
aerobic factors, including maximal oxygen uptake ( ̇V
O2max), running economy (RE) and the fraction of ̇V
O2max utilized over a given distance (Bassett and Howley
2000; Brandon 1995). Although the variability in distance
running performance can largely be explained by ̇VO2max
in heterogeneous groups of runners, RE and fractional uti-
lization are better capable of predicting performance in
runners homogenous for ̇VO2max (Conley and Krahenbuhl
1980). Specifically, in middle-distance events, a model that
included ̇VO2max and RE, was capable of explaining 96%
of the variance in performance in highly trained 800-m
and 1500-m runners (Ingham et al. 2008). It has recently
been suggested that expressing physiological parameters
in terms of aerobic energy cost ( ̇Eaero) provides greater
validity for quantifying exercise intensity compared to
traditional oxygen cost ( ̇Oaero)-based measurements (Beck
et al. 2018); however, these claims have not yet been fully
examined with experimental data.
The expression of aerobic factors in units of ̇Oaero is
limited because this measure does not account for differ-
ences in substrate utilization, which can vary substantially
between runners operating at the same oxygen uptake ( ̇V
O2) (Brooks and Mercier 1994; Fletcher et al. 2009). It has
been suggested that RE should, therefore, be quantified
as Ėaero, which provides a more accurate reflection of the
metabolic cost of exercise (Shaw et al. 2014). Previous
reports have confirmed that ̇Eaero provides a more sensitive
measure of RE compared to ̇Oaero across range of intensi-
ties in highly trained runners (Fletcher et al. 2009; Shaw
et al. 2014); however, this has not yet been established in
lesser trained populations of runners, such as adolescents.
̇Eaero appears to provide a more reliable measurement of
RE compared to ̇Oaero in high-performing adolescent run-
ners (Blagrove et al. 2017); however validity-related issues
associated with these measures have not previously been
scrutinized in this age group.
The physiological determinants of performance for
adolescents are similar to those of adult runners. A num-
ber of investigations have confirmed that ̇VO2max has a
moderate–good correlation (r = 0.5–0.9) with performance
over 1.5 km (Abe et al. 1998; Almarwaey et al. 2003),
3 km (Abe et al. 1998; Mahon et al. 1996; Unnithan et al.
1995), and 5 km (Abe et al. 1998; Cole et al. 2006; Cun-
ningham 1990) in young (10–18 years) groups of runners.
Measures of RE quantified in units of ̇VO2 also appear to
be related to middle-distance performance (Almarwaey
et al. 2003; Mayers and Gutin 1979; Unnithan et al. 1995).
Additionally, speed at ̇VO2max (s ̇VO2max) (Abe et al. 1998;
Almarwaey et al. 2003; Cole et al. 2006; Cunningham
1990) and fractional utilization calculated in ̇VO2 terms
have also been shown to significantly correlate with dis-
tance running performance in adolescents (Mahon et al.
1996; Unnithan et al. 1995). Despite these findings, for ̇E
aero to possess greater criterion validity compared to ̇Oaero,
it should be capable of predicting performance times with
greater accuracy. This direct comparative analysis of two
different approaches to quantifying aerobic-based determi-
nants of performance has not previously been performed
and is important for establishing validity of these metrics.
Moreover, the method used to partition groups of young
participants for differences in body size for variables
such as ̇VO2max and RE is also likely to influence findings
(Eisenmann et al. 2001). Previous studies have normalized
to body mass as a simple ratio (Abe et al. 1998; Almar-
waey et al. 2003; Mahon et al. 1996; Unnithan et al. 1995);
however, this is unlikely to appropriately partition out the
confounding influence of body size (Loftin et al. 2016).
It has been proposed that fractional utilization expressed as
the ratio between ̇Eaero and maximal aerobic energy expendi-
ture ( ̇Eaeromax) at lower intensities (respiratory exchange ratio
(RER) < 1.0) provides a numerically lower relative aerobic
intensity compared to fractional utilization quantified as % ̇V
O2max (Beck et al. 2018). This has important implications for
prescription of aerobic exercise intensity and for quantifying
the physiological outcomes to training or nutritional interven-
tions. Although this difference has been established in elite
race walkers (Beck et al. 2018), no papers have attempted to
compare these two approaches for other exercise modalities
and sub-elite populations. Moreover, small differences in the
predictive power of physiological determinants (expressed in
̇Eaero or ̇Oaero terms) on performance times may provide greater
deterministic accuracy when combined as part of a multiple-
factor regression model.
Consequently, the primary purpose of this study was to
examine the relationship between physiological variables,
quantified as both ̇Eaero and ̇Oaero, and race performances in a
group of competitive post-pubertal adolescent middle-distance
runners. The secondary aims were to investigate the influence
of running speed on RE quantified as both ̇Oaero and ̇Eaero, and
examine whether expressing relative aerobic intensity as % ̇V
O2max and % ̇Eaeromax produces a different slope of values
across a range of speeds. It was hypothesized that ̇Eaero would
provide a more valid means of expressing important aerobic
performance determinants compared to ̇Oaero.
1867
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
Methods
Participants
Following institutional level ethical approval and in accord-
ance with the Helsinki declaration, 56 competitive male
(n = 34) and female (n = 22) middle-distance (0.8–3 km)
runners (15–18 years) volunteered to take part in this study.
Participant descriptive statistics are displayed in Table 1. All
participants possessed at least 2 years of distance running
training and racing experience, were familiar with treadmill
running and considered middle-distance running to be their
main sport. Participants were informed of the requirements
and risks associated with the study and thereafter signed
consent to participate was obtained from a parent or guard-
ian, or the participant themselves if > 18 years.
Procedure
All trials were conducted in the same laboratory under simi-
lar environmental conditions (temperature 16–20 °C; relative
humidity, 29–54%; barometric pressure, 746–773 mmHg).
Participants were instructed to avoid strenuous exercise in
the 48 h preceding the trial, and arrive at least 2 h post-
prandial. Upon arrival at the laboratory, stature and sitting
height were measured with a stadiometer (SECA GmbH &
Co., Hamburg, Germany) to the nearest 0.01 m, and matu-
rity offset was predicted for each participant using published
formulae (Moore et al. 2015). Body mass was recorded
with digital scales (MPMS-230, Marsden Weighing Group,
Oxfordshire, UK) to the nearest 0.1 kg.
All exercise testing was performed on the same motor-
ised treadmill (HP Cosmos Pulsar 4.0, Cosmos Sports &
Medical GmbH, Munich, Germany). Throughout the test-
ing, participants breathed through a low-dead space mask
to monitor expired air via an open-circuit metabolic cart
(Oxycon Pro, Erich Jaeger GmbH, Germany). Before test-
ing, gas analysers were calibrated with known gas concen-
trations (16% O2; 5% CO2) and ventilation measurement
with a 3-L syringe. Participants completed a standard-
ized warm-up involving a 5-min run at 2 km h−1 slower
than the pre-determined start speed for their exercise test.
Each test involved a sub-maximal discontinuous incre-
mental test followed by an incremental continuous test
to volitional exhaustion. The sub-maximal test involved
5–7 × 3-min stages interspersed with 30-s rest periods for
extraction of a 20 µL capillary blood sample. The sam-
ple was immediately haemolysed in a micro-test tube and
tested for blood lactate (Biosen C-Line, EKF Diagnostic,
Ebendorfer Chaussee 3, Germany). The start speed of the
test was determined using participants’ best race times
and published recommendations (Jones 2006). Speed
was increased by 1 km h−1 every stage until lactate turn-
point (LTP) had been surpassed, which was defined as the
speed before a rise of > 1 mMol L−1 compared to the pre-
vious stage. The gradient of the treadmill remained at 1%
throughout the sub-maximal test (Jones and Doust 1996).
Following a 5-min passive recovery, participants ran
continuously at the speed corresponding to their LTP
(sLTP). At the end of each minute, the treadmill gradi-
ent was increased by 1% until volitional exhaustion was
reached (typically 6–8 min).
Physiological measures
Sub‑maximal measures
Prior to analysis of expired gases, data were filtered to
remove any values that were deemed to represent errant
breaths (Lamarra et al. 1987). The absence of a ̇VO2 slow
component was verified by calculating the difference
between the first 30 s of the final minute and the last 30 s. A
difference less than the minimal detectable change (MDC),
calculated as standard error of the mean × 1.96 ×
√
2, con-
firmed a ̇VO2 steady state had been achieved. The final 60 s
of each submaximal stage was averaged for ̇VO2, volume
of expired CO2 and RER. Updated non-protein quotient
equations (Peronnet and Massicotte 1991) and RER val-
ues were used to estimate ̇Eaero at each speed. Values for
the sLTP and the three speeds prior (sLTP − 1 km.h−1,
sLTP − 2 km.h−1, sLTP − 3 km.h−1) were used as the
measure of RE, and quantified as both ̇Oaero and ̇Eaero. For
each of the four submaximal speeds, the intensity relative
to each participants ̇VO2max or ̇Eaeromax was calculated
and expressed as a percentage. Fractional utilization at
the speed corresponding to lactate threshold (sLT) was
also quantified. sLT was defined as the final speed prior
to an initial rise (≥ 0.2 mmol L−1) of blood lactate from
baseline, which is greater than the typical error of meas-
urement at this speed in a similar cohort (10).
Table 1 Descriptive characteristics of the study participants
̇VO2max maximal oxygen uptake, sLT speed at lactate threshold, s ̇VO
2max speed at ̇VO2max
Measure
Males (n = 34)
Females (n = 22)
Age (year)
17 ± 1
17 ± 1
Stature (m)
1.76 ± 0.06
1.69 ± 0.06
Body mass (kg)
62.5 ± 6.4
52.7 ± 5.8
̇VO2max (ml kg−1 min−1)
70.1 ± 7.2
61.1 ± 6.4
sLT (km h−1)
13.4 ± 1.5
11.7 ± 1.3
s ̇VO2max (km h−1)
19.2 ± 1.5
17.0 ± 1.5
1868
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
Maximal measures
The highest average ̇VO2 attained within a 30-s period dur-
ing the maximal test was defined as a participant’s ̇VO2max.
Confirmation that ̇VO2max had been attained was identified
using an objective procedure (Midgley et al. 2009). A pre-
dicted ̇VO2max was calculated using the linear regression
line obtained from the ̇VO2 data between + 2 min following
the start of the test and − 2 min prior to exhaustion. A pla-
teau was confirmed if the difference between the predicted
and recorded ̇VO2max values was greater than 0.5 times the
regression gradient. ̇Eaeromax was obtained by multiply-
ing the ̇VO2max value (expressed in mL min−1) by 21.745
joules. s ̇VO2max was identified by substitution of values into
the linear regression equation representing the ̇VO2–speed
relationship from the sub-maximal running assessment.
Allometric scaling
It is well recognised that when expressing variables relative
to body size, the use of ratio scaling is inappropriate (Tan-
ner 1949). Consequently, when comparing youth perform-
ers or individuals of different sexes, allometric expression
of variables is more appropriate (Curran-Everett 2013). To
obtain allometrically scaled exponents for the population
under investigation, body mass and ̇VO2 data were log trans-
formed and linear regression lines compared for males and
females using an analysis of co-variance (ANCOVA) model.
Results revealed homogeneity of regression for the slopes of
all variables, thus a common scaling exponent was derived
on the logarithmic transformed data sets. The appropriate-
ness of the power function was confirmed using an absence
of relationships derived from the linear regression correla-
tions between body mass and ̇VO2 scaled values.
Performance measures
Participant’s best times over 0.8, 1.5 and 3 km during com-
petitive track races, within 60 days (41 ± 17 days) of labora-
tory testing, were converted to running speed as an index of
performance.
Statistical analysis
Data were analysed with IBM SPSS Statistics (v24) and
values are displayed as mean ± standard deviation (SD). A
p value of < 0.05 was used to denote statistical significance.
Normality in distribution of the dependent variables was
assessed using the Shapiro–Wilk statistic and homogene-
ity of variance with Levene’s test. Data from a number of
race distances did not conform to this assumption, thus run-
ning speeds were log-transformed prior to further analysis.
Normality associated with the standardized residual errors
was assessed using probability plots and confirmed objec-
tively using the standard residual statistic. Homoscedastic-
ity was assessed using scatterplots of the residual errors
and predicted values. Several variables displayed multi-
collinearity, defined as an r value > 0.7. Speed at LTP and
s ̇VO2max were, therefore, analysed as separate independent
variables with one-tailed Pearson correlation tests. For each
race distance, multiple linear regression models were used
to examine the combined influence of predictors expressed
in terms of the ̇Oaero measures, and predictors quantified as
̇Eaero. Zero-order correlation statistics were used to interpret
the relationship with each variable in the model. To compare
the correlation statistics for ̇Oaero-related measures against
those expressed as ̇Eaero, a 95% confidence interval (CI)
was calculated for each result. Correlation coefficients were
interpreted as ≤ 0.30 negligible correlation, 0.31–0.50 low
correlation, 0.51–0.70 moderate correlation, 0.71–0.90 high
correlation, > 0.90 very high correlation (Hinkle et al. 2003).
A one-way repeated measures analysis of variance
(ANOVA) was performed to evaluate the differences
between ̇Oaero and ̇Eaero across four relative running speeds.
Differences between % ̇VO2max and % ̇Eaeromax were
assessed using a two-way (measure × speed) ANOVA and
the differences between individual relative speeds was
analysed using a one-way ANOVA. Bonferroni post hoc
adjustments were used to detect any significant differences
between individual speeds or measures.
Results
Performance times for males and females are shown in
Table 2. Allometric scaling revealed exponents that approxi-
mated three-quarters for ̇VO2 at each speed [sLTP: b = 0.77
(95% CI 0.54–0.99), sLTP − 1 km h−1: b = 0.77 (95% CI
0.54–0.99), sLTP − 2 km h−1: b = 0.78 (95% CI 0.56–0.99),
sLTP − 3 km h−1: b = 0.84 (95% CI 0.64–1.05)] and ̇VO2max
[b = 0.74 (95% CI 0.48–1.00)]. Applying this power function
(b = 0.75) revealed an absence of any significant relationship
between body mass and scaled ̇VO2 across the intensities
assessed (r ≤ 0.14, p ≥ 0.36).
Table 2 shows a high level of similarity between the
correlation coefficients for the two methods used to quan-
tify aerobic energy expenditure. Multiple-regression
analysis revealed that the independent variables of mean
RE, fractional utilization at sLT, and ̇VO2max, accounted
for > 80% and > 70% of the variance in 3 km perfor-
mance in males and females, respectively (p < 0.001).
These three variables were also significant predictors of
0.8 km (p < 0.01) and 1.5 km (p < 0.001) performance in
males, but were poor predictors of 1.5 km performance in
females. s ̇VO2max and sLTP tended to correlate strongly
1869
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
Table 2 Performance time (mean ± standard deviation), coefficients (with 95% CI) of multiple regression, and Pearson correlations for physiological variables and running performance speed in
males and females
̇Oaero aerobic oxygen cost, ̇Eaero aerobic energy cost, ̇VO2max maximal oxygen uptake, ̇Eaeromax maximal aerobic energy expenditure, s ̇VO2max speed at ̇VO2max, sLTP speed at lactate turn
point
a p < 0.05, bp < 0.01, cp < 0.001
A Variables used in multiple regression analysis
Distance
Time (s)
Multiple
regression
adjusted r2
Mean running economyA
Fractional utilization at sLTA
̇VO2maxA
s ̇VO2max
sLTP
̇Oaero
̇Eaero
̇Oaero
̇Eaero
%̇V̇O2max
% ̇Eaeromax
Males
0.8 km (n = 21)
120.8 ± 8.8
0.40b 0.39b − 0.53a (− 0.78 to
− 0.13)
− 0.52a (− 0.78 to
− 0.12)
0.19 (− 0.26 to
0.57)
0.21 (− 0.25 to
0.59)
0.14 (− 0.31 to
0.55)
0.45 (− 0.01 to
0.72)
0.60b (0.20 to 0.81)
1.5 km (n = 34)
250.8 ± 17.9 0.56c 0.57c − 0.33 (− 0.60 to
0.01)
− 0.39a (− 0.64 to
− 0.06)
− 0.03 (− 0.31 to
0.36)
− 0.04 (− 0.29 to
0.38)
0.55b (0.27 to
0.75)
0.75c (0.55 to
0.87)
0.78c (0.60 to 0.89)
3 km (n = 21)
539.5 ± 43.5 0.84c 0.85c − 0.60b (− 0.82 to
− 0.23)
− 0.63b (− 0.83 to
− 0.27)
− 0.15 (− 0.52 to
0.33)
− 0.14 (− 0.50 to
0.34)
0.77c (0.52 to
0.91)
0.93c (0.77 to
0.97)
0.90c (0.77 to 0.96)
Females
0.8 km (n = 16)
136.6 ± 3.7
0.58b 0.44a − 0.66b (− 0.87 to
− 0.25)
− 0.40 (− 0.74 to
0.13)
0.02 (− 0.49 to
0.50)
0.07 (− 0.43 to
0.55)
0.52a (0.04 to
0.81)
− 0.01 (− 0.50 to
0.49)
0.22 (− 0.31 to
0.64)
1.5 km (n = 22)
281.4 ± 11.8 0.10
0.11
− 0.34 (− 0.68 to
0.08)
− 0.37 (− 0.69 to
0.05)
− 0.29 (− 0.64 to
0.15)
− 0.33 (− 0.66 to
0.11)
0.36 (− 0.08 to
0.68)
0.42 (0 to 0.72)
0.55a (0.17 to 0.79)
3 km (n = 16)
622.0 ± 36.3 0.79c 0.73c − 0.57a (− 0.83 to
− 0.11)
− 0.59a (− 0.84 to
− 0.14)
− 0.64b (− 0.86 to
− 0.22)
− 0.66b (− 0.87 to
− 0.25)
0.77c (0.45 to
0.92)
0.84c (0.59 to
0.94)
0.85c (0.62 to 0.95)
1870
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
with performance over longer distances but the relation-
ships were weaker for 0.8 km in both sexes (Table 2).
ANOVA revealed a significant decrease in ̇Oaero as run-
ning speed increased (F = 11.59, p < 0.001, Fig. 1). Post
hoc analysis revealed significant differences between ̇Oaero
at a number of individual speeds (Fig. 1) and two other
comparisons (sLTP vs sLTP − 1 km h−1, sLTP − 1 km h−1
vs sLTP − 2 km h−1) approached significance (p = 0.07).
A significant effect of running speed was also noted for
̇Eaero (F = 4.74, p = 0.015, Fig. 2). Post hoc inspection
identified a difference between sLTP and sLTP − 1 km
h−1 (p = 0.02); however, the difference between sLTP and
sLTP − 2 km h−1 was close to the threshold of significance
(p = 0.06).
A significant main effect between the slopes of the lines
was detected for % ̇VO2max and % ̇Eaeromax when plotted
against relative running speed (F = 5.38, p = 0.021); how-
ever, there was an absence of an interaction effect (meas-
ure × speed; F = 0.29, p = 0.834). One-way ANOVA analysis
was also not able to locate any difference between measures
at each relative speed.
Discussion
The primary aim of this study was to examine the relation-
ship between race performances and several important aero-
bic variables quantified as both ̇Eaero and ̇Oaero in adolescent
Fig. 1 Oxygen cost ( ̇Oaero)
for speed at lactate turnpoint
(LTP) and the three speeds
prior (n = 56). aSignificantly
different from speed at LTP
(p < 0.01), bsignificantly differ-
ent from speed at LTP-1 km h−1
(p = 0.01)
Fig. 2 Aerobic energy cost ( ̇E
aero) for speed lactate turnpoint
(LTP) and the three speeds prior
(n = 56). aSignificantly different
from speed at LTP (p = 0.02)
1871
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
middle-distance runners. Results indicate that ̇Eaero does not
provide a greater level of criterion validity compared to ̇O
aero-based measurements in this age group for the middle-
distance running events. The study also aimed to investigate
the validity of ̇Oaero and ̇Eaero as a means of quantifying RE.
Results showed differences in the manner ̇Oaero and ̇Eaero
change with increasing running speed, with ̇Oaero displaying
a decrease and ̇Eaero a curvilinear response. A further find-
ing was that the relationship between relative running speed
and the fraction of ̇VO2max or ̇Eaeromax that is accessed
also appears to differ, with the difference being greater at
lower intensities. These findings provide new insight into the
ongoing debate surrounding the most appropriate method of
expressing aerobic fitness parameters, which are typically
used to evaluate performance, health status and monitor
improvement.
Results of the multiple regression analysis show that
using ̇Eaero to quantify RE and fractional utilization, instead
of traditional ̇Oaero, provides no additional value in the pre-
diction of middle-distance running performance in adoles-
cents. To alter the strength of the relationship between ̇E
aero-based determinants and performance, a high-level of
inter-individual variability in substrate utilization is required.
This was not apparent as coefficient of variation (SD/mean)
for the RER values at each relative speed was ~ 4%. The
participants used in this study possessed somewhat homog-
enous physiological characteristics (Table 1), thus future
research could investigate a more heterogeneous sample
of runners, who are likely to differ more substantially in
terms of their consumption of substrates at the same rela-
tive speeds. Similarly, a relatively small range of running
speeds (sLTP to sLTP − 3 km h−1) was examined in the
present study and measurement stages were relatively short
(3 min), which resulted in mean RER values > 0.9. A larger
range of speeds and longer sampling duration would pro-
duce lower RER values (Van Loon et al. 2001) and may
have generated more substantial between-participant vari-
ability as maximal rates of lipid oxidation rates are known
to occur at ~ 65% ̇VO2max but is dependent upon training
status (Achten and Jeukendrup 2004). This would, there-
fore, alter the predictive power of variables quantified in ̇E
aero units. Nevertheless, it is also questionable that slower
running speeds would correlate well with middle-distance
performance given the large discrepancy between low-inten-
sity running and middle-distance race speed. Although ̇E
aero also accounts for the energy yield associated with work
performed during sub-maximal exercise compared to ̇Oaero
(Shaw et al. 2014), there are numerous other factors that also
govern these parameters, including use of stretch-shortening
cycle mechanisms, muscle activation in the musculotendi-
nous unit, running kinematics and anthropometric factors
(Barnes and Kilding 2015). Thus, it may also be the case
that the additional metabolic insight, which ̇Eaero provides,
is insufficient to alter the predictive capacity of these aerobic
parameters on performance. Within study designs that assess
participants at more than one point in time, expressing RE as
̇Eaero, rather than ̇Oaero, is likely to provide the most scientifi-
cally robust metric (Blagrove et al. 2017; Shaw et al. 2014).
To the author’s knowledge, this is the first study to apply
a multiple regression model to middle-distance performance
in adolescent runners, using acknowledged aerobic deter-
minants of performance (Bassett and Howley 2000; Ing-
ham et al. 2008). Results showed that a high level (~ 80%,
p < 0.001) of inter-individual variability in male and female
3 km performance could be explained by RE, fractional
utilization at sLT and ̇VO2max (Table 2). Moreover, ~ 40%
(p < 0.01) of the 0.8 km performance variability and 56%
(p < 0.001) of male 1.5 km performance could be explained
using these same variables. Surprisingly, this regression
model could only predict a small (10%) amount of the vari-
ability in female 1.5 km performance. This may be due to the
homogeneity of the performance times in the female (5%)
compared to the male sample (7%) over 1.5 km.
Previous studies have shown the importance of ̇VO2max
for middle-distance (1.5 km and 3 km) performance in
children and adolescent (Abe et al. 1998; Mahon et al.
1996; Unnithan et al. 1995), which is largely confirmed by
the results of this investigation (r = 0.55–0.77, p < 0.01).
Based upon the non-overlap of the 95% CI with the cor-
relation coefficients, it is also apparent that ̇VO2max
becomes more important as a determinant of performance
as race duration increases, which is in agreement with pre-
vious findings in adolescent (Almarwaey et al. 2003) and
adult runners (Ingham et al. 2008; Padilla et al. 1992). It
is likely that this pattern in results reflects the increasing
proportion of ̇VO2max that is attained as race duration
increases in middle-distance events (Brandon 1995). This
is also the case for s ̇VO2max and sLTP, both of which
show high (r > 0.84, p < 0.001) correlations with 3 km per-
formance in males and females but weaker correlations
at the shorter distances (Table 2). RE, as an independent
factor, is not thought to be important for middle-distance
running performance (Ingham et al. 2008) despite sev-
eral studies observing significant relationships in young
runners (Almarwaey et al. 2003; Mayers and Gutin 1979;
Unnithan et al. 1995). When RE was expressed as ̇Eaero,
it generally showed low–moderate negative relationships
(r = − 0.37 to − 0.63) with performance, which did not dif-
fer across race distances (Table 2). Relationships were sig-
nificant for male participants across all distances (p < 0.05)
and females only at 3 km (r = − 0.57, p < 0.05), which is
in agreement with the previous findings (Almarwaey et al.
2003). In adolescent middle-distance running, it, therefore,
appears that RE influences race performance, but explains
1872
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
a relatively small proportion of inter-individual variability.
It is possible that participants who have a low ̇VO2max
compensate by possessing better RE (Cunningham 1990).
This may explain the low relationship (r = 0.14) between
̇VO2max and 0.8 km performance in males but moderate
relationship (r = − 0.52, p < 0.05) between ̇Eaero and per-
formance over this distance.
Results demonstrate that the RE–speed relationship
differed depending upon the strategy used to quantify
RE. When expressed as ̇Oaero, running became less meta-
bolically expensive as a function of speed (F = 11.59,
p < 0.001, Fig. 1), which is in agreement with previous
findings (Iaia et al. 2009) but in contrast to others who
have shown no change (Fletcher et al. 2009; Shaw et al.
2014) or an increase (Fletcher et al. 2013) in ̇Oaero as speed
increases. This discrepancy between findings is likely due
to the range of speeds examined in each study and the
training status of participants. Similar to the study by Iaia
et al. (2009), the speeds selected in the present study rep-
resent the upper end of the range over which RE can be
measured with high validity (≤ LTP, RER < 1.0), whereas
others have utilized a lower range of relative intensities
(Fletcher et al. 2009; Shaw et al. 2014). Furthermore, pre-
vious studies used highly trained runners (Fletcher et al.
2009; Shaw et al. 2014), who were assessed at faster abso-
lute speeds compared to the young runners recruited in the
present study. When quantified as ̇Oaero (per unit distance),
a faster range of absolute speeds tends to produce a flatter
relationship compared to oxygen cost at slower absolute
speeds.
Conversely, when RE was quantified as ̇Eaero, a sub-
tle ‘U-shaped’ profile was apparent across the range of
speeds (Fig. 3), with a significant difference noted between
sLTP and sLTP − 1 km h−1 (p = 0.02) and a near-sig-
nificant difference between sLTP and sLTP − 2 km h−1
(p = 0.06). A curvilinear relationship between ̇Eaero and
speed has been observed in a number of studies (Black
et al. 2018; Rathkey and Wall‐Scheffler 2017; Steudel-
Numbers and Wall-Scheffler 2009; Willcockson and
Wall-Scheffler 2012), with the nadir representing the
most economical running speed. The least energetically
expensive speed (at sLTP − 2 km h−1) in the present study
was 13.4 ± 1.7 km h−1, which is similar to the 13 km h−1
(Black et al. 2018) and 12.6 km h−1 (Steudel-Numbers
and Wall-Scheffler 2009; Willcockson and Wall-Scheffler
2012) reported previously in similarly trained participants.
It is likely that other studies that have observed linear ̇E
aero–speed relationships have used a range of speeds that
did not capture the lowest point of the curve (Fletcher et al.
2009; Shaw et al. 2014) or used a lesser trained group of
runners (Black et al. 2018).
Crucially, the trend for an increase in ̇Eaero with faster
speeds above sLTP − 2 kmh−1 is in opposition to the rela-
tionship demonstrated between ̇Oaero and running speed. ̇E
aero represents a theoretically more valid measure of RE as
it estimates actual energy turnover, whereas ̇Oaero is sim-
ply a measure of the ̇VO2 per unit of running distance. The
increase in ̇Eaero as speed progressed from sLTP − 2 km h−1
towards sLTP, therefore, reflects the increase in RER value,
indicating an increased reliance on carbohydrate as an
energy source. As running speed increases, joint angular
velocities are greater and ground contact time reduces,
which requires a greater reliance on metabolically ineffi-
cient type II muscle fibres (Fletcher and MacIntosh 2017).
An increased recruitment of high threshold motor units is
likely to be the mechanism that drives the rate of carbohy-
drate utilization, and as the energy yield from carbohydrates
per mole of O2 is also greater than lipids (Jeukendrup and
Wallis 2005), this generates higher ̇Eaero at faster speeds. It
is, therefore, recommended that ̇Eaero should be used as a
measure of RE as this provides a more valid indicator of the
metabolic demand of running compared to ̇Oaero.
Fig. 3 Percentage utiliza-
tion of maximum oxygen
uptake and maximum aerobic
energy expenditure across four
sub-maximal relative speeds
(n = 56). sLTP speed at lactate
turnpoint
1873
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
Exercise intensity is often prescribed relative to an
individual’s ̇VO2max, thus expression of relative aerobic
intensity as a percentage of ̇Eaeromax would be more mean-
ingful. However, prescribing exercise intensity on either
basis has been criticised due to the heterogeneity at which
other important exercise thresholds (critical speed, anaer-
obic threshold, lactate threshold) occur (Baldwin et al.
2000; Scharhag-Rosenberger et al. 2010). Therefore, an
intensity expressed relative to either ̇VO2max or ̇Eaeromax
may represent a severe intensity (relative to critical power)
in one individual but provide a steady-state condition for
another individual. We attempted to account for this in
the present study by comparing the relationship between
% ̇VO2max or % ̇Eaeromax across a range of running speeds
expressed relative to each individual’s sLTP. A significant
main effect method of measurement (% ̇VO2max versus % ̇E
aeromax) was detected (F = 5.38, p = 0.021); however, no
differences were identified at individual relative speeds.
The divergent nature of the gradients (Fig. 3) as relative
intensity decreases suggests that at slower relative speeds,
the use of % ̇Eaeromax becomes more important. Therefore,
it is recommended that if exercise is prescribed based on
maximal aerobic values, intensity is expressed as a frac-
tion of ̇Eaeromax, rather than ̇VO2max. However, a superior
method for prescribing running intensity is to base calcula-
tions on sLTP (or a similar objective metabolic threshold),
which would reduce inter-individual differences in relative
intensity, thus providing a more valid strategy.
This study has several limitations that should be
acknowledged. First, physiological testing predominantly
took place during the pre-season or early competitive rac-
ing period, with the duration between a participants’ race
performance and laboratory testing typically 3–8 weeks.
Although every attempt was made to minimize this time
gap, small changes in the physiological profile of partici-
pants cannot be discounted, which may have influenced the
results. Second, participants performed laboratory testing
2 h post-prandial; however, it is less certain whether this
requirement was adhered to prior to races. Participants
possessed ≥ 2 years’ racing experience; therefore, it is
unlikely that subtle differences in pre-race routine con-
found the results to a large extent. Third, middle-distance
running performance is limited by anaerobic factors, in
addition to the aerobic determinants measured in this
study (Thompson 2017). These anaerobic variables were
not quantified in this investigation and are likely to explain
a large proportion of the variability in performance cur-
rently unaccounted for in the regression models. Moreo-
ver, investigating the determinants of longer race distances
(≥ 5 km), which have a greater reliance on aerobic sources
of energy, would also have been of interest in this age
group.
Conclusions
Expression of RE and fractional utilization in terms of ̇E
aero rather than ̇Oaero does not appear to alter the ability
of these determinants to predict middle-distance running
performance in adolescents. RE, fractional utilization at
sLT and ̇VO2max accounted for approximately 80% of the
variability in 3 km performance in adolescent males and
females. These variables could explain less (40–60%) of
the variation in performance over shorter race distances
and very little over 1.5 km in females. s ̇VO2max and sLTP
were confirmed as other important indicators of middle-
distance performance in adolescent runners with the
strength of relationships tending to be greater over longer
distances. Results also indicate markedly different pro-
files in the ̇Oaero–speed response compared to ̇Eaero–speed
relationship. It is recommended that RE is quantified in
̇Eaero units, which provides a more valid reflection of the
metabolic demand of running across a range of speeds.
Finally, there were differences observed in the slope of the
relationships between running speed and the proportion of
̇VO2max or ̇Eaeromax utilized at each speed, suggesting this
should be accounted for if prescribing exercise intensity
using this method.
Acknowledgements The authors would like to thank the participants
and their parents/guardians for the time they committed to this study.
The technical support provided by Ian Grant is also greatly appreciated.
Author contributions RB and PH conceived and designed research. RB
conducted experiments, analyzed data and wrote the manuscript. All
authors proof read, contributed to editing, and approved the manuscript.
Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of
interest.
Open Access This article is distributed under the terms of the Creative
Commons Attribution 4.0 International License (http://creativecom-
mons.org/licenses/by/4.0/), which permits unrestricted use, distribu-
tion, and reproduction in any medium, provided you give appropriate
credit to the original author(s) and the source, provide a link to the
Creative Commons license, and indicate if changes were made.
References
Abe D, Yanagawa K, Yamanobe K, Tamura K (1998) Assessment of
middle-distance running performance in sub-elite young run-
ners using energy cost of running. Eur J Appl Physiol Occup
Physiol 77:320–325. https ://doi.org/10.1007/s0042 10050 340
Achten J, Jeukendrup AE (2004) Optimizing fat oxidation through
exercise and diet. Nutrition 20:716–727
1874
European Journal of Applied Physiology (2019) 119:1865–1874
1 3
Almarwaey OA, Jones AM, Tolfrey K (2003) Physiological cor-
relates with endurance running performance in trained adoles-
cents. Med Sci Sports Exerc 35:480–487
Baldwin J, Snow RJ, Febbraio MA (2000) Effect of training status
and relative exercise intensity on physiological responses in
men. Med Sci Sports Exerc 32:1648–1654
Barnes KR, Kilding AE (2015) Running economy: measurement,
norms, and determining factors. Sports Med Open 1:8
Bassett DR Jr, Howley ET (2000) Limiting factors for maximum oxy-
gen uptake and determinants of endurance performance. Med
Sci Sports Exerc 32:70–84
Beck ON, Kipp S, Byrnes WC, Kram R (2018) Use aerobic energy
expenditure instead of oxygen uptake to quantify exercise intensity
and predict endurance performance. J Appl Physiol 125:672–674
Black MI, Handsaker JC, Allen SJ, Forrester SE, Folland JP (2018)
Is there an optimal speed for economical running? Int J Sports
Physiol Perform 13:75–81
Blagrove RC, Howatson G, Hayes PR (2017) Test–retest reliability of
physiological parameters in elite junior distance runners follow-
ing allometric scaling. Eur J Sport Sci 17:1231–1240. https ://doi.
org/10.1080/17461 391.2017.13643 01
Brandon LJ (1995) Physiological factors associated with middle dis-
tance running performance. Sports Med 19:268–277
Brooks GA, Mercier J (1994) Balance of carbohydrate and lipid uti-
lization during exercise: the "crossover" concept. J Appl Physiol
76:2253–2261
Cole AS, Woodruff ME, Horn MP, Mahon AD (2006) Strength, power,
and aerobic exercise correlates of 5-km cross-country running
performance in adolescent runners. Pediatr Exerc Sci 18:374–384
Conley DL, Krahenbuhl GS (1980) Running economy and distance
running performance of highly trained athletes. Med Sci Sports
Exerc 12:357–360
Cunningham LN (1990) Relationship of running economy, ventilatory
threshold, and maximal oxygen consumption to running perfor-
mance in high school females. Res Q Exerc Sport 61:369–374
Curran-Everett D (2013) Explorations in statistics: the analysis of ratios
and normalized data. Adv Physiol Educ 37:213–219. https ://doi.
org/10.1152/advan .00053 .2013
Eisenmann JC, Pivarnik JM, Malina RM (2001) Scaling peak VO2 to
body mass in young male and female distance runners. J Appl
Physiol 90:2172–2180
Fletcher JR, MacIntosh BR (2017) Running economy from a mus-
cle energetics perspective. Front Physiol 8:433. https ://doi.
org/10.3389/fphys .2017.00433
Fletcher JR, Esau SP, Macintosh BR (2009) Economy of running:
beyond the measurement of oxygen uptake. J Appl Physiol
107:1918–1922. https ://doi.org/10.1152/jappl physi ol.00307 .2009
Fletcher JR, Pfister TR, MacIntosh BR (2013) Energy cost of running
and Achilles tendon stiffness in man and woman trained runners.
Physiol Rep 1:e00178
Hinkle DE, Wiersma W, Jurs SG (2003) Applied statistics for the
behavioral sciences, 5th edn. Houghton Mifflin, Boston
Iaia FM, Hellsten Y, Nielsen JJ, Fernstrom M, Sahlin K, Bangsbo J
(2009) Four weeks of speed endurance training reduces energy
expenditure during exercise and maintains muscle oxidative
capacity despite a reduction in training volume. J Appl Physiol
106:73–80. https ://doi.org/10.1152/jappl physi ol.90676 .2008
Ingham SA, Whyte GP, Pedlar C, Bailey DM, Dunman N, Nevill AM
(2008) Determinants of 800-m and 1500-m running performance
using allometric models. Med Sci Sports Exerc 40:345–350. https
://doi.org/10.1249/mss.0b013 e3181 5a83d c
Jeukendrup A, Wallis G (2005) Measurement of substrate oxidation
during exercise by means of gas exchange measurements. Int J
Sports Med 26:S28–S37
Jones AM (2006) Middle- and long distance running. In: Winter EMJA,
Davison RC, Bromley PD, Mercer TH (eds) BASES sport and
exercise physiology testing guidelines: Vol. 1—sport testing. Tay-
lor & Francis e-Library, Abingdon, pp 147–155
Jones AM, Doust JH (1996) A 1% treadmill grade most accurately
reflects the energetic cost of outdoor running. J Sports Sci
14:321–327
Lamarra N, Whipp BJ, Ward SA, Wasserman K (1987) Effect of inter-
breath fluctuations on characterizing exercise gas exchange kinet-
ics. J Appl Physiol 62:2003–2012
Loftin M, Sothern M, Abe T, Bonis M (2016) Expression of VO2peak
in children and youth, with special reference to allometric scaling.
Sports Med 46:1451–1460
Mahon A, Del Corral P, Howe C, Duncan G, Ray M (1996) Physi-
ological correlates of 3-kilometer running performance in male
children. Int J Sports Med 17:580–584
Mayers N, Gutin B (1979) Physiological characteristics of elite prepu-
bertal cross-country runners. Med Sci Sports 11:172–176
Midgley AW, Carroll S, Marchant D, McNaughton LR, Siegler J (2009)
Evaluation of true maximal oxygen uptake based on a novel set
of standardized criteria. Appl Physiol Nutr Metab 34:115–123
Moore SA, McKay HA, Macdonald H, Nettlefold L, Baxter-Jones A,
Cameron N, Brasher P (2015) Enhancing a somatic maturity pre-
diction model. Med Sci Sports Exerc 47:1755–1764
Padilla S, Bourdin M, Barthelemy JC, Lacour JR (1992) Physiological
correlates of middle-distance running performance. A compara-
tive study between men and women. Eur J Appl Physiol Occup
Physiol 65:561–566
Peronnet F, Massicotte D (1991) Table of nonprotein respiratory quo-
tient: an update. Can J Sport Sci 16:23–29
Rathkey JK, Wall-Scheffler CM (2017) People choose to run at their
optimal speed. Am J Phys Anthropol 163:85–93
Scharhag-Rosenberger F, Meyer T, Gassler N, Faude O, Kindermann
W (2010) Exercise at given percentages of VO2max: heterogene-
ous metabolic responses between individuals. J Sci Med Sport
13:74–79. https ://doi.org/10.1016/j.jsams .2008.12.626
Shaw AJ, Ingham SA, Folland JP (2014) The valid measurement of
running economy in runners. Med Sci Sports Exerc 46:1968–
1973. https ://doi.org/10.1249/MSS.00000 00000 00031 1
Steudel-Numbers KL, Wall-Scheffler CM (2009) Optimal running
speed and the evolution of hominin hunting strategies. J Hum
Evol 56:355–360
Tanner J (1949) Fallacy of per-weight and per-surface area standards,
and their relation to spurious correlation. J Appl Physiol 2:1–15
Thompson MA (2017) Physiological and biomechanical mechanisms
of distance specific human running performance. Integr Comp
Biol 57:293–300
Unnithan V, Timmons J, Paton J, Rowland T (1995) Physiologic cor-
relates to running performance in pre-pubertal distance runners.
Int J Sports Med 16:528–533
van Loon LJ, Greenhaff PL, Constantin-Teodosiu D, Saris WH, Wagen-
makers AJ (2001) The effects of increasing exercise intensity on
muscle fuel utilisation in humans. J Physiol 536:295–304
Willcockson MA, Wall-Scheffler CM (2012) Reconsidering the effects
of respiratory constraints on the optimal running speed. Med Sci
Sports Exerc 44:1344–1350
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
| Quantification of aerobic determinants of performance in post-pubertal adolescent middle-distance runners. | 06-17-2019 | Blagrove, Richard C,Howatson, Glyn,Pedlar, Charles R,Hayes, Philip R | eng |
PMC6195805 |
SDC Figure 1. Group Allocation
Assessed for eligibility
(n = 37)
Excluded (n = 7)
Not meeting inclusion criteria
(n = 6)
Other reasons (n = 1)
Randomized (n = 30)
Allocated to intervention
(n = 15)
Received allocated
intervention (n = 15)
Allocation
Enrollment
Allocated to intervention
(n = 15)
Received allocated
intervention (n = 15)
Follow up
Discontinued intervention
due to change of location
or adherence (n = 2)
Discontinued intervention
(n =2) due to an injury
sustained during the
protocol or an injury
sustained in a non-
invention related event
Analysis
Analyzed (n = 13)
Analyzed (n = 13)
| Wrist-worn Accelerometry for Runners: Objective Quantification of Training Load. | [] | Stiles, Victoria H,Pearce, Matthew,Moore, Isabel S,Langford, Joss,Rowlands, Alex V | eng |
PMC8306057 | International Journal of
Environmental Research
and Public Health
Communication
Effects of Acute Hypoxia on Lactate Thresholds and
High-Intensity Endurance Performance—A Pilot Study
Martin Faulhaber 1,2,*, Katharina Gröbner 1, Linda Rausch 1
, Hannes Gatterer 3
and Verena Menz 1
Citation: Faulhaber, M.; Gröbner, K.;
Rausch, L.; Gatterer, H.; Menz, V.
Effects of Acute Hypoxia on Lactate
Thresholds and High-Intensity
Endurance Performance—A Pilot
Study. Int. J. Environ. Res. Public
Health 2021, 18, 7573. https://
doi.org/10.3390/ijerph18147573
Academic Editors: Zbigniew
Jastrz˛ebski, Guillermo Felipe López
Sánchez, Łukasz Radzimi´nski and
Maria Skalska
Received: 8 June 2021
Accepted: 13 July 2021
Published: 16 July 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed
under
the
terms
and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
Department of Sport Science, University Innsbruck, 6020 Innsbruck, Austria;
[email protected] (K.G.); [email protected] (L.R.); [email protected] (V.M.)
2
Austrian Society of Alpine and High-Altitude Medicine, 6414 Mieming, Austria
3
Institute of Mountain Emergency Medicine, Eurac Research, 3910 Bolzano, Italy; [email protected]
*
Correspondence: [email protected]; Tel.: +43-512-507-45893
Abstract: The present project compared acute hypoxia-induced changes in lactate thresholds (meth-
ods according to Mader, Dickhuth and Cheng) with changes in high-intensity endurance performance.
Six healthy and well-trained volunteers conducted graded cycle ergometer tests in normoxia and in
acute normobaric hypoxia (simulated altitude 3000 m) to determine power output at three lactate
thresholds (PMader, PDickhuth, PCheng). Subsequently, participants performed two maximal 30-min cy-
cling time trials in normoxia (test 1 for habituation) and one in normobaric hypoxia to determine mean
power output (Pmean). PMader, PDickhuth and PCheng decreased significantly from normoxia to hypoxia
by 18.9 ± 9.6%, 18.4 ± 7.3%, and 11.5 ± 6.0%, whereas Pmean decreased by only 8.3 ± 1.6%. Cor-
relation analyses revealed strong and significant correlations between Pmean and PMader (r = 0.935),
PDickhuth (r = 0.931) and PCheng (r = 0.977) in normoxia and partly weaker significant correlations
between Pmean and PMader (r = 0.941), PDickhuth (r = 0.869) and PCheng (r = 0.887) in hypoxia. PMader
and PCheng did not significantly differ from Pmean (p = 0.867 and p = 0.784) in normoxia, whereas
this was only the case for PCheng (p = 0.284) in hypoxia. Although investigated in a small and
select sample, the results suggest a cautious application of lactate thresholds for exercise intensity
prescription in hypoxia.
Keywords: anaerobic threshold; high altitude; maximal lactate steady state
1. Introduction
Endurance training sessions under hypoxic conditions are part of altitude training
concepts for competitive athletes [1], as well as in preventive and therapeutic settings [2,3].
The positive effects of hypoxia application on sport performance and health outcomes
have been extensively described in literature, but there are also negative reports [4,5] that
should not be ignored, especially since negative health effects (e.g., an increased mechanic
stress against the cerebral vessel wall) are also suspected [5]. The determination of exercise
intensity zones plays a key role in regulating training adaptations and preventing under or
over strain. Anaerobic threshold concepts are very popular to prescribe intensity zones
for endurance training; however, it remains unclear whether such concepts are still valid
under hypoxic conditions. Despite its practical relevance, scientific literature dealing with
anaerobic threshold concepts in hypoxia is scarce and, for example, ventilatory thresholds
seem to be more reduced compared to lactate thresholds [6]. Thus, it remains unclear if
intensity zones based on lactate or ventilatory threshold concepts are adequate tools for
training prescription in hypoxia.
Recently, Weckbach et al. [7] reported that peak power output (Pmax) and power
output at different lactate thresholds (LT), derived from incremental exercise testing, were
significantly reduced in acute hypoxia (2650 m) compared to low altitude. Interestingly,
the presented data revealed that the reduction in LT power output was more pronounced
compared to Pmax by 20 to 90% depending on the LT concept [7]. However, it remains
Int. J. Environ. Res. Public Health 2021, 18, 7573. https://doi.org/10.3390/ijerph18147573
https://www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2021, 18, 7573
2 of 8
unclear whether the reduction in LT also reflects a similar decrement in high-intensity
endurance performance. If hypoxia-induced changes in LT differ markedly from changes
in maximal steady-state performance, the validity of these threshold concepts for exercise
prescription under acute hypoxic conditions would be limited. The present pilot study
should contribute to filling this knowledge gap and, therefore, compared different LTs,
derived from incremental exercise testing, with high-intensity endurance performance in
normoxic and acute hypoxic conditions. Based on a previous study evaluating endurance
performance at 3200 m [8] and data of Weckbach et al. at 2650 m [7], we hypothesised that
power output at the lactate thresholds would be more impaired by an acute exposure to
hypoxia compared to endurance performance.
2. Materials and Methods
2.1. Participants and Study Design
Six (three females, three males) students (age: 25 ± 3 years; body height: 176 ± 3 cm;
body weight: 71 ± 5 kg; altitude of residence: 677 ± 134 m) from the Department of
Sport Science (University of Innsbruck) volunteered to participate in this pilot study. All
of them were regularly active for more than 3 h per week including various disciplines
(as is common for active sport students), but none of them were competitive endurance
athletes. Participants completed a routine health screening using an adapted physical
activity readiness questionnaire (PAR-Q) before inclusion in the study. Medical clarification
by a physician was undertaken if the PAR-Q identified specific issues that required further
investigation. Exclusion criteria were pre-existing acute or chronic diseases, pregnancy or
lactation and regular smoking of more than five cigarettes per day.
Participants were informed about the experimental details and gave informed consent
before commencing the study. The study was carried out in conformity with the ethical
standards laid down in the 1975 Declaration of Helsinki. Since the study was designed as
a pilot project the identical protocol of the subsequent main study was approved by the
Board for Ethical Questions in Science of the University of Innsbruck, Austria (Certificate
of good standing, 12/2021).
The study was designed as a within-subject design (without cross-over). Participants
conducted five cycle ergometer (Cyclus 2, RBM, Leipzig, Germany) tests in a fixed order
(Figure 1). Tests 1 and 2 were maximal incremental tests in normoxic (test 1) and hypoxic
conditions (test 2). Test 1 and 2 were separated by a recovery period of 7 to 10 days.
Tests 3 to 5 were maximal 30-min time trials in normoxic (tests 3 and 4) and hypoxic
conditions (test 5), and tests were separated by recovery periods of 2 to 7 days. Test 3
served for habituation and was not included into statistical analyses. Adjustments of the
cycle ergometer (e.g., saddle height) were fixed before the first test and kept constant for
the subsequent tests. All tests took place in the laboratories of the University of Innsbruck
(Department of Sport Science, 590 m). Tests under hypoxic conditions were conducted in
a normobaric hypoxic chamber (LowOxygen Systems, Berlin, Germany) adjusted at an
inspiratory fraction of oxygen of 15.4% corresponding to a simulated altitude of about
3000 m. The hypoxic system provides a high air flow keeping the inspiratory fraction
of oxygen constant and to avoid an excessive increase in inspiratory fraction of carbon
dioxide as reported in previous studies [9–11].
Int. J. Environ. Res. Public Health 2021, 18, x FOR PEER REVIEW
3 of 8
Figure 1. Experimental design including the sequence of the tests and the breaks between tests.
2.2. Tests and Measurements
2.2.1. Maximal Incremental Tests
Participants rested for about 5 min in a sitting position on the cycle ergometer before
resting parameters were taken. Workload started at 80 W for female and 100 W for male
participants and was increased by 20 W every 3 min until subjective exhaustion. Heart
t
(M430 P l
Vi
A
t i )
it
d Bl
d l
t t
t
ti
(S
Figure 1. Experimental design including the sequence of the tests and the breaks between tests.
Int. J. Environ. Res. Public Health 2021, 18, 7573
3 of 8
2.2. Tests and Measurements
2.2.1. Maximal Incremental Tests
Participants rested for about 5 min in a sitting position on the cycle ergometer before
resting parameters were taken. Workload started at 80 W for female and 100 W for male
participants and was increased by 20 W every 3 min until subjective exhaustion. Heart
rate (M430, Polar, Vienna, Austria) was monitored. Blood lactate concentrations (Super
GL Ambulance, Dr. Müller Gerätebau, Freital, Germany) were analysed from capillary
blood samples taken from the hyperaemised earlobe at the end of the resting phase, during
the last 30 s of each stage and about 3 min after cessation of the test. Pmax was defined
as the last completed work rate plus the fraction of time spent in the final uncompleted
work rate multiplied by 20 W [12]. Maximal heart rate (HRmax) was defined as the highest
5-s average, and maximal blood lactate concentration (BLAmax) was considered as the
value of the last sample about 3 min after test termination. Heart rate and blood lactate
values were transferred to an automated software (winlactat, Mesics, Münster, Germany)
to determine LTs. In accordance with the study of Weckbach et al. [7], we selected three
different methods for the detection of the LT: (a) fixed 4-mmol/L blood lactate concentration
according to Mader et al. [13], (b) lactate concentration of 1.5 mmol/L above the minimal
lactate equivalent according to Dickhut et al. [14], and (c) maximal perpendicular distance
from the blood lactate concentration curve to the line drawn from start- to endpoint (also
known as Dmax method) according to Cheng et al. [15]. Outcome parameters of the maximal
incremental tests were Pmax, HRmax as well as power output and heart rate at the three LT
(PMader, PDickhuth, PCheng and HRMader, HRDickhuth, HRCheng).
2.2.2. Maximal 30-min Time Trials
Time trials were conducted as described in detail in previous studies [8,16]. In brief:
testing began with warm-up periods of 5 min at 100/150 W (females/males) followed by
5 min at 150/200 W (females/males). Then, the cycle ergometer was shifted to a fixed
pedal force so that pedalling at 100 rpm produced about 70% of Pmax (determined by the
maximal incremental test in normoxia). Participants were encouraged to choose a maximal
pedalling rate that could be maintained for the respective test duration. Heart rate was
measured continuously (chest belt, Polar, Austria), and capillary blood samples were taken
from the hyperaemised earlobe after 7 and 27 min to determine blood lactate concentrations
(BLA7 and BLA27) (Super GL Ambulance, Dr. Müller Gerätebau, Germany). Outcome
parameters were mean power output (Pmean) and mean heart rate (HRmean), which were
automatically calculated by the software of the ergometer, BLA7 and BLA27.
2.3. Statistics
Statistical analyses were performed using SPSS 24.0 (IBM, Vienna, Austria). Data were
checked for normal distribution using the Shapiro–Wilk test. Since data were normally
distributed (except for BLA7 and BLA27), paired t-tests (Wilcoxon rank tests for BLA7 and
BLA27) were used to compare outcome parameters in normoxic versus hypoxic conditions.
In the next step, focusing on power output parameters separated for normoxic and hypoxic
conditions, Pearson correlation analyses between power output at the LT and Pmean were
performed, and potential differences were tested by paired t-tests. In addition, the hypoxia-
related reductions in power output at the LTs and in Pmean were compared with paired
t-tests to test whether lactate thresholds are affected differently by acute hypoxia compared
to high-intensity endurance performance. The level of significance was set at p < 0.05. Data
are presented as means ± SD.
3. Results
The results of the maximal incremental tests revealed a significant decrease in Pmax
by approximately 12% from normoxia to hypoxia (249 ± 25 versus 221 ± 36 W, p = 0.005).
HRmax (189 ± 10 versus 186 ± 8 bpm, p = 0.217) and BLAmax (11.9 ± 2.5 versus 13.1 ± 0.9,
p = 0.166) did not significantly differ from normoxia to hypoxia.
Int. J. Environ. Res. Public Health 2021, 18, 7573
4 of 8
Power output and heart rate at the different LT and parameters of the maximal 30-min
time trials are shown in Table 1. Power output decreased for all threshold concepts by about
12 to 19%, whereas heart rate values showed small (0 to 5%) and not significant changes
from normoxia to hypoxia. Pmean decreased significantly from normoxia to hypoxia by
approximately 8%, and HRmean was slightly lower in hypoxia compared to normoxia,
although not statistically significant. Blood lactate concentrations during the time trials
did not significantly differ between normoxia versus hypoxia (BLA7: 6.1 ± 2.2 versus
6.3 ± 2.2 mmol/L, p = 0.873; BLA27: 8.9 ± 2.5 versus 9.4 ± 2.0 mmol/L, p = 0.385).
Table 1. Power output and heart rate at different lactate thresholds and during maximal 30-min time
trials in normoxia and hypoxia. Values are means ± SD. p-Values refer to a comparison of normoxia
and hypoxia.
Normoxia
Hypoxia
Difference (%)
p-Value
Lactate thresholds based on stepwise maximal cycle ergometries
PMader (W)
194 ± 26
158 ± 33
−18.9 ± 9.6
0.004
HRMader (bpm)
167 ± 8
160 ± 12
−4.5 ± 6.8
0.154
PDickhuth (W)
179 ± 24
146 ± 22
−18.4 ± 7.3
0.001
HRDickhuth (bpm)
161 ± 8
153 ± 16
−5.1 ± 7.0
0.140
PCheng (W)
194 ± 30
172 ± 28
−11.5 ± 6.0
0.005
HRCheng (bpm)
167 ± 9
167 ± 11
+0.3 ± 8.0
0.977
Maximal 30-min time trials
Pmean (W)
195 ± 34
179 ± 32
−8.3 ± 1.6
<0.001
HRmean (bpm)
175 ± 12
170 ± 8
−2.7 ± 4.4
0.203
P, power output; HR, hear rate. Lactate thresholds were determined according to the methods of Mader et al.
(Mader), Dickhuth et al. (Dickhuth) and Cheng et al. (Cheng) [13–15]. Pmean, mean power output; HRmean, mean
heart rate.
Correlation analyses for normoxia (Figure 2a) data revealed strong and significant
correlations for power output at the three LT and Pmean (PMader: r = 0.935, p = 0.006;
PDickhuth: r = 0.931, p = 0.007; PCheng: r = 0.977, p = 0.001). Furthermore, PMader and PCheng
did not significantly differ from Pmean (p = 0.867 and p = 0.784 respectively), whereas
PDickhuth was significantly lower compared to Pmean (p = 0.045). With respect to hypoxic
conditions (Figure 2b), we also found significant correlations, but for PDickhuth and PCheng,
we found slightly weaker correlations (PMader: r = 0.941, p = 0.005; PDickhuth: r = 0.869,
p = 0.024; PCheng: r = 0.887, p = 0.019). PMader and PDickhuth were significantly lower
compared to Pmean (p = 0.007 and p = 0.005), whereas there was no significant difference for
PCheng (p = 0.284).
Comparing hypoxia-related impairments in power output at the three LT and in Pmean,
significant differences were found for PMader and PDickhuth but not for PCheng (Figure 3).
Int. J. Environ. Res. Public Health 2021, 18, 7573
5 of 8
Int. J. Environ. Res. Public Health 2021, 18, x FOR PEER REVIEW
5 of 8
Figure 2. Association between mean power output during the time trial and power output at three
different lactate thresholds (blue = PMader, red = PDickhuth, green PCheng) [11–13]. Data are presented for
normoxic (upper panel) und hypoxic conditions (lower panel). (a): normoxic conditions; (b):
hypoxic conditions
Comparing hypoxia-related impairments in power output at the three LT and in
Pmean, significant differences were found for PMader and PDickhuth but not for PCheng (Figure 3).
Figure 2. Association between mean power output during the time trial and power output at
three different lactate thresholds (blue = PMader, red = PDickhuth, green PCheng) [11–13]. Data are
presented for normoxic (upper panel) und hypoxic conditions (lower panel). (a): normoxic conditions;
(b): hypoxic conditions.
Int. J. Environ. Res. Public Health 2021, 18, 7573
6 of 8
Int. J. Environ. Res. Public Health 2021, 18, x FOR PEER REVIEW
6 of 8
Figure 3. Hypoxia-related changes in power output at the three lactate thresholds (PMader, PDickhuth,
PCheng) and in mean power output during the maximal 30-min time trial (Pmean) [13–15]. P-values
refer to a comparison to Pmean. Values are means ± SD.
4. Discussion
The presented data show that acute hypoxia impaired power output at different LT
and Pmean as expected. In accordance with our hypothesis, the results also demonstrate
that the reduction in Pmean was clearly lower (approximately 8%) compared to lactate
thresholds estimations (12% to 19%). Correlations between power output at the LT and
Pmean were found in normoxic as well as in acute hypoxic conditions. However, hypoxia-
related decreases in power output at two LT (PMader and PDickhuth) were significantly greater
compared to the hypoxia-related decrease in Pmean resulting in significant differences to
Pmean in hypoxia.
Although LT are often validated for specific exercise protocols, exercise modes and
populations, the observed strong correlations in normoxic conditions between power
output at the LT (PMader, PDickhuth, PCheng) and Pmean were well documented [17] and in
accordance with our findings. The method of Dickhuth et al. [14] was designed to detect
the first rise in blood lactate concentration and can be categorised as an aerobic lactate
threshold [17]. This observation was also reported in the review article of Faude et al. [17],
and, therefore, the significant underestimation of Pmean in our study was not surprising.
However, this underestimation of Pmean does not limit the application of this LT in exercise
prescription by model-specific intensity zones.
Regarding the hypoxia-induced changes in power output, the three LT showed more
pronounced reductions compared to Pmean. From a practical point of view, the application
of LT-based intensity zones, although fitting in normoxic conditions, can result in an
underestimation of endurance capacity and therefore sub-optimal or even ineffective
training loads when determined and applied in acute hypoxic conditions. Since the
method according to Cheng et al. [15] only slightly underestimated Pmean and,
furthermore, showed a strong correlation to Pmean under hypoxic conditions, it seems that
exercise intensity prescriptions based on this method may be more robust against
hypoxic-related underestimations. Based on the observation that resting lactate
concentrations are not influenced by moderate hypoxia (e.g., 3000 m) but are markedly
pronounced during exercise at the same absolute intensity level [18,19], the following
explanations may be reasonable: LT models, defined by a fixed blood lactate concentration
(i.e., 4 mmol/l) or adding a fixed value (i.e., 1.5 mmol/l) to an individual minimum may
be directly affected by changes in the absolute lactate values (e.g., under acute hypoxia).
Figure 3. Hypoxia-related changes in power output at the three lactate thresholds (PMader, PDickhuth,
PCheng) and in mean power output during the maximal 30-min time trial (Pmean) [13–15]. p-Values
refer to a comparison to Pmean. Values are means ± SD.
4. Discussion
The presented data show that acute hypoxia impaired power output at different LT
and Pmean as expected. In accordance with our hypothesis, the results also demonstrate that
the reduction in Pmean was clearly lower (approximately 8%) compared to lactate thresholds
estimations (12% to 19%). Correlations between power output at the LT and Pmean were
found in normoxic as well as in acute hypoxic conditions. However, hypoxia-related
decreases in power output at two LT (PMader and PDickhuth) were significantly greater
compared to the hypoxia-related decrease in Pmean resulting in significant differences to
Pmean in hypoxia.
Although LT are often validated for specific exercise protocols, exercise modes and
populations, the observed strong correlations in normoxic conditions between power
output at the LT (PMader, PDickhuth, PCheng) and Pmean were well documented [17] and in
accordance with our findings. The method of Dickhuth et al. [14] was designed to detect
the first rise in blood lactate concentration and can be categorised as an aerobic lactate
threshold [17]. This observation was also reported in the review article of Faude et al. [17],
and, therefore, the significant underestimation of Pmean in our study was not surprising.
However, this underestimation of Pmean does not limit the application of this LT in exercise
prescription by model-specific intensity zones.
Regarding the hypoxia-induced changes in power output, the three LT showed more
pronounced reductions compared to Pmean. From a practical point of view, the application
of LT-based intensity zones, although fitting in normoxic conditions, can result in an
underestimation of endurance capacity and therefore sub-optimal or even ineffective
training loads when determined and applied in acute hypoxic conditions. Since the
method according to Cheng et al. [15] only slightly underestimated Pmean and, furthermore,
showed a strong correlation to Pmean under hypoxic conditions, it seems that exercise
intensity prescriptions based on this method may be more robust against hypoxic-related
underestimations. Based on the observation that resting lactate concentrations are not
influenced by moderate hypoxia (e.g., 3000 m) but are markedly pronounced during
exercise at the same absolute intensity level [18,19], the following explanations may be
reasonable: LT models, defined by a fixed blood lactate concentration (i.e., 4 mmol/L) or
adding a fixed value (i.e., 1.5 mmol/L) to an individual minimum may be directly affected
by changes in the absolute lactate values (e.g., under acute hypoxia). In contrast, the LT
model by Cheng et al. [15] also considers the shape of the blood lactate curve and thus
Int. J. Environ. Res. Public Health 2021, 18, 7573
7 of 8
the individual lactate kinetics, which seems to be a crucial point in evaluating exercise
capacity [20].
To the best of our best knowledge, this is the first study comparing hypoxia-related
changes in lactate thresholds with changes in endurance performance (i.e., 30-min time-trial
performance). The inclusion of a first time trial for familiarisation improved the validity of
the hypoxia-related changes in the second and third time trial because it was shown that
time-trial performance improves from a first to a second test [16]. The small sample size
and the resulting susceptibility for individual outliers represent the main limitation of the
present study. For example, the correlation analysis in hypoxia revealed r = 0.887 for PCheng
versus Pmean (p = 0.019). However, the exclusion of one person (a statistical borderline
outlier) would result in r = 0.959, (p = 0.010, n = 5). In addition, this experiment was
conducted in a specific group of healthy subjects with an above-average fitness level and
results cannot be directly transferred either to elite athletes or to specific patient groups.
5. Conclusions
According to the hypothesis, power output at the lactate thresholds were more im-
paired by an acute exposure to hypoxia compared to high-intensity endurance performance
reaching statistical significance for the methods of Mader et al. and Dickhuth et al. In
conclusion, the application of LT for exercise intensity prescription in hypoxia, even when
determined under such conditions, may be prone to errors. The results of this pilot study
should provide a basis for future larger-scale investigations dealing with this topic in
different target groups.
Author Contributions: Conceptualisation, M.F. and V.M.; methodology, M.F.; formal analysis, K.G.
and M.F.; writing—original draft preparation, M.F. and L.R.; writing—review and editing, M.F., V.M.
and H.G. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: The study was conducted according to the guidelines of the
Declaration of Helsinki and approved by the Board for Ethical Questions in Science of the University
of Innsbruck, Austria (Certificate of good standing, 12/2021).
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: The data are not publicly available due to ethical considerations on
preserving the anonymity of study participants.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Millet, G.P.; Roels, B.; Schmitt, L.; Woorons, X.; Richalet, J.P. Combining Hypoxic Methods for Peak Performance. Sports Med.
2010, 40, 1–25. [CrossRef] [PubMed]
2.
Ramos-Campo, D.J.; Girard, O.; Pérez, A.; Rubio-Arias, J.Á. Additive stress of normobaric hypoxic conditioning to improve
body mass loss and cardiometabolic markers in individuals with overweight or obesity: A systematic review and meta-analysis.
Physiol. Behav. 2019, 207, 28–40. [CrossRef] [PubMed]
3.
Park, H.-Y.; Kim, J.; Park, M.-Y.; Chung, N.; Hwang, H.; Nam, S.-S.; Lim, K. Exposure and Exercise Training in Hypoxic Conditions
as a New Obesity Therapeutic Modality: A Mini Review. J. Obes. Metab. Syndr. 2018, 27, 93–101. [CrossRef] [PubMed]
4.
Gatterer, H.; Haacke, S.; Burtscher, M.; Faulhaber, M.; Melmer, A.; Ebenbichler, C.; Strohl, K.P.; Högel, J.; Netzer, N.C. Normobaric
Intermittent Hypoxia over 8 Months Does Not Reduce Body Weight and Metabolic Risk Factors–a Randomized, Single Blind,
Placebo-Controlled Study in Normobaric Hypoxia and Normobaric Sham Hypoxia. Obes. Facts 2015, 8, 200–209. [CrossRef]
[PubMed]
5.
Siebenmann, C.; Dempsey, J.A. Hypoxic Training Is Not Beneficial in Elite Athletes. Med. Sci. Sports Exerc. 2020, 52, 519–522.
[CrossRef] [PubMed]
6.
Ozcelik, O.; Kelestimur, H. Effects of acute hypoxia on the estimation of lactate threshold from ventilatory gas exchange indices
during an incremental exercise test. Physiol. Res. 2004, 53, 653–659. [PubMed]
7.
Weckbach, L.T.; Kassem, S.; Maier, F.; Hamm, W.; Schüttler, D.; Kellnar, A.; Lackermair, K.; Brunner, S. Impact of Acute Altitude
Exposure on Lactate Threshold. High Alt. Med. Biol. 2019, 20, 322–323. [CrossRef] [PubMed]
Int. J. Environ. Res. Public Health 2021, 18, 7573
8 of 8
8.
Burtscher, M.; Faulhaber, M.; Flatz, M.; Likar, R.; Nachbauer, W. Effects of Short-Term Acclimatization to Altitude (3200 m) on
Aerobic and Anaerobic Exercise Performance. Int. J. Sports Med. 2005, 27, 629–635. [CrossRef] [PubMed]
9.
Gatterer, H.; Philippe, M.; Menz, V.; Mosbach, F.; Faulhaber, M.; Burtscher, M. Shuttle-Run Sprint Training in Hypoxia for Youth
Elite Soccer Players: A Pilot Study. J. Sports Sci. Med. 2014, 13, 731–735. [PubMed]
10.
Faulhaber, M.; Pocecco, E.; Gatterer, H.; Niedermeier, M.; Huth, M.; Dünnwald, T.; Menz, V.; Bernardi, L.; Burtscher, M. Seven
Passive 1-h Hypoxia Exposures Do Not Prevent AMS in Susceptible Individuals. Med. Sci. Sports Exerc. 2016, 48, 2563–2570.
[CrossRef] [PubMed]
11.
Morawetz, D.; Dünnwald, T.; Faulhaber, M.; Gatterer, H.; Schobersberger, W. Impact of Hyperoxic Preconditioning in Normobaric
Hypoxia (3500 m) on Balance Ability in Highly Skilled Skiers: A Randomized, Crossover Study. Int. J. Sports Physiol. Perform.
2019, 14, 934–940. [CrossRef] [PubMed]
12.
Stepto, N.K.; Hawley, J.A.; Dennis, S.C.; Hopkins, W.G. Effects of different interval-training programs on cycling time-trial
performance. Med. Sci. Sports Exerc. 1999, 31, 736–741. [CrossRef] [PubMed]
13.
Mader, A.; Liesen, H.; Heck, H.; Philippi, H.; Rost, R.; Schuerch, P.; Hollmann, W. Zur Beurteilung der sportartspezifischen
Ausdauerleistungsfähigkeit im Labor. Sportarzt. Sportmed. 1976, 27, 80–88.
14.
Dickhuth, H.H.; Yin, L.; Niess, A.; Röcker, K.; Mayer, F.; Heitkamp, H.C.; Horstmann, T. Ventilatory, lactate-derived and
catecholamine thresholds during incremental treadmill running: Relationship and reproducibility. Int. J. Sports Med. 1999, 20,
122–127. [CrossRef] [PubMed]
15.
Cheng, B.; Kuipers, H.; Snyder, A.; Keizer, H.; Jeukendrup, A.; Hesselink, M. A New Approach for the Determination of
Ventilatory and Lactate Thresholds. Int. J. Sports Med. 1992, 13, 518–522. [CrossRef] [PubMed]
16.
Faulhaber, M.; Gatterer, H.; Haider, T.; Patterson, C.; Burtscher, M. Intermittent hypoxia does not affect endurance performance at
moderate altitude in well-trained athletes. J. Sports Sci. 2010, 28, 513–519. [CrossRef] [PubMed]
17.
Faude, O.; Kindermann, W.; Meyer, T. Lactate Threshold Concepts. Sports Med. 2009, 39, 469–490. [CrossRef] [PubMed]
18.
Friedmann, B.; Bauer, T.; Menold, E.; Bärtsch, P. Exercise with the Intensity of the Individual Anaerobic Threshold in Acute
Hypoxia. Med. Sci. Sports Exerc. 2004, 36, 1737–1742. [CrossRef] [PubMed]
19.
McLellan, T.; Jacobs, I.; Lewis, W. Acute altitude exposure and altered acid-base states. Graefe’s Arch. Clin. Exp. Ophthalmol. 1988,
57, 445–451. [CrossRef]
20.
Stegmann, H.; Kindermann, W.; Schnabel, A. Lactate Kinetics and Individual Anaerobic Threshold. Int. J. Sports Med. 1981, 2,
160–165. [CrossRef] [PubMed]
| Effects of Acute Hypoxia on Lactate Thresholds and High-Intensity Endurance Performance-A Pilot Study. | 07-16-2021 | Faulhaber, Martin,Gröbner, Katharina,Rausch, Linda,Gatterer, Hannes,Menz, Verena | eng |
PMC5880957 | ORIGINAL RESEARCH
Validation of a novel wearable, wireless technology to
estimate oxygen levels and lactate threshold power in the
exercising muscle
Parisa Farzam, Zack Starkweather & Maria A. Franceschini
Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts
Keywords
Athletic exercise training, blood lactate,
muscle oxygen saturation, near-infrared
spectroscopy (NIRS).
Correspondence
Parisa Farzam, Athinoula A. Martinos Center
for Biomedical Imaging, Massachusetts
General Hospital, Harvard Medical School
149, 13th Street, Boston, MA 02129.
Tel: +1- 617-726-9338
Fax: +1- 617-726-7422
E-mail: [email protected]
Funding Information
This research was sponsored by Dynometrics,
Inc.
Received: 20 February 2018; Accepted: 23
February 2018
doi: 10.14814/phy2.13664
Physiol Rep, 6 (7), 2018, e13664,
https://doi.org/10.14814/phy2.13664
Abstract
There is a growing interest in monitoring muscle oxygen saturation (SmO2),
which is a localized measure of muscle oxidative metabolism and can be
acquired continuously and noninvasively using near-infrared spectroscopy
(NIRS) methods. Most NIRS systems are cumbersome, expensive, fiber cou-
pled devices, with use limited to lab settings. A novel, low cost, wireless, wear-
able has been developed for use in athletic training. In this study, we evaluate
the advantages and limitations of this new simple continuous-wave (CW)
NIRS device with respect to a benchtop, frequency-domain near-infrared spec-
troscopy (FDNIRS) system. Oxygen saturation and hemoglobin/myoglobin
concentration in the exercising muscles of 17 athletic individuals were mea-
sured simultaneously with the two systems, while subjects performed an incre-
mental test on a stationary cycle ergometer. In addition, blood lactate
concentration was measured at the end of each increment with a lactate ana-
lyzer. During exercise, the correlation coefficients of the SmO2 and hemoglo-
bin/myoglobin concentrations between the two systems were over 0.70. We
also found both systems were insensitive to the presence of thin layers of vary-
ing absorption, mimicking different skin colors. Neither system was able to
predict the athletes’ lactate threshold power accurately by simply using SmO2
thresholds. Instead, the proprietary software of the wearable device was able
to predict the athletes’ lactate threshold power within half of one power incre-
ment of the cycling test. These results indicate this novel wearable device may
provide a physiological indicator of athlete’s exertion.
Introduction
Quantifying how muscles respond to physical exercise is
of great interest to athletes for improving performance
and mitigating the risk of injury. Traditionally, athletes
rely on measurements such as heart rate, blood lactate
concentration, or maximum oxygen uptake (VO2max).
These parameters are used to determine the intensity
levels at which athletes should be exerting themselves to
maximize athletic performance (Seiler and Kjerland 2006;
Bentley et al. 2007; Esteve-Lanao et al. 2007; Goodwin
et al. 2007). Although heart rate, blood lactate concentra-
tion, and VO2max can help guide an athlete’s training
regimen, these measurements are indicative of systemic
changes occurring in the body, with no specific informa-
tion about the working muscles.
Interest in examining muscle oxygen saturation (SmO2)
has been growing due to its ability to provide a localized
measurement continuously and noninvasively using near-
infrared spectroscopy (NIRS) (Chance et al. 1992; Belar-
dinelli et al. 1995; Hamaoka et al. 1996, 2011; Bhambhani
et al. 1997; Grassi et al. 1999, 2003; Wang et al. 2006;
Soller et al. 2008; Bailey et al. 2009; Ihsan et al. 2013;
Racinais et al. 2014; Boone et al. 2016; van der Zwaard
et al. 2016; Baker et al. 2017; Hammer et al. 2018; Perrey
and Ferrari 2018). NIRS techniques work by delivering
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
This is an open access article under the terms of the Creative Commons Attribution License,
which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
2018 | Vol. 6 | Iss. 7 | e13664
Page 1
Physiological Reports ISSN 2051-817X
light (in the 650–900 nm wavelength range) into the tis-
sue and measuring the diffused light to estimate the
absorption and scattering properties of the measured tis-
sue volume (Yodh and Chance 1995). The concentrations
of oxyhemoglobin (HbO) and deoxyhemoglobin (HbR)
in the tissue can be estimated from the measured absorp-
tion spectrum. Muscle oxygen saturation, also referred as
muscle oxygenation (SmO2), is then calculated by taking
the ratio of HbO to total hemoglobin concentration
(HbT). Previous studies have investigated the HbO, HbR,
HbT, or SmO2 trends during exercise to determine if
these NIRS parameters provide useful information to
guide athletic training (Belardinelli et al. 1995; Bhamb-
hani et al. 1997; Grassi et al. 1999, 2003; Wang et al.
2006; Bailey et al. 2009; Hamaoka et al. 2011; Racinais
et al. 2014). Correlations have been observed between the
threshold power found from VO2max or lactate data and
NIRS methods (Belardinelli et al. 1995; Grassi et al. 1999;
Wang et al. 2006; van der Zwaard et al. 2016). These
NIRS studies developed techniques to find an athlete’s
threshold power by analyzing the trends in HbO, HbR,
HbT, or SmO2 after an incremental power test (Belar-
dinelli et al. 1995; Grassi et al. 1999; Wang et al. 2006;
van der Zwaard et al. 2016). Other work has been per-
formed to understand muscle adaptations throughout an
athlete’s training by examining the HbO, HbR, HbT, or
SmO2 kinetics during specific exercise protocols (Bailey
et al. 2009; Ihsan et al. 2013). These studies support the
incorporation of NIRS technology into athletic training
but are limited to the laboratory setting due to the
required optical fibers, probes, and larger instruments to
obtain the measurements. The noninvasive monitoring of
muscle oxygenation could significantly benefit from small,
wearable, wireless, and accurate devices that can deliver
real-time feedback to athletes.
Currently, there are a small number of wearable, fiber-
less, NIRS devices used in the athletic market, such as the
Portamon (Artinis Medical System, Einsteinweg, The
Netherlands),
Moxy
Monitor
(Fortiori
Design,
LLC,
Hutchinson, MN, USA), and BSX Insight (BSX Athletics,
Austin, TX, USA). A review from Perrey and Ferrari
(2018) goes into detail about the different studies that
have examined the use of these NIRS devices in athletic
training and SmO2 monitoring. The Portamon and Moxy
devices can be manually strapped on to any muscle group
and have been used during a variety of activities, includ-
ing cycling, running, and strength training (Perrey and
Ferrari 2018). The BSX Insight is designed to be worn on
the gastrocnemius muscle within a custom-made com-
pression sleeve and differentiates itself by providing ath-
letes with their lactate threshold when following a given
protocol (Borges and Driller 2016; Driller et al. 2016).
These studies on wearable NIRS devices in the athletic
community show value in obtaining optical measure-
ments
during
sports,
however
the
validity
of
measurements from wearables would be enhanced by a
direct comparison with a bench-top, fiber-based, FDNIRS
system.
In this study, we compared the SmO2 of the quadriceps
muscle group recorded by a wearable, low-cost, continu-
ous-wave (CWNIRS) consumer device (Humon Beta,
Dynometrics, Inc.) against a benchtop fiber-based fre-
quency-domain near-infrared spectroscopy (FDNIRS) sys-
tem (MetaOx, ISS, Champaign, IL). The FDNIRS system
is considered the most robust and reliable reference com-
mercially available for comparison of oximeters (Kleiser
et al. 2016). Our goal was to examine the accuracy of this
wearable device and understand the limitations that arise
when using CWNIRS for muscle oxygenation measure-
ments. We also investigated the real-time feedback from
the Humon Beta device and reported the differences
between the optically derived threshold and blood lactate
threshold during an incremental cycling test.
Methods
Study population and measurement
protocol
Fifteen male and three female athletic subjects performed
an incremental step test on the cycle ergometer. For logis-
tical reasons, the test was carried out with the MetaOx
probe on the right leg rectus femoris and the Humon
Beta on the left leg rectus femoris, as illustrated in Fig-
ure 1. The rectus femoris was chosen as the muscle to
monitor in order to minimize fiber movement since this
was the area of the leg where the fibers could remain
most stable. Each session began with baseline measure-
ments for approximately two minutes, where subjects
were instructed to stay as still as possible. Following a
well-established incremental test protocol (Madden et al.
2013), the subjects began cycling at 30 W for 4 min, and
the power was increased in 30 W increments every 4 min
until voluntary exhaustion. Voluntary exhaustion was
determined when the subject requested to stop or when
he/she could no longer maintain the cycling cadence. In
the last minute of each power interval, the blood lactate
concentration was measured using a handheld lactate ana-
lyzer (Lactate Plus meter, Nova Biomedical, Waltham,
MA) by averaging three repeated blood samples. The sub-
jects were asked to keep a consistent cadence throughout
the cycling protocol, which was typically between 80 and
100 rotations per minute (RPM). For each subject, the
lactate threshold power was determined as the cycling
power at which the athlete’s blood lactate concentration
reaches 4 mmol/L (Faude et al. 2009; Madden et al.
2018 | Vol. 6 | Iss. 7 | e13664
Page 2
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
A New Wearable NIRS to Estimate Lactate Threshold
P. Farzam et al.
2013). One male subject was excluded from all data anal-
yses because of a Humon Beta malfunction. On this sub-
ject, sweat entered into the sensor and corrupted the
data, however, we verified that sweat did not enter on
any other Humon Beta device and did not affect any
other measurements. Thus, from now on, we just report
the values of the 17 subjects. In seven of the subjects, two
Humon Beta sensors were worn on the left rectus femoris
muscle to examine the heterogeneity of the measurements
within the same muscle. The devices were placed 2 cm
apart from one another, where one was proximal and the
other was distal along the muscle and the variability due
to location placement was assessed. We verified the sepa-
ration of the two devices was sufficient to prevent any
light leakage between them.
Before beginning the cycling on an upright stationary
cycle ergometer (Model E3, Kettler USA), basic informa-
tion was collected for each subject. This information
included age, height, weight, and gender. Body Mass
Index (BMI) was calculated as (kg/m2). In addition, the
subcutaneous adipose tissue thickness (SCATT) on the
rectus femoris of the right quadriceps was measured for
each subject using a skinfold caliper before the start of
the cycling test. Information about the subjects can be
found in Table 1. We also tested the vitals (heart rate,
blood pressure, body temperature, arterial oxygen satura-
tion) of all subjects to ensure that the subjects were
healthy prior to the onset of exercise.
The study protocol was reviewed and approved by the
Institutional Review Board (IRB) for Partners Healthcare,
the Partners Human Research Committee (PHRC). The
study method was designed and carried out in accordance
with PHRC requirements and the regulations that govern
human subjects research. All subjects interested in partici-
pating in the study went through a prescreening process
to ensure they were eligible. These subjects all exercised
more than 3 times per week, each session over 30 min
and their exercise routine typically included biking, run-
ning, swimming, or rowing. We refer to this population
as athletes throughout the text. They were all comfortable
exercising at a high intensity for a long period of time.
All eligible participants read and signed the approved
informed consent form before starting the measurement
session.
Instrument setup
The Humon Beta device (Fig. 1) uses two light sources in
the NIR window and three photodetectors to measure the
intensity of the light that has propagated through the tis-
sue. The sources and detectors are found behind individ-
ual polycarbonate windows, which come in contact with
the skin of the user. The photodetectors are located at
distances of 1.2, 1.8, and 2.4 cm from the light sources.
The acquisition rate is set to 4 Hz. The Humon Beta is
6.0 9 5.7 9 1.4 cm in size and has a slight curvature in
the plastic case to allow for easy contact with the skin on
Figure 1. The schematic of the measurement on an upright
stationary ergometer. The top drawing displays the location of
Humon wearable on the left leg and MetaOx probe on the right
leg. The distribution of the sources and detectors are presented for
both the MetaOx probe (source–detector separations: 1.5, 2.0, 2.5,
and 3.0 cm) and Humon Beta wearable (source–detector
separations: 1.2, 1.8, and 2.4 cm) in the bottom images.
Table 1. Subjects demographic.
All
Males
Females
Number
17
14
3
Right-footed
15
12
3
Age (years)
31 6
30 6
35 5
Weight (kg)
71 14
75 11
51 3
Height (cm)
176 11
180 7
156 3
BMI (kg/m2)
23 3
23 3
21 2
SCATT (mm)
5 2
5 2
7 1
Details about the subjects’ footedness, age, weight, height, BMI,
and subcutaneous adipose tissue thickness (SCATT) are shown,
including a gender breakdown for each category.
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
2018 | Vol. 6 | Iss. 7 | e13664
Page 3
P. Farzam et al.
A New Wearable NIRS to Estimate Lactate Threshold
the quadriceps. The wearable is attached to the athlete’s
quadriceps, using a strap that hooks through the device
and can be secured around the thigh with a hook-and-
loop fastener. The Humon Beta communicates with a
smartphone via Bluetooth, and a custom app displays the
workout progress in real time.
MetaOx is a hybrid device that consists of FDNIRS to
measure hemoglobin concentration and tissue oxygena-
tion and diffuse correlation spectroscopy (DCS) to mea-
sure blood flow (Boas et al. 2016). The details of the
system and data analysis are described by Carp et al.
(2017). For this work, we only considered the FDNIRS
data even though DCS data were also acquired. In sum-
mary, the FDNIRS components include 8 lasers in the
visible and NIR spectral region (672, 726, 759, 813, 690,
706, 784, and 830 nm) modulated at 110 MHz and 4
photomultiplier tubes detectors (PMT) modulated at
110.005 MHz to achieve heterodyne detection at 5 kHz.
The lasers are rapidly multiplexed in sequence and allow
the fast measurement (10 Hz) over all wavelengths. The
light is delivered to the tissue through a fiber optics bun-
dle coupled to the probe. The diffused photons are col-
lected by the probe’s fiber optics bundles located at
distances of 1.5, 2.0, 2.5, and 3.0 cm from the source (q),
(see Fig. 1) and delivered to the four PMTs.
Data analysis
The Humon Beta and MetaOx data were coregistered and
acquired continuously for the whole duration of the pro-
tocol. For MetaOx data analysis, the detected light ampli-
tude (AC) and phase shift at four separations and eight
wavelengths are calibrated on a silicone tissue mimicking
phantom with known optical properties to account for
the different gains of the four detectors (Farzam et al.
2013). After calibration, the frequency-domain solution of
the photon diffusion equation in the semi-infinite geome-
try is used to fit for the wavelength-dependent optical
properties of the tissue (Durduran et al. 2010). At each
wavelength, a linear fit is performed on (ln(AC(k)q2))
and phase over source-detector distances. Absorption (la)
and reduced scattering (ls0) coefficients at each wave-
length are calculated from the slopes of the linear fits
(Fantini et al. 1995). The measured absorption coefficient
is sum of the absorption of tissue chromophores (la (k)
= ∑ ei(k)ci), where the wavelength-dependent extinction
coefficient, ei(k), of the ith chromophore is obtained from
the literature (Prahl), and ci is the concentration of the
ith chromophore. The primary muscle chromophores in
the near-infrared are water, oxy-, deoxy-hemoglobin, and
myoglobin.
The
muscle
tissue
water
percentage
is
assumed to be 75% (Franceschini et al. 1997). Since myo-
globin cannot be distinguished from hemoglobin due to
the spectral overlap (Quaresima et al. 2004), its contribu-
tion
is
combined
into
the
calculated
oxy-
and
deoxy-hemoglobin
concentrations.
When we refer
to
hemoglobin throughout the text, we acknowledge that
this is a combination of hemoglobin and myoglobin. The
total hemoglobin concentration and oxygen saturation are
calculated as HbT = HbO + HbR and SmO2 = HbO/
HbT, respectively.
The Humon Beta is a CW device and only measures
light intensity at three separations and two wavelengths.
Hence, to estimate hemoglobin concentration and oxy-
genation, it needs to assume a fixed scattering coefficient.
The algorithms used to recover hemodynamic parameters
were not disclosed and are proprietary to Dynometrics
Inc. The HbO, HbR, HbT, and SmO2 results obtained
with the two devices were compared to determine the
drawbacks of using a limited set of wavelengths and fixed
scattering coefficients for a small, inexpensive wireless
NIRS device which cannot include as many features as
the benchtop frequency-domain system.
In addition, to further estimate the error introduced by
the simplifications in the wearable system, we compared
full MetaOx results with estimates obtained, using a sub-
set of the MetaOx intensity data, which best matches the
Humon Beta separations and wavelengths. In this way, we
can isolate the errors due to the reduced dataset from
contamination arising from other parameters such as dif-
ferences in sensor location and differences between left
and right leg muscles.
To test the effect of superficial thin layers, such as dif-
ferent skin tones, on the data collected by the Humon
Beta and MetaOx systems, neutral density (ND) filters
were placed over a silicone tissue mimicking phantom
with known optical properties and data were collected
with the two systems. We evaluated how the presence of
these superficial thin layers affect the recovered “effective
SmO2”, both with the Humon Beta and the MetaOx. We
call the calculated parameter “effective SmO2”, since the
phantom is made of silicon and does not contain blood.
For the MetaOx we also evaluated how the ND filter
affected the measured intensity and the slope of both
phase and intensity versus distance.
Moreover, in the 17 subjects we evaluate the effect of
subcutaneous adipose tissue on the estimated HbO, HbR,
HbT, and SmO2 by analyzing the MetaOx cycling data
using only the shorter (1.5 and 2.0 cm) or the larger (2.5
and 3.0 cm) separations. We estimated the difference in
the recovered HbO, HbR, HbT, and SmO2 at the two
separations throughout the whole cycling test, and the
correlation between SmO2 drop and measured SCATT.
Finally, we examined the relationship between the mea-
sured hemodynamic parameters and the blood lactate
measures and tested the accuracy of the Humon Beta for
2018 | Vol. 6 | Iss. 7 | e13664
Page 4
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
A New Wearable NIRS to Estimate Lactate Threshold
P. Farzam et al.
predicting the lactate threshold power. The Humon Beta
employs a proprietary algorithm to determine the lactate
threshold in real time. The data are processed and dis-
played on a smartphone application, where in addition to
SmO2 values, a corresponding exercise zone is shown.
Four zones are used to classify the subjects exercise state:
(1) the green zone represents a steady state, (2) the
orange zone indicates the athlete is approaching their
limit, (3) the red zone shows the athlete has hit/exceeded
their limit, and (4) the blue zone means the athlete is in
a recovery phase. For this analysis, we excluded the three
female subjects since the power increments were too large,
which resulted in the lactate threshold power and the end
of the exercise to be too close in time.
For HbO, HbR, HbT, and SmO2 comparisons between
the MetaOx and Humon Beta, we report the Pearson cor-
relation coefficient and the relative root mean square
error (RMSE). Nonparametric statistical tests were per-
formed when comparing features in the SmO2 curves
between the MetaOx and Humon Beta. All the data pro-
cessing and statistical analyses in this paper are performed
using MATLAB (MathWorks, USA), version R2017a, and
Statistics and Machine Learning Toolbox, version 11.1.
Results
Optical and physiological properties of the
leg muscle
Figure 2 shows the average of optical properties, that is,
the absorption coefficient (la) and reduced scattering
coefficient (ls0), measured by MetaOx across all the sub-
jects over 2 min of baseline before starting to pedal. The
error bars indicate the mean over all subjects and 95%
confidence interval of the mean. The dashed red line is
the fitted spectrum and the gray shaded area indicates the
95% confidence interval of the fitted spectrum. The
resulting average oxy- and deoxy-hemoglobin concentra-
tions are 37 12 (lmol/L) and 19 6 (lmol/L), respec-
tively. For the scattering, the average number density (a)
is 8.54 1.4 and the average effective particle size (b) is
0.75 0.24, for the relationship derived from the Mie
model (Jacques 2013).
The dashed red line is the mean of fitted spectrum and
the gray shaded area indicates the 95% confidence inter-
val of the mean.
SmO2 trend during cycling incremental
power test
We tested the accuracy of the HbO, HbR, HbT, and
SmO2 estimations from the Humon Beta data against the
MetaOx measurements. The data from all 17 subjects are
included for these calculations. A representative dataset
from the incremental power test is shown in Figure 3,
where the change of SmO2 over time is presented for
both MetaOx (dotted curve) and Humon Beta (solid
curve). The vertical lines indicate the time point that the
power on the bike was changed and the numbers at the
top of the plot between the lines show the power (Watts)
at which the subject was cycling. In general, at the begin-
ning of the exercise, the muscle oxygen saturation slightly
Figure 2. The measured absorption coefficient (la) and reduced scattering coefficient (ls’) of rectus femoris muscle, and their fitted spectrum.
The red error bars indicate the mean over all subjects and 95% confidence interval of the mean. The dashed red line is the fitted spectrum and
the gray shaded area indicates 95% confidence interval of the fitted spectrum.
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
2018 | Vol. 6 | Iss. 7 | e13664
Page 5
P. Farzam et al.
A New Wearable NIRS to Estimate Lactate Threshold
increases. As the power increases and the subject under-
goes high exertion levels, the oxygen saturation starts to
decrease. Finally, SmO2 sharply increases when the subject
starts the recovery phase.
Figure 4 is a scatter plot of the SmO2 values measured
by the two devices in all subjects during cycling. The solid
black line in Figure 4 indicates a line with slope of one
and zero intercept, which represents the trend if there
was a 1:1 relationship between the measured SmO2 by
Humon Beta and the MetaOx. The dashed line shows the
best linear fit to the data, with an intercept of ~10 and
R = 0.74. This linear relationship was found to be signifi-
cant (P value < 0.001 for slope and intercept). With
respect to the MetaOx, the Humon Beta overestimates
(2–3%) SmO2 at low values and underestimates (1%)
SmO2 at high values.
Quantifying differences between Humon
Beta and MetaOx measurements
The results of the comparison between Humon Beta and
MetaOx, MetaOx-subset and MetaOx, and Humon Beta
at two locations are reported in Table 2. In all three cases,
the correlation coefficient and the relative RMSE between
two measures are reported for HbO, HbR, HbT, and
SmO2, averaged over all subjects.
The results show a good agreement between Humon
Beta and MetaOx data. The differences are larger for
HbO, HbR, and HbT (19.4%, 25.7%, and 20.7%, respec-
tively) and smaller for SmO2 (3.4%). While the absolute
hemoglobin concentration values are different, a strong
correlation (correlation coefficient: 0.72-0.86) is present
between the time traces measured with the Humon Beta
and the MetaOx.
The MetaOx-subset assumes the same fixed scattering
coefficient and similar wavelengths and separations as the
Humon Beta. Table 2 shows that the errors due to the
simplified model results in a small deviation from the full
MetaOx
data.
The
average
relative
RMSE
between
MetaOx-subset
and
MetaOx
are
14.7%,
and
14.7%,
12.8%, for HbO, HbR, and HbT, respectively, and 3.6%
for SmO2 (correlation coefficient: 0.83–0.96).
To estimate differences due to differences in probe
location on the same rectus femoris, the data from the
two Humon Betas worn on the same leg (7 subjects) are
compared and the results are shown in Table 2, Humon
location 1 vs. Humon location 2.
Features within the SmO2 curves were also compared
between the MetaOx and Humon Beta measurements for
the 14 male subjects. Table 3 reports the average SmO2%
drop throughout the cycling tests, the average time at
which 50% of the recovery occurred, and the average
SmO2% overshoot relative to the average starting value
during the first power increment (30 W). A Wilcoxon
rank sum test was performed between the MetaOx and
Figure 3. SmO2 results for a representative subject (subject #4)
during the incremental cycling test. Humon Beta SmO2 (solid line)
and MetaOx SmO2 (dashed line) absolute values are 3–5% different
but the lines closely follow each other for the whole duration of
the exercise. The vertical lines indicate the different cycling power
periods with the power level indicated on the top.
Figure 4. : Scatter plot of the SmO2 values measured by Humon
Beta (y-axis) and MetaOx (x-axis) during exercise for all subjects.
Data were down sampled to 12 sec per point, to simplify the
figure. A strong linear relationship can be observed between the
SmO2 measured by both systems. The equation and correlation
coefficient of the best linear fit (dashed line) are reported in the
figure.
2018 | Vol. 6 | Iss. 7 | e13664
Page 6
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
A New Wearable NIRS to Estimate Lactate Threshold
P. Farzam et al.
Humon Beta data and the p values shown in the final
row
of
Table 3
indicate
that
these
values
are
not
significantly
different
between
the
two
systems
(P
values > 0.05).
Determining the lactate threshold power
The SmO2 data for the 14 male subjects were analyzed to
determine if the lactate threshold power could be found
using MetaOx and Humon Beta SmO2 changes. Neither
an SmO2 absolute threshold value or relative threshold
drop could identify the lactate threshold power accurately.
In particular, due to the large variability in SmO2 values
across subjects, we could not find an absolute threshold
value that could work for all subjects. Also, the difference
between maximum SmO2 and SmO2 at 4 mmol/L lactate
concentration (SmO2-drop) varied a lot across subjects. At
best, with either the MetaOx or the Humon Beta, by set-
ting the threshold drop equal to the average SmO2-drop
across subjects, we obtained an average power difference
with respect to the blood lactate measure of about 40 W
and a delay of about 5 min.
Instead, the Humon Beta proprietary algorithm used to
determine the lactate threshold showed good agreements
with the blood lactate measurements. Representative cases
of the Humon Beta SmO2 with the four exercise zones are
shown for subjects one and two in Figure 5. The beginning
of the red zone should correspond to the 4 mmol/L lactate
concentration threshold. While the Humon Beta algorithms
work well for subject #2, there is a relatively large delay on
the estimate of lactate power in subject #1. We evaluated the
accuracy of this threshold with respect to the blood lactate
concentration measures in all 14 male subjects.
Table 4 shows the difference between the threshold
power found by the blood lactate measurement (interpo-
lating the data acquired every 4 min) and the threshold
power determined by the Humon Beta proprietary algo-
rithm in the 14 male subjects. The time difference
between the two thresholds is also reported. Since the
power increments during the cycling test are 30 W, the
Humon Beta identified 4 threshold powers correctly and
was off by 1 power increment (30 W) in 10 subjects. The
absolute average power difference is 21.4 W and the abso-
lute average time difference is 2:32 min.
Table 2. Results comparisons.
Comparison
Parameter
Correlation coefficient
Relative RMSE (%)
Humon vs. MetaOx
HbO
0.80 0.21
19.4 14.4
HbR
082 0.19
25.7 17.6
HbT
0.72 0.33
20.7 15.6
SmO2
0.86 0.10
3.4 0.9
MetaOx-subset vs. MetaOx
HbO
0.89 0.10
14.7 10.9
HbR
0.96 0.08
14.7 9.9
HbT
0.83 0.16
12.8 8.7
SmO2
0.98 0.02
3.6 1.9
Humon loc.1 vs.Humon loc. 2
HbO
0.86 0.15
22.4 15.1
HbR
0.85 0.16
16.4 10.8
HbT
0.92 0.03
20.4 14.6
SmO2
0.87 0.14
3.6 3.0
The average of correlation coefficients and relative root mean square error (RMSE) of HbO, HbR, HbT, and SmO2 over all subjects comparing
two different datasets as described in the first column. The comparison of Humon Beta at two locations on the muscle is performed on seven
subjects, which had two Humon devices on the same muscle.
Table 3. SmO2 feature comparison.
SmO2 Drop (%)
Time to 50% Recovery (sec)
SmO2 Overshoot (%)
Humon
11.9 5.9
123.4 39.7
8.2 4.9
MetaOx
11.6 5.8
100.8 42.4
6.9 3.7
P value
0.87
0.18
0.60
The SmO2% drop throughout the incremental test (first column); The average time to reach 50% of the recovery (middle column), and the
average SmO2% overshoot (last column) are reported for the two systems. The p-values indicate there are not significant differences between
the measured parameters with the two devices.
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
2018 | Vol. 6 | Iss. 7 | e13664
Page 7
P. Farzam et al.
A New Wearable NIRS to Estimate Lactate Threshold
Influence of superficial layers on the
measured SmO2
Using a silicone tissue mimicking phantom with known
optical properties, we verified that neutral density (ND)
filters minimally affect the recovered effective SmO2 as
measured by the Humon Beta and MetaOx (Fig. 6 plots
A and B). By placing neutral density filters of 0.1, 0.2,
0.5, and 0.7 optical densities over the phantom, the
FDNIRS light intensity measured by the MetaOx is atten-
uated near 100% (Fig. 6C). In the linear fitting of inten-
sity, amplitude, and phase versus distance, the thin
absorbing layer affects the intercepts while the corre-
sponding slopes are minimally affected (Fig. 6D). As a
result, the recovered absorption and reduced scattering
coefficients of the phantom are not impacted by the pres-
ence of the thin superficial absorber. Therefore, the recov-
ered effective oxygen saturation (SmO2*) is not affected
as shown in Figure 6B, where the SmO2* changes with
respect to no filter are less than 2%. Similarly, Figure 6A
shows that the SmO2 estimated by the Humon Beta,
which also rely on multidistance algorithms, is not
affected by the presence of the ND filters.
Finally, to evaluate the effect of subcutaneous adipose
tissue on the estimated muscle hemoglobin parameters,
we compared
MetaOx
results at shorter and
larger
Figure 5. Two representative cases of Humon SmO2 with the estimated zones for (A) subject #2 and (B) subject #1. The measured blood
lactate concentration is plotted in the right y-axis (empty circles). The estimate 4 mmol/L blood lactate threshold is indicated by a red star. The
vertical lines indicate the different cycling power periods with the power level indicated on the top.
Table 4. Comparison of lactate threshold power estimated by
blood lactate concentration and Humon Beta proprietary algo-
rithm.
Subject #
Power
Difference (W)
Time Difference
(min:sec)
1
30
4:35
2
0
0:44
4
30
1:06
5
0
1:14
7
0
1:25
8
30
3:06
9
30
1:55
10
30
4:37
11
0
1:25
12
30
1:58
14
30
2:56
15
30
2:34
17
30
4:47
18
30
3:11
Absolute average difference
21.4 14.1
2:32 1:23
Difference between the threshold power found by the blood lac-
tate concentration measurements and by the Humon Beta algo-
rithm for the 14 male subjects, and the time difference between
the two. The positive values indicate power and time delays of the
Humon Beta with respect to the blood lactate. The last row
reports the absolute average values.
2018 | Vol. 6 | Iss. 7 | e13664
Page 8
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
A New Wearable NIRS to Estimate Lactate Threshold
P. Farzam et al.
separations. Using larger separations, we consistently
obtained higher hemoglobin concentration. Across all
subjects, the difference in total hemoglobin concentration
is about 17%. Instead, the difference in SmO2 at the two
separations is small (2.8% 1.4%). In addition, we
found the SmO2 drop at the end of the exercise to be on
average 11.8 5.2% and 12.7 5.5%, using the shorter
and larger separations, respectively. By performing a Wil-
coxon Rank Sum test, these drops were not found to be
significantly different (P > 0.05). This indicates that the
depth penetration is similar between the shorter and
longer separations used. To further assess the effect of the
subcutaneous adipose tissue on SmO2, we estimated the
correlation between the total SmO2 drop (using all
source–detector separations available in each system) and
the SCATT in all subjects. As shown in Figure 7, we con-
sistently
measured
smaller
SmO2
drops
for
thicker
Figure 6. Effect of a superficial thin absorbing layer on measured optical parameters. Four different neutral density filters were positioned
between the optical probe and a silicone phantom to mimic a thin attenuating superficial layer. Panels (A) and (B) report the difference
between the effective hemoglobin saturation (SmO2*, the SmO2 we would have estimated if there was blood in the phantom) measured by
Humon Beta and MetaOx with or without filter. The error bars represent the difference over five repetitions. Panel (C) shows the strong
attenuation of the detected light at 3.0 cm source–detector distance at a representative wavelength (690 nm) in the MetaOx system. Panel (D)
shows the corresponding amplitude and phase slope changes with the neutral density filters. Results are consistent at all wavelengths,
independent of the absorbance spectra of the ND filters. In fact, ND filters have constant absorption in the visible spectral range, but they
attenuate less in the near-infrared range. This difference in attenuation spectra does not affect the effective SmO2 calculated using both red
and near-infrared wavelengths, since at each wavelength the effect of the ND on the slopes is negligible.
Figure 7. Scatterplot of SmO2 drop during exercise versus
subcutaneous adipose tissue. With both devices, for thicker adipose
tissue we consistently measured smaller drops in SmO2.
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
2018 | Vol. 6 | Iss. 7 | e13664
Page 9
P. Farzam et al.
A New Wearable NIRS to Estimate Lactate Threshold
subcutaneous adipose layers (with a correlation coefficient
of 0.65).
Discussion
In this study, we recorded NIRS data from 17 athletes while
they were performing an incremental cycling test. The opti-
cally measured hemoglobin parameters (HbO, HbR, HbT,
and SmO2) responded to muscle physiological changes
during exercise. These results exhibit meaningful trends
that can provide insights to how muscles respond to exer-
tion. The average optical (absorption and reduced scatter-
ing coefficients) properties are illustrated in Figure 2. The
measured values are consistent with values reported in the
literature using similar NIRS methods to measure skeletal
muscles (Yu et al. 2005; Gurley et al. 2012; Shang et al.
2012; Mesquita et al. 2013; Baker et al. 2017).
Comparison of Humon Beta and MetaOx
estimates of SmO2
During exercise, the body supplies a higher level of blood
flow to the working muscles to provide oxygen (Yu et al.
2005; Joyner and Casey 2015). In the initial phase of
cycling, when the athletes were working at a lower power,
there was an overcompensation of blood, thus SmO2
slightly increased (Fig. 3 and Fig. 5). As the subjects
approached their lactate threshold power, the SmO2
decreased, indicating that the oxygen consumption in the
muscle had exceeded the oxygen supply. When the subjects
could no longer continue pedaling at high power, and
began the recovery phase, the SmO2 significantly increased
due to the high blood supply and sudden decrease of mus-
cle oxygen consumption. The agreement between the esti-
mated SmO2 between the Humon and MetaOx system is
consistent across all phases of exercise and all subjects as
shown in Table 2 by the high correlation coefficient (0.86).
Not only the trends, but also the absolute values on SmO2
are quite similar between the two systems, as shown by the
low average RMSE (3.4%). Figure 4 shows that the range of
variations of SmO2 measured with the Humon Beta system
is smaller than the range measured with the MetaOx. Using
the MetaOx subset model, we verified this difference is not
due to the fixed scattering assumption. In fact, the R2
between the two MetaOx models is 0.87 and the range of
variation is actually larger by fixing scattering than by cal-
culating it. Throughout the duration of the exercise, the
scattering coefficients also changed minimally (on average
less than 10%). The difference between the Humon and
MetaOx estimated SmO2 is probably due to the different
consumption of the right and left leg muscles. Eighty-eight
percent of our subjects were right-footed and the MetaOx
probe was measuring the dominant leg, therefore this may
cause some discrepancies between measurements. Bilateral
differences have been found by Hesford et al. (2012), who
examined asymmetry between the SmO2 of the left and
right quadriceps in ice skaters, and found differences
depending on the muscle side that was being exerted more.
The data from Table 3, supports the similarities between
the features of the SmO2 curves measured on the 14 male
subjects by the MetaOx and Humon Beta. We investigated
the relationship between recovery time and maximum
blood lactate concentration, and similar to Chance et al.
(1992), we found these variables are uncorrelated. Chance
et al. (1992) did find, however, that if individuals worked
at a fraction of their maximum power level, there was a
correlation between the recovery time and blood lactate
accumulation. We cannot verify this finding since in our
protocol we only measured recovery to maximum power
output.
Comparison of Humon Beta and MetaOx
estimates of hemoglobin concentration
While there was strong agreement between the SmO2 of
the two devices, we found larger differences in the hemo-
globin concentration values. This result is somehow
expected since the assumption of fixed scattering heavily
affects the hemoglobin concentrations absolute values
(Fantini et al. 1999). Nevertheless, we observed that the
model error (fixed scattering, no phase information, only
two wavelengths and only three separations vs. full
FDNIRS MetaOx data), is smaller than the effect of mus-
cle heterogeneity. As reported in Table 2, the RMSE
between Humon and MetaOx are similar to the RMSE
found comparing two Humon locations and double the
RMSE found comparing different MetaOx models. This
suggests muscle heterogeneity plays a larger role than dif-
ferences between CW and FDNIRS models. The hemoglo-
bin concentration spatial distribution throughout the
muscle has been previously reported (Hamaoka et al.
2011), and it needs to be considered, when assessing mus-
cle physiology with NIRS. Importantly, we found that
while hemoglobin content significantly differs across loca-
tions, its temporal changes during the incremental cycling
exercise are uniform across measured parts of the working
muscle. In fact, in all our comparisons, we found good
correlation coefficients for HbO, HbR, and HbT.
Determination of the threshold power
On average SmO2 values at exercise onset in the 14 male
subjects
were
67.8 4.0%
for
the
MetaOx
and
65.6 4.9% for the Humon Beta. SmO2 increased during
the initial power increments and at the lactate threshold
has decreased 6.2 4.6% and 6.3 3.7% from the
2018 | Vol. 6 | Iss. 7 | e13664
Page 10
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
A New Wearable NIRS to Estimate Lactate Threshold
P. Farzam et al.
maximum SmO2, measured by the MetaOx and Humon
Beta, respectively. Because of large variability across sub-
jects, we could not find an SmO2 threshold value that could
work for all subjects to estimate lactate threshold power
(the best attempt found an average power difference of
40 W and a delay of 5 minutes). This is in agreement with
previous NIRS works that show how the use of more com-
plex analysis techniques of SmO2 in addition to HbR,
HbO, and HbT are needed to better assess lactate threshold
power from NIRS (Belardinelli et al. 1995; Grassi et al.
1999; Wang et al. 2006; van der Zwaard et al. 2016). The
drop in SmO2 was larger when considering the maximum
SmO2 measured during the low power intervals as the ref-
erence value. SmO2 at rest, before starting pedaling, was on
average 3.5% lower than during the first power interval
(30 W). We believe that in addition to the increase of
blood flow with exercise onset which increases SmO2, the
muscle contraction and skin tightening during the pedaling
play a role here, reducing the contribution of superficial
adipose tissue to the measured parameters. For this reason,
to avoid this possible measurement artifact, in all our anal-
yses, we only considered differences while pedaling, not
with respect to rest.
The Humon Beta uses a combination of HbO, HbR,
HbT, and SmO2 changes to identify the lactate threshold.
We tested the agreement between the Humon estimated
threshold with the one identified by the lactate concentra-
tion measures and found a close agreement between the
two. The Humon Beta’s real-time threshold power on
average only differs 21.4 W and less than 3 min from the
invasive measure of blood lactate (Table 4). On the seven
subjects with two Humon Beta devices, we also verified
the time at which the threshold power was found was on
average less than one minute apart between the two loca-
tions. The threshold power algorithms are proprietary to
Dynometrics Inc. and are being refined as more data is
collected in controlled exercise settings like this one.
Finally, testing the effects that a light attenuating thin
superficial layer has on the NIRS data is crucial to ensure
dependable results across a wide range of individuals with
different melanin concentration in the skin. It is impor-
tant to note that ND filters have constant absorption in
the visible spectral range, but they attenuate less in the
near-infrared range. This difference in attenuation spectra
does not affect the effective SmO2 calculated using both
red and near-infrared wavelengths, since at each wave-
length the effect of the ND on the slopes is negligible.
Both the MetaOx and the Humon Beta recovered correct
effective SmO2 values across all the absorbing filters used.
The reason why the effect of the filter is negligible is that
both the MetaOx and the Humon Beta recover the optical
properties from the gradient over distance, not from the
absolute intensity values. This ensures confidence that the
Humon Beta can measure SmO2 in people of all skin col-
ors (Franceschini et al. 1998).
Limitations
It is important to address that when validating the
Humon Beta against the MetaOx, the NIRS data was
obtained on different legs of the subjects. While cycling
typically engages both legs uniformly, more differences
may arise in the data because the two systems were not
measuring the exact same location.
For the lactate power analysis, we had to exclude the
female subjects, since we realized this protocol is not well
suited for females. The power increments were too large,
which resulted in the lactate threshold power and end of
the exercise to be too close in time. We suggest smaller
power increments for each step to be used when measur-
ing female subjects who have smaller quadriceps com-
pared to males.
The subcutaneous adipose layer influences quantifica-
tion of hemoglobin parameters, especially at shorter
source–detector separations. We verified that using larger
(2.5–3.0 cm) source–detector separations relative to the
shorter (1.5–2.0 cm) we obtain an average of ~17%
higher hemoglobin concentration over the exercise dura-
tion. This is expected because of the larger penetration
depth at longer separations and because of the larger
hemoglobin concentration of muscle with respect to adi-
pose tissue. As a result, in the presence of a subcutaneous
adipose tissue layer, we measure higher hemoglobin con-
centrations at larger separations. Since fat and muscle
have similar oxygenation at rest, the difference in SmO2
at the two separations is small (2.8% 1.4%). We veri-
fied that during exercise, the difference in SmO2 at the
two sets of distances is constant and did not have statisti-
cally significant differences in the SmO2 drops by the end
of exercise. Nevertheless, at both sets of distances, the
presence of a subcutaneous adipose tissue layer affects the
measured SmO2 as shown by the correlation between the
SmO2 drop and the SCATT in Figure 7. For thicker
SCATT layers, there are smaller drops in SmO2 since the
volume measured includes less muscle and more fat. For
the specific application in athletes and fit individuals, the
SCATT is low, reducing the contamination by the adipose
tissue. While we only measured the SCATT over the rec-
tus femoris of the right leg, we also assumed that the
SCATT is symmetric and that this layer reported in Fig-
ure 7 affect similarly the MetaOx and Humon Beta data.
In this study, we chose to monitor the rectus femoris
rather than the vastus lateralis muscle (which is the com-
mon choice in other work examining NIRS measurements
in sports (Perrey and Ferrari 2018) to minimize optical
fiber movement throughout the cycling exercise. Further
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
2018 | Vol. 6 | Iss. 7 | e13664
Page 11
P. Farzam et al.
A New Wearable NIRS to Estimate Lactate Threshold
work needs to be done to examine the differences in
SmO2 measured in multiple muscle groups simultane-
ously, using several Humon Beta devices to help under-
stand what muscle group is the optimal choice for certain
types of activities. It is also crucial to determine if the
identification of an athletes’ threshold by NIRS is depen-
dent on the exact muscle chosen from the many working
muscles during a sport.
When comparing the NIRS measurements against the
lactate threshold, it is relevant to mention that multiple
methods exist to find the lactate threshold (Faude et al.
2009). By choosing a different threshold identification tech-
nique (besides 4 mmol/L), a new threshold power may be
identified, which may alter the values in Table 4. We chose
to determine the lactate threshold as 4 mmol/L because of
the widespread use and established studies that support this
method (Heck et al. 1985; Faude et al. 2009).
Conclusion
In summary, this study validates the performance of a low-
cost, wireless, wearable NIRS device against an advanced
benchtop device. During the incremental cycling test, the
wearable device provides similar results to the more expen-
sive FDNIRS technology. We verified that the assumptions
and simplifications of this CWNIRS system minimally
impact the SmO2 quantification and the recovery of
changes in the hemoglobin concentration trends. The main
deviations are accounted for by muscle heterogeneities.
While skin color does not affect the results, the main limi-
tation, common to all CW and FDNIRS systems, is the
reduced sensitivity to muscles in the presence of subcuta-
neous adipose tissue. Targeting athletes, who tend to be fit
individuals with thin adipose layers, provides a larger drop
in SmO2 readings than what can be achieved on people
with thicker adipose thicker layers. Finally, we demon-
strated the Humon Beta shows a good accuracy in predict-
ing the lactate threshold level. Thanks to the low cost, small
size, and low sensitivity to motion, the Humon Beta wear-
able and its accompanying smartphone application can
provide useful information about exercising muscle physi-
ology and help athletes optimize their training.
Acknowledgments
We thank Dr. Pamela Anderson and Dr. Daniel Wiese from
Dynometrics, Inc. for the scientific interaction and help
with the measurements and analysis of the Humon data.
Conflict of Interest
Dr. Franceschini has patents on the FDNIRS technology
used in this paper.
References
Bailey, S. J., D. P. Wilkerson, F. J. Dimenna, and A. M. Jones.
2009. Influence of repeated sprint training on pulmonary
O2 uptake and muscle deoxygenation kinetics in humans. J.
Appl. Physiol. (1985) 106:1875–1887.
Baker, W. B., Z. Li, S. S. Schenkel, M. Chandra, D. R. Busch,
E. K. Englund, et al., 3rd. 2017. Effects of exercise training
on calf muscle oxygen extraction and blood flow in patients
with peripheral artery disease. J. Appl. Physiol. (1985)
123:1599–1609. jap 00585 02017.
Belardinelli, R., T. J. Barstow, J. Porszasz, and K. Wasserman.
1995. Changes in skeletal muscle oxygenation during
incremental exercise measured with near infrared
spectroscopy. Eur. J. Appl. Physiol. Occup. Physiol. 70:487–
492.
Bentley, D. J., J. Newell, and D. Bishop. 2007. Incremental
exercise test design and analysis: implications for
performance diagnostics in endurance athletes. Sports Med.
37:575–586.
Bhambhani, Y. N., S. M. Buckley, and T. Susaki. 1997.
Detection of ventilatory threshold using near infrared
spectroscopy in men and women. Med. Sci. Sports Exerc.
29:402–409.
Boas, D. A., S. Sakadzic, J. Selb, P. Farzam, M. A.
Franceschini, and S. A. Carp. 2016. Establishing the diffuse
correlation spectroscopy signal relationship with blood flow.
Neurophotonics 3:031412.
Boone, J., K. Vandekerckhove, I. Coomans, F. Prieur, and J. G.
Bourgois. 2016. An integrated view on the oxygenation
responses to incremental exercise at the brain, the
locomotor and respiratory muscles. Eur. J. Appl. Physiol.
116:2085–2102.
Borges, N. R., and M. W. Driller. 2016. Wearable lactate
threshold predicting device is valid and reliable in runners.
J. Strength Cond. Res. 30:2212–2218.
Carp, S. A., P. Farzam, N. Redes, D. M. Hueber, and M. A.
Franceschini. 2017. Combined multi-distance frequency
domain and diffuse correlation spectroscopy system with
simultaneous data acquisition and real-time analysis.
Biomed. Opt. Express 8:3993–4006.
Chance, B., M. T. Dait, C. Zhang, T. Hamaoka, and F.
Hagerman. 1992. Recovery from exercise-induced
desaturation in the quadriceps muscles of elite competitive
rowers. Am. J. Physiol. 262:C766–C775.
Driller, M., D. Plews, and N. Borges. 2016. Wearable near
infrared sensor for determining an athlete’s lactate threshold
during exercise. NIR News 27:8–10.
Durduran, T., R. Choe, W. B. Baker, and A. G. Yodh. 2010.
Diffuse optics for tissue monitoring and tomography. Rep.
Prog. Phys. 73:pii: 076701.
Esteve-Lanao, J., C. Foster, S. Seiler, and A. Lucia. 2007.
Impact of training intensity distribution on performance in
endurance athletes. J. Strength Cond. Res. 21:943–949.
2018 | Vol. 6 | Iss. 7 | e13664
Page 12
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
A New Wearable NIRS to Estimate Lactate Threshold
P. Farzam et al.
Fantini, S., M. A. Franceschini, J. S. Maier, S. A. Walker, B.
Barbieri, and E. Gratton. 1995. Frequency-domain
multichannel optical detector for non-invasive tissue
spectroscopy and oximetry. Opt. Eng. 34:32–42.
Fantini, S., D. Hueber, M. A. Franceschini, E. Gratton, W.
Rosenfeld, P. G. Stubblefield, et al. 1999. Non-invasive
optical monitoring of the newborn piglet brain using
continuous-wave and frequency-domain spectroscopy. Phys.
Med. Biol. 44:1543–1563.
Farzam, P., P. Zirak, T. Binzoni, and T. Durduran. 2013.
Pulsatile and steady-state hemodynamics of the human
patella bone by diffuse optical spectroscopy. Physiol. Meas.
34:839–857.
Faude, O., W. Kindermann, and T. Meyer. 2009. Lactate
threshold concepts: how valid are they? Sports Med. 39:469–
490.
Franceschini, M. A., S. Fantini, A. Cerussi, B. Barbieri, B.
Chance, and E. Gratton. 1997. Quantitative spectroscopic
determination of hemoglobin concentration and saturation
in a turbid medium: analysis of the effect of water
absorption. J. Biomed. Optics 2:147–153.
Franceschini, M. A., S. Fantini, L. A. Paunescu, J. S. Maier,
and E. Gratton. 1998. Influence of a superficial layer in the
quantitative spectroscopic study of strongly scattering
media. Appl. Opt. 37:7447–7458.
Goodwin, M. L., J. E. Harris, A. Hernandez, and L. B.
Gladden. 2007. Blood lactate measurements and analysis
during exercise: a guide for clinicians. J. Diabetes Sci.
Technol. 1:558–569.
Grassi, B., V. Quaresima, C. Marconi, M. Ferrari, and P.
Cerretelli. 1999. Blood lactate accumulation and muscle
deoxygenation during incremental exercise. J. Appl. Physiol.
(1985) 87:348–355.
Grassi, B., S. Pogliaghi, S. Rampichini, V. Quaresima, M.
Ferrari, C. Marconi, et al. 2003. Muscle oxygenation and
pulmonary gas exchange kinetics during cycling exercise on-
transitions in humans. J. Appl. Physiol. (1985) 95:149–158
Gurley, K., Y. Shang, and G. Yu. 2012. Noninvasive optical
quantification of absolute blood flow, blood oxygenation,
and oxygen consumption rate in exercising skeletal muscle.
J. Biomed. Opt. 17:075010.
Hamaoka, T., H. Iwane, T. Shimomitsu, T. Katsumura, N.
Murase, S. Nishio, et al. 1996. Noninvasive measures of
oxidative metabolism on working human muscles by near-
infrared spectroscopy. J. Appl. Physiol. (1985) 81:1410–1417
Hamaoka, T., K. K. McCully, M. Niwayama, and B. Chance.
2011. The use of muscle near-infrared spectroscopy in sport,
health and medical sciences: recent developments. Philos
Trans A Math Phys Eng Sci 369:4591–4604.
Hammer, S. M., A. M. Alexander, K. D. Didier, J. R. Smith, J.
T. Caldwell, S. L. Sutterfield, et al. 2018. The noninvasive
simultaneous measurement of tissue oxygenation and
microvascular hemodynamics during incremental handgrip
exercise. J. Appl. Physiol. 124:604–614.
Heck, H., A. Mader, G. Hess, S. Mucke, R. Muller, and W.
Hollmann. 1985. Justification of the 4-mmol/l lactate
threshold. Int. J. Sports Med. 6:117–130.
Hesford, C. M., S. J. Laing, M. Cardinale, and C. E. Cooper.
2012. Asymmetry of quadriceps muscle oxygenation during
elite short-track speed skating. Med. Sci. Sports Exerc.
44:501–508.
Ihsan, M., C. R. Abbiss, M. Lipski, M. Buchheit, and G.
Watson. 2013. Muscle oxygenation and blood volume
reliability during continuous and intermittent running. Int.
J. Sports Med. 34:637–645.
Jacques, S. L. 2013. Optical properties of biological tissues: a
review. Phys. Med. Biol. 58:R37–R61.
Joyner, M. J., and D. P. Casey. 2015. Regulation of increased
blood flow (hyperemia) to muscles during exercise: a
hierarchy of competing physiological needs. Physiol. Rev.
95:549–601.
Kleiser, S., N. Nasseri, B. Andresen, G. Greisen, and M. Wolf.
2016. Comparison of tissue oximeters on a liquid phantom
with adjustable optical properties. Biomed. Opt. Express
7:2973–2992.
Madden, C., M. Putukian, E. McCarty, and C. Young. 2013.
Netter’s Sports Medicine E-Book. Elsevier Health Sciences,
Amsterdam, Netherlands.
Mesquita, R. C., M. Putt, M. Chandra, G. Yu, X. Xing, S. W.
Han, et al. 2013. Diffuse optical characterization of an
exercising patient group with peripheral artery disease. J.
Biomed. Opt. 18:57007.
Perrey, S., and M. Ferrari. 2018. Muscle oximetry in sports
science: a systematic review. Sports Med. 48:597–616.
Prahl, S. Optical Absoprtion of Hemoglobin http://omlc.org/
spectra/hemoglobin/.
Quaresima, V., M. Ferrari, M. A. Franceschini, M. L.
Hoimes, and S. Fantini. 2004. Spatial distribution of
vastus lateralis blood flow and oxyhemoglobin saturation
measured at the end of isometric quadriceps contraction
by multichannel near-infrared spectroscopy. J. Biomed.
Opt. 9:413–420.
Racinais, S., M. Buchheit, and O. Girard. 2014. Breakpoints in
ventilation, cerebral and muscle oxygenation, and muscle
activity during an incremental cycling exercise. Front.
Physiol. 5:142.
Seiler, K. S., and G. O. Kjerland. 2006. Quantifying training
intensity distribution in elite endurance athletes: is there
evidence for an “optimal” distribution? Scand. J. Med. Sci.
Sports 16:49–56.
Shang, Y., K. Gurley, B. Symons, D. Long, R. Srikuea, L. J.
Crofford, et al. 2012. Noninvasive optical characterization of
muscle blood flow, oxygenation, and metabolism in women
with fibromyalgia. Arthritis Res. Ther. 14:R236.
Soller, B. R., Y. Yang, S. M. Lee, C. Wilson, and R. D. Hagan.
2008. Noninvasive determination of exercise-induced
hydrodgen ion threshold through direct optical
measurement. J. Appl. Physiol. (1985) 104:837–844.
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
2018 | Vol. 6 | Iss. 7 | e13664
Page 13
P. Farzam et al.
A New Wearable NIRS to Estimate Lactate Threshold
Wang, L., T. Yoshikawa, T. Hara, H. Nakao, T. Suzuki, and S.
Fujimoto. 2006. Which common NIRS variable reflects
muscle estimated lactate threshold most closely? Appl.
Physiol. Nutr. Metab. 31:612–620.
Yodh, A., and B. Chance. 1995. Spectroscopy and imaging
with diffusing light. Phys. Today 48:34–41.
Yu, G., T. Durduran, G. Lech, C. Zhou, B. Chance, E. R. 3rd
Mohler, et al. 2005. Time-dependent blood flow and
oxygenation in human skeletal muscles measured with
noninvasive near-infrared diffuse optical spectroscopies. J.
Biomed. Opt. 10:024027.
van der Zwaard, S., R. T. Jaspers, I. J. Blokland, C. Achterberg,
J. M. Visser, A. R. den Uil, et al. 2016. Oxygenation
threshold derived from near-infrared spectroscopy: reliability
and its relationship with the first ventilatory threshold. PLoS
ONE 11:e0162914.
2018 | Vol. 6 | Iss. 7 | e13664
Page 14
ª 2018 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of
The Physiological Society and the American Physiological Society.
A New Wearable NIRS to Estimate Lactate Threshold
P. Farzam et al.
| Validation of a novel wearable, wireless technology to estimate oxygen levels and lactate threshold power in the exercising muscle. | [] | Farzam, Parisa,Starkweather, Zack,Franceschini, Maria A | eng |
PMC4892259 | Supplementary Information
Inferring muscle functional roles of the ostrich pelvic limb during walking and running using
computer optimization
Jeffery W. Rankin, Jonas Rubenson, John R. Hutchinson
Supplementary Information – Tables
Table S1 – Root mean square (RMS) and peak differences between the experimental and simulated joint angles (degrees)
obtained with computed muscle control in walking and running over the entire movement.
RMS Error Peak Error RMS Error Peak Error RMS Error Peak Error RMS Error Peak Error
Hip Flexion-Extension
0.002
0.061
0.001
0.039
0.007
0.148
0.002
0.044
Hip Ad-Abduction
0.002
0.106
0.001
0.034
0.005
0.164
0.002
0.065
Hip Medial-Lateral Rotation
0.005
0.139
0.003
0.076
0.012
0.347
0.003
0.054
Knee Flexion-Extension
0.006
0.201
0.002
0.047
0.024
0.497
0.002
0.069
Knee Ad-Abduction
0.005
0.130
0.002
0.103
0.014
0.361
0.004
0.071
Knee Medial-Lateral Rotation
0.005
0.236
0.003
0.071
0.010
0.264
0.005
0.122
Ankle Flexion-Extension
0.009
0.325
0.001
0.053
0.051
1.304
0.004
0.093
MTP Flexion-Extension
0.010
0.292
0.007
0.313
0.051
1.543
0.008
0.193
RCMCR
RCMCC
WCMCR
WCMCC
Joint Angle (Degrees)
Table S2 –Percent of muscle activity (as predicted by the simulations) that occurs during the swing phase. Predictions were
calculated by normalizing the entire gait cycle to 100% and then integrating the entire cycle and swing-phase-only portions
of the simulated muscle activation patterns.
Simulation Percent of Muscle
Activity in Swing
WSO
13.9
WCMCR
15.6
WCMCC
29.8
RSO
24.4
RCMCR
31.8
RCMCC
38.6
Supplementary Information
Inferring muscle functional roles of the ostrich pelvic limb during walking and running using
computer optimization
Jeffery W. Rankin, Jonas Rubenson, John R. Hutchinson
Supplementary Information – Figures
Figure S1 – Example fibre excursions of the digital flexor muscles during ostrich running. Due to having short muscle fibres,
these muscles undergo extensive fibre lengthening relative to their optimal fibre length during swing. As a result, passive
muscle forces generate a large ankle extension moment that cannot be counteracted by the ankle flexors (TCf,TCt, EDL)
alone and a reserve torque is required. Shaded area represents stance and the arrow indicates the point of peak ankle
reserve actuator torque.
| Inferring muscle functional roles of the ostrich pelvic limb during walking and running using computer optimization. | [] | Rankin, Jeffery W,Rubenson, Jonas,Hutchinson, John R | eng |
PMC9602838 | Citation: Tanous, D.; Motevalli, M.;
Wirnitzer, G.; Leitzmann, C.;
Rosemann, T.; Knechtle, B.; Wirnitzer,
K. Sex Differences in Training
Behaviors of 10 km to
Ultra-Endurance Runners (Part
A)—Results from the NURMI Study
(Step 2). Int. J. Environ. Res. Public
Health 2022, 19, 13238. https://
doi.org/10.3390/ijerph192013238
Academic Editors: Paul B.
Tchounwou, Stacy T. Sims and
Christopher T. Minson
Received: 18 August 2022
Accepted: 11 October 2022
Published: 14 October 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright:
© 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed
under
the
terms
and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
International Journal of
Environmental Research
and Public Health
Article
Sex Differences in Training Behaviors of 10 km to
Ultra-Endurance Runners (Part A)—Results from the NURMI
Study (Step 2)
Derrick Tanous 1,2
, Mohamad Motevalli 1,2
, Gerold Wirnitzer 3, Claus Leitzmann 4, Thomas Rosemann 5
,
Beat Knechtle 5,6,*
and Katharina Wirnitzer 1,2,7
1
Department of Sport Science, University of Innsbruck, 6020 Innsbruck, Austria
2
Department of Research and Development in Teacher Education, University College of Teacher Education,
Tyrol, 6010 Innsbruck, Austria
3
adventureV & change2V, 6135 Stans, Austria
4
Institute of Nutrition, University of Gießen, 35390 Gießen, Germany
5
Institute of Primary Care, University of Zurich, 8000 Zurich, Switzerland
6
Medbase St. Gallen Am Vadianplatz, 9000 St. Gallen, Switzerland
7
Research Center Medical Humanities, Leopold-Franzens University of Innsbruck, 6020 Innsbruck, Austria
*
Correspondence: [email protected]
Abstract: Training for running events is fundamental for successful participation in various running
events such as 10 km, half-marathon, marathon, or ultra-marathon distances. Training behaviors
are likely based on runner motivations and social constraints, particularly for females. Participants
completed a questionnaire following a cross-sectional approach, including questions on sociode-
mographics, general training behaviors, and periodization training strategies. The final sample
included 245 participants (141 females, 104 males), mostly from Germany (72%), Austria (18%), and
Switzerland (5%), with a median age of 39 years (IQR 17) and a BMI of 21.7 kg/m2 (IQR 3.5). Males
more often trained alone and independently, whereas females were most likely to follow an external
resource (p = 0.037). Non-parametric ANOVA revealed significant training differences between sexes
in daily training mileages and durations at each phase and stage (p < 0.05) as well as in weekly
training mileages and durations for general basic training and race-specific training (p < 0.05). Critical
sex differences in training behaviors may arise from physiological differences and social expectations,
which may be related to the distances they prefer to race at as well as their motivations for running
and racing. This study provides a wide overview of training behaviors for endurance runners or
professionals guiding healthy running performance.
Keywords: running; marathon; female; motive; recreational athlete; endurance exercise; habit
1. Introduction
Running has been well-established as a healthy physical activity for females and males
across adulthood [1–3]. While many similar training habits exist between female and
male endurance runners, a multitude of factors remain distinguishable between the sexes,
including primarily physiological and social differences [4–7]. Numerous studies have
investigated sex differences in the training behaviors of endurance runners [8–15]. To the
best of the authors’ knowledge, no investigation has analyzed sex-related running training
behaviors considering diverse distances, including 10 km (10 km) up to ultra-marathon
(UM) distances in one study.
Parallel to health benefits, frequent runners often follow various personal motivations
for running, primarily for leisure or sports performance [3,4,16]. Females generally report
higher running motives for psychological health reasons (body weight concern, a sense
Int. J. Environ. Res. Public Health 2022, 19, 13238. https://doi.org/10.3390/ijerph192013238
https://www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2022, 19, 13238
2 of 14
of belonging, life meaning, fulfillment/harmony), whereas males are mostly competition-
focused [17]. Training provides the structure for progress in meeting the individual’s
motivational goals [15,18] independent of sex.
Considering the anatomical (e.g., anthropometric, hormonal, maximum muscular
strength) [19,20] and societal (e.g., expectations, motives) differences between females and
males [6,21], further sex differences in runner training behaviors may arise, including
general training (training duration, external resource, other sports participation) and
specific training measures (weekly frequency, weekly and daily mileages and durations,
type of training from basic- to competition-specific, and various intensities) as well as the
distances they prefer to run (10 km, half-marathon, marathon/UM) [9,15,17,18]. Regardless
of the endurance runner’s motives, athletes typically adhere to periodization strategies
when training for running events, including recognizable training phases that may unfold
throughout the duration of up to a year before an event [22,23]. Different periods of training
(regeneration and transitional period, main preparatory period, competition period) are
followed to build and/or maximize beneficial training adaptations, mainly to improve the
maximal volume of oxygen consumption in runners to maintain a faster pace for better
finishing times [22–24].
Contrariwise to the sex differences in the training behaviors of endurance runners
identified by the previous literature [8–15], similarities between the sexes are also commonly
observed that appear contradictory [8,9,11–14]. However, this occurrence may be due to
the previous study designs, the strict inclusion of pre-specified race distances, and the
methodological heterogeneity between the studies at the population level [8–15]. Therefore,
this study is the first aiming to assess the relationships between training behaviors of female
and male recreational endurance runners over 10 km up to UM distances. The present
investigation hypothesized that there are critical training differences between the sexes
of recreational endurance runners of various distances (half-marathon, marathon/UM,
10 km).
2. Materials and Methods
The study protocol of the Nutrition and Running High Mileage (NURMI) Study [25] was
approved by the ethics board of St. Gallen, Switzerland on 6 May 2015 (EKSG 14/145). The
trial registration number is ISRCTN73074080 (retrospectively registered). Detailed information
about the methods of the NURMI Study Step 2 has been described elsewhere [3,4,7,26–30].
The NURMI Study was conducted in three steps following a cross-sectional design.
Endurance runners were mainly recruited from Austria, Germany, and Switzerland and
were contacted primarily via social media, websites of the organizers of marathon events,
online running communities, email lists, and runners’ magazines, as well as via magazines
for health, nutrition and lifestyle, trade fairs on sports, plant-based nutrition and lifestyle,
and through personal contacts. The characteristics of the subjects are presented in Table 1.
Participants completed an online survey within the NURMI Study Step 2 that was
available from February 2015 to December 2015 in German and English at www.nurmi-
study.com (accessed on 17 August 2022). Prior to completing the questionnaire on physical
and psychological health (including the basic assignment to an area of sports, motivation
and aim of running activities, and other sports to balance for running in order to better dif-
ferentiate between predominant health, leisure, or sport performance-orientated approach
to running), participants were provided with a written description of the procedures and
participants gave their informed consent to take part in the study.
For complete participation in the study, the following inclusion criteria were required:
(1) written informed consent, (2) at least 18 years of age, (3) questionnaire Step 2 completed,
and (4) completion of at least a half-marathon distance running event within the past two
years. However, as an additional criterion in this study, participants had to (5) select an
event as the NURMI running event, including either a half-marathon (HM) or marathon
(M) distance that they prepared for and subsequently finished in Step 3 (main NURMI
Study: Step 2 linked to Step 3) [25].
Int. J. Environ. Res. Public Health 2022, 19, 13238
3 of 14
An additional group of 91 highly motivated runners provided accurate and pragmatic
answers with plenty of high-quality data and had not successfully participated in either a
HM or M but in a 10 km race instead. To avoid an irreversible loss of these valuable data
sets, those who met all inclusion criteria but named a 10 km race as their running event
were kept as an additional race distance subgroup.
To control for measures of (1) running activity (history, motivation, training, racing,
etc.) and (2) diet, we included control questions in different survey sections. Incomplete,
inconsistent, and conflicting data sets were excluded from the data analysis. Moreover,
to control for a minimal status of health linked to a minimum level of fitness and further
enhance the reliability of data sets, the body mass index (BMI) approach followed the World
Health Organization (WHO) [31,32]. With a BMI ≥ 30, other health protective and weight
loss strategies alongside running are foremost necessary to safely reduce body weight.
Regarding the UM distance, the shortest UM reported was 50 km, and the longest
was 160 km. Figure 1 shows the categorization of participants according to sex and race
distance subgroups: HM, M/UM, and 10 km (M/UM data were pooled since the marathon
distance is within an ultra-marathon). Additionally, the involved reader is kindly referred
to the Part B publication for the subsequential linking of training behaviors with race
performances [33].
nviron. Res. Public Health 2022, 19, x FOR PEER REVIEW
3 of 15
(5) select an event as the NURMI running event, including either a half-marathon (HM)
or marathon (M) distance that they prepared for and subsequently finished in Step 3 (main
NURMI Study: Step 2 linked to Step 3) [25].
An additional group of 91 highly motivated runners provided accurate and prag-
matic answers with plenty of high-quality data and had not successfully participated in
either a HM or M but in a 10 km race instead. To avoid an irreversible loss of these valu-
able data sets, those who met all inclusion criteria but named a 10 km race as their running
event were kept as an additional race distance subgroup.
To control for measures of (1) running activity (history, motivation, training, racing,
etc.) and (2) diet, we included control questions in different survey sections. Incomplete,
inconsistent, and conflicting data sets were excluded from the data analysis. Moreover, to
control for a minimal status of health linked to a minimum level of fitness and further
enhance the reliability of data sets, the body mass index (BMI) approach followed the
World Health Organization (WHO) [31,32]. With a BMI ≥ 30, other health protective and
weight loss strategies alongside running are foremost necessary to safely reduce body
weight.
Regarding the UM distance, the shortest UM reported was 50 km, and the longest
was 160 km. Figure 1 shows the categorization of participants according to sex and race
distance subgroups: HM, M/UM, and 10 km (M/UM data were pooled since the marathon
distance is within an ultra-marathon). Additionally, the involved reader is kindly referred
to the Part B publication for the subsequential linking of training behaviors with race per-
formances [33].
Figure 1. Enrollment and Categorization of Participants by Sex. BMI—body mass index. HM—half-
marathon. M/UM—marathon/ultra-marathon. 10 km—10 kilometers.
Training behaviors of active female and male endurance running participants were
described by the following items linked to sex: running motivations (health, leisure, sport
performance); preferred time of day and preferred season for running activity (indoor vs.
outdoor); training duration for the main event; external training resource followed (none,
professional, other); participation in other sports activities to balance running (summer,
Figure 1. Enrollment and Categorization of Participants by Sex. BMI—body mass index. HM—half-
marathon. M/UM—marathon/ultra-marathon. 10 km—10 kilometers.
Training behaviors of active female and male endurance running participants were
described by the following items linked to sex: running motivations (health, leisure, sport
performance); preferred time of day and preferred season for running activity (indoor vs.
outdoor); training duration for the main event; external training resource followed (none,
professional, other); participation in other sports activities to balance running (summer,
winter); periodized running training (weekly frequency of training, daily and weekly
mileages and durations of training (km, hours) related to phase and stage of training,
respectively).
Int. J. Environ. Res. Public Health 2022, 19, 13238
4 of 14
The statistical software R version 3.6.2 Core Team 2019 (R Foundation for Statistical
Computing, Vienna, Austria) performed all statistical analyses. Exploratory analysis
was performed by descriptive statistics (median and interquartile range (IQR); mean and
standard deviation (SD)).
Significant differences in running activity (training behaviors) between race distance
and sex were calculated by using a non-parametric ANOVA. Chi-square test (χ2; nominal
scale) examined the association between variables, and the Wilcoxon test and/or Kruskal–
Wallis test (ordinal and metric scale) approximated F distributions with ordinary least
squares.
Differences in weekly and daily training with the mileages of female and male runners
are displayed as box plots.
The level of statistical significance was set at p ≤ 0.05.
3. Results
The survey was completed by 317 endurance runners. A total of 72 participants
were excluded from the initial sample due to satisfying the exclusion criteria (i.e., three
participants with a BMI ≥ 30) or disagreement with the inclusion criteria. After data
clearance, a total of 141 (58%) females and 104 (42%) males (n = 245) comprised the final
sample with a combined age of 39 (IQR 17) years, a body weight of 65 kg (IQR 14.2),
and a BMI of 21.7 kg/m2 (IQR 3.5) from several countries, including Austria (n = 44),
Germany (n = 177), Switzerland (n = 13), and some other countries (n = 11: Belgium, Brazil,
Canada, Italy, Luxemburg, Netherlands, Poland, Spain, United Kingdom). There were
154 NURMI-Runners (48% female) and 91 runners (74% female) competing at the 10 km
distance.
The female participants were significantly shorter (p < 0.001) and had a lower body
weight (p < 0.001) and BMI (p < 0.001) than the males. Regarding the highest academic
qualification achieved, participants reported a high school diploma or equivalent (53%
female), an A Levels or equivalent (58% female), a university degree or higher (59% female),
and some provided no answer (64% female). A significant difference (p = 0.044) was
identified between the sexes in terms of exercise focus, where females were largely focused
on leisure (57% female vs. 50% male), with a smaller proportion of females being sport
performance-oriented (30% female vs. 44% male). Participants reported their marital status
as single (32% of females vs. 20% of males), married or living with a partner (61% of females
vs. 75% of males), or divorced/separated (7% of females vs. 5% of males). Characteristics
of participants are further presented in Table 1, and more details on the characteristics of
participants are provided in Part B of the sequenced paper [33].
Table 1. Characteristics of Endurance Runners Displayed by Sex.
Total
Female
Male
Statistics
100% (245)
58% (141)
42% (104)
Age (years)
39 (IQR 17)
37 (IQR 16)
43 (IQR 18)
F(1, 243) = 7.03
p = 0.009 †
Body Weight (kg)
65 (IQR 14.2)
59.5 (IQR 10.9)
73 (IQR 11.9)
F(1, 243) = 191.23
p < 0.001 ‡
Height (m)
1.7 (IQR 0.1)
1.7 (IQR 0.1)
1.8 (IQR 0.1)
F(1, 243) = 228.04
p < 0.001 ‡
BMI (kg/m2)
21.7 (IQR 3.5)
20.9 (IQR 3.01)
22.8 (IQR 3.16)
F(1, 243) = 28.72
p < 0.001 ‡
Int. J. Environ. Res. Public Health 2022, 19, 13238
5 of 14
Table 1. Cont.
Total
Female
Male
Statistics
Academic
Qualification
No qualification
<1% (1)
<1% (1)
/
χ2(4) = 1.96
p = 0.744
High school diploma/
Technical qualification/
GCSE or equivalent
34% (83)
31% (44)
38% (39)
A Levels or equivalent
22% (53)
22% (31)
21% (22)
University degree/
Graduate degree
34% (83)
35% (49)
33% (34)
No answer
10% (25)
11% (16)
9% (9)
Country of
Residence
Austria
18% (44)
11% (16)
27% (28)
χ2(3) = 11.32
p = 0.010 *
Germany
72% (177)
79% (112)
62% (65)
Switzerland
5% (13)
4% (6)
7% (7)
Other
4% (11)
5% (7)
4% (4)
Exercise Focus
Health
Leisure
Sport performance
9% (23)
54% (133)
36% (89)
12% (17)
57% (81)
30% (43)
6% (6)
50% (52)
44% (46)
χ2(2) = 6.24
p = 0.044 *
Racing Distance
HM
M/UM
10 km
36% (89)
27% (65)
37% (91)
35% (49)
18% (25)
48% (67)
38% (40)
38% (40)
23% (24)
χ2(2) = 19.55
p < 0.001 ‡
Initial Running
Motivation
Health
Leisure
44% (108)
56% (137)
46% (65)
54% (76)
41% (43)
59% (61)
χ2(1) = 0.55
p = 0.459
Current Running
Motivation
Health
Leisure
Sport performance
19% (47)
46% (113)
35% (85)
21% (30)
48% (67)
31% (44)
16% (17)
44% (46)
39% (41)
χ2(2) = 2.06
p = 0.356
Note. *, †, or ‡ denote statistical significance at the levels p < 0.05, p < 0.01, or p < 0.001, respectively. Results are
presented as percentage (%), total numbers, and median (IQR). χ2 statistic calculated by Pearson’s Chi-squared
test and F statistic calculated by Kruskal–Wallis test. HM—half-marathon. M/UM—marathon/ultra-marathon.
10 km—10 kilometers.
No significant differences were found between the sexes for the initial running motiva-
tion (p = 0.459) or the current running motivation (p = 0.356). Seasonal running preferences
did not differ between females and males, including the preferred indoor (p = 0.312) or
outdoor running season (p = 0.727) and the ideal time of day for running, whether indoor
(p = 0.419) or outdoor (p = 0.592). From the total sample, most participants preferred out-
door running (n = 145; 57% female) in the springtime and most often during the morning
(n = 75; 57% female) but had no indoor running preference for a season (n = 178; 54%
female) or time of day (n = 168; 54% female).
Table 2 displays the general training behaviors for recreation runners for females
and males. A significant difference was identified (p = 0.037) in which males were more
likely to train alone (84% male vs. 71% female), whereas females were more likely to train
under the direction of a professional (24% female vs. 12% male) or another resource (11%
female vs. 4% male). Men were significantly more likely to participate in fell/trail running
(p = 0.021) and ski-touring (p = 0.029); no additional sex-based participation differences
were identified in other sports to balance running: cycling, swimming, rambling, triathlon,
skiing, cross-country skiing, or snowboarding. No significant differences were observed
between training duration for the main event and sex (p = 0.833) shown in Figure 2.
Int. J. Environ. Res. Public Health 2022, 19, 13238
6 of 14
Table 2. General Training Behaviors, including Total Duration, Resource, and Other Sports by Sex.
Total
Female
Male
Statistics
100% (245)
58% (141)
42% (104)
Training Duration for Main Event
1–2 months
3–4 months
4–6 months
7–8 months
9–10 months
>12 months
20% (46)
52% (122)
21% (48)
4% (9)
2% (5)
2% (4)
17% (22)
55% (73)
21% (28)
4% (5)
2% (3)
2% (2)
24% (24)
49% (49)
20% (20)
4% (4)
2% (2)
2% (2)
χ2(2) = 2.12; p = 0.833
Training Resource for Running Events
Alone and independently
76% (179)
71% (94)
84% (85)
χ2(2) = 6.57; p = 0.037 *
Under the direction of a professional
15% (36)
24% (18)
12% (12)
Other
8% (19)
11% (15)
4% (4)
Other Sports Activities to Balance for Running
Summer
Cycling
53% (130)
51% (72)
56% (58)
χ2(1) = 0.57; p = 0.451
Fell/Trail running
31% (75)
25% (35)
39% (40)
χ2(1) = 5.32; p = 0.021 *
Swimming
31% (75)
29% (41)
33% (34)
χ2(1) = 0.39; p = 0.535
Rambling
31% (75)
34% (47)
27% (28)
χ2(1) = 1.13; p = 0.287
Triathlon
19% (46)
17% (24)
21% (22)
χ2(1) = 0.69; p = 0.407
Winter
Skiing
14% (34)
13% (18)
16% (16)
χ2(1) = 0.35; p = 0.552
Cross country skiing
11% (26)
10% (14)
12% (12)
χ2(1) = 0.17; p = 0.681
Snowboarding
7% (16)
7% (10)
6% (6)
χ2(1) = 0.17; p = 0.682
Ski-touring
4% (9)
1% (2)
7% (7)
χ2(1) = 4.79; p = 0.029 *
Note. * denotes statistical significance at the level p < 0.05. Results are presented as percentage (%) and total
numbers. χ2 statistic calculated by Pearson’s Chi-squared test and F statistic calculated by Kruskal–Wallis test.
Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
6 of
Table 2. General Training Behaviors, including Total Duration, Resource, and Other Sports by S
Total
Female
Male
Statistics
100% (245)
58% (141)
42% (104)
Training Duration for Main Event
1–2 months
3–4 months
4–6 months
7–8 months
9–10 months
>12 months
20% (46)
52% (122)
21% (48)
4% (9)
2% (5)
2% (4)
17% (22)
55% (73)
21% (28)
4% (5)
2% (3)
2% (2)
24% (24)
49% (49)
20% (20)
4% (4)
2% (2)
2% (2)
χ2(2) = 2.12; p = 0.833
Training Resource for Running Events
Alone and independently
76% (179)
71% (94)
84% (85)
χ2(2) = 6.57; p = 0.037 *
Under the direction of a professional
15% (36)
24% (18)
12% (12)
Other
8% (19)
11% (15)
4% (4)
Other Sports Activities to Balance for Running
Summer
Cycling
53% (130)
51% (72)
56% (58)
χ2(1) = 0.57; p = 0.451
Fell/Trail running
31% (75)
25% (35)
39% (40)
χ2(1) = 5.32; p = 0.021 *
Swimming
31% (75)
29% (41)
33% (34)
χ2(1) = 0.39; p = 0.535
Rambling
31% (75)
34% (47)
27% (28)
χ2(1) = 1.13; p = 0.287
Triathlon
19% (46)
17% (24)
21% (22)
χ2(1) = 0.69; p = 0.407
Winter
Skiing
14% (34)
13% (18)
16% (16)
χ2(1) = 0.35; p = 0.552
Cross country skiing
11% (26)
10% (14)
12% (12)
χ2(1) = 0.17; p = 0.681
Snowboarding
7% (16)
7% (10)
6% (6)
χ2(1) = 0.17; p = 0.682
Ski-touring
4% (9)
1% (2)
7% (7)
χ2(1) = 4.79; p = 0.029 *
Note. * denotes statistical significance at the level p < 0.05. Results are presented as percentage
and total numbers. χ2 statistic calculated by Pearson’s Chi-squared test and F statistic calculated
Kruskal–Wallis test.
Figure 2. Sex-based differences in the prevalence of training duration (displayed in six categori
for main events. Data are presented by percentage.
Table 3 displays the periodized training phases by sex, including the regenerati
stage and transitional period (Phase A), the main preparatory period (Phase B), and t
main competition period (Phase C) based on the weekly training frequencies and t
weekly and daily training mileages and durations. No significant differences were o
served for weekly training frequency and sex, regardless of the phase (p > 0.05). Prepa
tory Stage 4 showed the greatest average difference in weekly mileage (+11.4 km/week
Figure 2. Sex-based differences in the prevalence of training duration (displayed in six categories) for
main events. Data are presented by percentage.
Table 3 displays the periodized training phases by sex, including the regeneration stage
and transitional period (Phase A), the main preparatory period (Phase B), and the main
competition period (Phase C) based on the weekly training frequencies and the weekly and
daily training mileages and durations. No significant differences were observed for weekly
training frequency and sex, regardless of the phase (p > 0.05). Preparatory Stage 4 showed
the greatest average difference in weekly mileage (+11.4 km/week for males; p = 0.030) and
duration (+1.71 h/day for males; p = 0.032). Significant differences were found between the
sexes in terms of daily training mileage and daily training duration for every phase and
stage (p < 0.05), with a highly significant difference identified in stage 1 (phase B) for daily
training mileage (+2.94 km/day for males; p < 0.001) and duration (+0.14 h/day for males;
Int. J. Environ. Res. Public Health 2022, 19, 13238
7 of 14
p < 0.001). Figures 3–5 depict the training mileages and weekly training frequency by sex
based on the training periodization phases.
Table 3. Periodization Training Behavior, including Frequency, Mileages, and Durations Displayed
by Sex.
Total
Female
Male
Statistics
100% (245)
58% (141)
42% (104)
Phase A—Regeneration Stage and Transitional Period
Weekly training frequency
Weekly training mileage (km)
Weekly training duration (hours)
3 (IQR 1)
22.4 ± 18.9
1.21 ± 1.03
2 (IQR 1)
20.6 ± 19
1.12 ± 1.03
3 (IQR 1)
24.6 ± 18.7
1.33 ± 1.02
F(1, 220) = 0.29; p = 0.590
F(1, 220) = 3.77; p = 0.053
F(1, 220) = 3.58; p = 0.060
Daily training mileage (km)
Daily training duration (hours)
7.06 ± 5.86
0.26 ± 0.21
6.24 ± 5.33
0.23 ± 0.19
8.12 ± 6.36
0.30 ± 0.23
F(1, 220) = 7.26; p = 0.008 †
F(1, 220) = 6.28; p = 0.013 *
Phase B—Main Preparatory Period
Preparatory Stage 1 (general basic training, mainly at low intensity)
Weekly training frequency
Weekly training mileage (km)
Weekly training duration (hours)
3 (IQR 2)
30.2 ± 24.6
4.63 ± 3.76
3 (IQR 2)
27.2 ± 23.1
4.16 ± 3.54
3 (IQR 2)
34.1 ± 25.9
5.22 ± 3.96
F(1, 220) = 0.01; p = 0.931
F(1, 220) = 4.40; p = 0.037 *
F(1, 220) = 4.67; p = 0.032 *
Daily training mileage (km)
Daily training duration (hours)
8.74 ± 6.61
0.4 ± 0.3
7.46 ± 6.13
0.34 ± 0.28
10.4 ± 6.86
0.48 ± 0.31
F(1, 220) = 14.93; p < 0.001 ‡
F(1, 220) = 14.70; p < 0.001 ‡
Preparatory Stage 2 (specific basic training, build-up training, low-to-moderate intensity)
Weekly training frequency
Weekly training mileage (km)
Weekly training duration (hours)
3 (IQR 2)
33.5 ± 27.5
4.82 ± 3.96
3 (IQR 2)
31 ± 26
4.46 ± 3.74
3 (IQR 3)
36.7 ± 29.2
5.28 ± 4.2
F(1, 220) = 0.02; p = 0.902
F(1, 220) = 2.15; p = 0.144
F(1, 220) = 2.21; p = 0.138
Daily training mileage (km)
9.42 ± 7.29
0.41 ± 0.32
8.39 ± 7.09
0.36 ± 0.31
10.8 ± 7.37
0.47 ± 0.32
F(1, 220) = 7.91; p = 0.005 †
F(1, 220) = 7.85; p = 0.006 †
Daily training duration (hours)
Preparatory Stage 3 (competition training, intervals, pace, moderate-to-high intensity)
Weekly training frequency
4 (IQR 2)
37.1 ± 31.1
5.65 ± 4.73
4 (IQR 1)
34.2 ± 29.5
5.2 ± 4.48
4 (IQR 3)
40.9 ± 32.9
6.22 ± 5
F(1, 220) = 0.04; p = 0.850
F(1, 220) = 2.41; p = 0.122
F(1, 220) = 2.57; p = 0.110
Weekly training mileage (km)
Weekly training duration (hours)
Daily training mileage (km)
9.98 ± 7.86
0.41 ± 0.32
9.01 ± 7.78
0.37 ± 0.32
11.24 ± 7.82
0.47 ± 0.32
F(1, 220) = 8.37; p = 0.004 †
F(1, 220) = 8.20; p = 0.005 †
Daily training duration (hours)
Preparatory Stage 4 (race-specific training, test competition, moderate-to-high intensity)
Weekly training frequency
4 (IQR 2)
39.5 ± 35.8
5.95 ± 5.39
4 (IQR 1)
34.6 ± 30.6
5.2 ± 4.6
4 (IQR 2)
46 ± 40.9
6.91 ± 6.15
F(1, 220) = 1.49; p = 0.223
F(1, 220) = 4.74; p = 0.030 *
F(1, 220) = 4.65; p = 0.032 *
Weekly training mileage (km)
Weekly training duration (hours)
Daily training mileage (km)
10.7 ± 8.31
0.5 ± 0.39
9.47 ± 7.93
0.45 ± 0.37
12.29 ± 8.56
0.58 ± 0.4
F(1, 220) = 10.13; p = 0.002 †
F(1, 220) = 9.58; p = 0.002 †
Daily training duration (hours)
Phase C—Competition Period (incl. tapering and interim race stages)
Weekly training frequency
3 (IQR 2)
32.2 ± 27.7
4.41 ± 3.8
3 (IQR 2)
28.8 (24.4)
3.95 (3.33)
4 (IQR 2)
36.6 (31.1)
5.01 (4.26)
F(1, 220) = 3.30; p = 0.071
F(1, 220) = 2.94; p = 0.088
F(1, 220) = 2.48; p = 0.117
Weekly training mileage (km)
Weekly training duration (hours)
Daily training mileage (km)
9.35 ± 8.7
0.41 ± 0.37
8.46 ± 8.33
0.37 ± 0.36
10.5 ± 9.07
0.45 ± 0.39
F(1, 220) = 5.21; p = 0.023 *
F(1, 220) = 4.47; p = 0.036 *
Daily training duration (hours)
Note. *, †, or ‡ denote statistical significance at the levels p < 0.05, p < 0.01, or p < 0.001, respectively. Results are
presented as median (IQR) and mean (SD). F statistic calculated by Kruskal–Wallis test. km—kilometer.
Int. J. Environ. Res. Public Health 2022, 19, 13238
8 of 14
Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
8 of 15
Note. *, †, or ‡ denote statistical significance at the levels p < 0.05, p < 0.01, or p < 0.001, respectively.
Results are presented as median (IQR) and mean (SD). F statistic calculated by Kruskal–Wallis test.
km—kilometer.
Figure 3. Box plots displaying sex differences in weekly training frequencies in different training
periodization phases and stages.
Figure 4. Box plots displaying sex differences in weekly training mileages (km) in different training
periodization phases and stages.
Figure 3. Box plots displaying sex differences in weekly training frequencies in different training
periodization phases and stages.
Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
8 of 15
Note. *, †, or ‡ denote statistical significance at the levels p < 0.05, p < 0.01, or p < 0.001, respectively.
Results are presented as median (IQR) and mean (SD). F statistic calculated by Kruskal–Wallis test.
km—kilometer.
Figure 3. Box plots displaying sex differences in weekly training frequencies in different training
periodization phases and stages.
Figure 4. Box plots displaying sex differences in weekly training mileages (km) in different training
periodization phases and stages.
Figure 4. Box plots displaying sex differences in weekly training mileages (km) in different training
periodization phases and stages.
Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
9 of 15
Figure 5. Box plots displaying sex differences in daily training unit (km/unit) in different training
periodization phases and stages.
4. Discussion
The objective of this study was to investigate sex differences in training behavior
among endurance runners; the present investigation was conducted to assess the relation-
ships between training behaviors of female and male recreational runners from 10 km up
U
d
h
f
d
f
l
h d
l
b d
h
d
Figure 5. Box plots displaying sex differences in daily training unit (km/unit) in different training
periodization phases and stages.
Int. J. Environ. Res. Public Health 2022, 19, 13238
9 of 14
4. Discussion
The objective of this study was to investigate sex differences in training behavior
among endurance runners; the present investigation was conducted to assess the relation-
ships between training behaviors of female and male recreational runners from 10 km up
to UM distances. The main findings were (1) female runners had a lower body weight and
height and thus a lower BMI than the males; (2) a significant sex difference was found in
main event racing distance; (3) significantly more females reported following an external
training resource but no sex difference was found in the training duration for the main event;
(4) weekly training frequencies were similar for females and males across periodization
phases (including all stages of Phase B); (5) various discrepancies were identified between
the sexes for weekly mileages and durations depending on each phase/stage; (6) males
were significantly more active at all phases/stages considering daily mileage and duration;
(7) males had a greater exercise focus on sport performance but no significant differences
were found between sexes for the initial or current running motivations. While similarities
exist between the training behaviors of female and male endurance runners [8–15], the
results of the present investigation uphold the hypothesis that there are critical training
differences between the sexes of recreational endurance runners of various distances (HM,
M/UM, 10 km).
Based on the previous literature concerning female and male runner anthropometric
differences [12], the present findings are consistent, indicating that females are generally
smaller than their male counterparts. Moreover, the participants’ anthropometrics (body
weight, height, BMI), especially of the females, highlight the general knowledge of recre-
ational runners being a slim, fit, and healthy population [2–4,28,34]. All the while, only 3 of
the initial participants (<0.01% of the sample) reported having obesity and were excluded
from this study due to the required WHO-based BMI criteria [31,32]; thus, the typical
healthy runner training and race preparation lifestyle is reflected by the participants’ BMI
within this study [2–4,28,34].
A significant difference was found in this study in the proportions of race distance
subgroups between the sexes, with the largest proportion of females being 10 km runners
and a remarkably larger proportion of males being M or UM runners. Until recent years,
female event participation in very long distances (HM, M, or UM) has been trailing that of
males [35], and this result may be a reflection that this difference between the sexes has not
been completely mitigated. Considering that the males were more heavily proportioned
among the longer-distance racing subgroups (HM, M/UM), and that the previous litera-
ture has shown considerable variation in runner training behaviors based on this factor
alone [15], it could be expected that there are significant critical sex differences in training
behavior of endurance runners.
Furthermore, it was found that significantly more females were training under the
direction of a professional (whether a sport scientist, doctor of sports medicine, or trainer)
or followed another resource rather than training alone and independently, which was
more common among the males. This finding could be related to a generally higher level
of health consciousness among females, while professional support is well-known to be
beneficial for the health of runners and especially injury prevention [3,4,36]. In addition,
while participation was similar in most other sports activities to balance running (cycling,
swimming, rambling, triathlon, skiing, cross country skiing, and snowboarding), a sex
difference for fell/trail running and ski-touring was detected. Regarding the finding of
significantly more male participation in fell/trail running, a previous study found the
opposite [37]; however, that report only analyzed runners of a specific trail race, and the
current sample was more general [25]. Ski-touring, on the other hand, predominantly takes
place in the backcountry with advanced-level terrain and increased avalanche risk; thus,
one possible explanation for more male ski-touring participation may be a higher risk-
seeking tendency among males [38]. Furthermore, no sex difference in the training duration
for the main event was found, which could be considered the overarching periodization
scheme [22]. This finding may be due to the fact that participating in an endurance running
Int. J. Environ. Res. Public Health 2022, 19, 13238
10 of 14
event requires a minimal preparatory period to safely complete each event, which is likely
dependent upon the event race distance (e.g., 10 km, HM, M, or UM) [15,39].
Regarding the breakdown of the training periodization into three distinct phases
(Phase A: Regeneration Stage and Transitional Period; Phase B: Preparatory Period; Phase
C: Main Competition Period), no sex difference in the weekly frequencies of training was
detected. Therefore, regardless of sex, participants trained regularly throughout each week
and training phase, possibly due to the fact that regular exercise is well-accepted as being
healthy and provides a plethora of benefits to each individual [34,40]. Within Phase A,
no sex differences in weekly training mileage or duration were found, which is likely
due to this particular phase having a main focus of recovery to avoid overloading as it
comes directly after participating in the previous main event [41]. Therefore, it would
be expected that the participants’ characteristics are somewhat unrelated to training in
this phase, as the major goal of this phase is proper regeneration and a smooth transition
to the main preparatory period [41]. However, significant differences between the sexes
regarding their daily training mileages and daily training durations in Phase A were found,
suggesting that males run a greater distance per day and also spend a greater amount
of time running per day than females during this period. This finding is possibly due to
the greater professional advice sought by females, as more resting is highly advisable for
achieving maximal recovery during this period, which may be a concern of overuse injury
or burnout for males who are seeking less professional support for training [41,42].
Within Phase B, there are four preparatory stages with distinct training characteristics.
Weekly training mileages and durations were similar for females and males during Prepara-
tory Stages 2 and 3, whereas significant sex differences were found for Stages 1 and 4.
Considering the training similarities between the sexes in Stage 2 (build-up training, includ-
ing specific basic training, and at a low-to-moderate intensity) and Stage 3 (intervals, pace,
specific competition training, and moderate-to-high intensity), it appears that the training
run distance and duration are unrelated to sex. Therefore, these particular types of weekly
exercises may be integral to the training plans of endurance runners regardless of sex,
which is consistent with previous reports [8,11]. For Stage 1 (general basic training at a low
intensity mainly) and Stage 4 (test competition, race-specific training, and moderate-to-high
intensity); however, the discrepancies between the sexes for weekly mileages and durations
are likely related to the previously mentioned result of more females racing in the shorter
distance (10 km) and more males racing in longer distances (M/UM) [15]. Furthermore,
and in connection, both the daily training mileages and daily training durations showed
significant differences between the sexes across all four stages of Phase B. Considering that
runners of longer distances have been shown to run further mileage while training [15],
these Phase B mileage differences between the sexes likely arise from the result that the
males race over longer distances. Regarding Stage 1, the general basic training at a low in-
tensity is dominant in the long-distance runner’s training plan (M/UM) [15,43,44], and the
present results show that males run remarkably more during this period with an additional
2.94 km per day (+0.14 h). Previous research has found performance and health benefits to
sustained loads of low-intensity training, including left ventricular hypertrophy and the
consequential enhancements to stroke volume, maximal volume of oxygen consumption,
and thus lower resting heart rate [45]. Indeed, moderate-to-high intensity training at Stage
4 would require extensively longer bouts of mileage and duration for the longer-distance
runner to meet their adaptive training thresholds for building mitochondrial density and
capillarization [44,46]. Previous research has not found any major link between sex and
volume of oxygen consumption training adaptations [46]. For Stage 4, the specific training
types (including test competition and race-specific training) are less likely to be related
to beneficial physical adaptations but are rather mental preparations, which are highly
important for the longer-distance runner, as very long distances (M/UM) require extensive
durations of focus [47,48]. Therefore, the training durations (weekly or daily) would appear
to be dependent upon the mileages (weekly or daily), as increases in the distance load
require more time to complete with a specified running intensity [15].
Int. J. Environ. Res. Public Health 2022, 19, 13238
11 of 14
Results for Phase C, which was considered the main competition period and included
tapering and interim race stage/s, show that there was no sex difference for weekly
training mileage or duration, which may be due to this phase including tapering, and the
consequential drastic reduction in weekly training volume for both females and males
compared to the prior phases [49]. However, considering that this phase also included
interim race stage/s, which are likely related to the main event race distance, there may
be a connection to the significant sex differences identified for daily training mileage and
duration [15].
Exercise motivations may underlie much of the participatory behavior in endurance
running [3,50]. The present results are in line with previous findings, that the male exercise
focus is more sport performance-oriented than females and that females are more focused
on health [50]. Regarding the finding of leisure as the female’s main exercise focus, this
result is somewhat inconsistent with previous literature [50], which suggests that females
are primarily health-focused. However, no significant sex difference was identified in terms
of the participants’ current running motivations, which could partly explain why there
were also some training similarities between the sexes within the current results. Therefore,
it is possible that a small fraction of participants considered themselves to be athletes of
another main sport instead of running.
Causal ascriptions have been suggested as the foundation for developing a theory
of motivation and emotion [51]. Therefore, it appears achievement-related projections,
which are plausibly socio-bounded, have a major influence on the exercise focus differences
among females and males [50,51]. In our sample, 96% of participants were from Austria,
Germany, and Switzerland, suggesting a rather homogenous group in terms of social
culture [52]. Those German-speaking countries are reported to have advanced economies
even compared to other western nations, and the results indicate no sex difference in
academic qualification, including a predominantly white-caucasian population with high-
income levels, high quality of life, and a high life expectancy [52,53].
Comparable to other studies that follow a cross-sectional design, the presented results
include some limitations that should be addressed when interpreting the findings. The
sample size was relatively small, and considering the sex-based approach of the present
investigation, there was an unequal distribution of sexes per se (58% females), and also
within the race distances, including more females (48% vs. 23% of males) as 10 km runners.
However, most of the participants were racing at the HM distance, marathon, or ultra-
marathon distance. As the results are based on self-reporting survey methods, over-
and under-reporting of answers are possible based on the sociological expectations of the
participants’ cultures; however, control questions (e.g., race distance) were used to minimize
this effect. Lastly, most participants were from Germany, Austria, and Switzerland, which
may limit the interpretation of some of our findings to Western and European running
cultures.
While the present investigation includes some limitations, the results have the potential
to add light to this specific gap in the current literature on sex differences in training
behaviors of 10 km up to ultra distance recreational runners. Upon careful consideration,
the findings may be particularly beneficial for athletic trainers, physical therapists, coaches,
team physicians, exercise physiologists, and endurance runner athletes to refine the vital
understanding of planning and applying an optimal, health-based training regimen for
successfully running races. Furthermore, the underrepresentation of females in previous
athletic samples studied regarding targeted sex-specific approaches and personalization
to training and performance requirements is evident [5,6,22]. In addition, future studies
should consider investigating sex differences of endurance runners by controlling for main
event race distance associated with runner motives.
5. Conclusions
In summary, this is the first study aiming to investigate sex differences in training
and race preparation behaviors of recreational runners of various distances, including HM,
Int. J. Environ. Res. Public Health 2022, 19, 13238
12 of 14
M/UM, and 10 km participants. The results indicate that: female runners are more likely
to train with an external resource; male runners train at a higher volume (greater daily
mileages and durations at every periodization phase; greater or null weekly mileages and
durations at each phase); females exercise with more leisure focus; and males concentrate
more on sports performance. It can be concluded that sex differences in training behaviors,
which may originate from physiological differences and social expectations, can be related
to the distances runners prefer to race and their motivations for running and racing. The
results of this study provide a wide overview of the fundamental training behaviors of
female and male recreational endurance runners of various distances (HM, M, UM, 10 km)
that may be remarkably supportive for carefully designing and implementing a thorough
training plan for endurance runners themselves or by their coaches, exercise physiologists,
athletic trainers, physical therapists, sport scientists, or sports medicine doctors.
Author Contributions: K.W. conceptualized and designed the study and the questionnaires together
with B.K. and C.L. K.W. conducted data analysis and D.T. and M.M. provided statistical expertise.
D.T., K.W., T.R. and B.K. drafted the manuscript. T.R., M.M., C.L. and K.W. critically reviewed it. G.W.
provided technical support and aided in data acquisition and management. All authors have read
and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: The study protocol is available online via https://springerplus.
springeropen.com/articles/10.1186/s40064-016-2126-4 (accessed on 17 August 2022) and was ap-
proved by the ethics board of St. Gallen, Switzerland on May 6, 2015 (EKSG 14/145). The study
was conducted in accordance with the ethical standards of the institutional review board, medical
professional codex, and with the 1964 Helsinki declaration and its later amendments as of 1996, the
Data Security Laws, and good clinical practice guidelines. Study participation was voluntary and
could be canceled at any time without the provision of reasons or negative consequences.
Informed Consent Statement: Informed consent was obtained from all individual participants
included in the study considering the data collected, used, and analyzed exclusively and only in the
context of the NURMI Study for scientific publication.
Data Availability Statement: The data sets generated during and/or analyzed during the current
study are not publicly available but may be made available upon reasonable request. Subjects will
receive a brief summary of the results of the NURMI Study if desired.
Acknowledgments: There are no professional relationships with companies or manufacturers who
will benefit from the results of the present study. Moreover, this research did not receive any specific
grant or funding from funding agencies in the public, commercial, or non-profit sectors.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Hespanhol Junior, L.C.; Pillay, J.D.; Van Mechelen, W.; Verhagen, E. Meta-Analyses of the Effects of Habitual Running on Indices
of Health in Physically Inactive Adults. Sports Med. 2015, 45, 1455–1468. [CrossRef] [PubMed]
2.
Lee, D.-C.; Pate, R.R.; Lavie, C.J.; Sui, X.; Church, T.S.; Blair, S.N. Leisure-Time Running Reduces All-Cause and Cardiovascular
Mortality Risk. J. Am. Coll. Cardiol. 2014, 64, 472–481. [CrossRef]
3.
Wirnitzer, K.; Boldt, P.; Wirnitzer, G.; Leitzmann, C.; Tanous, D.; Motevalli, M.; Rosemann, T.; Knechtle, B. Health status of
recreational runners over 10-km up to ultra-marathon distance based on data of the NURMI Study Step 2. Sci. Rep. 2022, 12, 1–16.
[CrossRef] [PubMed]
4.
Boldt, P.; Knechtle, B.; Nikolaidis, P.; Lechleitner, C.; Wirnitzer, G.; Leitzmann, C.; Wirnitzer, K. Sex Differences in the Health
Status of Endurance Runners: Results From the NURMI Study (Step 2). J. Strength Cond. Res. 2018, 33, 1929–1940. [CrossRef]
5.
Burrows, M.; Bird, S. The Physiology of the Highly Trained Female Endurance Runner. Sports Med. 2000, 30, 281–300. [CrossRef]
[PubMed]
6.
Barnekow-Bergkvist, M.; Hedberg, G.; Janlert, U.; Jansson, E. Physical activity pattern in men and women at the ages of 16 and 34
and development of physical activity from adolescence to adulthood. Scand. J. Med. Sci. Sports 2007, 6, 359–370. [CrossRef]
7.
Wirnitzer, K.; Motevalli, M.; Tanous, D.R.; Gregori, M.; Wirnitzer, G.; Leitzmann, C.; Rosemann, T.; Knechtle, B. Sex differences
in supplement intake in recreational endurance runners—Results from the NURMI Study (Step 2). Nutrients 2021, 13, 2776.
Available online: https://www.mdpi.com/2072-6643/13/8/2776/htm (accessed on 17 August 2022). [CrossRef] [PubMed]
Int. J. Environ. Res. Public Health 2022, 19, 13238
13 of 14
8.
Helgerud, J.; Ingjer, F. Sex differences in performance-matched marathon runners. Euro. J. Appl. Physiol. Occupat. Physiol. 1990, 61,
433–439. [CrossRef] [PubMed]
9.
Knechtle, B.; Knechtle, P.; Roseman, T.; Senn, O. Sex Differences in Association of Race Performance, Skin-Fold Thicknesses, and
Training Variables for Recreational Half-Marathon Runners. Percept. Mot. Ski. 2010, 111, 653–668. [CrossRef] [PubMed]
10.
Hewson, D.J.; Hopkins, W.G. Specificity of training and its relation to the performance of distance runners. Int. J. Sports Med.
1996, 17, 199–204. [CrossRef] [PubMed]
11.
Thuany, M.; Souza, R.; Hill, L.; Mesquita, J.; Rosemann, T.; Knechtle, B.; Pereira, S.; Gomes, T. Discriminant Analysis of
Anthropometric and Training Variables among Runners of Different Competitive Levels. Int. J. Environ. Res. Public Health 2021,
18, 4248. [CrossRef] [PubMed]
12.
Friedrich, M.; Rüst, C.A.; Rosemann, T.; Knechtle, P.; Barandun, U.; Lepers, R.; Knechtle, B. A Comparison of Anthropometric and
Training Characteristics between Female and Male Half-Marathoners and the Relationship to Race Time. As. J. Sports Med. 2013,
5, 10–20. [CrossRef]
13.
Billat, V.L.; Demarle, A.; Slawinski, J.; Paiva, M.; Koralsztein, J.P. Physical and training characteristics of top-class marathon
runners. Med. Sci. Sports Exerc. 2001, 33, 2089–2097. [CrossRef] [PubMed]
14.
Billat, V.; Lepretre, P.-M.; Heugas, A.-M.; Laurence, M.-H.; Salim, D.; Koralsztein, J.P. Training and Bioenergetic Characteristics in
Elite Male and Female Kenyan Runners. Med. Sci. Sports Exerc. 2003, 35, 297–304. [CrossRef]
15.
Knechtle, B.; Tanous, D.R.; Wirnitzer, G.; Leitzmann, C.; Rosemann, T.; Scheer, V.; Wirnitzer, K. Training and Racing Behavior of
Recreational Runners by Race Distance—Results from the NURMI Study (Step 1). Front. Physiol. 2021, 12, 620404. [CrossRef]
[PubMed]
16.
Masters, K.S.; Ogles, B.M.; Jolton, J.A. The Development of an Instrument to Measure Motivation for Marathon Running: The
Motivations of Marathoners Scales (MOMS). Res. Q. Exerc. Sport 1992, 64, 134–143. [CrossRef] [PubMed]
17.
Wa´skiewicz, Z.; Nikolaidis, P.T.; Gerasimuk, D.; Borysiuk, Z.; Rosemann, T.; Knechtle, B. What Motivates Successful Marathon
Runners? The Role of Sex, Age, Education, and Training Experience in Polish Runners. Front. Psychol. 2019, 10, 1671. [CrossRef]
[PubMed]
18.
Wirnitzer, K.; Motevalli, M.; Tanous, D.; Wirnitzer, G.; Leitzmann, C.; Wagner, K.-H.; Rosemann, T.; Knechtle, B. Training and
Racing Behaviors of Omnivorous, Vegetarian, and Vegan Endurance Runners—Results from the NURMI Study (Step 1). Nutrients
2021, 13, 3521. [CrossRef] [PubMed]
19.
Lynch, S.L.; Hoch, A.Z. The female runner: Gender specifics. Clin. Sports Med. 2010, 29, 477–498. [CrossRef]
20.
Hands, B.; Parker, H.; Larkin, D.; Cantell, M.; Rose, E. Male and female differences in health benefits derived from physical
activity: Implications for exercise prescription. J. Womens Health Issues Care. 2016, 5. [CrossRef]
21.
Nuzzo, J.L. Sex Difference in Participation in Muscle-Strengthening Activities. J. Lifestyle Med. 2020, 10, 110–115. [CrossRef]
[PubMed]
22.
Kenneally, M.; Casado, A.; Santos-Concejero, J. The Effect of Periodization and Training Intensity Distribution on Middle- and
Long-Distance Running Performance: A Systematic Review. Int. J. Sports Physiol. Perform. 2018, 13, 1114–1121. [CrossRef]
[PubMed]
23.
Boullosa, D.; Esteve-Lanano, J.; Casado, A.; Peyre-Tartaruga, L.A.; da Rosa, R.G.; Coso, J.D. Factors affecting training and physical
performance in recreation endurance runners. Sports 2020, 8, 35. [CrossRef] [PubMed]
24.
Legaz Arrese, A.; Izquierdo, D.M.; Galindo, J.R.S. Physiological Measures Associated with Marathon Running Performance in
High-Level Male and Female Homogeneous Groups. Endoscopy 2005, 27, 289–295. [CrossRef] [PubMed]
25.
Wirnitzer, K.; Seyfart, T.; Leitzmann, C.; Keller, M.; Wirnitzer, G.; Lechleitner, C.; Rüst, C.A.; Rosemann, T.; Knechtle, B.
Prevalence in running events and running performance of endurance runners following a vegetarian or vegan diet compared to
non-vegetarian endurance runners: The NURMI Study. SpringerPlus 2016, 5, 458. [CrossRef]
26.
Wirnitzer, K.; Motevalli, M.; Tanous, D.; Gregori, M.; Wirnitzer, G.; Leitzmann, C.; Hill, L.; Rosemann, T.; Knechtle, B. Supplement
intake in half-marathon, (ultra-)marathon and 10-km runners—results from the NURMI study (Step 2). J. Int. Soc. Sports Nutr.
2021, 18, 1–12. [CrossRef]
27.
Boldt, P.; Knechtle, B.; Nikolaidis, P.; Lechleitner, C.; Wirnitzer, G.; Leitzmann, C.; Rosemann, T.; Wirnitzer, K. Quality of life of
female and male vegetarian and vegan endurance runners compared to omnivores—results from the NURMI study (step 2). J.
Int. Soc. Sports Nutr. 2018, 15, 33. [CrossRef]
28.
Wirnitzer, K.; Boldt, P.; Lechleitner, C.; Wirnitzer, G.; Leitzmann, C.; Rosemann, T.; Knechtle, B. Health Status of Female and Male
Vegetarian and Vegan Endurance Runners Compared to Omnivores—Results from the NURMI Study (Step 2). Nutrients 2018, 11,
29. [CrossRef]
29.
Wirnitzer, K.; Motevalli, M.; Tanous, D.R.; Gregori, T.; Wirnitzer, G.; Leitzmann, C.; Hill, L.; Rosemann, T.; Knechtle, B. Supplement
intake in recreational vegan, vegetarian, and omnivorous endurance runners—Results from the NURMI Study (Step 2). Nutrients
2021, 13, 2741. Available online: https://www.mdpi.com/2072-6643/13/8/2741/htm (accessed on 17 August 2022). [CrossRef]
30.
Motevalli, M.; Wagner, K.-H.; Leitzmann, C.; Tanous, D.; Wirnitzer, G.; Knechtle, B.; Wirnitzer, K. Female Endurance Runners
Have a Healthier Diet than Males—Results from the NURMI Study (Step 2). Nutrients 2022, 14, 2590. [CrossRef]
31.
World Health Organization (WHO). Body Mass Index—BMI. Available online: https://www.euro.who.int/en/health-topics/
disease-prevention/nutrition/a-healthy-lifestyle/body-mass-index-bmi (accessed on 17 August 2022).
Int. J. Environ. Res. Public Health 2022, 19, 13238
14 of 14
32.
Word Health Organization (WHO). Noncommunicable Diseases: Risk Factors. 2010. Available online: http://www.who.int/
gho/ncd/risk_factors/bmi_text/en/ (accessed on 11 May 2018).
33.
Motevalli, M.; Tanous, D.; Wirnitzer, G.; Leitzmann, G.; Tanous, D.; Montevalli, M.; Rosemann, T.; Knechtle, B. Sex differences in
racing history of recreational 10 km to ultra runners (Part B)—Results from the NURMI Study (Step 2). IJERPH 2022. accepted for
publication/under production.
34.
Wirnitzer, K.C. Vegan Diet in Sports and Exercise—Health Benefits and Advantages to Athletes and Physically Active People: A
Narrative Review. Int. J. Sports Exerc. Med. 2020, 6, 165. [CrossRef]
35.
Scheerder, J.; Breedveld, K.; Borgers, J. Running across Europe: The Rise and Size of One of the Largest Sport Markets; Palgrave
Macmillan: Hampshire, UK, 2015.
36.
Lauersen, J.B.; Bertelsen, D.M.; Andersen, L.B. The effectiveness of exercise interventions to prevent sports injuries: A systematic
review and meta-analysis of randomised controlled trials. Br. J. Sports Med. 2014, 48, 871–877. [CrossRef]
37.
Navalta, J.W.; Montes, J.; Tanner, E.A.; Bodell, N.G.; Young, J.C. Sex and Age Differences in Trail Half Marathon Running. Int. J.
Exerc. Sci. 2018, 11, 281–289.
38.
Stoddart, M.C. Constructing masculinized sportscapes: Skiing, gender and nature in British Columbia, Canada. Int. Rev. Sociol.
Sport 2010, 46, 108–124. [CrossRef]
39.
van Dyck, D.; Cardon, G.; de Bourdeaudhuij, I.; de Ridder, L.; Willem, A. Who participates in running events? Socio-demographic
characteristics, psychosocial factors and barriers as correlates of non-participation–a pilot study in Belgium. Int. J. Environ. Res.
Public Health 2017, 14, 1315. [CrossRef]
40.
Bangsbo, J.; Blackwell, J.; Boraxbekk, C.-J.; Caserotti, P.; Dela, F.; Evans, A.; Jespersen, A.P.; Gliemann, L.; Kramer, A.; Lundbye-
Jensen, J.; et al. Copenhagen Consensus statement 2019: Physical activity and ageing. Br. J. Sports Med. 2019, 53, 856–858.
[CrossRef]
41.
Hausswirth, C.; Mujika, I. Recovery for Performance in Sport; Human Kinetics: Champaign, IL, USA, 2013.
42.
Van der Worp, M.P.; ten Haaf, D.S.M.; van Cingel, R.; de WIjer, A.; der Sanden, M.W.G.N.; Staal, J.B. Injuries in runners; a
systematic review on risk factor and sex differences. PLoS ONE. 2015, 10, e0114937. [CrossRef]
43.
Knechtle, B. Ultramarathon Runners: Nature or Nurture? Int. J. Sports Physiol. Perform. 2012, 7, 310–312. [CrossRef]
44.
Rüst, C.A.; Knechtle, B.; Knechtle, P.; Rosemann, T. Similarities and differences in anthropometry and training between recreational
male 100-km ultra-marathoners and marathoners. J. Sports Sci. 2012, 30, 1249–1257. [CrossRef]
45.
Scribbans, T.D.; Vecsey, S.; Hankinson, P.B.; Foster, W.S.; Gurd, B.J. The Effect of Training Intensity on VO2max in Young Healthy
Adults: A Meta-Regression and Meta-Analysis. Int. J. Exerc. Sci. 2016, 9, 230–247. [PubMed]
46.
Bacon, A.P.; Carter, R.E.; Ogle, E.A.; Joyner, M.J. VO2max trainability and high intensity interval training in humans: A
me-ta-analysis. PLoS ONE 2013, 8, e73182. [CrossRef] [PubMed]
47.
Jaenes, J.C.; Wilczy´nska, D.; Alarcón, D.; Peñaloza, R.; Casado, A.; Trujillo, M. The Effectiveness of the Psychological Intervention
in Amateur Male Marathon Runners. Front. Psychol. 2021, 12, 605130. [CrossRef] [PubMed]
48.
Hamstra-Wright, K.L.; Coumbe-Lilley, J.E.; Kim, H.; McFarland, J.A.; Bliven, K.C.H. The Influence of Training and Mental Skills
Preparation on Injury Incidence and Performance in Marathon Runners. J. Strength Cond. Res. 2013, 27, 2828–2835. [CrossRef]
49.
Murach, K.A.; Bagley, J.R. Less Is More: The Physiological Basis for Tapering in Endurance, Strength, and Power Athletes. Sports
2015, 3, 209–218. [CrossRef]
50.
Craft, B.B.; Carroll, H.A.; Lustyk, M.K.B. Gender Differences in Exercise Habits and Quality of Life Reports: Assessing the
Moderating Effects of Reasons for Exercise. Int. J. Lib. Arts Soc. Sci. 2014, 2, 65–76.
51.
Weiner, B. An attributional theory of achievement motivation and emotion. Psychol. Rev. 1985, 92, 548–573. [CrossRef]
52.
DACH Countries—Statistics & Facts: STATISTA. 2021. Available online: https://www.statista.com/topics/4623/dach-countries/
#dossierKeyfigures (accessed on 15 June 2022).
53.
Wirnitzer, K.; Motevalli, M.; Tanous, D.; Wirnitzer, G.; Leitzmann, C.; Pichler, R.; Rosemann, T.; Knechtle, B. Who Is Running in the
D-A-CH Countries? An Epidemiological Approach of 2455 Omnivorous, Vegetarian, and Vegan Recreational Runners—Results
from the NURMI Study (Step 1). Nutrients 2022, 14, 677. [CrossRef]
| Sex Differences in Training Behaviors of 10 km to Ultra-Endurance Runners (Part A)-Results from the NURMI Study (Step 2). | 10-14-2022 | Tanous, Derrick,Motevalli, Mohamad,Wirnitzer, Gerold,Leitzmann, Claus,Rosemann, Thomas,Knechtle, Beat,Wirnitzer, Katharina | eng |
PMC8518541 | 2002 |
Scand J Med Sci Sports. 2021;31:2002–2009.
wileyonlinelibrary.com/journal/sms
Received: 26 April 2021 | Revised: 20 June 2021 | Accepted: 30 June 2021
DOI: 10.1111/sms.14016
O R I G I N A L A R T I C L E
The socio- economic impact of running- related injuries: A large
prospective cohort study
Tjerk S. O. Sleeswijk Visser1,2
| Marienke van Middelkoop3
| Tryntsje Fokkema2,4
|
Robert- Jan de Vos1
This is an open access article under the terms of the Creat ive Commo ns Attri bution License, which permits use, distribution and reproduction in any medium, provided the original
work is properly cited.
© 2021 The Authors. Scandinavian Journal of Medicine & Science In Sports published by John Wiley & Sons Ltd.
Trial registration number: NTR number: NL5843
1Department of Orthopedic Surgery
and Sports Medicine, Erasmus MC
University Medical Center, Rotterdam,
The Netherlands
2Leiden University Medical Center,
Leiden, The Netherlands
3Department of General Practice,
Erasmus MC University Medical Center,
Rotterdam, The Netherlands
4Department of General Practice and
Elderly Care Medicine, University
Medical Centre Groningen, University of
Groningen, Groningen, The Netherlands
Correspondence
Robert- Jan de Vos, Department of
Orthopedic Surgery and Sports Medicine,
Erasmus MC University Medical Centre,
Dr. Molewaterplein 40, 3015 GD
Rotterdam, The Netherlands.
Email: [email protected]
Funding information
The INSPIRE trial was funded by the
Netherlands Organization for Health
Research and Development (ZonMW),
grant number 536001001
Objectives: To evaluate the impact of running- related injuries (RRIs) on activities of
daily living (ADL), work, healthcare utilization, and estimated costs.
Design: Prospective cohort study with data from a randomized controlled trial.
Methods: Adult recreational runners who registered for a running event (distances
5 to 42 km) were included in this study. Minimum follow- up duration was 3 months
(preparation, event participation, and post- race period). Injuries were registered using
a standardized definition. Primary outcome measure was a standardized 5- item sur-
vey on limitations in ADL. The survey data were categorized to the number of injured
runners with complete/moderate/no limitations. This outcome was expressed as the
percentage of injured runners with any limitation (complete or moderate limitations
amalgamated). Secondary outcomes were work absenteeism, the number of health-
care visits per injured runner, and estimated direct medical and indirect costs per
participant and per RRI.
Results: 1929 runners (mean [SD] age 41 [12] years, 53% men) were included in
this study and 883 runners (46%) sustained a RRI during the course of the study.
Injured runners reported the highest limitations (% with any limitation) of RRIs dur-
ing the first week of injury on sports and leisure activities (70%) and transportation
activities (23%). 39% of the injured runners visited a healthcare professional. Work
absenteeism due to the RRI was reported in 5% of the injured runners. The total mean
estimated costs were €74 per RRI and €35 per participant.
Conclusions: Injured runners are mainly limited in their transportation activities and
during sports and leisure. While the estimated costs of RRIs are not high when ex-
pressed per participant, the absolute costs may be substantial due to the popularity of
running.
K E Y W O R D S
activities of daily living, epidemiology, running, work
| 2003
SLEESWIJK VISSER Et aL.
1 | INTRODUCTION
Physical activity has proven to be a cost- effective way to en-
hance overall health and reduce morbidity and mortality.1- 4
Running is an increasingly popular way to improve physical
and mental well- being.5- 7 In 2019, close to 2 million people
(11% of the Dutch population) performed weekly running ac-
tivities in the Netherlands.8
Musculoskeletal injuries are a prominent disadvantage of
running, with training errors being frequently suggested to
be a major cause of injury.9,10 Most running injuries are due
to overuse and are located at the knee, lower leg, ankle, and
foot.11,12 The incidence of running- related injuries (RRIs)
varies among different populations (eg, cross- country run-
ners, novice runners, and long- distance runners) of runners
and can be up to 85% in novice runners training for an event.9
Even though RRIs are frequent, not much is known about
the impact of these injuries on socio- economic outcome mea-
sures. The impact of RRIs on activities of daily living (ADL)
has, for example, never been described in literature.
Healthcare utilization, direct medical costs, and indirect
costs due to absenteeism from work are outcome measures to
estimate the impact of a disease.13 A few studies have reported
the economic burden of RRIs, which varies between €83 and
€174 per RRI and €13 and €105 per participant training for
an event.14- 16 These ranges are large and this may be due to
the fact that these results were based on small study samples
or only novice runners were included.14,15 This makes it dif-
ficult to extrapolate these findings to the general recreational
running population.16,17 Therefore, it is relevant to evaluate
the economic burden of RRIs in a large heterogeneous run-
ning population. Knowledge of the social impact, the specific
areas affected by RRIs, and the experienced pain and disabil-
ity could aid in the design of tailored treatment practices. The
magnitude of the economic burden of RRIs is important to
know, as it affects the urgency of RRIs in scientific agendas.
Therefore, the primary objective of this study is to assess
the impact of RRIs on activities of daily living in runners
training for an event. Secondary objectives are to evaluate the
experienced pain and the effect of RRIs on work absentee-
ism, healthcare utilization, and estimated direct and indirect
costs.
2 | METHODS
2.1 | Study design
The study was designed at the Erasmus MC University
Medical Centre (Rotterdam, the Netherlands) and was part
of a randomized control trial (The INSPIRE trial), which
evaluated the effect of an online prevention program on
the number of RRIs among recreational runners. A detailed
study protocol has been published elsewhere.18 The Medical
Ethics Committee of the Erasmus MC University Medical
Centre Rotterdam, the Netherlands, approved the study pro-
tocol (MEC 2016– 292). The trial was registered before com-
mencement (NTR number: NL5843).
For the randomized trial, patients in the intervention
group had access to an online injury prevention program,
whereas the control group did not receive this information.
There were no differences in injury proportion between both
groups, and therefore, we regarded this study population as a
large cohort. The results of this randomized controlled trial
have been published elsewhere.19
In the RCT and the current study, an RRI was defined
as an injury of the muscles, joints, tendons, and/or bones in
the lower back or lower extremities that was caused by run-
ning with at least one of the following criteria: (1) the injury
caused a reduction in running distance, speed, duration, or
frequency for at least 1 week; (2) the injury led to a visit to
a medical specialist and/or physiotherapist; and/or (3) med-
ication was necessary to reduce symptoms as a result of the
injury.
2.2 | Participants
Potentially eligible participants were runners of 18 years or
older who registered for one of 3 running events in 2017.
These running events included the LadiesRun Rotterdam
(5, 7.5 or 10 km), the NN Marathon Rotterdam (10.6 or
42.2 km), and the NN City Pier City The Hague (5, 10 or
21.1 km). If runners expressed their interest to participate
during online registration for the event, they were provided
with more information, and if still interested, they were as-
sessed for eligibility. Participants were included if they met
the inclusion criteria (18 years or older, registration at least
2 months before the running event, knowledge of the Dutch
language, and access to email). After providing digital in-
formed consent, participants could immediately complete the
baseline survey.
2.3 | Procedures
Patients were asked to complete an online survey (using the
secure application LimeSurvey) on 4 different time points; (i)
at baseline (≥2 months before the running event, (ii) 2 weeks
before the running event, (iii) 1 day after the running event,
and (iv) 1 month after the running event. At baseline, runners
were asked to complete questions on demographics (sex, age,
length, and weight), training characteristics during the past
year (running frequency, duration, and speed), and lifestyle
(smoking, alcohol use). The baseline survey also inquired
whether the runner had suffered an RRI in the past 12 months.
2004 |
SLEESWIJK VISSER Et aL.
The 3 follow- up survey consisted of questions about the cur-
rent state of previously reported RRIs, the occurrence of new
RRIs, the impact of new RRIs on ADL, and work absentee-
ism and health care utilization due to the RRI. Injured run-
ners were asked to specify injury location (back, buttock, hip,
groin, upper leg, knee, shin, calf, Achilles, ankle, foot, or toe)
and injury onset (gradual or acute).
2.4 | Outcome measures
2.4.1 | Primary outcome measure
Impact of RRIs on activities of daily living (ADL) was meas-
ured at all 3 follow- up time points (2 weeks before the run-
ning event, 1 day after, and 1 month after the running event),
using a 5- item survey. Only participants who sustained one
or multiple new RRIs were asked to complete this survey.
Injured runners completed this survey only once per follow-
up time point, independently of the number of RRIs they
sustained. This survey has not been validated but has been
used in previous studies on RRIs.20,21 The survey consists
of 5 questions on the following dimensions: (1) daily activi-
ties (eg, getting up, washing, getting dressed), (2) household
activities (eg, cleaning, vacuuming), (3) activities at work/
school, (4) transportation activities (eg, driving, cycling,
walking), and (5) sports and leisure activities. Each domain
consists of 3 response options: no limitations, moderate limi-
tations, and complete limitations. Injured runners were asked
to indicate their ability to perform activities of daily living in
the first week after the injury. For every follow- up time point,
injured runners completed this survey, resulting in an expres-
sion of the number (%) of injured runners with complete,
moderate, or no limitations per domain. The results of these
three separate follow- up time points were combined and ex-
pressed as the total number (%) of injured runners with com-
plete, moderate, or no limitations per domain. Results were
also expressed as the number of injured runners with any
limitation (complete or moderate limitations amalgamated).
We also compared the impact on ADL in the first week after
the injury between RRIs with an acute and gradual onset.
2.4.2 | Secondary outcome measures
Impact on ADL per RRI location
We compared the impact on ADL between different injury
locations by subdividing the RRIs in 5 clustered injury loca-
tions: (1) lower back, (2) buttock/hip/groin, (3) upper leg/
knee, (4) lower leg (shin/Achilles/ankle), and (5) foot/toe.19
If injured runners sustained more than 1 RRI, which origi-
nated from different clustered injury locations (eg, if a runner
sustained an RRI to the groin and an RRI to the ankle), they
were excluded from this part of the analysis. This is because
in these cases, it was not possible to adequately assess which
RRI specifically led to an impact on ADL. If injured runners
sustained multiple RRIs, but these injuries were all located in
the same clustered injury location, they were included in this
part of the analysis. For every follow- up time point, injured
runners completed this survey, resulting in an expression of
the number (%) of injured runners with any limitation (com-
plete or moderate limitations amalgamated). Injured runners
were asked to indicate the amount of pain (on a visual ana-
logue scale; VAS 0– 10) during rest and running in the week
preceding the completion of the survey. We also compared
the mean pain during rest and running between the clustered
injury locations and between acute injuries and gradual onset
RRIs.
Work absenteeism
Work absenteeism was assessed by the number of lost days at
work/school due to an RRI and was measured at all 3 follow-
up time points. Only injured runners were asked to complete
this part of the survey. Work absenteeism was expressed as
the mean number of days of absence from work per injured
runner.
Healthcare utilization
Injured runners were asked whether they had used health care
due to an RRI. Healthcare utilization was assessed by asking
the total number of healthcare visits per type of healthcare
provider. Healthcare utilization was expressed as the mean
healthcare consumption (number of visits) and mean medical
costs per injured runner, per type of healthcare provider.
Costs
The estimated costs were divided into 2 categories: costs from
healthcare utilization (direct costs) and costs as a result of ab-
senteeism from work (indirect costs). We established produc-
tivity costs per hour and the costs per treatment/visit based on
a guideline for economic evaluations in health care, published
by the Dutch Healthcare Authority.22,23 We determined the di-
rect costs by multiplying the total number of visits/treatments
with the estimated medical costs for those visits/treatments.
The specific costs used for the economic evaluation are pre-
sented in Appendix S1. Mean direct and indirect costs due to
an RRI were calculated per RRI and per participant (the mean
of all participants and not only injured runners).
2.5 | Statistical analysis
Presence of a normal distribution of data was assessed using
the Shapiro- Wilk test. Normally distributed data are presented
as mean with standard deviation (SD) and non- normally dis-
tributed data as median with interquartile range (IQR). We
| 2005
SLEESWIJK VISSER Et aL.
presented costs (in €) as mean with standard deviation (SD).
Differences in direct, indirect, and total costs between men
and women and between acute and gradual onset RRIs were
compared using a Mann- Whitney U test. Missing data (par-
ticipants who did not complete at least one follow- up sur-
vey) were excluded from the analyses for the study purposes
described in this manuscript. For the analyses, a p value of
<0.05 was considered statistically significant. We used SPSS
software (V.24.0.0.1; SPSS) for statistical analysis.
3 | RESULTS
3.1 | Participants
In total, 2378 participants were included in the randomized
control trial of whom 1929 participants (81%) completed at
least one follow- up survey and were included in this study.
The mean (SD) age was 4212 years with the majority being
male (53%). 883 (46%) participants reported at least one RRI
during the course of this study. Most injuries (61%) had an
acute origin. 714 of the 883 (81%) injured runners completed
the surveys for our primary and secondary outcome meas-
ures. The participant characteristics are displayed in Table 1.
3.2 | Primary outcome— Activities of
daily living
Injured runners reported the highest limitations (any limita-
tion) of RRIs during the first week of their injury on sports
and leisure activities (70%) and transportation activities
(23%). Lower frequencies of limitations were reported for
daily activities (10%), household activities (12%), and activi-
ties at work/school (9%).
Injured runners with acute onset RRIs reported higher
limitations (any limitation) of RRIs during the first week of
their injury compared to injured runners with gradual onset
RRIs on daily activities (11% vs. 6%), household activities
(16% vs. 7%), activities at work/school (12% % vs 5%), trans-
portation activities (25% vs. 18%), and sports and leisure ac-
tivities (75% vs 60%). Figure 1 shows the impact on ADL of
acute and gradual onset RRIs.
3.3 | Secondary outcomes
3.3.1 | Impact on ADL per RRI location
Injured runners with RRIs located at the lower back and
lower leg reported higher limitations (any limitation) of RRIs
during the first week of their injury compared to the over-
all average on household activities (42% and 13% vs. 12%),
activities at work/school (25% and 10% vs 9%), transporta-
tion activities (42% and 26% vs. 23%), and sports and leisure
activities (71% and 75% vs 67%).
The impact on ADL per clustered injury location is shown
in Figure 2. Mean (SD) pain (VAS 0– 10) score during rest
was higher in lower back (4.8 [2.8]) injuries compared to but-
tock/hip/groin (3.7 [2.4]), upper leg (3.5 [2.4]), lower leg (3.3
[2.9]), and foot injuries (3.2 [2.4]). Mean pain (VAS 0– 10)
score during running was lower in lower back injuries (5.1
[3.1]) compared to buttock/hip/groin (6.0 [2.7]), upper leg
(5.9 [2.7]), lower leg (6.0 [2.9]), and foot injuries (6.4 [2.7]).
Mean (SD) pain (VAS 0– 10) scores during rest and running
were similar for acute injuries (3.7 [2.4] and 5.7 [2.7]) and
gradual onset RRIs (3.4 [2.4]) and 6.0 [3.0]).
3.4 | Work absenteeism
Work absenteeism due to an RRI was reported in 5% of the
injured runners. Within this group of injured runners, the
mean (SD) number of days of absence from work due to an
RRI was 3.5 (3.5).
3.5 | Healthcare utilization
39% of the injured runners visited a healthcare profes-
sional and 8% initiated self- care. The mean (SD) number of
TABLE 1 Descriptive statistics of participants
Characteristics (n = 1929)
Mean (SD)
Personal characteristics
Age (years)
41.9 (12.1)
Sex (Male/Female)
1020/909
BMI (kg/m2)
23.6 (2.9)
Injury- related factors
Injury proportion
883 (45.8%)
Injury mechanism (acute/gradual onset); %
61/39
Previous RRI (preceding 12 months) n (%)
994 (51.5%)
Reported RRI at baseline (yes)
415 (21.5%)
Sports- related factors
Running duration (hours/week)
3.1 (3.7)
Running experience (years)
6.8 (8.1)
Lifestyle- related factors
Smoking (yes) (%)
76 (3.9%)
Alcohol use (glasses per week)
4.2 (4.8)
Days with >30 mins of physical activity (days/
week)
5.8 (2.0)
Note: Values are displayed in frequency means (standard deviation).
Abbreviations: BMI, body mass index; RRI, Running- related injury; SD,
standard deviation.
2006 |
SLEESWIJK VISSER Et aL.
healthcare visits was 1.4 (4.5) per injured runner. A visit to
a physiotherapist was reported by 32% of the injured run-
ners. 4% visited a general practitioner and 2% reported hav-
ing visited a medical specialist. 76% of the total number of
healthcare visits consisted of physiotherapist visits. Table 2
demonstrates the frequencies of healthcare visits per type of
healthcare provider.
3.6 | Estimated direct and indirect costs
The majority (82%) of the estimated total healthcare costs con-
sisted of physiotherapy treatments (Table 2). The estimated
total healthcare costs were €39 (SD 139) per RRI and €18 (SD
97) per participant, accounting for the entire study population
of participants who completed at least one follow- up survey
FIGURE 1
Impact on activities of
daily living (ADL) during first week of
injury (gradual vs. acute onset injuries).
Displayed values are percentages of any
(moderate and severe) limitation
0
10
20
30
40
50
60
70
80
Daily activites
Household activities
Activities at
work/school
Transportation
activities
Sports/Leisure
activities
Impact on activities of daily living (ADL) during the first
week of injury
Gradual
Acute
FIGURE 2
Impact on activities of
daily living (ADL) during first week of
injury (specified per injury location).
Displayed values are percentages of any
(moderate and severe) limitation
0
10
20
30
40
50
60
70
80
Daily activities
Household activities
Activities at
work/school
Transportation
activities
Sports and leisure
activities
Impact on activities of daily living (ADL) during first week
of injury.
Lower back
Buttock/hip/groin
Upper leg/knee
Lower leg
Foot/toe
Overall average
TABLE 2 Healthcare utilization and medical costs per injured runner, per type of healthcare provider (n = 714 injured runners)
Healthcare provider
Patients using health
care, no. (%)
Mean healthcare consumption
(% of all healthcare visits)
Mean (SD)
Medical costs
Primary care (visits)
General practitioner
26 (3.6%)
0.06 (8.6%)
€2.11 (13.0)
Physical therapist
231 (32.4%)
1.2 (76.2%)
€39.88 (141.7)
Othera
31 (4.3%)
0.1(10.2%)
€3.25 (19.4)
Secondary care (visits)
Medical specialist (eg, Sports medicine
physician/orthopedic surgeon)
15 (2.1%)
0.04 (5.0%)
€3.48 (27.7)
Total
1.4 (100)
€48.74 (154.6)
Abbreviation: SD, standard deviation.
aAnother healthcare provider (eg, masseur, osteopath, podiatrist, alternative healthcare provider).
| 2007
SLEESWIJK VISSER Et aL.
(N = 1929). The estimated costs due to absenteeism from work
were €35 (SD 267) per RRI and €16 (SD 183) per participant.
Total estimated direct and indirect costs were €74 (SD 329)
per RRI and €35 (SD 227) per participant (Table 3). Costs
from work absenteeism accounted for 48% of the total costs.
Estimated direct costs for acute and gradual onset RRIs
were €41 (SD 166) and €37 (SD 88), respectively. This
difference was not statistically significant (U = 96 950.5,
Z = −0.38, p = 0.71). The estimated indirect costs for acute
RRIs (mean €59) were significantly higher (U = 92 953.0,
Z = −4.2, p = 0.00) than for RRIs with a gradual onset (mean
€2). There was no significant difference in total costs for
acute RRIs (mean €100) compared to RRIs with a gradual
onset (mean €39) (U = 97 563.0, Z = −0.17, p = 0.87).
No statistically significant difference in estimated direct and
indirect costs was found between males and females (Table 3).
4 | DISCUSSION
This study showed that the largest impact of RRIs is on sports
and leisure (70%) and transportation activities (23%), while the
impact on other activities of daily living was relatively low.
The percentage of injured runners with any limitation in ADL
was higher in RRIs located at the lower back and lower leg
compared to the other clustered injury locations. Work absen-
teeism due to an RRI was reported in 5% of the injured run-
ners. The total mean number of healthcare visits was 1.4 per
injured runner, and the total mean estimated costs were €74
per RRI and €35 per participant. Acute injuries initially led to
more limitations in ADL and higher estimated total costs.
4.1 | Activities of daily living
Activities of daily living were mainly affected in the domains
sports and leisure activities and transportation activities. Still
approximately 1 in 10 injured runners experienced limita-
tions in daily and household activities or activities in work/
school. To better understand the impact of RRIs on activi-
ties of daily living, we evaluated this impact for both acute
and gradual onset RRIs and for different injury locations. We
found that acute onset RRIs and lower back and lower leg in-
juries in particular led more frequently to limitations in daily
life. This limiting effect only partly correlated with pain score
during rest, which was relatively high in lower back injuries.
The restricting effect of lower back pain on ADL has been
demonstrated in previous studies.24,25 For lower leg injuries
such as shin bone, Achilles tendon, and ankle joint injuries,
this has never been described using this approach. Healthcare
providers can take this into account when educating injured
runners with lower back or lower leg injuries about the poten-
tial consequences of their injury. This information can also
be used by healthcare policy makers in the design of tailored
management plans or preventive measures.
4.2 | Healthcare utilization, work
absenteeism, and costs
Healthcare utilization mainly consisted of physiotherapy
visits, which is in line with existing literature.13- 16 The total
costs of RRIs were estimated at €35 per participant and €74
per RRI. We compared these costs to other studies describing
the economic burden of RRIs in runners preparing for and
participating in a running event. Two studies estimated the
economic burden to be around €173 per RRI.14,15 In contrast
to our study, only small and selected (trailrunners and run-
ners participating in events ≤10 miles) running populations
were included. This could have led to a less accurate estima-
tion of the economic burden of RRIs in these studies and less
generalizability to the overall running population. A large
prospective cohort study estimated the cost of RRIs to be €13
per participant.16 However, this study had a follow- up of only
Overall
Mean (SD)
medical costs
Mean (SD) indirect costs
(absenteeism from paid
work)
Cost per RRI, total
(n = 898)
€74.29 (328.6)
€38.99 (138.5)
€35.29 (266.7)
Acute onset (n = 523)
€99.75 (420.6)
€40.74 (165.6)
€59.00 (346.6)
Gradual onset (n = 375)
€38.78 (100.0)
€36.56 (88.0)
€2.22 (32.1)
p Value
0.87
0.71
0.0
Cost per participant, total
(n = 1929)
€34.58 (227.7)
€18.15 (96.9)
€16.43 (182.8)
Males (n = 1020)
€33.98 (225.7)
€20.35 (120.0)
€13.63 (169.2)
Females (n = 909)
€35.25 (230.3)
€15.68 (59.3)
€19.57 (196.9)
p Value
0.48
0.45
0.49
Note: Mann- Whitney U test values comparing costs between acute and gradual onset RRIs and males and
females.
TABLE 3 Direct and indirect costs
per RRI (n = 901) and per participant
(n = 1929)
2008 |
SLEESWIJK VISSER Et aL.
6 weeks. This could explain why the estimated costs in the
current study are higher as long- standing injuries will have
had more impact on our study.16,26,27 Expressed per partici-
pant the total costs of RRIs appear to be relatively low. When
considering the popularity of running, which is practiced by
close to 2 million people in the Netherlands, the absolute
costs of RRIs may be substantial. This underlines the need
for optimized preventive measures.
Previous studies reported contradictory findings on the
difference in costs between males and females and between
acute and gradual onset RRIs.14- 16 The costs per RRI in the
current study were substantially higher for injuries with an
acute onset than injuries with a gradual beginning. This dif-
ference can be explained by the difference in costs from ab-
senteeism from work. The finding of higher indirect costs for
acute RRIs is in line with a large study on RRIs among nov-
ice runners but in contrast with other studies.14- 16 Overuse
injuries are supposed to have more impact over time, which
was demonstrated in several studies.16,26,27 If we would have
followed the runners for a longer time, costs of overuse inju-
ries may have increased. It could be hypothesized that RRIs
with an acute onset are accompanied with higher absenteeism
from work because the severity of symptoms is higher in the
initial phase of the injury, while the severity of injuries with
a gradual onset is spread out over time. This might lead to
less people being absent from work. We found that injured
runners with acute onset RRIs experience more limitations
on ADL in the first week of injury, which could support this
hypothesis. The difference between both injury groups could
also originate from the way we measured indirect costs in this
study, as we only included costs from absenteeism from work
and did not ask participants about a decrease in work pro-
ductivity. A decrease in work productivity could lead to sub-
stantial costs, which has been shown in several studies on the
impact of overuse injuries.28- 30 It could be that— while acute
onset injuries lead to higher absenteeism in the short term—
once back at work the work productivity in people who suf-
fered this type of injury is back to normal. Subsequently,
indirect costs could be similar in both injury groups if they
were measured over a longer period of time and measured
more accurately (including work productivity measures).
4.3 | Strengths and limitations
This is the first study to report the impact of RRIs on ADL.
We were able to show this impact for different injury loca-
tions. Furthermore, this prospective cohort study included a
large heterogeneous running population, which increases the
generalizability of these findings. There are also some limi-
tations to this study. We assessed our primary outcome by
using a non- validated survey, which could decrease the reli-
ability of the results. However, we used an online survey with
limited response options, which was completed in the same
way by all injured runners at all 3 follow- up time points. This
guaranteed the internal consistency of the survey, and it is
therefore likely that these results are reliable. Secondly, we
asked patients about their limitations in daily life during the
first week of injury retrospectively. This could have induced
recall bias and may have led to inaccuracy of the results on
this specific outcome measure.
4.4 | Recommendations for future research
Future research could focus on the impact of RRIs on qual-
ity of life, hereby using validated questionnaires (eg, the
EuroQol questionnaire [EQ- 5D]). This will provide more
information on the social impact of RRIs and the specific do-
mains which are affected. In addition, it would be interesting
to perform an economic evaluation of RRIs with the addi-
tion of work productivity and a longer follow- up period in
order to evaluate if this affects the direct and indirect costs
of overuse injuries. Next to this, it would be helpful if more
economic evaluations per type of sports are performed to be
able to adequately compare the costs of RRIs with different
sports.
5 | PERSPECTIVE
This study showed that runners suffering from an RRI are
mainly limited in their sports and leisure and transportation
activities and these limitations are particularly substantial in
lower back and lower leg injuries. The total costs for run-
ners training for an event were €74 per RRI and €35 per
participant. Even though the estimated costs of RRIs are
not high when expressed per participant, the absolute costs
may be substantial due to the popularity of running and be-
cause long- standing RRIs may further increase the costs with
longer follow- up time. Consequently, this study emphasizes
the need for optimized preventive measures.
CONFLICT OF INTEREST
The authors declare there is no conflict of interest.
DATA AVAILABILITY STATEMENT
The authors can confirm that all relevant data are included in
the article and/or its Supplementary Information Files.
ORCID
Tjerk S. O. Sleeswijk Visser
https://orcid.
org/0000-0002-4483-1936
Marienke van Middelkoop
https://orcid.
org/0000-0001-6926-0618
Tryntsje Fokkema
https://orcid.org/0000-0002-3767-2770
| 2009
SLEESWIJK VISSER Et aL.
Robert- Jan de Vos
https://orcid.
org/0000-0003-0372-0188
REFERENCES
1. Hatziandreu EI, Koplan JP, Weinstein MC, Caspersen CJ, Warner
KE. A cost- effectiveness analysis of exercise as a health promotion
activity. Am J Public Health. 1988;78(11):1417- 1421.
2. Frew EJ, Bhatti M, Win K, et al. Cost- effectiveness of a community-
based physical activity programme for adults (Be Active) in the
UK: an economic analysis within a natural experiment. Br J Sports
Med. 2014;48(3):207- 212.
3. Chakravarty EF, Hubert HB, Lingala VB, Fries JF. Reduced dis-
ability and mortality among aging runners: a 21- year longitudinal
study. Arch Intern Med. 2008;168(15):1638- 1646.
4. Samitz G, Egger M, Zwahlen M. Domains of physical activity and
all- cause mortality: systematic review and dose- response meta-
analysis of cohort studies. Int J Epidemiol. 2011;40(5):1382- 1400.
5. van der Worp MP, de Wijer A, van Cingel R, Verbeek AL, van der
Nijhuis Sanden MW, Staal JB. The 5- or 10- km Marikenloop Run:
a prospective study of the etiology of running- related injuries in
women. J Orthop Sports Phys Ther. 2016;46(6):462- 470.
6. Lee D- C, Brellenthin AG, Thompson PD, Sui X, Lee I- M, Lavie
CJ. Running as a key lifestyle medicine for longevity. Prog
Cardiovasc Dis. 2017;60(1):45- 55.
7. Stamatakis E, Chaudhury M. Temporal trends in adults' sports par-
ticipation patterns in England between 1997 and 2006: the Health
Survey for England. Br J Sports Med. 2008;42(11):901- 908.
8. Sportdeelname wekelijks. Het aandeel van de Nederlandse bevolk-
ing van 4 jaar en ouder dat één keer per week of vaker sport. https://
www.sport enbew egeni ncijf ers.nl/kerni ndica toren/ sport deeln ame-
wekel ijks. Centraal Bureau voor de Statistiek ism het Rijksinstituut
voor Volksgezondheid en Milieu. 2019. Accessed on 14- 12- 2020
9. Kluitenberg B, van Middelkoop M, Diercks R, van der Worp H.
What are the differences in injury proportions between different
populations of runners? A systematic review and meta- analysis.
Sports Med. 2015;45(8):1143- 1161.
10. Nielsen RO, Buist I, Sørensen H, Lind M, Rasmussen S. Training
errors and running related injuries: a systematic review. Int J
Sports Phys Ther. 2012;7(1):58- 75.
11. Walther M, Reuter I, Leonhard T, Engelhardt M. Verletzungen
und Überlastungsreaktionen im Laufsport. Der Orthopäde.
2005;34(5):399- 404.
12. Lopes AD, Hespanhol Júnior LC, Yeung SS, Costa LO. What are
the main running- related musculoskeletal injuries? A systematic
review. Sports Med. 2012;42(10):891- 905.
13. Smits DW, Huisstede B, Verhagen E, et al. Short- term absenteeism
and health care utilization due to lower extremity injuries among
novice runners: a prospective cohort study. Clin J Sport Med.
2016;26(6):502- 509.
14. Hespanhol Junior LC, van Mechelen W, Verhagen E. Health and
economic burden of running- related injuries in Dutch trailrun-
ners: a prospective cohort study. Sports Med. 2017;47(2):367- 377.
15. Hespanhol Junior LC, van Mechelen W, Postuma E, Verhagen E.
Health and economic burden of running- related injuries in runners
training for an event: a prospective cohort study. Scand J Med Sci
Sports. 2016;26(9):1091- 1099.
16. Hespanhol Junior LC, Huisstede BM, Smits DW, et al. The
NLstart2run study: Economic burden of running- related injuries
in novice runners participating in a novice running program. J Sci
Med Sport. 2016;19(10):800- 804.
17. Videbæk S, Bueno AM, Nielsen RO, Rasmussen S. Incidence of
running- related injuries per 1000 h of running in different types of
runners: a systematic review and meta- analysis. Sports Medicine.
2015;45(7):1017- 1026.
18. Anonymous. Details omitted for double- blind reviewing; 2017.
19. Anonymous. Details omitted for double- blind reviewing; 2019.
20. Kluitenberg B, van Middelkoop M, Diercks RL, et al. The
NLstart2run study: health effects of a running promotion program
in novice runners, design of a prospective cohort study. BMC
Public Health. 2013;13(1):685.
21. Van Middelkoop M, Kolkman J, Van Ochten J, Bierma- Zeinstra
SM, Koes B. Prevalence and incidence of lower extremity in-
juries in male marathon runners. Scand J Med Sci Sports.
2008;18(2):140- 144.
22. IJzerman MJ. Richtlijn voor het uitvoeren van economische eval-
uaties in de gezondheidszorg. Door Zorginstituut Nederland. In
samenwerking met het ministerie van VWS; 2016.
23. Hakkaart- van Roijen LVdLN, Bouwmans C, Kanters T,
Kostenhandleiding TSS. Methodologie van kostenonderzoek en
referentieprijzen voor economische evaluaties in de gezondhe-
idszorg. In opdracht van Zorginstituut Nederland. Geactualiseerde
versie; 2015.
24. Dutmer AL, Schiphorst Preuper HR, Soer R, et al. Personal and so-
cietal impact of low back pain: the Groningen spine cohort. Spine.
2019;44(24):E1443- E1451.
25. Froud R, Patterson S, Eldridge S, et al. A systematic review and
meta- synthesis of the impact of low back pain on people's lives.
BMC Musculoskelet Disord. 2014;15(1):50.
26. Clarsen B, Rønsen O, Myklebust G, Flørenes TW, Bahr R. The
Oslo Sports Trauma Research Center questionnaire on health prob-
lems: a new approach to prospective monitoring of illness and in-
jury in elite athletes. Br J Sports Med. 2014;48(9):754- 760.
27. Pluim BM, Loeffen FG, Clarsen B, Bahr R, Verhagen EA. A one-
season prospective study of injuries and illness in elite junior ten-
nis. Scand J Med Sci Sports. 2016;26(5):564- 571.
28. De Vries AJ, Koolhaas W, Zwerver J, et al. The impact of patellar
tendinopathy on sports and work performance in active athletes.
Res Sports Med. 2017;25(3):253- 265.
29. Van den Heuvel SG, IJmker S, Blatter BM, de Korte EM. Loss
of productivity due to neck/shoulder symptoms and hand/arm
symptoms: results from the PROMO- study. J Occup Rehabil.
2007;17(3):370- 382.
30. Martimo KP, Shiri R, Miranda H, Ketola R, Varonen H, Viikari-
Juntura E. Self- reported productivity loss among workers
with upper extremity disorders. Scand J Work Environ Health.
2009;35(4):301- 308.
SUPPORTING INFORMATION
Additional supporting information may be found online in
the Supporting Information section.
How to cite this article: Sleeswijk Visser TSO, van
Middelkoop M, Fokkema T, de Vos R- J. The socio-
economic impact of running- related injuries: A large
prospective cohort study. Scand J Med Sci Sports.
2021;31:2002– 2009. https://doi.org/10.1111/sms.14016
| The socio-economic impact of running-related injuries: A large prospective cohort study. | 07-11-2021 | Sleeswijk Visser, Tjerk S O,van Middelkoop, Marienke,Fokkema, Tryntsje,de Vos, Robert-Jan | eng |
PMC8997730 |
Citation: Wu, J.; Zhang, L.; Yang, H.;
Lu, C.; Jiang, L.; Chen, Y. The Effect of
Music Tempo on Fatigue Perception
at Different Exercise Intensities. Int. J.
Environ. Res. Public Health 2022, 19,
3869. https://doi.org/10.3390/
ijerph19073869
Academic Editors: Mark Reybrouck,
Piotr Podlipniak and David Welch
Received: 15 February 2022
Accepted: 22 March 2022
Published: 24 March 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright:
© 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed
under
the
terms
and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
International Journal of
Environmental Research
and Public Health
Article
The Effect of Music Tempo on Fatigue Perception at Different
Exercise Intensities
Jianfeng Wu 1
, Lingyan Zhang 2, Hongchun Yang 1, Chunfu Lu 1,*, Lu Jiang 2 and Yuyun Chen 2
1
Industrial Design and Research Institute, Zhejiang University of Technology, Hangzhou 310023, China;
[email protected] (J.W.); [email protected] (H.Y.)
2
School of Design and Architecture, Zhejiang University of Technology, Hangzhou 310023, China;
[email protected] (L.Z.); [email protected] (L.J.); [email protected] (Y.C.)
*
Correspondence: [email protected]
Abstract: Background: This study aimed to clarify the effect of music tempo on runners’ perception
of fatigue at different exercise intensities and while listening to music of different tempos through
running experiments. Methods: This study used a within-subject two-factor experimental design with
music tempo (fast music, slow music, no music) and exercise intensity (high intensity, low intensity)
as independent variables and the time to fatigue perception (TFP), the difference in heart rate (HR)
and the difference in the median frequency (MF) of surface electromyography (sEMG) signals as
observation indexes. Eighteen participants completed a total of 108 sets of running experiments.
Results: (1) The main effect of music tempo on the TFP was significant (p < 0.001). (2) The main effect
of exercise intensity on the TFP was significant (p < 0.001), and the main effect on the difference in HR
was significant (p < 0.001). (3) The interaction effect of music tempo and exercise intensity on the TFP
was significant (p < 0.05). Conclusions: Exercisers’ subjective perception of fatigue was affected by
music tempo and the interaction between music tempo and exercise intensity, and exercisers’ objective
fatigue perception was influenced mostly by exercise intensity. The findings of this study provide
guidance for runners’ choice of music at different intensities of exercise. Whether it is low-intensity
exercise or high-intensity exercise, listening to fast music while exercising can help runners perform
better mentally and physically during their runs.
Keywords: running; music tempo; exercise intensity; fatigue perception; heart rate; surface
electromyographic signals; median frequency
1. Introduction
Runners’ fatigue perception varies while running. Fatigue perception refers to the sub-
jective intensity of perception of tension, discomfort and fatigue during physical exercise [1].
Exercise fatigue usually manifests as soreness in muscles, an increased heart rate and de-
creased cognitive performance [2,3]. Although the continuous accumulation of fatigue will
affect people’s exercise intentions and performance [4] and excessive fatigue will lead to
physical injury [5,6], appropriate exercise fatigue can improve physical performance [7].
Therefore, to achieve a better fitness effect, runners usually adjust their exercise intensity
continuously according to their subjectively perceived state after perceiving exercise fatigue
during exercise [8]. Listening to music to reduce the sense of boredom and fatigue during
running and to enhance motivation has become a common behavior during running [9].
In terms of whether music interferes with the perception of fatigue in exercisers, there
is a consensus among academics that the effects of music can be explained from three
perspectives: emotional regulation, attention diversion and fatigue recovery. Emotion
plays an important role in regulating motivation in sports. Music can cause exercisers
to experience positive emotions during exercise [10], alleviate their perception of fatigue
during running and enhance the pleasure and sense of participation in running [11]. Using
parallel information processing theory, Rejeski points out that the bandwidth of human
Int. J. Environ. Res. Public Health 2022, 19, 3869. https://doi.org/10.3390/ijerph19073869
https://www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2022, 19, 3869
2 of 18
attention processing narrows during exercise. In the absence of the external stimuli of music,
runners may pay more attention to their running behaviors and feelings, thus increasing
their perception of fatigue, while in conditions with music, runners shift their attention
from unpleasant physical feelings to music, reducing their perception of fatigue and other
negative feelings [12]. It has also been pointed out that the tempo of music can influence
the internal movement tempo of the body [13], and music with a good tempo can induce
specific movement patterns [14] and enhance the arousal level of the exerciser, raising the
threshold of fatigue perception and speeding up recovery from exercise fatigue [15,16].
Further studies have shown that exercise fatigue perception is influenced by both
music tempo and exercise intensity. Different tempos stimulate different emotional states in
listeners, resulting in different effects on runners’ fatigue perception [17]. Exercise intensity
is another frequently discussed influencing factor. It has been argued that music can play
a role in reducing the perception of fatigue at different exercise intensities [18]. However,
some scholars believe that music only plays a regulatory role at specific intensities [19].
Subsequently, many scholars have conducted studies on music tempo and exercise in-
tensity, trying to find the optimal combination of exercise intensity and music tempo for
modulating fatigue perception. For example, Maddigan et al. concluded that fast music
could improve the performance of exercisers during high-intensity exercise and reduce
fatigue perception [20], while Yamamoto et al. found that neither fast music nor slow music
changed the average power output of exercisers during high-intensity exercise [11], i.e.,
different music tempos did not disperse individuals’ perception of fatigue. Some scholars
have also discussed how movement perception is influenced by the interaction of music
tempo and movement-related fatigue, but the key lies in the music tempo, the consistency
of exercisers’ preference and in the music and movement tempos being synchronous or
asynchronous. No study compared music tempo to runner fatigue perception at different
intensities of exercise. In addition, some scholars have discussed whether exercisers’ fatigue
perception is affected by the interaction between music tempo and exercise intensity, but
the studies focused on the consistency of music tempo with exercisers’ preference [21] or
the influence of music tempo being synchronous or asynchronous with exercise tempo [22],
and there is no study comparing the effect of music tempo on runners’ fatigue perception
at different exercise intensities for the time being.
The aforementioned studies provide references for the positive effects of music on
exercise fatigue perception, but there is still some space for exploration on how to choose
the appropriate tempo of music for different intensities of exercise. For example, how
does musical tempo affect fatigue perception of exercisers during different intensities of
exercise? Is there an interaction effect between music tempo and exercise intensity on
runners’ fatigue perception? All these problems need our attention, and the effects of music
tempo on fatigue perception at different exercise intensities still need to be further explored.
To clarify the effect of music tempo on runners’ fatigue perception at different exercise
intensities, this study conducted experiments on running and analyzed and discussed the
changes in runners’ fatigue perception at different exercise intensities and with different
music tempo conditions by evaluating runners’ fatigue perception. The results provide
guidance for individual fitness practitioners choosing music to listen to during exercise at
different intensities.
2. Methods
2.1. Subjects
We conducted an a priori analysis of the required sample size in the study using
G*power, with the presumption of the presence of a medium effect size of f = 0.25 [23],
a statistical test power = 0.8 and a significance level α = 0.05. The results of the analysis
indicated that a sample size of 18 would be sufficient to achieve a medium effect size
interaction effect. To obtain more generalizable research conclusions, the experimental
subjects were ordinary college students who did not have regular fitness habits and were
not guided by scientific theories or methodological knowledge of running fitness. They
Int. J. Environ. Res. Public Health 2022, 19, 3869
3 of 18
had no diseases of the muscular, skeletal, respiratory or cardiovascular systems. They were
not allergic to alcohol and had no cuts or scratches on the thigh muscles. All subjects had
no significant differences in physical indicators, were between 20 and 30 years old, 170 and
180 cm in height and 55 and 75 kg.
A total of 18 healthy males were recruited as experimental subjects after the above
rigorous screening. The basic information of the experimental subjects is shown in Table 1.
Participants were required to confirm compliance with the following requirements prior to
each experiment: (1) Rest at least 8 h the night before each experiment and refrain from
participating in any other physical activity other than the exercise tasks of this experiment
to prevent cardiopulmonary or muscle function injury or abnormalities. (2) Maintain a
previously habitual daily diet and ensure that the experiment is performed at least one hour
after the meal, and avoid eating, drinking alcohol or consuming excessive water within
one hour prior to the experiment.
Table 1. Basic information of the subjects (M ± SD).
Number of
Subjects
Age (Years)
Height (cm)
Weight (kg)
Resting Heart Rate
(Beats/min)
18
23.95 ± 1.49
173.78 ± 1.54
64.39 ± 4.55
78.97 ± 8.15
2.2. Experimental Design
The experiment was a two-factor within-subject experimental design with the inde-
pendent variables being music tempo (fast tempo, slow tempo, no music) and exercise
intensity (high intensity, low intensity). The two independent variables led to a total of six
experimental protocols. The dependent variables were the time to fatigue perception (TFP),
heart rate (HR) and the median frequency (MF) of surface EMG signals. For each of the
three music conditions, the experiment was completed for high- and low-intensity running.
To prevent interference between different groups of experiments, the interval between
each running experiment was 48 h for each subject, and the running experiments for each
subject were scheduled at the same time of day (from 9:00 a.m. to 11:00 a.m.) to control for
possible effects of diurnal patterns [24]. Fresh air circulation in the room was maintained
during the experiment, and the ambient temperature was kept at 23 ± 2 ◦C to reduce the
environmental load, promote normal body heat dissipation in the subjects during the exper-
iment and to avoid electrode shedding or short circuits caused by excessive sweating. Each
participant signed an informed consent form for the experiment before participating. The
study was conducted in accordance with the Declaration of Helsinki and approved by the
Ethics Committee of the Industrial Design Institute of Zhejiang University of Technology
(protocol code 0903/2021, date of approval 15 September 2021) (Supplementary Materials
Files S1 and S2).
2.3. Independent Variable
2.3.1. Music Tempo
Music can be divided into fast tempo music and slow tempo music according to the
speed of the tempo. The beat of fast tempo music is 150–160 bpm, and the beat is strong.
The beat of slow tempo music is 90–100 bpm, and such music is characterized by a narrow
range, soft sound and soothing changes [25]. In addition to the requirements of music
tempo, this study also followed the following principles of song selection: (1) The songs
chosen had a cheerful music style and obvious tempo and were able to induce positive
emotions in the subjects. (2) Music without any lyrics was selected to avoid the interference
of lyric content. (3) Songs with complex tunes were excluded. According to the principle
of music selection, 15 pieces of fast tempo music and 15 pieces of slow tempo music
were selected.
All subjects who participated in the experiment were invited to score the valence
and arousal effect of each piece of music, and the music with higher validity and arousal
Int. J. Environ. Res. Public Health 2022, 19, 3869
4 of 18
effect among the fast music and slow music was selected for inclusion in the final song list.
Finally, 15 pieces of music were obtained for the running experiment. Among them, 9 were
slow tempo pieces and 6 were fast tempo music pieces. Details of the selected music are
shown in Table 2.
Table 2. Details of music selection.
Music Tempo
Tracks
Duration/min
Beat/bpm
Slow music
Falcom Sound Team Jdk
4′23′′
90
Grass Harvest
3′14′′
96
The des Alizes
3′40′′
100
Harunouta
3′03′′
90
Sakurairo Contrail
2′28′′
90
Springtime Affair
2′49′′
95
Regrettably, You Know
2′04′′
96
A Tiny Sunshine
1′57′′
99
Sakura Residential Area
2′05′′
97
Fast music
Cigarette Daydreams
3′12′′
150
Hero
3′34′′
150
Shanghai Alice Magic Orchestra
3′36′′
152
Toy War
1′55′′
160
Where to Jun
3′57′′
160
Dream Land Days
3′17′′
155
2.3.2. Exercise Intensity
Exercise intensity refers to the amount of exercise an individual can perform per unit
of time and is a very important indicator in physical exercise and training. The American
College of Sports Medicine suggests that adults should exercise at an exercise intensity
of 50–85% of heart rate reserve (%HRR) to improve their cardiopulmonary function [26].
In view of the fact that this study included people who usually exercise less, the exercise
intensity during the experiment was within the recommended range of %HRR, with
50–60% HRR for low exercise intensity and 70–80% HRR for high exercise intensity.
To facilitate the experiment, the mean treadmill speed was adjusted according to
the %HRR interval to explore the relationship between %HRR and exercise intensity and
determine the treadmill speed to use for the running experiment [27]. Half of the subjects
(7 in total) were randomly selected to participate in the treadmill speed determination
experiment. Each subject was required to perform two separate running experiments
at 50–60% HRR and 70–80% HRR, respectively. To avoid the effect of fatigue, the two
running experiments were separated by more than 24 h. The experimental steps were
as follows: (1) the resting HR of the subjects was measured to determine the target HR
range; (2) the subjects wore an HR belt and then performed a 3 min warm-up exercise
on the treadmill; (3) an incremental running experiment was conducted, in which the
subjects started running at a speed of 5.5 km/h and the speed was increased by 0.5 km/h
every 2 min; (4) after each subject reached the upper limit of the target HR, the subject ran
continuously at that speed and stopped after 5 min. During the 5 min, the change in target
HR of the subjects was observed, and the HR of the subjects was kept at the upper limit of
the target HR range by increasing or decreasing the speed of the treadmill as appropriate.
The target HR was calculated by Equation (1):
Target HR = target intensity %HRR × (HRmax − resting HR) + resting HR
(1)
Based on the average speed of the treadmill while the subjects were in the target HR
range, the treadmill speed corresponding to low-intensity exercise (50–60% HRR) for this
experiment was 7 km/h, and the treadmill speed corresponding to high-intensity exercise
(70–80% HRR) was 9 km/h.
Int. J. Environ. Res. Public Health 2022, 19, 3869
5 of 18
2.3.3. Music Tempo with Exercise Intensity
The subjects performed constant-load running exercises on a home motorized tread-
mill in the general fitness mode at different intensities under the conditions of music or
no music. A total of 6 exercise regimens were used for subjects within the 3 × 2 group:
no music × low intensity, no music × high intensity, slow tempo × low intensity, slow
tempo × high intensity, fast tempo × low intensity, fast tempo × high intensity. To prevent
the interference of order effects, the Latin square design method was used to arrange the
experimental order in this experiment.
2.4. Dependent Variable
2.4.1. Time to Fatigue Perception
In this study, Borg’s scale for rating of perceived exertion (RPE) was used to measure
the subjective exertion of exercisers during running. Scores on the RPE scale ranged from
6 (no exertion at all) to 20 (exercise limit), as shown in Table 3. The RPE scale is a subjective
assessment of an individual’s perception of muscular exertion, physical tension, discomfort
or fatigue during exercise and reflects the individual’s perception of his or her fatigue
state [28]. When the RPE reaches 15, the exerciser shows shortness of breath and significant
muscle fatigue. Therefore, to ensure the safety of the experiment, RPE = 15 was chosen
as the index for the experiment. Running was stopped when the RPE value reported by
the subject reached 15, and the time from the start of running to the time when the RPE
reached 15, i.e., the time to fatigue perception, was recorded.
Table 3. Borg’s scale for rating of perceived exertion.
Score
Subjective
Exercise Intensity
Subjective
Exercise Fatigue
Score
Subjective
Exercise Intensity
Subjective
Exercise Fatigue
6
No exertion at all
Not hard at all
14
-
7
Extremely light
Extremely relaxed
15
Hard (heavy)
Tired
8
- 1
16
-
9
Very light
Very relaxed
17
Very hard
Very tired
10
-
18
-
11
Light
Relaxed
19
Extremely hard
Extremely tired
12
-
20
Maximal exertion
Trying one’s best
13
Somewhat hard
A little tired
1-represents the fatigue status between the two levels.
2.4.2. Instantaneous HR
HR can be used to objectively evaluate exercise fatigue and is the easiest indicator to
use to assess the intensity of current exercise and exercise fatigue [29]. Exercise fatigue
leads to a decrease in the HR regulation function, HR increases with fatigue and HR is
dynamic during exercise. In this study, the difference in HR before and after running was
used to characterize the degree of fatigue of exercisers. Two heart rate measurements were
taken in each set of experiments; the first time the subject’s resting HR was measured before
the run. The second time was measured during the running process, where the subject was
wearing a heart rate belt, and we collected the HR signal of the subject during the whole
running process. However, in the data processing, only the heart rate signal of the subject
5 s before the end of the run was selected and averaged as an indicator of the immediate
post-run HR.
2.4.3. Surface Electromyography Signal
Electromyography (EMG) signals are indicative of biological electrical signals gener-
ated by the contraction of human muscles. Surface EMG (sEMG) measures the compre-
hensive electrical effect of the conduction of human muscle electrical signals that can be
sensed on the skin surface [30]. The sEMG signal is commonly used for the evaluation of
neuromuscular function because of its real-time, sensitive and flexible characteristics. It
Int. J. Environ. Res. Public Health 2022, 19, 3869
6 of 18
has important practical and research value in the fields of sports science, clinical medicine,
ergonomics, etc. There have been many studies linking sEMG signals to fatigue [31]. The
sEMG uses the surface electrode bipolar conduction method, and the derived EMG signal
is increased by a signal amplifier and then enters a converter for signal conversion and
storage in a computer. It has been shown that the frequency domain index of the sEMG
signal is more sensitive to the muscle fatigue state of runners [32]. Frequency domain anal-
ysis mainly relies on fast Fourier transformation (FFT) to obtain the frequency spectrum
or power spectrum of the EMG signal to reflect the variation of the EMG signal in the
frequency dimension [33]. The median frequency (MF) is one of the most commonly used
indicators in the frequency domain analysis of surface EMG signals, and the MF value is
more stable for assessing the muscle fatigue state [34,35]. MF refers to the middle value of
muscle fiber discharge frequency during skeletal muscle contraction and high frequency
discharge is the main expression of excitation of fast muscle fibers, while slow muscle fibers
are dominated by low-frequency potential activity [36]. Therefore, the difference in static
MF before and after running was used as the index of local muscle fatigue in this study.
The formula for MF is shown in Equation (2):
MF = 1
2
Z ∞
0
PSD( f )d f
(2)
where PSD(f) is the myoelectric power spectral density.
Running mainly mobilizes the leg muscle groups; therefore, in this experiment, the
sEMG signals of the local muscles in the runners’ legs were collected, and the rectus
femoris (RF) and vastus medialis (VM), which display obvious changes in sEMG signals
and are very stable during movement changes, were examined [37]. The electrode positions
for the RF and VM are shown in Figure 1. Figure 1a shows the electrode positions for
the RF, Figure 1b shows the electrode positions for the VM. The black electrode is the
reference electrode.
2.4.3. Surface Electromyography Signal
Electromyography (EMG) signals are indicative of biological electrical signals gener-
ated by the contraction of human muscles. Surface EMG (sEMG) measures the compre-
hensive electrical effect of the conduction of human muscle electrical signals that can be
sensed on the skin surface [30]. The sEMG signal is commonly used for the evaluation of
neuromuscular function because of its real-time, sensitive and flexible characteristics. It
has important practical and research value in the fields of sports science, clinical medicine,
ergonomics, etc. There have been many studies linking sEMG signals to fatigue [31]. The
sEMG uses the surface electrode bipolar conduction method, and the derived EMG signal
is increased by a signal amplifier and then enters a converter for signal conversion and
storage in a computer. It has been shown that the frequency domain index of the sEMG
signal is more sensitive to the muscle fatigue state of runners [32]. Frequency domain
analysis mainly relies on fast Fourier transformation (FFT) to obtain the frequency spec-
trum or power spectrum of the EMG signal to reflect the variation of the EMG signal in
the frequency dimension [33]. The median frequency (MF) is one of the most commonly
used indicators in the frequency domain analysis of surface EMG signals, and the MF
value is more stable for assessing the muscle fatigue state [34,35]. MF refers to the middle
value of muscle fiber discharge frequency during skeletal muscle contraction and high
frequency discharge is the main expression of excitation of fast muscle fibers, while slow
muscle fibers are dominated by low-frequency potential activity [36]. Therefore, the dif-
ference in static MF before and after running was used as the index of local muscle fatigue
in this study. The formula for MF is shown in Equation (2):
PSD f df
=
0
( )
2
1
MF
(2)
where PSD(f) is the myoelectric power spectral density.
Running mainly mobilizes the leg muscle groups; therefore, in this experiment, the
sEMG signals of the local muscles in the runners’ legs were collected, and the rectus fem-
oris (RF) and vastus medialis (VM), which display obvious changes in sEMG signals and
are very stable during movement changes, were examined [37]. The electrode positions
for the RF and VM are shown in Figure 1. Figure 1a shows the electrode positions for the
RF, Figure 1b shows the electrode positions for the VM. The black electrode is the refer-
ence electrode.
(a)
(b)
Figure 1. Muscle and electrode location. (a) Shows the electrode positions for the RF, (b) shows the
electrode positions for the VM.
The RF and VM sEMG signals of the subject’s dominant leg were collected. The sub-
ject was first asked to perform thigh flexion and extension movements to find the target
muscle location. Once the target muscle was found, the location of the target muscle was
marked [38]. Subsequently, cotton dipped in alcohol was used to clean the skin of surface
dirt and remove sweat, sebum and other impurities on the skin surface. The electrodes
were pasted after the skin was dry, and surface hairs were removed if necessary [39]. The
purpose of taking the above measures was to reduce the impedance effect of the skin,
Figure 1. Muscle and electrode location. (a) Shows the electrode positions for the RF, (b) shows the
electrode positions for the VM.
The RF and VM sEMG signals of the subject’s dominant leg were collected. The
subject was first asked to perform thigh flexion and extension movements to find the target
muscle location. Once the target muscle was found, the location of the target muscle was
marked [38]. Subsequently, cotton dipped in alcohol was used to clean the skin of surface
dirt and remove sweat, sebum and other impurities on the skin surface. The electrodes
were pasted after the skin was dry, and surface hairs were removed if necessary [39]. The
purpose of taking the above measures was to reduce the impedance effect of the skin,
enhance the adhesion of the electrode patch to the skin and obtain the best recording effect.
Following the surface EMG for non-invasive assessment of muscles (SENIAM) guidelines,
bipolar sEMG electrodes were placed along the longitudinal midline of the muscle (in the
direction of the muscle fibers) on the muscle abdomen at the muscle–tendon junction, and
the point spacing of the electrode patch was 2–3 cm [40,41]. After placement of the sensor
and the reference electrode, a test was performed to determine whether the electrodes
were placed properly on the muscle and connected to the equipment so that a reliable
Int. J. Environ. Res. Public Health 2022, 19, 3869
7 of 18
sEMG signal could be recorded [42]. EMG signal acquisition software (Acknowledge4.2,
Biopac) was connected to the sensor, and then the subject was asked to perform flexion
and extension movements. It was then observed whether the corresponding EMG signal
in the interface of the acquisition software produced obvious changes. If the signal did
not show obvious changes, or the signal showed abnormal values, the muscle selection
position was further calibrated, and the electrode patch was checked to determine whether
the connection with skin was firm [38]. Testing continued until the signal was stable. The
above steps were followed before each acquisition of sEMG signals. The electrodes used
for testing were Ag/AgCl electrocardiographic electrodes. The EMG100c signal amplifier
was then secured to the subject’s lower leg with a strap.
2.5. Experimental Apparatus
The experimental apparatus included a treadmill, a pair of Bluetooth headphones, a set
of barbell pieces with different weights, an HR belt, an RPE scale and an MP150 telemetry
physical recorder and its accessories. The treadmill had a rated power of 1100 W and a belt
area of 1350 × 480 mm, which was used to provide running conditions with a constant load.
Airpod Bluetooth headphones were used to play music at a constant volume of 50% of
the maximum volume, or approximately 65 dB. Barbells were hung on the legs before and
after running to collect sEMG signals during static muscle exertion. A Polar HR monitor
(Polar Electro Oy, Kempele, Finland) with a sampling rate of 125 Hz was used to acquire
the real-time HR of the subjects during running. Participants wore the Polar heart rate
band on their chest, which synced with their iPhone via Bluetooth. The MP150 telemetry
physiological recorder (Biopac Inc., Goleta, CA, USA) was used to acquire sEMG signals
at a sampling rate of 2048 Hz, equipped with two EMG100c signal amplifiers and several
disposable ECG electrode patches (Shanghai Huyou Medical Electrode Co., Ltd., Shangai,
China). The electrodes were used to connect the target muscle to the EMG signal amplifier.
The bipolar electrode (Ag/AgCl) was attached with a 10 mm diameter gel medium for
reducing the impedance between the electrode and the skin. The size of the electrode
patches was trimmed to 3 cm before use so that they could meet the electrode patch 2–3 cm
spacing distance.
2.6. Experimental Procedure
The experimental procedure for this study is shown in Figure 2. Each subject was
required to follow the experimental procedure and complete a total of six sets of the
running experiment.
2.6.1. Acquisition of Resting HR
The HR of the subject was captured and recorded in real-time with an HR band. To
exclude the influence of the initial state among different subjects, the resting HR of the
subjects was collected before the experiment began. After the subjects sat in a comfortable
position for 5 min, their HR was collected for 1 min, and the average value was calculated
as the resting HR of the subjects [28].
2.6.2. Pre-Run Safety Instructions and Warmup
Before the formal experiment started, the subjects were introduced to the experiment
contents and procedures in detail and received instructions on the use of the treadmill,
correct running posture, a breathing adjustment method that could be used during running
and the use of the RPE scale to ensure the safety of the subjects and the smooth operation
of the experiment. All subjects underwent a 5-min warmup exercise before the experiment
to avoid the occurrence of muscle damage during the experiment.
Int. J. Environ. Res. Public Health 2022, 19, 3869
8 of 18
Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
8
Figure 2. Experiment flow chart.
2.6.1. Acquisition of Resting HR
The HR of the subject was captured and recorded in real-time with an HR ban
exclude the influence of the initial state among different subjects, the resting HR
subjects was collected before the experiment began. After the subjects sat in a comfo
position for 5 min, their HR was collected for 1 min, and the average value was calcu
as the resting HR of the subjects [28].
2.6.2. Pre-run Safety Instructions and Warmup
Before the formal experiment started, the subjects were introduced to the exper
contents and procedures in detail and received instructions on the use of the trea
correct running posture, a breathing adjustment method that could be used during
ning and the use of the RPE scale to ensure the safety of the subjects and the smoot
eration of the experiment. All subjects underwent a 5-min warmup exercise befor
experiment to avoid the occurrence of muscle damage during the experiment.
2.6.3. Acquisition of sEMG before Running
Static sEMG signals of the abovementioned muscles were collected before and
running to complement the instability of the sEMG signals during running [28]. A
barbell pieces of different weights was prepared for hanging on the dominant leg b
and after running to collect sEMG signals during static muscle exertion. Static sEM
nals were acquired as follows: the subject’s leg was subjected to a certain percenta
weight-bearing and knee extension so that the leg muscles were in a state of exertion
then data were recorded for the RF and VM with an MP150 telemetry physical rec
for 30 s [43]. In this case, the weight-bearing level was determined based on the
weight of each subject: weight-bearing level = body weight (kg) × 10% [44]. The EMG
Figure 2. Experiment flow chart.
2.6.3. Acquisition of sEMG before Running
Static sEMG signals of the abovementioned muscles were collected before and after
running to complement the instability of the sEMG signals during running [28]. A set of
barbell pieces of different weights was prepared for hanging on the dominant leg before
and after running to collect sEMG signals during static muscle exertion. Static sEMG
signals were acquired as follows: the subject’s leg was subjected to a certain percentage of
weight-bearing and knee extension so that the leg muscles were in a state of exertion, and
then data were recorded for the RF and VM with an MP150 telemetry physical recorder for
30 s [43]. In this case, the weight-bearing level was determined based on the body weight
of each subject: weight-bearing level = body weight (kg) × 10% [44]. The EMG amplitudes
of the RF and VM on the dominant side of the leg were recorded. To ensure signal stability
and data reliability, the middle 20 s of the 30-s static EMG signal were extracted, and the
20-s MF values were calculated with a 2048-point window.
The experimental procedure is shown in Figure 3. Figure 3a shows the electrode posi-
tioning for the target muscle and the actual situation of wearing and using the instrument;
Figure 3b shows the setup for measuring sEMG signals when loading and stretching the
knee; and Figure 3c shows the running experimental procedure.
2.6.4. Running Experiment and Data Collection
Each runner ran under low-intensity and high-intensity conditions, with no music,
fast music and slow music. The subjects started exercising at 7 km/h for the low-intensity
exercise condition and at 9 km/h for the high-intensity exercise condition. The subjects
were observed and asked about their current fatigue at 1-min intervals, and the RPE values
were recorded. The subjects stopped running when their subjective RPE reached 15. The
time taken for the RPE value to reach 15 was recorded, and the HR signal was collected for
5 s before the end of the run. Immediately after the subjects stopped running, the weighted
knee extension experiment was performed, and the sEMG signals of the RF and VM were
Int. J. Environ. Res. Public Health 2022, 19, 3869
9 of 18
collected from the subjects. Throughout the experiment, the experimenter observed the
subjects’ movement status and ensured their safety.
Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
9 of 18
The experimental procedure is shown in Figure 3. Figure 3a shows the electrode po-
sitioning for the target muscle and the actual situation of wearing and using the instru-
ment; Figure 3b shows the setup for measuring sEMG signals when loading and stretch-
ing the knee; and Figure 3c shows the running experimental procedure.
(a)
(b)
(c)
Figure 3. (a) The electrode position for the target muscles and the actual situation of wearing and
using the instrument; (b) the setup for measuring sEMG signals when loading and stretching the
knee; and (c) the running experimental procedure.
2.6.4. Running Experiment and Data Collection
Each runner ran under low-intensity and high-intensity conditions, with no music,
fast music and slow music. The subjects started exercising at 7 km/h for the low-intensity
exercise condition and at 9 km/h for the high-intensity exercise condition. The subjects
were observed and asked about their current fatigue at 1-min intervals, and the RPE val-
ues were recorded. The subjects stopped running when their subjective RPE reached 15.
The time taken for the RPE value to reach 15 was recorded, and the HR signal was col-
lected for 5 s before the end of the run. Immediately after the subjects stopped running,
the weighted knee extension experiment was performed, and the sEMG signals of the RF
and VM were collected from the subjects. Throughout the experiment, the experimenter
observed the subjects’ movement status and ensured their safety.
After the experiment, the HR band, electrode pads and other experimental equip-
ment were removed, and the subjects were asked if there was any discomfort and were
instructed to perform appropriate stretching and recovery exercises to regulate HR recov-
ery. Then, the next experiment was scheduled.
2.7. Data Processing and Analysis
All results are presented as group means and standard deviations. The normality of
the data distribution was confirmed using the Shapiro–Wilk test. To determine the effect
of the intervention on the dependent variable, a two-way analysis of variance (ANOVA)
for repeated measures was used to calculate between-group differences. If the interaction
between music tempo and exercise intensity was significant, Bonferroni post hoc tests
were calculated. The effect size was calculated as partial eta squared (η2). The criteria for
classifying Cohen d were as follows: small (0 < d < 0.5), medium (0.5 ≤ d < 0.8) and large
(d ≥ 0.8) [23]. The significance level was 0.05. All statistical analyses were performed using
SPSS 26.0 (SPSS Inc., Chicago, IL, USA).
3. Results
3.1. Examination of the Effects of Music Tempo and Exercise Intensity
A two-way repeated-measures ANOVA was conducted for TFP and the differences
in HR and MF to determine whether there were main and interaction effects for music
Figure 3. (a) The electrode position for the target muscles and the actual situation of wearing and
using the instrument; (b) the setup for measuring sEMG signals when loading and stretching the
knee; and (c) the running experimental procedure.
After the experiment, the HR band, electrode pads and other experimental equipment
were removed, and the subjects were asked if there was any discomfort and were instructed
to perform appropriate stretching and recovery exercises to regulate HR recovery. Then,
the next experiment was scheduled.
2.7. Data Processing and Analysis
All results are presented as group means and standard deviations. The normality of
the data distribution was confirmed using the Shapiro–Wilk test. To determine the effect
of the intervention on the dependent variable, a two-way analysis of variance (ANOVA)
for repeated measures was used to calculate between-group differences. If the interaction
between music tempo and exercise intensity was significant, Bonferroni post hoc tests
were calculated. The effect size was calculated as partial eta squared (η2). The criteria for
classifying Cohen d were as follows: small (0 < d < 0.5), medium (0.5 ≤ d < 0.8) and large
(d ≥ 0.8) [23]. The significance level was 0.05. All statistical analyses were performed using
SPSS 26.0 (SPSS Inc., Chicago, IL, USA).
3. Results
3.1. Examination of the Effects of Music Tempo and Exercise Intensity
A two-way repeated-measures ANOVA was conducted for TFP and the differences
in HR and MF to determine whether there were main and interaction effects for music
tempo and exercise intensity, and the results are shown in Table 4. The main effect of music
tempo on TFP was significant (p < 0.001), but the main effects of the differences in HR and
MF were not significant (p > 0.05). The main effect of exercise intensity on TFP and the
difference in HR was significant (p < 0.001), but the main effect for the difference in MF was
not significant (p > 0.05). The interaction effect of music tempo and exercise intensity was
significant for TFP (p < 0.05) but not for the difference in HR or MF (p > 0.05).
Int. J. Environ. Res. Public Health 2022, 19, 3869
10 of 18
Table 4. Examination of the effects of music tempo and exercise intensity.
Parameter
High
Low
Music Tempo
Exercise
Intensity
Music Tempo
× Exercise
Intensity
Fast
Slow
None
Fast
Slow
None
p-Value
η2
p-Value
η2
p-Value
η2
TFP (min)
7.18 ± 2.36
6.23 ± 2.44
5.40 ± 1.94
12.68 ± 6.46
10.79 ± 4.86
9.06 ± 4.36
0.000
0.632
0.000
0.540
0.031
0.207
HR
difference 1
69.56 ± 10.66
67.39 ± 11.14
71.11 ± 10.18
57.89 ± 10.30
61.83 ± 10.81
62.06 ± 8.95
0.077
0.140
0.000
0.796
0.075
0.141
MFRF
difference 2
−0.64 ± 4.98
−2.02 ± 2.95
0.53 ± 4.45
−0.86 ± 6.34
−0.42 ± 6.81
−0.96 ± 5.11
0.672
0.023
0.967
0.000
0.210
0.088
MFVM
difference 2
−0.75 ± 5.91
−0.89 ± 3.81
1.01 ± 4.2
0.38 ± 3.81
−1.08 ± 5.6
−0.79 ± 4.41
0.609
0.029
0.670
0.011
0.323
0.064
1 HR difference value is post-run HR−resting HR. 2 MF difference is MF before running−MF after running. RF: rectus femoris. VM: vastus medialis.
Int. J. Environ. Res. Public Health 2022, 19, 3869
11 of 18
3.2. Time to Fatigue Perception
The effects of music tempo on TFP at different exercise intensities are shown in Figure 4.
The combined results show that the TFP of the group with music was significantly greater
than that of the group without music. Specifically, in the low-intensity exercise experiment,
the TFP in the no music condition (9.06 min) was shorter than the TFP in the slow music
condition (10.79 min) and the fast music condition (12.68 min). During high-intensity
exercise, the TFP in the no music condition (5.40 min) was shorter than the TFP in the slow
tempo music condition (6.23 min) and the TFP in the fast tempo music condition (7.18 min).
difference 2
0.64 ± 4.98
2.02 ± 2.95 0.53 ± 4.45
0.86 ± 6.34 0.42 ± 6.81 0.96 ± 5.11
0.672 0.023 0.967 0.000 0.210
0.088
MFVM
difference 2 −0.75 ± 5.91 −0.89 ± 3.81 1.01 ± 4.2
0.38 ± 3.81 −1.08 ± 5.6 −0.79 ± 4.41
0.609 0.029 0.670 0.011 0.323
0.064
1 HR difference value is post-run HR−resting HR. 2 MF difference is MF before running−MF after
running. RF: rectus femoris. VM: vastus medialis.
3.2. Time to Fatigue Perception
The effects of music tempo on TFP at different exercise intensities are shown in Figure
4. The combined results show that the TFP of the group with music was significantly
greater than that of the group without music. Specifically, in the low-intensity exercise
experiment, the TFP in the no music condition (9.06 min) was shorter than the TFP in the
slow music condition (10.79 min) and the fast music condition (12.68 min). During high-
intensity exercise, the TFP in the no music condition (5.40 min) was shorter than the TFP
in the slow tempo music condition (6.23 min) and the TFP in the fast tempo music condi-
tion (7.18 min).
Figure 4. Time to fatigue perception for different experimental conditions.
Since the interaction effect of music tempo and exercise intensity on the TFP was sig-
nificant, a further Bonferroni post hoc test was performed to show the results (Table 5).
The results of the Bonferroni post hoc test showed that the difference in the effect on TFP
between no music and slow music at different exercise intensities was statistically signif-
icant (p < 0.01), the difference in the effect on TFP between no music and fast music was
Figure 4. Time to fatigue perception for different experimental conditions.
Since the interaction effect of music tempo and exercise intensity on the TFP was
significant, a further Bonferroni post hoc test was performed to show the results (Table 5).
The results of the Bonferroni post hoc test showed that the difference in the effect on
TFP between no music and slow music at different exercise intensities was statistically
significant (p < 0.01), the difference in the effect on TFP between no music and fast music
was statistically significant (p < 0.001) and the difference in the effect on TFP between slow
music and fast music was statistically significant (p < 0.001).
Table 5. Post hoc test of TFP under different experimental conditions.
Exercise Intensity
Pairwise Comparisons
No Music vs.
Slow Music vs.
Fast Music vs.
Low intensity
Slow music: p = 0.004
No music: p = 0.004
No music: p = 0.000
Fast music: p = 0.000
Fast music: p = 0.000
Slow music: p = 0.000
High intensity
Slow music: p = 0.000
No music: p = 0.000
No music: p = 0.0000
Fast music: p = 0.000
Fast music: p = 0.000
Slow music: p = 0.000
3.3. HR Changes
The effect of music tempo on the differences in HR at different exercise intensities
is shown in Figure 5. As seen from the difference in HR between the before and after
measurements, the difference in HR in the high-intensity exercise group was higher than
that in the low-intensity group overall. This is evident in the following: during low-
intensity exercise, the no music group had the highest HR variation (62.06 beats/min),
followed by the slow music group (61.83 beats/min) and finally the fast music group
(57.89 beats/min); during high-intensity exercise, the no music group had the highest HR
variation (71.11 beats/min), followed by the fast music group (69.56 beats/min) and finally
the slow music group (67.39 beats/min).
Int. J. Environ. Res. Public Health 2022, 19, 3869
12 of 18
,
g
y
g
p
g
in the low-intensity group overall. This is evident in the following: during low-intensity
exercise, the no music group had the highest HR variation (62.06 beats/min), followed by
the slow music group (61.83 beats/min) and finally the fast music group (57.89 beats/min);
during high-intensity exercise, the no music group had the highest HR variation (71.11
beats/min), followed by the fast music group (69.56 beats/min) and finally the slow music
group (67.39 beats/min).
Figure 5. Differences in HR before and after running under different experimental conditions.
3.4. sEMG Changes
The results of the MF difference of EMG signals of the RF and VM of the subjects
under the six exercise protocols are shown in Figure 6a,b. From the figure, it can be seen
that the MF difference values of both the RF and VM have two different directions of pos-
itive and negative results, failing to show a regular variation in surface EMG signal
changes of the RF and VM.
Figure 5. Differences in HR before and after running under different experimental conditions.
3.4. sEMG Changes
The results of the MF difference of EMG signals of the RF and VM of the subjects under
the six exercise protocols are shown in Figure 6a,b. From the figure, it can be seen that the
MF difference values of both the RF and VM have two different directions of positive and
negative results, failing to show a regular variation in surface EMG signal changes of the
RF and VM.
Int. J. Environ. Res. Public Health 2022, 19, x FOR PEER REVIEW
12 of 18
(a)
(b)
Figure 6. (a) MF differences of rectus femoris before and after running under different experi-
mental conditions; (b) MF differences of vastus medialis before and after running under different
experimental conditions.
4. Discussion
4.1. The Effect of Music Tempo on Runners’ Subjective Perception of Fatigue at Different Exer-
cise Intensities
The TFP results indicate that music can effectively influence runners’ perception of
fatigue during running, a finding that is consistent with previous studies [45,46]. The sig-
nificant main effect of music tempo (p < 0.001) also confirms the positive effect of music
on fatigue perception (0.50 ≤ effect size d < 0.8). The results show that music tempo had a
moderate effect on the time of fatigue perception. This result is consistent with Nethery’s
study, where listening to music reduced subjective fatigue induced by exercise at different
exercise intensities [18]. The mechanism of the effect can be explained by attentional lim-
itation theory and selective sensory filtering theory [47,48]. That is, listening to music dur-
ing exercise can reduce the excitability of the sympathetic nervous system, thereby reduc-
ing subjective fatigue perception and enhancing exercise tolerance. This study further ex-
plored the influence of music tempo on the TFP at different exercise intensities. This also
suggests that in the absence of musical stimuli, participants may be more focused on their
efforts and feel fatigued more quickly.
The significant main effect of exercise intensity (p < 0.001) indicates that there was a
significant difference between the effects of high and low exercise intensity on runners’
subjective perception of fatigue, and there was a moderate effect of exercise intensity on
subjective fatigue perception (0.50 ≤ effect size d < 0.8), which is consistent with common
Figure 6. (a) MF differences of rectus femoris before and after running under different experimental
conditions; (b) MF differences of vastus medialis before and after running under different experimen-
tal conditions.
4. Discussion
4.1. The Effect of Music Tempo on Runners’ Subjective Perception of Fatigue at Different
Exercise Intensities
The TFP results indicate that music can effectively influence runners’ perception of
fatigue during running, a finding that is consistent with previous studies [45,46]. The
significant main effect of music tempo (p < 0.001) also confirms the positive effect of music
on fatigue perception (0.50 ≤ effect size d < 0.8). The results show that music tempo had a
moderate effect on the time of fatigue perception. This result is consistent with Nethery’s
study, where listening to music reduced subjective fatigue induced by exercise at different
exercise intensities [18]. The mechanism of the effect can be explained by attentional
limitation theory and selective sensory filtering theory [47,48]. That is, listening to music
during exercise can reduce the excitability of the sympathetic nervous system, thereby
reducing subjective fatigue perception and enhancing exercise tolerance. This study further
explored the influence of music tempo on the TFP at different exercise intensities. This also
suggests that in the absence of musical stimuli, participants may be more focused on their
efforts and feel fatigued more quickly.
Int. J. Environ. Res. Public Health 2022, 19, 3869
13 of 18
The significant main effect of exercise intensity (p < 0.001) indicates that there was
a significant difference between the effects of high and low exercise intensity on runners’
subjective perception of fatigue, and there was a moderate effect of exercise intensity on
subjective fatigue perception (0.50 ≤ effect size d < 0.8), which is consistent with common
knowledge about the positive correlation between exercise intensity and fatigue perception.
Meanwhile, the significant interaction effect of music tempo and exercise intensity indicated
that the effects of music tempo and exercise intensity (p < 0.05) on runners’ subjective
perception of fatigue were influenced by each other, but the effect is small. However, the
effect size result (0 < effect size d < 0.5) shows that music tempo and exercise intensity
interacted to a lesser extent. The results of further post hoc analyses indicate that whether in
high intensity or low intensity exercise, listening to fast music can prolong the running time
and reduce the subjective fatigue of runners compared to listening to slow music. During
high-intensity exercise, the effect of fast music and slow music on the TFP is significant
(p < 0.001), indicating that fast music was to some extent more effective than slow music in
reducing fatigue perception. This conclusion is also supported by previous studies, such
as Cental’s study, which mentioned that listening to fast music increased overall exercise
tolerance and the neuromuscular fatigue threshold [18]. Furthermore, post-experiment
interviews with all subjects revealed that listening to fast music resulted in a more uplifted
state of mind and a more positive mood, which effectively distracted them and reduced the
perception of fatigue. Some subjects also believed that fast music made them unconsciously
adjust their running pace, which helped to counteract the perception of fatigue. The
combined results of the experiments show that fast music effectively reduced the subjective
perception of fatigue at different exercise intensities in runners.
4.2. The Effect of Music Tempo on HR Differential at Different Exercise Intensities
The ANOVA results show that the difference in HR before and after running was
only affected by the main effect of exercise intensity (p < 0.001) and reached an effect
level close to the high intensity (0.50 ≤ effect size d < 0.8). It indicates that HR changes
were largely influenced by the exercise intensity. This finding indicates that there is a
significant difference in the magnitude of HR variation in runners at different intensities of
exercise, and this finding is consistent with the common knowledge that there is a strong
correlation between the magnitude of HR variation in runners and the intensity of exercise.
In addition, a study by Szabo et al. provides evidence for the idea that music not only has
an effect on fatigue perception during progressive exercise, but can still provide effects for
a period of time at higher intensities, depending on the quality of intrinsic arousal [49]. The
main reason is that the greater intensity of exercise causes a greater cardiac load, and the
maintenance of cardiac output depends on an increase in HR; the fatigue of the body also
increases, resulting in a higher HR immediately after running [50].
However, the main effect of music tempo was not significant (p > 0.05), and the
level of effect size was small (0 < effect size d < 0.5), indicating that the magnitude of HR
change during exercise was influenced by music tempo to a lesser extent. This finding is
inconsistent with some current studies which indicate that the magnitude of HR variation
during exercise is influenced by music tempo [51]. This finding is also inconsistent with
the significant main effect of music tempo on the TFP, probably because the TFP is a
subjective index of fatigue that reflects runners’ self-evaluation of their physical state
and is significantly influenced by mental aspects. In contrast, under control conditions,
the HR variation of the subjects mainly depends on subjects’ physical fatigue [52]. As
some studies pointed out, the effect of music on fatigue perception would be reflected in
the shift of runners’ attention to the fatigue state; that is, it would be shown first in the
shift in perception of mental fatigue and then in the perception of physical fatigue [53,54].
However, the duration of this experiment was determined by the runners’ subjective
fatigue, and the runners ended the run when their subjective fatigue reached the level of
RPE = 15, which led to the intervention effect of music tempo on the runners’ physical
fatigue perception not being fully demonstrated in this experiment. This further led to
Int. J. Environ. Res. Public Health 2022, 19, 3869
14 of 18
a nonsignificant interaction effect between music tempo and exercise intensity (p > 0.05,
0 < effect size d < 0.5). However, there are studies that have come to similar conclusions
as the present study, such as Edworthy and Waring and Dyck et al., who also did not find
a link between music tempo changes and HR changes [17,55]. HR and music tempo can
be considered as interacting oscillatory systems that will start at the same period, but this
alignment strategy only works at the beginning of the experiment.
4.3. Effect of Music Tempo on the Difference in MF at Different Exercise Intensities
The differences in MF for the RF and VM for different exercise regimens varied widely
and did not show regular changes. The results show that the main effects of music tempo
and exercise intensity on both MFRF difference and MFVM difference were not significant
(p > 0.05) and that both had small effect sizes (0 < effect size d < 0.5). The interaction
effects of music tempo and exercise intensity on MFRF and MFVM differences were also not
significant (p > 0.05) and both effect sizes were small (0 < effect size d < 0.5). Meanwhile,
the experimental results for the differences in MF were negative, which indicates decreased
local muscle fatigue after running, which is inconsistent with the results of most previous
studies that used the difference in MF as an indicator [56,57]. The reason may be due to
the different distribution patterns of fast and slow muscle fibers in different individuals
and muscles, which are genetically determined and almost impossible to change later by
exercise [58]. In the post-experiment communication with the subjects, we learned that
runners may use different force generation methods for different running conditions, which
also indirectly influenced the experimental results. Meanwhile, the ANOVA results also
show that the main effects of both exercise intensity and music tempo were not significant,
indicating that the change in muscle fatigue before and after running was not significant.
However, in the post-experiment questioning of the subjects, we found that the subjects all
had some soreness and swelling in their lower limb muscles after finishing the running
experiment. This phenomenon was not monitored during the experiment, and the sEMG
signal changes for the two muscles before and after running were not regular; there were
more negative differences in MF, which shows that the MF values in this experiment did
not show a decreasing trend but an increasing trend. The reasons for this result in this study
may be twofold. (1) The experimental test time was short, and the runners reached the
specified value of subjective fatigue in a short period of time, during which the local muscle
tissue continuously recruited fast muscle fibers to participate in the exercise so that the
muscle firing frequency increased, and the neural excitability continued during the running
process [59,60]. (2) The energy consumption during muscle contraction during running
did not exceed the energy recovery during relaxation, and the lactic acid accumulation was
not significant, so the driving strategy of the central nervous system for the muscles and
the conduction speed of the muscle fibers were not affected [61]. Other scholars have also
found different trends in MF frequency domain indicators during exercise [62]. Combining
the above views to see the limitations of using the decrease in the MF frequency domain
index pre- to post-exercise can determine the fatigue of local muscles during dynamic
exercise such as running.
4.4. Pratical Applications
This study aimed to determine the effect of music tempo on runners’ fatigue perception
at different exercise intensities and provided a physiological and psychological explanation
of the role of music tempo in resisting fatigue perception by observing time of fatigue
perception, heart rate changes and sEMG signal changes during the running experiment.
The combined results of the three indicators revealed that the use of music with different
tempos at different exercise intensities caused runners to exhibit different mental and
physical performances. Specifically, runners’ TFP was influenced by the music tempo, and
in addition, runners’ TFP was also influenced by the interaction between music tempo and
exercise intensity. The change in HR of runners during running was mainly influenced
by the exercise intensity. Whether it is low-intensity exercise or high-intensity exercise,
Int. J. Environ. Res. Public Health 2022, 19, 3869
15 of 18
listening to fast music may mitigate individuals’ perception of fatigue to some extent. This
was also pointed out in the study by Centala et al. The effect of listening to fast music
during exercise would be better than slow music and could regulate mental emotions and
reduce the perception of fatigue during exercise [63].
In addition, this study also provides many practical implications. Running is an
endurance sport, and runners will feel fatigue to different degrees during running, es-
pecially for many non-professional runners, who often do not persist in running due to
insufficient exercise, limited physical strength and lack of perseverance. The results of this
study can help runners to develop a more beneficial exercise program. For example, both
low-intensity exercisers and high-intensity exercisers are advised to listen to fast music
while exercising, which would help them maintain a better mental state, make the exercise
process less tedious and better control the physiological state, including heart rate changes.
This research could lead to better effects of music on perceived fatigue, allowing runners to
have better mental and physical performance while running.
4.5. Limitations
There were some limitations to this study. First, the subject population was not
subdivided, and none of the runners had undergone professional training. Second, we only
examined male subjects. Future studies could use sex as an independent variable to discuss
the effects of music tempo on subjects of different sexes and could also include people
of all ages and from different occupational backgrounds to make the study results more
generalizable. Third, the observation indexes used in this study were relatively limited,
and more measurement techniques could be integrated in the future to more accurately
assess the fatigue perception status of runners. Fourth, based on this study, future studies
could refine the characteristics of the music, such as the cultural background of the music,
or use music with lyrics to explore other methods of music intervention during exercise to
promote the popularity of the research results.
5. Conclusions
In this paper, the effect of music tempo on runners’ subjective and objective fatigue
perception at different exercise intensities was investigated through running experiments.
The influence of music on runners’ perception of fatigue was explored, and the interaction
between music tempo and exercise intensity was considered. The results of the study
showed that (1) there were significant main effects and interaction effects of music tempo
and exercise intensity on the TFP; (2) in terms of the difference in HR, the main effect of
exercise intensity was significant, the main effect of music tempo was not significant and
the interaction effect of music tempo and exercise intensity was not significant; and (3) in
terms of the difference in the MF of sEMG, the main effect and interaction effect of music
tempo and exercise intensity were not significant, and the results did not show regular
changes. The combined results of the study indicate that fast music can effectively reduce
the perception of fatigue of runners during running. The results of this study provide
a scientific basis for ordinary runners to select evidence-based for running, help them
effectively overcome their perception of fatigue during exercise and optimally modify their
perception of fatigue with music.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10
.3390/ijerph19073869/s1, File S1: Ethic reviews; File S2: Informed Consent.
Author Contributions: Conceptualization, J.W.; methodology, L.J. and H.Y.; investigation, Y.C.;
resources, C.L.; formal analysis, L.J. and L.Z.; writing—original draft preparation, L.J.; writing—
review and editing, L.Z.; supervision, J.W. and H.Y. All authors have read and agreed to the published
version of the manuscript.
Funding: This research was funded by the Key Industrial Projects of Zhejiang Province Science and
Technology Department, China, grant number 2019C03124.
Int. J. Environ. Res. Public Health 2022, 19, 3869
16 of 18
Institutional Review Board Statement: The study was conducted in accordance with the Declaration
of Helsinki and approved by the Ethics Committee of the Industrial Design Institute of Zhejiang
University of Technology (protocol code 0903/2021, date of approval 15 September 2021).
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Pageaux, B. Perception of effort in exercise science: Definition, measurement and perspectives. Eur. J. Sport Sci. 2016, 16, 885–894.
[CrossRef] [PubMed]
2.
Taylor, J.L.; Amann, M.; Duchateau, J.; Meeusen, R.; Rice, C.L. Neural contributions to muscle fatigue: From the brain to the
muscle and back again. Med. Sci. Sports Exerc. 2016, 48, 2294.
3.
Azevedo, R.; Silva-Cavalcante, M.D.; Gualano, B.; Lima-Silva, A.; Bertuzzi, R. Effects of caffeine ingestion on endurance
performance in mentally fatigued individuals. Eur. J. Appl. Physiol. 2016, 116, 2293–2303. [CrossRef] [PubMed]
4.
Shin, M. Interactive effect of listening to music and mid- and post-exercise affective change on exercise intention. Psychol. Health
Med. 2021, 1–7. [CrossRef]
5.
Gholami, M.; Napier, C.; Patiño, A.G.; Cuthbert, T.J.; Menon, C. Fatigue monitoring in running using flexible textile wearable
sensors. Sensors 2020, 20, 5573. [CrossRef]
6.
Clansey, A.C.; Hanlon, M.; Wallace, E.S.; Lake, M.J. Effects of fatigue on running mechanics associated with tibial stress fracture
risk. Med. Sci. Sports Exerc. 2012, 44, 1917–1923. [CrossRef] [PubMed]
7.
Swain, D.P.; Franklin, B.A. Comparison of cardioprotective benefits of vigorous versus moderate intensity aerobic exercise. Am. J.
Cardiol. 2006, 97, 141–147. [CrossRef] [PubMed]
8.
Brown, D.M.; Bray, S.R. Effects of mental fatigue on exercise intentions and behavior. Ann. Behav. Med. 2019, 53, 405–414.
9.
Nikol, L.; Kuan, G.; Ong, M.; Chang, Y.-K.; Terry, P.C. The heat is on: Effects of synchronous music on psychophysical parameters
and running performance in hot and humid conditions. Front. Psychol. 2018, 9, 1114.
10.
Meng, Q.; Jiang, J.; Liu, F.; Xu, X. Effects of the musical sound environment on communicating emotion. Int. J. Environ. Res. Public
Health 2020, 17, 2499. [CrossRef]
11.
Yamamoto, T.; Ohkuwa, T.; Itoh, H.; Kitoh, M.; Terasawa, J.; Tsuda, T.; Kitagawa, S.; Sato, Y. Effects of pre-exercise listening to
slow and fast tempo music on supramaximal cycle performance and selected metabolic variables. Arch. Physiol. Biochem. 2003,
111, 211–214. [PubMed]
12.
Rejeski, W.J. Perceived exertion: An active or passive process? J. Sport Exerc. Psychol. 1985, 7, 371–378.
13.
Faulkner, M.; McNeilly, A.; Davison, G.; Rowe, D.; Hewitt, A.; Nevill, A.; Duly, E.; Trinick, T.; Murphy, M. Music tempo: A tool for
regulating walking cadence and physical activity intensity in overweight adults? Int. J. Environ. Res. Public Health 2021, 18, 7855.
[PubMed]
14.
Crust, L. Perceived importance of components of asynchronous music during circuit training. J. Sports Sci. 2008, 26, 1547–1555.
[CrossRef]
15.
Stork, M.J.; Kwan, M.Y.; Gibala, M.J.; Martin Ginis, K. Music enhances performance and perceived enjoyment of sprint interval
exercise. Med. Sci. Sports Exerc. 2015, 47, 1052–1060.
16.
Karageorghis, C.I.; Priest, D.-L. Music in the exercise domain: A review and synthesis (Part I). Int. Rev. Sport Exerc. Psychol. 2012,
5, 44–66. [CrossRef]
17.
Edworthy, J.; Waring, H. The effects of music tempo and loudness level on treadmill exercise. Ergonomics 2006, 49, 1597–1610.
[CrossRef]
18.
Nethery, V.M. Competition between internal and external sources of information during exercise: Influence on RPE and the
impact of the exercise load. J. Sports Med. Phys. Fit. 2002, 42, 172–178.
19.
Pujol, T.J.; Langenfeld, M.E. Influence of music on Wingate Anaerobic Test performance. Percept. Mot. Skills. 1999, 88, 292–296.
20.
Maddigan, M.E.; Sullivan, K.M.; Halperin, I.; Basset, F.A.; Behm, D.G. High tempo music prolongs high intensity exercise. PeerJ
2019, 6, e6164. [CrossRef]
21.
Karageorghis, C.; Jones, L.; Low, D. Relationship between exercise heart rate and music tempo preference. Res. Q. Exerc. Sport
2006, 77, 240–250. [PubMed]
22.
Hutchinson, J.; Sherman, T. The relationship between exercise intensity and preferred music intensity. Sport. Exerc. Perform. 2014,
3, 191.
23.
Cohen, J. Statistical Power Analysis for the Behavioral Sciences; Routledge: New York, NY, USA, 2013.
24.
Scheer, F.A.J.L.; Hu, K.; Evoniuk, H.; Kelly, E.E.; Malhotra, A.; Hilton, M.F.; Shea, S.A. Impact of the human circadian system,
exercise, and their interaction on cardiovascular function. Proc. Natl. Acad. Sci. USA 2010, 107, 20541–20546. [CrossRef] [PubMed]
25.
Karageorghis, C.; Jones, L.; Stuart, D.P. Psychological effects of music tempi during exercise. Int. J. Sports Med. 2008, 29, 613–619.
[CrossRef]
Int. J. Environ. Res. Public Health 2022, 19, 3869
17 of 18
26.
Da Cunha, F.A.; Farinatti, P.T.V.; Midgley, A.W. Methodological and practical application issues in exercise prescription using the
heart rate reserve and oxygen uptake reserve methods. J. Sci. Med. Sport 2011, 14, 46–57.
27.
Chai, G.; Wang, Y.; Wu, J.; Yang, H.; Tang, Z.; Zhang, L. Study on the recognition of exercise intensity and fatigue on runners
based on subjective and objective information. Healthcare 2019, 7, 150. [CrossRef]
28.
Williams, N. The Borg rating of perceived exertion (RPE) scale. Occup. Med. 2017, 67, 404–405. [CrossRef]
29.
Kim, T.; Han, J.; Lee, J.; Choi, Y. The effect of polarized training on the athletic performance of male and female cross-country
skiers during the general preparation period. Healthcare 2021, 9, 851. [CrossRef]
30.
Finni, T.; Hu, M.; Kettunen, P.; Vilavuo, T.; Cheng, S. Measurement of EMG activity with textile electrodes embedded into clothing.
Physiol. Meas. 2007, 28, 1405. [CrossRef]
31.
Ražanskas, P.; Verikas, A.; Viberg, P.-A.; Olsson, M.C. Predicting physical parameters in fatiguing bicycling exercises using muscle
activation timing. Biomed. Signal Process. Control 2017, 35, 19–29.
32.
Yuan, A.G.; Shi, S.R.; Cai-Yun, H.E. Characteristics on EMG during exercise-induced muscle fatigue and arguments of its
mechanism. Sichuan Sport. Sci. 2006, 3, 36–39.
33.
Wu, J.; Li, X.; Liu, W.; Wang, Z.J. sEMG signal processing methods: A review. J. Phys. Conf. Ser. 2019, 1237, 032008.
34.
Wu, D.M.; Xin, S.; Zhang, Z.C. Feature collection and analysis of surface electromyography signals. J. Clin. Rehabilit. Tissue Eng.
Res. 2010, 14, 8073–8076.
35.
Cifrek, M.; Medved, V.; Tonkovi, S.; Ostoji, S. Surface EMG based muscle fatigue evaluation in biomechanics. Clin. Biomech. 2009,
24, 327–340.
36.
Kupa, E.J.; Roy, S.H.; Kandarian, S.C.; de Luca, C.J. Effects of muscle fiber type and size on EMG median frequency and conduction
velocity. J. Appl. Physiol. 1995, 79, 23–32.
37.
Wu, J.; Wu, Q.; Sun, S. Research on classification algorithm of reduced support vector machine for low limb movement recognition.
China Mech. Eng. 2011, 22, 433–438.
38.
Hermens, H.J.; Freriks, B.; Disselhorst-Klug, C.; Rau, G. Development of recommendations for SEMG sensors and sensor
placement procedures. J. Electromyogr. Kinesiol. 2000, 10, 361–374. [CrossRef]
39.
Wolf, S.L. Essential considerations in the use of EMG biofeedback. Phys. Ther. 1978, 58, 25–31.
40.
Pakosz, P.; Domaszewski, P.; Konieczny, M.; B ˛aczkowicz, D. Muscle activation time and free-throw effectiveness in basketball. Sci.
Rep. 2021, 11, 7489. [CrossRef]
41.
Stegeman, D.; Hermens, H. Standards for Surface Electromyography: The European Project Surface EMG for Non-Invasive Assessment of
Muscles (SENIAM); Roessingh Research and Development: Enschede, The Netherlands, 2007; Volume 10, pp. 8–12.
42.
Hermens, H.; Freriks, B. The State of the Art on Sensors and Sensor Placement Procedures for Surface Electromyography: A Proposal for
Sensor Placement Procedures; Roessingh Research and Development: Enschede, The Netherlands, 1997.
43.
Cahyadi, B.; Khairunizam, W.; Muhammad, M.N.; Zunaidi, I.; Majid, S.; Rudzuan, M.; Bakar, S.A.; Razlan, Z.M.; Mustafa,
W. Analysis of EMG based arm movement sequence using mean and median frequency. In Proceedings of the 2018 5th
IEEE International Conference on Electrical Engineering, Computer Science and Informatics (EECSI), Malang, Indonesia,
16–18 October 2018; pp. 440–444.
44.
Taimela, S.; Webber, C.L.; Airaksinen, O. Lumbar paraspinal muscle fatigability in repetitive isoinertial loading: EMG spectral
indices, Borg scale and endurance time. Eur. J. Appl. Physiol. Occup. Physiol. 1997, 76, 236–242. [CrossRef]
45.
Crust, L.; Clough, P.J. The influence of tempo and personality in the endurance response to motivational asynchronous music.
J. Sport. Sci. 2006, 24, 187–195.
46.
Hutchinson, J.C.; Sherman, T.; Davis, L.; Cawthon, D.; Reeder, N.B.; Tenenbaum, G. The influence of asynchronous motivational
music on a supramaximal exercise bout. Int. J. Sport Psychol. 2011, 42, 135–148.
47.
Razon, S.; Basevitch, I.; Land, W.; Thompson, B.; Tenenbaum, G. Perception of exertion and attention allocation as a function of
visual and auditory conditions. Psychol. Sport Exerc. 2009, 10, 636–643.
48.
Morillon, B.; Hackett, T.; Kajikawa, Y.; Schroeder, E. Predictive motor control of sensory dynamics in auditory active sensing.
Curr. Opin. Neurobiol. 2015, 31, 230–238.
49.
Szabo, A.; Small, A.; Leigh, M. The effects of slow-and fast-tempo classical music on progressive cycling to voluntary physical
exhaustion. J. Sports Med. Phys. Fit. 1999, 39, 220.
50.
Arazi, H.; Keihaniyan, A.; EatemadyBoroujeni, A.; Oftade, A.; Takhsha, S.; Asadi, A.; Ramirez-Campillo, R. Effects of heart rate
vs. speed-based high intensity interval training on aerobic and anaerobic capacity of female soccer players. Sports 2017, 5, 57.
[CrossRef]
51.
Lee, S.; Kimmerly, D.S. Influence of music on maximal self-paced running performance and passive post-exercise recovery rate.
J. Sport. Med. Phys. Fit. 2016, 56, 39–48.
52.
Darbandy, M.; Rostamnezhad, M.; Hussain, S.; Khosravi, A.; Nahavandi, S.; Sani, Z. A new approach to detect the physical
fatigue utilizing heart rate signals. Res. Cardiovasc. Med. 2020, 9, 23–27.
53.
Ekkekakis, P. Pleasure and displeasure from the body: Perspectives from exercise. Cogn. Emot. 2003, 17, 213–239. [CrossRef]
54.
Bigliassi, M.; Karageorghis, C.; Bishop, D.; Alexander, V.; Michael, J. Cerebral effects of music during isometric exercise: An fMRI
study. Int. J. Psychophysiol. 2018, 133, 131–139.
55.
Dyck, E.; Six, J.; Soyer, E.; Denys, M.; Bardijn, I.; Leman, M. Adopting a music-to-heart rate alignment strategy to measure the
impact of music and its tempo on human heart rate. Music. Sci. 2017, 21, 390–404.
Int. J. Environ. Res. Public Health 2022, 19, 3869
18 of 18
56.
Jia, W.; Zhou, R. The analysis of muscle fatigue based on sEMG. In Proceedings of the 2013 IEEE International Conference on
Mechatronic Sciences, Electric Engineering and Computer (MEC), Shenyang, China, 20–22 December 2013; pp. 862–865.
57.
Klich, S.; Kawczy´nski, A.; Pietraszewski, B.; Zago, M.; Chen, A.; Smoter, M.; Hassanlouei, H.; Lovecchio, N. Electromyographic
evaluation of the shoulder muscle after a fatiguing isokinetic protocol in recreational overhead athletes. Int. J. Environ. Res. Public
Health 2021, 18, 2516. [PubMed]
58.
Methenitis, S.; Karandreas, N.; Spengos, K.; Zaras, N.; Stasinaki, A.; Gerasimos, T. Muscle fiber conduction velocity, muscle fiber
composition, and power performance. Med. Sci. Sport. Exerc. 2016, 48, 1761–1771.
59.
Girard, O.; Mendez-Villanueva, A.; Bishop, D. Repeated-sprint ability—Part I: Factors contributing to fatigue. Sport. Med. 2011,
41, 673–694.
60.
Muñoz-López, A.; de Hoyo, M.; Sañudo, B. Muscle contractile properties measured at submaximal electrical amplitudes and not
at supramaximal amplitudes are associated with repeated sprint performance and fatigue markers. Int. J. Environ. Res. Public
Health 2021, 18, 11689. [CrossRef]
61.
Tang, J.; Li, M. Monitoring and application of physical load intensity in exercise. Chin. J. Tissue Eng. Res. 2012, 16, 3784–3788.
62.
Ament, W.; Verkerke, G.J.; Bonga, G.J.J.; Hof, A.L. Electromyogram median power frequency in dynamic exercise at medium
exercise intensities. Eur. J. Appl. Physiol. Occup. Physiol. 1996, 74, 180–186. [CrossRef]
63.
Centala, J.; Pogorel, C.; Pummill, S.W.; Malek, M.H. Listening to fast-tempo music delays the onset of neuromuscular fatigue.
J. Strength Cond. Res. 2020, 34, 617–622. [CrossRef]
| The Effect of Music Tempo on Fatigue Perception at Different Exercise Intensities. | 03-24-2022 | Wu, Jianfeng,Zhang, Lingyan,Yang, Hongchun,Lu, Chunfu,Jiang, Lu,Chen, Yuyun | eng |
PMC6679305 | sensors
Article
Use of Machine Learning and Wearable Sensors to
Predict Energetics and Kinematics of
Cutting Maneuvers
Matteo Zago 1,2,3,*, Chiarella Sforza 4
, Claudia Dolci 4, Marco Tarabini 3,5
and
Manuela Galli 1,3
1
Dipartimento di Elettronica, Informazione e Bioingegneria, Politecnico di Milano, 20133 Milano, Italy
2
Fondazione Istituto Farmacologico Filippo Serpero, 20159 Milano, Italy
3
E4Sport Lab, Politecnico di Milano, 20133 Milano, Italy
4
Dipartimento di Scienze Biomediche per la Salute, Università degli Studi di Milano, 20133 Milano, Italy
5
Dipartimento di Meccanica, Politecnico di Milano, 20129 Milano, Italy
*
Correspondence: [email protected]; Tel.: +39-02-2399-3351
Received: 14 June 2019; Accepted: 9 July 2019; Published: 12 July 2019
Abstract: Changes of directions and cutting maneuvers, including 180-degree turns, are common
locomotor actions in team sports, implying high mechanical load.
While the mechanics and
neurophysiology of turns have been extensively studied in laboratory conditions, modern inertial
measurement units allow us to monitor athletes directly on the field. In this study, we applied
four supervised machine learning techniques (linear regression, support vector regression/machine,
boosted decision trees and artificial neural networks) to predict turn direction, speed (before/after
turn) and the related positive/negative mechanical work. Reference values were computed using an
optical motion capture system. We collected data from 13 elite female soccer players performing a
shuttle run test, wearing a six-axes inertial sensor at the pelvis level. A set of 18 features (predictors)
were obtained from accelerometers, gyroscopes and barometer readings. Turn direction classification
returned good results (accuracy > 98.4%) with all methods. Support vector regression and neural
networks obtained the best performance in the estimation of positive/negative mechanical work
(coefficient of determination R2 = 0.42–0.43, mean absolute error = 1.14–1.41 J) and running speed
before/after the turns (R2 = 0.66–0.69, mean absolute error = 0.15–018 m/s). Although models can
be extended to different angles, we showed that meaningful information on turn kinematics and
energetics can be obtained from inertial units with a data-driven approach.
Keywords: supervised learning; changes of direction; IMU; mechanical work
1. Introduction
Changes of direction (CoD) and cutting maneuvers are basic locomotor actions in team sports,
implying high physiological and mechanical load [1–3]. High-intensity and abrupt sidestepping is
the most frequent cause for non-contact ligamentous injuries at the knee level, involving primarily
anterior cruciate ligament lesions, and secondarily meniscal or medial collateral ligament strains [4].
The amount of deceleration required in sidestep cutting is related to the angle and speed of approach
and has been associated to the likelihood of knee injuries [5,6]. CoDs also have a high associated
metabolic cost, impacting on the energetic requirements of exercise [7,8].
The mechanics and neurophysiology of CoDs have been accurately described primarily in
laboratory conditions, unveiling joint kinematics and loads as a function of the running angle
and technique [1,9–11], foot-landing strategies [12], muscular activations [3,13], and response to
Sensors 2019, 19, 3094; doi:10.3390/s19143094
www.mdpi.com/journal/sensors
Sensors 2019, 19, 3094
2 of 11
fatigue [14,15]. In addition, we recently proposed an algorithm to estimate the energy cost of running
with repeated 180 degree-CoDs [16]: the external mechanical energy associated to the trajectory of a
body’s center of mass was combined with the knee flexion angle and ground contacts to provide an
estimation of the positive (concentric) and negative (eccentric) muscular work.
Although promising, this method was limited by the obtainment of full-body kinematics with
optical motion capture systems, being therefore confined to laboratory conditions. Rather, gathering
reliable information on the side, speed and energetics of 180-degree sidestep CoDs in realistic
on-the-field conditions would better help in monitoring the energetic, physiological and mechanical
load, as well as to prevent overuse injuries.
An emerging trend, quantifying sport actions with body-worn inertial measurements units (IMU),
enables the assessment of athletes in ecologic conditions [17,18]. The metrological issues related to
the use of wearable sensors for sport performance assessments have been the focal point of different
research works: even though magneto-inertial technology allows monitoring the performance of
athletes of all levels, especially when complemented with a sensor fusion network, there is a need
for further research on the ease of use and error compensation to provide coaches and practitioners
with informative and concise metrics [19–22]. The use of inertial units also raises technical issues
to extract meaningful data from a broad class of signals (acceleration, angular velocity, magnetic
field orientation, etc.) which are often prone to noise, non-linearities, and measurement inaccuracies.
These characteristics might practically limit the usability of results in specific conditions. In the case of
changes of direction, obtaining running speed analytically from one inertial unit and then applying
linear equations to estimate the related energy cost in [8] appears practically unfeasible due to inherent
biases and drifts.
A way to overcome these limitations is to apply machine learning techniques to IMU data.
Supervised machine learning algorithms take a known set of input data (called predictors) and know
responses and train a model to generate predictions from new data [23]. These techniques have been
applied in team sports to quantify movement patterns during training and competition, like physical
output and tackling impacts in rugby and Australian football [18,24], player load [25] or deceleration
before turns in soccer [26]. However, the estimation of the energetics associated with 180-degree cutting
actions in team sports has not been investigated. This study intends to introduce the application of
machine learning models to detect direction, speed and external mechanical work associated with
180-degree CoDs, by using only data coming from a single inertial unit. We hypothesize that a unique
sensor placed close to the core (pelvis) could capture the key information on athletes’ actions during
these tasks. Our complementary aim is to show that the combination of regression analysis technique
and easily available sensors can provide coaches and practitioners with a wealth of information about
such crucial game actions.
2. Materials and Methods
2.1. Experimental Procedures and Equipment
All tests were performed in the morning within two weeks after the end of the regular season.
The experimental setting was a full motion capture laboratory equipped with an eight-camera system
(sampling frequency: 100 Hz; Smart-Dx, BTS Bioengineering, Milano, Italy). A set of 14 reflective
markers (diameter: 15 mm) were positioned on the skin in the following anatomical landmarks:
tragi, acromia, olecranons; radius styloid processes; greater trochanters; femoral lateral epicondyles;
and lateral malleoli (additional markers were added for further biomechanical investigations, but they
were not considered in the current study). A six-axes IMU (GaitUp Physilog 5, Lausanne, Switzerland)
was fixed to the shorts with a plastic clip close to the sacrum marker. Inertial sensor settings were:
sampling frequency 512 Hz, measurement range ±2000 degrees/s (gyroscope) and ±16 g (accelerometer).
The x-axis of the sensor reference frame pointed backwards, the y-axis upwards and the z-axis to the
subjects’ left. The unit also included a barometer with a sampling frequency of 64 Hz.
Sensors 2019, 19, 3094
3 of 11
Ambient temperature was 22–24 ◦C. Participants wore minimal sports clothing and running shoes.
They were first acquainted with the experimental procedures; after a 10 min warm up supervised by a
professional strength and conditioning coach, participants completed a 5 m shuttle-run test (Figure 1)
at the average speed of 70% of their maximal aerobic speed, as detailed in [7]. Maximal aerobic speed
is the lowest running speed at which the maximum oxygen uptake occurs, and it was estimated with
aerobic power tests (Yo-Yo intermittent recovery test [27]) throughout the season. Average running
speed was 2.5 ± 0.2 m/s. Athletes had to keep running to exhaustion, i.e., when they could not reach
the end lane by the acoustic signal—pacing the shuttle rhythm—for two consecutive times.
Sensors 2019, 19, x FOR PEER REVIEW
3 of 11
run test (Figure 1) at the average speed of 70% of their maximal aerobic speed, as detailed in [7].
Maximal aerobic speed is the lowest running speed at which the maximum oxygen uptake occurs,
and it was estimated with aerobic power tests (Yo-Yo intermittent recovery test [27]) throughout the
season. Average running speed was 2.5 ± 0.2 m/s. Athletes had to keep running to exhaustion, i.e.,
when they could not reach the end lane by the acoustic signal—pacing the shuttle rhythm—for two
consecutive times.
Figure 1. Experimental setting and turn (180-degree change of direction) action. In the right picture,
the orientation of the sensor reference frame is displayed.
2.2. Study Design and Participants
This observational case-series study involved 13 female soccer players (age 23.6 ± 3.3 years, body
mass: 59.0 ± 7.3 kg, height: 1.66 ± 0.05 m, body mass index: 21.2 ± 1.9 kg/m2), playing for elite clubs
competing in the first and second Italian league. All participants were judged by a medical doctor
not to have any restriction to sports practice, had no injuries in the 12 months preceding the test, and
signed a written informed consent after a detailed explanation of the aims, benefits and risks of this
study. The study was approved by the Institutional Ethics Committee (n. 1/2016) and was conducted
according to the Declaration of Helsinki.
2.3. Data Processing and Features Engineering
Custom routines were developed within Matlab (v. 2018b, The Mathworks Inc., Natick MA, USA).
The three-dimensional coordinates of body center of mass (CoM) were obtained using the segmental
centroid method, specifically validated for sports applications [28–30], after applying a low-pass, zero-
lag second order Butterworth filter (cut-off frequency: 15 Hz) to the raw marker trajectories. CoD events
were easily identified with the peaks of the CoM position in the running direction (CoMx). Mass-specific
CoM external energy was computed according to classical physiology texts [31]:
𝐸௫௧ = 1
2 𝒗ெ
ଶ
+ 𝑔ℎெ
(1)
where vCoM is the norm of instantaneous CoM speed, and hCoM is its vertical height. Positive (negative)
mechanical work (W+/−) was then obtained as the sum of positive (negative) changes of Eext [31] in the
two second window across the turn. Also, for each turn, we computed CoM approach speed (1 s before
the turn, vbefore), and CoM speed during acceleration (1 s after the turn, vafter). These four variables,
alongside the side of the pivoting limb (right or left), constitute the set of known targets (responses).
Their distribution is illustrated in Figure 2: as negative external work is required to decelerate the CoM
before the turn, and positive work is needed to accelerate it back in the new direction, W− was
represented relative to vbefore, and W+ to vafter.
Figure 1. Experimental setting and turn (180-degree change of direction) action. In the right picture,
the orientation of the sensor reference frame is displayed.
2.2. Study Design and Participants
This observational case-series study involved 13 female soccer players (age 23.6 ± 3.3 years,
body mass: 59.0 ± 7.3 kg, height: 1.66 ± 0.05 m, body mass index: 21.2 ± 1.9 kg/m2), playing for elite
clubs competing in the first and second Italian league. All participants were judged by a medical
doctor not to have any restriction to sports practice, had no injuries in the 12 months preceding the
test, and signed a written informed consent after a detailed explanation of the aims, benefits and risks
of this study. The study was approved by the Institutional Ethics Committee (n. 1/2016) and was
conducted according to the Declaration of Helsinki.
2.3. Data Processing and Features Engineering
Custom routines were developed within Matlab (v. 2018b, The Mathworks Inc., Natick, MA, USA).
The three-dimensional coordinates of body center of mass (CoM) were obtained using the segmental
centroid method, specifically validated for sports applications [28–30], after applying a low-pass,
zero-lag second order Butterworth filter (cut-off frequency: 15 Hz) to the raw marker trajectories.
CoD events were easily identified with the peaks of the CoM position in the running direction (CoMx).
Mass-specific CoM external energy was computed according to classical physiology texts [31]:
Eext = 1
2v2
CoM + ghCoM
(1)
where vCoM is the norm of instantaneous CoM speed, and hCoM is its vertical height. Positive (negative)
mechanical work (W+/−) was then obtained as the sum of positive (negative) changes of Eext [31] in the
two second window across the turn. Also, for each turn, we computed CoM approach speed (1 s before
the turn, vbefore), and CoM speed during acceleration (1 s after the turn, vafter). These four variables,
alongside the side of the pivoting limb (right or left), constitute the set of known targets (responses).
Their distribution is illustrated in Figure 2: as negative external work is required to decelerate the
CoM before the turn, and positive work is needed to accelerate it back in the new direction, W− was
represented relative to vbefore, and W+ to vafter.
Sensors 2019, 19, 3094
4 of 11
Sensors 2019, 19, x FOR PEER REVIEW
4 of 11
Figure 2. Negative (left) and positive (right) external mechanical energy computed in the two second
window across the turn as a function of running speed before (left) and after (right) the turn. The
distribution of the variables is also displayed on the edges.
Eighteen features (predictors) were extracted from IMU data. We first had to detect the same
CoD events in the IMU readings: smoothed (Butterworth, zero-lag low-pass filter, fcut = 0.5 Hz)
angular velocity around the y (vertical) axis was particularly suitable for this, as it showed clear peaks
in correspondence to turns (Figure 3). Matching between events timing was obtained by computing
the cross-correlation between the resampled and rectified gyroscope and CoM trajectory and shifting
them in time by an offset equal to the lag corresponding to maximum cross-correlation (Figure 3).
The average CoD events detection error was 0.001 ± 0.102 s. Likewise for optical data, a two second
window across this event was considered for the following computations. Static biases on each
channel of the inertial sensors were obtained with a 30 min test with the unit kept still. We then
subtracted these values from the accelerometer and gyroscope readings, which were additionally
filtered with a fourth order Butterwort filter (cut-off frequency: 128 Hz).
Figure 3. Events detection, based on center of mass (CoM) horizontal position (red curve, referred to
the origin of the laboratory global reference system) and raw/filtered gyroscope rotation around the
vertical axis (gray and blue, respectively). The autocorrelation function between the two allowed us
to synchronize the two measurement systems.
Figure 2.
Negative (left) and positive (right) external mechanical energy computed in the two
second window across the turn as a function of running speed before (left) and after (right) the turn.
The distribution of the variables is also displayed on the edges.
Eighteen features (predictors) were extracted from IMU data. We first had to detect the same CoD
events in the IMU readings: smoothed (Butterworth, zero-lag low-pass filter, fcut = 0.5 Hz) angular
velocity around the y (vertical) axis was particularly suitable for this, as it showed clear peaks in
correspondence to turns (Figure 3). Matching between events timing was obtained by computing
the cross-correlation between the resampled and rectified gyroscope and CoM trajectory and shifting
them in time by an offset equal to the lag corresponding to maximum cross-correlation (Figure 3).
The average CoD events detection error was 0.001 ± 0.102 s. Likewise for optical data, a two second
window across this event was considered for the following computations. Static biases on each channel
of the inertial sensors were obtained with a 30 min test with the unit kept still. We then subtracted
these values from the accelerometer and gyroscope readings, which were additionally filtered with a
fourth order Butterwort filter (cut-off frequency: 128 Hz).
Sensors 2019, 19, x FOR PEER REVIEW
4 of 11
Figure 2. Negative (left) and positive (right) external mechanical energy computed in the two second
window across the turn as a function of running speed before (left) and after (right) the turn. The
distribution of the variables is also displayed on the edges.
Eighteen features (predictors) were extracted from IMU data. We first had to detect the same
CoD events in the IMU readings: smoothed (Butterworth, zero-lag low-pass filter, fcut = 0.5 Hz)
angular velocity around the y (vertical) axis was particularly suitable for this, as it showed clear peaks
in correspondence to turns (Figure 3). Matching between events timing was obtained by computing
the cross-correlation between the resampled and rectified gyroscope and CoM trajectory and shifting
them in time by an offset equal to the lag corresponding to maximum cross-correlation (Figure 3).
The average CoD events detection error was 0.001 ± 0.102 s. Likewise for optical data, a two second
window across this event was considered for the following computations. Static biases on each
channel of the inertial sensors were obtained with a 30 min test with the unit kept still. We then
subtracted these values from the accelerometer and gyroscope readings, which were additionally
filtered with a fourth order Butterwort filter (cut-off frequency: 128 Hz).
Figure 3. Events detection, based on center of mass (CoM) horizontal position (red curve, referred to
the origin of the laboratory global reference system) and raw/filtered gyroscope rotation around the
vertical axis (gray and blue, respectively). The autocorrelation function between the two allowed us
to synchronize the two measurement systems.
Figure 3. Events detection, based on center of mass (CoM) horizontal position (red curve, referred to
the origin of the laboratory global reference system) and raw/filtered gyroscope rotation around the
vertical axis (gray and blue, respectively). The autocorrelation function between the two allowed us to
synchronize the two measurement systems.
Sensors 2019, 19, 3094
5 of 11
The first feature F1 was the sum of the root mean square of changes in acceleration and deceleration
per second, also known as ’player load’, a metric commonly used to calculate the load or activity level
of athletes in team sports [25]:
player load =
1023
X
t=1
r
(ax,t+1 − ax,t)2 +
ay,t+1 − ay,t
2 + (az,t+1 − az,t)2
(2)
Other accelerometer features were the trapezoidal numerical integral of each axes positive and
negative acceleration (F2−7). Features F8−10 were the integral of each gyroscope axis. Features F11−13
and F14−16 were the root mean square, the skewness and the kurtosis of the norm of the accelerometer
and gyroscope readings during the two second turn window. Skewness describes the symmetry of the
acceleration and angular velocity signal distributions and is given by:
skewness = E(x − x)3
σ3
(3)
The kurtosis quantifies the extent to which the acceleration and angular velocity signals are peaked
or flat with respect to a normal distribution:
kurtosis = E(x − x)4
σ4
(4)
In Equations (3) and (4), E is the expected value, x is the mean and σ is the standard deviation of
the signal in the two second window [32]. These two features quantify the degree of distortion with
respect to a normal distribution of acceleration or angular velocity data series: for instance, an abrupt
braking action would contain more negative than positive acceleration values, and so it would be
highly skewed.
The last two features F17 and F18 were obtained from the filtered barometer output (low-pass
tenth order Butterworth filter, cut-off frequency: 1 Hz) and were the difference between the mean sea
level altitude (computed from ambient pressure) at the CoD event and one second before or after the
CoD, respectively. Table 1 provides an overview of features and responses. Before further processing,
outliers were removed when examples were outside the variable’s mean ± 3 standard deviations,
and the coefficient of variation (CV) was computed for each variable.
2.4. Regression and Classification Models
For the prediction of W+, W−, vbefore and vafter, we applied four supervised machine learning
regression techniques:
1.
Multiple linear regression, modeling the linear relationship between predictors and the response
(dependent) variables.
2.
Support vector regression (SVR): this technique is based on support vector machines (SVM),
which in turn construct hyperplanes to define decision boundaries in a multi-dimensional space.
SVR computes the parameter of a function f(x), where x is the matrix of predictors, fitting the
input data with the most ε-deviation from the target y (response). As SVR is particularly suited to
handle non-linear tasks, in this study we chose a Gaussian kernel.
3.
Boosted trees (BT): classification or regression models are in the form of a tree structure, which
is built top-down from a root node, and involves partitioning data into subsets that contain
common features based on the level of information gain, i.e., a decrease in entropy after a dataset
is separated [33]. Boosted trees are an extension of decision trees that aggregate an ensemble
of decision trees into a unique result, which reduces the chance of overfitting. The number of
learners (trees) set in this study was 40.
Sensors 2019, 19, 3094
6 of 11
4.
Artificial neural networks (ANN): a feedforward network consisting of an input, a hidden and an
output layer was designed. Neurons (n = 40) in the hidden layer process the input features in
accordance to hyperbolic tangent sigmoid functions. The output layer is a single neuron which
returns the estimated (predicted) response. The back-propagation learning algorithm was used
to update the weights and biases of the ANN. Input data was split into three subsets: 70% for
training, 15% for testing and 15% for validation.
Classification models to determine turn direction (right or left) matched the previous techniques
and included: (1) linear discriminant analysis, (2) SVM, (3) BT, (4) ANN.
Table 1. Overview of the five responses (Ri, where R5 is a categorical variable) and of the 18 features
(Fi) used to train the models.
Variable
Description
Unit
Mean
SD
CV
Min
Max
R1
Positive work
J/kg
8.49
2.35
0.28
1.18
18.44
R2
Negative work
J/kg
8.17
2.73
0.33
1.02
18.91
R3
Speed before turn
m/s
2.60
0.38
0.15
1.49
3.69
R4
Speed after turn
m/s
2.59
0.33
0.13
1.57
3.44
R5
Turn side
cat.
-
-
-
F1
Player load
A.U.
190.6
62.6
0.33
67.3
407.4
F2
Positive accx integral
m/s
325.8
86.0
0.26
125.7
584.0
F3
Positive accy integral
m/s
528.00
96.1
0.18
267.4
846.3
F4
Positive accz integral
m/s
238.7
67.6
0.28
82.0
435.1
F5
Negative accx integral
m/s
−328.7
89.4
0.27
−587.1
−107.8
F6
Negative accy integral
m/s
−508.8
89.2
0.18
−810.1
−263.4
F6
Negative accz integral
m/s
−251.7
79.0
0.31
−504.5
−93.4
F7
Norm acc RMS
m/s2
7.59
3.57
0.47
1.19
21.04
F8
Acceleration skewness
-
4.95
1.35
0.27
2.08
9.00
F9
Acceleration kurtosis
-
35.8
18.5
0.52
7.6
98.4
F10
Gyroscopex integral
rad
9.9·103
1.4·104
1.44
−2.9·104
4.7·104
F11
Gyroscopey integral
rad
0.9·103
6.7·104
n.a.
−8.5·104
8.6·104
F12
Gyroscopez integral
rad
−45.3
3.8·104
n.a.
−6.0·104
6.1·104
F13
Gyroscope norm RMS
rad/s
2.4·103
1.7·103
0.69
0.3·103
7.9·103
F14
Gyroscope skewness
-
4.64
1.61
0.35
0.41
9.31
F15
Gyroscope kurtosis
-
32.8
20.1
0.61
2.41
104.0
F17
∆baro, before
m
−0.39
0.26
0.65
−1.08
0.36
F18
∆baro, after
m
0.41
0.26
0.62
−0.35
1.12
∆baro: altitude difference between before/after 1 s. acc: acceleration; A.U. arbitrary units; cat.: categorical variable.
CV: coefficient of variation; RMS: root mean square; SD: standard deviation; n.a.: not applicable (bimodal variable,
CV > 50).
2.5. Validation
To evaluate the predictive accuracy, models 1–3 underwent a standard 10-fold cross-validation
procedure: data were randomly partitioned into two sets, the first was used for training, while the
second was used for validation. This process was repeated 10 times, by randomly selecting the training
and validation portions [34]. The means of the 10 classification accuracy rates were taken as an unbiased
estimate of the model for the complete dataset. Root mean square error (RMSE), mean absolute error
and coefficient of determination (R2) were computed as performance metrics for regression models.
Classification models were evaluated in terms of accuracy, sensitivity, specificity and area under the
receiver operating characteristic curve (AUC).
3. Results
Th shuttle run test duration was on average 158.4 ± 65.1 s, with 72 ± 30 turns per participant.
Overall, we collected 937 cutting maneuvers. Outlier removal led us to exclude 32 of them. Feature
means, standard deviations and ranges are shown in Table 1: while the variability of speed was below
Sensors 2019, 19, 3094
7 of 11
15%, that of positive/negative mechanical work reached 33%; the highest variability in predictors was
contained by gyroscope integrals and RMS, the lowest in the acceleration integrals.
Turn direction was detected with good accuracy (98.4%, Table 2) by linear discriminant analysis
(sensitivity = 97.3%, specificity = 99.6%, AUC = 1.00), while with the other methods (boosted trees,
SVM and ANN), we obtained perfect classification (accuracy = 100%, AUC = 1.00).
The best performance in predicting the mechanical work in decelerations and accelerations was
achieved by SVR models, with a moderate R2 and an error of about 15% (Table 2). ANNs best
predicted incoming and sprint speed with a moderate to substantial R2, and an error lower than 10%
(0.15–0.18 m/s). Feature importance for BT is reported in Figure 4.
Sensors 2019, 19, x FOR PEER REVIEW
7 of 11
The best performance in predicting the mechanical work in decelerations and accelerations was
achieved by SVR models, with a moderate R2 and an error of about 15% (Table 2). ANNs best
predicted incoming and sprint speed with a moderate to substantial R2, and an error lower than 10%
(0.15–0.18 m/s). Feature importance for BT is reported in Figure 4.
Figure 4. Feature importance returned by boosted trees models when predicting positive/negative
mechanical work (Wp and Wn, top), and approach/sprint running speed (vbefore and vafter, bottom).
Refer to Table 1 for the description of features Fi.
4. Discussion
The combination of accelerometer-, gyroscope- and barometer-based features and relatively
simple machine learning models enabled us to estimate key kinematic and energetic characteristics
of 180-degree turns with an error of about 15%. Features were obtained from a single pelvis-mounted
unit, and no prior calibration procedure was required. In addition, we limited signal processing to a
few basic steps, with only standard low-pass filtering being applied to sensor readings: this
potentially increases the generalizability to different units and vendors.
Table 2. Performance of regression models. In bold, the best model for each response, in terms of root
mean square error (RMSE).
Response
Regression Model
R2
RMSE
MAE
Positive work (J)
Multilinear regression
0.36
1.96
1.30
Support vector regression
0.43
1.85
1.14
Boosted trees
0.39
1.91
1.22
Artificial neural network
0.39
2.40
1.80
Figure 4. Feature importance returned by boosted trees models when predicting positive/negative
mechanical work (Wp and Wn, top), and approach/sprint running speed (vbefore and vafter, bottom).
Refer to Table 1 for the description of features Fi.
4. Discussion
The combination of accelerometer-, gyroscope- and barometer-based features and relatively
simple machine learning models enabled us to estimate key kinematic and energetic characteristics of
180-degree turns with an error of about 15%. Features were obtained from a single pelvis-mounted
unit, and no prior calibration procedure was required. In addition, we limited signal processing to a
few basic steps, with only standard low-pass filtering being applied to sensor readings: this potentially
increases the generalizability to different units and vendors.
Sensors 2019, 19, 3094
8 of 11
Table 2. Performance of regression models. In bold, the best model for each response, in terms of root
mean square error (RMSE).
Response
Regression Model
R2
RMSE
MAE
Positive work (J)
Multilinear regression
0.36
1.96
1.30
Support vector regression
0.43
1.85
1.14
Boosted trees
0.39
1.91
1.22
Artificial neural network
0.39
2.40
1.80
Negative work (J)
Multilinear regression
0.35
2.27
1.66
Support vector regression
0.42
2.14
1.41
Boosted trees
0.41
2.16
1.49
Artificial neural network
0.44
2.34
1.84
Speed before (m/s)
Multilinear regression
0.53
0.26
0.21
Support vector regression
0.65
0.23
0.17
Boosted trees
0.60
0.24
0.19
Artificial neural network
0.66
0.23
0.18
Speed after (m/s)
Multilinear regression
0.47
0.25
0.20
Support vector regression
0.65
0.20
0.15
Boosted trees
0.62
0.21
0.16
Artificial neural network
0.69
0.19
0.15
MAE: mean absolute error; R2: coefficient of determination.
4.1. Turn Direction
As CoDs and sidestepping maneuvers impose a high muscular and mechanical load on lower limb
structures, knowing the intensity and direction of these actions might help in preventing unilateral
overloading and in turn potential injuries [2,35]. Turn side (direction) was satisfactorily estimated even
with linear discriminant analysis, which makes the adoption of BT, SVM and ANNs redundant for this
specific task.
In fact, the combination of gyroscope features F11 (integral of the angular velocity around the pelvic
anteroposterior axis) and most importantly F12 (integral around the pelvic mediolateral axis) returned
almost perfect classifications. In other words, pelvis rotation captures the most relevant information
about turning direction and could be easily implemented in existing global positioning system (GPS)
activity trackers [36]. This sensor location allows for the implementation of smart apparel, that unlike
smartwatches or wristbands, can be worn during games. A potential improvement to this model could
be the classification among multiple classes of directional changes, i.e., 45–90 degree, 90–120 degree
and 120–180 degree CoDs. Further, it is likely that the classification performance would decrease when
positioning the sensor on other less favorable—limited to turn direction classification—segments such
as feet or shanks.
4.2. Turn Speed and Mechanical Energy
Turning speed is related to the energy dissipated during the deceleration (braking) phase, and thus
impacts on the risk of ligamentous lesions, especially at the knee level [37]. Once a directional change
was detected, the proposed models were successfully able to estimate turn speed with an error below
0.2 m/s, which is comparable to 10 Hz GPS error in common sports applications [38,39]. However, rapid
directional changes usually degrade GPS accuracy [38]: obtaining speed during CoDs still represents a
challenge for such systems [40], whose weaknesses can be mitigated by adding or integrating data
from inertial sensors [36,41].
The best prediction performance of ANNs showed that a linear relationship between predictors
and response (i.e., multiple linear regression) was outperformed by non-linear models: ANNs can detect
complex nonlinear interactions between inputs and targets [42]. However, the lack of transparency in
the mathematical models of ANNs hinders the interpretation of their output. Decision trees showed
slightly lower performance, but the model structure of the entire procedure can be followed and
Sensors 2019, 19, 3094
9 of 11
interpreted: in BT models, prediction appeared highly dependent on positive mediolateral acceleration
(F4) for sprinting speed, and acceleration skewness for approach speed (Figure 4).
A similar reasoning applies to Wp and Wn, which were better predicted by SVR with Gaussian
kernel. Mechanical energy during a CoD is an indirect measure of its associated metabolic cost.
Computed positive/negative external CoM work was in line with previous investigations from different
groups [8,16], which ensures the predictions were constructed on a solid basis. The prediction error
(1.1–1.4 J) can be considered acceptable as it allows us to get a realistic measure of the amount of energy
involved in the braking/propulsive actions, which is highly dependent on running speed.
4.3. Limitations and Perspectives
Turning event detection was particularly easy in the featured experimental protocol,
which involved repeated 180-degree turns with just few running steps in between: further developments
should include various angles of approach, and potentially a less homogeneous sample to increase the
model’s generalizability. Compared to the reference optical system, CoD events were determined with
a variability of about 0.1 s, which is of the same order of magnitude of the stance phase of the pivoting
limb (~0.3 s [14]). However, we chose to compute features over a two second span; thus, any overlaps
should be in the range of 5% of the whole window. Although data were processed off-line, once a
turn is detected it would be relatively straightforward to apply the regression algorithms to moving
windows containing updated data streams.
More complex features could be also added to exploit the IMU three-dimensional orientation
(i.e., quaternions): however, (i) accurate long-term orientation tracking based on a unique inertial
sensor is not trivial and (ii) we intentionally limited feature processing to variables that could be easily
computed on a portable device. In addition, the proposed models did not rely on anthropometric
information (height, weight, BMI, etc.), that were purposefully excluded from predictions.
Even if a more general scope should be adopted for practical on-the-field implementations,
this paper showed that the extraction of meaningful information on turn kinematics and energetics
is highly viable with a data-driven approach using commercially-available units and established
regression and classification techniques. The current study represents a further step towards the
accurate, ecological quantification of the key features of changes of direction in team sports.
Author Contributions: Conceptualization, M.Z. and C.S.; methodology, M.G. and M.T.; formal analysis, M.Z.;
data curation, M.Z.; resources, C.D., C.S. and M.T.; supervision, M.G. and C.S.; project administration, C.D.;
writing—original draft preparation, M.Z.; writing—review and editing, M.G., C.S. and M.T.
Funding: University of Milan (Piano di sostegno alla ricerca 2015–17).
Acknowledgments: The Authors are grateful to Filippo Bertozzi and Francesca Salaorni for the precious
collaboration and the recruitment of participants, respectively.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Hader, K.; Mendez-Villanueva, A.; Ahmaidi, S.; Williams, B.K.; Buchheit, M. Changes of direction during
high-intensity intermittent runs: Neuromuscular and metabolic responses. BMC Sports Sci. Med. Rehabil.
2014, 11, e0149839. [CrossRef] [PubMed]
2.
Alentorn-Geli, E.; Myer, G.D.; Silvers, H.J.; Samitier, G.; Romero, D.; Lazaro-Haro, C.; Cugat, R.R.;
Lázaro-Haro, C.; Cugat, R.R. Prevention of non-contact anterior cruciate ligament injuries in soccer players.
Part 1: Mechanisms of injury and underlying risk factors. Knee Surg. Sports Traumatol. Arhrosc. 2009, 17,
705–729. [CrossRef] [PubMed]
3.
Besier, T.F.; Lloyd, D.G.; Ackland, T.R. Muscle activation strategies at the knee during running and cutting
maneuvers. Med. Sci. Sports Exerc. 2003, 35, 119–127. [CrossRef]
4.
Read, P.J.; Oliver, J.L.; De Ste Croix, M.B.A.; Myer, G.D.; Lloyd, R.S. Neuromuscular Risk Factors for Knee
and Ankle Ligament Injuries in Male Youth Soccer Players. Sports Med. 2016, 46, 1059–1066. [CrossRef]
[PubMed]
Sensors 2019, 19, 3094
10 of 11
5.
Vanrenterghem, J.; Venables, E.; Pataky, T.; Robinson, M.A. The effect of running speed on knee mechanical
loading in females during side cutting. J. Biomech. 2012, 45, 2444–2449. [CrossRef] [PubMed]
6.
McLean, S.G.; Huang, X.; Van Den Bogert, A.J. Association between lower extremity posture at contact
and peak knee valgus moment during sidestepping: Implications for ACL injury. Clin. Biomech. 2005, 20,
863–870. [CrossRef]
7.
Ciprandi, D.; Lovecchio, N.; Piacenza, M.; Limonta, E.; Esposito, F.; Sforza, C.; Zago, M. Energy Cost of
Continuous Shuttle Running. J. Strength Cond. Res. 2017, 32, 2265–2272. [CrossRef]
8.
Zamparo, P.; Bolomini, F.; Nardello, F.; Beato, M. Energetics (and kinematics) of short shuttle runs. Eur. J.
Appl. Physiol. 2015, 115, 1985–1994. [CrossRef]
9.
Cortes, N.; Onate, J.; van Lunen, B. Pivot task increases knee frontal plane loading compared with sidestep
and drop-jump. J. Sports Sci. 2011, 29, 83–92. [CrossRef]
10.
Zago, M.; Codari, M.; Grilli, M.; Bellistri, G.; Lovecchio, N.; Sforza, C. Determinants of the half-turn with the
ball in sub-elite youth soccer players. Sports Biomech. 2016, 15, 234–244. [CrossRef]
11.
Zago, M.; Piovan, A.G.; Annoni, I.; Ciprandi, D.; Iaia, F.M.; Sforza, C. Dribbling determinants in sub-elite
youth soccer players. J. Sports Sci. 2016, 34, 411–419. [CrossRef] [PubMed]
12.
David, S.; Mundt, M.; Komnik, I.; Potthast, W. Understanding cutting maneuvers—The mechanical
consequence of preparatory strategies and foot strike pattern. Hum. Mov. Sci. 2018, 62, 202–210. [CrossRef]
[PubMed]
13.
Maniar, N.; Schache, A.G.; Cole, M.H.; Opar, D.A. Lower-limb muscle function during sidestep cutting.
J. Biomech. 2019, 82, 186–192. [CrossRef] [PubMed]
14.
Zago, M.; Esposito, F.; Bertozzi, F.; Tritto, B.; Rampichini, S.; Galvani, C.; Galli, M.; Sforza, C. Kinematic
effects of repeated turns while running. Eur. J. Sport Sci. 2019, 5, 1–10. [CrossRef] [PubMed]
15.
Cortes, N.; Greska, E.; Kollock, R.; Ambegaonkar, J.; Onate, J.A. Changes in lower extremity biomechanics
due to a short-term fatigue protocol. J. Athl. Train. 2013, 48, 306–313. [CrossRef]
16.
Zago, M.; Esposito, F.; Rausa, G.; Limonta, E.; Corrado, F.; Rampichini, S.; Sforza, C. Kinematic algorithm to
determine the energy cost of running with changes of direction. J. Biomech. 2018, 76, 189–196. [CrossRef]
[PubMed]
17.
Camomilla, V.; Bergamini, E.; Fantozzi, S.; Vannozzi, G. Trends supporting the in-field use of wearable
inertial sensors for sport performance evaluation: A systematic review. Sensors 2018, 18, 873. [CrossRef]
18.
Cust, E.E.; Sweeting, A.J.; Ball, K.; Robertson, S. Machine and deep learning for sport-specific movement
recognition: A systematic review of model development and performance. J. Sports Sci. 2019, 37, 568–600.
[CrossRef]
19.
Mukhopadhyay, S.C. Wearable sensors for human activity monitoring: A review. IEEE Sens. J. 2015, 15,
1321–1330. [CrossRef]
20.
Aroganam, G.; Manivannan, N.; Harrison, D. Review on Wearable Technology Sensors Used in Consumer
Sport Applications. Sensors 2019, 19, 1983. [CrossRef]
21.
Ahamed, N.U.; Benson, L.C.; Clermont, C.A.; Pohl, A.J.; Ferber, R. New Considerations for Collecting
Biomechanical Data Using Wearable Sensors: How Does Inclination Influence the Number of Runs Needed
to Determine a Stable Running Gait Pattern? Sensors 2019, 19, 2516. [CrossRef] [PubMed]
22.
Kim, W.; Kim, M. On-line detection and segmentation of sports motions using a wearable sensor. Sensors
2018, 18, 913. [CrossRef] [PubMed]
23.
Zhang, Y.; Ma, Y. Application of supervised machine learning algorithms in the classification of sagittal gait
patterns of cerebral palsy children with spastic diplegia. Comput. Biol. Med. 2019, 106, 33–39. [CrossRef]
[PubMed]
24.
Gastin, P.B.; Mclean, O.C.; Breed, R.V.P.; Spittle, M. Tackle and impact detection in elite Australian football
using wearable microsensor technology. J. Sports Sci. 2014, 32, 947–953. [CrossRef] [PubMed]
25.
Dalen, T.; JØrgen, I.; Gertjan, E.; Havard, H.G.; Ulrik, W. Player load, acceleration, and deceleration during
forty-five competitive matches of elite soccer. J. Strength Cond. Res. 2016, 30, 351–359. [CrossRef] [PubMed]
26.
Nedergaard, N.J.; Kersting, U.; Lake, M. Using accelerometry to quantify deceleration during a high-intensity
soccer turning manoeuvre. J. Sports Sci. 2014, 32, 1897–1905. [CrossRef] [PubMed]
27.
Bangsbo, J.; Iaia, F.M.; Krustrup, P. The Yo-Yo intermittent recovery test: A useful tool for evaluation of
physical performance in intermittent sports. Sports Med. 2008, 38, 37–51. [CrossRef]
Sensors 2019, 19, 3094
11 of 11
28.
Mapelli, A.; Zago, M.; Fusini, L.; Galante, D.; Colombo, A.; Sforza, C. Validation of a protocol for the
estimation of three-dimensional body center of mass kinematics in sport. Gait Posture 2014, 39, 460–465.
[CrossRef]
29.
Zago, M.; Motta, A.F.; Mapelli, A.; Annoni, I.; Galvani, C.; Sforza, C. Effect of leg dominance on the
center-of-mass kinematics during an inside-of-the-foot kick in amateur soccer players. J. Hum. Kinet. 2014,
42, 51–61. [CrossRef]
30.
Zago, M.; Mapelli, A.; Shirai, Y.F.; Ciprandi, D.; Lovecchio, N.; Galvani, C.; Sforza, C. Dynamic balance in
elite karateka. J. Electromyogr. Kinesiol. 2015, 25, 894–900. [CrossRef]
31.
Willems, P.A.; Cavagna, G.A.; Heglund, N.C.; Umana, F.; Chelmsford, S. External, internal and total work in
human locomotion. J. Exp. Biol. 1995, 393, 379–393.
32.
Wang, W.H.; Hsu, Y.L.; Chung, P.C.; Pai, M.C. Predictive Models for Evaluating Cognitive Ability in Dementia
Diagnosis Applications Based on Inertia-and Gait-Related Parameters. IEEE Sens. J. 2018, 18, 3338–3350.
[CrossRef]
33.
Safavian, S.R.; Landgrebe, D. A Survey of Decision Tree Classifier Methodology. IEEE Trans. Syst. Man
Cybern. 1991, 21, 660–674. [CrossRef]
34.
Little, M.A.; Varoquaux, G.; Saeb, S.; Lonini, L.; Jayaraman, A.; Mohr, D.C.; Kording, K.P. Using and
understanding cross-validation strategies. Perspectives on Saeb et al. Gigascience 2017, 5, gix020. [CrossRef]
[PubMed]
35.
Stevens, T.G.A.; De Ruiter, C.J.; Van Maurik, D.; Van Lierop, C.J.W.; Savelsbergh, G.J.P.; Beek, P.J. Measured
and estimated energy cost of constant and shuttle running in soccer players. Med. Sci. Sports Exerc. 2015, 47,
1219–1224. [CrossRef] [PubMed]
36.
Mertens, J.C.; Boschmann, A.; Schmidt, M.; Plessl, C. Sprint diagnostic with GPS and inertial sensor fusion.
Sports Eng. 2018, 21, 441–451. [CrossRef]
37.
Besier, T.F.; Lloyd, D.G.; Cochrane, J.L.; Ackland, T.R. External loading of the knee joint during running and
cutting maneuvers. Med. Sci. Sports Exerc. 2001, 33, 1168–1175. [CrossRef]
38.
Beato, M.; Bartolini, D.; Ghia, G.; Zamparo, P. Accuracy of a 10 Hz GPS Unit in Measuring Shuttle Velocity
Performed at Different Speeds and Distances (5–20 M). J. Hum. Kinet. 2016, 54, 15–22. [CrossRef]
39.
Varley, M.C.; Fairweather, I.H.; Aughey, R.J. Validity and reliability of GPS for measuring instantaneous
velocity during acceleration, deceleration, and constant motion. J. Sports Sci. 2012, 30, 121–127. [CrossRef]
40.
Rawstorn, J.C.; Maddison, R.; Ali, A.; Foskett, A.; Gant, N. Rapid directional change degrades GPS distance
measurement validity during intermittent intensity running. PLoS ONE 2014, 9, e93693. [CrossRef]
41.
Aughey, R.J. Applications of GPS Technologies to Field Sports Robert. Int. J. Sports Physiol. Perform. 2011, 6,
295–310. [CrossRef] [PubMed]
42.
Chau, T. A review of analytical techniques for gait data. Part 2: Neural network and wavelet methods.
Gait Posture 2001, 13, 102–120. [CrossRef]
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
| Use of Machine Learning and Wearable Sensors to Predict Energetics and Kinematics of Cutting Maneuvers. | 07-12-2019 | Zago, Matteo,Sforza, Chiarella,Dolci, Claudia,Tarabini, Marco,Galli, Manuela | eng |
PMC8523042 | Sedentary
Step length (m)
Cadence (steps/s)
S4 Fig
0.5
1.0
1.5
2.0
2.5
2
3
4
5
1 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
2 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
3 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
16 (F)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
5 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
6 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
7 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
18 (F)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
9 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
10 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
11 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
17 (F)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
8 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
14 (F)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
19 (F)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
20 (F)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
15 (F)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
12 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
4 (M)
0.5
1.0
1.5
2.0
2.5
2
3
4
5
13 (M)
| Spatiotemporal inflection points in human running: Effects of training level and athletic modality. | 10-18-2021 | Goto, Yuta,Ogawa, Tetsuya,Kakehata, Gaku,Sazuka, Naoya,Okubo, Atsushi,Wakita, Yoshihiro,Iso, Shigeo,Kanosue, Kazuyuki | eng |
PMC5330462 | RESEARCH ARTICLE
Oxygen uptake kinetics and energy system’s
contribution around maximal lactate steady
state swimming intensity
Jailton Grego´rio Pelarigo1,2,3,4☯*, Leandro Machado3,4‡, Ricardo Jorge Fernandes3,4‡,
Camila Coelho Greco5‡, João Paulo Vilas-Boas3,4☯
1 University Catholic Center of Quixada´–UNICATO´ LICA, Quixada´, Ceara´, Brazil, 2 Metropolitan College of
Grande Fortaleza–FAMETRO, Fortaleza, Ceara´, Brazil, 3 Centre of Research, Education, Innovation and
Intervention in Sport, Faculty of Sport, University of Porto, Porto, Portugal, 4 Porto Biomechanics Laboratory,
LABIOMEP, University of Porto, Porto, Portugal, 5 Human Performance Laboratory, Physical Education
Department, São Paulo State University, Rio Claro, São Paulo, Brazil
☯ These authors contributed equally to this work.
‡ These authors also contributed equally to this work.
* [email protected]
Abstract
The purpose of this study was to examine the oxygen uptake ( _VO2) kinetics and the energy
systems’ contribution at 97.5, 100 and 102.5% of the maximal lactate steady state (MLSS)
swimming intensity. Ten elite female swimmers performed three-to-five 30 min submaximal
constant swimming bouts at imposed paces for the determination of the swimming velocity
(v) at 100%MLSS based on a 7 x 200 m intermittent incremental protocol until voluntary
exhaustion to find the v associated at the individual anaerobic threshold. _VO2 kinetics (cardi-
odynamic, primary and slow component phases) and the aerobic and anaerobic energy con-
tributions were assessed during the continuous exercises, which the former was studied for
the beginning and second phase of exercise. Subjects showed similar time delay (TD)
(mean = 11.5–14.3 s) and time constant (τp) (mean = 13.8–16.3 s) as a function of v, but
reduced amplitude of the primary component for 97.5% (35.7 ± 7.3 mL.kg.min-1) compared
to 100 and 102.5%MLSS (41.0 ± 7.0 and 41.3 ± 5.4 mL.kg.min-1, respectively), and τp
decreased (mean = 9.6–10.8 s) during the second phase of exercise. Despite the slow com-
ponent did not occur for all swimmers at all swim intensities, when observed it tended to
increase as a function of v. Moreover, the total energy contribution was almost exclusively
aerobic (98–99%) at 97.5, 100 and 102.5%MLSS. We suggest that well-trained endurance
swimmers with a fast TD and τp values may be able to adjust faster the physiological
requirements to minimize the amplitude of the slow component appearance, parameter
associated with the fatigue delay and increase in exhaustion time during performance, how-
ever, these fast adjustments were not able to control the progressive fatigue occurred
slightly above MLSS, and most of swimmers reached exhaustion before 30min swam.
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
1 / 12
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Pelarigo JG, Machado L, Fernandes RJ,
Greco CC, Vilas-Boas JP (2017) Oxygen uptake
kinetics and energy system’s contribution around
maximal lactate steady state swimming intensity.
PLoS ONE 12(2): e0167263. doi:10.1371/journal.
pone.0167263
Editor: Juan Sastre, Universitat de Valencia, SPAIN
Received: February 16, 2016
Accepted: November 11, 2016
Published: February 28, 2017
Copyright: © 2017 Pelarigo et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All relevant data are
within the paper and its Supporting Information
files.
Funding: This investigation was supported by
grants of the Capes Foundation, Ministry of
Education of Brazil (BEX: 0536/10-5): JGP and
project PTDC/DES/101224/2008 (FCOMP-01-
0124-FEDER-009577): JPVB RF.
Competing interests: The authors have declared
that no competing interests exist.
Introduction
An important aspect of aerobic endurance performance is the ability to sustain the highest per-
centage of maximal oxygen uptake (% _VO2max) as long as possible. In this sense, coaches and
swimmers have used the % _VO2max in different submaximal intensities to control, prescribe
and improve sports training [1]. Additionally, scientists have shown that the _VO2 kinetics
analysis may help to understand the physiological adjustments produced over time by the ath-
letes in several sports, allowing them to maintain a high % _VO2max in a physiological steady-
state during aerobic endurance performance [2–4].
Meanwhile, the scientific community has mainly described the _VO2 kinetics in three differ-
ent intensity domains during continuous exercise. First, the moderate domain is described as
the exercise intensities in which a state steady for _VO2 is achieved within 3 min of constant
exercise [5]. Subsequently, the heavy domain is described as the exercise intensities in which
_VO2 slow component should be evident, causing a delay on the achievement of the _VO2
steady-state during exercise [2]. Last, the severe domain is described as the exercise intensities
in which _VO2 is elevated compared to rest values and continue to increase over time, leading
to attain the _VO2max [6, 7].
Maximal lactate steady state (MLSS) is considered one of the main relevant parameters for
prescription and improvement of aerobic endurance performance, once it has been assumed
as the limit intensity at which, during prolonged and submaximal exercise, the metabolic
energy is produced mainly by the aerobic metabolism of pyruvate and glycolysis [8, 9]. More-
over, MLSS is identified as the maximal intensity that can be maintained over time without the
lactate production exceeding removal more than 1 mmol.L-1, and considered gold-standard
method for the evaluation of aerobic capacity [10–12].
Once maximal velocity where a steady-state is found represents a fundamental physiologi-
cal border, subtle changes in this intensity could likely modify _VO2 kinetics response. For
instance, when the exercise is performed at intensities slightly below MLSS, a physiological
steady state is sustained for both blood lactate concentration [La-] and _VO2 as a function of
time [6, 7, 13]. On the other hand, at intensities above the MLSS, a significant increase in [La-]
and _VO2 is likely to be observed throughout time [3, 7, 8, 12], leading to fatigue and voluntary
exhaustion [3, 4, 14]. Moreover, the swimming MLSS determination needs a short time of
interruption for the blood collection during the 10th minute of exercise for the analysis of [La-
], and then, a resumption of exercise to complete the test. Thus, it seems to be fundamental to
examine the behavior of _VO2 kinetics not only the beginning of exercise, but too after the
resumption of exercise throughout exercise to better understanding of the entire process of the
swimmer physiological response along the exercise.
_VO2 kinetics has been studied in different sports over the last decades [2, 6, 15], and there
are relevant number of researches based on [La-] and gas exchange at intensities related to
MLSS [8, 13, 14]. However, no study has evaluated _VO2 kinetics at (and around) the MLSS
intensity. Thus, our purpose was to examine _VO2 kinetics and the energy systems’ contribu-
tion at 97.5, 100 and 102.5%MLSS in swimming. It was hypothesized that at 97.5%MLSS, _VO2
kinetics adjustments may not be so evident such as 100 and 102.5%MLSS. It was further
hypothesized that even at the 100%MLSS intensity, swimmers may also have to adjust _VO2
kinetics during the exercise, once this intensity would lead to voluntary exhaustion over time.
On the other hand, at the intensity of 102.5%MLSS, _VO2 kinetics may be compromised by
fatigue, requiring faster time adjustments for time delay and time constant, and higher _VO2
amplitudes either for primary or slow components compared to lower exercise intensities. We
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
2 / 12
further intended to assess _VO2 kinetics of the second phase of exercise, starting after the collec-
tion of [La-] and resumption of exercise (from 10th min to the exercise end—final exercise),
hypothesizing that these parameters could be faster than without previous exercise. Moreover,
as MLSS may be maintained for long time period without continuous [La-] accumulation, as
well as a submaximal exercise, the energy supply should be mainly supported through the aer-
obic system for the swimming intensities of ± 2.5% around MLSS.
Material and methods
Ten elite female swimmers volunteered and gave written informed consent (or parent/guard-
ian when subjects were under 18yrs) to participate in the present study, which was approved
by the Ethics Committee of Faculty of Sport from the University of Porto and performed
according to the Declaration of Helsinki. The swimmers were (mean ± SD) 17.6 ± 1.9 years of
age, 1.70 ± 0.05 m height, 61.3 ± 5.8 kg body mass, 15.5 ± 2.9% body fat mass, and 54.9 ± 6.7
mL.kg.min-1 _VO2max, specialized in middle- and long-distance swimming events. The subjects
had, at the least, seven years of experience as competitive swimmers and their mean perfor-
mance over a 400m freestyle swim was 88.0 ± 3.4% of the short course word record.
The test sessions were performed in a 25 m indoor swimming pool. Air humidity was main-
tained nominally between 40–60%, and pool water temperature between 27–28˚C. Swimmers
were advised to refrain from intense training at least 24 h before the experimental sessions.
The tests were conducted within a seven day period, at the same time of the day (± 2 h), mini-
mizing the circadian rhythm effects. Previously to the test sessions, swimmers performed a
1000 m warm-up at low/moderate intensity. The tests were performed in front crawl, with in-
water starts and open turns, without relevant underwater glides. A 24 h interval was imposed
between all tests.
Initially, swimmers performed an intermittent incremental protocol until voluntary
exhaustion to find the velocity (v) corresponding to the individual anaerobic threshold
(IAnT). The distance covered in each step was 200 m, with v increases of 0.05 m.s-1 and 30 s
rest intervals between each swim [16]. According to these authors, the predetermined v of the
last step was defined as the currently best expected performance for the subjects’ 400 m front
crawl, and then used to define all the v steps for the incremental test. The IAnT was assessed
by the relationship between [La-] and v using a curve fitting method, and considered the inter-
ception point between linear and exponential regressions to determine the accurate v where
[La-] increased exponentially [16, 17].
Subsequently, each swimmer performed three-to-five 30 min submaximal constant swim-
ming bouts at imposed paces to determine the highest v where a MLSS was achieved (100%
MLSS). The first trial was performed at the v corresponding to IAnT; and, if a steady state or a
decrease in [La-] was observed, further subsequent trials with 2.5% higher velocities were per-
formed until no [La-] steady state could be maintained [14]. Following this study, if the first
trial resulted in a clearly identifiable increase of the [La-], and/or could not be sustained due to
exhaustion, further trials were conducted with reduced velocities. MLSS was defined as the
[La-] that increased by no more than 1 mmol.l-1 between the 10th and 30th min of the test [9].
Earlobe capillary blood samples (5 μL) were collected: (a) at rest and in the first 30 s after
each step of the incremental test, immediately after exhaustion, and at each 2 min of recovery
(until the [La-] recovery peak was found); and (b) at rest, 10 and 30th min (or voluntary
exhaustion) of each continuous bout (Lactate Pro, Arkray, Inc., Kyoto, Japan).
The v was set and maintained using a visual underwater pacer (GBK-Pacer, GBK Electron-
ics, Aveiro, Portugal), with lights located each 2.5 m apart by a light strip on the bottom of the
pool. Swimmers followed the flashing lights to maintain the predetermined velocities. and
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
3 / 12
were instructed to keep their heads above each visual signal. Exhaustion was defined when the
swimmers remained 5 m behind the lights.
_VO2 was measured by a telemetric portable gas analyzer (K4b2, Cosmed, Italy) in both
tests, connected to the swimmer by a low hydrodynamic resistance respiratory snorkel and
valve system (New AquaTrainer1, Cosmed, Italy). This system has been previously validated
[18] and used in similar studies [15]. The device was calibrated for minute ventilation ( _VE)
with a calibrated syringe (3 L) and the O2 and CO2 analyzers with standard calibration gases
(16% O2 and 5% CO2) before each test. In all tests, _VO2 data were analyzed and errant breaths
occurred by swallow water and/or saliva, sighs and coughs were excluded. Afterwards, _VO2
values were measured in mean ± 3 SD and outside values were removed. Subsequently, the
breath-by-breath data were linearly interpolated to provide five-by-five s values, and smoothed
using three breath averages [15, 19]. Heart rate (HR) was monitored and registered continu-
ously by a HR monitor system (Polar Vantage NV, Polar electro Oy, Kempele, Finland) and
transferred in real time, through a telemetric signal, to the K4b2 device. The HR values were
also averaged every 5 s intervals.
The average _VO2 values were analyzed by a nonlinear least squares algorithm to fit the data
through MatLab 7.0 Software (MathWorks, Natick, MA). The mathematical model consisted
of two (cardiodynamic and primary components) or three (cardiodynamic, primary and slow
components) exponential models. An F-Test (p < 0.05) was used to evaluate whether the two
or three exponentials models provided the best fit to each data set.
_VO2 ðtÞ ¼ _VO2 baseline þ Ac ½1 using multivariate ANOVA and examined by the intensity and previous exercise effects. The v
and [La-] values were performed using the univariate ANOVA. All analyses were conducted
for repeated measures, complemented with the Bonferroni correction post-hoc test with a sig-
nificance level of p < 0.05.
Results
All swimmers performed 30 min when swimming at 97.5 and 100%MLSS, but eight swimmers
were not able to maintain the predetermined v during 30 min at 102.5%MLSS, reaching volun-
tary exhaustion at 19.3 ± 4.9 min. The average v and % _VO2max values were different in between
the three swim intensities, with 97.5%MLSS slowest and lowest, and 102.5%MLSS fastest and
highest (F2,18 = 2560.200, p < 0.001, p
2 = 0.996; F2,18 = 15.538, p < 0.001, p
2 = 0.633, respec-
tively) (Table 1). [La-] and HR values for the three swim intensities are also shown in Table 1
with a higher values at 102.5%MLSS compared to 97.5 and 100%MLSS for [La-] (F2,18 =
18.123, p < 0.001, p
2 = 0.668), and at 102.5%MLSS compared to 97.5%MLSS for HR (F2,18 =
7.222, p < 0.005, p
2 = 0.445).
_VO2 kinetics parameters are presented in Table 2. Ap tended to increase with the swimming
intensity (v) during the initial exercise, but differences were only noticed comparing 100 and
102.5%MLSS to 97.5%MLSS (F2,18 = 8.249, p < 0.05, p
2 = 0.478). Meanwhile, Ap was similar at
final exercise for the three swim conditions (F2,18 = 1.167, p = 0.334, p
2 = 0.115). On the other
hand, Ap decreased as a function of previous exercise for the three swims bouts. TDp, τp and
MRT were similar as function of v at initial exercise and final exercise during the three swim-
ming conditions. However, when analyzed the swim bouts as a function of previous exercise,
TDp decreased for the 97.5%MLSS, but the values remained similar for 100 and 102.5%MLSS;
τp decreased for all swim intensities, and MRT decreased for the 97.5 and 102.5%MLSS, but
remained similar for 100%MLSS.
The both measured _VO2baseline at initial exercise (F2,18 = 2.389, p = 0.120, p
2 = 0.210) and
final exercise (F2,18 = 1.034, p = 0.376, p
2 = 0.103) were similar in between the three swim con-
ditions, but _VO2baseline increased as a function of previous exercise (initial to final exercise) for
all continuous intensities (F1,9 = 68.311, p < 0.001, p
2 = 0.884). Ac was similar as a function of v
for both initial exercise (F2,18 = 0.134, p = 0.876, p
2 = 0.015) and final exercise (F2,18 = 1.974,
p = 0.168, p
2 = 0.180). Moreover, at 97.5%MLSS, Ac was lower comparing initial and final exer-
cise, but values remained similar for 100 and 102.5%MLSS.
As of _VO2 kinetics was observed for all tested swimming intensities and testing phases (ini-
tial and final exercise) only in two out of ten subjects. In one subject As was not observed. The
As was observed for 6 swimmers during initial exercise and 8 swimmers during final exercise
at 97.5%MLSS, for 6 swimmers during initial exercise and 7 swimmers during final exercise at
100%MLSS, and for 9 swimmers during initial exercise and 5 swimmers during final exercise
Table 1. Mean (SD) values of swimming velocity (v), blood lactate concentrations ([La-]), heart rate
(HR), and percentage of maximal oxygen uptake (%VO2max) are shown at 97.5, 100 and 102.5% of the
maximal lactate steady state (MLSS) (N = 10).
97.5%MLSS
100%MLSS
102.5%MLSS
v (m.s-1)
1.21 (0.07)
1.24 (0.07)1
1.27 (0.07)1,2
[La-] (mmol.L-1)
1.48 (0.39)
1.89 (0.77)
2.97 (0.87)1,2
HR (beats.min-1)
167.1 (15.0)
173.6 (9.7)
179.3 (9.2)1
% _VO2max (%)
78.9 (8.7)
84.7 (3.8)1
90.9 (4.6)1,2
1,2 Values different from 97.5 and 100%MLSS, respectively (p < 0.05).
doi:10.1371/journal.pone.0167263.t001
_
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
5 / 12
at 102.5%MLSS. The As values are presented in Table 3. As tended to increase with swimming
intensity during initial exercise, but keeping constant during final exercise whatever the inten-
sity considered; however no statistical analysis was applied, once the occurrence of the As was
apparently chaotic among swimmers both considering swimming intensities and phases of
testing (initial and final exercise).
The relative energy contribution for each one of the three swim intensity bouts is shown in
Fig 1. The aerobic energy contribution decreased (F2,18 = 15.254, p < 0.001, p
2 = 0.629) and the
anaerobic energy increased (F2,18 = 15.254, p < 0.001, p
2 = 0.629) at 102.5%MLSS compared to
97.5 and 100%MLSS.
Discussion
The purposes of this study were to examine the _VO2 kinetics responses during constant-veloc-
ity swims at intensities of 97.5, 100 and 102.5%MLSS, the effect of previous exercise on the
parameters of _VO2 kinetics, and the contribution of the energetic systems at the three
Table 2. Mean (SD) values of VO2 kinetics parameters at velocities of 97.5, 100 and 102.5% of the maximal lactate steady state (MLSS) for the
beginning of exercise until the break of swim for blood collection (initial exercise), and the second phase of exercise, starting after blood collec-
tion (final exercise) (N = 10).
97.5%MLSS
100%MLSS
102.5%MLSS
Initial exercise
Final exercise
Initial exercise
Final exercise
Initial exercise
Final exercise
VO2 baseline (mL.kg-1.min-1)
7.2 (2.1)
16.0 (5.3)a
6.0 (1.0)
17.4 (5.7)a
6.4 (0.8)
18.8 (5.8)a
Ac (mL.kg-1.min-1)
16.4 (5.9)
10.4 (4.9)a
16.1 (7.1)
14.2 (5.4)
15.1 (6.5)
14.9 (5.7)
Ap (mL.kg-1.min-1)
35.7 (7.3)
26.3 (7.4)a
41.0 (7.0)1
28.3 (5.2)a
41.3 (5.4)1
29.8 (5.5)a
TDp (s)
14.3 (5.5)
12.0 (5.3)a
12.4 (8.1)
11.9 (4.9)
11.5 (6.8)
11.1 (4.7)
τp (s)
16.3 (5.4)
10.8 (4.7)a
13.8 (4.5)
9.7 (4.5)a
16.0 (5.8)
9.6 (5.3)a
MRT (s)
30.6 (5.2)
22.8 (5.4)a
26.2 (6.8)
21.6 (4.6)
27.4 (8.5)
20.7 (5.2)a
Statistical analyses were described by intensity and previous exercise effect.
1 Values different from 97.5%MLSS for initial exercise.
a Values different from initial exercise (p < 0.05).
doi:10.1371/journal.pone.0167263.t002
Table 3. Individual and mean (SD) values of the amplitude of slow component (As) at velocities of 97.5, 100 and 102.5% of the maximal lactate
steady state (MLSS) for the beginning of exercise until the break of swim for blood collection (initial exercise), and the second phase of exercise,
starting after blood collection (final exercise) (N = 10).
As (mL.kg-1.min-1)
97.5%MLSS
100%MLSS
102.5%MLSS
swimmer
Initial exercise
Final exercise
Initial exercise
Final exercise
Initial exercise
Final exercise
1
1.7
2.9
2.3
3.8
4.5
1.6
2
2.3
0.7
4.4
0
3.7
0
3
1.1
0
2.6
0
4.4
0
4
0
0
0
0
0
0
5
4.2
0.9
2.8
0.8
2.9
0
6
0
0.8
0
0.8
1.9
2.2
7
0
1.2
0
0.9
7.2
1.1
8
0
1.7
2.8
1.1
4.5
0
9
2.5
1.3
2.6
1.5
6.1
0.8
10
1.4
0
0
1.5
5.1
1.8
Mean (SD)
2.2 (1.1)
1.4 (0.8)
2.9 (0.8)
1.5 (1.1)
4.5 (1.6)
1.5 (0.6)
doi:10.1371/journal.pone.0167263.t003
_
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
6 / 12
conditions. The main original findings were that increasing exercise intensity resulted in
greater primary amplitude of the _VO2 kinetics, in accordance with previous results in running
[22]. As demonstrated by other studies [23–25], the previous exercise may increase the ampli-
tude of the primary component and accelerate _VO2 kinetics (i.e., MRT) during the subsequent
exercise. There was a significant increase in the anaerobic contribution when swimming above
MLSS. However, the aerobic energetic system contribution corresponded to ~99% of the total
energy demand of the exercise in all exercise conditions analyzed in this study.
In sports science, _VO2 kinetics have added the understanding of physiological adjustments
over time [2–4], such as muscle metabolism and systemic oxygen transport [26]. Moreover,
one of the most relevant exercise intensities in swimming for aerobic training, prescription
and evaluation is the v at which MLSS is obtained, being considered the direct and gold-stan-
dard method for the evaluation of aerobic capacity [8, 10–12, 14]. Thus, both aspects ( _VO2
kinetics and MLSS) are decisive for the understanding of energy supply and oxidative metabo-
lism supporting muscular exercise. Therefore, our purpose was to examine the amplitude and
time adjustments of _VO2 kinetics during swims at intensities of 97.5, 100 and 102.5%MLSS,
exploring the effects of small prescriptions variations on swimming oxidative physiology.
The main findings were: (a) Ap tended to increase with swimming v for the initial phase of
exercise, despite differences were only noticed comparing 100 and 102.5%MLSS to 97.5%
MLSS. Meanwhile, Ap was similar at the final phase of exercise during the three swim condi-
tions. However, Ap decreased as a function of previous exercise for the three swim intensities;
(b) TDp, τp and MRT were similar irrespective of v both at initial and final exercise; (c) regard-
ing the effect of previous exercise comparing initial and final exercise for the three swimming
intensities, TDp decreased for the 97.5%MLSS, but was similar for 100 and 102.5%MLSS, τp
decreased for all swim intensities, and MRT decreased for the 97.5 and 102.5%MLSS, but was
similar for 100%MLSS; (d) although As was not evident for all swimmers during the three
swimming conditions, it tended to increase with intensity during initial exercise, remaining
constant during final exercise; (e) Ac was similar both for the initial and final exercise compar-
ing the three swim intensities, but was lower during final exercise compared to initial exercise
at 97.5%MLSS, and was similar at 100 and 102.5%MLSS; (f) aerobic and anaerobic energy con-
tributions were different at 102.5%MLSS compared to lower swim velocities; (g) at the three
Fig 1. Mean ± SD of aerobic and anaerobic energy relative contribution values at velocities corresponding to 97.5, 100
and 102.5% of the maximal lactate steady state (MLSS).
doi:10.1371/journal.pone.0167263.g001
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
7 / 12
swim intensities, the aerobic contribution values were higher than 98% of the total energy
input.
The _VO2 values in the present study were directly measured breath-by-breath throughout
time for the three swim intensities. Subsequently, the _VO2 data were fitted through mathemat-
ical modelling as previously applied in swimming for maximal and submaximal exercises [15,
19, 27–30]. Some studies have reported _VO2 kinetics at intensities near the maximal v where a
steady state in swimming is found (MLSS) [27–29], however we are unaware of a study that
has evaluated and compared _VO2 kinetics at or around the MLSS in swimming. Most of previ-
ous studies reported in sports science [2, 15, 19, 27–31] have studied _VO2 kinetics at maximal
and submaximal intensities, demonstrating the fundamental role of _VO2 kinetics to under-
stand the physiological mechanisms underpinning the dynamics of the aerobic response at dif-
ferent exercise intensities. Thus, the understanding of the _VO2 kinetics throughout time may
aid the evaluation of aerobic capacity and prescription of specific training sets during these
fundamental training intensities around MLSS.
The 100%MLSS v values reported in this study are in accordance with those reported in
previous ones [13, 14, 32], in spite of the fact that most of the swimmers examined in the previ-
ous studies were male when compared with the female subjects of the present study. Despite
higher v values at a given relative intensity are expected to be higher for male than female
counterparts of similar training level [33], the sex similitude comparing our results with litera-
ture could likely be explained by a higher technical and biomechanical proficiency of our
female swimmers when compared to the male swimmers of the previous studies. Indeed, the
% _VO2max at 100%MLSS (85 ± 4%) observed in the present study for women is similar to previ-
ously reported data for men (86.1% _VO2peak) [34], suggesting similar levels of aerobic capacity
development, even the _VO2max= _VO2peak being higher in the previous study (mean = ~83 mL.
kg-1.min-1) when compared with our results (54.9 ± 6.7 mL.kg-1.min-1). Meanwhile, the mean
HR value at 100%MLSS was 174 ± 10 beats.min-1 in the present study, values which were simi-
lar to the previous reported in literature [32, 34], as expected by the comparable age of
samples.
Moreover, the [La-] at 100%MLSS (1.89 ± 0.77 mmol.L-1) in the present study were lower
when compared to swimming literature (2.8–3.3 mmol.L-1) [14, 34, 35]. These lower [La-] val-
ues may be explained by sex differences for similar levels of aerobic capacity development,
with expected lower values for women due to lower body mass and lean muscle mass com-
pared to men [36]. Furthermore, women have showed lower testosterone concentration com-
pared to men [37] during aerobic endurance exercise [33, 36], suggesting different metabolic
contributions between carbohydrates and fat during long-distance exercise [33, 38], and sup-
porting comparable lower [La-].
Since the early research on _VO2 kinetics [39] until up to date, the time constant (τ) has
been studied in sports science in the attempt to comprehend the physiological adjustments
during the non-steady state period at the beginning of exercise due to the increase of metabolic
demand. In the present study, the τp values were similar between intensity levels for the initial
exercise phase (mean = 15.4 ± 5.2 s) and final exercise phase (mean = 10.0 ± 4.7 s), but the val-
ues decreased with previous exercise for the three swim conditions. This is particularly relevant
for training practice, underlining the influence of previous exercise on the subsequent meta-
bolic dynamics. In all studied exercise intensities, the τp in the present study showed similar
values compared than those previously reported in swimming (~15–20 s) [27–29], cycling [40,
41], rowing [42], and running [43, 44]. Thus, those values reported for intensities up to and
above the MLSS seem to behave similarly as expected, based on the previous knowledge on the
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
8 / 12
_VO2 kinetics during different intensity domains for well-trained athletes. Indeed, a faster
attainment of a steady state and a reduction in the oxygen deficit are associated to the fatigue
delay and increase in exhaustion time, being well trained athletes able to perform at higher
intensities with lower requirements of anaerobic energy during the transition from rest to
exercise [5]. Hence, the lower τp values reported in this study when compared to previously
published ones regarding physiological adaptations induced by aerobic endurance training
confirm the highly endurance training status and specialization (endurance athletes) of our
swimmers [5, 44].
Partially in contrast with previous literature that showed the existence of the As at these
exercise intensities [2, 4, 5], in the present study it has shown to occur chaotically during the
three swimming conditions, with very diverse individual occurrence profiles; however, observ-
ing the sample data a tendency to As increase as a function of intensity was observed (2.2 ± 1.1,
2.9 ± 0.8 and 4.5 ± 1.6 mL.kg-1.min-1, respectively for 97.5, 100 and 102.5%MLSS), but only
during initial exercise, not during the final phase after metabolic adaptation already occurred.
Besides, only two swimmers showed As occurrence in all trials both at the initial and final exer-
cise phases, and one swimmer did not show any As during all the swimming efforts and phases.
It is worthy to emphasize the curiosity of that particular swimmer being a national record
holder (800 and 1500m) and the best endurance swimmer of the sample. These partially con-
tradictory findings could be explained, at least in part, by the specific physiological adaptations
occurred through the highly endurance training status for our swimmers, such as faster τp
[44], possible increase in the mitochondrial content of the cell [45], beyond also possible alter-
ations in the mitochondrial sensitivity to the respiration regulators [46], and the fact of these
endurance athletes might have mainly type I muscle fibers [45]. Thus, our endurance swim-
mers with fast _VO2 kinetics would be able to adjust faster the physiological requirements for
aerobic performance during the high intensity aerobic exercises, minimizing the As demand.
In addition, the appearance of the As is normally explained by a phenomena that may be atten-
uated in our very specialized sample, namely the recruitment of type II fibers with fatigue [47],
after which the magnitude of As has been correlated with the rise of [La-] [2, 45]. Thereby, the
absence of significant As in the present study may be likely explained by the high-level of
endurance training of the sample [48].
Moreover, to reinforce the predominance of aerobic energy system during the three swim
conditions around MLSS, the present study determined the total energy contribution at each
one of the studied exercise intensities. At all swimming intensities up to and above MLSS, the
aerobic energy contribution was higher than 98% of the total energy contribution; however
there were significant differences between the highest and the lower v regarding aerobic and
anaerobic energy contributions. This study was the first study to show the energy contribution
during intensities at and around MLSS directly measured breath-by-breath in swimming,
which highlights that even at intensities above MLSS; the total energy contribution was mainly
and almost exclusively controlled by the oxidative bioenergetics system.
Conclusions
The present study showed that well-trained endurance swimmers with a fast component of
_VO2 kinetics, i.e. an abrupt and fast increase in _VO2 response, and low [La-] may be able to
adjust faster the physiological requirements during intensities up to and slightly above MLSS
to minimize the appearance of the slow component of _VO2 and reduce the oxygen deficit,
both parameters are associated to the fatigue delay and the increase in exhaustion time, key
factors to endurance performance. however, these fast adjustments were not able to control
the progressive fatigue occurred slightly above MLSS, and most of swimmers reached
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
9 / 12
exhaustion before 30min swam. Moreover, the data shows that the aerobic energy contribution
at intensities up to and slightly above MLSS plays a fundamental role controlling almost exclu-
sively the required energy supply.
Supporting information
S1 File. Values of physiological parameters at 97.5, 100 and 102.5% of the maximal lactate
steady state (MLSS) (N = 10).
(PDF)
S2 File. Values of VO2 kinetics parameters at 97.5, 100 and 102.5% of the maximal lactate
steady state (MLSS) (N = 10).
(PDF)
Acknowledgments
This investigation was supported by grants of the Capes Foundation, Ministry of Education of
Brazil (BEX: 0536/10-5), and project PTDC/DES/101224/2008 (FCOMP-01-0124-FEDER-
009577).
Author Contributions
Conceptualization: JGP JPVB CCG RJF.
Data curation: JGP LM JPVB CCG RJF.
Formal analysis: JGP LM JPVB.
Funding acquisition: JPVB RJF.
Investigation: JGP JPVB CCG RJF LM.
Methodology: JGP JPVB LM CCG RJF.
Project administration: JGP JPVB RJF.
Resources: JGP RJF JPVB.
Software: LM JGP JPVB.
Supervision: JGP JPVB CCG RJF LM.
Validation: JGP LM JPVB CCG RJF.
Visualization: JGP JPVB CCG RJF LM.
Writing – original draft: JGP JPVB RJF CCG LM.
Writing – review & editing: JGP JPVB CCG RJF LM.
References
1.
Bosquet L, Leger L, Legros P. Methods to determine aerobic endurance. Sports Med. 2002; 32
(11):675–700. PMID: 12196030
2.
Gaesser GA, Poole DC. The slow component of oxygen uptake kinetics in humans. Exerc Sport Sci
Rev. 1996; 24(1):35–71.
3.
Poole DC, Ward SA, Gardner GW, Whipp BJ. Metabolic and respiratory profile of the upper limit for pro-
longed exercise in man. Ergonomics. 1988; 31(9):1265–79. Epub 1988/09/01. doi: 10.1080/
00140138808966766 PMID: 3191904
_
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
10 / 12
4.
Poole DC, Richardson RS. Determinants of oxygen uptake. Implications for exercise testing. Sports
Med. 1997; 24(5):308–20. Epub 1997/11/22. PMID: 9368277
5.
Burnley M, Jones A. Oxygen uptake kinetics as a determinant of sports performance. European Journal
of Sport Science. 2007; 7(2):63–9.
6.
Whipp BJ, Wasserman K. Oxygen uptake kinetics for various intensities of constant-load work. J Appl
Physiol. 1972; 33(3):351–6. Epub 1972/09/01. PMID: 5056210
7.
Whipp BJ, Ward SA. Physiological determinants of pulmonary gas exchange kinetics during exercise.
Med Sci Sports Exerc. 1990; 22(1):62–71. Epub 1990/02/01. PMID: 2406547
8.
Beneke R. Maximal lactate steady state concentration (MLSS): experimental and modelling
approaches. Eur J Appl Physiol. 2003; 88(4–5):361–9. doi: 10.1007/s00421-002-0713-2 PMID:
12527964
9.
Heck H, Mader A, Hess G, Mucke S, Muller R, Hollmann W. Justification of the 4-mmol/l lactate thresh-
old. Int J Sports Med. 1985; 6(3):117–30. doi: 10.1055/s-2008-1025824 PMID: 4030186
10.
Beneke R, von Duvillard SP. Determination of maximal lactate steady state response in selected sports
events. Med Sci Sports Exerc. 1996; 28(2):241–6. PMID: 8775160
11.
Faude O, Kindermann W, Meyer T. Lactate threshold concepts: how valid are they? Sports Med. 2009;
39(6):469–90. doi: 10.2165/00007256-200939060-00003 PMID: 19453206
12.
Billat VL, Sirvent P, Py G, Koralsztein JP, Mercier J. The concept of maximal lactate steady state: a
bridge between biochemistry, physiology and sport science. Sports Med. 2003; 33(6):407–26. PMID:
12744715
13.
Baron B, Dekerle J, Depretz S, Lefevre T, Pelayo P. Self selected speed and maximal lactate steady
state speed in swimming. J Sports Med Phys Fitness. 2005; 45(1):1–6. PMID: 16208283
14.
Pelarigo JG, Denadai BS, Greco CC. Stroke phases responses around maximal lactate steady state in
front crawl. J Sci Med Sport. 2011; 14(2):168 e1–e5. Epub 2010/10/12.
15.
Sousa AC, Vilas-Boas JP, Fernandes RJ. VO2 Kinetics and Metabolic Contributions Whilst Swimming
at 95, 100, and 105% of the Velocity at VO2max. Biomed Res Int. 2014; 1(1):1–9. Epub 2014/07/22.
16.
Fernandes RJ, Billat VL, Cruz AC, Colaco PJ, Cardoso CS, Vilas-Boas JP. Does net energy cost of
swimming affect time to exhaustion at the individual’s maximal oxygen consumption velocity? J Sports
Med Phys Fitness. 2006; 46(3):373–80. PMID: 16998440
17.
Machado L, Almeida M, Morais P, Fernandes R, Vilas-Boas JP, editors. Assessing the individual anaer-
obic threshold: the mathematical model. Xth International Symposium of Biomechanics and Medicine in
Swimming; 2006; Porto, Portugal2006.
18.
Baldari C, Fernandes RJ, Meucci M, Ribeiro J, Vilas-Boas JP, Guidetti L. Is the new AquaTrainer® snor-
kel valid for VO2 assessment in swimming? Int J Sports Med. 2013; 34(4):336–44. doi: 10.1055/s-
0032-1321804 PMID: 23041962
19.
Fernandes RJ, de Jesus K, Baldari C, Sousa AC, Vilas-Boas JP, Guidetti L. Different VO2max time-
averaging intervals in swimming. Int J Sports Med. 2012; 33(12):1010–5. doi: 10.1055/s-0032-1316362
PMID: 22791619
20.
di Prampero PE. Energetics of muscular exercise. Rev Physiol Biochem Pharmacol. 1981; 89(1):143–
222. Epub 1981/01/01.
21.
Stevens JP. Applied multivariate statistics for the social sciences. Fourth edition ed. Mahwah: Law-
rence Erlbaum Associates; 2002.
22.
Carter H, Pringle JS, Jones AM, Doust JH. Oxygen uptake kinetics during treadmill running across exer-
cise intensity domains. Eur J Appl Physiol. 2002; 86(4):347–54. PMID: 11990749
23.
Burnley M, Doust JH, Carter H, Jones AM. Effects of prior exercise and recovery duration on oxygen
uptake kinetics during heavy exercise in humans. Exp Physiol. 2001; 86(3):417–25. PMID: 11429659
24.
Sousa A, Ribeiro J, Sousa M, Vilas-Boas JP, Fernandes RJ. Influence of Prior Exercise on VO2 Kinetics
Subsequent Exhaustive Rowing Performance. PLoS One. 2014; 9(1):0084208.
25.
Jones AM, Koppo K, Burnley M. Effects of prior exercise on metabolic and gas exchange responses to
exercise. Sports Med. 2003; 33(13):949–71. PMID: 14606924
26.
Xu F, Rhodes EC. Oxygen uptake kinetics during exercise. Sports Med. 1999; 27(5):313–27. PMID:
10368878
27.
Reis JF, Alves FB, Bruno PM, Vleck V, Millet GP. Effects of aerobic fitness on oxygen uptake kinetics in
heavy intensity swimming. Eur J Appl Physiol. 2012; 112(5):1689–97. doi: 10.1007/s00421-011-2126-6
PMID: 21879352
28.
Reis JF, Alves FB, Bruno PM, Vleck V, Millet GP. Oxygen uptake kinetics and middle distance swim-
ming performance. J Sci Med Sport. 2012; 15(1):58–63. Epub 2011/08/02. doi: 10.1016/j.jsams.2011.
05.012 PMID: 21802360
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
11 / 12
29.
Pessoa Filho DM, Alves FB, Reis JF, Greco CC, Denadai BS. VO2 kinetics during heavy and severe
exercise in swimming. Int J Sports Med. 2012; 33(9):744–8. doi: 10.1055/s-0031-1299753 PMID:
22592546
30.
Reis VM, Marinho DA, Policarpo FB, Carneiro AL, Baldari C, Silva AJ. Examining the accumulated oxy-
gen deficit method in front crawl swimming. Int J Sports Med. 2010; 31(6):421–7. doi: 10.1055/s-0030-
1248286 PMID: 20301045
31.
Jones AM, Burnley M. Oxygen uptake kinetics: an underappreciated determinant of exercise perfor-
mance. Int J Sports Physiol Perform. 2009; 4(4):524–32. PMID: 20029103
32.
Dekerle J, Pelayo P, Clipet B, Depretz S, Lefevre T, Sidney M. Critical swimming speed does not repre-
sent the speed at maximal lactate steady state. Int J Sports Med. 2005; 26(7):524–30. doi: 10.1055/s-
2004-821227 PMID: 16195984
33.
Greco CC, Pelarigo JG, Figueira TR, Denadai BS. Effects of gender on stroke rates, critical speed and
velocity of a 30-min swim in young swimmers. J Sports Sci Med. 2007; 6(4):441–7. PMID: 24149476
34.
Dekerle J, Nesi X, Lefevre T, Depretz S, Sidney M, Marchand FH, et al. Stroking parameters in front
crawl swimming and maximal lactate steady state speed. Int J Sports Med. 2005; 26(1):53–8. doi: 10.
1055/s-2004-817854 PMID: 15643535
35.
Figueiredo P, Naza´rio R, Sousa M, Pelarigo JG, Vilas-Boas JP, Fernandes R. Kinematical Analysis
along Maximal Lactate Steady State Swimming Intensity. J Sports Sci Med. 2014; 13(3):610–5. PMID:
25177189
36.
Crewther B, Cronin J, Keogh J. Possible stimuli for strength and power adaptation: acute metabolic
responses. Sports Med. 2006; 36(1):65–78. PMID: 16445311
37.
Deschenes MR, Kraemer WJ. Performance and physiologic adaptations to resistance training. Am J
Phys Med Rehabil. 2002; 81(11 Suppl):S3–16. Epub 2002/11/01. doi: 10.1097/01.PHM.0000029722.
06777.E9 PMID: 12409807
38.
Tarnopolsky MA, Atkinson SA, Phillips SM, MacDougall JD. Carbohydrate loading and metabolism dur-
ing exercise in men and women. J Appl Physiol. 1995; 78(4):1360–8. Epub 1995/04/01. PMID:
7615443
39.
Margaria R, Edwards HT, Dill DB. The possible mechanisms of contracting and paying the oxygen debt
and the roˆle of lactic acid in muscular contraction1933 1933-11-30 00:00:00. 689–715 p.
40.
Berger NJ, Jones AM. Pulmonary O2 uptake on-kinetics in sprint- and endurance-trained athletes. Appl
Physiol Nutr Metab. 2007; 32(3):383–93. Epub 2007/05/19. doi: 10.1139/H06-109 PMID: 17510672
41.
Koppo K, Bouckaert J, Jones AM. Effects of training status and exercise intensity on phase II VO2 kinet-
ics. Med Sci Sports Exerc. 2004; 36(2):225–32. Epub 2004/02/10. doi: 10.1249/01.MSS.0000113473.
48220.20 PMID: 14767244
42.
Ingham SA, Carter H, Whyte GP, Doust JH. Comparison of the oxygen uptake kinetics of club and olym-
pic champion rowers. Med Sci Sports Exerc. 2007; 39(5):865–71. Epub 2007/05/01. doi: 10.1249/mss.
0b013e31803350c7 PMID: 17468587
43.
Borrani F, Candau R, Millet GY, Perrey S, Fuchslocher J, Rouillon JD. Is the VO2 slow component
dependent on progressive recruitment of fast-twitch fibers in trained runners? J Appl Physiol. 2001; 90
(6):2212–20. PMID: 11356785
44.
Carter H, Jones AM, Barstow TJ, Burnley M, Williams C, Doust JH. Effect of endurance training on oxy-
gen uptake kinetics during treadmill running. J Appl Physiol. 2000; 89(5):1744–52. PMID: 11053321
45.
Holloszy JO, Coyle EF. Adaptations of skeletal muscle to endurance exercise and their metabolic con-
sequences. J Appl Physiol Respir Environ Exerc Physiol. 1984; 56(4):831–8. Epub 1984/04/01. PMID:
6373687
46.
Dudley GA, Tullson PC, Terjung RL. Influence of mitochondrial content on the sensitivity of respiratory
control. J Biol Chem. 1987; 262(19):9109–14. Epub 1987/07/05. PMID: 3597408
47.
Poole DC, Schaffartzik W, Knight DR, Derion T, Kennedy B, Guy HJ, et al. Contribution of exercising
legs to the slow component of oxygen uptake kinetics in humans. J Appl Physiol. 1991; 71(4):1245–60.
Epub 1991/10/01. PMID: 1757346
48.
Billat V, Binsse V, Petit B, Koralsztein JP. High level runners are able to maintain a VO2 steady-state
below VO2max in an all-out run over their critical velocity. Arch Physiol Biochem. 1998; 106(1):38–45.
Epub 1998/10/23. doi: 10.1076/apab.106.1.38.4396 PMID: 9783059
Oxygen kinetics and aerobic endurance performance
PLOS ONE | DOI:10.1371/journal.pone.0167263
February 28, 2017
12 / 12
| Oxygen uptake kinetics and energy system's contribution around maximal lactate steady state swimming intensity. | 02-28-2017 | Pelarigo, Jailton Gregório,Machado, Leandro,Fernandes, Ricardo Jorge,Greco, Camila Coelho,Vilas-Boas, João Paulo | eng |
PMC8198575 | International Journal of
Environmental Research
and Public Health
Article
Speed, Change of Direction Speed and Reactive Agility in
Adolescent Soccer Players: Age Related Differences
Slobodan Andraši´c 1
, Marko Guši´c 2
, Mima Stankovi´c 3
, Draženka Maˇcak 2
, Asim Bradi´c 4, Goran Sporiš 4
and Nebojša Trajkovi´c 3,*
Citation: Andraši´c, S.; Guši´c, M.;
Stankovi´c, M.; Maˇcak, D.; Bradi´c, A.;
Sporiš, G.; Trajkovi´c, N. Speed,
Change of Direction Speed and
Reactive Agility in Adolescent Soccer
Players: Age Related Differences. Int.
J. Environ. Res. Public Health 2021, 18,
5883. https://doi.org/10.3390/
ijerph18115883
Academic Editors: Caio Victor Sousa
and Samuel da Silva Aguiar
Received: 31 March 2021
Accepted: 27 May 2021
Published: 30 May 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed
under
the
terms
and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
Faculty of Economics, University of Novi Sad, 24000 Subotica, Serbia; [email protected]
2
Faculty of Sport and Physical Education, University of Novi Sad, 21000 Novi Sad, Serbia;
[email protected] (M.G.); [email protected] (D.M.)
3
Faculty of Sport and Physical Education, University of Niš, 18000 Niš, Serbia; [email protected]
4
Faculty of Kinesiology, University of Zagreb, 10000 Zagreb, Croatia; [email protected] (A.B.);
[email protected] (G.S.)
*
Correspondence: [email protected]
Abstract: There are a plethora of studies investigating agility in soccer; however, studies have rarely
presented the reaction time in differentiating age groups in adolescent soccer players. We investigated
age differences in reactive agility, speed, and change of direction speed (CODs), in a group of highly
trained adolescent soccer players. A total of 75 adolescent male soccer players (aged 14–19 years)
were recruited. The players were grouped based on their age to under 15 (U15; n = 27), under 17
(U17; n = 25), and under 19 (U19; n = 23) players. Players were tested for 5 m, 10 m, and 20 m sprint,
CODs speed test, Illinois test, and reactive agility test (total and reaction time). Only the reactive
agility test with a live tester (RAT live) and RAT live reaction time (RAT live RT) distinguished U19
from both groups, U17 (RAT live, p < 0.01; RAT RT live, p < 0.01) and U15 (RAT live, p < 0.01; RAT
RT live, p < 0.01). Groups did not have different times for 5 m sprint, RAT light and RAT RT light,
F = 0.472, 2.691, 1.023, respectively, p > 0.05. Moreover, a significantly slower average performance of
sprint 20, CODs left and right, and Illinois was also observed in U15 as compared to U17 and U19
(p < 0.05). We can conclude that results in agility tests that include live testers can be a significant
factor that differentiates between adolescent soccer players considering their age.
Keywords: agility; differences; youth; performance; football
1. Introduction
The demands of modern soccer have changed significantly and increased in recent
years. Nowadays, players are required to have more power and to cover greater distances,
with more frequent changes in intensity [1–3]. Most high-intensity activities (sprints)
take place during decisive moments, such as tackling, offensive and defensive actions,
as well as goal-scoring opportunities [4–7]. As soccer is considered to be a sport that
requires that attackers evade their opponents’ pressures or tackles, and defenders reduce
space on the field in order to limit attacking movements or potentially achieve a turnover,
having good change of direction speed (CODs) and agility is beneficial [8]. In the last few
decades, change of direction speed and reactive agility were considered to be the same
skill [9]. However, nowadays, pre-planned agility may be defined as sprints with change
of direction, while the reactive agility (RA) is classified as sprints with directional changes
in response to a stimulus [10,11]. Therefore, RA is based on greater levels of motor control,
when compared to pre-planned CODs [12].
Reactive agility and CODs are one of the most important skills required for soccer
success [13]. Moreover, reactive agility tests (RATs) are able to differentiate the key per-
formance indicators presented as the skill levels among soccer players [14]. On the other
side, there are still some doubts about the test design and type of stimulus presented in
Int. J. Environ. Res. Public Health 2021, 18, 5883. https://doi.org/10.3390/ijerph18115883
https://www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2021, 18, 5883
2 of 7
the existing literature regarding reactive agility tests. In addition, most of the studies only
measured total times instead of reaction times. As RATs are designed to evaluate both
physical, and technical and cognitive abilities, there is an absolute need to further examine
reactive agility assessment in soccer [15].
Some research shows that older adolescent players tend to complete the RATs quicker
than younger players due to their higher fitness level, CODs, and anticipatory skill, which
makes them play on a much higher level [13]. Several studies investigated the perceptual
abilities of higher-level players, and it was shown that there are fundamental perceptual
and cognitive differences between them and lower-level players [16–18]. Additionally,
Tretloci et al. [19] showed that field-based tests including vertical jumps, change of direction
speed, and reactive agility can differentiate between under 16 elite and sub-elite soccer
players. This was confirmed by Trajkovic et al. [20], showing that the skilled players
performed better in reactive agility tests, speed, and CODs compared to amateur players.
Moreover, the authors stated that reactive agility tests with live opponent stimuli can
be a significant factor that differentiates between adolescent soccer players considering
their level.
At this point, there are studies that determine the difference in physical performance
between levels of play. However, there are not many studies that show the difference
between age groups in CODs and RA [13,21]. Additionally, the categories used in those
studies do not match the categories used here.
Moreover, according to the authors’ knowledge, there are no studies that compare
light and live stimulus in RATs in soccer players to assess whether these approaches
differentiate between different age groups. Therefore, the purpose of this study was to
determine possible age-related differences in speed, CODs and reactive agility in a group
of trained adolescent soccer players.
2. Materials and Methods
2.1. Subjects
A priori, the G*power 3.1 power analysis software was used to determine that the
required sample size is n = 72 given the critical F(69) = 3.13, eta2 = 0.13 p = 0.05, 1 − β = 0.8,
and number of groups = 3. A total of 75 adolescent male soccer players (aged 14–19 years),
who participate at the highest level of competition in Serbia at their age, were recruited
for this research (Table 1). The players were grouped based on their age to under 15 (U15;
n = 27), under 17 (U17; n = 25), and under 19 (U19; n = 23) players. Only field players
were tested, with goalkeepers excluded. Written informed consent was obtained from the
players and their parents. Moreover, the ethics board of the Faculty of Sport and Physical
Education provided the approval of the research experiment (Ethical Board Approval No:
2019/31). Players were recruited if they had at least 5 years of experience in playing soccer;
had a general training history (more than three times per week) in the previous 12 months;
were currently training for soccer (more than 7 h per week); and did not have existing
medical conditions that would compromise study participation.
Table 1. Physical characteristics for U15 (n = 25), U17 (n = 27), and U19 (n = 23).
U15
U17
U19
Age
14.7 ± 0.6
16.2 ± 0.7
18.8 ± 0.7
Height
178.06 ± 5.82
180.49 ± 6.56
179.12 ± 5.45
Weight
69.06 ± 10.82
72.89 ± 8.48
70.70 ± 8.22
Experience (years)
5.1 ± 2.7
6.5 ± 2.9
7.3 ± 1.4
Training (min·week-1)
447 ± 126
475 ± 167
487 ± 130
U15—younger than 15 years; U17—younger than 17 years; U19—younger than 19 years; min·week—minutes
per week.
Int. J. Environ. Res. Public Health 2021, 18, 5883
3 of 7
2.2. Procedure
Testing was conducted at the beginning of the annual training season to limit differ-
ences in training status between players. All players followed a similar training program
under the supervision of their respective coaches. All performance tests were conducted
on the same day. Test sessions were undertaken between 09:00 and 13:00 h following at
least 8 h of sleep and 48 h of rest. All performance tests were performed on an outdoor
facility with artificial grass in favorable weather conditions (no wind or rain). Before
testing, a 20 min standardized warm-up was conducted, which consisted of low-intensity
running, acceleration runs, skipping and hopping exercises. Players were familiar with all
test procedures.
Height and weight measurements were taken in the morning. Height was measured
with a fixed stadiometer (+0.1 cm, Holtain Ltd., Crosswell, UK), and body mass with a
digital balance (+0.1 kg, ADE Electronic Column Scales, Hamburg, Germany). The same
researcher conducted all the measurements.
Running speed. The running speed of players was determined using the time to 5, 10
and 20 m using infrared timing gates, 20-m sprint effort with photocell gates (Microgate,
Polifemo Radio Light, Bolzano, Italy) placed 0.4 m above the ground, with an accuracy of
0.001 s. The timer was automatically activated as participants crossed the first gate at the
starting line with split times at 5 m and 10 m. Players were instructed to run as quickly
as possible over the 20-m distance from a standing start (crouched start positioned 0.5 m
behind the timing lights). Acceleration was evaluated using the time to cover the first 5 m
of the 20-m test. Participants performed two trials with at least 3 min of rest between them.
The best performance of the two tests was used for analysis.
Change of direction speed test (CODs). The pre-planned agility test [22] is used to
evaluate CODs. Participants were asked to sprint as fast as possible for 5 m through a
triggered timing gate (start gate), make a 45◦ cut and sprint 5 m to the left and right through
a target gate. In this test, participants knew the cut direction. Running time was recorded
using photocell gates (Microgate, Polifemo Radio Light, Bolzano, Italy) placed 0.4 m above
the ground, with an accuracy of 0.001 s at the start and finish gates. The best time of three
attempts on the left and right side was considered for further analysis.
The reactive agility test (RAT) was performed according to the protocol described
previously by Chaouachi et al. [23]. In the current study, the RAT involved a decision-
making element provided by a live tester (RAT live) acting as an opponent and light stimuli
used instead of testers (RAT light). During RAT live, the tester had 4 options for each
condition: preplanned and randomly ordered (i.e., 8 trials). All these conditions were
provided to each player in 2 series (5–8 min between sets rest) in a random order. Players
were instructed to recognize the cues as soon as possible. Running time was recorded
using photocell gates (Microgate, Polifemo Radio Light, Bolzano, Italy) placed 0.4 m above
the ground, with an accuracy of 0.001 s. Total time (RAT TT live) and reaction time (RAT
RT live) were recorded for each trial, and the best performance was considered for the
analysis. The same conditions were used for RAT light, but this time the Witty SEM lights
were used instead the testers. When the participants pass the first gate, the signal shows
right or left direction. The participants must react to the visual signal, change direction
and pass the third gate. Similar to RAT live, the total time (RAT TT light) and response
movement time (RAT RT light) were recorded for each trial, and the best performance was
taken for analysis.
Illinois agility test: The length of the field is 10 m, while the width (distance between
the start and finish points) is 5 m. Four cones were placed in the center of the testing area
at a distance of 3.3 m from one another. Four cones were used to mark the start, finish
and two turning points. The subjects started the test lying face down, with their hands at
shoulder level. The trial started on the “go” command, and the subjects began to run as
fast as possible. The trial was completed when the players crossed the finish line without
having knocked any cones over. Best time out of three trials was used for analysis [24].
Int. J. Environ. Res. Public Health 2021, 18, 5883
4 of 7
2.3. Statistical Analysis
The analysis of the data obtained from the study was performed by SPSS software,
version 23.0 (SPSS Inc., Chicago, IL, USA). Data are reported as mean ± SD unless otherwise
stated. The Kolmogorov–Smirnov test was conducted to verify if all data met the normality
test assumption. Test–retest reliability was assessed for all tests using a one-way intra-class
correlation coefficient (ICC) based on average measurements (ICC 1,k).
All analyses of variance (ANOVA) were performed on log-transformed data; for the
sake of clarity, however, they are reported non-transformed. Age-based comparisons of
study outcomes were made with one-way ANOVA (U15, U17, and U19). When ANOVA
showed a significant group effect, between group differences were allocated by using post
hoc Bonferroni tests. Eta squared (η2) is reported as a measure of effect size and defined as
small (0.01), medium (0.06), and large (0.14) according to Cohen. The level of significance
was set at p < 0.05.
3. Results
3.1. Sample Characteristics
Table 1 shows the physical characteristics of the players according to age group.
Height and weight were similar across the groups (p > 0.05); therefore, adjustments were
not performed.
3.2. Study Outcomes in Relation to the Age Group of Soccer Players
On average, the U15 needed significantly more time than U19 to perform all tests,
except sprint 5 m (p = 0.63), RAT light (p = 0.08), and RAT RT light (p = 0.37), which were
similar across the age groups. A significantly slower average performance of sprint 20,
CODs left and right, and Illinois was also observed in U15 as compared to U17 and U19.
However, the U17 had a significantly slower mean performance only of RAT live and RAT
RT live than the U19. Visit Table 2 for detailed results from a one-way analysis of variance.
Table 2. Times for 5 m, 10 m sprint, and 20 m sprint, and agility performance for the U15 (n = 25),
U17 (n = 27), and U19 (n = 23) soccer players.
Outcomes
U15
U17
U19
A One-Way ANOVA
F(1, 144)
η2
Sprint 5 m
1.16 ± 0.23
1.13 ± 0.19
1.11 ± 0.12
0.47
0.02
Sprint 10 m
1.93 ± 0.13 a
1.85 ± 0.18
1.83 ± 0.11
3.28 *
0.09
Sprint 20 m
3.38 ± 0.23 b
3.18 ± 0.27
3.16 ± 0.31
5.17 **
0.16
CODs left
2.25 ± 0.15 b
2.15 ± 0.17
2.12 ± 0.21
3.61 *
0.07
CODs right
2.27 ± 0.14 b
2.16 ± 0.19
2.13 ± 0.16
4.87 **
0.11
Illinois
15.82 ± 0.76 b
15.24 ± 0.53
14.93 ± 0.58
12.47 **
0.25
RAT light
2.69 ± 0.13
2.61 ± 0.16
2.60 ± 0.19
2.69
0.07
RAT live
2.58 ± 0.10
2.58 ± 0.11
2.48 ± 0.06 c
8.99 **
0.20
RAT RT light
1.44 ± 0.16
1.41 ± 0.16
1.38 ± 0.11
1.02
0.03
RAT RT live
1.49 ± 0.08
1.38 ± 0.10
1.31 ± 0.14 c
16.27 **
0.31
Values are mean ± SD. U15—younger than 15 years; U17—younger than 17 years; U19—younger than 19 years;
CODs left—change of direction left; CODs right—change of direction right; RAT light—reactive agility test with
witty SEM visual signals; RAT RT light—reaction time during RAT live; RAT live—reactive agility test with testers;
RAT RT light—reaction time during RAT light; * significant age group effect at p ≤ 0.05; ** significant age group
effect at p ≤ 0.01; a U15 and U19 significantly different at p ≤ 0.05; b U15 significantly different at p ≤ 0.05; c U19
significantly different at p ≤ 0.05.
4. Discussion
The present study aimed to determine the difference in several performance indica-
tors relevant for soccer performance in adolescent players of different age groups. The
main finding of this study was that the reactive agility test with live testers was able to
differentiate U19 players from other age groups. Moreover, the U15 group showed slower
average performance in sprint 20, CODs left and right, and Illinois compared to U17 and
Int. J. Environ. Res. Public Health 2021, 18, 5883
5 of 7
U19. A possible explanation for these results could be found in the fact that modern soccer
training, with frequent changes in tactics based on the characteristics of the opponent, has
led to increased adaptability of player roles, especially in young players [9]. Moreover, it is
not so uncommon in youth training to change a player’s position in response to different
game situations, developing a large range of technical solutions useful for their future
soccer-playing career [25].
In the literature, sprinting ability over short (5 m) and longer distances (20 m) is
considered to require separate and specific biomechanical and neuromuscular qualities
and, therefore, training techniques [26]. When considering the U15, U17, and U19 players,
we found a difference between groups for 10 m and 20 m distance, which is in line with the
abovementioned fact. The current findings [14] show that there is no difference between
U17 and U19 for 20 m sprints which is, as well, presented in this study. In addition, we
found that U15 players had worse results than U17 and U19 players. Generally, most
studies that investigated age difference came to the conclusion that older age groups had
better results than younger groups [14,27]. These differences could be associated with the
maturity stage of players, which can affect U15 sprint performance more [28] compared to
U17 and U19. Moreover, post-pubertal players have accelerated gains in strength supported
by rapid gains in muscle mass [29], which may also contribute to the mentioned differences.
Our findings suggest that U19 and U17 players, who performed with similar results,
were statistically significantly better at CODs left, CODs right and the Illinois test than U15
players. These results are similar to the ones found by [9], where U16 and U18 players
gave better results in the Illinois modified test. Additionally, their study showed that there
was no difference in the CODs left and CODs right test between U16 and U18 players,
but there were differences found between U14 players and the aforementioned groups.
A possible reason for non-existent differences between U19 and U17 players is that the
greatest individual differences in biological maturation were found in players 11–16 years
old [30]. Contrary to this study [14], Poljskic et al. concluded that U19 players were
significantly better that U17 players. Taking everything into account, the superiority of
U19 players in agility performances may be observed as a direct consequence of their long
involvement in systematic soccer training and higher performance level due to the CODs.
Our results show that U19 players scored better in the RAT live test than U15 and U17
players. Additionally, we found that U19 players had better results than U15 and U17 in
reaction time with live stimuli. Therefore, in the current study, RATs clearly discriminated
U19 players from U15 and U17. A possible reason could be found in better anticipatory
skills in older adolescent players who have more games and experience behind them. They
have better ability to recognize relevant cues of testers, as previously demonstrated for
team sport athletes [31,32]. Our results are similar to Fiorilli et al. [13], where U16 and U18
players made better results than U14 in the reactive agility test. Moreover, Poljskic et al. [14]
showed that U19 players had better results than U17 in RAG (reactive agility) test. On
the other hand, there is a study that found no significant difference between juniors (<18)
and seniors (>18) in specific reactive agility tests [33]. However, reactive agility is being
developed until the late adolescent age, when it can reach its peak, which could be the
reason for the discrepancy in the results.
It has been stated recently that intervention programs may have to be different for
different age stages [34]. According to our results, we could speculate that modern training
is similar for all age categories in adolescent soccer players. Nowadays, the training
contains sport-specific stimuli rather than generic and high-intensity training for physical
skills. Soccer players, who have anticipatory expertise and make decisions much faster, are
able to recognize and react promptly to a stimulus. Younger players with less experience
may need more time to respond to a stimulus before having the proper reaction in the
shortest time possible in order to avoid being executed by the opponent.
The main limitation of this study is that the attribution of physical ability could be to
talent or previous training. In our study, the players were interviewed about their current
training load (weekly time) and previous experience (years engaged in soccer). Moreover,
Int. J. Environ. Res. Public Health 2021, 18, 5883
6 of 7
they were from the same squad, with the same programs conducted in all categories.
Therefore, we could speculate if different approaches to training could contribute to differ-
ences in other variables. Future studies should examine players from different teams and
academies. Moreover, the maturity level was not introduced and taken into account due
to the fact that the majority of studies have focused largely on players 11–16 years of age,
where individual differences in biological maturation are perhaps the greatest. Another
limitation is the possibility of the circadian rhythm’s influence on performance [35] due to
the time the testing was conducted (9:00 to 13:00 h).
5. Conclusions
Reactive agility and COD speed are key skills required for soccer success, based on
greater levels of motor control. We found that the reactive agility test with a live tester
can be a significant factor that differentiates between older and younger adolescent soccer
players. Moreover, our findings prove that field-based tests including speed, change of
direction speed, and reactive agility are sufficiently sensitive to differentiate between a
group of adolescent soccer players. Further studies are needed to confirm these results.
Author Contributions: Conceptualization, S.A. and N.T.; methodology, M.G.; software, D.M.; vali-
dation, S.A. and N.T.; formal analysis, D.M.; investigation, S.A.; resources, G.S.; data curation, S.A.,
M.G. and N.T.; writing—original draft preparation, S.A., M.S., and N.T.; writing—review and editing,
N.T. and D.M.; visualization, A.B. and M.S.; supervision, N.T.; project administration, S.A.; funding
acquisition, G.S. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: The study was conducted according to the guidelines of the
Declaration of Helsinki, and approved by the Institutional Review Board of the Faculty of Sport and
Physical Education, University of Novi Sad (Ref. No. 30-08-01/2018).
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author. The data are not publicly available because they contain information that
could compromise the privacy of research participants.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Bradley, P.S.; Sheldon, W.; Wooster, B.; Olsen, P.; Boanas, P.; Krustrup, P. High-intensity running in English FA Premier League
soccer matches. J. Sports Sci. 2009, 27, 159–168. [CrossRef] [PubMed]
2.
Di Salvo, V.; Gregson, W.; Atkinson, G.; Tordoff, P.; Drust, B. Analysis of high intensity activity in Premier League soccer. Int. J.
Sports Med. 2009, 30, 205–212. [CrossRef] [PubMed]
3.
Carling, C.; Bloomfield, J.; Nelsen, L.; Reilly, T. The Role of Motion Analysis in Elite Soccer. Sports Med. 2008, 38, 839–
862. [CrossRef]
4.
Faude, O.; Koch, T.; Meyer, T. Straight sprinting is the most frequent action in goal situations in professional football. J. Sports Sci.
2012, 30, 625–631. [CrossRef]
5.
Rienzi, E.; Drust, B.; Reilly, T.; Carter, J.E.L.; Martin, A. Investigation of anthropometric and work-rate profiles of elite South
American international soccer players. J. Sports Med. Phys. Fit. 2000, 40, 162.
6.
Rumpf, M.C.; Lockie, R.G.; Cronin, J.B.; Jalilvand, F. Effect of different sprint training methods on sprint performance over
various distances: A brief review. J. Strength Cond. Res. 2016, 30, 1767–1785. [CrossRef]
7.
Sasaki, S.; Koga, H.; Krosshaug, T.; Kaneko, S.; Fukubayashi, T. Biomechanical Analysis of Defensive Cutting Actions During
Game Situations: Six Cases in Collegiate Soccer Competitions. J. Hum. Kinet. 2015, 46, 9–18. [CrossRef]
8.
Young, W.; Rogers, N. Effects of small-sided game and change-of-direction training on reactive agility and change-of-direction
speed. J. Sports Sci. 2013, 32, 307–314. [CrossRef]
9.
Fiorilli, G.; Iuliano, E.; Mitrotasios, M.; Pistone, E.M.; Aquino, G.; Calcagno, G.; Di Cagno, A. Are Change of Direction Speed and
Reactive Agility Useful for Determining the Optimal Field Position for Young Soccer Players? J. Sports Sci. Med. 2017, 16, 247–253.
10.
Brughelli, M.; Cronin, J.; Levin, G.; Chaouachi, A. Understanding Change of Direction Ability in Sport. Sports Med. 2008, 38,
1045–1063. [CrossRef] [PubMed]
11.
Salaj, S.; Markovic, G. Specificity of Jumping, Sprinting, and Quick Change-of-Direction Motor Abilities. J. Strength Cond. Res.
2011, 25, 1249–1255. [CrossRef]
Int. J. Environ. Res. Public Health 2021, 18, 5883
7 of 7
12.
Spasic, M.; Krolo, A.; Zenic, N.; Delextrat, A.; Sekulic, D. Reactive Agility Performance in Handball; Development and Evaluation
of a Sport-Specific Measurement Protocol. J. Sports Sci. Med. 2015, 14, 501–506.
13.
Fiorilli, G.; Mitrotasios, M.; Iuliano, E.; Pistone, E.M.; Aquino, G.; Calcagno, G.; Di Cagno, A. Agility and change of direction in
soccer: Differences according to the player ages. J. Sports Med. Phys. Fit. 2016, 57, 1597–1604.
14.
Pojskic, H.; Åslin, E.; Krolo, A.; Jukic, I.; Uljevic, O.; Spasic, M.; Sekulic, D. Importance of Reactive Agility and Change of Direction
Speed in Differentiating Performance Levels in Junior Soccer Players: Reliability and Validity of Newly Developed Soccer-Specific
Tests. Front. Physiol. 2018, 9. [CrossRef]
15.
Dolan, K. Reactive Agility, Core Strength, Balance, and Soccer Performance. Ithaca College Theses, 2013. Paper 14.: Retrieved on
the 16 March 2021. Available online: http://digitalcommons.ithaca.edu/ic_theses (accessed on 29 May 2021).
16.
Williams, A.M.; Ward, P.; Smeeton, N.; Allen, D. Developing Anticipation Skills in Tennis Using On-Court Instruction: Perception
versus Perception and Action. J. Appl. Sport Psychol. 2004, 16, 350–360. [CrossRef]
17.
Williams, A.M.; Davids, K.; Williams, J.G. Visual Perception and Action in Sport E&F Spon; Routledge: London, UK, 1999.
18.
Mann, D.T.; Williams, A.M.; Ward, P.; Janelle, C.M. Perceptual-Cognitive Expertise in Sport: A Meta-Analysis. J. Sport Exerc.
Psychol. 2007, 29, 457–478. [CrossRef]
19.
Trecroci, A.; Longo, S.; Perri, E.; Iaia, F.M.; Alberti, G. Field-based physical performance of elite and sub-elite middle-adolescent
soccer players. Res. Sports Med. 2019, 27, 60–71. [CrossRef] [PubMed]
20.
Trajkovi´c, N.; Sporiš, G.; Kristiˇcevi´c, T.; Madi´c, D.M.; Bogataj, Š. The Importance of Reactive Agility Tests in Differentiating
Adolescent Soccer Players. Int. J. Environ. Res. Public Health 2020, 17, 3839. [CrossRef]
21.
Horicka, P.; Simonek, J.; Brodani, J. Diagnostics of reactive and running agility in young football players. Phys. Act. Rev. 2018, 6,
29–36. [CrossRef]
22.
Dellal, A.; Wong del, P.; Moalla, W.; Chamari, K. Physical and technical activity of soccer players in the French First League-with
special reference to their playing position. Int. Sport J. 2010, 11, 278–290.
23.
Chaouachi, A.; Chtara, M.; Hammami, R.; Chtara, H.; Turki, O.; Castagna, C. Multidirectional Sprints and Small-Sided Games
Training Effect on Agility and Change of Direction Abilities in Youth Soccer. J. Strength Cond. Res. 2014, 28, 3121–3127. [CrossRef]
24.
Daneshjoo, A.; Mokhtar, A.H.; Rahnama, N.; Yusof, A. Effects of the 11+ and Harmoknee Warm-up Programs on Physical
Performance Measures in Professional Soccer Players. J. Sports Sci. Med. 2013, 12, 489–496.
25.
Deprez, D.; Fransen, J.; Boone, J.; Lenoir, M.; Philippaerts, R.; Vaeyens, R. Characteristics of high-level youth soccer players:
Variation by playing position. J. Sports Sci. 2015, 33, 243–254. [CrossRef] [PubMed]
26.
Harris, N.K.; Cronin, J.B.; Hopkins, W.G.; Hansen, K.T. Relationship Between Sprint Times and the Strength/Power Outputs of a
Machine Squat Jump. J. Strength Cond. Res. 2008, 22, 691–698. [CrossRef] [PubMed]
27.
Krolo, A.; Gilic, B.; Foretic, N.; Pojskic, H.; Hammami, R.; Spasic, M.; Uljevic, O.; Versic, S.; Sekulic, D. Agility Testing in Youth
Football (Soccer)Players; Evaluating Reliability, Validity, and Correlates of Newly Developed Testing Protocols. Int. J. Environ.
Res. Public Health 2020, 17, 294. [CrossRef] [PubMed]
28.
Waldron, M.; Murphy, A. A comparison of physical abilities and match performance characteristics among elite and subelite
under-14 soccer players. Pediatr. Exerc. Sci. 2013, 25, 423–434. [CrossRef] [PubMed]
29.
Lloyd, R.S.; Oliver, J.L. The youth physical development model: A new approach to long-term athletic development. Strength
Cond. J. 2012, 34, 61–72. [CrossRef]
30.
Deprez, D.; Buchheit, M.; Fransen, J.; Pion, J.; Lenoir, M.; Philippaerts, R.M.; Vaeyens, R. A Longitudinal Study Investigating the
Stability of Anthropometry and Soccer-Specific Endurance in Pubertal High-Level Youth Soccer Players. J. Sports Sci. Med. 2015,
14, 418–426.
31.
Gabbett, T.J.; Kelly, J.N.; Sheppard, J.M. Speed, Change of Direction Speed, and Reactive Agility of Rugby League Players. J.
Strength Cond. Res. 2008, 22, 174–181. [CrossRef]
32.
Paul, D.J.; Gabbett, T.J.; Nassis, G.P. Agility in Team Sports: Testing, Training and Factors Affecting Performance. Sports Med.
2016, 46, 421–442. [CrossRef]
33.
Sekulic, D.; Gilic, B.; Foretic, N.; Spasic, M.; Uljevi´c, O.; Verši´c, Š. Fitness profiles of professional futsal players: Identifying
age-related differences. Biomed. Hum. Kinet. 2020, 12, 212–220. [CrossRef]
34.
Yanci, J.; Arcos, A.L.; Salinero, J.J.; Mendiguchia, J.; Gil, E.; Santesteban, D.; Grande, I. Effects of different agility training programs
among first-grade elementary school students. Coll. Antropol. 2015, 39, 87–92. [PubMed]
35.
Reilly, T.; Atkinson, G.; Edwards, B.; Waterhouse, J.; Farrelly, K.; Fairhurst, E. Diurnal Variation in Temperature, Mental and
Physical Performance, and Tasks Specifically Related to Football (Soccer). Chronobiol Int. 2007, 24, 507–519. [CrossRef] [PubMed]
| Speed, Change of Direction Speed and Reactive Agility in Adolescent Soccer Players: Age Related Differences. | 05-30-2021 | Andrašić, Slobodan,Gušić, Marko,Stanković, Mima,Mačak, Draženka,Bradić, Asim,Sporiš, Goran,Trajković, Nebojša | eng |
PMC4552464 | RESEARCH ARTICLE
Open Access
App use, physical activity and healthy
lifestyle: a cross sectional study
Joan Martine Dallinga1,2*, Matthijs Mennes1, Laurence Alpay2†, Harmen Bijwaard2†
and Marije Baart de la Faille-Deutekom1,2
Abstract
Background: Physical inactivity is a growing public health concern. Use of mobile applications (apps) may be a
powerful tool to encourage physical activity and a healthy lifestyle. For instance, apps may be used in the
preparation of a running event. However, there is little evidence for the relationship between app use and change
in physical activity and health in recreational runners. The aim of this study was to determine the relationship
between the use of apps and changes in physical activity, health and lifestyle behaviour, and self-image of short
and long distance runners.
Methods: A cross sectional study was designed. A random selection of 15,000 runners (of 54,000 participants) of a
16 and 6.4 km recreational run (Dam tot Damloop) in the Netherlands was invited to participate in an online survey
two days after the run. Anthropometrics, app use, activity level, preparation for running event, running physical
activity (RPA), health and lifestyle, and self-image were addressed. A chi-squared test was conducted to analyse
differences between app users and non-app users in baseline characteristics as well as in RPA, healthy lifestyle and
perceived health. In addition, a multivariate logistic regression analysis was performed to determine if app use
could predict RPA, perceived health and lifestyle, and self-image.
Results: Of the 15,000 invited runners, 28 % responded. For both distances, app use was positively related to RPA
and feeling healthier (p < 0.05). Also, app use was positively related to feeling better about themselves, feeling like
an athlete, motivating others to participate in running, and losing weight (p < 0.01). Furthermore, for 16 km runners
app use was positively related to eating healthier, feeling more energetic and reporting a higher chance to
maintain sport behaviour (p < 0.05).
Conclusions: These results suggest that use of mobile apps has a beneficial role in the preparation of a running
event, as it promotes health and physical activity. Further research is now needed to determine a causal
relationship between app use and physical and health related behaviour.
Background
Benefits of physical activity have often been studied and
include improved health and reduced mortality rates [1–3].
However, actually becoming physically active is a challenge
for many. In the Netherlands research shows that 41 %
percent of all adults do not comply with the Dutch Public
Health Physical Activity Guideline (at least 30 min of mod-
erate to vigorous physical activity during at least 5 days of
the week) [4]. Moreover, only 20 % of Dutch adults meet the
Strenuous Intensity Physical Activity Guideline of at least
three times a week 20 min of vigorous exercise [4]. Physical in-
activity is a growing public health concern in the Netherlands
as well as in other Western countries. Significant health prob-
lems such as increased morbidity and mortality attributable to
cardiovascular disease, diabetes, cancers and increased risk of
depression may arise if the amount of physical activity in the
general population does not increase [5–9].
There is need for innovative ways to promote physical
activity and a healthy lifestyle. One promising develop-
ment is the use of smartphones during exercise. Use of
mobile applications (apps) may be a powerful tool to
encourage physical activity and health [10, 11]. Apps are
* Correspondence: [email protected]
†Equal contributors
1School of Sports and Nutrition, Amsterdam University of Applied Sciences,
Dr. Meurerlaan 8, 1067 SM Amsterdam, The Netherlands
2Faculty of Health, Sports and Social Work, Inholland University of Applied
Sciences, Blijdorplaan 15, 2015 CE Haarlem, The Netherlands
© 2015 Dallinga et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Dallinga et al. BMC Public Health (2015) 15:833
DOI 10.1186/s12889-015-2165-8
accessible, have a large reach, and have multiple function-
alities, such as interactive possibilities and feedback op-
portunities [12, 13]. Although more than 17,000 health
and fitness apps have been developed and are available for
the public [12], the literature considering the relationship
of app use and health and physical activity is scarce.
However, preliminary evidence is promising [11, 14, 15].
Two reviews and one meta-analysis demonstrated positive
effects of mobile phone interventions, interventions with
mobile technology, and interventions with remote and
web interventions in healthy, inactive and overweight indi-
viduals [11, 14, 15]. The mobile phone interventions were
often combined with additional education, self-reporting
of frequency and type of use of the program or telephone
calls. The positive effects of these interventions included
increased physical activity (expressed by total time,
number of occasions of physical activity and energy ex-
penditure), cardiovascular fitness and reduced overweight
[11, 14, 15]. Small to moderate effect sizes were reported
[14, 15]. Nevertheless, in these three reviews few interven-
tions were included that used apps. Moreover, in some
studies additional interventions were provided next to the
mobile phone and app interventions, therefore based on
those studies no conclusions can be drawn regarding to
the isolated effects of apps on physical activity. Another
recent review demonstrated modest effects of app based
interventions on physical activity expressed by step count
[16]. It should be noted that the apps were often com-
bined with external pedometers, small sample sizes were
included, small increases in step counts and a short
duration of interventions was presented [16]. However, a
recent study has shown promising results of the isolated
effect of app use [17]. This study demonstrated that use of
a Web-based app on lifestyle indicators decreased weight
and increased physical activity of people [17]. Moreover,
app users presented a higher chance to maintain a healthy
lifestyle [17]. In summary, few studies have examined the
effect of app use on changes in physical activity and
health.
In recreational running the use of apps is high and
emerging and several apps have been developed to assist
individuals in their running exercise. Previous research
has shown that recreational running or participation in a
running mass event could also be a potential health and
physical activity promoting activity [18–20]; Chatton and
Kayser showed that participants in a 16 km run were more
active than the general population and better in shape
[18]. Additionally, in the preparation for a 5 and a 10 km
run participants increased physical activity [19, 20]. A
majority of participants train in preparation for running
events; some of them exercise individually and some of
them in a running group [21, 22]. Potentially, app use
could assist runners to increase motivation, to increase
activity level and set goals during the preparation for a
running event. Perhaps the use of apps could assist
runners to increase running physical activity and to live
and feel healthier. Therefore, the aim of this study was to
determine the relationship between the use of apps and
changes in physical activity and health and lifestyle behav-
iour of short and long distance runners. More specific, we
were interested in training volume, alcohol intake, smok-
ing behaviour, and lifestyle (e.g. weight loss and eating
behaviour).
Methods
Study design and participants
A cross sectional study was designed to analyse the rela-
tionship between app use and physical activity, health and
lifestyle of recreational runners. On September 21st 2014
the 30th Dam tot Damloop, a running event, was orga-
nized in Amsterdam, the Netherlands. The organization
of the running event randomly selected and invited 15,000
runners out of 54,410 participants (16 and 6.4 km) to
participate in an online survey. Runners of all levels were
invited to participate. Participation in the run was either
on an individual basis, with a company or for a charity.
Inclusion criteria were (a) ≥18 years and (b) signed in-
formed consent. Exclusion criteria were (a) participating
in both distances or (b) leaving all questions unanswered
after informed consent.
Two days after participation to the event, an email invi-
tation including a link to the online survey was sent to the
random selection of participants. After one week, a
reminder was sent to the participants who had not
responded yet. This online survey was based on a previ-
ously developed survey [23], with additional items for this
specific running event. An additional file presents the sur-
vey questions (see Additional file 1). In the introduction of
the survey the purpose of the study was explained and
confidentiality was guaranteed. Furthermore, it was ascer-
tained that participation was voluntary and that the
participant was allowed to quit at any time. Responding to
the questionnaire took approximately 15 min. The ethical
approval was not required in the Netherlands, however
the research was conducted in line with the Helsinki
Declaration.
Key measures
Dependent variables
Running physical activity (RPA) was collected. Participants
were invited to report on two occasions (before their
training phase (baseline) and during training phase) how
many kilometres per week they ran (<5 km a week,
between 5 and 10, between 10 and 20 a week, between 20
and 30 a week and more than 30 km per week). In
addition, the survey included questions regarding health
and lifestyle. Alcohol consumption (glasses per week) and
frequency of smoking (per day) was asked before their
Dallinga et al. BMC Public Health (2015) 15:833
Page 2 of 9
training phase and during training phase. Additionally,
participants were requested to indicate whether participa-
tion in the run affected their health (no effect, feel much
healthier, feel healthier, feel less healthy, feel much less
healthy). Moreover, participants indicated if the run influ-
enced their body weight, diet, and energy level (totally
agree, agree, neutral, disagree, totally disagree). To gain
insight in self-image, participants were asked whether the
run influenced their perception concerning a healthy
lifestyle (totally agree, agree, neutral, disagree, totally dis-
agree). Items included were: performing sports is good for
me, chance of maintaining physical activity, feeling better
about oneself, no change in lifestyle, and feeling tired
more often.
Potential prediction variables
Participants indicated if they used an app or other train-
ing tool. Additionally, we collected information about
several variables that needed to be controlled for (gen-
der, age, body mass index (BMI)). To calculate age, date
of birth was asked. Subsequently, age was calculated by
subtracting the year of birth from 2014. BMI (kg/m2)
was used as a proxy of body composition and calculated
as self-reported body weight (kg) divided by the square
of height (m). We used the WHO categories for the
classification: BMI < 18.5 means underweight, [18.5, 25)
equals normal weight, [25, 30) means overweight and ≥
30 corresponds to obese [24]. In addition, information
was collected to determine the participant’s preparation
for the event, fitness state, and experience with running/
sports. As an indication, exercise frequency, in number
of training sessions per year, was requested [25]. Partici-
pants were asked whether they had participated before
in this running event (and if so the number of previous
participations) to estimate
experience with running
events. The training period that participants scheduled
to prepare for this running event was asked as well. Par-
ticipants could choose between categories: no training/
barely, 1–5 weeks, 6–11 weeks, 12 weeks or more, no
specific training/train all year and don’t know/no an-
swer. Participants indicated self-reported finishing time
in hours and minutes as well.
Data reduction
The difference in RPA between baseline and training
phase was calculated. For all participants it was assessed
whether the RPA was increased or not. Furthermore, the
difference in consumption of alcohol and smoking be-
tween baseline and training phase was calculated. Calcula-
tions were performed to examine if these two factors were
decreased or not. For the outcome of perceived health it
was determined whether participants felt healthier or not.
Answers on theses concerning healthy lifestyle and self-
image were reduced from five to two categories; we
calculated if the participants agreed or answered neutral/
disagreed with the theses about these topics.
Statistical analysis
SPSS version 20.0 was used for all calculations. For both
distances, means and standard deviations (SD’s) were
calculated for age, BMI and exercise frequency. The data
was checked for outliers. For the categorical variables,
frequency and percentage were calculated.
We used a chi-squared test to determine differences
between app users and non-app users in baseline charac-
teristics as well as in physical activity, healthy lifestyle and
perceived health during training phase. In addition, a
multivariate logistic regression analysis was performed to
determine if app use could predict changes in RPA, health
and lifestyle, and self-image. Outcome variables were ef-
fects on RPA (increased, not increased), health (healthier/
not healthier), alcohol consumption (more/not more),
smoking (more/not more), eat healthier (agree/disagree),
energy level (agree/disagree), performing sports is good
for me (agree/disagree), chance of maintaining physical
activity (agree/disagree), feeling better about oneself
(agree/disagree),
no
change
in
lifestyle
(agree/
disagree), lose weight (agree/disagree), and feel tired
more often (agree/disagree). In these logistic regres-
sion analyses, we controlled for age, gender, BMI, kilo-
metres per week before preparation and exercise
frequency in last year. Separate analyses were per-
formed for the 16 and the 6.4 km. The alpha level was
set at α ≤ 0.05 a priori.
Results
Of all invited runners 4307 (28 %) agreed to participate in
the survey, of which 2838 runners participated in the
16 km and 1341 in the 6.4 km. Table 1 presents the sub-
ject characteristics of male and female 16 and 6.4 km run-
ners. Hundred-twelve participants participated in both
distances and 507 participants reported too much missing
values and were therefore excluded. The type of apps used
by participants is shown in Fig. 1. Most participants used
Runkeeper (44.4 %) in their preparation. The category
‘other apps” was the second largest app type chosen by
participants (16.9 %), these were the apps that were not
mentioned in the answer options.
Differences app and non-app users
Baseline characteristics A significant association was
found between app use and gender for both distances.
More app users were female (16 km: Chi-squared = 4.90,
p = 0.027; 6.4 km:
Chi-squared = 9.61, p = 0.002). In
addition, app users were significantly younger compared
to non-app users (16 km: t = -12.09, df = 2456.56, p <
0.001; 6.4 km: t = −4.24, df = 879.92, p < 0.001) and
trained less often in a year (16 km: t = −5.58, df =
Dallinga et al. BMC Public Health (2015) 15:833
Page 3 of 9
2542.24, p < 0.001; 6.4 km: t = −2.44, df = 969.84, p =
0.015). In the 6.4 km runners, app use was associated
with BMI category (Chi-squared = 7.45, p = 0.024); app
users were more often overweight. We found a signifi-
cant association between app use and kilometres per
week that participants ran before the preparation phase
(16 km: Chi-squared = 87.48, p < 0.001; 6.4 km: Chi-
squared = 16.10, p = 0.003). In general, it seemed that app
users trained fewer kilometres before they had started the
preparation for the running event, compared to non-app
users. A significant association between app use and
duration of training period was found as well (16 km: Chi-
squared = 69.36, p < 0.001; 6.4 km: Chi-squared = 30.16,
p < 0.001). For the 16 km, there were more app users who
trained 12 weeks or more and who did not schedule a spe-
cific training period for this event compared to the non-
app users. For the 6.4 km, app users trained more often 6
to 11 weeks and 12 weeks or more compared to non-app
users, whereas non-app users more often did not train or
trained barely compared to app users.
Table 1 Subject characteristics of 16 and 6.4 km runners
16 km
6.4 km
Males
Females
Males
Females
Variable
M ± SD
M ± SD
M ± SD
M ± SD
Age (years)
42.19 ± 10.73
37.11 ± 10.26
42.01 ± 11.39
36.33 ± 10.31
Training sessions per year (n/year)
120.91 ± 56.81
121.31 ± 55.39
101.17 ± 57.90
99.97 ± 56.08
N (%)a
N (%)a
N (%)a
N (%)a
BMI category
Underweight
20 (1.0)
46 (2.2)
5 (0.5)
40 (4.1)
Normal weight
756 (36.1)
646 (30.9)
97 (10.0)
443 (45.5)
Overweight
481 (23.0)
143 (6.8)
140 (14.4)
249 (25.6)
Use of app during training
Yes
736 (28.5)
543 (21.0)
160 (13.5)
537 (45.3)
No
830 (32.1)
477 (18.4)
140 (11.8)
349 (29.4)
Duration training period
No training/ barely
114 (4.4)
38 (1.5)
37 (3.1)
97 (8.2)
1–5 weeks
129 (5.0)
88 (3.4)
46 (3.9)
100 (8.4)
6–11 weeks
183 (7.1)
125 (4.8)
33 (2.8)
117 (9.9)
12 weeks or more
225 (8.7)
218 (8.4)
38 (3.9)
129 (10.9)
No separate training period
909 (35.2)
546 (21.1)
142 (12.0)
433 (36.5)
Don’t know/no answer
6 (0.2)
3 (0.1)
5 (0.4)
9 (0.8)
Kilometres before
< 5 km/week
229 (9.1)
134 (5.3)
86 (7.6)
328 (28.9)
5–10 km/week
318 (12.6)
332 (13.2)
96 (8.5)
314 (27.7)
10–20 km/week
473 (18.8)
307 (12.2)
68 (6.0)
165 (14.5)
20–30 km/week
301 (11.9)
162 (6.4)
25 (2.2)
34 (3.0)
> 30 km/week
202 (8.0)
64 (2.5)
12 (1.1)
7 (0.6)
aTotal N varies due to missing values
.0
10.0
20.0
30.0
40.0
50.0
Get Running-app
DtD 2014 app
Adidas miCoach
App + Renate Wennemars
Myasics
Strava
Endomundo
Runtastic
Nike + iPod / I Phone app
Other
RunKeeper
Percentage of participants (%)
Fig. 1 Apps used in preparation for the 16 and 6.4 km recreational run
Dallinga et al. BMC Public Health (2015) 15:833
Page 4 of 9
Outcome variables Table 2 shows the differences be-
tween app users and non-app users in RPA, perceived
health and lifestyle, and self-image. App users increased
more often their RPA, felt healthier, ate healthier (6.4 km
no significant difference), felt more energetic, felt that they
had a higher chance of maintaining sport behaviour, felt
better about themselves, felt more like an athlete, changed
their lifestyle, stimulated others to perform sport and lost
weight.
Predictive ability of app use
Table 3 presents results of the logistic regression analyses
for each distance, corrected for age, gender, BMI, kilo-
metres per week before preparation and frequency of
participation in this running event. Logistic regression
analyses showed that for both 16 and 6.4 km runners, app
use was positively related to RPA and feeling healthier. In
addition, the app use was related to feeling better about
themselves, feeling more like an athlete, motivating others
to participate in running, and losing weight. Also, for the
16 km runners using apps was related to eating healthier,
feeling more energetic and reporting a higher chance to
maintain sport behaviour.
Discussion
Our main finding was that app use was positively related
to RPA, feeling healthier, changing lifestyle and self-image.
Also, use of apps was positively related to stimulating
Table 2 Differences between app users and non-app users in RPA, perceived health and lifestyle, and self-image
16 km
6.4 km
App use
No app use Chi2
P
App use
No app use Chi2
P
N (%)
N (%)
N (%)
N (%)
RPA
Decreased/same
624 (23.7)
821 (31.1)
55.49 < 0.001
467 (39.1)
369 (30.9)
17.22 < 0.001
Increased
689 (26.1)
504 (19.1)
246 (20.6)
112 (9.4)
Perceived health
Not healthier
497 (18.2)
722 (27.5)
72.71 < 0.001
294 (23.5)
268 (21.4)
18.36 < 0.001
Healthier
863 (31.6)
646 (23.7)
443 (35.4)
246 (19.7)
Smoking behavioura
More/equal
164 (43.3)
111 (29.3)
0.11
0.814
91 (52.3)
52 (29.9)
2.16
0.208
Less
64 (16.9)
40 (10.6)
24 (13.8)
7 (4.0)
Alcohol consumptionb
More/equal
901 (41.5)
897 (41.3)
1.63
0.211
441 (54.4)
296 (36.5)
0.28
0.619
Less
201 (9.3)
173 (8.0)
46 (5.7)
27 (3.3)
Eat healthier
Agree
496 (18.4)
420 (15.6)
10.71
0.001
221 (18.0)
129 (10.5)
3.76
0.052
Disagree
843 (31.3)
932 (34.6)
502 (40.8)
377 (30.7)
Feel more energetic
Agree
923 (34.3)
731 (27.2)
65.17 < 0.001
467 (38.1)
281 (22.9)
9.95
0.002
Disagree
412 (15.3)
623 (23.2)
255 (20.8)
223 (18.2)
Chance of maintaining sport behaviour
Agree
949 (35.3)
868 (32.3)
13.30 < 0.001
538 (44.0)
339 (27.7)
7.33
0.007
Disagree
389 (14.5)
481 (17.9)
183 (15.0)
163 (13.3)
I know that performing sport is not my thing Agree
21 (0.8)
28 (1.0)
0.97
0.387
12 (1.0)
14 (1.1)
1.82
0.226
Disagree
1313 (49.1)
1316 (49.1)
711 (58.0)
488 (39.8)
Feel better about myself
Agree
859 (32.0)
646 (24.1)
74.19 < 0.0001 492 (40.1)
257 (21.0)
37.60 < 0.0001
Disagree
475 (17.7)
703 (26.2)
229 (18.7)
248 (20.2)
Feel more like an athlete
Agree
605 (22.5)
422 (15.7)
55.40 < 0.0001 343 (28.0)
168 (13.7)
24.68 < 0.0001
Disagree
731 (27.2)
926 (34.5)
377 (30.8)
335 (27.4)
Changed lifestyle
Agree
913 (34.1)
796 (29.7)
25.01 < 0.0001 502 (40.9)
302 (24.6)
12.76 < 0.001
Disagree
421 (15.7)
550 (20.5)
220 (17.9)
204 (16.6)
Stimulating others to perform sport
Agree
657 (24.5)
566 (21.1)
14.65 < 0.001
384 (31.3)
217 (17.7)
12.02
0.001
Disagree
676 (25.2)
784 (29.2)
339 (27.6)
287 (23.4)
Losing weight
Agree
543 (20.2)
399 (14.8)
36.72 < 0.0001 270 (22.0)
125 (10.2)
21.61 < 0.0001
Disagree
794 (29.5)
955 (35.5)
453 (36.9)
380 (30.9)
Feel tired more often
Agree
97 (3.6)
84 (3.1)
1.17
0.282
52 (4.3)
38 (3.1)
0.08
0.824
Disagree
1237 (46.1)
1266 (47.2)
668 (54.7)
463 (37.9)
aThe participants who did not smoke were excluded
bThe participants who did not drink alcohol were excluded
Dallinga et al. BMC Public Health (2015) 15:833
Page 5 of 9
others to become active. Moreover, app use in 16 km
runners was positively related to feeling more energetic,
eating healthier and maintaining the sport behaviour. The
odds ratios ranged from 1.24 to 1.89. Additionally, for
RPA the explained variance was 41 % and 38 % for 16 km
and 6.4 km respectively. These findings are of high
importance considering that for app users the weekly
training volume prior to the preparation phase was lower
than non-app users.
These results corroborate with the findings of other
studies, in which app use seemed to have increased
physical activity and a healthy lifestyle [11, 14–17]. In
contrast to those studies, the focus in this study was on
mobile app use only. It should be noted that we did not
analyse the effect of app use, but we examined the use of
mobile apps in relation to physical activity, perceived
health and self-image. This relationship between app use
and perceived health and self-image in the preparation
of a running event has not been considered in previous
studies. Analysing this relationship is relevant, since it
provides insight in innovative and accessible ways to
encourage physical activity and a healthier life.
Although most results were comparable for 16 and
6.4 km runners, a few differences were found. In 16 km
runners, app use was related to eating healthier, feeling
more energetic and a higher chance to maintain sport
behaviour. The relationships between app use and these
variables did not reach significance level in the 6.4 km
runners. The “fun run” character of the 6.4 km may be a
first explanation for the differences found. Compared to
Table 3 Results of multivariate logistic regression with outcome measure RPA, perceived health and lifestyle
App use
Distance
OR (95 % CI)a
P
R2b
RPA
16 km
1.43 (1.16–1.75)
0.001
0.41
6.4 km
1.89 (1.34–2.65)
<0.001
0.38
Health
16 km
1.59 (1.33–1.90)
<0.0001
0.10
6.4 km
1.33 (1.02–1.73)
0.038
0.10
Alcohol consumption
16 km
1.06 (0.83–1.35)
0.651
0.04
6.4 km
1.57 (0.86–2.85)
0.143
0.03
Smoking behaviour
16 km
1.09 (0.71–1.69)
0.691
0.06
6.4 km
2.06 (0.80–5.30)
0.134
0.05
Eat healthier
16 km
1.24 (1.03–1.48)
0.022
0.02
6.4 km
1.24 (0.93–1.66)
0.150
0.04
Feel more energetic
16 km
1.68 (1.40–2.01)
<0.0001
0.08
6.4 km
1.13 (0.99–1.70)
0.055
0.05
I know that performing sport is not my thing
16 km
0.92 (0.44–1.75)
0.701
0.02
6.4 km
0.47 (0.19–1.03)
0.058
0.12
Chance of maintaining sport behaviour
16 km
1.24 (1.03–1.50)
0.021
0.02
6.4 km
1.31 (0.98–1.74)
0.067
0.02
Feel better about myself
16 km
1.75 (1.47–2.09)
<0.0001
0.07
6.4 km
1.84 (1.41–2.40)
<0.0001
0.07
Feel more like an athlete
16 km
1.69 (1.41–2.01)
<0.0001
0.05
6.4 km
1.67 (1.28–2.18)
<0.001
0.06
Did not change lifestyle
16 km
0.70 (0.58–0.83)
<0.0001
0.02
6.4 km
0.70 (0.53–0.92)
0.010
0.06
Motivated others to participate
16 km
1.43 (1.20–1.69)
<0.0001
0.02
6.4 km
1.45 (1.12–1.87)
0.005
0.03
Lost weight
16 km
1.57 (1.31–1.89)
<0.0001
0.06
6.4 km
1.72 (1.29–2.30)
<0.0001
0.09
Feel tired more often
16 km
1.03 (0.73–1.46)
0.877
0.04
6.4 km
0.70 (0.44–1.12)
0.140
0.03
aControlled for gender, age, BMI, training sessions per year and weekly training distance before training phase
bNagelkerke R2 [39]
Dallinga et al. BMC Public Health (2015) 15:833
Page 6 of 9
the 16 km run, participation in a 6.4 km run may not
require a long preparation phase and lifestyle changes.
In addition, we found that in the training phase most
16 km runners trained 10–20 km per week (37.3 %) and
20–30 km per week (27.3 %), whereas the largest part of
6.4 km runners trained 5–10 km per week (42.0 %) and
10–20 km per week (26.3 %). Thus another possible
explanation might be that the differences in weekly
training distance of 16 and 6.4 km runners combined
with a shorter preparation resulted in the inconsistent
findings. Previous literature has shown that running
improves aerobic fitness and cardiovascular function at
rest [26]. In a review, a fairly strong dose–response rela-
tionship between weekly training volume and cardiore-
spiratory fitness was shown for inactive and healthy
middle aged and elderly people [27]. This may explain
why the physical fitness of the 16 km runners increased
more compared to 6.4 km runners, resulting in a higher
perceived energy level. Potentially, there is a link be-
tween weekly training volume and eating behaviour as
well. To support this suggestion, Williams et al. showed
that a larger weekly running distance promoted a health-
ier eating pattern [28]. In addition, in that study a rela-
tionship was found between weekly running distance
and years spend in running, which might provide an
explanation for our finding that app use was related to a
higher chance to maintain the physical activity of the
longer distance runners compared to the shorter dis-
tance runners.
Previous studies have shown that participating in run-
ning events can encourage physical activity [26, 29]. How-
ever, maintaining an active lifestyle is difficult for many
[30]. Moreover, the gap between intention for being phys-
ically active and actually being active is large [31]. In many
of behaviour change models, such as the Fogg behaviour
model and the attitude, social influence and efficacy (ASE)
model, the behavioural intention is assumed to be most
important in changing behaviour [32, 33]. It would be
interesting to determine the impact of an app on behav-
iour determinants such as self-efficacy, attitude and social
influence. In addition, given that behaviour change theor-
ies (BCTs) are often relatively absent in apps, it would be
valuable to find out which of these theories are taken into
account in the app [34].
This study showed that the intention to maintain the
running behaviour was higher for the app users, therefore
app use may assist in decreasing drop-out of running and
encouraging physical activity. This is a very interesting
finding, since apps were more often used by overweight
participants and the participants in the 6.4 km run (who
trained less often). For these two groups physical activity
may need to be encouraged. Furthermore, a very interest-
ing finding was that app users more often encouraged
others to engage in running compared to non-app users.
This could be explained by the fact that some apps
contain features to interact with others, such as following
and supporting their activities [13]. This interaction com-
bined with the use of social media might motivate others
to be more active [35]. These findings suggest that the use
of mobile apps can contribute to the promotion of run-
ning and prevention of drop-out. Our findings may be
related to the new phenomenon of quantified self, which
means that people are measuring their health conditions
via wearables [36]. This new trend may actually be an
underlying element in the findings of this study.
Furthermore, when we look at practical implications,
we suggest that app use could be an additional stimulus
to the training program, because it provides an easy and
accessible tool to promote physical activity and a healthy
lifestyle. Given that the use of smart phones increases
[37, 38], a large amount of individuals can be reached
with health and fitness apps. Sport organizations and
employers may therefore recommend the use of apps in
the preparation of a running event. For instance, large
recreational running events often include a business run,
in which business teams can compete. The use of apps
may encourage employers to train more and live health-
ier. This data shows that app use is related to increased
physical activity and improved health. Moreover, fre-
quency of app use is higher in inexperienced and over-
weight participants. We could hypothesize that these
group of runners have some comparable characteristics
as inactive individuals. Therefore, our results could po-
tentially be transferred to inactive individuals.
Some limitations of this study need to be addressed.
At first, a self-reported, non-validated survey was used.
Second, a causality between app use and the outcome
variables cannot be determined. It remains unclear what
would be the cause and what would be the result; did
app use increase physical activity or did physical activity
encourage app use. The involvement of other underlying
causes should be considered as well. Randomized con-
trolled studies need to be performed to determine a
causal relationship. The third limitation was that several
types of apps were included. The most used app was
Runkeeper, but also a number of other apps were used.
It would be interesting to find out why people choose
certain apps and which features make an app popular.
Apps differ in their features and may differ in their
effectiveness as well. Therefore, the possibility that the
relationships found might be different for each app has
to be kept in mind, because the way apps present infor-
mation and provide feedback differs. As a fourth limita-
tion low explained variances for app use in relation to
most of the health and lifestyle outcomes were found.
Therefore we have to keep in mind that other factors,
such as psychological factors, contributed to the runner’s
lifestyle and self-image as well. At last, this study included
Dallinga et al. BMC Public Health (2015) 15:833
Page 7 of 9
individuals that were already active and motivated to
participate to a running event. However, considering the
problem of increased inactivity, it would be even more
interesting to conduct research on potential of app use in
promoting a healthy lifestyle in inactive individuals includ-
ing long-term consequences. Further research is needed
to determine which features would need to be included in
such an app.
Conclusion
In conclusion, our results showed that recreational run-
ners who used an app are more likely to be more physic-
ally active and feel and live healthier. These results suggest
that use of mobile apps has a beneficial role in the prepar-
ation of a running event, as it promotes health. Further
research is now needed to determine a causal relationship
between app use and physical and health related out-
comes. More specific, a randomized controlled trial (RCT)
needs to be developed and conducted. For instance, the
effect of one app such as Runkeeper could be examined
on weekly training distance and lifestyle. Another example
would be to develop and evaluate a physical activity and
health promotion app in a group of inactive individuals.
To gain insight in long-term effects, a follow-up survey
should be included as well.
Additional files
Additional file 1: Online Survey Dam tot Damloop. This file contains
the questions that were stated in the online survey used in this study.
The file starts with an introduction in which the purpose is explained and
confidentiality is guaranteed. (PDF 239 kb)
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
JD participated in the design of the study, carried out the statistical analyses,
interpreted the data and drafted the manuscript. MM was involved in the
analysis and interpretation of the data and helped to draft the manuscript.
LA was involved in the analysis and interpretation of the data and critically
revised the manuscript. HB was involved in the analysis and interpretation of
the data and critically revised the manuscript. MB participated in the design
of the study, acquisition of data, and analysis and interpretation of these
data. In addition, MB critically revised the manuscript. All authors read and
approved the manuscript. Also, all authors agree to be accountable for all
aspects of the work in ensuring that questions related to the accuracy or
integrity of any part of the work are appropriately investigated and resolved.
Acknowledgements
The authors would like to thank Le Champion for the cooperation. Our
gratitude also goes to Cees Vervoorn for his contributions to the study
development. This publication was supported by the Dutch national
program COMMIT.
Received: 29 April 2015 Accepted: 18 August 2015
References
1.
Chakravarty EF, Hubert HB, Krishnan E, Bruce BB, Lingala VB, Fries JF.
Lifestyle risk factors predict disability and death in healthy aging adults. Am
J Med. 2012;125(2):190–7.
2.
Leitzmann MF, Park Y, Blair A, Ballard-Barbash R, Mouw T, Hollenbeck AR, et
al. Physical activity recommendations and decreased risk of mortality. Arch
Intern Med. 2007;167(22):2453–60.
3.
Garber CE, Blissmer B, Deschenes MR, Franklin BA, Lamonte MJ, Lee IM, et al.
American College of Sports Medicine position stand. Quantity and quality of
exercise for developing and maintaining cardiorespiratory, musculoskeletal,
and neuromotor fitness in apparently healthy adults: guidance for prescribing
exercise. Med Sci Sports Exerc. 2011;43(7):1334–59.
4.
Hildebrandt VH, Bernaards C, Stubbe J. Trendrapport bewegen en
gezondheid 2010/2011. Leiden: TNO; 2013
5.
World Health Organization. Global health risks: mortality and burden of
disease attributable to selected major risks. Geneva: World Health
Organization; 2009.
6.
Church TS, LaMonte MJ, Barlow CE, Blair SN. Cardiorespiratory fitness and
body mass index as predictors of cardiovascular disease mortality among
men with diabetes. Arch Intern Med. 2005;165(18):2114–20.
7.
McKercher C, Sanderson K, Schmidt MD, Otahal P, Patton GC, Dwyer T, et al.
Physical activity patterns and risk of depression in young adulthood: a
20-year cohort study since childhood. Soc Psychiatry Psychiatr Epidemiol.
2014;49(11):1823–34.
8.
Smart N, Marwick TH. Exercise training for patients with heart failure: a
systematic review of factors that improve mortality and morbidity. Am J
Med. 2004;116(10):693–706.
9.
Schottenfeld D, Beebe-Dimmer JL, Buffler PA, Omenn GS. Current perspective
on the global and United States cancer burden attributable to lifestyle and
environmental risk factors. Annu Rev Public Health. 2013;34:97–117.
10.
Glynn LG, Hayes PS, Casey M, Glynn F, Alvarez-Iglesias A, Newell J, et al.
Effectiveness of a smartphone application to promote physical activity in
primary care: the SMART MOVE randomised controlled trial. Br J Gen Pract.
2014;64(624):e384–91.
11.
Stephens J, Allen J. Mobile phone interventions to increase physical
activity and reduce weight: a systematic review. J Cardiovasc Nurs.
2013;28(4):320–9.
12.
Middelweerd A, Mollee JS, van der Wal C, Brug J, Te Velde SJ. Apps to
promote physical activity among adults: a review and content analysis. Int J
Behav Nutr Phys Act. 2014;11(1):97.
13.
West JH, Hall PC, Hanson CL, Barnes MD, Giraud-Carrier C, Barrett J. There’s
an app for that: content analysis of paid health and fitness apps. J Med
Internet Res. 2012;14(3):e72.
14.
Foster C, Richards J, Thorogood M, Hillsdon M. Remote and web 2.0
interventions for promoting physical activity. Cochrane Database Syst Rev.
2013;30(9):CD010395.
15.
Fanning J, Mullen SP, McAuley E. Increasing physical activity with mobile
devices: a meta-analysis. J Med Internet Res. 2012;14(6):e161.
16.
Bort-Roig J, Gilson ND, Puig-Ribera A, Contreras RS, Trost SG. Measuring and
influencing physical activity with smartphone technology: a systematic
review. Sports Med. 2014;44(5):671–86.
17.
Naimark SJ, Madar Z, Shahar RD. The Impact of a Web-Based App (eBalance)
in Promoting Healthy Lifestyles: Randomized Controlled Trial. J Med Internet
Res. 2015;17(3):e56.
18.
Chatton A, Kayser BE. Self-reported health, physical activity and socio-
economic status of middle-aged and elderly participants to a popular road
running race in Switzerland: better off than the general population? Swiss
Med Wkly. 2013;143:1–9.
19.
Stevinson C, Hickson M. Exploring the public health potential of a mass
community participation event. J Public Health (Oxf). 2014;36(2):268–74.
20.
Lane A, Murphy N, Bauman A, Chey T. Randomized controlled trial to
increase physical activity among insufficiently active women following their
participation in a mass event. Health Educ J. 2010;69(3):287–96.
21.
Celie F, Faes M, Hopman M, Stalenhoef AF, Rikkert MGO. Running on age in
a 15-km road run: minor influence of age on performance. Eur Rev Aging
Phys Act. 2010;7(1):43–7.
22.
Marti B, Abelin T, Minder CE. Relationship of training and life-style to 16-km
running time of 4000 joggers. The '84 Berne “Grand-Prix” Study. Int J Sports
Med. 1988;9(2):85–91.
23.
Bottenburg M, Hover P. Evenementenlopers in beeld. Utrecht/’s-
Hertogenbosch: Universiteit Utrecht/W.J.H. Mulier Instituut; 2009.
Dallinga et al. BMC Public Health (2015) 15:833
Page 8 of 9
24.
World Health Organization. [http://apps.who.int/bmi/
index.jsp?introPage=intro_3.html] March 4,2015
25.
Mulier instituut: richtlijn sportdeelname onderzoek. [http://
www.sportdeelname.nl/methodieken/rso/] 4 Oct,2015
26.
Oja P, Titze S, Kokko S, Kujala UM, Heinonen A, Kelly P, et al. Health benefits
of different sport disciplines for adults: systematic review of observational
and intervention studies with meta-analysis. Br J Sports Med.
2015;49(7):434–40.
27.
Oja P. Dose response between total volume of physical activity and health
and fitness. Med Sci Sports Exerc. 2001;33(6 Suppl):S428–37. discussion
S452-3.
28.
Williams PT. Relationship of distance run per week to coronary heart disease
risk factors in 8283 male runners: the National Runners’ Health Study. Arch
Intern Med. 1997;157(2):191–8.
29.
Funk D, Jordan J, Ridinger L, Kaplanidou K. Capacity of mass participant
sport events for the development of activity commitment and future
exercise intention. Leisure Sciences. 2011;33(3):250–68.
30.
Marcus BH, Forsyth LH, Stone EJ, Dubbert PM, McKenzie TL, Dunn AL, et al.
Physical activity behavior change: issues in adoption and maintenance.
Health Psychol. 2000;19(1S):32.
31.
Rhodes RE, Bruijn G. How big is the physical activity intention–behaviour
gap? A meta‐analysis using the action control framework. Br J Health
Psychol. 2013;18(2):296–309.
32.
Fogg BJ. A behavior model for persuasive design. Proceedings of the 4th
International Conference on Persuasive Technology. 2009;40:7.
33.
de Vries H, Dijkstra M, Kuhlman P. Self-efficacy: the third factor besides
attitude and subjective norm as a predictor of behavioural intentions.
Health Educ Res. 1988;3(3):273–82.
34.
Direito A, Dale LP, Shields E, Dobson R, Whittaker R, Maddison R. Do
physical activity and dietary smartphone applications incorporate
evidence-based behaviour change techniques? BMC Public Health.
2014;14(1):646.
35.
Laranjo L, Arguel A, Neves AL, Gallagher AM, Kaplan R, Mortimer N, et al.
The influence of social networking sites on health behavior change: a
systematic review and meta-analysis. J Am Med Inform Assoc.
2015;22(1):243–56.
36.
Swan M. Sensor mania! the internet of things, wearable computing,
objective metrics, and the quantified self 2.0. J Sens Actuator Netw.
2012;1(3):217–53.
37.
Entner R. Smartphones to overtake feature phones in US by 2011. [http://
www.nielsen.com/us/en/insights/news/2010/smartphonesto-overtake-
feature-phones-in-u-s-by-2011.html], Aug 24, 2015. Nielsen Wire. 2010.
38.
Van den Bighelaar S, Akkermans M. Gebruik en gebruikers van sociale
media. The Hague: Centraal Bureau voor de Statistiek; 2013.
39.
Nagelkerke NJ. A note on a general definition of the coefficient of
determination. Biometrika. 1991;78(3):691–2.
Submit your next manuscript to BioMed Central
and take full advantage of:
• Convenient online submission
• Thorough peer review
• No space constraints or color figure charges
• Immediate publication on acceptance
• Inclusion in PubMed, CAS, Scopus and Google Scholar
• Research which is freely available for redistribution
Submit your manuscript at
www.biomedcentral.com/submit
Dallinga et al. BMC Public Health (2015) 15:833
Page 9 of 9
| App use, physical activity and healthy lifestyle: a cross sectional study. | 08-28-2015 | Dallinga, Joan Martine,Mennes, Matthijs,Alpay, Laurence,Bijwaard, Harmen,Baart de la Faille-Deutekom, Marije | eng |
PMC6283599 | RESEARCH ARTICLE
Optimizing beat synchronized running to
music
Jeska BuhmannID1*, Bart Moens1, Edith Van Dyck1, Dobromir Dotov2, Marc Leman1
1 IPEM, Ghent University, Ghent, Belgium, 2 LIVELab, McMaster University, Hamilton, Canada
* [email protected]
Abstract
The use of music and specifically tempo-matched music has been shown to affect running
performance. But can we maximize the synchronization of movements to music and does
maximum synchronization influence kinematics and motivation? In this study, we explore
the effect of different types of music-to-movement alignment strategies on phase coher-
ence, cadence and motivation. These strategies were compared to a control condition
where the music tempo was deliberately not aligned to the running cadence. Results show
that without relative phase alignment, a negative mean asynchrony (NMA) of footfall timings
with respect to the beats is obtained. This means that footfalls occurred slightly before the
beat and that beats were anticipated. Convergence towards this NMA or preferred relative
phase angle was facilitated when the first music beat of a new song started close to the
step, which means that entrainment occurred. The results also show that using tempo and
phase alignment, the relative phase can be manipulated or forced in a certain angle with a
high degree of accuracy. Ensuring negative angles larger than NMA (step before beat)
results in increased motivation and decreasing cadence. Running in NMA or preferred rela-
tive phase angles results in a null effect on cadence. Ensuring a positive phase angle with
respect to NMA results in higher motivation and higher cadence. None of the manipulations
resulted in change in perceived exhaustion or a change in velocity. Results also indicate
that gender plays an important role when using forced phase algorithms: effects were more
pronounced for the female population than for the male population. The implementation of
the proposed alignment strategies and control of beat timing while running opens possibili-
ties optimizing the individual running cadence and motivation.
Introduction
Sports and exercise activities are generally believed to benefit from music listening. Under par-
ticular conditions music has been shown to capture attention, raise spirits, trigger a range of
emotions, alter or regulate mood, evoke memories, increase work output, heighten arousal,
induce states of higher functioning, reduce inhibitions, and encourage rhythmic movement [1,
2]. Effects of music during exercise can even be enhanced when certain types of music are con-
sidered [3, 4], especially when a certain level of synchrony between the musical stimuli and the
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
1 / 21
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Buhmann J, Moens B, Van Dyck E, Dotov
D, Leman M (2018) Optimizing beat synchronized
running to music. PLoS ONE 13(12): e0208702.
https://doi.org/10.1371/journal.pone.0208702
Editor: Ramesh Balasubramaniam, University of
California Merced, UNITED STATES
Received: February 19, 2018
Accepted: November 21, 2018
Published: December 6, 2018
Copyright: © 2018 Buhmann et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All data files and
scripts are available via https://doi.org/10.5281/
zenodo.1467592.
Funding: This research has been supported by
BeatHealth (contract #610633), a collaborative
project funded by the European Commission under
the Seventh Framework Programme: https://
cordis.europa.eu/project/rcn/110990_en.html. The
funders had no role in study design, data collection
and analysis, decision to publish, or preparation of
the manuscript.
Competing interests: The authors have declared
that no competing interests exist.
listener’s movements occurs [5–8]. A much-researched topic concerns the synchrony between
music and running, measured by beat and footfall as markers of the rhythm that drives syn-
chronized running. However, synchronizing one’s footfall with the musical beat is not a
straightforward endeavor. Some people easily synchronize, while others need an instruction to
synchronize [9]. Some music facilitates movement (such as groove [10]) and running (such as
activating music), while other music is less effective [3, 4]. In short, the use of music and specif-
ically tempo-matched music has been shown to affect running performance, but it is an open
question whether we can maximize the synchronization of movements to music and whether
this maximum synchronization somehow influences kinematics and motivation.
In this context, we define synchronization as the stable maintenance over time of the senso-
rimotor coupling between beat and footfall. We define entrainment as the process that realizes
the sensorimotor coupling, more specifically the process that brings the (perceived) beat and
the (performed) footfall towards a stable timing. Rather than through pull and push forces (as
in a dynamical system) it is straightforward to assume that prediction error minimization is a
mechanism for entrainment [11]. For an in depth explanation of the factors that determine the
strength of the sensorimotor coupling and entrainment, see Leman [12].
The tempo of running is expressed in steps per minute (SPM), i.e. cadence, and for music it
is expressed in beats per minute (BPM). The tempo of running, or cadence, refers to the fre-
quency of steps, whereas speed (expressed in km/h) refers to the total distance travelled in a
certain amount of time. In previous studies, the preferred exercise intensity was often mea-
sured before the actual experiment and the music tempo was generally rather coarsely matched
to a subject’s spontaneous or comfort tempo (e.g., within a 10% range of the assessed cadence)
[13, 14]. However, such an approach disregards the fact that the comfort tempo of an individ-
ual might be different at the time of the test, or that it might fluctuate during the test period.
As a result, the contrast between the tempo of the music and the tempo of the exercise perfor-
mance is likely to become too considerable to enable spontaneous entrainment. For that rea-
son, we measure comfort tempo in the first part of each running task. In addition, a study by
Van Dyck et al. [15] unveiled that uninstructed synchronization of running cadence to musical
tempo occurs spontaneously when the tempo of the music does not deviate more than 2.5%
from the initial running cadence. This finding highlights that conditions of entrainment yield
affordances for sensorimotor adaptation to be effective, provided that the tempo of the music
can be matched to the person’s comfort tempo with high accuracy. Other research that
addressed the particular relations between entrainment conditions and sensorimotor adapta-
tion is, however, scarce [12, 16].
Another aspect that is likely to be important in terms of spontaneously manipulating run-
ning cadence is the anticipation effect reported in sensorimotor synchronization studies
(SMS). For example, people typically tend to tap a little bit before the auditory stimulus, which
is indicated by a slightly negative phase angle of the tap relative to the click, also referred to as
negative mean asynchrony (NMA) [11]. They do this, presumably, because the sensing of the
tapping has a longer delay than the sensing of the click [17]. Synchronization at brain level
therefore means tapping before the click. Being able to continuously manipulate the exact
moment of the beat might therefore prove to be a very accurate way to match and influence
people’s running capacity.
In the present study we focus on very precise sensorimotor (beat-footfall) alignment strate-
gies during running. In fact, some of those strategies manipulate only the timing of the beat
and it is expected that the timing of the footfall is somehow influenced by this manipulation.
Moens et al. [16] describe four different strategies that manipulate sensorimotor alignment.
However, these alignment strategies were not yet compared with each other within a single
study, nor with the same musical test samples. Therefore, the purpose of this study is to
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
2 / 21
contribute to a high definition sensorimotor alignment strategy that accounts for each person’s
sensorimotor ability. In particular, we aim at comparing different beat-footfall alignments
using a randomized experimental setup where different strategies for supporting such align-
ments are applied. We hypothesized that strategies employing phase alignment to adjust the
time of the musical beat relative to the time of the footfall during running would result in the
highest level of sensorimotor synchronization and have a bigger effect on kinematics (cadence
and speed) compared to strategies using period alignment instead. In addition, we expect that
a high degree of synchronization would have an effect on motivation.
While our main aim is to discover whether different music-to-movement alignment strate-
gies affect kinematics (cadence and speed) and/or motivation in distinct ways, we also want to
address possible gender differences in entrainment capacity. The study by Van Dyck et al. [15]
unveiled higher levels of entrainment for female compared to male runners. Other research
shows that, when people are requested to rate the motivational qualities of musical excerpts,
women pay closer attention to the rhythmical qualities of the stimuli compared to their male
counterparts [18]. In addition, musical preference seems to affect women differently than
men: when female runners listen to highly preferred stimuli, they tend to perform better than
when listening to non-preferred music. In comparison, musical preference does not seem to
affect the performance of male exercisers [19]. From the above-described findings, we
expected to uncover differences between men and women regarding their ability to entrain
and synchronize with the beats of the music. Such differences would be particularly relevant in
terms of designing a more individualized music-technology approach.
Materials and methods
Subjects
To establish sample size, a power analysis for a repeated-measures design was conducted using
GPower 3.1.9.2 [20]. Based on a small effect size (.25) with alpha set at .05 and power at .95, it
was estimated that at least 28 participants would be required. In total, 36 healthy, adult partici-
pants (19 males) took part in the study. All participants were recreational runners (Mage =
31.22 years; SDage = 8.13 years) and indicated to be capable of running 30 minutes continu-
ously. Of all participants, 38.89% were trained in music (Pearson Chi-Square test showed no
significant relation between gender and musical background, χ2(1) = 1.22, p = .27). In addi-
tion, about half of them (55.56%) reported to generally run without music, 22.22% indicated to
usually run with music, and 22.22% ran both with and without musical accompaniment. The
study was approved by the Ethics Committee of the Faculty of Arts and Philosophy of Ghent
University and was in accordance with the statements of the Declaration of Helsinki. Written
informed consent was obtained from all the participants before the start of the experiment.
Experimental design
Stimuli.
A music database consisting of music tracks in the tempo range of 120–200 BPM
(the range of natural running cadence) was created. The database included musical stimuli
from a previous running experiment [15]. Using the Brunel Music Rating Inventory-2 (BMRI-
2) test [21], all stimuli were rated as highly motivational for running. Additional tracks were
selected to ensure complete coverage of the tempo range. In total, 43 tracks with clear beat infor-
mation were selected. The tempo stability throughout each entire track was validated and intros
lacking clear beats were cut from the stimuli using Audacity (http://audacity.sourceforge.net).
BeatRoot [22] was applied to track the beats and tempo of each music track, while Adobe Audi-
tion (http://www.adobe.com) was used to normalize perceived loudness and minimize possible
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
3 / 21
imbalances in sound pressure level using the ITU-R BS.1770-3 standard at -23 LUFS which is
commonly used in audio broadcasts [23].
Apparatus.
Data was collected using a 7" tablet (Panasonic FZ-M1) running Windows
8.1, which was strapped to a backpack. In addition, a pair of sensors, headphones, and a man-
agement computer was employed. The tablet operated as the main hub that handled incoming
sensor data and provided the musical stimuli.
To detect footfall instants, participants were equipped with two iPods (4th generation); one
attached at each ankle. Using the Sensor Monitor Pro application on the iPods, data from
accelerometers and gyroscopes was streamed wirelessly to the tablet at a sampling rate of 100
Hz. Step timings were extracted from the signal using an approach based on Pappas et al. [24].
Speed measurements were performed using a sonar system (MaxBotix LV-MaxSonar-EZ:
MB1010) connected to the tablet through a Teensy 3.1 micro-controller. It detected marker
rods of 1.90 m high, placed at a regular interval of 10 m around the running track. Through
computation of the time it took to cover each interval, absolute speed was determined. The
analogue signal was sampled at 30 Hz and digitized using the Teensy.
The wireless connection between the tablet, iPods, and management computer was provided
through a Wi-Fi router (TP-Link M5360), firmly strapped to the backpack, ensuring reliable
communication between all crucial components. The management computer was applied to
initiate the experimental sessions and to monitor sensor data in real-time. Musical tempi were
manipulated using a phase vocoder based on the technology of E´lastique Pitch of ZPlane.de
[25]. The phase vocoder manipulates tempo in real-time without modifying pitch using a com-
bination of frequency and time-domain methods. The system logged all data and calculations in
real-time. Music tempo was adapted based on the selected alignment strategy (for the imple-
mentation of the music alignment strategies, see Moens et al. [16]). Finally, the aligned music
was sent back to the participant using Sennheiser HD60 headphones connected to the tablet.
Procedure.
All experiments took place in the Flanders Sports Arena of Ghent, Belgium.
After being equipped, participants were asked to run on a 200 m running track for five minutes
continuously, and this for six consecutive times. In each of the six 5-minute runs, a different
alignment strategy was tested and it was ensured that all orders could occur only once.
Participants were instructed to run at their own comfortable pace. No information was dis-
tributed concerning the purpose of the experiment and all participants ran in solo conditions.
After each 5-minute run, participants were allowed to take a break for several minutes in
which they rated their perceived exertion (RPE) on the Borg Scale [26]. This way we could
determine whether fatigue was influenced by the different music alignment strategies. In addi-
tion, they rated the level of physical enjoyment on the 8-item version of the Physical Activity
Enjoyment Scale (PACES) [27, 28], a single factor scale to assess the level of enjoyment during
a physical activity in adults across exercise modalities.
Each of the 5-minute runs started with 25 seconds of silence, followed by five musical
excerpts of equal length (55 s) with an original tempo approaching the average cadence of the
last seven footsteps. Musical tempo was then manipulated based on the selected alignment
strategy, presented in the following section.
Conditions or alignment strategies.
In the control alignment strategy (S0), music and
running performance behave in a completely allochronic fashion (in our case: music is played
20 BPM faster/slower than the assessed running cadence). This strategy is used as a control
condition to compare against five other music-to-movement alignment strategies.
Two strategies (S1, S2) involve the alignment of the music tempo to the runner’s cadence.
The tempo matching occurs either at the beginning of a song only (S1), or continuously
throughout the exercise (S2). Tempo-matching alone, however, does not consider the exact
matching of the musical beats to the footfall instants; hence it neglects the phase (or the time
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
4 / 21
between musical beat and footfall) while it maintains the period (for the musical beat and foot-
fall, independently). In the adaptive tempo conditions, participants were still able to change
their timing with respect to the beats as it takes several steps to determine the runners’ cadence.
Three strategies (S3, S4, S5) involve the alignment of the music tempo of the runner’s cadence
using relative phase angle manipulation. The relative phase angle expresses the footfall-beat time
as a segment of the previous beat-period, which is defined as having 360˚. Steps and beats are
recurring events, hence cyclic in nature. Hence, the difference in timing between a beat and the
nearest footfall can be expressed with a relative phase angle (-180˚ to +180˚), where a 0˚ angle
indicates that the beat and footfall instant coincide exactly. Alignment strategies that employ
such phase manipulation minimize the relative phase angle between the beat and footfall instant,
and therefore, they drive the alignment to any desired relative phase angle. In S3 the relative
phase angle is minimized to 0˚ once, at the beginning of the exercise. The tempo of the music is
however adapted continuously. The final two strategies (S4, S5) involve continuous relative phase
angle adaptation starting at the beginning of the exercise. S4 guides the runner towards a prede-
fined phase angle by adjusting the phase and hence tempo of the music at each step, thus at dis-
crete timing intervals. In previous research, feedback from participants indicated that such a
music adaptation sometimes felt unnatural or forced due to sudden tempo changes. Hence a new
strategy is introduced (S5) based on adaptive oscillators [29], which results in continuous relative
phase alignment towards a predefined relative phase angle using smoother tempo adaptations.
The first experiment in the present study intended that S4 and S5 guided the runner
towards a 0˚ relative phase angle, or a perfect synchrony between footfalls and musical beats.
However, after the experiment it became clear that an inaccuracy in the calibration led both
alignment strategies towards -70˚ instead of 0˚. This resulted in musical beats occurring after
the footfall, while they were intended to occur simultaneously. For clarity, we henceforth refer
to S4 and S5 as strategies with a configurable relative phase angle. A second (follow-up) experi-
ment was then performed to assess the influence of our initially intended S4 [0˚] and S5 [0˚]
strategies, which is reported subsequently. Based on the results of the first experiment, we also
added an additional condition, which forced a +30˚ angle (beat before the step), and this is
also reported later in this paper. In the discussion we consider the results from both experi-
ments. A summary of the different strategies is provided in Table 1.
Although we are aware of the phenomenon of NMA in SMS studies, little is known about
the magnitude of such a NMA for running and whether this is similar for all participants. We
therefore decided to use 0˚ as a target relative phase and discuss our results taking into account
the knowledge on NMA.
Measurements
Cadence and speed. We examined the effect of the different alignment strategies on kine-
matic parameters such as cadence and speed. Average cadence and speed values during music
playback are compared to those in the preceding 25 seconds of silence. The resulting depen-
dent variables are expressed in percentages, where zero indicates no difference, while a nega-
tive or positive value indicates a respective decrease or increase in cadence or speed compared
to the silent part of the condition.
Synchronization and phase angles.
The level of synchronicity with the music, or rather,
the stability and timing of the relation between a runner’s footfall and the musical beat, is typi-
cally represented by the resultant vector or R. The length of this vector is a measure of tempo
entrainment, ranging from zero to one with one representing perfect entrainment [30]. In addi-
tion, the angle of the resultant vector represents the average relative phase angle and reveals
whether footfall instants occur before the beat is played (negative phase) or after (positive phase).
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
5 / 21
The resultant vector represents an average of a time period, and is calculated based on a distribu-
tion of individual phase angles or timing differences between each step and the closest beat [31].
Each relative phase angle is calculated with the formula below, where St refers to a step at time t.
B1 is the time of the beat that occurred before St and B2 is the time of the first beat after St.
φ ¼ 360 St Results
Synchronization
A Friedman’s ANOVA showed a main effect of the strategy on resultant vector length, χ2(5) =
147.705, p < .001. Wilcoxon tests were used to follow up this finding. A Bonferroni correction
was applied and so all effects are reported at a .003 level of significance. Results reveal that all
six strategies differ with respect to the resultant vector length, except S2 versus S3 and S4
[-70˚] versus S5 [-70˚]. All comparisons are summarized in Table 2.
The circular Wheeler-Watson Mardia test was used to verify that all phase angle distribu-
tions were significantly different and thus not homogenous (W = 5113.6, p < .001), meaning
the alignment strategies had a significant influence on the relative phase angles.
Cadence
One of the dependent variables of interest is the change in cadence from running in silence to
running with music. A 2x6 mixed-design ANOVA test with gender (male, female) as between-
subjects variable and condition (S0 to S5 [-70˚]) as within-subjects variable revealed a signifi-
cant main effect of the strategy on the change in cadence, F(5,170) = 16.46, p < .001. Contrasts
revealed that for S4 [-70˚], F(1, 34) = 17.23, p < .001, r = .58, and S5 [-70˚], F(1, 34) = 29.48, p
< .001, r = .68, running cadence decreased significantly more (M = -1.81%, SE = 0.37, and M
= -2.15%, SE = 0.34 respectively) compared to S0 (M = -0.53%, SE = 0.27).
There was no significant main effect of gender, indicating that on average there were no dif-
ferences in change in cadence between male (M = -1.06%, SE = 0.15) and female participants
(M = -0.93%, SE = 0.22), F(1,34) < 1, p = .78, r = .05.
However, an interaction effect between the strategy and the gender of the participant was
observed, F(5, 170) = 4.97, p < .001, indicating that the change in cadence differed between
men and women for different strategies. Contrasts were performed, revealing interaction
effects between gender x S0 x S4 [-70˚], F(1, 34) = 9.10, p = .005, r = .46, and gender x S0 x S5
[-70˚], F(1, 34) = 6.40, p = .016, r = .40. This indicated that although, for both males and
Table 2. Significant differences in phase coherence (resultant vector length R).
Comparisons (Mdn)
z
pa
r
S0 (0.03) vs. S1 (0.65)
-5.232
< .001
-.62
S0 (0.03) vs. S2 (0.80)
-5.232
< .001
-.62
S0 (0.03) vs. S3 (0.80)
-5.232
< .001
-.62
S0 (0.03) vs. S4 [-70˚] (0.94)
-5.232
< .001
-.62
S0 (0.03) vs. S5 [-70˚] (0.94)
-5.232
< .001
-.62
S1 (0.65) vs. S2 (0.80)
-3.268
.001
-.39
S1 (0.65) vs. S3 (0.80)
-3.991
< .001
-.47
S1 (0.65) vs. S4 [-70˚] (0.94)
-5.232
< .001
-.62
S1 (0.65) vs. S5 [-70˚] (0.94)
-5.232
< .001
-.62
S2 (0.80) vs. S3 (0.80)
-0.644
.519
-.08
S2 (0.80) vs. S4 [-70˚] (0.94)
-5.185
< .001
-.61
S2 (0.80) vs. S5 [-70˚] (0.94)
-5.059
< .001
-.60
S3 (0.80) vs. S4 [-70˚] (0.94)
-5.122
< .001
-.60
S3 (0.80) vs. S5 [-70˚] (0.94)
-5.001
< .001
-.59
S4 [-70˚] (0.94) vs. S5 [-70˚] (0.94)
-0.055
.956
-.01
a p values were calculated with Wilcoxon signed-rank tests comparing all six alignment strategies with each other.
https://doi.org/10.1371/journal.pone.0208702.t002
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
7 / 21
females, cadence decreased substantially during S4 [-70˚] and S5 [-70˚] compared to S0, this
decrease was more pronounced for female runners.
Speed
No main effect of the type of strategy on change in speed was uncovered, F(5, 165) = 1.02, p =
.407, nor was there a significant main effect of gender, F(1, 33) = 1.36, p = .25, r = .20. Besides,
there was no interaction effect between strategy and gender, F(5, 165) = 2.16, p = .061.
Motivation and perceived exertion
No significant differences were revealed for perceived exertion (RPE) with a Friedman’s
ANOVA, χ2(5) = 6.64, p = .25.
Wilcoxon signed-rank tests (comparing each strategy with S0) were performed on the
scores of the PACES scale. A Bonferroni correction was applied and so all effects are reported
at a .01 level of significance. The motivational scores were higher for S5 [-70˚] (Mdn = 71.38)
compared to S0 (Mdn = 67.25), z = -2.662, p = .008, r = -.31. None of the other strategies dis-
played significant differences in motivation when compared to the allochronic strategy (S0).
Table 3 summarizes the motivational comparisons. The median value for S5 [-70˚] is not that
different from S1-4. The reason, however, that ratings for S5 [-70˚] are significantly different
from S0, while the other alignment strategies are not, is due to differences in ranking: a higher
percentage of the ratings where in favor of S5 [-70˚] over S0 than in the other comparisons.
Follow-up experiment
The data analysis of the first experiment showed surprising results of S4 and S5, leading to an
investigation of the apparatus’ calibration. As we noted earlier, S4 and S5 were initially aimed to
obtain a 0˚ phase synchronization. However, the system incorrectly forced a -70˚ relative phase
angle. We believed that the non-intended forced -70˚ angle was the main cause of the observed
cadence decrease and motivational increase for these strategies. Therefore, it was decided to do
a follow-up experiment introducing correct 0˚ strategies. As the procedure and apparatus is
almost identical, we only elaborate on the differences and results for cadence and motivation.
Materials and methods
Tests took place at the same location of the first experiment. We recruited 11 of the initial par-
ticipants (6 female, 5 male, Mage = 39.27 years; SDage = 9.82 years) to have a similar population
and to be able to compare both experiments.
Strategies S0, S1 and S2 were identical to the initial experiment. The -70˚ strategies (S3 [-70˚],
S4 [-70˚], S5 [-70˚]) were replaced by their intended counterparts (S3 [0˚] S4 [0˚] and S5 [0˚]),
where the beats coincide with the footfalls as initially intended. One additional strategy was added
Table 3. Differences in motivation (PACES ratings).
Comparisons (Mdn)
z
pa
r
S0 (67.25) vs. S1 (69.88)
-.718
.487
-.08
S0 (67.25) vs. S2 (73.19)
-1.987
.048
-.23
S0 (67.25) vs. S3 (73.31)
-1.327
.196
-.15
S0 (67.25) vs. S4 [-70˚] (71.75)
-1.581
.109
-.18
S0 (67.25) vs. S5 [-70˚] (71.38)
-2.662
.008
-.31
a p values were calculated with Wilcoxon signed-rank tests comparing five alignment strategies with the allochronic
control condition (S0).
https://doi.org/10.1371/journal.pone.0208702.t003
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
8 / 21
to explore possible inverse effects of the negative phase angle, namely S5 [+30˚]. This strategy
placed the beat slightly before the footfall, implying that the musical beats preceded the runner.
For methodological reasons, this S5 [+30˚] strategy occurred at the end of the experiment and
was optional, to exclude influences of potential dropouts on our comparison with the earlier
experiment. A summary of the different strategies is provided in Table 1.
Results
All 11 participants completed all seven conditions. In total, two out of 77 trials were invalid
due to sensor errors. Given our initial experimental results, this section focuses on phase
angles, cadence, and motivation.
Phase angles and synchronization.
As for the initial experiment, a repeated-measures
ANOVA showed a main effect of the strategy on resultant vector length, F(6,54) = 103.794, p
< .001. The circular Wheeler-Watson Mardia test was used to verify that all phase angle distri-
butions were significantly different and thus not homogeneous (W = 1664.7, p < .001), mean-
ing that the alignment strategies had a significant influence on the relative phase angles.
Cadence.
In order to confirm that the retrieved effect on cadence from the initial experi-
ment was due to the phase angle being targeted at -70˚, we expected to find no effects on
cadence for S4 and S5 targeted at 0˚. Therefore, as in the initial experiment, a 2x6 mixed-
design ANOVA test with gender (male, female) as between-subjects variable and condition
(S0 to S5 [0˚]; excluding S5 [+30˚]) as within-subjects variable was performed. Indeed, no
main effect of the strategy on the change in cadence was revealed, F(5,40) = 1.210, p = .322. In
addition, no significant main effect of gender was observed, F(1,8) = 4.128, p = .077, r = .58,
nor an interaction effect between the strategy and the gender of the participant, F(5, 40) =
1.149, p = .351.
When we included S5 [+30˚], and performed a 2x7 mixed-design ANOVA, we did find a
main effect on cadence, F(6,48) = 3.693, p = .004, and a small but significant gender effect, F
(1,8) = 5.412, p = .048, r = .64. No gender x strategy interaction effect was found, F(6,48) =
1.306, p = .273.
Concerning the impact on cadence of different target phase angles in the configurable
phase angle strategy (S5), additional tests were performed. To compare S5 [-70˚] from our ini-
tial experiment with S5 [0˚] from the follow-up experiment an independent samples t-test was
executed. This test revealed that the decrease in cadence for S5 [-70˚] was significantly larger
(M = -2.15%, SE = 0.34) than for S5 [0˚], (M = -0.09%, SE = 0.51), t(45) = -3.034, p = .004, r =
.41. Furthermore, a paired samples t-test was used to compare S5 [+30˚] with S5 [0˚] from the
follow-up experiment. Results showed that the change in cadence for S5 [+30˚] was signifi-
cantly larger and even positive (M = 0.68%, SE = 0.59) compared to S5 [0˚], (M = -0.40%,
SE = 0.44), t(9) = -2.970, p = .016, r = .70.
Motivation.
Wilcoxon signed-rank tests (comparing each strategy with S0) were per-
formed on the scores of the PACES scale. A Bonferroni correction was applied and so all
effects are reported at a .008 level of significance. Only the motivational scores for S5 [+30˚]
showed a trend towards being significantly higher (Mdn = 44.00) than for S0 (Mdn = 38.00), z
= -2.661, p = .008, r = -.57. No other differences were found. Table 4 summarizes the motiva-
tional comparisons.
Discussion
The goal of the present study is whether we maximize the synchronization of perceived musical
beat and footfall during running, and whether this maximum synchronization influences kine-
matics and motivation. Table 5 summarizes the results, showing the different phase angle
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
9 / 21
distributions for conditions and experiments. We then consider the effects of synchronization,
cadence, and motivation. The last section discusses possible gender differences for these effects.
Preferred phase angles, synchronization, and phase locking
The data from the different alignment strategies reveal differences and preferences in synchro-
nization behavior. Fig 1 provides a visual summary of the different strategies that were tested
in terms of timing information of the step versus the beat (i.e. the relative phase angle) and the
synchronization stability or phase coherence (i.e. resultant vector length).
In Fig 1, condition S0 clearly visualizes what the control condition was designed for, i.e.
that runners don’t synchronize, nor phase-lock (because they cannot). S0 can thus provide a
baseline or comparison for later tests.
Since strategy S1, S2, and S3 did not use continuous relative phase angle manipulation, they
gave a good idea of the (average over song) relative phase angle when runners ran with music
at their (previously measured) preferred tempo. Once the song starts, the tempo remains con-
stant (S1), or is continuously adapted to match the runners’ preferred tempo, which may
change during the song (S2 and S3). These conditions allowed for self-selected relative phase
Table 4. Differences in motivation (PACES ratings) for follow-up experiment.
Comparisons (Mdn)
z
pa
r
S0 (38.00) vs. S1 (42.00)
-1.721
.085
-.37
S0 (38.00) vs. S2 (42.00)
-1.736
.083
-.37
S0 (38.00) vs. S3 (41.00)
-0.949
.343
-.20
S0 (38.00) vs. S4 [0˚] (41.00)
-1.961
.050
-.42
S0 (38.00) vs. S5 [0˚] (42.00)
-1.661
.097
-.35
S0 (38.00) vs. S5 [+30˚] (44.00)
-2.661
.008
-.57
a p values were calculated with Wilcoxon signed-rank tests comparing six alignment strategies with the allochronic
control condition (S0).
https://doi.org/10.1371/journal.pone.0208702.t004
Table 5. Circular descriptives of the phase angle distributions for all strategies. Initial experiment data is abbreviated as I.E., while follow-up experiment data is abbre-
viated as F.U.E. For each alignment strategy the distribution of relative phase angles is described, using the following parameters: the mean angle ø representing the mean
direction, the resultant vector length R and the circular variance CV, angular deviation s (dispersion around the mean), circular skewness b (asymmetry) and the circular
kurtosis k (peak load). See [30] for more information.
Alignment Strategy
S0
S1
S2
S3
S4–70˚
S4 0˚
S5–70˚
S5 0˚
S5 30˚
Short name
Allochronic Music
Fixed Tempo
Continuous
Tempo
Adaptation
Continuous
Tempo
Adaptation, in
phase start
Forced Phase
Coherence—
Algorithmic
Forced Phase Coherence—
Adaptive Oscillator
Forced phase angle start / goal
-70˚
0˚
-70˚
0˚
-70˚
0˚
+30˚
Experiment
I.E.
F.U.E.
I.E.
F.U.E.
I.E.
F.U.E.
I.E.
F.U.E.
I.E.
F.U.E.
I.E.
F.U.E.
F.U.E.
N Steps
26326
7246
26445
7285
26275
7451
26199
6709
25980
7466
25773
7419
6772
N Songs
180
55
180
55
180
55
180
50
180
55
180
55
50
Mean relative phase angle ø
138˚
165˚
-13.7˚
-17.7˚
-8.04˚
3.61˚
-22.1˚
-39.2˚
-70.3˚
-1.95˚
-75.5˚
-9.66˚
17.4˚
Resultant Vector Length
j
R
j
0.0034
0.0085
0.424
0.402
0.196
0.136
0.671
0.667
0.911
0.912
0.925
0.936
0.895
Circular Variance CV
0.997
0.991
0.576
0.598
0.804
0.864
0.329
0.333
0.0891
0.0877
0.0747
0.0645
0.105
Angular Deviation s
1.4
1.4
1.1
1.1
1.3
1.3
0.81
0.82
0.42
0.42
0.39
0.36
0.46
Circular Skewness b
0.00048
-0.0006
0.02
0.03
-0.014
-0.035
-0.031
0.024
-0.0049
0.0074
0.0011
0.02
0.011
Circular Kurtosis k
-0.0016
0.0026
0.22
0.19
0.062
0.04
0.46
0.36
0.74
0.74
0.79
0.82
0.74
https://doi.org/10.1371/journal.pone.0208702.t005
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
10 / 21
angles. Since S2 and S3 used an adaptive tempo strategy, there was less freedom to deviate
from the initial relative phase angle between footfall and beat (for S2 this relative phase angle is
random, for S3 it is around -70˚ or 0˚ for initial and follow-up experiment).
Of particular interest was condition S1, where, due to the fixed tempo of the song, the high-
est degree of freedom was implemented with regard to the selection of the relative phase angle.
Each song started at a random relative phase angle (random time with respect to footfall), but
the distribution ended up with a clear clustering around negative relative phase angles (-14º
and -18 º for initial and follow-up experiment respectively). In addition, there was a large
spread in resultant vector lengths: in some cases people’s footfalls showed a constant timing
relative to the timing of the beats for a song (the dots on or near the circle perimeter, represent-
ing a high resultant vector length), in other cases this relation was less stable, or even almost
absent (the dots close to the center of the circle, representing a low resultant vector length).
This indicates that there was a convergence during the song from the starting angle towards
the final angle. The average resulting relative phase angle, as indicated by the arrows in Fig 1
Fig 1. Circular scatter plots of resultant vectors. Each dot represents the resultant vector (both angle and length) per song per participant. A negative angle indicates that
during that specific song, the footfall (averaged over the song) occurred before the musical beat (runner is first), while a positive angle indicates that the footfall occurred
after the beat (music is first). The distance from the center indicates the resultant vector length. The closer the dot is to the circle perimeter, the higher the phase coherence,
i.e. the more steps were taken towards this average phase angle. The arrows indicate the resultant vector length and average phase angle of the complete distribution of
songs, thus providing a general overview of all participants’ behavior to all songs, in a specific condition. The color and form of each dot indicate whether the data point
was from the initial or the follow-up experiment, which is denoted in the legend of the figures.
https://doi.org/10.1371/journal.pone.0208702.g001
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
11 / 21
and the resultant vector length value (length of the arrow in Fig 1, see also Table 5), show a pre-
ferred relative phase angle around -15˚. This confirms the preference for a slight negative
mean asynchrony, and that runners prefer the footfalls occurring just prior to the beats.
In S2, the tempo of the music was continuously matched to the tempo of the runner. The
figure shows high amounts of synchronization (runners consistently keeping their initial rela-
tive phase) but with a slight clustering around the same negative relative phase angles. This
indicates that phase attraction towards the preferred relative phase angle found in S1 still hap-
pened with adaptive tempo conditions, but it is less pronounced and it occurs only when start-
ing close to this preferred angle.
S3 further elaborated on this finding. The blue dots on Fig 1.S3 indicate data from the fol-
low-up experiment, where each song started around 0˚, that is, the beat is perfectly in sync
with the footfall. These participants deviated towards -40˚, thus increasing their relative phase
angle. For the initial experiment, we discovered a reverse effect: participants started in -70˚
and finished in on average -22˚, thus reducing their relative phase angle. This indicates that
there is a ‘phase attractor’ between -20˚ and -40˚, and that when the starting relative phase was
close to these angles, participants tended to get phase-locked around their preferred relative
phase angle.
Note that this attractor effect was not clearly visible in S2 as there was no clear clustering
around these relative phase angles. In this condition, the starting relative phase angle was ran-
dom. To explain why runners starting in these random relative phase angles were not attracted
to their own preferred angle, we refer to a mathematical model which is used in movement
coordination: the HKB model [34]. This model shows an attraction basin around the preferred
relative phase angle, but if a relative phase angle deviates too far from this preferred relative
phase angle, the attraction is not present. This could explain why there are high synchroniza-
tion scores, or resultant vector lengths, at all different relative phase angles. Additionally, when
participants were eventually attracted towards this preferred angle, another issue arose: since
the songs lasted only 55 seconds, it is likely that runners did not have enough time to converge
towards their preferred relative phase angle. Fig 1.S2, visualizes this phenomenon with lower
resultant vector lengths between -90º and 0˚ (compared to S1), which can be attributed to this
time-consuming convergence. A transition from one relative phase angle to a completely dif-
ferent relative phase angle results in a wide distribution and thus a lower resultant vector
length. We assume that if the song would have been longer, the same clustering as in S1 and S3
might have been discovered in S2 for starting relative phase angles close to the preferred rela-
tive phase angles.
S1 and S3, both the initial and the follow-up experiments, clearly revealed a negative mean
asynchrony (NMA), i.e. the participants showed a tendency to anticipate the beats (from -40˚
to -14˚). This is in accordance with other studies concerning sensorimotor synchronization.
On average, the runners put their feet down 40 to 15 ms before they perceive the musical beat.
This asynchrony could be explained by the fact that the tactile signal takes longer than the
audio signal to reach the brain. To synchronize in the brain, the tactile signal needs to occur
before the audio signal, hence the asynchrony of about -15º in relative phase. This is in line
with the nerve-conduction hypothesis, where peripheral processing time is dependent on the
distance of feedback (tactile or auditory) to the brain [17] and the sensory accumulator model
assuming that synchrony is established at the level of central representations [35]. Another
group of explanations is based on the onset computation or P-center hypothesis [36]: if an
event (e.g., a tap or a footstep) is extended over time, the perceptual center (P-center) differs
from the onset of the event. For tapping, it is suggested that rather than the initial surface con-
tact, the moment of peak force is the meaningful target in timing control [37].
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
12 / 21
In terms of attraction towards the NMA or preferred phase angle, we found that our results
from S1—S3 are in line with Haken-Kelso-Bunz (HKB) model [34]. This model depicts inter-
limb coordination and phase transitions between different states. It shows a strong attraction
basin around the 0˚ phase angle and a secondary attractor near 180˚ (antiphase), inducing phase
transitions from the current angle towards the closest attractor. The model also shows that the
strength of this attraction force decreases with increasing movement frequency, and at higher
tempi (such as running at a higher cadence than 150 SPM) the secondary antiphase attractor dis-
appears and only relative phases close to 0˚ get attracted. This is remarkably similar to the find-
ings of S1 through S3 (although with a rotation towards the NMA or preferred relative phase
angle of the participants): (1) if the starting phase differs enough from the NMA, no attraction is
present when both systems operate at the same frequency (music is matched to cadence) and (2)
when the starting phase is close to the NMA, participants converge towards their NMA.
S4 and S5 were designed to keep a constant relation between the moments of the footfall
and the beat. This is clearly illustrated in Fig 1 by the concentrations of dots on the circle
perimeter near the target phase angle that was implemented in these strategies: -70˚ in the ini-
tial experiment (in red), 0˚ (in blue) and +30˚ (in green) in the follow-up experiment. This
indicates that the software works as configured, and that we can force users towards a certain
relative phase angle. Later sections of the discussion will look into possible kinematic and
motivational effects of certain relative phase angles.
To summarize, we found that the three conditions allowing self-selected relative phase angles
show that the preferred angle does not approximate 0˚, but is slightly negative–following the
NMA displayed in sensorimotor synchronization experiments [11, 38]. We also found indica-
tions for an attraction force towards this preferred relative phase angle, especially when the initial
relative phase was close to the preferred relative phase. When the initial starting relative phase
was not near to this attractor, a stable relative phase was kept at this angle. Hence, to reach the
preferred self-selected relative phase angle, the starting relative phase is important. If we want to
design future music-to-movement alignment strategies that support the runners’ natural and
individual interactive behavior with music, not only should the technique be capable of selecting
music that matches the preferred tempo of the runner, it should also be able to apply the preferred
relative phase angle (as measured for instance with a fixed tempo strategy like S1 and fine-tuned
with a strategy similar to S3) as a target relative phase angle in strategies similar to S4 and S5.
Effect on cadence
The combined results of the initial and the follow-up experiment indicate that strategy S1, S2,
and S3 have no significant effect on cadence when compared to the allochronic music strategy
(S0). Runners can be phase-locked, i.e. maintain a stable footfall-to-beat relation (seen best in
S1, S2 and S3), and in general running with these strategies reflects a natural preference for a
slightly negative phase angle relation (NMA). In contrast, our main results do show an effect
of the alignment strategy on cadence with S4 and S5. This is further elaborated on with the fol-
low-up experiment: the effect appears to be dependent on the target phase angle settings in the
strategy. Fig 2 shows the resulting cadence change based on the forced relative phase angle.
For this discussion, we group several conditions based on the forced relative phase angle:
(0) control or asynchronous music, (1) self-selected relative phase angles, (2) forced at -70˚, (3)
forced at 0˚, and (4) forced at +30˚. Fig 2 visualizes the effect of forced relative phase angles on
cadence.
There was no difference between self-selected angles and the control condition in terms of
the cadence change, meaning participants did not show any difference in running behavior in
their self-selected angle compared to allochronic music. There was however an effect of 0˚:
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
13 / 21
cadence seemed to increase to a minor extent, while this increase was more pronounced at
+30˚. At -70˚, we noticed a clear decrease in cadence. Thus, to have a neutral null-effect, the
optimal relative phase angle would be the preferred relative phase angle as found in S1 and S3,
similar to our earlier findings on synchronization.
The decrease in cadence observed in S4 and S5 at -70º could be explained by the fact that
the beats were continuously perceived approximately 70 ms after the moment of the footfall.
The participant was therefore constantly urged to lengthen the duration of his or her next step
to reach the preferred -20˚ attractor. We theorize that, in order to deal with the perceived sen-
sorimotor error, the brain continuously adapts in order to try to minimize the perceived senso-
rimotor error (i.e., obtain NMA). Although that error persists (due to our matching
algorithm), the entrainment (or brain-driven sensorimotor error minimization) persists as
well, and the outcome is a decrease in cadence. Conversely, the introduction of the beat prior
to the footfall instant (positive relative phase angle) might rather induce a feeling of ’being
late’, which could in turn stimulate the runner to speed up. This was indeed observed in the
follow-up experiment, where S5 at +30˚ resulted in an increase in cadence compared to S0.
To further explore these results, a correlation analysis was performed to see a potential rela-
tion between the relative phase angle and the cadence change. This analysis only considers the
data of conditions where phase manipulation was used (S4 [-70˚], S4 [0˚], S5 [-70˚], S5 [0˚]
and S5 [+30˚]).
A Kolmogorov-Smirnov test indicated that the data were not normally distributed (p <
.01), hence a Spearman’s Rho correlation test was performed. We assumed that participants
would have tried to reach their preferred relative phase angle around -20˚, hence a one-tailed
test was performed as this assumption implies that a negative relative phase angle would result
Fig 2. Cadence change compared to initial silence over time. Conditions are grouped based on the forced relative phase angle to improve visual representation. Self-
selected relative phase angles include S1 through S3, Forced at -70˚ are S4 [-70˚] and S5 [-70˚], forced at 0˚ include S4 [0˚] and S5 [0˚], and finally forced at +30˚ is S5
[+30˚]. Self-selected relative phase angles induce little to no cadence change compared to the control condition, while the -70˚, 0˚ and +30˚ conditions have an effect over
time. Error bars represent 1 SE. The error bars are noticeably higher in the 0˚ and +30˚ conditions, as the amount of participants (N = 11) for the follow-up experiment
was lower than the amount of participants in the initial experiment (N = 39).
https://doi.org/10.1371/journal.pone.0208702.g002
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
14 / 21
in a lower cadence and vice versa. We uncovered a positive correlation between the relative
phase angle and cadence change (r = .293, p < .001, n = 681), showing that a connection
between cadence change and the forced relative phase angle indeed exists.
Furthermore, the influence of the relative phase angle appeared to build up over time. Fig 2
shows the evolution of cadence change for different strategies over the five songs. When look-
ing at the average cadence change over time, it was evident that the influence was not immedi-
ate, but increased over time. The lowering-cadence effect seemed to stop around 2.2%
slowdown (similar to the tempo basin found by Van Dyck et al. [15]), while the cadence
increasing effect was less pronounced. This shows that the relative phase angle manipulation
effect was ’limited’ towards the influence on cadence.
Motivation
In the initial experiment, as well as in the follow-up experiment, similar results were obtained
when comparing the enjoyment ratings of the different strategies to S0. The strategies that did
not significantly affect running cadence (S1, S2, S3, S4 [0˚], and S5 [0˚]) were not rated differ-
ently from S0. However, in the two conditions that affected cadence most (S5 [-70˚] and S5
[+30˚]) a small but significantly higher enjoyment rate, or a tendency towards such an effect,
was observed. According to literature on gaming [39] and exergames [40] the balance between
skill and challenge is crucial for being intrinsically self-rewarding. When challenged at a level
that the gamer perceives as pleasant, he or she is more likely to have a positive experience. As
such, it could be suggested that a small deviation from a preferred cadence, such as realized by
continuously manipulating the relative phase angle to a non-preferred phase angle (S5–70˚
and +30˚), might imply a more challenging or rewarding running exercise. A challenge that is
not too demanding for the runner’s skills (note that no significant increases in RPE were
observed) could result in a more pleasant and rewarding experience than when the runner is
not challenged or manipulated to change his or her preferred running behavior. According to
Fritz et al. [41] moving in synchrony with music can evoke a sense of agency, or a feeling of
’being in control’. Combined with a certain amount of physical exertion, this sense of agency
might contribute to a perceived positivity bias [42]. As such, our results seem to suggest that
finding a new phase angle balance induces such feelings of agency. Similar to the skill-chal-
lenge balance [39, 40], being in and out of balance with a slightly demanding phase angle
seems to more positively impact enjoyment levels compared to maintaining a preferred, less
demanding, phase angle. Alternatively, error minimization towards NMA can be seen as a reg-
ulatory mechanism to homeostasis (an equilibrium state consisting of predictive, expressive
and effort processes) [12].
Fig 3 summarizes our findings concerning the influence of forced relative phase angles. We
note 3 regions: the neutral ’self-selected’ or negative mean asynchrony near -20˚ (footfalls
slightly before beats), the cadence increasing larger angles between -10˚ and 30˚ (steps simulta-
neous or after the beats) and the lower angles between -40˚ and -90˚ (steps noticeably before
the beats). The effects of running in these regions are shown in the legend.
Gender differences
Several studies gave direct and indirect proof or reason to expect differences in how men run,
entrain, and synchronize with beats of the music, compared to women [15, 18, 19]. And
although no main effect of gender was found on the change in cadence, some interesting inter-
action effects were uncovered that are worth discussing. Below, we compare the running
behavior of men and women from the perspective of preferred phase angle and synchroniza-
tion, cadence, and motivation.
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
15 / 21
Gender differences: Preferred phase angle and synchronization.
Analysis showed that
there is no significant difference between genders for average phase angles in all different con-
ditions. However, in the fixed music condition (S1), where participants had the most freedom
to self-select relative phase angles (and converge to their NMA), we noticed a slight difference
in distributions for males versus females. The relative phase angle distribution was slightly
more asymmetrical and wider than that for females, indicating a larger spread in relative phase
angles and a less consistent synchronization. When looking at the phase coherence or synchro-
nization of these angles (expressed by the resultant vector length), more pronounced differ-
ences emerge: women tend to reach higher synchronization scores than men in the fixed
music (S1) condition. Fig 4 shows the distribution of the resultant vector lengths for S1.
Analysis with independent t-tests confirmed that there was indeed a significant difference
in the resultant vector length for S1 (fixed tempo) for males (M = 0.57, SD = 0.27) compared
to females (M = 0.68, SD = 0.27), t(233) = -2.964, p = .003. Females thus obtained higher syn-
chronization scores. No other conditions showed such differences; perhaps because these con-
ditions (S2-S5) allowed less freedom to self-select phase angles.
Using a resultant vector length of .75 as a threshold to determine running in phase coher-
ence or not (similar to [4]), we see 52% of songs in phase coherence for women versus 23% for
men in the S1 (fixed tempo) condition.
Gender differences: Cadence.
A study on running to music [15] reported differences
between male and female participants: women showed significantly higher levels of tempo
entrainment than men. This is in line with research by Karageorghis [18], who found that
women pay more attention to the rhythmic characteristics of music than men do. These find-
ings make it interesting to look at the before mentioned forced-phase influence on cadence
with the gender as a grouping factor.
Fig 3. Summary of the effects of running in a forced relative phase angle.
https://doi.org/10.1371/journal.pone.0208702.g003
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
16 / 21
The main results showed an interaction effect between condition (alignment strategy) and
gender (F(5, 170) = 4.97, p < .001) for the cadence change, indicating that the change in cadence
differed between men and women for the different strategies. More in depth analysis showed a
more pronounced change in cadence between the strategies for female runners (see Fig 5). It
seems that, compared to their male counterparts, women demonstrated lower levels in cadence
change with S4 [-70˚] and S5 [-70˚], and higher ones when running with S1, S2, and S3.
A correlation test was performed on the forced relative phase angle conditions to test the
significance of the forced phase angle on cadence change. Fig 6 illustrates the dataset. Both
male and female distributions were non-normal (p < .01) as determined by the Kolmogorov-
Smirnov test. The Spearman’s Rho correlation test showed a difference in correlation coeffi-
cients for the different genders. While the male population showed a slightly positive and sig-
nificant correlation between relative phase angle and cadence change (r = .124, p < .01,
N = 363), the female population demonstrated a much higher correlation coefficient at a
higher level of significance (r = .449, p < .001). These results indicate that females might
indeed be more susceptible towards the influence of relative phase angles on cadence.
Gender differences: Motivation.
With respect to gender, our interest was to find out if
there are differences in entrainment capacity and effects on cadence. For completeness sake,
we also report that no significant differences were observed for the enjoyment ratings
(PACES) between men and women: not in general, and not when the data were inspected per
strategy.
Fig 4. Resultant vector length or phase coherence for all songs in the S1 (fixed music) condition. A high level of phase coherence is observed for
most of the women, whereas the flatter distribution of the resultant vector length for men indicates that a higher percentage of men did not maintain a
phase-locked running behavior over the duration of a song.
https://doi.org/10.1371/journal.pone.0208702.g004
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
17 / 21
Conclusions
Music-to-movement alignment strategies make it possible for everyone to keep running in
synchrony with a musical stimulus, without consciously manipulating your own behavior to
Fig 5. Cadence change compared to initial silence, split on gender. Conditions are grouped based on the forced relative phase angle to improve visual representation.
Self-selected angles include S1 through S3, Forced at -70° are S4 [-70˚] and S5 [-70˚], Forced at 0° include S4 [0˚] and S5 [0˚], and finally Forced at +30° is S5 [+30˚]. Both
the control and self-selected phase angles induced little to no cadence change, while the -70˚ and +30˚ conditions had an effect that was more pronounced for females.
Running at 0˚ phase angle seemed to have neutral effects for the males but a cadence increasing effect for females.
https://doi.org/10.1371/journal.pone.0208702.g005
Fig 6. Correlations between forced relative phase angle and cadence change. The effect is more visible for the female than the male population.
https://doi.org/10.1371/journal.pone.0208702.g006
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
18 / 21
align with the music. Our study shows that strategies implementing a continuous relative
phase alignment achieve the highest level of synchronization (as intended) and are therefore
perfectly capable of spontaneously manipulating running behavior. A change in cadence is in
fact induced by deviating the relative phase angle continuously from a preferred phase angle
that is typically slightly negative: a more negative phase angle (step before beat) causes a slow-
down in cadence and a positive phase angle (step after beat) leads to an increase in cadence.
Not only was cadence affected by phase angles that diverged from the runners’ preferred
phase angles, also enjoyment levels were affected. Although differences in enjoyment were not
very obvious, small, slightly significant increases in enjoyment were reported for the strategies
that impacted cadence. Spontaneously guiding runners a little away from their preferred run-
ning behavior could add just enough challenge to their exercise, resulting in a more rewarding
experience [39, 40].
The general preference for the footfalls to occur prior to the beats suggests that future
music alignment strategies could be improved by considering the negative mean asynchrony
(NMA) rather than 0˚ as a reference point. A question for future research might be whether
the size of the NMA is task- and/or user dependent. For example, is human anticipation of a
beat different when running compared to tapping? Are there differences between people and
why? With respect to the latter, our results indicate that gender plays a role when using forced
phase algorithms: effects of phase coherence and change in cadence were more pronounced
for females compared to their male counterparts.
Practical applications
Music-to-movement alignment strategies enable us to continuously and closely follow a per-
son’s behavioral response to music. This is of great value for sports and rehabilitation pro-
grams where music-based biofeedback is employed to improve individual performance [43].
In the future, the aim is to further improve our alignment strategies and introduce musical
beats either slightly before or after the predicted footfalls. Such strategies could open up possi-
bilities to spontaneously (and imperceptibly) optimize cadence and step size [44].
Acknowledgments
The authors also wish to thank the Flanders Sports Arena of Ghent, for giving us access to
their indoor running track.
Author Contributions
Conceptualization: Bart Moens, Dobromir Dotov, Marc Leman.
Formal analysis: Jeska Buhmann.
Funding acquisition: Marc Leman.
Investigation: Jeska Buhmann, Edith Van Dyck.
Methodology: Jeska Buhmann, Bart Moens, Edith Van Dyck.
Project administration: Edith Van Dyck.
Software: Bart Moens, Dobromir Dotov.
Supervision: Edith Van Dyck.
Writing – original draft: Jeska Buhmann.
Writing – review & editing: Jeska Buhmann, Bart Moens, Edith Van Dyck, Marc Leman.
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
19 / 21
References
1.
Karageorghis CI, Priest D-L. Music in the exercise domain: a review and synthesis (Part I). Int Rev
Sport Exerc Psychol. 2012; 5(1):44–66. https://doi.org/10.1080/1750984X.2011.631026 PMID:
22577472
2.
Priest D-L, Karageorghis CI, Sharp NCC. The characteristics and effects of motivational music in exer-
cise settings: The possible influence of gender, age, frequency of attendance, and time of attendance. J
Sports Med Phys Fitness. 2004; 44:77–86. PMID: 15181394
3.
Leman M, Moelants D, Varewyck M, Styns F, van Noorden L, Martens J-P. Activating and relaxing
music entrains the speed of beat synchronized walking. PLoS ONE. 2013; 8(7):e67932. https://doi.org/
10.1371/journal.pone.0067932 PMID: 23874469
4.
Buhmann J, Desmet F, Moens B, Van Dyck E, Leman M. Spontaneous velocity effect of musical
expression on self-paced walking. PLoS ONE. 2016; 11(5):e0154414. https://doi.org/10.1371/journal.
pone.0154414 PMID: 27167064
5.
Terry PC, Karageorghis CI, Saha AM, D’Auria S. Effects of synchronous music on treadmill running
among elite triathletes. J Sci Med Sport. 2012; 15(1):52–7. https://doi.org/10.1016/j.jsams.2011.06.003
PMID: 21803652
6.
Bood RJ, Nijssen M, van der Kamp J, Roerdink M. The power of auditory-motor synchronization in
sports: enhancing running performance by coupling cadence with the right beats. PLoS ONE. 2013; 8
(8):e70758. https://doi.org/10.1371/journal.pone.0070758 PMID: 23951000
7.
Lim HBT, Karageorghis CI, Romer LM, Bishop DT. Psychophysiological effects of synchronous versus
asynchronous music during cycling. Med Sci Sports Exerc. 2014; 46(2):407–13. https://doi.org/10.
1249/MSS.0b013e3182a6378c PMID: 24441216
8.
Ramji R, Aasa U, Paulin J, Madison G. Musical information increases physical performance for synchro-
nous but not asynchronous running. Psychol Music. 2016; 44(5):984–95. https://doi.org/10.1177/
0305735615603239
9.
Mendonc¸a C, Oliveira M, Fontes L, Santos J. The effect of instruction to synchronize over step fre-
quency while walking with auditory cues on a treadmill. Human movement science. 2014; 33:33–42.
https://doi.org/10.1016/j.humov.2013.11.006 PMID: 24576706
10.
Janata P, Tomic ST, Haberman JM. Sensorimotor coupling in music and the psychology of the groove.
J Exp Psychol Gen. 2012; 141(1):54–75. https://doi.org/10.1037/a0024208 PMID: 21767048
11.
Repp BH, Su Y-H. Sensorimotor synchronization: a review of recent research (2006–2012). Psycho-
nomic bulletin & review. 2013; 20(3):403–52.
12.
Leman M. The expressive moment: How interaction (with music) shapes human empowerment. Cam-
bridge, MA: MIT Press; 2016.
13.
Franěk M, van Noorden L, Rezˇny´ L. Tempo and walking speed with music in the urban context. Front
Psychol. 2014; 5:1361. https://doi.org/10.3389/fpsyg.2014.01361 PMID: 25520682
14.
Alter DA, O’Sullivan M, Oh PI, Redelmeier DA, Marzolini S, Liu R, et al. Synchronized personalized
music audio-playlists to improve adherence to physical activity among patients participating in a struc-
tured exercise program: a proof-of-principle feasibility study. Sports Med Open. 2015; 1:23. https://doi.
org/10.1186/s40798-015-0017-9 PMID: 26284164
15.
Van Dyck E, Moens B, Buhmann J, Demey M, Coorevits E, Dalla Bella S, et al. Spontaneous entrain-
ment of running cadence to music tempo. Sports Med Open. 2015; 1:15. https://doi.org/10.1186/
s40798-015-0025-9 PMID: 26258007
16.
Moens B, Muller C, van Noorden L, Franěk M, Celie B, Boone J, et al. Encouraging spontaneous syn-
chronisation with D-Jogger, an adaptive music player that aligns movement and music. PLoS ONE.
2014; 9(12):e114234. https://doi.org/10.1371/journal.pone.0114234 PMID: 25489742
17.
Paillard J. Quelques donnes psychophysiologiques au dclenchement de la commande motrice [Some
psychophysiological data relating to the triggering of motor commands]. LAnne Psychologique 48:
2847.[aBH] Paillard J., ed.(1991) Brain and space. Oxford University Press.[aBH]; 1949.
18.
Karageorghis CI, Terry PC, Lane AM. Development and initial validation of an instrument to assess the
motivational qualities of music in exercise and sport: The Brunel Music Rating Inventory. J Sports Sci.
1999; 17(9):713–24. https://doi.org/10.1080/026404199365579 PMID: 10521002
19.
Cole Z, Maeda H. Effects of listening to preferential music on sex differences in endurance running per-
formance. Percept Mot Skills. 2015; 121(2):390–8. https://doi.org/10.2466/06.PMS.121c20x9 PMID:
26447745
20.
Faul F, Erdfelder E, Lang A-G, Buchner A. G* Power 3: A flexible statistical power analysis program for
the social, behavioral, and biomedical sciences. Behav Res Methods. 2007; 39(2):175–91. https://doi.
org/10.3758/BF03193146 PMID: 17695343
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
20 / 21
21.
Karageorghis CI, Priest D-L, Terry PC, Chatzisarantis NLD, Lane AM. Redesign and initial validation of
an instrument to assess the motivational qualities of music in exercise: The Brunel Music Rating Inven-
tory-2. J Sports Sci. 2006; 24(8):899–909. https://doi.org/10.1080/02640410500298107 PMID:
16815785
22.
Dixon S. Evaluation of the audio beat tracking system beatroot. J New Music Res. 2007; 36(1):39–50.
https://doi.org/10.1080/09298210701653310
23.
International Telecommunication Union. ITU-R BS.1770-3. Algorithms to measure audio programme
loudness and true-peak audio level. 2012.
24.
Pappas IP, Popovic MR, Keller T, Dietz V, Morari M. A reliable gait phase detection system. IEEE
Transactions on neural systems and rehabilitation engineering. 2001; 9(2):113–25. https://doi.org/10.
1109/7333.928571 PMID: 11474964
25.
Zplane. E´ lastique Time-Stretching and Pitch-Shifting SDKs 2002 [cited 2013 9 Mar].
26.
Borg G. Borg’s perceived exertion and pain scales. Champaign, IL: Human Kinetics; 1998.
27.
Kendzierski D, DeCarlo KJ. Physical activity enjoyment scale: two validation studies. J Sport Exerc Psy-
chol. 1991; 13(1):50–64. https://doi.org/10.1123/jsep.13.1.50
28.
Mullen SP, Olson EA, Phillips SM, Szabo AN, Wo´jcicki TR, Mailey EL, et al. Measuring enjoyment of
physical activity in older adults: invariance of the physical activity enjoyment scale (paces) across
groups and time. Int J Behav Nutr Phys Act. 2011; 8:103. https://doi.org/10.1186/1479-5868-8-103
PMID: 21951520
29.
Hove MJ, Suzuki K, Uchitomi H, Orimo S, Miyake Y. Interactive rhythmic auditory stimulation reinstates
natural 1/f timing in gait of Parkinson’s patients. PLoS ONE. 2012; 7(3):e32600. https://doi.org/10.1371/
journal.pone.0032600 PMID: 22396783
30.
Berens P. CircStat: a MATLAB toolbox for circular statistics. J Stat Softw. 2009; 31(10):1–21.
31.
Mormann F, Lehnertz K, David P, Elger CE. Mean phase coherence as a measure for phase synchroni-
zation and its application to the EEG of epilepsy patients. Physica D. 2000; 144(3):358–69. https://doi.
org/10.1016/S0167-2789(00)00087-7
32.
Fisher NI. Statistical analysis of circular data: Cambridge University Press; 1995.
33.
Mardia KV. A multi-sample uniform scores test on a circle and its parametric competitor. Journal of the
Royal Statistical Society Series B (Methodological). 1972:102–13.
34.
Haken H, Kelso JS, Bunz H. A theoretical model of phase transitions in human hand movements. Bio-
logical cybernetics. 1985; 51(5):347–56. PMID: 3978150
35.
Aschersleben G, Prinz W. Delayed auditory feedback in synchronization. Journal of motor behavior.
1997; 29(1):35–46. https://doi.org/10.1080/00222899709603468 PMID: 20037008
36.
Morton J, Marcus S, Frankish C. Perceptual centers (P-centers). Psychological Review. 1976; 83
(5):405.
37.
Du Y, Clark JE, Whitall J. Timing at peak force may be the hidden target controlled in continuation and
synchronization tapping. Experimental brain research. 2017; 235(5):1541–54. https://doi.org/10.1007/
s00221-017-4918-3 PMID: 28251338
38.
Repp BH. Sensorimotor synchronization: A review of the tapping literature. Psychonomic bulletin &
review. 2005; 12(6):969–92.
39.
Corcos A. Being enjoyably challenged is the key to an enjoyable gaming experience: an experimental
approach in a first-person shooter game. Socioaffective neuroscience & psychology. 2018; 8
(1):1474668.
40.
Lyons EJ. Cultivating engagement and enjoyment in exergames using feedback, challenge, and
rewards. Games for health journal. 2015; 4(1):12–8. https://doi.org/10.1089/g4h.2014.0072 PMID:
26181675
41.
Fritz TH, Hardikar S, Demoucron M, Niessen M, Demey M, Giot O, et al. Musical agency reduces per-
ceived exertion during strenuous physical performance. Proc Natl Acad Sci U S A. 2013; 110
(44):17784–9. https://doi.org/10.1073/pnas.1217252110 PMID: 24127588
42.
Fritz TH, Schneider L, Villringer A. The Band Effect-Physically Strenuous Music Making Increases
Esthetic Appreciation of Music. Front Neurosci. 2016; 10:448. https://doi.org/10.3389/fnins.2016.00448
PMID: 27799893
43.
Maes P-J, Buhmann J, Leman M. 3Mo: A Model for Music-Based Biofeedback. Front Neurosci. 2016;
10:548. https://doi.org/10.3389/fnins.2016.00548 PMID: 27994535
44.
Reenalda J, Maas MTF, de Koning JJ. The Influence of Added Mass on Optimal Step Length in Run-
ning. Int J Sports Physiol Perform. 2016; 11(7):920–6. https://doi.org/10.1123/ijspp.2015-0182 PMID:
26791558
Optimizing beat synchronized running to music
PLOS ONE | https://doi.org/10.1371/journal.pone.0208702
December 6, 2018
21 / 21
| Optimizing beat synchronized running to music. | 12-06-2018 | Buhmann, Jeska,Moens, Bart,Van Dyck, Edith,Dotov, Dobromir,Leman, Marc | eng |
PMC3912221 | A Healthy Brain in a Healthy Body: Brain Network
Correlates of Physical and Mental Fitness
Linda Douw1,2*, Dagmar Nieboer3, Bob W. van Dijk4, Cornelis J. Stam4, Jos W. R. Twisk3
Abstract
A healthy lifestyle is an important focus in today’s society. The physical benefits of regular exercise are abundantly clear, but
physical fitness is also associated with better cognitive performance. How these two factors together relate to
characteristics of the brain is still incompletely understood. By applying mathematical concepts from ‘network theory’,
insights in the organization and dynamics of brain functioning can be obtained. We test the hypothesis that neural network
organization mediates the association between cardio respiratory fitness (i.e. VO2 max) and cognitive functioning. A healthy
cohort was studied (n = 219, 113 women, age range 41–44 years). Subjects underwent resting-state eyes-closed magneto-
encephalography (MEG). Five artifact-free epochs were analyzed and averaged in six frequency bands (delta-gamma). The
phase lag index (PLI) was used as a measure of functional connectivity between all sensors. Modularity analysis was
performed, and both within and between-module connectivity of each sensor was calculated. Subjects underwent a
maximum oxygen uptake (VO2 max) measurement as an indicator of cardio respiratory fitness. All subjects were tested with
a commonly used Dutch intelligence test. Intelligence quotient (IQ) was related to VO2 max. In addition, VO2 max was
negatively associated with upper alpha and beta band modularity. Particularly increased intermodular connectivity in the
beta band was associated with higher VO2 max and IQ, further indicating a benefit of more global network integration as
opposed to local connections. Within-module connectivity showed a spatially varied pattern of correlation, while average
connectivity did not show significant results. Mediation analysis was not significant. The occurrence of less modularity in the
resting-state is associated with better cardio respiratory fitness, while having increased intermodular connectivity, as
opposed to within-module connections, is related to better physical and mental fitness.
Citation: Douw L, Nieboer D, van Dijk BW, Stam CJ, Twisk JWR (2014) A Healthy Brain in a Healthy Body: Brain Network Correlates of Physical and Mental
Fitness. PLoS ONE 9(2): e88202. doi:10.1371/journal.pone.0088202
Editor: Renaud Lambiotte, University of Namur, Belgium
Received October 5, 2013; Accepted January 9, 2014; Published February 3, 2014
Copyright: 2014 Douw et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was funded by grants from the Dairy Foundation, the Netherlands Heart Foundation, the Dutch Prevention Fund, Heineken BV, the Ministry
of Public Health, Wellbeing and Sport (VWS), the Scientific Board of Smoking and Health, the VU University and the VU University Medical Center. The funders had
no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have the following interests: This study was funded in part by Heineken BV. Co-author Prof. CJ Stam is a PLOS ONE Editorial
Board member. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials.
* E-mail: [email protected]
Introduction
A healthy lifestyle is a major focus in today’s society. Regular
exercise and adequate physical fitness have proven to be important
for the immune system, metabolism, prevention of infectious
disease, skeletal functioning, and risk of cancer [1–6]. In addition
to these physical benefits, cardiorespiratory fitness is also related to
better cognitive functioning [7]. Several neural factors have been
reported to mediate the relationship between mental and physical
fitness, including increased neural vascularization [8], increased
production of brain derived neurotrophic factor (BDNF; [9]),
increased hippocampal volume [10], and higher levels of N-
acetylaspartate [11], although none of these mediators fully
explain the reported associations.
Another framework that has elucidated the neural correlates of
the association between cognition and physical fitness is resting-
state functional connectivity, as measured with functional mag-
netic resonance imaging (fMRI). The resting-state, during which
no task is present and alert relaxation is achieved, can be
characterized by several highly robust networks [12,13], of which
the default mode network (DMN) is the most stable and best
studied example [14,15]. This network seems to be the functional
backbone of the brain [16], and is implicated in almost all
neurological diseases. With respect to cardiorespiratory fitness,
higher connectivity within the DMN (as measured by seeding the
posterior cingulate cortex and examining its significantly correlat-
ed regions) is associated with better fitness level, and DMN
connectivity mediates the association between physical fitness and
cognitive functioning [17]. Moreover, a one-year aerobic training
intervention in older adults improves functional connectivity
within several resting-state networks, including the DMN and
the fronto-parietal network, which is thought to be important for
working memory [18]. Conversely, overweight adults show
increased DMN connectivity, which normalizes after a six month
exercise program [19]. The important role of functional connec-
tivity in the relationship between physical fitness and cognition was
confirmed in another study by Voss and colleagues, showing that
the association between exercise and connectivity is related to
BDNF, insulin-like growth factor type 1 (IGF-1), and vascular
PLOS ONE | www.plosone.org
1
February 2014 | Volume 9 | Issue 2 | e88202
1 Department of Neurology, Vrije Universiteit Medical Center, Amsterdam, The Netherlands, 2 Department of Anatomy and Neurosciences, Vrije Universiteit Medical
Center, Amsterdam, The Netherlands, 3 Department of Epidemiology and Biostatistics, Vrije Universiteit Medical Center, Amsterdam, The Netherlands, 4 Department of
Clinical Neurophysiology and Magnetoencephalography Center, Vrije Universiteit Medical Center, Amsterdam, The Netherlands, 5 Department of Epidemiology and
Biostatistics, EMGO Institute for Health and Care Research, VU University Medical Center, Amsterdam, the Netherlands
endothelial
growth
factor
(VEGF),
which
are
markers
for
neuroplasticity [20].
However, fMRI is an indirect measure of neural functioning, as
it measures the slowly operating process of blood oxygenation.
Functional connectivity can also be determined frommagnetoen-
cephalography (MEG), which is a much more direct measure of
neural activity. Furthermore, functional connectivity in general
can be used as a starting point for more extensive, higher-order
analysis of the entire brain network using graph theory [21–23].
This type of analysis has shown that the brain network is very
comparable to many simpler biological and sociological systems
[24]. This elegantly theory-governed but still data-driven property
has made the application of network theory to the brain a very
interesting endeavor. For instance, the brain network is a ‘small-
world’, combining local segregation with global integration [24–
27]. Brain network topology is to a large extent genetically
determined [28–30] and is disturbed in several neurological
diseases [21,31].
The functional brain network also correlates with global
cognitive functioning and intelligence [32,33], indicating network
theory may add relevant information on neural correlates of
functioning above connectivity alone. Important information
about the integrity of the (brain) network can be extracted by
looking at modularity. Modules are clusters of nodes, or brain
areas, that are highly connected to each other, but much less to
nodes outside of their own module [34]. In the brain, five to seven
modules can be discerned, which correspond to major functional
systems [35]. Moreover, the role that specific brain areas play both
within their module and in connecting other modules has proven
relevant to brain functioning [36–38].
In this study, physical fitness, intelligence, and their neural
correlates in terms of network modularityare investigated. We
aimed to prove that VO2 max, a measure of cardio respiratory
fitness,is related to modular network topology based on MEG in a
large group of healthy subjects. Furthermore, we hypothesized that
intelligence is associated with physical fitness mediated through
brain network topology in terms of modular organization.
Materials and Methods
Ethics statement
This study was approved by the Medical Ethical Institutional
Review Board of the VU University Medical Center. All subjects
gave written informed consent before participation. This study was
carried out in accordance with the Declaration of Helsinki.
Subjects
All subjects participated in a prospective longitudinal study,
originally investigating natural development of growth, health and
lifestyle of adolescents, the Amsterdam Growth and Health
Longitudinal Study (AGHLS). This cohort study started in 1976
with four annual measurements and continued with an extensive
number of assessments with five to seven year time intervals[39].
All participants were born between 1961 and 1965 and were
residents of the Netherlands. First- and second-year pupils from
two equally large secondary schools were recruited. In 2006, MEG
recordings of the remaining 344 participants (who were all
between 41 and 44 years old) were obtained, in addition to the
health parameters that were investigated at each time-point of the
AGHLS [40]. These data are not publicly available at this point.
Physical fitness
Physical fitness was measured with a maximal running test on a
treadmill (Quinton 18–54; Quinton, Bothell, Wash), with a speed
of 8 km/h and increasing slope (every 2 minutes) and with direct
measurements of oxygen uptake (Ergoanalyzer; Jaeger, Bunnik,
the Netherlands). Maximum oxygen consumption (VO2 max) was
used as a measure of physical fitness (Kemper, 1995). This
measurement was performed approximately six years before MEG
recording.
Cognitive performance
Subjects underwent a cognitive test battery at the time of MEG
recording, to assess full-scale intelligence. The test battery
administered included the shortened Groninger Intelligence Test
(GIT [41]), which is a commonly used Dutch intelligence test.
Three subtests of the entire GIT were used, constituting the short
version of the test to assess intelligence [42]. Completion of the test
took approximately 45 minutes per subject.
Magnetoencephalography
Magnetic fields were recorded for five minutes while subjects
were seated inside a magnetically shielded room (Vacuumsch-
melze GmbH, Hanau, Germany), using a 151-channel whole-
head MEG system (CTF SystemsInc., Port Coquitlam, BC,
Canada). A third-order software gradient was used after online
band-pass filtering between 0.25 and 125 Hz. Sample frequency of
recording was 625 Hz. Fields were measured during a no-task,
eyes-closed condition of five minutes. At the beginning and ending
of each recording, the head position relative to the coordinate
system of the helmet was recorded by passing small alternating
currents through three head position coils attached to the left and
right pre-auricular points and the nasion on the subject’s head.
For each subject, the first five artifact-free epochs of 4096
samples (6.554 s) were selected by one of the authors [BWvD]. All
data analyses were performed using BrainWave [CJS, version
0.9.58, available from http://home.kpn.nl/stam7883/brainwave.
html]. Before calculating connectivity and network topology,
epochs were band-pass filtered into the commonly used frequency
bands delta (0.5–4 Hz), theta (4–8 Hz), lower alpha (8–10 Hz),
upper alpha (10–13 Hz), beta (13–30 Hz), and gamma (30–
45 Hz). All further analyses were performed for these bands
separately. The average relative power in the six abovementioned
frequency bands was calculated in each subject using a fast Fourier
transform as described in [43].
Phase Lag Index (PLI)
As a measure of functional connectivity, the phase lag index
(PLI) was used[44], which calculates the asymmetry of the
distribution of (instantaneous) phase differences between two
time-series. This asymmetry can be obtained from a time series of
phase differences DW (tk), k = 1… N samples:
PLI~ vsign sin D
tk
ð Þ
ð
Þ
½
w
j
j
The phase difference DQ is defined in the interval [2p, p] and
,. denotes the mean value. Volume conduction causes a zero
phase lag between two time-series, but the presence of a consistent,
non-zero,
phase
lag
between
two
time-series
reflects
true
interactions that are unaffected by volume conduction or common
sources.
Modularity, between and within module connectivity
First, to describe modularity in the whole-brain network we
used a version of previously described approaches[45], adapted for
weighted networks [37,46]:
Brain Networks and Physical Fitness
PLOS ONE | www.plosone.org
2
February 2014 | Volume 9 | Issue 2 | e88202
Qw
m ~
X
m
s~1
ls
L{
ds
2L
2
where m is the number of modules, ls is the sum of the weights of
all links in module s, L is the total sum of all weights in the
network, and ds is the sum of the strength of all nodes (i.e. the
summed weights per node) in module s. In short, the relation
between intra- and intermodular connections determines the
strength of each module. This measure describes modularity by
summing the relative strength of all the network modules, which
takes both within and between module connections into account.
A strongly modular network has modularity value close to 1, while
modularity is closer to 0 (but not absent) in a random network.
Finding the optimal modular organization in a network is a
computationally intensive problem. Simulated annealing can
overcome part of this issue, and was used in the current study
[45]. Initially, each of the N nodes was randomly assigned to one
of m possible clusters, where the initial m was taken as the square
root of N. At each step, one of the nodes was chosen at random,
and assigned a different randomly chosen module number from
the interval [1,N]. Modularity was calculated before and after this.
The cost C was defined as {Qw
m. The new partitioning was
preserved with probability p: if the final cost Cf was lower or equal
to the initial cost Ci (indicating no added cost when preserving the
partition), p was equal to 1. If Cf was higher than Ci, p was
calculated as follows:
p~ e{
cf { ci
T
The temperature T was 1 initially, and was lowered once every
100 steps as follows: Tnew = 0.995 Told. In total, the simulated
annealing algorithm was run for 106 steps. The partition with the
strongest modular organization (highest Q) was identified sepa-
rately for each epoch of every person for all the different frequency
bands, and subjected to further graph analysis.
Once the modular organization in a network has been
determined, the topological role of individual nodes can be
described in greater detail: nodes can be mainly involved in
communication with other nodes in the same module, but can also
preferably interact with other modules (see figure1). This aspect is
quantified by two properties: the within-module degree (Zi), and
the participation coefficient (PC) [45]. The within-module degree
measures the connectivity of the node within the module
compared to the other nodes in the same module, and thus
describes the relative importance within the module. It was
defined as follows:
zw
i ~ kw
i mi
ð
Þ{ k
w mi
ð
Þ
skw mi
ð
Þ
Here, mi is the module containing node i, kw
i mi
ð
Þ is the within
module strength of node i (the sum of all weights of the links
between i and all other nodes in mi), and k
w ( mi ) and skw mi
ð
Þ are
the respective mean and standard deviation of the within-module
mi degree distribution.
The participation coefficient expresses how strongly a node is
connected to other modules, and the weighted version is defined
as:
PCi ~1{
X
m[M
kw
i m
ð Þ
kw
i
M is the set of modules, and kw
i is the sum of all weights of the
links between i and all nodes in module m. The within module
degree and the participation coefficient determine the identity of a
node in the modular network structure.
Statistical analysis
Statistical analyses were performed using PASW Statistics
package (version 20.0) and Matlab version r2012b. Differences
between men and women regarding VO2max, IQ, and average
head surface were tested using Student’s t-tests. The association
between intelligence and VO2 max was analyzed using a linear
regression with IQ as the dependent variable and VO2 max as
independent variable, adjusted for sex. The association between
VO2 max and modular network topology was investigated with
regression analysis in which modularity indices per frequency band
were the dependent variables and VO2 max the dependent
variable, adjusted for sex, head surface size, and relative power per
frequency band. Similar analyses were used to explore average
between-module connectivity. Finally, the relationships between
VO2 max, network modularity, and intelligence were further
studied using mediation analyses. Mediation analysis investigates
whether a third parameter underlies an observed relationship
between two variables, meaning that the third variable governs the
Figure 1. Schematic representation of modularity and modular
connectivity. Note. In (a), two modules can be discerned. These
modules show high within-module connectivity, but low between-
module connectivity. (b) depicts two nodes in the network that are
characterized by high within-module connectivity, while (c) shows a
node with very high between-module connectivity.
doi:10.1371/journal.pone.0088202.g001
Brain Networks and Physical Fitness
PLOS ONE | www.plosone.org
3
February 2014 | Volume 9 | Issue 2 | e88202
association between the other two. In our study, we hypothesized
that the association between VO2 max and intelligence is
mediated by brain network modularity. This mediation model
was tested using the INDIRECT PASW statistics plug-in [47].
Direct and indirect effects between the dependent and indepen-
dent variables as well as the mediator were tested with regression
analyses (adjusted for significant abovementioned covariates), after
which 95% CIs were calculated for the total indirect effects using
bootstrapping (5,000 samples) as an unbiased means of testing
whether the mediation model was valid. The presence of a
mediation effect signifies that instead of having a direct causal
effect between the independent variable (VO2 max) and depen-
dent variable (IQ), the mediator (modular network topology) plays
an important role in the association between these two variables.
Results
Subject characteristics
At this time-point in the AGHLS study, 344 healthy subjects
participated. Our strict inspection of artifacts in the MEG
recordings caused exclusion of 79 subjects. Fourteen subjects were
excluded after examination of their intelligence scores, because
they performed well below average (,75). Of the remaining 251
subjects, VO2max measurements were performed in 219 subjects,
in whom all subsequent analyses were performed (see table1 for
subject characteristics). Subjects were on average 42 years old
(range 41–44). With respect to IQ, men and women did not differ
(t(217) = 1.005, p = 0.316). Men did have higher VO2 max
(t(217) = 14.006, p,0.001) and greater head surface in cm2
(t(217) = 8.979, p,0.001). To ascertain that network topology
results were not confounded by head surface size, this variable was
used as a covariate in all analyses. Four MEG sensors were
malfunctioning at the time of data collection, and these were
excluded from further analysis in all subjects.
The last VO2 max measurement took place six years before
MEG recording, when subjects were approximately 36 years old.
In order to investigate whether this gap could induce large changes
in physical fitness, we examined data from previous measurements
in the AGAHLS cohort. These measurements were performed at
13, 14, 15, 16, 21, 27, 29 and 32 years of age in subgroups of the
total cohort (with group sizes varying between 70 and 227
subjects). When looking at the consistency of VO2 max over these
time points, there is strong consistency within subjects over time
(see supplementary figure1), with an average correlation coefficient
R = 0.773 from one time point to the next. When comparing the
first adult measurement at 21 years old and the measurement used
in the remainder of this study at 36 years old (94 subjects
overlapping), the correlation coefficient is 0.791. Furthermore,
subjects who experienced major health burdens possibly influenc-
ing their lifestyle were excluded, which also ensures the stability
and consistency of the VO2 max measurements up to MEG and
IQ measurements six years later.
Physical fitness, intelligence and brain modularity
The previously reported association between physical fitness
and intelligence was confirmed: VO2 max was a significant
predictor of intelligence in a linear regression model (B = 0.322,
95% CI [0.049 0.594], p = 0.021). We then set out to investigate
our hypothesis concerning the association between physical fitness
and brain network topology. Lower modularity in the upper alpha
and beta bands was related to higher VO2 max (upper alpha band
B = 21.81, 95% CI [23.31 20.315], p = 0.018; beta band
B = 21.167, 95% CI [21.753 25.81], p = 0.017), adjusted for
sex, head surface, and relative power per frequency band (see
table2 for results of all frequency bands).
In order to confirm that these associations were indeed due to
network topology instead of global connectivity levels, we
performed an ANOVA with VO2 max as dependent variable
and both modularity and average connectivity in the upper alpha
and beta bands as independent variables. While the upper alpha
and beta connectivity indices did not yield significant results
(p = 0.529 and p = 0.869, resp.), modularity indices were signifi-
cantly related to VO2 max (p = 0.016 and p = 0.012, resp.). The
number of modules in the upper alpha and beta bands was not
associated with VO2 max, indicating that it was not the number of
modules that mattered, but the connectivity patterns within and
between those modules.
Modular connectivity
We then investigated the associations of between and within
module connectivity with VO2 max in these two frequency bands.
Results show that in the beta band, higher VO2max was related to
increased between-module connectivity (B = 0.674, 95% CI [0.101
1.246], p = 0.021), indicating indeed that physical fitness is
Table 1. Subject characteristics.
Total group (N = 219)
Men (N = 106)
Women (N = 113)
Mean age in years (SD)
42 (0.7)
42 (0.7)
42 (0.7)
IQ score (SD)
108 (13)
109 (13)
108 (13)
Head surface in cm2 (SD)
231 (19)**
242 (17)
222 (15)
VO2max (SD)
46 (8.6)**
52 (7.0)
40 (5.5)
Note ** = p,.01, significant gender difference.
doi:10.1371/journal.pone.0088202.t001
Table 2. Associations between band-specific modularity and
VO2 max.
B
95% CI (B)
p-value
Delta band modularity
0.435
[-.080 0.167]
0.487
Theta band modularity
20.390
[21.64 0.086]
0.540
Loweralphamodularity
0.509
[20.121 2.23]
0.561
Upperalphamodularity
21.81
[23.31 23.15]
0.018*
Betamodularity
21.17
[21.75 25.81]
0.017*
Gamma modularity
211.1
[280.9 58.6]
0.754
Note. * = p,0.05. Sex, relative power in each frequency band, and skull size
were entered as covariates in each regression.
doi:10.1371/journal.pone.0088202.t002
Brain Networks and Physical Fitness
PLOS ONE | www.plosone.org
4
February 2014 | Volume 9 | Issue 2 | e88202
important for better intermodular integration. Moreover, an
ANOVA with both between-module and average connectivity as
covariates shows significant results for between-module connec-
tivity only (p = 0.039 and p = 0.919, resp.), further underlining the
added
value
of modularity-based
connectivity
over
regular
connectivity alone.
Figure2a shows significant associations between VO2 max and
between-module connectivity per channel for both upper alpha
and beta bands, after correcting for the number of tests performed
with the false discovery rate (FDR, q,0.05 [48]). These maps
confirm the analysis of averaged between-module connectivity,
and show positive correlations between VO2 max and between-
module connectivity throughout the brain in the beta band, but
not the upper alpha band.
Due to the nature of the within-module calculation (i.e. within-
subject z-score is computed), no global average can be computed
for this measure. However, figure2b displays significant within-
module connectivity associations with VO2 max in the upper
alpha and beta bands, indicating that higher within-module
connectivity in the central areas is positively associated with VO2
max, while the within-module connectivity within lateral temporal
areas is negatively associated with physical fitness.
Modularity and between-module connectivity as VO2
max – IQ mediators
Finally, the associations between intelligence, VO2 max, and
brain modularity were analyzed using mediation analyses. Our
hypothesis was that better physical fitness leads to better cognitive
performance and thus higher IQ later, through the mediating
effect of brain network modularity (see figure3). This hypothesis
was not confirmed. Although separate regressions of the associ-
ations between both VO2 max and network characteristics and
intelligence were significant, the mediation effects, as evidenced by
significance levels and 95% confidence intervals through 5,000
bootstrapping samples, were not (see table3). This indicates that
although modularity, VO2 max and IQ are interrelated, the
association between VO2max and intelligence is not statistically
explained by modularity. Exploratory mediation analyses using
different dependent, independent, and mediating variables also did
not yield significant results.
Discussion
Physical fitness and cognitive functioning are related. We show
that this relation is also associated with topology of the functional
brain network during the resting-state. Decreased upper alpha and
beta band modularity were related to higher VO2 max, with
higher beta between-module connectivity being associated with
better physical fitness. Average functional connectivity did not
show this association with VO2 max. The association between
Figure 2. Significant sensor-specific associations between
modular connectivity and VO2 max. Note. (a) shows an FDR-
corrected t-map of significant associations between alpha band and
beta band (left and right panel, resp.) between-module connectivity
and VO2 max, while (b) shows the same for within-module connectivity.
Warm colors indicate positive associations, cool colors refer to negative
associations.
doi:10.1371/journal.pone.0088202.g002
Figure 3. Graphical representation of hypothesized mediation
effect. Note. A mediating effect of brain network topology on the
association between VO2 max (maximum oxygen uptake during an
effort test) and intelligence quotient (IQ) was hypothesized.
doi:10.1371/journal.pone.0088202.g003
Table 3. Mediation analyses of network topology on the
association between physical fitness and intelligence.
Upper alpha band modularity (total 95% CI
[20.042 0.055])
Beta
p
VO2max - upper alpha modularity
20.216
0.018*
VO2max - IQ total
0.187
0.046*
VO2max - IQ direct
0.185
0.052
Upper alpha band modularity mediation
20.010
0.885
Beta band modularity (total 95% CI [20.057
0.058])
Beta
p
VO2max - beta modularity
20.197
0.018*
VO2max - IQ total
0.187
0.046*
VO2max - IQ direct
0.188
0.049*
Beta band modularity mediation
0.004
0.959
Beta band PC (total 95% CI [20.032 0.074])
Beta
p
VO2max - beta PC
0.206
0.021*
VO2max - IQ total
0.187
0.046*
VO2max - IQ direct
0.184
0.053
Beta band PC mediation
0.013
0.854
Note. * p,0.05, CI = total confidence interval of indirect effects, based on 5,000
bootstrap samples.
Adjusted for sex, head surface, and relative band power. PC = participation
coefficient.
doi:10.1371/journal.pone.0088202.t003
Brain Networks and Physical Fitness
PLOS ONE | www.plosone.org
5
February 2014 | Volume 9 | Issue 2 | e88202
cardiorespiratory
fitness
and
intelligence
was
however
not
statistically mediated by network characteristics.
Modularity refers to the extent to which the brain can be
subdivided into coherent subsystems. Although such a modular
organization is generally beneficial for brain functioning [35],
having consistently tight connectivity within modules may be
detrimental. Our results show negative correlations between
modularity and both mental and physical functioning, indicating
that higher levels of within-module connectivity versus between-
module connectivity may be related to decreased functioning. An
MEG study comparing modularity during several conditions of a
working memory task reports decreasing modularity, i.e. increas-
ing intermodular communication, as effort increases [49]. A study
compiling a large number of task-related fMRI and PET studies
also shows the importance of the modular organization of the
brain for cognitive functioning [38]. However, how modularity
relates to healthy functioning during the resting-state has not been
reported.
Another resting-state study using modularity reports increased
delta and theta band modularity in Alzheimer’s patients when
compared to healthy controls, which was related to poorer
performance on a fluency task [37]. Furthermore, an fMRI study
during
task
performance
did
not
find
changes
in
overall
modularity over consecutive learning sessions, but does report
that the flexibility of particular nodes, i.e. the number of times that
each node in the network changes its belonging to specific
modules, was related to better performance [50]. That is, having a
highly dynamic modular structure, as opposed to a fixed modular
division, was related to better functioning. These task-based
findings
concerning
network
flexibility
have
recently
been
replicated, localizing these multi tasking nodes mainly in the
fronto-parietal network [51]. Our findings indicate that the
resting-state is characterized by lower modularity and increased
between-module, possibly long-range connections in brighter and
fitter individuals. It would be interesting to investigate the
transition from resting-state to any task, which may indicate that
the resting-state modular flexibility of the brain network is similar
to task-based dynamics.
The effects of modularity and between-module connectivity
were present in the upper alpha and beta bands. These frequency
bands havebeen studied extensively with respect to cognitive tasks,
albeit mostly with respect to power and not connectivity or
network properties. The (upper) alpha band has been related to
attention and working memory [52–55], while the beta band has
been implicated in learning, novelty detection, and reward
evaluation, indicating that this oscillation might be an important
mechanism for directing attention towards a novel stimulus [56–
59]. A previous study used electroencephalography (EEG) to
investigate connectivity and network efficiency during a task in
active versus sedentary subjects [60]. Results show that in the beta
band, active subjects show greater connectivity and network
efficiency than sedentary subjects. Similar results were obtained
when using coherence as a measure of connectivity, also in the
alpha and beta band [61]. None of these studies investigated
resting-state network topology.
With respect to the previously described study investigating
modularity during increasingly difficult cognitive conditions [49],
most effects of neural reconfiguration were found in the beta band,
which the authors ascribe to the need for higher long-range
synchronization, increased intermodular connectivity, and thus
loss of modularity in this frequency band during tasks. This
hypothesis, as well as our results, are corroborated by computa-
tional and animal work, showing that beta oscillations provide
excellent support for long-distance synchronization [59,62]. The
beta band may speculatively be at the heart of communication
between hub areas in the brain, which regulate higher-order
functioning
of
the
brain
network
and
therefore
relate
to
intelligence and cardio respiratory fitness, although more studies
are needed to confirm this hypothesis.
Previous studies have only reported associations between
resting-state functional connectivity, cognitive functioning, and
physical fitness. Particularly higher connectivity within the default
mode network (DMN) has been related to increased cardiorespi-
ratory
fitness,
while
DMN
connectivity
also
mediates
the
association between VO2 max and cognitive functioning [17].
After a 1-year exercise intervention in older adults, both the DMN
and the fronto-parietal network show higher connectivity than a
control group [18], further building on the causal relationships
that might exist between physical fitness and functional connec-
tivity. Our results partly corroborate these findings, and indicate
that there might be differential associations with particular types of
connectivity: in our investigation of a very large cohort of healthy
subjects with a direct measurement of neural activity, particularly
increased between-module connectivity was related to superior
cardiorespiratory fitness and intelligence. Also, our lack of findings
with respect to average functional connectivity indicate that
network analysis contributes valuable information to the associa-
tion between fitness and intelligence, and advocates for investiga-
tion of the brain network as a whole instead of only focusing on
connectivity between particular spatially determined areas.
Although circumstantial evidence is available, the definite
direction of the association between increased cardio respiratory
fitness, functional brain network organization, and cognition is still
uncertain, and our results in a large sample do not support the
hypothesis that better physical condition leads to better intelli-
gence through brain network topology. Several aerobic interven-
tion studies, which usually randomize between an exercise
program and a control intervention of for instance light stretching,
have reported increased cognitive functioning afterwards, but a
number of studies failed to find a cognitive effect of increased
cardio respiratory fitness [7,63]. Our study was not aimed at
addressing this issue, and mediation analyses were not significant.
Additionally, measurement of VO2 max took place approximately
six years prior to intelligence testing and MEG recording. Our
analysis of VO2 max at previous time points suggests that this
measurement is a relatively stable measure of physical fitness, and
all subjects with disease burden influencing their lifestyle were
excluded. Finally, the presence of associations between VO2 max
and intelligence six years later suggest that we are indeed looking
at a robust indication of physical fitness. However, we cannot
ascertain that this interval between measurements did not
influence our results. Future longitudinal studies are needed to
shed light on the causal relations between cardiorespiratory fitness,
intelligence, and network topology, while investigation of anatom-
ical brain connections may also yield further insights into this issue.
Increased physical fitness is associated with better functional
brain network topology. The step from exercise to functional brain
network may be difficult to understand. On a cellular level, better
physical fitness has often been associated with increases in BDNF
[64,65], and possibly with IGF-1 and VEGF [66]. A recent study
suggests that these exercise-induced cellular changes are indeed
related to functional connectivity, by comparing BDNF, IGF-1
and VEGF levels in two groups of participants undergoing either
an aerobic or non-aerobic intervention [20]. The link between
cellular biology and network functioning as measured with MEG
has recently also been addressed in a study of protein expression
and epilepsy in brain tumor patients [67]. We were able to show a
direct association between epilepsy-related protein expression and
Brain Networks and Physical Fitness
PLOS ONE | www.plosone.org
6
February 2014 | Volume 9 | Issue 2 | e88202
between-module connectivity of the tumor area, further indicating
that these network patterns may be the intermediate between
molecules and behavior. Future studies are needed to further
explore how cellular changes as a consequence of exercise lead to
changes in functional connectivity.
Several limitations of the current study should be recognized.
First of all, as previously mentioned, measurement of cardiorespi-
ratory fitness was performed several years before MEG recording
and intelligence testing took place. The influence of the lag
between measurements in the current study design on the reported
results is unknown. Secondly, this study was performed on the
sensor-level, since no anatomical MRI scans (which are necessary
to perform accurate source reconstruction in MEG data) were
available. This limits the spatial specificity of our results, and
prohibits further investigation of specific spatial network proper-
ties. Third, the spatial resolution of MEG is limited. Although
MEG is much less sensitive to volume conduction and disturbing
effects of the skull and scalp than EEG, common sources still pose
a serious problem for coupling analysis. However, the phase lag
index is a particularly strict measure of functioning connectivity,
because it excludes all non-zero lagged correlations [44,68].
In conclusion, we show that functional brain network organi-
zation may mediate the association between cardiorespiratory
fitness and intelligence. Less tightly connected, more intercon-
nected functional modular topology in the upper alpha and
particularly beta band may promote long-range connectivity in the
resting-state, which relates to both increased physical and mental
fitness.
Supporting Information
Figure S1
Temporal consistency of VO2 max measure-
ments
in
the
AGAHLS
cohort.
Note.
(a)
depicts
the
correlations between VO2 max measurements at each neighboring
time point in the AGAHLS study, the first two measurements
being performed at 13 and 14 years old (YO). The number of
overlapping subjects between time points is indicated in paren-
theses. In (b), the first adult VO2 max measurement at 21 years old
is correlated to the last measurement at 36 years old, which we
used in this study (correlation coefficient of 0.791, p,0.001).
(TIFF)
Acknowledgments
The authors would like to thank K. Plugge, P. Ris and N. Sijsma for their
help in recording MEG in all subjects.
Author Contributions
Conceived and designed the experiments: LD DN BWD CJS JWRT.
Performed the experiments: LD BWD JWRT. Analyzed the data: LD DN
BWD CJS JWRT. Contributed reagents/materials/analysis tools: BWD
CJS. Wrote the paper: LD DN BWD CJS JWRT.
References
1. Oguma Y, Sesso HD, Paffenbarger RS, Lee I-M (2002) Physical activity and all
cause mortality in women: a review of the evidence. BrJSports Med 36: 162–
172.
2. Barlow CE, LaMonte MJ, Fitzgerald SJ, Kampert JB, Perrin JL, et al. (2006)
Cardiorespiratory fitness is an independent predictor of hypertension incidence
among initially normotensive healthy women. AmJEpidemiol 163: 142–150.
doi:10.1093/aje/kwj019.
3. Blair SN, Kampert JB, Kohl HW, Barlow CE, Macera CA, et al. (1996)
Influences of cardiorespiratory fitness and other precursors on cardiovascular
disease and all-cause mortality in men and women. JAMA 276: 205–210.
4. Evenson KR, Stevens J, Cai J, Thomas R, Thomas O (2003) The effect of
cardiorespiratory fitness and obesity on cancer mortality in women and men.
Med Sci Sports Exerc 35: 270–277. doi:10.1249/01.MSS.0000053511.02356.
72.
5. Kodama S, Saito K, Tanaka S, Maki M, Yachi Y, et al. (2009) Cardiorespiratory
fitness as a quantitative predictor of all-cause mortality and cardiovascular events
in healthy men and women: a meta-analysis. JAMA 301: 2024–2035.
doi:10.1001/jama.2009.681.
6. Lynch J, Helmrich SP, Lakka TA, Kaplan GA, Cohen RD, et al. (1996)
Moderately intense physical activities and high levels of cardiorespiratory fitness
reduce the risk of non-insulin-dependent diabetes mellitus in middle-aged men.
Arch Intern Med 156: 1307–1314.
7. Ho¨tting K, Ro¨der B (2013) Beneficial effects of physical exercise on
neuroplasticity and cognition. Neurosci Biobehav Rev. doi:10.1016/j.neu-
biorev.2013.04.005.
8. Bullitt E, Rahman FN, Smith JK, Kim E, Zeng D, et al. (2009) The effect of
exercise on the cerebral vasculature of healthy aged subjects as visualized by MR
angiography. AJNR AmJNeuroradiol 30: 1857–1863. doi:10.3174/ajnr.A1695.
9. Knaepen K, Goekint M, Heyman EM, Meeusen R (2010) Neuroplasticity -
exercise-induced response of peripheral brain-derived neurotrophic factor: a
systematic review of experimental studies in human subjects. Sports Med 40:
765–801. doi:10.2165/11534530-000000000-00000.
10. Erickson KI, Prakash RS, Voss MW, Chaddock L, Hu L, et al. (2009) Aerobic
fitness is associated with hippocampal volume in elderly humans. Hippocampus
19: 1030–1039. doi:10.1002/hipo.20547.
11. Erickson KI, Weinstein AM, Sutton BP, Prakash RS, Voss MW, et al. (2012)
Beyond vascularization: aerobic fitness is associated with N-acetylaspartate and
working memory. Brain Behav 2: 32–41. doi:10.1002/brb3.30.
12. Rosazza C, Minati L (2011) Resting-state brain networks: literature review and
clinical applications. Neurol Sci 32: 773–785. doi:10.1007/s10072-011-0636-y.
13. Damoiseaux JS, Rombouts SA, Barkhof F, Scheltens P, Stam CJ, et al. (2006)
Consistent resting-state networks across healthy subjects. Proc Natl Acad
SciUSA 103: 13848–13853.
14. Buckner RL, Vincent JL (2007) Unrest at rest: default activity and spontaneous
network correlations. Neuroimage 37: 1091–6; discussion 1097–9. doi:10.1016/
j.neuroimage.2007.01.010.
15. Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, et al. (2001)
A default mode of brain function. Proc Natl Acad SciUSA 98: 676–682.
16. Buckner RL, Sepulcre J, Talukdar T, Krienen FM, Liu H, et al. (2009) Cortical
hubs revealed by intrinsic functional connectivity: mapping, assessment of
stability, and relation to Alzheimer’s disease. JNeurosci 29: 1860–1873.
17. Voss MW, Erickson KI, Prakash RS, Chaddock L, Malkowski E, et al. (2010)
Functional connectivity: a source of variance in the association between
cardiorespiratory fitness and cognition? Neuropsychologia 48: 1394–1406.
doi:10.1016/j.neuropsychologia.2010.01.005.
18. Voss MW, Prakash RS, Erickson KI, Basak C, Chaddock L, et al. (2010)
Plasticity of brain networks in a randomized intervention trial of exercise
training in older adults. Front Aging Neurosci 2: 1–17. doi:10.3389/
fnagi.2010.00032.
19. McFadden KL, Cornier M-A, Melanson EL, Bechtell JL, Tregellas JR (2013)
Effects of exercise on resting-state default mode and salience network activity in
overweight/obese adults. Neuroreport 24: 866–871. doi:10.1097/
WNR.0000000000000013.
20. Voss MW, Erickson KI, Prakash RS, Chaddock L, Kim JS, et al. (2013)
Neurobiological markers of exercise-related brain plasticity in older adults. Brain
Behav Immun 28: 90–99. doi:10.1016/j.bbi.2012.10.021.
21. Stam CJ, van Straaten ECW (2012) The organization of physiological brain
networks. Clin Neurophysiol 123: 1067–1087.
22. Bassett DS, Bullmore ET (2009) Human brain networks in health and disease.
Curr Opin Neurol 22: 340–347.
23. Sporns O (2013) Network attributes for segregation and integration in the
human brain. Curr Opin Neurobiol. doi:10.1016/j.conb.2012.11.015.
24. Watts DJ, Strogatz SH (1998) Collective dynamics of ‘‘small-world’’ networks.
Nature 393: 440–442.
25. Bassett DS, Meyer-Lindenberg A, Achard S, Duke T, Bullmore E (2006)
Adaptive reconfiguration of fractal small-world human brain functional
networks. Proc Natl Acad SciUSA 103: 19518–19523. doi:10.1073/
pnas.0606005103.
26. Stam CJ (2004) Functional connectivity patterns of human magnetoencephalo-
graphic recordings: a ‘‘small-world’’ network? Neurosci Lett 355: 25–28.
27. Sporns O, Zwi JD (2004) The small world of the cerebral cortex. Neuroinfor-
matics 2: 145–162.
28. Fornito A, Zalesky A, Bassett DS, Meunier D, Ellison-Wright I, et al. (2011)
Genetic influences on cost-efficient organization of human cortical functional
networks. JNeurosci 31: 3261–3270. doi:10.1523/JNEUROSCI.4858-10.2011.
29. Smit DJ, Stam CJ, Posthuma D, Boomsma DI, de Geus EJ (2008) Heritability of
‘‘small-world’’ networks in the brain: A graph theoretical analysis of resting-state
EEG functional connectivity. Hum Brain Mapp 29: 1368–1378.
30. Schutte NM, Hansell NK, de Geus EJC, Martin NG, Wright MJ, et al. (2013)
Heritability of Resting State EEG Functional Connectivity Patterns. Twin Res
Hum Genet: 1–8. doi:10.1017/thg.2013.55.
Brain Networks and Physical Fitness
PLOS ONE | www.plosone.org
7
February 2014 | Volume 9 | Issue 2 | e88202
31. Bullmore E, Sporns O (2009) Complex brain networks: graph theoretical
analysis of structural and functional systems. Nat Rev Neurosci 10: 186–198.
32. Van den Heuvel MP, Stam CJ, Kahn RS, Hulshoff Pol HE (2009) Efficiency of
functional brain networks and intellectual performance. JNeurosci 29: 7619–
7624.
33. Douw L, Schoonheim MM, Landi D, van der Meer ML, Geurts JJ, et al. (2011)
Cognition is related to resting-state small-world network topology: an
magnetoencephalographic study. Neuroscience 175: 169–177.
34. Newman MEJ (2006) Modularity and community structure in networks. Proc
Natl Acad SciUSA 103: 8577–8582. doi:10.1073/pnas.0601602103.
35. Meunier D, Lambiotte R, Bullmore ET (2011) Modular and hierarchically
modular organization of brain networks. Front Neurosci 4: 200.
36. Van Diessen E, Hanemaaijer JI, Otte WM, Zelmann R, Jacobs J, et al. (2013)
Are high frequency oscillations associated with altered network topology in
partial epilepsy? Neuroimage 82C: 564–573. doi:10.1016/j.neuro-
image.2013.06.031.
37. De Haan W, van der Flier WM, Koene T, Smits LL, Scheltens P, et al. (2012)
Disrupted modular brain dynamics reflect cognitive dysfunction in Alzheimer’s
disease. Neuroimage 59: 3085–3093. doi:10.1016/j.neuroimage.2011.11.055.
38. Crossley NA, Mechelli A, Ve´rtes PE, Winton-Brown TT, Patel AX, et al. (2013)
Cognitive relevance of the community structure of the human brain functional
coactivation network. Proc Natl Acad SciUSA 110: 11583–11588. doi:10.1073/
pnas.1220826110.
39. Kemper HC (2004) Amsterdam growth and health longitudinal study: a 23-year
follow-up from teenager to adult about lifestyle and health. Basel, Switzerland:
Karger.
40. Wijnstok NJ, Hoekstra T, van Mechelen W, Kemper HCG, Twisk JWR (2013)
Cohort profile: the Amsterdam Growth and Health Longitudinal Study.
IntJEpidemiol 42: 422–429. doi:10.1093/ije/dys028.
41. Luteijn F, Ploeg F Van der (1983) Groninger intelligentie test.
42. Kooreman A, Luteijn F (1987) Groninger Intelligentie Test GIT: Schriftelijke
verkorte vorm.
43. Stam CJ, de Haan W, Daffertshofer A, Jones BF, Manshanden I, et al. (2009)
Graph theoretical analysis of magnetoencephalographic functional connectivity
in Alzheimer’s disease. Brain 132: 213–224.
44. Stam CJ, Nolte G, Daffertshofer A (2007) Phase lag index: Assessment of
functional connectivity from multi channel EEG and MEG with diminished bias
from common sources. Hum Brain Mapp 28: 1178–1193.
45. Guimera R, Nunes Amaral LA (2005) Functional cartography of complex
metabolic networks. Nature 433: 895–900.
46. Stam CJ (2010) Use of magnetoencephalography (MEG) to study functional
brain networks in neurodegenerative disorders. JNeurol Sci 289: 128–134.
47. Preacher KJ, Hayes AF (2008) Asymptotic and resampling strategies for
assessing and comparing indirect effects in multiple mediator models. Behav Res
Methods 40: 879–891.
48. Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: A practical
and powerful approach to multiple testing. JRStat Soc SerB 57: 289–300.
49. Kitzbichler MG, Henson RN a, Smith ML, Nathan PJ, Bullmore ET (2011)
Cognitive effort drives workspace configuration of human brain functional
networks. JNeurosci 31: 8259–8270. doi:10.1523/JNEUROSCI.0440-11.2011.
50. Bassett DS, Wymbs NF, Porter MA, Mucha PJ, Carlson JM, et al. (2011)
Dynamic reconfiguration of human brain networks during learning. Proc Natl
Acad SciUSA 108: 7641–7646.
51. Cole MW, Reynolds JR, Power JD, Repovs G, Anticevic A, et al. (2013) Multi-
task connectivity reveals flexible hubs for adaptive task control. Nat Neurosci 16:
1348–1355. doi:10.1038/nn.3470.
52. Basar E, Basar-Eroglu C, Karakas S, Schurmann M (1999) Are cognitive
processes manifested in event-related gamma, alpha, theta and delta oscillations
in the EEG? Neurosci Lett 259: 165–168.
53. Klimesch W, Schack B, Sauseng P (2005) The functional significance of theta
and upper alpha oscillations. Exp Psychol 52: 99–108.
54. Stam CJ (2000) Brain dynamics in theta and alpha frequency bands and working
memory performance in humans. Neurosci Lett 286: 115–118.
55. Gootjes L, Bouma A, Van Strien JW, Scheltens P, Stam CJ (2006) Attention
modulates hemispheric differences in functional connectivity: evidence from
MEG recordings. Neuroimage 30: 245–253.
56. Kisley MA, Cornwell ZM (2006) Gamma and beta neural activity evoked during
a sensory gating paradigm: effects of auditory, somatosensory and cross-modal
stimulation. Clin Neurophysiol 117: 2549–2563. doi:10.1016/j.clinph.2006.
08.003.
57. Haenschel C, Baldeweg T, Croft RJ, Whittington M, Gruzelier J (2000) Gamma
and beta frequency oscillations in response to novel auditory stimuli: A
comparison of human electroencephalogram (EEG) data with in vitro models.
Proc Natl Acad SciUSA 97: 7645–7650. doi:10.1073/pnas.120162397.
58. Martin C (2004) Learning Modulation of Odor-Induced Oscillatory Responses
in the Rat Olfactory Bulb: A Correlate of Odor Recognition? JNeurosci 24:
389–397. doi:10.1523/JNEUROSCI.3433-03.2004.
59. White JA, Banks MI, Pearce RA, Kopell NJ (2000) Networks of interneurons
with fast and slow gamma-aminobutyric acid type A (GABAA) kinetics provide
substrate for mixed gamma-theta rhythm. Proc Natl Acad SciUSA 97: 8128–
8133. doi:10.1073/pnas.100124097.
60. Kamijo K, O9Leary KC, Pontifex MB, Themanson JR, Hillman CH (2010) The
relation of aerobic fitness to neuroelectric indices of cognitive and motor task
preparation.
Psychophysiology
47:
814–821.
doi:10.1111/j.1469-
8986.2010.00992.x.
61. Hogan M, Kiefer M, Kubesch S, Collins P, Kilmartin L, et al. (2013) The
interactive effects of physical fitness and acute aerobic exercise on electrophys-
iological coherence and cognitive performance in adolescents. Exp Brain Res
229: 85–96. doi:10.1007/s00221-013-3595-0.
62. Tallon-Baudry C, Mandon S, Freiwald WA, Kreiter AK (2004) Oscillatory
synchrony in the monkey temporal lobe correlates with performance in a visual
short-term memory task. Cereb Cortex 14: 713–720. doi:10.1093/cercor/
bhh031.
63. Angevaren M, Aufdemkampe G, Verhaar HJJ, Aleman A, Vanhees L (2008)
Physical activity and enhanced fitness to improve cognitive function in older
people without known cognitive impairment. Cochrane database Syst Rev:
CD005381. doi:10.1002/14651858.CD005381.pub2.
64. Vaynman S, Ying Z, Gomez-Pinilla F (2004) Hippocampal BDNF mediates the
efficacy of exercise on synaptic plasticity and cognition. EurJNeurosci 20: 2580–
2590. doi:10.1111/j.1460-9568.2004.03720.x.
65. Ding Q, Vaynman S, Akhavan M, Ying Z, Gomez-Pinilla F (2006) Insulin-like
growth factor I interfaces with brain-derived neurotrophic factor-mediated
synaptic plasticity to modulate aspects of exercise-induced cognitive function.
Neuroscience 140: 823–833. doi:10.1016/j.neuroscience.2006.02.084.
66. Cotman CW, Berchtold NC, Christie L-A (2007) Exercise builds brain health:
key roles of growth factor cascades and inflammation. Trends Neurosci 30: 464–
472. doi:10.1016/j.tins.2007.06.011.
67. Douw L, Groot M de, van Dellen E, Aronica E, Heimans JJ, et al. (2013) Local
MEG networks: the missing link between protein expression and epilepsy in
glioma
patients?
Neuroimage
75:
195–203.
doi:10.1016/j.neuro-
image.2013.02.067.
68. Peraza LR, Asghar AUR, Green G, Halliday DM (2012) Volume conduction
effects in brain network inference from electroencephalographic recordings using
phase lag index. JNeurosci Methods 207: 189–199. doi:10.1016/j.jneu-
meth.2012.04.007.
Brain Networks and Physical Fitness
PLOS ONE | www.plosone.org
8
February 2014 | Volume 9 | Issue 2 | e88202
| A healthy brain in a healthy body: brain network correlates of physical and mental fitness. | 02-03-2014 | Douw, Linda,Nieboer, Dagmar,van Dijk, Bob W,Stam, Cornelis J,Twisk, Jos W R | eng |
PMC9794057 |
1
S1 Table. CREDES checklist.
Items of reporting
Reported on
page
Purpose and rationale. The purpose of the study should be clearly defined and
demonstrate the appropriateness of the use of the Delphi technique as a
method to achieve the research aim. A rationale for the choice of the Delphi
technique as the most suitable method needs to be provided.
4
Expert panel. Criteria for the selection of experts and transparent information
on recruitment of the expert panel, sociodemographic details including
information on expertise regarding the topic in question, (non)response and
response rates over the ongoing iterations should be reported.
5, 8-9
Description of the methods. The methods employed need to be
comprehensible; this includes information on preparatory steps (How was
available evidence on the topic in question synthesised?), piloting of material
and survey instruments, design of the survey instrument(s), the number and
design of survey rounds, methods of data analysis, processing and synthesis
of experts’ responses to inform the subsequent survey round and
methodological decisions taken by the research team throughout the process.
4-7
Procedure. Flow chart to illustrate the stages of the Delphi process, including a
preparatory phase, the actual ‘Delphi rounds’, interim steps of data processing
and analysis, and concluding steps.
Fig 1
Definition and attainment of consensus. It needs to be comprehensible to the
reader how consensus was achieved throughout the process, including
strategies to deal with non-consensus.
4
Results. Reporting of results for each round separately is highly advisable in
order to make the evolving of consensus over the rounds transparent. This
includes figures showing the average group response, changes between
rounds, as well as any modifications of the survey instrument such as deletion,
addition or modification of survey items based on previous rounds.
8-11,
Table 2,
S5-S7 Tables
Discussion of limitations. Reporting should include a critical reflection of
potential limitations and their impact of the resulting guidance.
14-15
Adequacy of conclusions. The conclusions should adequately reflect the
outcomes of the Delphi study with a view to the scope and applicability of the
resulting practice guidance.
16
Publication and dissemination.
-
| Factors associated with high-level endurance performance: An expert consensus derived via the Delphi technique. | 12-27-2022 | Konopka, Magdalena J,Zeegers, Maurice P,Solberg, Paul A,Delhaije, Louis,Meeusen, Romain,Ruigrok, Geert,Rietjens, Gerard,Sperlich, Billy | eng |
PMC4370475 | RESEARCH ARTICLE
Sprint Conditioning of Junior Soccer Players:
Effects of Training Intensity and Technique
Supervision
Thomas Haugen1,2*, Espen Tønnessen1, Øyvind Øksenholt3, Fredrik Lie Haugen3,
Gøran Paulsen1,3, Eystein Enoksen3, Stephen Seiler2
1 Norwegian Olympic Sports Program (Olympiatoppen), Sognsveien 228, 0840 Oslo, Norway, 2 Faculty of
Health and Sport Sciences, University of Agder, Gimlemoen 25, 4630 Kristiansand, Norway, 3 Norwegian
School of Sport Sciences, Sognsveien 220, 0806 Oslo, Norway
* [email protected]
Abstract
The aims of the present study were to compare the effects of 1) training at 90 and 100%
sprint velocity and 2) supervised versus unsupervised sprint training on soccer-specific
physical performance in junior soccer players. Young, male soccer players (17 ±1 yr,
71 ±10 kg, 180 ±6 cm) were randomly assigned to four different treatment conditions over a
7-week intervention period. A control group (CON, n=9) completed regular soccer training
according to their teams’ original training plans. Three training groups performed a weekly
repeated-sprint training session in addition to their regular soccer training sessions per-
formed at A) 100% intensity without supervision (100UNSUP, n=13), B) 90% of maximal
sprint velocity with supervision (90SUP, n=10) or C) 90% of maximal sprint velocity without
supervision (90UNSUP, n=13). Repetitions x distance for the sprint-training sessions were
15x20 m for 100UNSUP and 30x20 m for 90SUP and 90UNSUP. Single-sprint performance
(best time from 15x20 m sprints), repeated-sprint performance (mean time over 15x20 m
sprints), countermovement jump and Yo-Yo Intermittent Recovery Level 1 (Yo-Yo IR1)
were assessed during pre-training and post-training tests. No significant differences in per-
formance outcomes were observed across groups. 90SUP improved Yo-Yo IR1 by a mod-
erate margin compared to controls, while all other effect magnitudes were trivial or small. In
conclusion, neither weekly sprint training at 90 or 100% velocity, nor supervised sprint train-
ing enhanced soccer-specific physical performance in junior soccer players.
Introduction
The importance of sprinting in professional soccer is well established and the need for speed is
clear [1–4]. According to track-and-field statistics [5], trends over time from large retrospective
data collections in soccer players [3,4] and the experience of practitioners [6], sprint perfor-
mance is resistant to training enhancement. Athletes can spend years training to improve a few
hundredths of a second over short distances [5]. Numerous intervention studies have been
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
1 / 13
a11111
OPEN ACCESS
Citation: Haugen T, Tønnessen E, Øksenholt Ø,
Haugen FL, Paulsen G, Enoksen E, et al. (2015)
Sprint Conditioning of Junior Soccer Players: Effects
of Training Intensity and Technique Supervision.
PLoS ONE 10(3): e0121827. doi:10.1371/journal.
pone.0121827
Academic Editor: Oyvind Sandbakk, Norwegian
University of Science and Technology, NORWAY
Received: November 10, 2014
Accepted: February 4, 2015
Published: March 23, 2015
Copyright: © 2015 Haugen et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any
medium, provided the original author and source are
credited.
Data Availability Statement: All relevant data are
within the paper.
Funding: The authors have no support or funding to
report.
Competing Interests: The authors have declared
that no competing interests exist.
performed over the years in order to enhance soccer-specific sprinting. A recent review reveals
that sprinting under assisted, resisted and normal conditions, maximal and explosive strength
training, plyometric training and high-intensity running have been investigated in different
combinations, but no specific training method has so far emerged as superior [1]. Time effi-
ciency is an important constraining aspect of team-sport conditioning and extensive off-field
interventions will most likely be rejected by team coaches, independent of intervention
efficacy [1].
The term ‘direct supervision’ refers to training situations in which a supervisor or training
expert is present at all times [7,8]. The supervisor oversees training activities as they occur and
provides direction, instruction, feedback and assistance. The importance of guidance and feed-
back during practice is well known in motor skill learning and performance enhancements
may happen immediately in such settings [9]. Mazzetti et al. [7] and Coutts et al. [8] concluded
that the presence of a training expert was beneficial for maximal strength development over
time. To the authors’ knowledge, the effect of supervised sprint-training sessions in soccer
players has not been investigated. According to motor skill learning theories, errors increase
with the speed of the movement [9]. Technical training of typically rapid or ballistic move-
ments should be interfered with by using specific drills, large amount of repetitions and an in-
tensity where the athletes are able to control the movements (proper execution not interfered
by fatigue). If the movement is slowed down slightly, the same generalized motor program can
be used as in the normal-speed version [9]. In contrast, the vast majority of studies involving
sprint-training interventions for soccer players make no recommendations other than that
sprint velocity should be maximal throughout [1]. Available evidence in endurance and
strength training demonstrates that high, but sub-maximal intensity loading effectively stimu-
lates adaptation through the interaction between high intensity and larger accumulated work
that can be achieved before the onset of fatigue, compared to maximal efforts [10,11]. This
makes it tempting to speculate similar effects on sprinting. Anecdotal evidence in support of
this is observed in the sprint-training philosophy developed by the athletic sprint pioneer
coach Carlo Vittori in the mid-1970s [12]. His successful athletes performed repeated-sprint
training sessions with an intensity as low as 90% of maximal sprint speed during initial pre-sea-
son conditioning in order to improve sprint endurance (later termed repeated-sprint perfor-
mance). Inspection of training diaries reveals that internationally-competing sprinters perform
sprint training with varying intensity through all parts of the season (unpublished material,
Norwegian Olympic Federation). However, the lowest effective sprinting intensity for stimulat-
ing adaptation is so far not established in the research literature. Recently, Haugen et al. [13]
observed that repeated 20-m sprints at 90% intensity did not enhance sprint performance dur-
ing a soccer season. It was suggested that such training should be performed at other times of
the season to avoid training-related constraints due to the high volume of overall soccer condi-
tioning. The aims of the present study were therefore to compare the effects of 1) training at 90
and 100% sprint velocity and 2) supervised versus unsupervised sprint training at 90% sprint
velocity on soccer-specific physical performance capacities in junior soccer players’ early in
pre-season.
Materials and Methods
Ethics statement
This study was conducted in accordance with the declaration of Helsinki. All participants pro-
vided written, voluntary informed consent before participation. Written parental consent was
also provided for participants < 18 yr old. The human subjects review committee of the Faculty
for Health and Sport, University of Agder, approved the study.
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
2 / 13
Experimental approach to the problem
In this randomized controlled trial, participants were randomly assigned to four different treat-
ment conditions. A control group (CON) completed regular soccer training according to their
teams’ original early pre-season training plans. Three training groups performed a weekly re-
peated-sprint training session in addition to their regular soccer training sessions, which was
performed at A) 100% intensity without supervision (100UNSUP), B) 90% of maximal sprint
speed with supervision (90SUP) or C) 90% of maximal sprint speed without supervision
(90UNSUP). Based on sample size limitations and motor learning principles identified in the
introduction, the present study was not performed with a factorial design (i.e. an additional
“100SUP” group). The duration of the intervention period was 7 weeks. To evaluate the treat-
ment conditions (independent variables), the following soccer-specific performance tests (pri-
mary dependent variables) were assessed prior to and after the intervention period: 15x20 m
repeated-sprint, countermovement jump (CMJ) and Yo-Yo Intermittent Recovery 1 (Yo-Yo
IR1). To investigate possible mechanistic influences regarding adaptations to sprint training,
the following secondary dependent variables were assessed during the 15x20 m repeated-sprint
pre- and post-training tests: Heart rate, blood lactate concentration, step length and step rate.
Finally, sprint times for all training sessions were assessed for intensity control (90SUP and
90UNSUP) and to examine weekly changes in repeated-sprint performance (100UNSUP).
Participants
Fifty-two male junior soccer players, aged 16–19 years, volunteered to participate. The athletes
were playing in the highest junior division level for four different clubs (n = 6,13,16 and 17) in
Norway. Each participant had a minimum two years of soccer-specific conditioning experi-
ence. During the intervention period, the participants were requested to refrain from perform-
ing any other off-field physical training regimes in terms of speed, strength and/or endurance.
All participants were free of injuries prior to preliminary testing. None of the athletes had pre-
vious experience with specialized repeated-sprint training.
To eliminate the influence of varying overall soccer conditioning, the participants were ini-
tially distributed by club and then allocated to one of the four intervention conditions by a co-
author not directly involved in testing or the training intervention. The 14 participants ran-
domly assigned to each of the three training groups were required to complete at least six out
of seven training sessions during the intervention period in addition to all performance tests in
order to be included in further analyses. The 10 allocated CON participants were required to
perform at least 80% of planned sessions and complete all pre- and post-training tests. We
chose a slightly uneven distribution of subjects based on 1) the expectation of increased drop-
out risk generally observed in any intervention and 2) the expectation of lower variability of
outcome in CON exposed to testing only and an unchanged training routine.
One participant each from CON, 100UNSUP and 90SUP dropped out due to illness during
training or testing. Two participants from 90SUP and one from 90UNSUP dropped out due to
injuries sustained outside the sprint-training intervention. A final player from 90SUP group
dropped out due to Achilles tendon strain, possibly associated with the sprint intervention.
Thus, 45 of 52 participants completed the study with the following sample sizes (club distribu-
tion in brackets): CON = 9 (0,3,3,3), 100UNSUP = 13 (0,4,5,4), 90UNSUP = 13 (1,3,5,4) and
90SUP = 10 (2,3,3,2). Physical and training characteristics of these participants are presented
in Table 1.
Regular soccer training sessions typically commenced with warm-up activities like
short-passing and coordination exercises with the ball, followed by more intensive
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
3 / 13
change-of-direction exercises with and without ball. The main part of the soccer practice con-
sisted of small-sided and more full-sized team compositions, ranging from 3 vs. 3 to 7 vs. 11.
Testing procedures
The pre- and post-training tests were conducted at the Norwegian Olympic Training Centre
on two separate days, with two days in between. All participants completed the tests in the
same order and at the same time of day. Regarding nutrition, hydration, sleep and physical ac-
tivity, the athletes were instructed to prepare as they would for a regular soccer match, includ-
ing no high-intensity training the last two days before testing. They were also instructed to use
identical footwear and kit for each of the tests. Test day one consisted of CMJ and 15x20 m re-
peated- sprint testing. On test day two, the athletes completed the Yo-Yo IR1 test. Prior to test-
ing on test-day 1, participants completed a 25 min standardized treadmill warm-up consisting
of a 10-min general warm-up at 60–75% of age-predicted maximum heart rate, 3 sets of 4 exer-
cise drills (high knees, back kick, sideway and backwards running) and finally 2–3 repetitions
of 40-m runs with a progressive increase in speed. Prior to testing on test-day 2, participants
warmed up with a 10-min easy jog at 60–75% of age-predicted heart rate followed by the initial
60–90 s of the Yo-Yo IR1 test.
CMJ test. Immediately after warm up, each athlete was weighed on a force platform for
system calibration before performing three trials of CMJ (vertical jump) separated by 1 min re-
covery. The best result for each player was retained for analysis. To isolate leg extensor muscles
and minimize technical elements, all jumps were performed with hands placed on the hips.
The tests were performed on an AMTI force platform (OR6–5–1, Watertown, USA). Calcula-
tion of jump height is formerly described in Haugen et al. [3].
Sprint test. A 15x20 m repeated-sprint test with starts each 60 s was performed directly
after the CMJ test. Distance and recovery were chosen in line with mean frequency and typical
distance of all-out sprints reported from match analyses [14]. Procedures and equipment are
formerly described in Haugen et al. [13]. Best 20-m time was used in order to determine maxi-
mal single-sprint capacity, while mean time for the 15 sprints was used to determine repeated-
sprint performance. Heart rate was measured continuously during the test (Polar RS400, Kem-
pele, Finland). A blood sample was acquired via finger stick to quantify the blood lactate
concentration (BLa-) immediately after the last sprint (LactatePro LT-1710, Arkay KDK,
Kyoto, Japan).
All sprint tests were captured by a video camera (Sony HDR-HC9E)) mounted on a tripod
in line with the finish line and 3 m from the sprinter’s running lane. Video recordings were
analysed in ProSuite, version 5.5 (Dartfish, Switzerland) to determine step count and derive av-
erage step length (SL). For precision, the digital ruler in the analyser window was used to
Table 1. Physical and training characteristics at inclusion.
Group
n
Age (yr)
BM (kg)
Height (cm)
Weekly training sessions
Games per week (n)
Tot. vol. (hwk-1)
CON
9
17 ±1
72 ±11
181 ±6
4.4 ±2.3
0.4 ±0.4
6.8 ±3.3
100UNSUP
13
17 ±1
66 ±9*
178 ±6
4.4 ±2.3
0.3 ±0.7
6.6 ±3.8
90UNSUP
13
17 ±1
72 ±6
183 ±5
4.5 ±2.4
0.4 ±1.0
7.0 ±3.5
90SUP
10
17 ±1
72 ±8
178 ±7
4.4 ±1.6
0.4 ±0.9
6.8 ±2.9
Values are mean ± SD. BM = Body mass, Tot. vol. = Total training volume. Training values are based on self-reported weekly averages during the
intervention period. There were no significant differences among the groups for any of the variables, except for body mass (*100UNSUP < 90UNSUP,
p = 0.04).
doi:10.1371/journal.pone.0121827.t001
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
4 / 13
interpolate the last step across the finish line. For example; if the 13th and 14th ground contact
occurred 0.8 m in front of and 1.2 m beyond the finish line, respectively, the recorded number
of steps was registered as 13.4. Mean SL was calculated by dividing the number of steps by the
distance (in this case: 20 m13.4–1 = 1.49 m). Mean step rate (SR) was calculated from mean ve-
locity and mean SL. Prior to the present study, this measurement method was validated by roll-
ing out thin paper at the finish line area in order to measure the distance between the visible
spike shoe marks from competitive sprinters. The absolute difference across twenty sprint
comparisons never exceeded 0.1 steps. Thus, the maximal margin of error for step counts over
20 m is 0.7–0.8% for athletes using 13–15 steps.
Yo-Yo IR1 test.
The Yo-Yo IR1 test was performed indoors on artificial turf. Two test
leaders supervised the tests. The athletes were divided in small groups that completed the test
consecutively, such that each supervisor was responsible for 5 athletes during the test. Set-up
and procedures were in line with the guidelines by Krustrup et al. [15], who have reported a
test-retest CV < 5%. The test score is reported in total distance covered until exhaustion.
Intervention program
The training intervention took place from the end of October to mid December, corresponding
to early pre-season in the Norwegian soccer annual cycle. The sprint-training sessions were
performed at the same time and day for each training group throughout the intervention peri-
od and no regular soccer training sessions were performed on the same day as the sprint train-
ing took place. Athletes in 100UNSUP performed 15x20 m maximal sprints with starts each
60 s once a week. Groups 90SUP and 90UNSUP performed one weekly training session con-
sisting of a larger dose of 30x20 m sprints at 90% of maximal sprint velocity (based on the best
20-m sprint time obtained during the pre-training test) with starts each 60 s.
Two sprint-training experts, with extensive national-level coaching experience, supervised
the 90SUP group during the intervention. Three key sprint-technical elements and correspond-
ing verbal instructions were emphasized during the training sessions:
• Optimal upper-body angle relative to the ground during the initial steps in order to create
higher horizontal propulsive forces through more effective utilization of hip and knee exten-
sors [16,17]. The athletes were instructed to assume a start position with forward lean in the
upper body and a lowered centre of gravity and to gradually become more upright through-
out the acceleration.
• Minimize horizontal braking forces [16]: Athletes with apparently too high braking forces
were encouraged to assume a more favourable configuration at the point of ground contact
with the foot plant closer to the perpendicular line from the centre of mass. This can be
achieved by hitting the ground with a bent knee (relevant during acceleration) or with the
centre of mass at a large vertical distance above the ground (relevant during maximal
sprinting).
• Produce a stiff rebound during ground contact in order to minimize degeneration of hori-
zontal propulsive forces [18–20]: Identified “heal runners” were encouraged to pre- activate
dorsiflexion muscles prior to foot plant and stiffen the ankle joint during ground contact, al-
lowing them to utilize the elasticity in the plantar flexors for greater force development.
These instructions were emphasized during the warm up drills.
After video analysis of the first training session, the two sprint-training experts prepared an
individual capacity profile for all participants in the 90SUP group. Each athlete was presented
with one technical task at a time, in accordance with general feedback principles [9]. Players
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
5 / 13
with obvious technical limitations were provided with more verbal instructions than technical-
ly well-performing athletes.
In the absence of previously published studies, a 1:2 repetition ratio between 100% and 90%
sprinting was chosen. Several measurements were assessed in order to compare the two repeat-
ed-sprint training sessions used. Firstly, session rated perceived exertion (RPE) was recorded
for all athletes after the repeated sprints performed in pre-training testing and the first training
session. Written and verbal instructions regarding its use were provided in advance [21]. More-
over, heart rate was measured continuously during the first training session for all athletes who
ran at 90% sprint intensity, in addition to BLa- immediately after their last sprint. These were
compared to corresponding data assessed during pre-training tests. Mean SL and SR for the
first sprint-training session were calculated by identical procedures as for the pre- and post-
training tests. Finally, all training group athletes performed 3x20 m maximal sprints with starts
each 60 s 48 hours after the first training session for a performance recovery check. The mean
time for these three sprints was compared with corresponding sprints from the pre-
training test.
Electronic timing was continuously used to control running speed and adjust intensity ac-
cording to each player’s “target time”. Target time for the 90SUP and 90UNSUP participants
were derived from the best single-sprint time achieved during preliminary testing by multiply-
ing mean velocity over the 20-m distance by 0.9. No feedback other than sprint time informa-
tion (for intensity control purposes) was provided by a timekeeper for the 90UNSUP and
100UNSUP groups after each run. Fig. 1 shows intensity distribution for the two 90% groups
(90SUP and 90UNSUP pooled together) during all training sessions. More than 90% of all
sprints were completed with intensities between 87 and 93% of maximal sprint velocity. All
sprints for 100UNSUP during the training sessions were completed with an intensity > 97%
(mean ± SD: 98.2 ± 0.8%) when related to the best single-sprint within each training session.
Thus, treatment conditions in 90SUP and 90UNSUP were strictly separated and did not over-
lap with 100UNSUP. For simplicity, we continue to use the terms “100UNSUP” or “maximal
intensity”.
Statistical analysis
All statistical analyses were carried out using SPSS 17.0 for Windows (SPSS Inc., Chicago, IL,
USA). Level of significance was set to p<0.05. The General Linear Model with Repeated Mea-
sures followed by Bonferroni adjustment for multiple comparisons was used to examine re-
peated-sprint performance development (mean sprint time) for 100UNSUP across tests and
training sessions. The same model was used for 90SUP and 90UNSUP (both groups pooled
together) to compare effort-related variables in maximal and sub-maximal sprinting. Analy-
sis of covariance (ANCOVA) adjusting for the pre-training test value and randomization
stratification factor (club) was used to examine within-group and between-group mean
changes. The differences were judged by using estimated marginal means (EMM).
Bonferroni corrections were used to adjust p-values for multiple testing. Effect magnitudes
were calculated and interpreted categorically according to the guidelines by Hopkins et al.
[22]. The first 6 sprints from the pre-training test (for all included participants) were used to
calculate typical variation for sprint time, SL and SR. Effect size of the within-group changes
for mean sprint time were based on mean change and typical variation. The results are ex-
pressed as mean ±SD and 95% confidence intervals (95% CIs) were calculated for
all measures.
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
6 / 13
Results
Table 2 shows effort-related variables for the two repeated-sprint training sessions used in the
present intervention. No differences in RPE were observed between the sessions. Mean sprint
time for the 3x20-m sprints performed 48 hours after the first training session was not signifi-
cantly different when compared to the corresponding pre-training sprint test. Sprinting at 90%
velocity was accompanied with reduced HR peak (17%; very large effect; p<0.001), BLa- (55%;
large effect; p<0.001) and SR (11%; very large effect; p<0.001) compared to maximal sprinting.
Fig 1. Intensity distribution for the sprint training groups during all training sessions. Best sprint from
pre-training testing was set as reference (100%) for 90SUP and 90UNSUP, while best sprint within each
training session was set as reference (100%) for 100UNSUP.
doi:10.1371/journal.pone.0121827.g001
Table 2. Effort-related variables in maximal (100%) and sub-maximal (90%) sprinting.
Sprint session
15x20m (100% intensity)
30x20m (90% intensity)
Δ sprint time 48 h (s)
0.00 ±0.02
0.01 ±0.02
Session RPE
3.8 ±1.2
4.0 ± 1.1
HR peak (beats min-1)
170 ±10
141 ±10*
BLa- (mmolL-1)
4.4 ±1.8
2.0 ±0.7*
SL (m)
1.55 ±0.08
1.56 ±0.09
SR (stepss-1)
4.36 ±0.18
3.87 ±0.22*
Δ sprint time 48 h = sprint time 48 hours after the first training session minus corresponding pre-training
sprint test time (mean of first 3 sprints for each time point), RPE = rated perceived exertion, HR peak = peak
heart rate, BLa- = blood lactate concentration, SL = step length, SR = step rate
* = significantly different from 100% sprinting (p<0.001).
doi:10.1371/journal.pone.0121827.t002
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
7 / 13
No significant within-group differences for the analyzed performance parameters were ob-
served, except that 90SUP improved Yo-Yo IR1 performance from pre- to post-training (258
m; 17,3%; p<0.01). No significant between-group differences were observed (p<0.05). 90SUP
improved Yo-Yo IR1 performance by a moderate margin compared to all other groups, while
all other between-group differences were small or trivial (Table 3 and Fig. 2).
Achievement of best sprint performance was randomly distributed across the 15 sprints in
all groups during both pre- and post-training tests. Typical variation for sprint time, SL and SR
was 0.025 s (CV 1.0%), 0.028 m (CV 1.8%) and 0.08 stridess-1 (CV 1.9%), respectively. In
CON, a variation in mean sprint time of ±0.04 s was observed between the pre- and post-train-
ing tests. Corresponding variation for SL and SF was 0.06 m and 0.19 stridess-1, respectively.
In 100UNSUP, significant differences from pre- to post-training tests were observed for
BLa- (1.5 mmolL-1; 35.7%; p<0.001), SL (-0.04 m; 2.6%; p = 0.020) and SF (0.13 stepss-1;
3.0%; p = 0.019). BLa- increased significantly in 100UNSUP compared to CON from pre- to
post-training (p = 0.008) (Table 4). No other within- or between-group differences were
Table 3. Between-group changes (mean and 95% CIs) versus controls in physical performance from pre- to post-training.
Intervention group
Best sprint time (s)
Mean sprint time (s)
CMJ (cm)
Yo-Yo IR1 (m)
100UNSUP
-0.03 (-0.07 to 0.00)
-0.03 (-0.06 to 0.01)
1.0 (-0.6 to 2.6)
-34 (-272 to 205)
90UNSUP
-0.03 (-0.07 to 0.01)
-0.02 (-0.06 to 0.02)
0.4 (-1.3 to 2.1)
-1 (-120 to 117)
90SUP
-0.02 (-0.06 to 0.02)
-0.03 (-0.07 to 0.01)
1.8 (0.0 to 3.6)
131 (-108 to 369)
The differences vs. control group are assessed by estimated marginal mean. Minus (-) indicates lower values post-training compared with the control
group (assessed by estimated marginal means). CMJ = countermovement jump, Yo-Yo IR1 = Yo-Yo intermittent recovery level 1. No significant between-
group differences were observed.
doi:10.1371/journal.pone.0121827.t003
Fig 2. Individual changes in 15x20 m mean sprint time from pre- to post-training tests.
doi:10.1371/journal.pone.0121827.g002
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
8 / 13
observed. The change in BLa- within 100UNSUP was moderate while the other effect magni-
tudes between- or within-groups were trivial or small.
Fig. 3 shows the development of repeated-sprint performance (mean sprint time) for
100UNSUP during the intervention period, including pre- and post-training tests. Weekly
changes in group mean values up to 0.05 s were observed.
Discussion
In the present study, weekly repeated-sprint training sessions performed at maximal or with
90% intensity were not sufficient to improve soccer-specific physical performance in junior
soccer players, when compared to a matched control group assumed to maintain a constant
training pattern. Moreover, no differences in performance outcomes were observed between
supervised and unsupervised sprint-training groups training at 90% maximal sprinting veloci-
ty. Apparently, the relative work loads elicited by the current intervention strategies were not
sufficient to create appropriate adaptations during the early pre-season soccer period.
To the authors’ knowledge, this is the first study to compare the effects of sprint training at
90 vs. 100% sprint intensity or supervised vs. unsupervised sprint training. Our findings con-
firm the assumption that sprint performance is resistant to training enhancement, even among
junior soccer players during the early pre-season period where total training load is reduced.
Since treatment allocation was adjusted for club participation, the current results were not
influenced by varying overall soccer conditioning across groups (Table 1). Age distribution was
consistent across groups (Table 1) and body mass did not change significantly in any of the
groups (Table 4). The moderate group sample sizes may mask possible significant outcomes.
However, based on the trivial to moderate effect magnitudes, our findings do not support a rec-
ommendation to perform the present training regimes under otherwise identical conditions.
Fig 3. 95% confidence intervals of mean sprint time for 100UNSUP during the intervention.
doi:10.1371/journal.pone.0121827.g003
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
9 / 13
Despite the absence of significant differences in the experimental training interventions, the
present study may outline directions for future related studies.
Effort matched sprint training
The two training sessions used were equally rated in terms of session RPE (Table 2). Further-
more, recovery status after two days was not different for the maximal and sub-maximal train-
ing groups. Based on these observations, we find it reasonable to conclude that the two
repeated-sprint training sessions were effort matched. Blood lactate values obtained after re-
peated sprints at 90% intensity were below what has been considered “lactate threshold intensi-
ty” (2.5–4.0 mmolL-1) in endurance training [23]. In contrast, repeated sprinting at maximal
intensity was accompanied with BLa- at or above the typical lactate threshold range described
for endurance athletes. Even though BLa- values obtained from sprint and endurance training
are not directly comparable, the present data suggest metabolic pathway partitioning differ-
ences between 90% and maximal sprinting. Small but meaningful differences in muscle recruit-
ment cannot be excluded as contributory to the observed increase in lactate accumulation at
maximal sprinting.
Effects of training at maximal and sub-maximal intensity
The present results revealed only trivial and non-significant changes in soccer-related sprinting
from pre- to post-training for 100UNSUP (Table 3, Fig. 2). Previously, Tønnessen et al. [24]
observed unaltered sprint velocity over 0–20 m sprint and improved velocity over 20–40 m as a
result of weekly repeated 40-m sprints at maximal or near maximal intensity. This suggests
that players are more disposed to adaptations over somewhat longer but less soccer-specific
sprint distances. Soccer players perform a high number of brief accelerations during training
and games [14]. Thus, one could argue that most players have likely maximized their 0–20 m
sprint (acceleration) potential during regular soccer conditioning. While sprint performance
remained unchanged in 100UNSUP (Table 3), SL and SR changed significantly from pre- to
post-training (Table 4). These changes were greater than the observed typical variation. Our
findings are somewhat in contrast to previously published studies stating that individual
achievement of sprint velocity corresponds to an optimal self-selected step length/step rate
ratio [25] and that a different ratio will produce a lower velocity, so-called negative interaction
[26]. In the present study, maximal sprint training induced a significant shift in the step
length/frequency relationship “selected” by the athletes, with step frequency increasing 3%
from 4.33 to 4.46 steps.s-1 and step length declining correspondingly. These changes cannot be
ascribed to supervision as this group did not receive sprint technique feedback or instruction.
Moreover, 100UNSUP demonstrated an increase in BLa- after 15x20 m maximal sprinting in
the post-training test, suggesting possible changes in anaerobic energy release, buffering
Table 4. Between group changes (mean and 95% CIs) versus controls for underlying performance variables between pre- and post-training.
Intervention group
Body mass (kg)
HRpeak (beatsmin-1)
BLa- (mmolL-1)
SL (m)
SR (stepss-1)
100UNSUP
0.3 (-0.8 to 1.5)
5 (-1 to 12)
1.9 (0.7 to 3.2)*
0.00 (-0.07 to 0.06)
0.06 (-0.13 to 0.25)
90UNSUP
-0.3 (-1.4 to 0.8)
2 (-5 to 8)
1.1 (-0.1 to 2.3)
0.04 (-0.02 to 0.10)
-0.09 (-0.28 to 0.10)
90SUP
-0.3 (-1.5 to 0.9)
4 (-3 to 11)
1.5 (0.2 to 2.9)
0.03 (-0.03 to 0.10)
-0.04 (-0.24 to 0.17)
The differences vs. control group are assessed by estimated marginal mean. Minus (-) indicates lower values post-training compared with the control
group (assessed by estimated marginal means). HR = heart rate, BLa- = blood lactate concentration, SL = step length, SR = step rate
* = significantly different (Bonferroni adjusted) from CON (p = 0.01).
doi:10.1371/journal.pone.0121827.t004
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
10 / 13
characteristics or muscle recruitment pattern. However, these possible physiological changes
were not accompanied with enhanced performance.
Based on both current and previous findings [13], it cannot be concluded that weekly train-
ing at 90% velocity is a sufficient sprinting intensity for stimulating adaptation over short
sprint distances. Blood lactate and peak heart rate values observed in the present junior soccer
players indicate relatively low metabolic stress (Table 2). It is possible that sub-maximal sprint
training is more appropriate for typical competitive athletics sprinting distances (100–200 m)
compared to 0–20 m accelerations. 20-m sprints are comprised of high to maximal acceleration
from a resting state and continuing through the timed distance. In this condition, energy de-
mands during the acceleration phase greatly exceed those at peak velocity [17]. The change in
kinetic energy (½mv2) is proportional to the square of the change in velocity, such that the
90% sprint condition is associated with a nearly 20% reduction in kinetic energy change (and
presumably, muscular energetic demand) compared to maximal sprinting velocity. Due to this
non-linearity, a 5% reduction in short sprint velocity during repeated-sprint training over
short distances would correspond to 90% workloads in strength training and endurance train-
ing and might give a more optimal balance of stress, injury risk reduction and adaptive signal
retention. This possibility remains to be explored.
Effects of supervised training
The present study revealed no significant training effects when supervised and unsupervised
sprint training at 90% sprint velocity were compared (Table 3 and Fig. 2). However, the 90SUP
group improved Yo-Yo IR1 performance by a moderate margin compared to the other groups.
Since Haugen et al. [13] reported unchanged VO2 max after seven weeks of repeated-sprint
training at 90% intensity, it is reasonable to assume that locomotion efficiency during high-in-
tensity running has improved in 90SUP. The lack of effects on maximal and repeated-sprint
performance may have been affected by the possibility that sprint training at 90% sprint speed
is below the lowest effective sprinting intensity for stimulating adaptation. Future studies
should therefore explore the effect of supervised training with a gradual increase in intensity
from sub-maximal to maximal sprint velocity. Another argument for such a gradual increase
in velocity also becomes relevant if one assumes that the athletes gradually enhance sprint per-
formance over the training period. We chose not to control and adjust for possible sprint per-
formance enhancement in 90SUP and 90UNSUP throughout the present intervention period
to avoid a mix of different treatment conditions.
Theoretically, the lack of effects with supervised sprint training may be due to poor coaching
quality such that the athletes were not able to translate the instructions into practice. However,
both training experts used in the present study had many years of experience coaching athletics
performers on both national and international levels. Mazzetti et al. [7] and Coutts et al. [8]
showed that the presence of a training expert was beneficial for maximal strength and power
development over time. In contrast to the present study, the training experts in those studies
were allowed to adjust the total training load during the interventions. Based on these observa-
tions, one could argue that the effect of expert supervision during training is optimized when
combined with greater flexibility in the day-to-day training prescription.
Training-related constraints
Common challenges in applied studies of this nature are related to constraints with overall
team conditioning [1,13,27]. Current analyses confirm these constraints, even though the
study was conducted early pre-season where total training load is typically reduced. In CON,
we observed ±0.04 s absolute individual variation in mean sprint time between pre- and post-
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
11 / 13
training tests. More importantly, weekly changes in group mean values up to 0.05 s (nearly 2%)
were observed in 100UNSUP (Fig. 3), despite consistent frequency and volume of games and
training sessions during the intervention period (Table 1). This weekly or seasonal variation is
considerably higher than the observed typical variability. The present findings emphasize the
need for more detailed information about overall conditioning load, accepting that soccer-spe-
cific movements (i.e. brief accelerations, high sprinting velocities or changes of directions) are
impossible to assess accurately in groups of players with current technology [28,29]. In princi-
ple, the present study could have been accomplished in a more controlled experimental envi-
ronment, omitting the concurrent soccer training. However, such an approach severely limits
the external validity. If improvement of sprinting performance is the primary goal for certain
players, future studies should explore the effects of more frequent sprint-training sessions and
longer intervention periods, perhaps in combination with other training forms. Future studies
should also explore whether it is more effective to structure the players’ weekly soccer training
rather than introducing an additional physical conditioning regime. A theoretically perfect
conditioning program for certain capabilities may limit other important qualities and vice
versa. Coaches and conditioning experts have to balance their training methods and exercises
in order to optimize different skills in relation to their contribution to overall soccer perfor-
mance. Based on the varying individual responses to each of the performed treatments (Fig. 2)
and the absence of evidence supporting the choice of specific training methods at the group
level, it is essential to diagnose each individual and develop training interventions that target
their key physiological and technical weaknesses.
Author Contributions
Conceived and designed the experiments: TH ET ØØ FLH EE SS. Performed the experiments:
TH ET ØØ FLH EE GP. Analyzed the data: TH ET ØØ FLH GP SS. Contributed reagents/ma-
terials/analysis tools: TH ET GP EE SS. Wrote the paper: TH ET ØØ FLH GP EE SS.
References
1.
Haugen T, Tønnessen E, Hisdal J, Seiler S. The role and development of sprinting speed in soccer. Int
J Sports Physiol Perform. 2013; 9: 432–441.
2.
Faude O, Koch T, Meyer T. Straight sprinting is the most frequent action in goal situations in profession-
al soccer. J Sports Sci. 2012; 30: 625–631. doi: 10.1080/02640414.2012.665940 PMID: 22394328
3.
Haugen T, Tønnessen E, Seiler S. Speed and countermovement-jump characteristics of elite female
soccer players, 1995–2010. Int J Sport Physiol Perform. 2012; 7: 340–349. PMID: 22645175
4.
Haugen T, Tønnessen E, Seiler S. Anaerobic performance testing of professional soccer players
1995–2010. Int J Sport Physiol Perform. 2013; 8: 148–156. PMID: 22868347
5.
Haugen T, Tønnessen E, Seiler S. 9.58 and 10.49: Nearing the citius end for 100 m? Int J Sports Phy-
siol Perform. 2015; 10: 269–272. doi: 10.1123/ijspp.2014-0350 PMID: 25229725
6.
Miller J. Burst of speed. South Bend (IN), Icarus Press; 1984.
7.
Mazzetti SA, Kraemer WJ, Volek JS, Duncan ND, Ratamess NA, Gomez AL, et al. The influence of di-
rect supervision of resistance training on strength performance. Med Sci Sports Exerc. 2000; 32:
1175–1184. PMID: 10862549
8.
Coutts AJ, Murphy AJ, Dascombe BJ. Effect of direct supervision of a strength coach on measures of
muscular strength and power in young rugby league players. J Strength Cond Res. 2004; 18: 316–323.
PMID: 15142000
9.
Schmidt RA, Wrisberg CA. Motor learning and performance: A situation based learning approach. 4th
edition. Human Kinetics; 2008.
10.
Kraemer WJ, Adams K, Cafarelli E, Dudley GA, Dooly C, Feigenbaum MS, et al. Progression models in
resistance training for healthy adults. Med Sci Sports Exerc. 2002; 34: 364–380. PMID: 11828249
11.
Seiler S, Jøranson K, Olesen BV, Hetlelid KJ. Adaptations to aerobic interval training: interactive effects
of exercise intensity and total work duration. Scand J Med Sci Sports 2013; 23: 74–83. doi: 10.1111/j.
1600-0838.2011.01351.x PMID: 21812820
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
12 / 13
12.
Vittori C. The European school in sprint training: The experiences in Italy. New Studies in Athletics
1996; 11: 85–92.
13.
Haugen T, Tønnessen E, Leirstein S, Hem E, Seiler S. Not quite so fast: Effect of training at 90% sprint
speed on maximal and repeated sprint ability in soccer players. J Sports Sci. 2014; 32: 1979–1986.
PMID: 25385308
14.
Di Salvo V, Pigozzi F, González-Haro C, Laughlin MS, De Witt JK. Match performance comparison in
top English soccer leagues. Int J Sports Med. 2013; 34: 526–532. doi: 10.1055/s-0032-1327660 PMID:
23184481
15.
Krustrup P, Mohr M, Amstrup T, Rysgaard T, Johansen J, Steensberg A, et al. The yo-yo intermittent re-
covery test: physiological response, reliability and validity. Med Sci Sports Exerc. 2003; 35: 697–705.
PMID: 12673156
16.
Harland MJ, Steele JR. Biomechanics of the sprint start. Sports Med. 1997; 23: 11–20. PMID: 9017856
17.
di Prampero PE, Fusi S, Sepulcri L, Morin JB, Belli A, Antonutto G. Sprint running: a new energetic ap-
proach. J Exp Biol. 2005; 208: 2809–2816. PMID: 16000549
18.
Chelly SM, Denis C. Leg power and hopping stiffness: relationship with sprint running performance.
Med Sci Sports Exerc. 2001; 33: 326–333. PMID: 11224825
19.
Kuitunen S, Komi PV, Kyröläinen H. Knee and ankle joint stiffness in sprint running. Med Sci Sports
Exerc. 2002; 34: 166–173. PMID: 11782663
20.
Girard O, Micallef JP, Millet GP. Changes in spring-mass model characteristics during repeated running
sprints. Eur J Appl Physiol. 2011; 111: 125–134. doi: 10.1007/s00421-010-1638-9 PMID: 20824280
21.
Foster C. Monitoring training in athletes with reference to overtraining syndrome. Med Sci Sports
Exerc. 1998; 30: 1164–1168. PMID: 9662690
22.
Hopkins WG, Marshall SW, Batterham AM, Hanin J. Progressive statistics for studies in sports medi-
cine and exercise science. Med Sci Sports Exerc. 2009; 41: 3–13. doi: 10.1249/MSS.
0b013e31818cb278 PMID: 19092709
23.
Tokmakidis SP, Léger LA, Pilianidis TC. Failure to obtain a unique threshold on the blood lactate con-
centration curve during exercise. Eur J Appl Physiol Occup Physiol. 1998; 77: 333–342. PMID:
9562362
24.
Tønnessen E, Shalfawi S, Haugen T, Enoksen E. The effect of 40-m repeated sprint training on maxi-
mum sprinting speed, repeated sprint endurance, vertical jump and aerobic capacity in young elite
male soccer players. J Strength Cond Res. 2011; 25: 2364–2370. doi: 10.1519/JSC.
0b013e3182023a65 PMID: 21869624
25.
Salo AI, Bezodis IN, Batterham AM, Kerwin DG. Elite sprinting: are athletes individually step-frequency
or step-length reliant? Med Sci Sports Exerc. 2011; 43: 1055–1062. doi: 10.1249/MSS.
0b013e318201f6f8 PMID: 20980924
26.
Hunter JP, Marshall RN, McNair PJ. Interaction of step length and step rate during sprint running. Med
Sci Sports Exerc. 2004; 36: 261–271. PMID: 14767249
27.
Soares-Caldeira LF, de Souza EA, de Freitas VH, de Moraes SM, Leicht AS, Nakamura FY. Effects of
additional repeated sprint training during preseason on performance, heart rate variability, and stress
symptoms in futsal players: a randomized controlled trial. J Strength Cond Res. 2014; 28: 2815–2826.
doi: 10.1519/JSC.0000000000000461 PMID: 24662230
28.
Castellano J, Alvarez-Pastor D, Bradley PS. Evaluation of research using computerised tracking sys-
tems (Amisco and Prozone) to analyse physical performance in elite soccer: a systematic review.
Sports Med. 2014; 44: 701–712. doi: 10.1007/s40279-014-0144-3 PMID: 24510701
29.
Cummins C, Orr R, O'Connor H, West C. Global positioning systems (GPS) and microtechnology sen-
sors in team sports: a systematic review. Sports Med. 2013; 43: 1025–1042. doi: 10.1007/s40279-013-
0069-2 PMID: 23812857
Repeated-Sprint Training in Soccer Players
PLOS ONE | DOI:10.1371/journal.pone.0121827
March 23, 2015
13 / 13
| Sprint conditioning of junior soccer players: effects of training intensity and technique supervision. | 03-23-2015 | Haugen, Thomas,Tønnessen, Espen,Øksenholt, Øyvind,Haugen, Fredrik Lie,Paulsen, Gøran,Enoksen, Eystein,Seiler, Stephen | eng |
PMC7538888 | ARTICLE
Human running performance from real-world
big data
Thorsten Emig
1✉ & Jussi Peltonen
2
Wearable exercise trackers provide data that encode information on individual running
performance. These data hold great potential for enhancing our understanding of the complex
interplay between training and performance. Here we demonstrate feasibility of this idea
by applying a previously validated mathematical model to real-world running activities
of ≈ 14,000 individuals with ≈ 1.6 million exercise sessions containing duration and distance,
with a total distance of ≈ 20 million km. Our model depends on two performance parameters:
an aerobic power index and an endurance index. Inclusion of endurance, which describes the
decline in sustainable power over duration, offers novel insights into performance: a highly
accurate race time prediction and the identification of key parameters such as the lactate
threshold, commonly used in exercise physiology. Correlations between performance indices
and training volume and intensity are quantified, pointing to an optimal training. Our findings
hint at new ways to quantify and predict athletic performance under real-world conditions.
https://doi.org/10.1038/s41467-020-18737-6
OPEN
1 Université Paris-Saclay, CNRS, Laboratoire de Physique Théorique et Modèles Statistiques, 91405 Orsay, France. 2 Polar Electro Oy, Professorintie 5, 90440
Kempele, Finland. ✉email: [email protected]
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
1
1234567890():,;
S
keletal evidence suggests that endurance running may have
evolved 2 million years ago1. It probably originated as a
hunting skill but has later developed to competition, dating
back to ancient Olympic Games ~720 BC2 and exercise form for
mass population. Over the years, endurance running has under-
gone substantially change. Recent decades have witnessed an ever
growing exercising population which uses wearable sensors to
bring together astonishing volumes of data for speed, distance,
heart rate, accelerations, and more3–5. For example, endurance
athletes like runners and cyclists currently upload from GPS
enabled sensors more than a billion activities per year world-
wide6. In principle, these data provide an exciting opportunity to
monitor human physiology noninvasively under real-world
conditions outside the laboratory. Measuring the physiological
response to physical activity can provide important insights for a
variety of populations ranging from elite athletes to recreational
exercisers to patients in rehabilitation7,8. However, the analysis of
big data sets of large, heterogeneous groups of individuals poses a
substantial challenge due to the quality of the data itself9,10, lack
of effective theoretical models11, and influence of environmental
factors like weather conditions12,13. The important, robust
properties of an individual’s physiology can be overshadowed by
details specific to the conditions of recording. Thus, there is a
demand for universal theoretical models that have been validated
for noise-free exercise data and can be applied under noisy real-
world conditions to derive meaningful physiological and perfor-
mance information14.
To date, exercise physiologists conventionally use laboratory
testing to determine parameters that measure fitness and per-
formance potential15. A strength of laboratory testing is that it
can distinguish between cardiovascular limit, maximal rate of
oxygen consumption (VO2max), neuromuscular effects, and run-
ning economy16,17. Together VO2max and running economy
determine maximal aerobic speed, which is the slowest speed at
which VO2max occurs. Maximal aerobic speed correlates with race
speed on shorter distances but alone cannot predict race times for
longer distances such as the marathon. Exercise thresholds have
been used in exercise testing to quantify metabolism. However,
the determination of such thresholds, like the lactate threshold, in
the laboratory is somewhat limited. Typical laboratory testing is
short-lasting and does not always fully capture time and distance
dependent reduction in running economy18,19. For example, only
sparse results exist for the endurance limited fractional utilization
of maximal aerobic power (MAP) and its dependence on exercise
duration20. Moreover laboratory testing is expensive and not
available to most of the population. The undeniable fact that the
best test of running performance is an actual race and not
laboratory tests highlights the need for models specifically con-
structed to extract performance indices of an athlete from their
regular exercise performance. For these reasons, models that can
utilize data from wearable devices and turn those into meaningful
performance parameters may offer a cost effective alternative
approach to laboratory testing. However, it must stressed that this
type of approach does not elucidate the physiological and bio-
mechanical mechanisms that control performance. It is an
adjunct to the methods which are already used, providing addi-
tional insight into running and the potential training factors
influencing performance and it does not replace the insights that
we can gain from laboratory testing.
Several empirical and physiological models have been put
forward for explaining running world records in terms of a few
physiological parameters. The noted physiologist Hill empirically
proposed a hyperbola to describe the maximal power output as a
function of exercise duration21. Also a purely mechanical
approach, based on the runners equation of motion, has been
proposed22. These approaches predict that the average racing
velocity tends to be a constant value with increasing race distance
which contradicts observation. While more recent approaches
have combined physiology and observations to propose more
realistic logarithmic relations between maximal power output and
duration23, these models depend on many parameters that vary
among individuals24. Recently we have developed a universal
running model which builds on concepts in exercise physiology,
depends only a minimal set of key performance indices that are
required to predict race performance, contains no additional
individual-dependent quantities and has been validated with
running world-records14. Here, we show that it is also possible to
obtain novel insights into individual’s running performance by
applying this model to big exercise datasets.
Exercise data are a valuable source of information about
individual long-term training protocols. Endurance training leads
to a wide spectrum of physiological responses. However, in
practice, training is prescribed often only by anecdotal evidence
and personal experience. This might be due to a lack of knowl-
edge of statistically significant correlations between the relevant
physiological parameters and training characteristics for large
groups of individuals with different fitness status. Here, we
demonstrate the feasibility to extract key performance indices
from real-world running exercise data recorded with wearable
exercise trackers. We apply our method to runners during their
training season before a marathon race. Our universal running
model characterizes a runner’s performance with two indices that
measure (1) endurance (endurance index) and (2) the velocity
requiring MAP output (aerobic power index). The main aim of
our work is to demonstrate the feasibility of extracting perfor-
mance indices from real-world racing results in a big population
of runners and to use these indices to predict accurate race times
and evaluate the effect and efficiency of training. Our approach
represents a potentially powerful platform to enlarge dramatically
the number of tested subjects in sports science by extending
performance index acquisition from conventional laboratory
testing to real-world conditions with the aid of mathematical
modeling and wearable technology.
Results
Universal performance model. In previous work we have
developed a model that can be used to extract aerobic perfor-
mance indices from race data14. To summarize, this model
expresses exercise intensity on a relative power scale p, which
varies between zero, corresponding to basal metabolic rate, and
unity at MAP generation. MAP is expected to correspond to
maximal oxygen uptake VO2,max but this analogy needs not to be
assumed in our approach. A linear relation p(v) maps running
velocity v to relative power with p(vm) = 1 defining vm as
an aerobic power index associated with MAP beyond which
anaerobic energy supply can yield p > 1 for a short time only.
Anaerobic supply contributes to maximal exercise shorter than a
crossover time tc which in our model is the longest time over
which MAP can be sustained. An important prediction of our
model is that the maximal value of the relative power p that a
runner can maintain declines logarithmically with duration, with
a rate γl, assuming that the durations are longer than tc. This
finding is in agreement with a finding of A.V. Hill who observed
this form of decline in running world records21. For more details
on our model, see the “Methods” section. Here, we use this
universal, i.e., subject independent model for human running
performance, to extract aerobic performance indices from fin-
ishing times of runners worldwide by matching them with model
predictions14. The analyzed data set comes from an exercise
tracking platform that contains precise records of distance and
duration (and hence average velocity) of running activities
ARTICLE
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
2
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
of ≈19K individuals, who ran a total distance of 32M km over a
period of 3.5 years. The data were recorded by the individuals
with a GPS digital sports watch (V800, Polar Electro Oy, Oulu,
Finland)25, and uploaded to the platform. Maximal performance
of an individual was measured by the fastest finishing time for the
four most common racing distances 5000 m, 10,000 m, half-
marathon (21,097.5 m) and marathon (42,195 m) within a racing
season, which is defined as the 180 days preceding the marathon
race (see “Methods” section for detection of racing activities).
The velocity corresponding to our parameter vm is difficult to
measure in laboratory settings since VO2,max can be achieved
over a wide range of sub-maximal intensities because of an
upward drift of oxygen uptake with exercise duration18,19. In
general, our model can determine vm from the crossover of the
race–time–distance relation at time tc, and hence is free from
this complications. The simplest version of the model assumes a
fixed time tc. Model predictions for sub-MAP performances do
not depend on this fixed time since other choices lead only to
consistently renormalized values for vm and γl (which are then
no longer associated strictly with MAP but with a slightly
different power). In agreement with the application of our
model to running records on both the super- and sub-MAP
branches14 and laboratory testings26, we choose tc = 6 min in
the following. Combining running economy and the decline of
the fractional utilization of maximal power output with race
duration, the fastest time T(d) over a distance d is given by the
universal expression
TðdÞ ¼ tc
γl
d
dc
1
W1 d
dc
expð1=γlÞ
γl
h
i
for d ≥ dc ;
ð1Þ
where we defined dc = vmtc, and W−1 is a real branch of
the Lambert W-function which is defined as the multi-valued
inverse of the function w ! w expðwÞ27. W−1(z) is real valued
for −1/e ≤ z < 0 which is fulfilled for all distances d that we
consider (see the “Methods” section for more detail). Note that
T(dc) = tc, i.e., dc is the distance that can be maximally raced in
the time tc. The condition d ≥ dc is always satisfied for the race
distances considered here. We note that Eq. (1) is an exact
solution of our model. It can be also obtained from earlier
descriptions of the energetics of endurance running28–30 when
the fractional utilization of MAP is described by our prediction
of a slow, logarithmic decay, and a linear increase of the energy
cost of running with velocity is assumed.
The model parameters, called performance indices, quantify
different aspects of performance and provide a unique insight
into basic determinants of fitness in a large population of
runners over a wide range of exercise capacities and over long
time scales. The velocity vm measures combined running
economy and MAP and is known to be a better predictor of
performance than VO2,max alone31. We define the endurance
index as El ¼ expð0:1=γlÞ, which encodes that 90% of vm can
be maintained for an extended time Eltc > tc. The pair of
performance indices vm, El is sufficient to account for racing
velocity variations for distances from dc (typically one mile in
our data set) to the marathon. For example, when analyzing
consistent running records of individuals, we found strong
evidence that they follow the same universal scaling law of
Eq. (1) as running world (or national) records do, with mean
errors below 1%14. Here, our model estimates are based on an
individual’s fastest times for the four fixed racing distances, 5 k,
10 k, half-marathon, and marathon. Unfortunately, we cannot
determine from the available data set if performance was
achieved during an actual racing event. For our approach
however, it is only required that the recorded performance
corresponds to the maximal effort over a given running distance
achieved during the racing season.
Exercise data. An overview of the data analysis design is provided
in Fig. 1. All available subjects and activities in the data set of the
exercise tracking platform were grouped by SID and marathon
date, combining all individual running activities during the
180 days before the marathon, defining a season. For each season,
activities with the fastest time for the four fixed race distances
defined a racing season. We imposed the condition that each
racing season contains at least two races. If a season contained 30
or more total running activities they were defined as training
season. For consistency certain data filters were applied to all
activities and races (see the “Methods” section for more detail).
Two variants of racing season were defined, with the marathon
included and excluded. A total of ~25,000 racing seasons with
the marathon included and ~10,000 racing seasons without the
marathon, and ~22,000 training seasons were analyzed (see
Table 1 for a summary of the available data and performed
analyses).
Accuracy of performance prediction. For all individuals, we
estimated their performance indices vm and γl for each racing
season by matching race events to Eq. (1) by minimizing the
relative prediction error for the race times. The probability den-
sities of these indices are shown in Fig. 2. For all racing seasons
with three and more races (N = 12,309), the mean error between
model prediction and actual race time was only 2.0%. This sug-
gests that our model captures correctly determinants of aerobic
endurance
performance.
Correlations
between
performance
indices and marathon finishing times are presented in Fig. 3. To
investigate the predictive power of our model in more detail, we
applied our model also to the racing season with the marathon
performance excluded (see Fig. 4). This allowed us to estimate
the marathon finishing time from the performances on shorter
distances only. As a function of performance indices, in the most
likely parameter range the model predicted the marathon per-
formance with an overall accuracy of better than 10%. Only for
very small (or large) endurance El, estimated times tended to be
too slow (or fast) which indicates that sub-marathon distances
were raced inconsistently, leading to an under (or over) estima-
tion of El. Given all the possible uncertainties in marathon racing
that are beyond the control of this study (e.g., weather, course
profile, and motivation of the athlete), our predictions for the
marathon finishing times are rather satisfying.
Maximal velocity for 1 h. Analysis of ~25,000 racing seasons
reveals a normally distributed velocity vm and an exponential decay
of the probability density for the endurance El (see Fig. 2). Inter-
estingly, VO2,max in a study on 450 elite soccer players has also been
found to obey a normal distribution32. Note that vm also measures
running economy, which varies considerably among individuals
and modulates performance24. In exercise physiology, the ability of
a runner to maintain a certain effort is often characterized in terms
of thresholds, of which a common example is lactate threshold. In
our approach, however, there is a continuous relationship between
power output and velocity, and the change of this relation with
duration appears to be a natural measure for endurance capability.
Hence, as a practical measure for endurance, we define in our
model the velocity v1hU ¼ vm½1 0:1 log ð60 min =tcÞ=log ðElÞ
that a runner can maintain for 1 h, corresponding to the maximal
fractional utilization of MAP for 1 h. While any duration
could be chosen here, we used 1 h in analogy to running coaches
defining threshold velocity as the effort that can be maintained for
about 1 h33. The 1h utilization ratio p1hU = v1hU/vm had been
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
ARTICLE
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
3
estimated previously from laboratory measurements and races for a
smaller group of 18 male long distance runners to be approximately
0.82 ± 0.0534. Strikingly, our findings from the running data
for ~14,000 subjects corroborate this range without any invasive
measurements, as demonstrated in Fig. 2c. Moreover, our obser-
vation of exponentially small but finite probability for larger El
explains observed values p1hU ≈0.9 in some well trained long dis-
tance runners.
We also computed the marathon race time from our model
and compared it to the actual marathon time Tm for all racing
seasons, see Fig. 3. Our model predicts theoretical curves of
constant Tm in the plane of performance indices (shown as
dashed lines in Fig. 3a). We found that the actual race times are
ordered according to these curves. This shows that our selected
physiological
profiles,
computed
from
sub-marathon
and
marathon best performances, are highly correlated with Tm. It
is important to understand that the position of a marathon
performance in the parameter space is determined by all races
and hence reflects relative importance of the indices vm and El.
This demonstrates the crucial importance of taking into account
endurance in addition to MAP and running economy when
assessing performance of long distance runners.
Importance of endurance. Our findings demonstrate the strong
sensitivity of performance to endurance. For example, a runner
with a velocity of vm = 5 m s−1 can improve his/her marathon
time from 3 h 27 min 38 s to 2 h 53 min 8 s by doubling endur-
ance from El = 3 to El = 6 (corresponding to a change in the one-
hour utilization from 79 to 87% of VO2max), without any change
in VO2,max or running economy. We also find that faster runners
Exercise tracking platform
Raw data
Model
Training
Observations
Subjects
(N=18,993)
completed at least one marathon
[SID, gender]
Activities
(N=2,487,037; d=32,091,664 km)
during period of 180 days before M-date,
distance ≥ 1km, velocity≤7.8m/s
[SID, M-date, time, distance]
Athletes (w/o M)
(N=6749)
at least 2 races/season
Athletes (with M)
(N=14,304)
at least 2 races/season
Race seasons (w/o M)
[SID, M-date]
Race seasons (with M)
(Nraces=64,045)
(Nraces=21,184)
[SID, M-date]
Estimate M-time
Physiol. param. (w/o M)
(NM=9714) [vm, El]
(NM=24,858) [vm, El]
Physiol. param. (with M)
Correlations between
Performance and Training
Training parameters
(NT=21,605)
[dtrain, ptrain, TRIMP]
Training athletes
(N=12,233)
athlete has full training season and
at least 2 races with mean model error< 5%
Training seasons (180 days before M-date)
(activities: N=1,616,004; d=19,959,214 km)
full season: 30 or more activities
[SID, M-date]
SID
Fig. 1 Flowchart of the exercise data analysis. SID: subject identifier, M: marathon, M-date: date of marathon, d: total running distance, “race season”:
fastest times of an athlete for at least two of the distances 5 km, 10 km, half-marathon, and marathon (±3% to account for GPS tolerance), Nraces: total
number of races, NM: number of successful model fits, NT: number of analyzed training seasons for which physiological parameters vm, El could be obtained
and predicted actual race times within a mean error below 5%, “full training season”: at least 30 activities during the 180 days before M-date.
Table 1 Summary of data sets analyzed.
Data
Available
Fit with marathon
Fit w/o marathon
Training seasonc
# Subjects
18,993
14,304
6749
12,233
# Activitiesa (distance ≥ 1 km)
2,487,037
1,616,004
Total distance [km]
32,091,664
19,959,214
Mean distance/activity [km]
12.9
12.4
# Racing eventsb
85,993
64,045
21,184
54,620
# Race/training seasons
24,858
9714
21,605
All data were collected through the PolarFlow web service48.
aAfter removal of unrealistic average velocities (faster than world record).
bDistances are 5 km, 10 km, half-marathon, and marathon depending on the model fit (w or w/o marathon).
cSeasons with # runs ≥ 30.
ARTICLE
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
4
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
tend to race more consistently over all race distances than slower
runners, highlighted by the dependence of the prediction error
ΔTm on the marathon finishing time (see Fig. 4b). For example,
within our fastest group of runners with a marathon time below
160 min, the prediction error was typically less than ±2.5%. This
observation supports our explanation for the observed uncer-
tainty in the endurance parameter El.
Correlation with training. Finally, we compared physiological
profiles to running activities within a training season. There exist
a few studies of the relation between training volume and
intensity, improvements of aerobic fitness and performance35.
For example, it has been stated that running at velocity vm might
represent an optimal stimulus for improving endurance36. There
is also evidence supporting that a relatively large percentage of
low-intensity training over a long period improves performance
during highly intense endurance events37,38. It has been argued
that running velocity at lactate threshold is the best physiological
predictor for distance running performance39.
To investigate the effect of training distance and speed, relative
to the velocity vm, we selected consistent racing seasons defined
by having a mean race time prediction error below 5%. Figure 5a
shows that as the total training distance dtrain of the training
season increases, vm increases on average linearly, with a weak
saturation trend at largest dtrain. Several studies have demon-
strated an increased vm due to endurance training35. A faster
velocity vm can be achieved by a better running economy and/or
an increase in MAP. We hypothesize that longer training distance
has generated improved running economy, in agreement with
earlier observations in a group of eleven well-trained long
0.0
0.1
0.2
0.3
0.4
0.5
1
3
5
7
9
11
13
Endurance El
Probability density
b
0.0
0.1
0.2
0.3
0.4
0.5
2
3
4
5
6
7
vm [m s–1]
Probability density
a
0
1
2
3
4
5
6
0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00
Relative power p1hU = v1hU/vm
Probability density
c
Fig. 2 Probability density of model parameters. The crossover velocity vm which is the smallest velocity that elicits maximal aerobic power MAP and the
endurance El are obtained by applying our model to the fastest performances of a subject for the four distances 5 km, 10 km, Halfmarathon, and Marathon
of a racing season. For these distributions, a total of 24,858 racing seasons have been analyzed. a The velocity vm is approximately normally distributed
with a mean of 4.4 m s−1. b The probability density for the endurance El resembles an exponential decay. c The probability density for the relative power for
1h utilization (1hU) peaks at about 82% of MAP.
b
a
12
Endurance El
10
8
6
4
2
12
Endurance El
10
8
6
4
2
2
3
4
5
6
7
6:00
5:30
4:30
5:00
4:00
3:40
3:20
3:00
2:40
2:20
2:00
6:00
5:30
4:30
5:00
4:00
3:40
3:20
3:00
2:40
Velocity vm [m s–1]
2
3
4
5
6
7
Velocity vm [m s–1]
Tm [min]
330
40
30
20
10
290
250
210
170
n
Fig. 3 Correlation between performance indices and marathon race time (model estimates for 24,504 racing seasons are shown here). a Visualization
of the marathon race time Tm in the (vm, El) parameter plane. Performance indices are obtained from individual's best performances during the racing
season. Color changes from fast (magenta) to slower (blue) finishing times (see color legend for time in minutes). Parameter pairs (vm, El) along the
dashed curves yield the same marathon race time indicated at the top of the graph (in hh:mm format). b Color coded visualization of the number n of racing
seasons analyzed as function of the parameters (vm, El).
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
ARTICLE
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
5
ΔTm
< −20 %
0.06
0.04
0.02
0.00
−20 to −15 %
−15 to −10 %
−10 to −5 %
−5 to 0 %
0 to 5 %
5 to 10 %
10 to 15 %
15 to 20 %
> 20 %
Actual Tm [min]
Tm < 160
160 < Tm < 180
180 < Tm < 200
200 < Tm < 220
220 < Tm < 240
a
b
Endurance El
Probability density
12
10
8
6
4
2
2
3
4
5
6
7
–20 –15 –10 –5
0
5
10
15
20
Velocity vm [m s–1]
ΔTm [%]
6:00
5:30
4:30
5:00
4:00
3:40
3:20
3:00
2:40
Fig. 4 Estimate of Marathon race time from the racing season (for 9410 seasons). a Visualization of the relative difference ΔTm between actual and
estimated marathon race time Tm (in percent of race time) as function of crossover velocity vm and endurance El. Magenta (blue) color indicates a faster
(slower) than estimated finish. b Probability density of race time differences color coded according to groups of different race time intervals.
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
3
5
7
9
11
0.5
0.6
0.7
0.8
3
5
7
9
11
5
10
15
20
25
30
Training TRIMP [×103]
Endurance El
Endurance El
Training intensity ptrain = vtrain / vm
Training intensity ptrain = vtrain / vm
3.0
3.5
4.0
4.5
5.0
5.5
6.0
6.5
1000
2000
3000
dtrain [km]
vm [m s–1]
vm [m s–1]
a
b
c
d
El = 1.83 + 0.101 ptrain e5.19 ptrain
Fig. 5 Correlations between performance indices and training characteristics. We have measured the distance and time for each running activity during a
training season for a total number of 21,605 seasons. The graphs show the observed relations between performance indices (obtained from a model fit to
the racing season) and different measures of training volume and intensity. The magenta line indicates the average, the gray region one standard deviation,
and the light magenta and blue shaded areas represent the standard error of the mean and standard deviation, respectively, as obtained from bootstrap
resampling with replacement (see “Methods” section for more details). a Increase of crossover velocity vm with total distance dtrain of training runs.
b Relation between crossover velocity vm and relative training intensity ptrain ¼ vtrain=vm where vtrain is the average training velocity. c Increase and
saturation of endurance El with training impulse (TRIMP). d Exponential growth of endurance El with relative training intensity.
ARTICLE
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
6
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
distance runners40. Our analysis provides a statistically signifi-
cant, quantitative relation between training distance and speed at
MAP, vm, for ~22,000 training seasons. Another explanation for
this relation could be that fitter runners with a larger MAP and
hence higher vm log more kilometer during their training.
Unfortunately, we could not measure vm at the beginning and the
end of the training season independently from two different
racing seasons or time trials. We also found a linear decrease of
vm with the mean relative training intensity between 50% and
about 90% of vm, as shown in Fig. 5b. Our findings can be
interpreted as faster runners train typically at lower relative
intensities which is consistent with high-intensity performance
improvement due to low-intensity training. The range of training
velocities increases with larger vm which reflects a wider range of
accessible intensities between minimal (jogging) and maximal
speed. For example, a runner with vm = 4 m s−1 typically (within
one standard deviation) trains between 64 and 84% of vm or
MAP, while a runner with vm = 5 m s−1 trains typically up to
66% of vm so that both runners have an almost identical upper
pace ~5 min km−1 for the majority of their runs. Slow runners
must train at a relative high intensity if they want to avoid a
transition to walking. It is important to realize that these typical
ranges do not include fast, high-intensity workouts which account
only for a small fraction of total training volume. However, high-
intensity sessions involve also resting phases that can reduce the
average velocity when timer is not stopped, potentially explaining
observed intensities below ~50% of vm.
Optimal training impulse. We found strong evidence that
combined effect of training volume and intensity, known as
TRaining IMPulse (TRIMP)41, enhances endurance only up to a
limit. Previously, it was found in recreational long distance run-
ners that individual TRIMP correlates with 5000 m and 10,000 m
track performances42. We computed TRIMP by summing the
TRIMP points of all runs of the training season. For each run,
TRIMP points were assigned according to the duration of the run
and its relative average velocity v=vm (see “Methods” section for
details). We analyzed the quantitative relation between endurance
El and total TRIMP of a training season (see Fig. 5c). We
observed an initial linear increase of El with TRIMP, a plateau
around El = 7.5 ± 2 for TRIMP ~25,000, and a statistically sig-
nificant final drop which may be due to over-training. This result
suggests that there is an optimal TRIMP per training season, and
the corresponding maximal endurance enables a close to optimal
marathon race time for a given velocity vm (see Fig. 3a). Finally,
we probed the definition of TRIMP itself to determine if it
implements the best relation between endurance and training
intensity. We found a striking agreement between the exponential
dependence of El on vtrain=vm and the original definition of
TRIMP based on the rise of blood lactate with intensity, as
demonstrated in Fig. 5d. Our findings for thousands of runners
show that relations between training mode and performance
indices that are usually only accessible by invasive and resource-
consuming laboratory testing can be obtained reliably from
running activity data.
Discussion
Recent advances in wearable sensor technology have enabled real-
time and noninvasive measurement of physiological data during
exercise. However, if we are to employ these data to better
understand interplay between exercise, performance and human
health, we must develop new models that are adapted to extract
from the raw data quantities that are most relevant for health and
performance assessment. In this work we have taken this
approach for long distance running to estimate physiological
model indices such as MAP and endurance, and examined their
correlations with training volume and intensity by analyzing
exercise data of ~14,000 marathon runners worldwide. We found
that our recent universal model for a logarithmic relation between
fractional utilization of maximal power and exercise duration14 is
crucial for going beyond previous approaches which ignored this
relation, and for defining a parameter measuring endurance. This
is an important complement to physiological testing in the
laboratory where the required maximal effort is unpractical to
achieve for distances over 20 km. Indeed, our results provide
evidence of the possibility to extract precise indicators for
performance and fitness status from long-duration real-world
exercise tracking data. Using automated digital exercise tracking
goes beyond previous outside-lab studies that relied often on
frequently inaccurate self-reports of exercise. The probability
distributions of the extracted performance indices show large
variances, implying that studies with only a few individuals might
produce misleading results, missing the large interindividual
variability of response to exercise.
Our work has also some limitations: For each activity, only
total distance and duration was available in the data set. This
could lead to biased estimates of the mean velocity, for example
due to periods of rest or stopping with the device timer not
stopped. For the detected correlations between performance
indices and training the direction of any cause–effect relationship
remained open: for example, training with a higher total TRIMP
might produce better endurance, but higher endurance could also
enable runners to follow training modes with a higher TRIMP. To
resolve this relationship, additional data filters need to be devel-
oped to select groups of runners with similar initial performance
which subsequently follow different training modes. However, the
observed correlations can be of practical importance. They can be
useful for estimating realistic expectations for a race for less
experienced runners from their training intensity and volume. In
addition, our observation that endurance peaks at a given training
load (TRIMP) should help preventing over-training, i.e., unpro-
ductive increase in training that can cause injury and other health
problems. It should also be stressed that real-world data always
lack the controlled environment of laboratory based testing. For
example, the energy cost of running has been measured very
accurately in laboratory conditions43–46 and the theoretical
approaches derived from these experiments have motivated the
development of our model.
Our work implies several directions for future research. The
combination of effective models and real-world exercise data
holds great potential for a change in our theoretical description
and understanding of human response to physical activity over
longer periods of time, optimal exercise dosing and training, early
injury detection and prevention, and elite athlete performance.
Approaches similar to ours could be used to develop standards
for cardiorespiratory fitness based on the probability distribution
of performance indices in populations with certain characteristics.
More
detailed,
time-resolved
activity
data
for
heart
rate,
mechanical power output and others could be integrated in our
model to improve accuracy and to extract other performance
indices. Further applications of our approach include the detec-
tion of the usage of performance enhancers in professional sports,
the early identification of talented athletes, and even the effect of
sports equipment like new running shoe technology on perfor-
mance indices47.
Methods
Exercise tracking platform. Exercise data were obtained from Polar Flow web
service48, which is an exercise tracking platform that allows users to upload various
exercise data, including running distance and velocity from GPS watches. Meta
data and activity data of users are linked anonymously through user identification.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
ARTICLE
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
7
Selection of subjects and activities. Users of the exercise tracking platform were
selected as subjects for this study under the conditions that they had completed a
run over the marathon distance (42,195 m) in the period between 1 Jul 2015 and 31
Dec 2018, and used the same GPS watch (Polar V800) for activity recording to
assure comparable accuracy of GPS based distance recording. We analyzed the
running data of ~19,000 individuals who completed ~2.5M activities with a total
distance of ~32M km (see Table 1 for details). For each individual all running
activities in the 180 days before a completed marathon race were grouped together
with the marathon race and the groups labeled uniquely by a subject identifier
(SID) and the marathon date (M-date). Note that an individual may have have
completed multiple marathons during the studied period. For each of those groups,
labeled by the pair (SID, M-date), a race season was defined as the fastest runs of all
activities over the four race distances 5km, 10km, half-marathon (21,097.5m) and
marathon (42,195 m), if distances were available. A tolerance of ±3% was allowed
in the distance selection to account for GPS inaccuracy, and average race velocities
were determined by assuming the actual race distances (which are more reliable
than GPS recordings). We applied conditions that race velocities must increase
with decreasing race distance and must be slower than current world record
velocities. Inconsistent race seasons were identified by violation of these conditions
and excluded from further analysis. Race seasons were defined both with and
without the marathon race included. A valid race season must contain at least two
different race distances. For each race season with a successful performance model
fit with mean race time error below 5% (see section below) a corresponding
training season was defined as all running activities with a total distance ≥1000 m
in the 180 days before the marathon. Runs with apparent velocities ≥7.8 m s−1
(world record for 1000 m) were excluded. Only training seasons with 30 or more
runs were considered so that runner had trained at least once per week and training
seasons with longer interruptions were excluded.
Performance model. We mathematically describe running performance by a
minimal model based on a relative power scale14. The model is formulated in terms
of relative quantities to eliminate irrelevant, subject dependent quantities. The
nominal power expenditure P(v) that is required to run at a constant velocity v, the
so-called running economy, determines the relative power as
pðvÞ ¼ PðvÞ Pb
Pm Pb
¼ v
vm
;
ð2Þ
where we introduced a basal power Pb that is obtained by linearly extrapolating the
running economy to zero velocity and a crossover power Pm that we expect to be
close to the MAP associated with maximal oxygen uptake VO2max. This power Pm
defines a crossover velocity vm that is close to the velocity that permits exercise with
maximal time at MAP. For velocities v > vm the energy cost of running cannot be
determined from oxygen uptake alone due to anaerobic energy supply.
The running performance of an athlete is not only determined by p(v) (which is
fixed by running economy and VO2max) but depends crucially on the average
power Pmax that can be maximally generated over a duration T over which it can be
sustained. To run at the average velocity vmax that can be maximally sustained over
the time T, the nominal power P(vmax) = Pmax(T) is required, establishing a relation
between vmax and T. It has been shown14 that Pmax(T) can be obtained from a self-
consistency relation which states that the time average of the instantaneously
utilized power Pmax(T − t) equals the sum of Pmax(T) and a supplemental power.
This supplemental power has aerobic and anaerobic contributions and accounts for
an upward shift in the power that is required to complete a run with a given
average velocity, for example, due to deteriorating running economy or muscle
fatigue. The existence of an upward shift has been observed experimentally and it is
essential since its absence would yield a duration independent Pmax, which
contradicts the fact that a given power cannot be sustained for an arbitrary
duration. The solution of the self-consistency equation yields
PmaxðTÞ ¼ Pm Pllog T
tc
for T ≥ tc ;
ð3Þ
where Pl measures the supplemental power supply and tc is a crossover time scale
separating different anaerobic and aerobic forms of supplemental power. It can be
shown that for T < tc, Pmax is given by Eq. (3) with Pl replaced by another constant.
By inverting PmaxðTÞ and using the power–velocity relation of Eq. (2), we get the
maximal time TmaxðvÞ ¼ tc exp½ðvm vÞ=ðγlvmÞ over which an average velocity v
can be sustained. Here, the constant γl = Pl/(Pm − Pb) measures endurance
El ¼ expð0:1=γlÞ, see main text. The shortest time T(d) for covering a distance d
follows from solving T ¼ Tmaxðv ¼ d=TÞ for T, yielding Eq. (1). It is important for
the application to a large, inhomogeneous group of subjects that this model is
universal in the sense that it only depends on three parameters vm, tc, and γl and
does not depend directly on any additional, subject-dependent parameters.
Performance data analysis. We tested whether or not meaningful performance
indices can be deduced only from the racing performance of individuals, employing
the performance model described before. For each racing season, uniquely labeled
by a pair (SID, M-date), two model parameters, vm and γl, were computed from
Eq. (1) applied to all races in the racing season. In general, the time tc must be
obtained from the crossover between anaerobic and aerobic regimes, and hence
from races that involve both means of energy supply, i.e., events with finishing time
shorter and longer than tc. Explicit comparison to racing results and laboratory testing
has shown that tc = 6 min is a good approximation on average, and this estimate was
used in our data analysis14. We numerically minimized the sum of the squared
relative differences between the actual race time and the one predicted by Eq. (1). The
nonlinear fitting was based on a Levenberg–Marquardt type algorithm with multiple
starting values to minimize probability to converge only to local minimum, and with
support for lower and upper parameter bounds. Parameter bounds were chosen as 2
m s−1 ≤ vm ≤ 7 m s−1, 0.039 ≤ γl ≤ 0.135 corresponding to 2.1 ≤ El ≤ 13.014. Fits that
converged onto these bounds were excluded from further analysis.
Training data analysis. To quantify training of individuals during the 180-day
period before a marathon, we must establish measures based on duration and
distances of activities within the training season. We considered an optimal set of
three variables that measure quantity, quality, and a combination of quantity and
quality. Training volume was quantified by total running distance dtrain of a
training season. To account for possibly varying physiological adaptions during
different training modes, training intensity ptrain ¼ vtrain=vm was measured by the
average running velocity vtrain in relation to the characteristic velocity vm that was
determined for each race season independently. Finally, the overall training load
was evaluated by the TRIMP scale, which is frequently employed in exercise
physiology and the design of training. TRIMP is a measure for both volume and
intensity of exercise. We assigned to each activity of a training season a TRIMP
number using the definition TRIMP ¼ Ttrainκ1ðv=vmÞ expðκ2v=vmÞ for activity of
duration Ttrain and average velocity v with κ1 = 0.64, κ2 = 1.92 for male subjects,
and κ1 = 0.86, κ2 = 1.67 for female subjects49. The total training TRIMP number
was then obtained by summing the individual TRIMP numbers of all activities
within a training season. Usually TRIMP is defined in terms of the average heart
rate reserve during exercise which is expected to be well approximated by the ratio
v=vm. We are interested in the relation between physiological model parameters vm
and El, and training variables. To measure these relations, we grouped training
variables into bins of widths Δdtrain = 300 km, Δptrain = 0.025 and ΔTRIMP = 2000.
The standard error of the mean and of the standard deviation of vm and El within
each bin was estimated by bootstrap resampling with replacement and computa-
tion of the standard deviation from 1000 bootstrap replicates.
Reporting summary. Further information on research design is available in the Nature
Research Reporting Summary linked to this article.
Data availability
The data that support the findings of this study are available from Polar Electro Oy but
restrictions apply to the availability of these data, which were used under the license
for the current study, and so are not publicly available. Data are, however, available
from the authors upon reasonable request and with permission of Polar Electro Oy
([email protected]).
Code availability
The code (R-script) is available from the Zenodo website https://doi.org/10.5281/
zenodo.4008806.
Received: 18 January 2020; Accepted: 8 September 2020;
References
1.
Lieberman, D. E. & Bramble, D. M. The evolution of marathon running.
Sports Med. 37, 288–290 (2007).
2.
Newby, Z. Athletics in the Ancient World (Bristol Classical Press, 2006).
3.
Althoff, T. et al. Large-scale physical activity data reveal worldwide activity
inequality. Nature 547, 336 (2017).
4.
Pantelopoulos, A. & Bourbakis, N. G. A survey on wearable sensor-based
systems for health monitoring and prognosis. IEEE Trans. Syst. Man Cybern.
Part C 40, 1–12 (2010).
5.
Bandodkar, A. J. & Wang, J. Non-invasive wearable electrochemical sensors: a
review. Trends Biotechnol. 32, 363–371 (2014).
6.
2019 year in sport data report. https://blog.strava.com/press/.
7.
Mazzeo, R. et al. Exercise and physical activity for older adults. Med. Sci. Sports
Exerc. 30, 992–1008 (1998).
8.
Gibala, M. J., Little, J. P., MacDonald, M. J. & Hawley, J. A. Physiological
adaptations to low-volume, high-intensity interval training in health and
disease. J. Physiol. 590, 1077–1084 (2012).
9.
Rawstorn, J. C., Maddison, R., Ali, A., Foskett, A. & Gant, N. Rapid directional
change degrades GPS distance measurement validity during intermittent
intensity running. PLoS ONE 9, 1–6 (2014).
10. Scott, M. T. U., Scott, T. J. & Kelly, V. G. The validity and reliability of global
positioning systems in team sport: a brief review. J. Strength Cond. Res. 30,
1470–1490 (2016).
ARTICLE
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
8
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
11. Sreedhara, V. S. M., Mocko, G. M. & Hutchison, R. E. A survey of
mathematical models of human performance using power and energy. Sports
Med. 5, 54 (2019).
12. Tatterson, A. J., Hahn, A. G., Martini, D. T. & Febbraio, M. A. Effects of heat
stress on physiological responses and exercise performance in elite cyclists.
J. Sci. Med. Sport 3, 186–193 (2000).
13. Vihma, T. Effects of weather on the performance of marathon runners. Int. J.
Biometeorol. 54, 297–306 (2010).
14. Mulligan, M., Adam, G. & Emig, T. A minimal power model for human
running performance. PLoS ONE 13, 1–26 (2018).
15. Hughson, R., Orok, C. & Staudt, L. A high velocity treadmill running test to
assess endurance running potential. Int. J. Sports Med. 5, 23–25 (1984).
16. Kipp, S., Kram, R. & Hoogkamer, W. Extrapolating metabolic savings in
running: Implications for performance predictions. Front. Physiol. 10, 79 (2019).
17. Joyner, M. J. & Coyle, E. F. Endurance exercise performance: the physiology of
champions. J. Physiol. 586, 35–44 (2008).
18. Sproule, J. Running economy deteriorates following 60 min of exercise at 80%
vo2max. Eur. J. Appl. Physiol. 77, 366–371 (1998).
19. Thomas, D. Q., Fernhall, B. & Granat, H. Changes in running economy during
a 5-km run in trained men and women runners. J. Strength Cond. Res. 13,
162–167 (1999).
20. Medbo, J. I. et al. Anaerobic capacity determined by maximal accumulated o2
deficit. J. Appl. Physiol. 64, 50–60 (1988).
21. Hill, A. V. The physiological basis of ahletic records. Lancet 206, 481–486 (1925).
22. Keller, J. B. A theory of competitive running. Phys. Today 26, 43–47 (1973).
23. Peronnet, F. & Thibault, G. Mathematical analysis of running performance
and world running records. J. Appl. Physiol. 67, 453–465 (1989).
24. Batliner, M. E., Kipp, S., Grabowski, A. M., Kram, R. & Byrnes, W. C. Does
metabolic rate increase linearly with running speed in all distance runners?
Sports Med. Int. Open 2, E1–E8 (2018).
25. Caminal, P. et al. Validity of the Polar V800 monitor for measuring heart rate
variability in mountain running route conditions. Eur. J. Appl. Phys. 118,
669–677 (2018).
26. Billat, V., Binsse, V., Petit, B. & Koralsztein, J. J. High level runners are able to
maintain a vo2 steady-state below vo2max in an all-out run over their critical
velocity. Arch. Physiol. Biochem. 106, 38–45 (1998).
27. Corless, R., Gonnet, G., Hare, D., Jeffrey, D. & Knuth, D. On the lambert w
function. Adv. Comp. Math. 5, 329–359 (1996).
28. di Prampero, P. E., Atchou, G., Brueckner, J. C. & Moia, C. The energetics of
endurance running. Eur. J. Appl. Phys. 55, 259 (1986).
29. di Prampero, P. E. et al. Energetics of best performances in middle-distance
running. J. Appl. Physiol. 74, 2318 (1993).
30. Lazzer, S. et al. The energetics of ultra-endurance running. Eur. J. Appl. Phys.
112, 1709 (2012).
31. Daniels, J. T. A physiologist’s view of running economy. Med. Sci. Sports
Exerc. 17, 332 (1985).
32. Manari, D. et al. Vo2max and vo2at: athletic performance and field role of
elite soccer players. Sport Sci. Health 12, 221–226 (2016).
33. Daniels, J. Daniels’ Running Formula (Human Kinetics, 2013), 3rd edn.
34. Farrell, P. A., Wilmore, J. H., Coyle, E. F., Billing, J. E. & Costill, D. L. Plasma
lactate accumulation and distance running performance. Med. Sci. Sports 11,
338–344 (1979).
35. Jones, A. & Carter, H. The effect of endurance training on parameters of
aerobic fitness. Sports Med. 29, 373–386 (2000).
36. Billat, V., Renoux, J., Pinoteau, J., Petit, B. & Koralsztein, J. Reproducibility of
running time to exhaustion at vo2max in subelite runners. Med. Sci. Sports
Exerc. 26, 254–257 (1994).
37. Esteve-Lanao, J., San Juan, A., Earnest, C., Foster, C. & Lucia, A. How do
endurance runners actually train? Relationship with competition performance.
Med. Sci. Sports Exerc. 37, 496–504 (2005).
38. Esteve-Lanao, J., Foster, C., Seiler, S. & Lucia, A. Impact of training intensity
distribution on performance in endurance athletes. J. Strength Cond. Res. 21,
943–949 (2007).
39. Bassett, D. & Howley, E. Limiting factors for maximum oxygen uptake and
determinants of endurance performance. Med. Sci. Sports Exerc. 32, 70–84
(2000).
40. Kubo, K., Tabata, T., Ikebukuro, T., Igarashi, K. & Tsunoda, N. A longitudinal
assessment of running economy and tendon properties in long-distance
runners. J. Strength Cond. Res. 24, 1724–1731 (2010).
41. Banister, E. W. & Calvert, T. W. Planning for future performance:
implications for long term training. Can. J. Appl. Sport Sci. 5, 170 (1980).
42. Manzi, V., Iellamo, F., Impellizzeri, F., D’Ottavio, S. & Castagna, C. Relation
between individualized training impulses and performance in distance
runners. Med. Sci. Sports Exerc. 41, 2090–2096 (2009).
43. Margaria, R., Cerretelli, P., Aghemo, P. & Sassi, G. Enery cost of running.
J. Appl. Physiol. 18, 367–370 (1963).
44. di Prampero, P. E. The energy cost of human locomotion on land and in
water. Int. J. Sports Med. 7, 55–72 (1986).
45. Ferretti, G., Bringard, A. & Perini, R. An analysis of performance in human
locomotion. Eur. J. Appl. Physiol. 111, 391–401 (2011).
46. Tam, E. et al. Energetics of running in top-level marathon runners from
kenya. Eur. J. Appl. Physiol. 112, 3797–3806 (2012).
47. Hoogkamer, W. et al. A comparison of the energetic cost of running in
marathon racing shoes. Sports Med. 48, 1009–1019 (2018).
48. Polar Flow. https://flow.polar.com/ (2019).
49. Borresen, J. & Lambert, M. I. The quantification of training load, the training
response and the effect on performance. Sports Med. 39, 779–795 (2009).
Acknowledgements
The support by Polar Electro in obtaining the exercise data from their data base is greatly
acknowledged.
Author contributions
T.E. designed the study and performed the numerical analysis. T.E. and J.P. wrote
the paper.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/s41467-
020-18737-6.
Correspondence and requests for materials should be addressed to T.E.
Peer review information Nature Communications thanks Guido Ferretti and the other,
anonymous, reviewer(s) for their contribution to the peer review of this work. Peer
reviewer reports are available.
Reprints and permission information is available at http://www.nature.com/reprints
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit http://creativecommons.org/
licenses/by/4.0/.
© The Author(s) 2020
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18737-6
ARTICLE
NATURE COMMUNICATIONS | (2020) 11:4936 | https://doi.org/10.1038/s41467-020-18737-6 | www.nature.com/naturecommunications
9
| Human running performance from real-world big data. | 10-06-2020 | Emig, Thorsten,Peltonen, Jussi | eng |
PMC2346553 | The Aerodynamic Signature of Running Spiders
Je´roˆ me Casas1*, Thomas Steinmann1, Olivier Dangles2,3
1 University of Tours, Institut de Recherches sur la Biologie de l’Insecte, UMR CNRS 6035, 37200 Tours, France, 2 IRD, UR 072, LEGS, UPR 9034, CNRS, 91198 Gif-sur-Yvette,
France, 3 Universite´ Paris-Sud 11, 91405 Orsay, France
Abstract
Many predators display two foraging modes, an ambush strategy and a cruising mode. These foraging strategies have been
classically studied in energetic, biomechanical and ecological terms, without considering the role of signals produced by
predators and perceived by prey. Wolf spiders are a typical example; they hunt in leaf litter either using an ambush strategy
or by moving at high speed, taking over unwary prey. Air flow upstream of running spiders is a source of information for
escaping prey, such as crickets and cockroaches. However, air displacement by running arthropods has not been previously
examined. Here we show, using digital particle image velocimetry, that running spiders are highly conspicuous
aerodynamically, due to substantial air displacement detectable up to several centimetres in front of them. This study
explains the bimodal distribution of spider’s foraging modes in terms of sensory ecology and is consistent with the escape
distances and speeds of cricket prey. These findings may be relevant to the large and diverse array of arthropod prey-
predator interactions in leaf litter.
Citation: Casas J, Steinmann T, Dangles O (2008) The Aerodynamic Signature of Running Spiders. PLoS ONE 3(5): e2116. doi:10.1371/journal.pone.0002116
Editor: Andrew Iwaniuk, Smithsonian Institution, United States of America
Received January 7, 2008; Accepted March 31, 2008; Published May 7, 2008
Copyright: 2008 Casas et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work is part of the research conducted within the Cricket Inspired perCeption and Autonomous Decision Automata (CICADA) project (IST-2001-
34718) and within the Customized Intelligent Life Inspired Arrays (CILIA) project (FP6-IST-016039). These projects are both funded by the European Community
under the ‘‘Information Society Technologies-IST’’ Program, Future and emergent Technologies (FET).
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: [email protected]
Introduction
Many predatory species can switch between foraging modes,
usually alternating between an ambush and a cruising mode in
water,
soil
or
vegetation.
Much
care
has
been
taken
in
evolutionary ecology to evaluate the relative advantages of
foraging strategies in terms of energetics, biomechanics, success
rate and impact on the ecosystem [1–7]. However, the relationship
between the sensory processes involved in signal production by a
predator attacking with one of both strategies and the corre-
sponding signal perception by its escaping prey is unknown for
most systems. The outcome of this relationship is likely to play an
important role in defining the most appropriate predatory foraging
mode. For instance, wolf spiders pursue their cricket prey on the
bare soil and in leaf litter using two attack strategies [8–10].
Spiders attack prey using either an extremely slow-motion
approach, corresponding almost to the ambush strategy, or by
running over at relatively high speed (up to 40 cm/s, cruising
strategy) [10]. Spiders attack at intermediate speeds much less
frequently; biotests using a piston mimicking the attack of a spider
showed that a cricket’s chances of survival were highest for attacks
at intermediate speed (20 cm/s) [10]. Although crickets and many
other detritivorous and herbivorous arthropods are sometimes
caught unaware by a spider’s fast strike, they often escape with fast
movements. Information contained in air signals upstream from
running spiders can be used by prey in these fast escape reactions.
Indeed, crickets, cockroaches and other orthropteroid insects are
equipped with air-flow sensors (filiform hairs) at the rear end of
their abdomen [11]. They possess many short hairs, serving as
acceleration sensors, and fewer long hairs (velocity sensors) on
their cerci [12]. These mechanosensors are among the most
sensitive sensors in the animal kingdom, with action potentials
triggered by less than one tenth the energy of a photon [13];
indeed, the orthropteroid escape system, and in particular fluid flow
sensing using filiform hairs, has maintained textbook-example status
over many years [14–17]. Thus, we hypothesised that spiders use the
two different hunting strategies to cope with optimal air-flow
detection by crickets. One strategy (ambush) substantially reduces
the distance at which the prey can perceive the attack, while the
other strategy (cruising) reduces the escape probability by over-
whelming the prey sensory capabilities. The high speed ensures that
the encounter occurs faster than the escape response.
The aims of this study were therefore: (1) to quantify the air flow
in front of a running spider using digital particle imaging
velocimetry (DPIV), and (2) to assess these complex flow patterns
in the context of attack and escape strategies by predators and
prey. Very little is known about air movements upstream from a
running arthropod, limiting potential evaluation of the ecological
and evolutionary importance of air-flow sensing for many
predator-prey interactions. Near-field fluid movement cues are
used by many invertebrate species to obtain information about
potential predators, prey or mates, in both terrestrial and aquatic
ecosystems. In particular, several recent studies have led to greater
understanding of the physics of near-field fluid motion in animal
locomotion and sensing in open enclosures. Such technological
and conceptual advances have opened up the arena for similar
studies on running animals [18–23].
Results
We recorded the air flow produced by wolf spiders (Pardosa
[lugubris] sp., most likely P. lugubris (Walkenaer)) running in a small
wind tunnel (Figure 1). As spiders dislike the intense laser light
sheet, we obtained 14 runs from six different individuals with the
PLoS ONE | www.plosone.org
1
May 2008 | Volume 3 | Issue 5 | e2116
horizontal set-up, but only two runs with the vertical set-up. These
were not used in the following quantitative analysis, but gave
useful information on several other qualitative aspects of the flow,
described below. The mean velocity of the spiders recorded in the
horizontal set-up was 9.44 cm/s (SD = 65.51; N = 14). This lies
within the range of attack speeds observed under unconstrained
hunting behaviour [10]. One spider ran at a high speed of 40 cm/
s. This was an outlier in the velocity distribution, and so was not
used to calculate the mean. Running spiders displaced air in front
of and above their body trunk (Figures 2 & 3). Pockets of high
velocity produced by moving legs could be distinguished and
substantially extended the region of flow influenced by the spider
(Figure 2). Front legs still produce a forward air movement when
moving downwards, as they do not move back and forth (Figure 3B
and cartoon on Figure 2). The air field within the first centimetre
upstream from a spider varies considerably from run to run
because it is not possible to synchronise the PIV clock with the leg
kinematics. Thus, depending on the exact moment of flow field
mapping, a leg may or may not have a large effect on the flow in
its near vicinity (see cartoon, Figure 2). This also explains the
absence of relationship observed between the air velocity at 6 mm
away from the spider and the spider’s body velocity, and our
subsequent decision to pool individual runs for a statistical
analysis. The air flow upstream from a running spider declines
smoothly with distance (Figure 4); a constrained regression, using
the function given in (Eq. 2) and the independently measured
mean spider’s velocity as a fixed parameter, lead to a good fit over
the whole range of distances (R2 = 0.80).
Discussion
The air field upstream from a running spider is disturbed over a
large distance of several body lengths. The need for prey to
perceive attacking predators from as large a distance as possible,
using the minimal amount of energy, means that this information
is of biological importance. Indeed, previous experimental studies
on the air flow produced by attacking toads shooting out their
toungs [24] and independent theoretical studies [25] suggested
that cockroaches may recognise the wind signature of a predator
by the low frequency components in the far field. The most sensitive
hairs of crickets are the longest ones (.1000 microns), working near
the thermal noise level [13]. Electrophysiological studies estimate
their minimal threshold at Vthresh= 30 mm/s. Thus, using the
expected flow velocity upstream from a running spider from the
fitted model, this threshold should be attained at around 3 cm in
front of a spider. This distance, obtained using the observed mean
speed, will vary as a function of the speed of the spider. Crickets seem
to make full use of this information, with their largest escape
distances being 2.4 cm in front of a spider and 2.1 cm in front of a
piston device mimicking the kinematics of the attack [10]. This is
most impressive, given the time taken for processing information in
the abdominal terminal ganglion, the insect brain, and from leg
movements [26]. Thus, the cricket’s entire escape system, including
sensory and locomotive control, is indeed optimised to pick up the
slightest air movements by the best sensors.
The implications of our results for the foraging modes of spiders
are twofold. First, spiders markedly increase their likelihood of
successful attacks by launching fast strikes, at the same time
decreasing the potential escape time (time between danger
perception by a cricket and encounter by a spider) in a non-
linear fashion (Figure 5). While low speed movements imply high
potential escape times, the distance at which prey can perceive
predatory signals is so short that prey are nearly within reach of
spiders (ambush strategy). Second, their highest speeds may
correspond to the lowest potential prey escape time [26]. Such
attack speeds are between 25–35 cm/s, corresponding well with
Figure 1. Digital particle image velocity (DPIV) measurements of a running spider. In the horizontal position, the laser light sheet is
focussed 3 mm above the floor, at mid-height of the spider, just below the bottom eye row level. The yellow portion represents the camera’s field of
view. Spiders were gently triggered to run using a stick inserted through a small hole at the entrance of the wind tunnel.
doi:10.1371/journal.pone.0002116.g001
Aerodynamics of Spiders
PLoS ONE | www.plosone.org
2
May 2008 | Volume 3 | Issue 5 | e2116
the higher speeds distribution observed during spider-cricket
interactions. Higher hunting speeds are seldom observed, as they
do not increase the capture rate but are energetically expensive.
Thus, our quantification of air flow upstream from a running
predator extends the interpretation of the two foraging modes in
terms of sensory processes, beyond the classical description in
energetic and biomechanical terms. Future studies dedicated to
body
and
leg
kinematics
should
be
prioritised,
since
our
understanding of this subject is substantially poorer than that of
wing and leg kinematics in insects, and their influence on the
upstream flow. The role of acceleration, body posture and height
over the substrate [27,28] as well as the nature of the substrate,
aspects which we have neglected here, are also expected to have a
major impact on the flow field upstream from the spider.
Many other invertebrate predators, including several other
arachnid groups, carabid, cincidelid and staphylinid beetles, hunt
prey using the same two strategies as those used by spiders. At the
same time, many prey living in litter harbour well-developed cerci
bearing filiform hairs triggered by slight air movements. These
include primitive and modern insects such as bristletails, firebrats,
springtails, cockroaches and crickets; indeed, most prey-spider
interactions observed today are the same as they were some
400 million years ago [29,30]. For example, cockroaches have been
extremely successful and thrive in tropical leaf litter despite strong
predator pressure. Our findings demonstrate a significant role of the
physical information contained in slight air currents in interspecific
interactions among terrestrial arthropods and suggest a tight sensory
coevolution between both opponents. Lurking predators may mostly
hide and wait for their prey, but the final strike produces conspicuous
signals that prey exploit for their survival.
Materials and Methods
DPIV
Our measurement set-up was composed of a sealed glass box
(106262 cm), seeded with 0.2 mm oil particles. Oil particles (Di-
Ethyl-Hexyl-Sebacat, 0.5 L, TPAS, Dresden, Germany) were
generated using an aerosol generator (ATM 230, ACIL, Chatou,
France). The laser (NewWave Research Solo PIV 2, Nd:YAG, dual
pulsed; Dantec Dynamics A/S, Skovlunde, Denmark) illuminated
the flow produced by the spider’s displacement through glass. The
laser sheet (width = 17 mm, thickness at focus point = 50 mm) was
operated at low power (3 mJ at 532 nm) to minimise glare. A target
area (17630 mm) was then imaged onto the CCD array of a digital
camera (Photron FastCam X1280 PCI 4K) using a Macro Lens
(Nikon, AF Nikkor, 60 mm, f : 2.8). The CCD captured separate
image frames (128061024 px). Once a sequence of two light pulses
was recorded, the images were divided into small subsections which
were cross-correlated with each other using flow map software (Flow
Manager 4.4. Dantec Dynamics A/S, Skovlunde, Denmark). The
Figure 2. Horizontal flow field and close-up view of the flow around a running spider. The sequence in (A) highlights the pockets of high
air-flow velocity created by leg strokes superimposed on the air movements created by the body trunk movement. Neither the tips of the spider’s
legs, nor their associated flow patterns, are visible as they are located below the light sheet. The time delay between two images is 500 ms; the spider
was running at a speed of 5.7 cm/s. The cartoon, adapted from [9], highlights the relative position of legs to body trunk. An overlay of two images
(first image in white, second image in grey) of the moving spider, separated by 500 ms, is shown in (B). The zone of flow velocities above the
measurable range is in black. The running speed was 10.5 cm/s.
doi:10.1371/journal.pone.0002116.g002
Aerodynamics of Spiders
PLoS ONE | www.plosone.org
3
May 2008 | Volume 3 | Issue 5 | e2116
correlation was achieved using an interrogation area of 32632
pixels, allowing us to obtain valid measurements down to a particle
displacement of 0.1 pixels. Using the equation,
sV~ sDx
Dtime
~ spixdr
Dtime
ð1Þ
with sDx the minimal displacement measurable (m), Dtime = 33 ms,
the time separating two image record and dr = 27 mm the spatial
resolution, one obtains the lowest detectable speed of 0.082 mm/s,
and of 5.4 mm/s for a time interval of 500 ms. Conversely, with the
maximal measurable particle displacement of 32 pixels, the maximal
detectable speed is 2.62 cm/s for a time interval of 33 ms, and
17.3 cm/s for a time interval of 500 ms.
Estimation of spider’s velocity and profile extraction
Pardosa (Koch) is the most speciose genus among Holarctic wolf
spider genera. Several species groups have been recognized, based
on characteristics of the copulatory organs [31]. Based upon
identification of mature males from our collecting sites, Pardosa
lugubris (Walkenaer) was the most common species. However, this
species was recently shown to incorporate distinct cryptic species
whose immature individuals are, to date, impossible to differen-
tiate (Kronestedt 2007). In our experiments, we used only
immature spiders because they naturally spend much of their
time hunting for prey and not seeking for partners. The mean
body size was 3.6 mm (S.D. = 0.2 mm, N = 6). The body size was
obtained by measuring the largest width of the prothorax, to which
we added the lengths of coxa and the trochanter, as these three
body parts act aerodynamically as a single unit. In the studied
spiders, this unit was wider than the abdomen. During a single
time interval of 33 ms, a spider travelled a distance of 5 mm when
moving at a speed of 15 cm/s. There are therefore no data
available on flow velocity for the 5 mm space next to the body
surface. The distance from the body for which no information was
available was greater for greater speeds.
In the horizontal set-up, we took care that the laser light sheet is
focused 3 mm above the floor, at mid-height of the spider, just
below the bottom eye row level. However, we cannot ascertain
that the laser light sheet, which is diverging with an angle of 24u
Figure 3. Vertical flow field and close-up view of the flow around a running spider. The sequence in (A) highlights the high air-flow
velocity above the spider’s body. The time delay between two images is 500 ms; the spider was running at a speed of 3.7 cm/s. An overlay of two
images (first image in white, second image in grey) of a moving spider, separated by 500 ms, is shown in (B). The horizontal component of the air flow
in the near vicinity of the legs is always directed forward, as front legs do not move back and forth (see cartoon in Figure 2). The running speed was
21 cm/s.
doi:10.1371/journal.pone.0002116.g003
Aerodynamics of Spiders
PLoS ONE | www.plosone.org
4
May 2008 | Volume 3 | Issue 5 | e2116
from the focal point, did not affect the spider, or during the low
phase of the body oscillations. Whatever the amount of light
spiders did get, it was much below the intensity of the bulk of the
laser sheet, as we would otherwise see the eyes within the light
sheet. We observed a tendency to avoid the laser light sheet rapidly
in the vertical set-up.
We recorded 14 runs made by six Pardosa [lugubris] sp. spiders
with the horizontal set-up. Measurements were only made when
the spider velocity was assumed constant for several centimetres
and
spiders
were
running
straight.
The
constant
velocity
assumption is derived from the measurements in [10] reporting
an acceleration phase restricted to one centimetre, followed by a
constant velocity. We therefore positioned the field of view of the
camera at least 2–3 centimetres away from the entrance of the
tunnel. The spider’s velocity was determined by measuring the
average velocity of the spider’s body on a run. A run was restricted
5
10
15
20
25
0
1
2
3
4
5
6
7
Distance from body (mm)
Velocity (mm/s)
Figure 4. Flow velocity upstream of running spiders. The observed speeds (mean and standard deviation; dots and error bars, respectively),
and the fit of the statistical function (Eq. 2) are represented.
doi:10.1371/journal.pone.0002116.g004
Figure 5. Spider’s attack speed and cricket escape time. The potential escape time for a cricket (red line) is expressed as a function of the
spider’s attack speed. At slow attack speeds, the distance at which crickets can perceive spiders is limiting (ambush strategy), whereas at high
hunting speeds, the escape time becomes limiting (cruising strategy). The potential escape time is defined as the time interval between predator
perception by a cricket and hit by a spider running at a given speed. It is based on the distance, for a given speed, at which the threshold of 30 mm/s
for danger perception is attained [13]. The minimal recorded escape time for crickets is around 0.2 ms (horizontal bar, [26]). The distribution of
observed attack speeds and the five successful attacks (stars) were obtained from observations of real attacks, at constant speeds, during cricket-
spider interactions [10].
doi:10.1371/journal.pone.0002116.g005
Aerodynamics of Spiders
PLoS ONE | www.plosone.org
5
May 2008 | Volume 3 | Issue 5 | e2116
to the pairs of images (varying from one to five pairs) for which the
images were of quality high enough for a faithful quantification of
air flow. We extracted velocity profiles from the vector fields for
each measurement. Profiles were evaluated along the upstream
axis. In order to describe the flow velocity as faithfully as possible,
we fitted the data with a flexible statistical function:
V~Vbody
1
6
A
A
2 zx z 1
12
B2
A
2 zx
2 z 1
24
C3
C
2 zx
3
!
ð2Þ
With x being the distance to the spider’s body (m), A = 0.0007,
B = 20.0011 and C = 0.0179 and Vbody, the spider’s body velocity
(0.0944 m/s).
Acknowledgments
We thank the consortium members of CICADA and CILIA teams for
discussions, M. Greenfield, G. Krijnen, C. Lazzari, E. McCauley and J.
Mogdans, and two reviewers for comments on the MS.
Author Contributions
Conceived and designed the experiments: JC TS OD. Performed the
experiments: TS. Analyzed the data: JC TS. Contributed reagents/
materials/analysis tools: JC TS. Wrote the paper: JC TS OD.
References
1. Schoener TW (1971) Theory of feeding strategies. Ann Rev Ecol. Sys 2:
369–404.
2. Curio E (1976) The ethology of predation. Berlin: Springer Verlag. 250 p.
3. Huey RB, Pianka ER (1981) Ecological consequences of foraging mode. Ecology
62: 991–999.
4. Kramer DL, McLaughlin RL (2001) The behavioural ecology of intermittent
locomotion. Am Zool 41: 137–153.
5. Cooper WE (2005) The foraging mode controversy: both continuous variation
and clustering of foraging movements occur. J Zool 267: 179–190.
6. Schmitz O (2007) Predator diversity and trophic interactions. Ecology 88:
2415–2426.
7. Schmitz OJ (2008) Effects of predator hunting mode on grassland ecosystem
function. Science 319: 952–954.
8. Riechert SE (1992) Spiders as representative ’Sit and wait’ predators. In:
Crawley MJ, ed. Natural enemies. London: Blackwell. pp 313–328.
9. Foelix RF (1996) Biology of Spiders. Oxford: Oxford University Press. 330 p.
10. Dangles O, Ory N, Steinmann T, Christides JP, Casas J (2006) Spider’s attack
vs. cricket’s escape: velocity modes determine success. Anim Behav 72: 603–610.
11. Edwards JS, Palka J (1974) The cerci and abdominal giant fibres of the house
cricket Acheta domesticus. I Anatomy and physiology of normal adults. Proc R Soc
Lond B 185: 83–103.
12. Shimozawa T, Kanou M (1984) Varieties of filiform hairs: range fractionation by
sensory afferents and cercal interneurons of a cricket. J Comp Physiol A 155:
485–493.
13. Shimozawa T, Murakami J, Kumagai T (2003) Cricket wind receptors: thermal
noise for the highest sensitivity known. In: Barth FG, Humphrey JAC,
Secomb T, eds. Sensors and sensing in biology and angineering. Berlin:
Springer Verlag. pp 145–157.
14. Camhi JM (1984) Neuroethology: Nerve cells and the natural behaviour of
animals. Sunderland: Sinauer Associates. 432 p.
15. Tautz J (1989) Medienbewegung in der Sinneswelt der Arthropoden. Stuttgart:
Gustav Fischer Verlag. 59 p.
16. Gnatzy W (1996) Digger wasp vs. cricket: neuroethology of a predator-prey
interaction. Stuttgart: Gustav Fischer Verlag. 92 p.
17. Dangles O, Casas J, Coolen I (2006) Textbook cricket goes to the field: the
ecological scene of the neuroethological play. J Exp Biol 209: 393–398.
18. Dickinson MH, Farely CT, Full RJ, Koehl MAR, Kram R, et al. (2000) How
animals move: an integrative view. Science 288: 100–106.
19. Koehl MAR, Koseff JR, Crimaldi JP, Cooper T, McCay M, et al. (2001) Lobster
sniffing filters the spatiotemporal information in a turbulent odour plume.
Science 294: 1948–1951.
20. Taylor GK, Nudds RL, Thomas ALR (2003) Flying and swimming animals
cruise at a strouhal number tuned for high power efficiency. Nature 425:
707–711.
21. Warrick DR, Tobalske BW, Powers DR (2005) Aerodynamics of the hovering
hummingbird. Nature 435: 1094–1097.
22. Steinmann T, Casas J, Krijnen G, Dangles O (2006) Air-flow sensitive hairs:
boundary layers in oscillatory flows around arthropod appendages. J Exp Biol
209: 4398–4408.
23. Lauder GV, Madden PGA (2008) Advances in comparative physiology from
high-speed imaging of animal and fluid motion. Annu Review Physiol; Doi:
10.1146/annurev.physiol.70.113006.100438.
24. Plummer MR, Camhi JM (1981) Discrimination of sensory signals from noise in
the escape system of the cockroach: the role of wind acceleration. J Comp
Physiol A 142: 347–357.
25. Rinberg D, Davidowitz ÆH (2003) Wind spectra and the response of the cercal
system in the cockroach. J Comp Physiol A 189: 867–876.
26. Tauber E, Camhi JF (1995) The wind-evoked escape behaviour of the cricket
Gryllus bimaculatus: integration of behavioural elements. J Exp Biol 198:
1895–1907.
27. Full RF, Tu MS (1991) Mechanics of a rapid running insect: two-, four-, and six-
legged locomotion. J Exp Biol 156: 215–231.
28. Nishikawa K, Biewener AA, Aerts P, Ahn AN, Chiel HJ, et al. (2007)
Neuromechanics: an integrative approach for understanding motor control.
Integr Comp Biol 47: 16–54.
29. Labandeira CC (2002) Paleobiology of predators, parasitoids, and parasites:
death and accommodation in the fossil record of continental invertebrates. Paleo
Soc Papers 8: 211–249.
30. Grimaldi D, Engel MS (2005) Evolution of the insects. Cambridge: Cambridge
University Press. 755 p.
31. Kronestedt T (2007) A new species of wolf spider from the Pyrenees, with
remarks on other species in the Pardosa pullata-group (Araneae, Lycosidae).
Zootaxa 1650: 25–40.
Aerodynamics of Spiders
PLoS ONE | www.plosone.org
6
May 2008 | Volume 3 | Issue 5 | e2116
| The aerodynamic signature of running spiders. | 05-07-2008 | Casas, Jérôme,Steinmann, Thomas,Dangles, Olivier | eng |
PMC3439386 | Alveolar-Membrane Diffusing Capacity Limits
Performance in Boston Marathon Qualifiers
Kaleen M. Lavin1, Allison M. Straub1, Kathleen A. Uhranowsky1, James M. Smoliga3,
Gerald S. Zavorsky1,2*
1 Human Physiology Laboratory, Marywood University, Scranton, Pennsylvania, United States of America, 2 The Commonwealth Medical College, Scranton, Pennsylvania,
United States of America, 3 Department of Physical Therapy, High Point University, High Point, North Carolina, United States of America
Abstract
Purpose: (1) to examine the relation between pulmonary diffusing capacity and marathon finishing time, and (2), to
evaluate the accuracy of pulmonary diffusing capacity for nitric oxide (DLNO) in predicting marathon finishing time relative
to that of pulmonary diffusing capacity for carbon monoxide (DLCO).
Methods: 28 runners [18 males, age = 37 (SD 9) years, body mass = 70 (13) kg, height = 173 (9) cm, percent body fat = 17 (7)
%] completed a test battery consisting of measurement of DLNO and DLCO at rest, and a graded exercise test to determine
running economy and aerobic capacity prior to the 2011 Steamtown Marathon (Scranton, PA). One to three weeks later, all
runners completed the marathon (range: 2:22:38 to 4:48:55). Linear regressions determined the relation between finishing
time and a variety of anthropometric characteristics, resting lung function variables, and exercise parameters.
Results: In runners meeting Boston Marathon qualification standards, 74% of the variance in marathon finishing time was
accounted for by differences in DLNO relative to body surface area (BSA) (SEE = 11.8 min, p,0.01); however, the relation
between DLNO or DLCO to finishing time was non-significant in the non-qualifiers (p = 0.14 to 0.46). Whereas both DLCO
and DLNO were predictive of finishing time for all finishers, DLNO showed a stronger relation (r2 = 0.30, SEE = 33.4 min,
p,0.01) compared to DLCO when considering BSA.
Conclusion: DLNO is a performance-limiting factor in only Boston qualifiers. This suggests that alveolar-capillary membrane
conductance is a limitation to performance in faster marathoners. Additionally, DLNO/BSA predicts marathon finishing time
and aerobic capacity more accurately than DLCO.
Citation: Lavin KM, Straub AM, Uhranowsky KA, Smoliga JM, Zavorsky GS (2012) Alveolar-Membrane Diffusing Capacity Limits Performance in Boston Marathon
Qualifiers. PLoS ONE 7(9): e44513. doi:10.1371/journal.pone.0044513
Editor: Mauricio Rojas, University of Pittsburgh, United States of America
Received March 30, 2012; Accepted August 3, 2012; Published September 11, 2012
Copyright: 2012 Lavin et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The authors have no funding or support to report.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: [email protected]
Introduction
In 2010, nearly one half million runners in the United States
completed a marathon, representing about 0.2% of the U.S.
population over 18 years of age. Many marathoners aspire to
qualify for the Boston Marathon, participation in which is
restricted to a relatively small percentage of runners by age and
gender-graded qualification standards. In 2011, approximately
40,000 runners qualified for the 2012 or 2013 Boston Marathon,
representing about 10% of all runners who finished a marathon in
the United States. These statistics are readily available to the
public online at MarathonGuide.com.
Due to its aura and relative popularity, marathon running has
been examined in several scientific studies, with particular
attention given to the metabolic [1] and physiological [2,3,4,5,6]
correlates of running a fast marathon. Although these correlates
are multifactorial and widely debated [7], it is well established that
aerobic capacity ( _VO2max) is an important determinant of
marathon performance. Approximately 40 to 77% of the variance
in marathon performance is attributable to _VO2max [5,8,9,10,11].
In addition to aerobic capacity, marathon-specific endurance is
related to performance, such that the ability to sustain a higher
percentage of _VO2max is correlated with a faster marathon [12].
On average, top marathoners (,136 minutes for men, ,158
minutes for women) run at 85 to 90% _VO2max [5], while those in
the 156 to 240 minute range run at approximately 75 to 85% of
_VO2max [10,13].
Aerobic capacity is dependent on the integrated function of
major organ systems, including the heart, lung, and skeletal muscle
[14]. Unlike the heart and skeletal muscle, the lung does not
readily adapt to endurance training [15,16], possibly limiting
_VO2max [15]. Even in those who are highly aerobically fit, heavy
exercise may cause arterial oxygen pressure to drop #80 mm Hg,
while the alveolar-to-arterial oxygen pressure difference may
increase to $25 mmHg [17]. In this way, arterial oxyhemoglobin
saturation is reduced, leading to a decrease in
_VO2max and,
subsequently, endurance performance [18,19,20].
Recently, alveolar-membrane diffusing capacity (measured at
rest) has been shown to be very closely related to _VO2max in fit and
PLOS ONE | www.plosone.org
1
September 2012 | Volume 7 | Issue 9 | e44513
obese individuals [21,22] and to longevity in heart disease patients
[23]. Specifically, when measured at rest, pulmonary diffusing
capacity for nitric oxide (DLNO, mL NO.min21.mmHg21) – a
surrogate for alveolar-membrane diffusing capacity for carbon
monoxide (DmCO) [24,25] – has been shown to be related to
aerobic capacity in fit men and women, such that for every 1 unit
increase in DLNO, _VO2max increases by 0.3 mL O2. kg21. min21
[21]. The ratio of DLNO to DmCO is debated. It has been said
that DmCO = DLNO 4 2.42, or, more recently, DmCO =
DLNO 4 2.06 to 2.26. As long as the ratio is kept consistent
within a study, any percent change in DmCO or pulmonary
capillary blood volume is still valid. Furthermore, alveolar
membrane conductance is the main pulmonary diffusing capacity
component representative of fitness, with the exception of
pulmonary capillary blood volume (VC) and the blood transfer
conductance (H) for CO (HCO). There are mixed data as to
whether DLCO or DLNO is a more valid predictor of aerobic
capacity, but overall diffusion capacity does appear to be
significantly correlated with aerobic capacity in fit subjects
[21,22]. Nitric oxide (NO) has been shown to bind more strongly
than CO to hemoglobin [26], leading to a higher value for
membrane conductance and a diffusion measurement more
reflective of total membrane diffusion. An additional benefit of
measuring DLNO simultaneously with DLCO is a reduction in
time and effort of the procedure.
As there is a relation between DLNO and
_VO2max, and
between _VO2max and marathon running performance, it follows
that DLNO may be related to marathon performance, such that
pulmonary diffusing capacity introduces a limitation that influ-
ences marathon performance. However, it is likely that this
correlation may be observed in only well-trained marathoners,
who are more likely to experience pulmonary limitations to oxygen
availability. Pulmonary limitations have been observed in elite
athletes, and as many as 50% of highly-trained individuals
experience low oxygen concentrations in the blood (hypoxemia),
potentially due to diffusion limitation [27]. Whereas hypoxemia is
most frequently observed at exercise intensities near maximal
exertion, it is possible that endurance events such as a marathon
place constraints on the working lung muscles regardless of the
submaximal speed at which most runners compete. Indeed,
Amann et al. [28] found that pulmonary limitations were capable
of significantly decreasing performance in a 5 kilometer cycling
time trial. This effect is likely intensified during a marathon,
which, although run at a slower pace, is more than eight times as
long. Another study shows that seasoned runners experience
significant decreases in diffusing capacity following completion of a
marathon [29], suggesting an important role for the lungs in an
event that requires submaximal speed but maximal overall effort.
Based on these observations, the purpose of this study was to
compare the relation between DLNO and DLCO (indexed to
body surface area, BSA) and marathon running performance. It
was hypothesized that pulmonary diffusion limitation would exist
in faster runners, defined herein as those meeting qualification
standards for the Boston Marathon (Boston Qualifiers, BQ).
Qualification for this prestigious event is dependent on one’s
performance relative to age and gender standards, thus eliminating
the concerns associated with grouping subjects based on marathon
time alone (e.g. creating a younger or predominantly male group).
Furthermore, the Steamtown Marathon is a certified qualification
course and frequently qualifies over 20% of its participants for the
Boston event. This local event therefore provides the opportunity
to study differences between sub-elite and more recreational
marathoners using a less ambiguous distinction than running pace.
In analyzing these data, the relation between marathon
finishing time and DLNO or DLCO in both groups was examined
using a linear regression model. No significant relation between
these factors was expected in non-qualifiers (non-BQ), whereas
qualifiers (BQ) were expected to show a significant correlation
between finishing time and DLNO/DLCO. Additionally, it was
hypothesized that DLNO would show a stronger relation with
marathon finishing time than would DLCO, in agreement with
previous findings [21,30].
Methods
Twenty-eight endurance trained subjects (18 males, 10 females),
reported for preliminary testing 2 to 3 weeks before the 2011
Steamtown Marathon in Scranton, PA. Institutional Review
Board-approved informed written consent and a Physical Activity
Readiness Questionnaire were obtained from all subjects before
participation. Anthropometrics (body mass, height, BSA) and age
were obtained, and percent body fat (BF %) was measured using
Dual Energy X-ray Absorptiometry (DEXA) (Lunar ProdigyTM,
GE Medical Systems, Madison, WI).
Pulmonary function tests, consisting of spirometry maneuvers to
identify
obstructive
or
restrictive
patterns,
were
conducted
according to established guidelines [31]. The maneuver to
determine DLNO and DLCO was also performed according to
established guidelines [32], with a 5 to 6 second breath-hold [21].
Because this one-step maneuver allows simultaneous measurement
of DLNO and DLCO, pulmonary capillary blood volume (VC)
was then calculated as follows: DmCO was computed as DLNO 4
2.42. The 1/HCO was determined from Roughton and Forster
[33] as (0.73+0.058 ? PAO2) ? (14.6/[Hb]), where alveolar oxygen
pressure (PAO2) was 100 mmHg, and the hemoglobin concentra-
tion [Hb] was set as 14.6 g. dL21 for males, and 13.4 g. dL21 for
females [32]. As such, 1/HCO was 1.310 for males and 1.427 for
females. VC was then obtained by solving for it using the following
equation [33]:
1
DLCO ~
1
DmCO z
1
HCO:Vc
Reference equations were then used to compare each marathon
runner’s lung function to normative data from the standard
population [34,35].
After the lung function tests were completed, running economy
testing was performed on treadmill at a 2% grade to simulate
outdoor running conditions at three different sub-maximal
running speeds, each lasting 5 minutes. The treadmill was
calibrated before the first subject was tested. Throughout testing,
heart rate (HR) was measured using a Polar heart rate monitor
(Model S610, Polar Electro USA, Lake Success, NY). Metabolic
data were collected using breath-by-breath analysis (Sensormedics
Vmax 229D, Viasys, CA).
Assuming that steady state exercise had been achieved within
the first three minutes, means for metabolic data for the last two
minutes of each stage were computed. Running economy for each
stage was computed as the _VO2 required to travel one kilometer
and expressed as mL O2.min21.km21. Mean running economy for
the three speeds was then computed to serve as a measure of
overall running economy.
The final stage of the running economy test protocol was
followed immediately by a graded exercise test, in which treadmill
speed was increased by 0.5 mph every minute until volitional
exhaustion. Several cardiorespiratory parameters, including max-
imum respiratory exchange ratio (RERmax), maximum heart rate
DLNO Predicts Marathon Finishing Time
PLOS ONE | www.plosone.org
2
September 2012 | Volume 7 | Issue 9 | e44513
(HRmax), maximum expired ventilation ( _VEmax), and
_VO2max
were measured, and the treadmill speed at which _VO2max was
obtained was recorded (v _VO2max). If this speed was sustainable for
less than a full minute, the highest speed sustained for 60 seconds
was also recorded. Marathon-specific endurance (%V _VO2max) was
calculated by dividing the runner’s mean speed for the Steamtown
marathon by v _VO2max, with higher values indicating performance
at a greater relative physiologic intensity. Since _VO2 could not be
directly measured during the race, average _VO2 for the marathon
was calculated from the slope of the regression line between speed
and _VO2 for each subject, using the three speeds of the running
economy test. Average speed for the marathon was then entered
into each subject’s own equation to solve for oxygen consumption.
Univariate ANCOVAs or independent t-tests (with a confidence
interval of 95%) were used to determine whether differences in
finishing time, _VO2max (both absolute, in L.min21, and relative, in
mL.kg21.min21), and lung function parameters existed between
BQ and non-BQ groups. Age and gender served as covariates for
the
ANCOVA.
Step-wise
multiple
linear
regressions
were
conducted to determine variables most closely related to finishing
time for the entire sample, as well as for each group separately.
DLNO and DLCO normalized to body surface area (BSA) were
included in regression analyses to account for the effects of body
mass and height on lung size. Other predictor variables entered
into regression analysis include gender, age, body mass, body fat
percentage,
_VO2max (L O2. min21), running economy (mL
O2. kg21. km21), DLNO, DLCO, and marathon specific endur-
ance. To address the significance of diffusing capacity, the relation
between DLNO or DLCO and finishing time was further explored
in a bivariate regression analysis for both BQ and non-BQ groups.
The data were analyzed by SPSS Version 19.0, (SPSS Inc.,
Chicago, IL). Statistical significance was declared when p,0.05.
Results
A total of 392 Steamtown Marathon finishers (22% of marathon
participants) met qualification standards for the 2012 or 2013
Boston Marathon. Of the 28 subjects in this study, 10 runners (6
males, 4 females) (36%) qualified for Boston. Anthropometric
measurements (Table 1) indicated that significant differences exist
between BQ’s and non-BQ’s with respect to age and body fat
percentage (p,0.05).
The graded exercise test to volitional exhaustion lasted 10.1(1.2)
minutes including the final 5-min running economy bout. The
_VO2max in L O2. min21 from that test was not different between
groups; however, relative
_VO2max (mL O2. kg21. min21) was
significantly different between the two groups (p,0.01) such that
BQ’s had a mean (standard deviation)
_VO2max about 11(2.5)
mL O2. kg21. min21 greater than that of non-BQ’s. There was a
non-significant trend (p = 0.08) for body mass to differ between
groups. During the graded exercise test, BQ’s attained a maximal
treadmill speed 20% faster than non-qualifiers (p,0.05). Addi-
tionally, BQ’s completed the marathon at a higher percentage of
v _VO2max (7565%) than did non-BQ’s (6764%, p,0.001). For
both groups combined, there was a significant bivariate relation
between _VO2max (mL O2. kg21. min21) and marathon finishing
time (adjusted r2 = 0.47, SEE = 37.5 min, p,0.05).
Percent of predicted values for a lung function tests were not
significantly different between groups (Table 2). Three subjects in
each group had a DLNO greater than the upper limit of normal
(ULN). Two BQ’s had a DLCO above the ULN, while 5 non-
BQ’s had a DLCO that surpassed the ULN. Chi-square analysis
reveals that there is not a significant difference in the proportion of
subjects with an abnormally high diffusion capacity (DLCO or
DLNO) between groups (data not included).
Mean finishing time for all subjects was 220.0 minutes,
(range = 142.6 to 288.9). Weight loss from the marathon was
comparable between the two groups [1.0 (1.4) kg for BQ’s; 0.9
(0.8) kg for non-BQ’s, p = 0.746]. Ten of the twenty-eight
participants (6 males, 4 females) qualified for the Boston Marathon
(average
time = 180.0+23.1 min);
the
average
time
for
the
remaining 18 (12 males, 6 females) was 242.2+28.3 minutes.
Finishing time was significantly faster in BQ’s when controlling for
age and gender (p,0.01). Step-wise linear regression determined
that finishing time for all subjects was dependent on maximum
treadmill
speed
and
specific
endurance
(adjusted
r2 = 0.97,
SEE = 6.9 minutes p,0.05); however, 80% of the variance in
finishing time is accounted for by differences in maximum
treadmill
speed
sustained
for
60
seconds
alone
(adjusted
r2 = 0.80, SEE = 17.7 minutes, p,0.05).
In BQ’s, the strongest relation identified was between finishing
time and DLNO normalized to BSA. For non-BQ’s, finishing time
was best predicted by maximum treadmill velocity sustained for 60
seconds and specific endurance, where 74% of the variance in
finishing time is accounted for by differences in maximum
treadmill velocity. The relation between DLNO normalized to
BSA and finishing time was not significant for non-BQ’s
(p = 0.127). A significant difference was found between the
correlation coefficients of the two linear regressions (two-tailed
z = 22.15, p = 0.03). When regression lines of DLNO normalized
to BSA versus finishing time plotted on the same axes, the
regressions intersect at a point corresponding to a finishing time of
178.1 minutes (Figure 1), suggesting that the relation between
DLNO normalized to BSA and marathon time begins to change
around 237 m. min21 pace (6 minutes, 47 seconds per mile).
DLCO normalized to BSA was also correlated to finishing time in
BQ subjects, to a lesser extent. For non-BQ, no significant relation
between the variables is evident (p = 0.46). These regressions
intersect at the point corresponding to 184.3 minutes (Figure 2).
Discussion
The novelty of this study lies in that it shows that the relation
between pulmonary factors (measured at rest) and marathon
performance may differ between athletes of different skill level. In
particular, this study was able to isolate an approximate time point
at which the relation between pulmonary diffusing capacity for
nitric oxide and marathon finishing time changes for trained
endurance athletes, pinpointing a pace at which lung function
becomes limiting to performance. Whereas runners qualifying for
the Boston Marathon, because of their overall faster pace, are
limited by DLNO, non-qualifiers probably experience a more
mechanical limitation, such as leg turnover (related to maximum
treadmill velocity).
The primary purpose of this study was to compare the
correlation of DLNO and DLCO to marathon running perfor-
mance. The results demonstrate that there was a significant slope
(indicating a strong correlation) between DLNO and DLCO
(normalized to BSA) versus marathon finishing time only in
runners that qualified for the Boston Marathon, with these
variables showing a stronger predictive relation to finishing time
than either _VO2max or running economy. DLNO was shown to be
the
strongest
predictor
of finishing
time,
such
that
every
1 mL. min21. mmHg21.m22 increase in DLNO at rest projects
that finishing time will decrease by about 1.4 minutes (with a range
of 0.8 to 2.1 min). These results strongly suggest that alveolar-
capillary membrane conductance may be performance-limiting in
DLNO Predicts Marathon Finishing Time
PLOS ONE | www.plosone.org
3
September 2012 | Volume 7 | Issue 9 | e44513
runners that complete a marathon in 3 hours or faster, as shown
by the intersection of regression lines for BQ and non-BQ groups
(Figure 1 and 2). These figures also demonstrate that DLNO
relative to BSA is a more accurate predictor of finishing time than
DLCO, as the former correlation shows a larger adjusted r2 and a
lower standard error of the estimate.
The
physiological
mechanism
closely
relating
DLNO
to
marathon performance in BQ’s is speculative, given that these
subjects ran at approximately 75% of _VO2max, a value consistent
elsewhere in those with similar running abilities [10,13]. Though
arterial oxygen pressure and the alveolar-to-arterial oxygen
pressure difference was not measured throughout the race in this
study, others have shown hypoxemia is not induced in fit athletes
running half of a full marathon at ,75%
_VO2max [36].
Nonetheless, we suggest that, in well-trained runners, there is a
_VO2 threshold at which pulmonary diffusion limits oxygen
consumption. In other words, these individuals run the marathon
at a speed at which _VO2 is high enough that gas diffusion at the
alveolar-capillary membrane becomes a physiological bottleneck,
and those with greater alveolar-capillary membrane conductance
are able to maintain greater arterial oxygen saturation. Similarly,
non-BQ’s likely complete the marathon at a
_VO2 at which
pulmonary diffusion is not limiting; this may explain the lack of
relation between DLNO and performance in this group. Thus,
individuals
who
have
superior
alveolar-capillary
membrane
conductance (high DLNO measurements), and yet do not reach
a ‘‘heart’’ or ‘‘muscle limitation,’’ (i.e., non-BQ’s) would not have
Table 1. Anthropometric Data for Boston Qualifiers and non-Qualifiers.
BQ (n = 10)
Non-BQ (n = 18)
Total (n = 28)
ANTHROPOMETRICS
Age (yr)*
33 (9)
40 (7)
37 (9)
22–50
29–52
22–52
Weight (kg)
64.1 (11.1)
73.2(13.3)
69.9 (13.2)
42.2–84.5
45.0–103.0
42.2–103.0
Height (cm)
170.7 (9.0)
174.0 (9.2)
172.8 (9.1)
154.0–186.7
156.0–187.0
154.0–187.0
Body Fat (%)*
13.1 (7.12)
19.1 (6.7)
17.0 (7.3)
5.1–24.1
9.0–34.5
5.1–34.5
CARDIOPULMONARY VARIABLES AT MAXIMAL EXERCISE
_VO2max (L/min)
3.75 (0.64)
3.56 (0.64)
3.63 (0.64)
2.71–4.75
2.25–4.52
2.25–4.75
_VO2max (mL/kg/min)*
59.4 (8.3)
48.7 (5.0)
52.5 (8.1)
49.1–73.1
38.3–60.4
38.3–73.1
RERmax
1.17 (0.07)
1.16 (0.06)
1.16 (0.06)
1.06–1.30
1.06–1.28
1.06–1.30
VEmax (L/min)
119.91 (21.10)
112.03 (18.46)
114.85 (19.43)
82.6–141.13
81.00–149.30
81.00–149.30
HRmax (bpm)
187 (12)
178 (12)
181 (12)
169–202
152–198
152–202
RUNNING PERFORMANCE
Running Economy (mL/kg/km){
194.5 (13.0)
205.8 (19.1)
202.0 (18.1)
180.8–224.5
172.9–247.6
172.9–247.6
Running Economy (mL/kg/min)
43.2 (5.9)
36.3 (3.7)
38.8 (5.6)
36.0–52.0
30.0–44.2
30.0–52.0
Maximum Treadmill Speed for 60 seconds (m/min)*
316 (34)
266 (31)
284 (40)
271–362
228–316
228–362
Marathon Finishing Time (min)
180.0 (23.1)
242.2 (28.3)
220.0 (40.0)
142.0–203.0
200.4–289.0
142.0–289.0
% _VO2max for Marathon
76.4 (6.9)
74.8 (7.2)
75.3 (7.0)
63.0–82.4
63.0–88.3
63.0–88.3
Specific Endurance (% _V VO2max)*
75.31 (5.44)
66.47 (3.97)
69.62 (6.20)
63.16–81.72
56.26–72.04
56.26–81.72
Data are reported as mean (SD) values and range.
*denotes significant difference (p,0.05) between BQ and non-BQ subjects. Controlling for age and gender using an ANCOVA did not affect the outcome of statistical
analyses.
{Running Economy calculated at average speed for group; BQ = 222.2 (30.0) m/min, non-BQ = 177.2 (20.4) m/min.
doi:10.1371/journal.pone.0044513.t001
DLNO Predicts Marathon Finishing Time
PLOS ONE | www.plosone.org
4
September 2012 | Volume 7 | Issue 9 | e44513
any performance advantage over other individuals who have lower
alveolar-capillary membrane conductance. In fact, studies have
demonstrated that marathon running causes a significant drop in
pulmonary diffusing capacity [29,37]. About 30% of the drop in
DLNO (normalized to BSA) post-exercise is accounted for by
marathon finishing time (p = 0.046) [29]. Thus, for every 1 minute
improvement
in
marathon
time,
DLNO
is
reduced
by
1.2 mL.min21.mmHg21.m22 [29]. The diminished DLNO with
marathon running can be expected, since _VO2max accounts for
about 40 to 77% of the variance in marathon performance
[5,8,9,10,11] and about 40% of the change in DLNO from pre- to
post-exercise [38]. Therefore, it is possible that individuals with
larger alveolar-capillary membrane conductance at the start of the
marathon have a physiological advantage: diffusion impairments
that may arise during the race will likely not decrease diffusion
capacity to problematic levels.
Only one other study to date has examined the association
between pulmonary diffusion and marathon finishing time using
both DLNO and DLCO as predictors [29]. While Manier and
colleagues did not intend to examine this association, the data set
was available in the publication. In BQ runners (n = 9, mean
finishing time = 177.0615.0 min), the relation between DLNO
indexed to BSA and finishing time was present, but it was
not as strong as in the current study (adjusted r2 = 0.30,
SEE = 12.5 min,
p = 0.073).
In
Manier’s
study,
for
every
1 mL.min21.mmHg21.m22 increase in DLNO at rest, marathon
finishing time was 0.8 minutes faster (ranging from 1.8 minutes
faster to 0.1 minute slower) [29]. Combining these data with those
from the present study suggests that 30 to 74% of the variance in
DLNO (mL. min21. mmHg21.m22) at rest is related to marathon
finishing time in BQ’s, while no such relation exists in non-BQ’s.
Additionally, controlling for age and gender does not affect this
correlation in either study separately or collectively [29]. Combin-
ing the data from these independent studies further supports that
pulmonary diffusing capacity is an important contributor to
marathon performance in well-trained runners. It is also important
to note that while the BQ group represents a well-trained
population and some of the subjects in this study performed at a
very high level, international class runners tend to have even higher
values of _VO2max and, possibly, an even greater dependence on
alveolar-capillary membrane conductance.
Although DLNO normalized to BSA was only related to
finishing time in BQ’s, DLNO did not significantly differ between
BQ and non-BQ whether indexed to BSA (p = 0.078) or not
(p = 0.80). DLNO is usually higher in fit subjects [21] and in the
present study, DLNO was significantly higher than predicted
whether using norms from Zavorsky and colleagues (113%
predicted, p,0.01) [34] or Aguilaniu and colleagues (107%
predicted, p = 0.011) [39]. Several subjects in both BQ and non-
BQ groups had values above the upper limit of normative data for
a variety of pulmonary function parameters. As such, we can
conclude that BQ and non-BQ were of comparable respiratory
fitness. These findings suggest that endurance training itself may
Table 2. Pulmonary function measurements for Boston Qualifiers and non-Qualifiers.
BQ (n = 10)
Non-BQ (n = 18)
Total (n = 28)
Mean
Percent Predicted
Mean
Percent Predicted
Mean
Percent Predicted
FVC (L)
5.1 (0.9)
110 (8)*
5.0 (1.0)
105 (14)
5.0 (0.9)
106 (13)*
3.8–6.4
96–124
3.4–7.2
79–134
3.4–7.2
79–134
FEV1 (L)
4.0 (0.7)
108 (11)*
3.8 (0.7)
100 (11)
3.9 (0.7)
103 (11)
2.9–5.4
96–128
2.5–5.1
86–129
2.5–5.4
86–129
FEV1/FVC
81.2 (10.6)
99 (8)
77.2 (6.3)
96 (9)
78.6 (6.2)
97 (8)
73.4–90.2
89–113
64.9–87.2
77–110
64.9–90.2
77–113
PEF (L)
9.0 (2.2)
102 (11)
10.1 (2.0)
111 (10)*
9.7 (2.1)
108 (11)*
6.0–12.9
81–118
7.0–13.1
97–137
6.0–13.1
81–137
FEF25–75 (L/s)
5.2(1.5)
133 (32)*
4.8 (1.3)
129 (29)*
5.0 (1.3)
131 (29)*
3.2–8.3
91–192
2.7–6.8
84–176
2.7–8.3
84–192
DLCO
34.7 (5.9)
113 (15)*
34.3 (6.6)
114 (14)*
34.3 (6.73)
113 (14)*
25.9–43.4
91–135
22.6–48.6
93–146
22.6–48.6
91–146
DLCO/BSA
20.0 (3.1)
–
18.3 (2.6)
–
18.9 (2.8)
–
15.5–23.9
–
13.6–23.6
–
13.6–23.9
–
DLNO
179 (27)
113 (12)*
176 (34)
113 (13)*
175 (35)
113 (13)*
140–212
97–130
124–256
94–149
120–256
94–149
DLNO/BSA
103 (14)
–
93 (13)
–
97 (14)
–
85–123
–
72–124
–
72–124
–
VC (mL)
90 (15)
116 (19)*
89 (13)
117 (15)*
90 (14)
117 (16)*
72–113
95–151
64–118
95–144
64–118
95–151
FVC: forced vital capacity; FEV1: forced expiratory volume within 1 sec; FEV1/FVC: fraction of inspired air expired within 1 sec; PEF: peak expiratory flow; FEF25–75: forced
expiratory flow during 25–75% of 6-second exhale; DLCO: pulmonary diffusing capacity for carbon monoxide, in mL/min/mmHg; DLCO/BSA: DLCO relative to body
surface area, in mL/min/mmHg/m2; DLNO: pulmonary diffusing capacity for nitric oxide, in mL/min/mmHg; DLNO/BSA: DLNO relative to body surface area, in mL/min/
mmHg/m2; VC: pulmonary capillary blood volume.
Data are reported as mean (SD) values and range.
*Denotes a significant difference in observed parameter relative to predicted (p,0.05).
doi:10.1371/journal.pone.0044513.t002
DLNO Predicts Marathon Finishing Time
PLOS ONE | www.plosone.org
5
September 2012 | Volume 7 | Issue 9 | e44513
improve alveolar-capillary membrane conductance above that of
untrained individuals, but improvements in DLNO likely plateau
well before that of the heart or skeletal muscle.
Generally speaking, the lungs become limiting at a running pace
of 6:47 minutes per mile or 236 m.min21 (a 3 hour marathon).
Therefore, athletes performing at or around this pace should be
aware of the potential significance of this limitation and its
ramifications for performance. It is unknown whether any specific
training practices can be implemented to improve pulmonary
diffusion and therefore improve marathon performance.
This study is limited by its small sample size; a higher power to
detect differences in group means would likely be achieved by
recruiting more participants. Nevertheless, it is frowned upon to
conduct a post-hoc power analysis after data collection has occurred
[40,41], thus we did not perform one. Instead, confidence intervals
replace power calculations after a study is completed [40,41]. We
have provided confidence intervals in Figures 1 and 2. Addition-
ally, small group sizes result in large variances; as such, the
standard errors of the estimate for both regressions are large and
overlapping, obscuring estimation of a clear range of intersection
at which DLNO indexed to BSA begins to predict finishing time.
Larger sample sizes would also allow us to divide runners into
more specific categories by time, possibly delineating a clearer
relation between DLNO indexed to BSA and finishing time with
increasing running speed. Additionally, normalization to BSA
transforms DLNO into a variable with multiple units, possibly
complicating analysis. Mean height and body mass themselves
were not significantly different between groups; however, the
combination of these variables appears to be important when
related to pulmonary variables. It remains possible that BSA
introduces variation in the data set due to its relationship with heat
dissipation, largely dependent on stature [42]. This might have
impacted finishing time, especially in slower runners finishing as
the ambient temperature increased on race day, from about 8uC
(96% humidity) at the 8:00 AM start to 21uC (51% humidity) at
the finish line by 1:00 PM.
As pulmonary diffusion has been shown to decrease during long-
duration submaximal exercise [29,37], the efficacy of interventions
which may counter these negative effects should be explored (e.g.,
anti-inflammatory drugs or antioxidants). Our results could be
strengthened by measuring DLNO immediately after completion of
the marathon, with a greater change in DLNO representative of
presence of a limitation. Nevertheless, measurement of DLNO at
rest may still underestimate severity of diffusion limitation. As such,
future studies could also measure DLNO at submaximal speeds
during running economy testing and extrapolate these data to
marathon race pace, allowing a more accurate estimation of the
impact of diffusing capacity on running performance.
The genetic basis of DLNO could be further studied to
determine how this parameter might change as one adapts to
training. Furthermore, longitudinal and interventional studies are
recommended to determine if any specific type of training can
optimize pulmonary diffusion capacity and therefore improve
endurance running performance. More extensive understanding of
the relationship outlined in this study will allow us to confirm the
validity of diffusing capacity for nitric oxide as a fitness predictor.
In conclusion, this study found that DLNO indexed to BSA is a
better predictor of marathon finishing time in runners qualifying for
the Boston Marathon than are more commonly used variables, such
as _VO2max or running economy, but this relation was not observed
for non-BQ’s. This suggests that alveolar-capillary membrane
conductance can be pulmonary limitation in well-trained runners.
Acknowledgments
The authors would like to thank the Steamtown Marathon Race
Committee, including Assistant Race Director Jim Cummings, for help
in recruiting participants for this study. Additionally, we acknowledge the
Steamtown Marathon Medical Director, Tim Rowland, MD, for providing
available space at the medical triage area at the finish line.
Figure 1. Regression showing relation between DLNO/BSA and
marathon finishing time. For Boston Qualifiers (solid line), r2 = 0.74,
SEE = 11.8, p,0.01, showing a significant correlation. Y-intercept is
198.8+18.7 (95% confidence interval ranges from 155.7 to 242.0); slope
of the line is 20.532+0.103, with a 95% confidence interval range of
20.8 to 20.3. For non-qualifiers (dashed line), p = 0.14. Y-intercept is
135.7+26.5 (95% confidence interval ranges from 79.5 to 191.8); the
slope of the line is 20.175+0.109, with a confidence interval range of
20.4 to 0.05. The point of intersection for these lines is 178.07 minutes
(2:58:11). There is a significant difference between the correlation
coefficients of the two regressions (2-tailed z = 22.15, p = 0.03).
doi:10.1371/journal.pone.0044513.g001
Figure 2. Regression showing relation between DLCO/BSA and
marathon finishing time. For Boston Qualifiers (solid line), r2 = 0.69,
SEE = 12.8, p,0.01, showing a significant correlation. Y-intercept of the
line is 40.6+4.5, with a 95% confidence interval range of 30.2 to 51.0.
The slope of the line is 20.11+0.03, with a confidence interval of 20.18
to 20.06. For non-qualifiers (dashed line), p = 0.172. Y-intercept is
25.7+5.2, confidence interval ranges from 14.6 to 36.8. The slope of the
line is 20.03+0.02, with a 95% confidence interval ranging from 20.08
to 0.02. The point of intersection for these lines is 179.22 minutes
(2:59:14). The correlation coefficients of these lines are significantly
different (2-tailed z = 21.99, p = 0.046).
doi:10.1371/journal.pone.0044513.g002
DLNO Predicts Marathon Finishing Time
PLOS ONE | www.plosone.org
6
September 2012 | Volume 7 | Issue 9 | e44513
Author Contributions
Conceived and designed the experiments: GSZ JMS. Performed the
experiments: KML AMS KAU. Analyzed the data: KML GSZ.
Contributed reagents/materials/analysis tools: GSZ. Wrote the paper:
KML GSZ AMS.
References
1. Rapoport BI (2010) Metabolic factors limiting performance in marathon
runners. PLoS Comput Biol 6: e1000960.
2. Coyle EF (2007) Physiological regulation of marathon performance. Sports Med
37: 306–311.
3. Joyner MJ, Ruiz JR, Lucia A (2011) The two-hour marathon: who and when?
J Appl Physiol 110: 275–277.
4. La Torre A, Vernillo G, Agnello L, Berardelli C, Rampinini E (2011) Is it time to
consider a new performance classification for high-level male marathon runners?
J Strength Cond Res 25: 3242–3247.
5. Billat VL, Demarle A, Slawinski J, Paiva M, Koralsztein JP (2001) Physical and
training characteristics of top-class marathon runners. Med Sci Sports Exerc 33:
2089–2097.
6. Sjodin B, Svedenhag J (1985) Applied physiology of marathon running. Sports
Med 2: 83–99.
7. Stellingwerff T, Jeukendrup AE, Perrey S, Shephard RJ, Schubert MM, et al.
(2011) Commentaries on Viewpoint: The two-hour marathon: Who and when?
J Appl Physiol 110: 278–293.
8. Hagan RD, Upton SJ, Duncan JJ, Gettman LR (1987) Marathon performance
in relation to maximal aerobic power and training indices in female distance
runners. Br J Sports Med 21: 3–7.
9. Florence S, Weir JP (1997) Relationship of critical velocity to marathon running
performance. Eur J Appl Physiol Occup Physiol 75: 274–278.
10. Maughan RJ, Leiper JB (1983) Aerobic capacity and fractional utilisation of
aerobic capacity in elite and non-elite male and female marathon runners.
Eur J Appl Physiol Occup Physiol 52: 80–87.
11. Loftin M, Sothern M, Koss C, Tuuri G, Vanvrancken C, et al. (2007) Energy
expenditure and influence of physiologic factors during marathon running.
J Strength Cond Res 21: 1188–1191.
12. Millet GY, Banfi JC, Kerherve H, Morin JB, Vincent L, et al. (2011)
Physiological and biological factors associated with a 24 h treadmill ultra-
marathon performance. Scand J Med Sci Sports 21: 54–61.
13. Maron MB, Horvath SM, Wilkerson JE, Gliner JA (1976) Oxygen uptake
measurements during competitive marathon running. J Appl Physiol 40: 836–
838.
14. ATS/ACCP (2003) ATS/ACCP Statement on cardiopulmonary exercise
testing. Am J Respir Crit Care Med 167: 211–277.
15. Dempsey JA, Fregosi RF (1985) Adaptability of the pulmonary system to
changing metabolic requirements. Am J Cardiol 55: 59D–67D.
16. Wagner PD (2005) Why doesn’t exercise grow the lungs when other factors do?
Exerc Sport Sci Rev 33: 3–8.
17. Dempsey JA, Hanson PG, Henderson KS (1984) Exercise-induced arterial
hypoxaemia in healthy human subjects at sea level. Journal of Physiology,
London 355: 161–175.
18. Harms CA, McClaran SR, Nickele GA, Pegelow DF, Nelson WB, et al. (2000)
Effect of exercise-induced arterial O2 desaturation on VO2max in women. Med
Sci Sports Exerc 32: 1101–1108.
19. Romer LM, Haverkamp HC, Lovering AT, Pegelow DF, Dempsey JA (2006)
Effect of exercise-induced arterial hypoxemia on quadriceps muscle fatigue in
healthy humans. Am J Physiol Regul Integr Comp Physiol 290: R365–375.
20. Koskolou MD, McKenzie DC (1994) Arterial hypoxemia and performance
during intense exercise. Eur J Appl Physiol 68: 80–86.
21. Zavorsky GS, Wilson B, Harris JK, Kim DJ, Carli F, et al. (2010) Pulmonary
diffusion and aerobic capacity: Is there a relation? Does obesity matter? Acta
Physiol (Oxf) 198: 499–507.
22. de Bisschop C, Martinot JB, Leurquin-Sterk G, Faoro V, Guenard H, et al.
(2012) Improvement in lung diffusion by endothelin A receptor blockade at high
altitude. J Appl Physiol 112: 20–25.
23. Guazzi M, Pontone G, Brambilla R, Agostoni P, Reina G (2002) Alveolar–
capillary membrane gas conductance: a novel prognostic indicator in chronic
heart failure. Eur Heart J 23: 467–476.
24. Tamhane RM, Johnson RL, Jr., Hsia CC (2001) Pulmonary membrane diffusing
capacity and capillary blood volume measured during exercise from nitric oxide
uptake. Chest 120: 1850–1856.
25. Ceridon ML, Beck KC, Olson TP, Bilezikian JA, Johnson BD (2010) Calculating
alveolar capillary conductance and pulmonary capillary blood volume:
comparing the multiple- and single-inspired oxygen tension methods. J Appl
Physiol 109: 643–653.
26. Borland CD, Higenbottam TW (1989) A simultaneous single breath measure-
ment of pulmonary diffusing capacity with nitric oxide and carbon monoxide.
Eur Respir J 2: 56–63.
27. Powers SK, Dodd S, Lawler J, Landry G, Kirtley M, et al. (1988) Incidence of
exercise induced hypoxemia in elite endurance athletes at sea level. Eur J Appl
Physiol Occup Physiol 58: 298–302.
28. Amann M (2012) Pulmonary system limitations to endurance exercise
performance in humans. Exp Physiol 97: 311–318.
29. Manier G, Moinard J, Techoueyres P, Varene N, Guenard H (1991) Pulmonary
diffusion limitation after prolonged strenuous exercise. Respir Physiol 83: 143–
153.
30. Puri S, Baker BL, Dutka DP, Oakley CM, Hughes JM, et al. (1995) Reduced
alveolar-capillary membrane diffusing capacity in chronic heart failure. Its
pathophysiological relevance and relationship to exercise performance. Circu-
lation 91: 2769–2774.
31. Miller MR, Hankinson J, Brusasco V, Burgos F, Casaburi R, et al. (2005)
Standardisation of spirometry. Eur Respir J 26: 319–338.
32. Macintyre N, Crapo RO, Viegi G, Johnson DC, van der Grinten CP, et al.
(2005) Standardisation of the single-breath determination of carbon monoxide
uptake in the lung. Eur Respir J 26: 720–735.
33. Roughton FJ, Forster RE (1957) Relative importance of diffusion and chemical
reaction rates in determining rate of exchange of gases in the human lung, with
special reference to true diffusing capacity of pulmonary membrane and volume
of blood in the lung capillaries. J Appl Physiol 11: 290–302.
34. Zavorsky GS, Cao J, Murias JM (2008) Reference values of pulmonary diffusing
capacity for nitric oxide in an adult population. Nitric Oxide 18: 70–79.
35. Hankinson JL, Odencrantz JR, Fedan KB (1999) Spirometric reference values
from a sample of the general U.S. population. Am J Respir Crit Care Med 159:
179–187.
36. Hanson P, Claremont A, Dempsey J, Reddan W (1982) Determinants and
consequences of ventilatory responses to competitive endurance running. J Appl
Physiol 52: 615–623.
37. Miles DS, Doerr CE, Schonfeld SA, Sinks DE, Gotshall RW (1983) Changes in
pulmonary diffusing capacity and closing volume after running a marathon.
Respir Physiol 52: 349–359.
38. Zavorsky GS, Lands LC (2005) Lung diffusion capacity for nitric oxide and
carbon monoxide is impaired similarly following short-term graded exercise.
Nitric Oxide 12: 31–38.
39. Aguilaniu B, Maitre J, Glenet S, Gegout-Petit A, Guenard H (2008) European
reference equations for CO and NO lung transfer. Eur Respir J 31: 1091–1097.
40. Hoenig JM, Heisey DM (2001) The abuse of power: The pervasive fallacy of
power calculations for data analysis. American Statistician 55: 19–24.
41. Wilkinson L, Inference TFS (1999) Statistical methods in psychology journals -
Guidelines and explanations. American Psychologist 54: 594–604.
42. Dennis SC, Noakes TD (1999) Advantages of a smaller bodymass in humans
when distance-running in warm, humid conditions. Eur J Appl Physiol Occup
Physiol 79: 280–284.
DLNO Predicts Marathon Finishing Time
PLOS ONE | www.plosone.org
7
September 2012 | Volume 7 | Issue 9 | e44513
| Alveolar-membrane diffusing capacity limits performance in Boston marathon qualifiers. | 09-11-2012 | Lavin, Kaleen M,Straub, Allison M,Uhranowsky, Kathleen A,Smoliga, James M,Zavorsky, Gerald S | eng |
PMC6928244 | Physiological Reports. 2019;7:e14313.
| 1 of 7
https://doi.org/10.14814/phy2.14313
wileyonlinelibrary.com/journal/phy2
1 |
INTRODUCTION
In 2007 the National Institutes of Health launched the Human
Microbiome Project (HMP), an interdisciplinary research
initiative seeking to characterize the contribution of human
gut microbiota to health and disease (Turnbaugh et al.,
2007). Subsequent findings have demonstrated compelling
relationships between human gut microbiome composition
and many leading causes of death worldwide including car-
diovascular disease (Wang et al., 2011), diabetes (Larsen et
al., 2010), and cancer (Ahn et al., 2013). Although the gut
microbiome is suggested to exhibit exceptional plasticity
(Gomez et al., 2019), a detailed understanding of the fac-
tors determining human microbiome assembly is lacking
(Relman, 2015).
DOI: 10.14814/phy2.14313
O R I G I N A L R E S E A R C H
Rapid gut microbiome changes in a world-class ultramarathon
runner
Gregory J. Grosicki1
| Ryan P. Durk2 | James R. Bagley2
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original
work is properly cited.
© 2019 The Authors. Physiological Reports published by Wiley Periodicals, Inc. on behalf of The Physiological Society and the American Physiological Society.
1Biodynamics and Human Performance
Center, Georgia Southern University,
Savannah, Georgia
2San Francisco State University, San
Francisco, California
Correspondence
Gregory J. Grosicki, Biodynamics and
Human Performance Center, Georgia
Southern University (Armstrong Campus),
11935 Abercorn Street, Savannah, GA,
31419.
Email: [email protected]
Funding information
Rossi Family Foundation
Abstract
The human gut microbiome is a dynamic ecosystem with prolific health connotations.
Physical activity is emerging as a potent regulator of human microbiome composition.
This study examined changes in the gut microbiome of a world-class ultramarathon
runner before and after competing in the Western States Endurance Run (WSER), a
163 km mountain footrace. Anthropometrics and body composition were assessed
and the ultramarathoner's submaximal and maximal performance profiles were eval-
uated. Gut microbiome analyses were performed at four time-points: 21 weeks and
2 weeks before and 2 hours and 10 days after WSER. Aerobic power (VO2max)
was 4.24 L/min (66.7 ml kg−1 min−1), and running economy (51.1 ml kg−1 min−1
at 268 m/min) and lactate threshold (~83% VO2max) values were comparable to
that of highly trained distance runners. Two hours post-race, considerable changes
in the ultrarunners’ gut microbiome were observed. Alpha diversity (Shannon
Diversity Index) increased from 2.73 to 2.80 and phylum-level bacterial composition
(Firmicutes/Bacteroidetes ratio) rose from 4.4 to 14.2. Underlying these macro-level
microbial alterations were demonstrable increases in select bacterial genera such as
Veillonella (+14,229%) and Streptococcus (+438%) concomitant with reductions in
Alloprevotella (−79%) and Subdolingranulum (−50%). To our knowledge, this case
study shows the most rapid and pronounced shifts in human gut microbiome compo-
sition after acute exercise in the human literature. These findings provide yet another
example of how exercise can be a powerful modulator of human health.
K E Y W O R D S
endurance exercise, gut microbiota, ultramarathon, Veillonella
2 of 7 |
GROSICKI et al.
Recently, our group (Durk et al., 2018) and others (Allen
et al., 2018; Keohane et al., 2019; Scheiman et al., 2019) have
shown a link between physical activity and human microbi-
ome composition, aiding in the delineation of a signature mi-
crobiome response to exercise training. Increases in bacterial
diversity and a proliferation of taxa responsible for the pro-
duction of short chain fatty acids, such as butyrate, are among
the most pervasively observed microbial alterations with exer-
cise (Mailing, Allen, Buford, Fields, & Woods, 2019). While
these changes are generally regarded as beneficial to the host,
a comprehensive understanding of exercise training-induced
microbial modifications and their systemic physiological
implications remains to be delineated. Moreover, even less
is known regarding the human gut microbiome response to
an acute exercise bout. Recently, we had the unique oppor-
tunity to track the gut microbiome of a world-class ultra-
marathon runner who finished in the top-10 at the Western
States Endurance Run, a 163 km mountain footrace featuring
~5,486 m climbing and ~7,010 m of descent. Substantially re-
duced splanchnic blood flow over the course of the ~16-hr
event in concert with tremendous energetic demands (13,000–
16,000 kcal) (Cuddy, Slivka, Hailes, & Dumke, 2009) pro-
vides a unique lens through which to study the plasticity of our
microbial inhabitants under extreme conditions.
2 |
METHODS
2.1 | Participant
The participant was a 32 yr old male world-class ultramara-
thon runner, studied over the course of a 21-week period dur-
ing the 2019 race season. He began running competitively
in high school, and in college he was a two-time NCAA
Division II Cross-Country All-American. He since raced in
the Olympic marathon trials and is a two-time champion at
the Javelina Jundred.
2.2 | Approval and screening
This project was approved by the Institutional Review Board
at Georgia Southern University. On the day of the first study
visit, project objectives and risks and procedures were ver-
bally explained to the participant, after which the participant
provided written informed consent to participate. A visual rep-
resentation of the study methodology is provided in Figure 1.
2.3 | Training quantification
Daily training logs recorded via a Suunto Ambit3 Peak GPS
watch provided by the ultramarathoner were utilized to
quantify weekly and cumulative training duration, mileage,
and pace.
2.4 | Body composition and bone
mineral content
Height and body mass were measured on a wall-mounted
stadiometer and calibrated digital scale, respectively. Body
composition was assessed at 21 and 5 weeks prior to the
Western States Endurance Run (WSER) (Figure 1) using
dual-energy X-ray absorptiometry (DEXA) (Lunar iDXA,
GE Healthcare). The DEXA machine was outfitted with
enCORE version 16 and the machine was calibrated im-
mediately prior to each scan per manufacturer instructions
(laboratory coefficient of variation < 0.07%). Scans were
obtained after an overnight fast (>10 hr) and after voiding
of the bladder using “standard thickness” mode based on the
subject's size characteristics.
2.5 | Submaximal and maximal oxygen
consumption (VO2max) exercise testing
Submaximal exercise testing was performed on a slat
belt treadmill (4Front, Woodway) 30 min after ingesting a
150 kcal (23 g carbohydrate, 6 g fat, 1 g protein) solid snack
(GU Energy Labs). Testing was performed in a tempera-
ture-controlled environment (22°C, 63% relative humid-
ity and 766.1 mmHg) using a protocol from McMiken and
Daniels (1976) to facilitate running economy comparisons to
a well-characterized cohort of high-caliber distance runners
(Conley & Krahenbuhl, 1980; Costill, Thomason, & Roberts,
1973). Testing stages were 6 min in duration and expired gas
was monitored throughout the test via indirect calorimetry
(TrueOne2400, Parvo Medics) while heart rate was monitored
via wireless telemetry (Polar H10, Polar USA). At the com-
pletion of each testing stage, blood lactate was evaluated via
finger-stick (Lactate Scout+, EKF Diagnostics). Submaximal
testing was performed at a 0% grade using incremental steps
in velocity (80, 134, 215, 268, 295 and 311 m/min) and pro-
ceeded until blood lactate exceeded 4 mmol/l. Gas values
(e.g., O2 consumption, respiratory exchange ratio, etc.) were
quantified using the average of two 30 s values during the last
minute of each testing stage and rating of perceived exertion
(RPE, Borg scale) (Borg, 1970) was assessed in the last 15 s
of each stage.
Based on a recent report of performance testing in a
world-class cyclist (Bell, Furber, Someren, Anton-Solanas,
& Swart, 2017), a 15 min rest period was provided between
submaximal and maximal (VO2max) treadmill exercise test-
ing. The VO2max test was administered in 2-min stages at 0%
grade, beginning at 80 m/min and working up to the velocity
|
3 of 7
GROSICKI et al.
at lactate threshold (295 m/min). After 2 min at this velocity,
the grade was increased to 4% and then another 2% for every
2 min thereafter until volitional exhaustion. Oxygen consump-
tion, heart rate and RPE were monitored using the same equip-
ment and testing procedures as described above. Attainment
of VO2max was verified by (a) achievement of greater than
or equal to 90% age-predicted maximal heart rate, (b) a respi-
ratory exchange ratio (RER)> 1.1, and (c) a final RPE > 17.
2.6 | Gut microbiome analyses
Gut microbiome analyses were performed at four time-points
(21 and 2 weeks pre-WSER, 2 hours and 10 days post-WSER),
as shown in Figure 1. Stool samples were self-collected by the
ultramarathoner using a commercially available kit (Ubiome
Explorer) in accordance with the specifications laid out by
the NIH Human Microbiome Project (McInnes, 2010). All
samples, besides the 2 hr post sample, were taken at ap-
proximately the same time of day before eating or exercis-
ing (0800). Following a bowel movement, a sterile swab was
used to transfer a small amount of fecal matter into a vial con-
taining a lysis and stabilization buffer that preserves the ge-
netic material for transport at ambient temperatures. Samples
were sent to Ubiome laboratories (Ubiome) (Bik et al., 2018)
and lysed by bead-beating prior to DNA extraction in a class
1,000 clean room using a guanidine thiocyanate silica col-
umn-based purification method via a liquid-handling robot.
PCR amplification of the 16S rRNA genes was performed
with primers containing universal primers amplifying the V4
variables region (515F: GTGCCAGCMGCCGCGGTAA and
806R: GGACTACHVGGGTWTCTAAT) (Caporaso et al.,
2011). In addition, the primers contained Illumina tags and
barcodes. Samples were barcoded with a unique combination
of forward and reverse indexes allowing for simultaneous
processing of multiple samples. PCR products were pooled,
column-purified, and size selected through microfluidic DNA
fractionation (Minalla & Dubrow, 2001). Consolidated librar-
ies were quantified using quantitative real-time PCR using the
Kapa iCycler qPCR kit (Bio-Rad) on a BioRad Myio before
loading into the sequencer. Sequencing was performed in a
pair-end modality on a NextSeq 500 platform (Illumina) ren-
dering 2 × 150 bp pair-end sequences. These DNA sequencing
techniques were then used to generate data outputs (.csv file)
that provided a comprehensive bacterial taxonomic profile.
Shannon diversity index (i.e., alpha diversity) was computed
using PAST: Paleontological statistics software package
for education and data analysis (version 3.25) (Hammer
et al., 2001).
3 |
RESULTS
3.1 | Training quantification
For the 6 weeks prior to the WSER-specific training block,
the athlete was running ~115 km per week over mostly
flat ground with little strength training. Over the course
FIGURE 1
Visual representation
of study design. Anthropometric and
physiological measurements including
cardiorespiratory fitness (treadmill) were
first taken 21 weeks prior to Western States
Endurance Run (WSER; 163 km mountain
footrace from Squaw Valley, CA to Auburn,
CA) and are reported in the text as Baseline
results. Body composition (scale) and gut
microbiome composition (fecal symbol)
were intermittently evaluated throughout the
observational period. Relevant events prior
to the race such as a training camp (tent) and
shorter/preparatory races (race flags) have
been highlighted for temporal interpretation
4 of 7 |
GROSICKI et al.
of the 21-week training block, the participant increased
volume slightly (~124 km at an average pace of 201 m/
min) and supplemented with an additional 120 min per
week of strength and stretching exercises, accumulating
15,331 total minutes of training time. The largest vol-
ume of training consisted of an eight-day training camp in
northern California 4–5 weeks prior to WSER (Figure 1),
during which the ultramarathon runner accumulated 1,037
total minutes of training time (987 min running) and cov-
ered 177 km with 6,016 m elevation gain. To prepare for
WSER, the athlete competed in a 50 k trail race at week 4,
much of which was on the WSER trail, as well as a 100 k
sponsor-endorsed road race. During the 10-day post-race
observation period, the athlete refrained from any struc-
tured exercise for seven days before beginning light jog-
ging (~5:30/km) for 30–60 min duration.
3.2 | Body composition and bone
mineral content
Baseline body mass and height were 64.0 kg and 170 cm,
respectively, and body fat mass was 14.8%. Regional dis-
tribution of baseline body fat mass was 17.2%, 12.8% and
13.7% for the arms, legs and trunk, respectively. Bone min-
eral content was 2.6 kg and total body bone mineral den-
sity was 1.074 g/cm2 (NHANES/Lunar T-score = −1.3)
(WHO Study Group, 1994). After 15 weeks of WSER-
specific training, body mass increased slightly (+0.5 kg)
and total body fat was reduced to 14.2% due to compo-
sitional shifts in the arms and legs (16.5% and 12.7%, re-
spectively). Although bone mineral content was identical
between visits, bone mineral density improved to 1.086 g/
cm2 (T-score = −1.1), with the most notable changes oc-
curring in the spine and pelvis (Table 1).
3.3 | Submaximal and maximal oxygen
consumption (VO2max) exercise testing
Submaximal testing lasted 24 min, concluding after comple-
tion of a 6-min stage at 311 m/min (~5:10 min/mile) where
blood lactate reached 6.1 mmol/l and relative oxygen con-
sumption (VO2) was 61.8 ml kg−1 min−1. At velocities of
268 and 295 m/min, VO2 was 51.1 and 55.3 ml kg−1 min−1,
running economy values comparable or even superior to
those reported by Costill, Thomason and Roberts (51.7 and
59.0 ml kg−1 min−1 at corresponding velocities) in highly
trained distance runners (Costill et al., 1973).
Maximal exercise testing was terminated upon voli-
tional exhaustion (RPE = 20) at 295 m/min at an 8% grade.
Maximal oxygen consumption (VO2max) was 4.24 L/min
(66.7 ml kg−1 min−1) and maximal heart rate and RER values
were 186 bpm and 1.19, respectively. Maximal ventilation was
130.7 L/min. Based on these maximal values, lactate threshold
was estimated to occur at ~83% VO2max (RER = 0.89).
3.4 | Gut microbiome analyses
Microbial diversity (Shannon Diversity Index) oscillated
throughout the investigation, decreasing following 19-week
of highly specific race preparation but then increasing post-
event (Figure 2a). Firmicutes/Bacteroidetes ratio, a macro-
level indicator of microbial composition, was relatively stable
pre-WSER (~5:1), but nearly tripled 2 hr post-race due to a
69% reduction in Bacteroidetes relative abundance (Figure
2b). Meanwhile, the relative proportion of Proteobacteria in-
creased by more than fivefold 2 hr post-race, largely owing
to a 29-fold increase in Haemophilus (Figure 2c). Other no-
table changes in gut microbiome composition post-WSER
included the proliferation of Veillonella (+14,229%) and
Streptococcus (+438%) genera concomitant with reductions
in Alloprevotella (−79%) and Subdolingranulum (−50%).
4 |
DISCUSSION
Manipulating the human microbial ecosystem has pro-
lific health implications and emerging therapeutic poten-
tial (Khanna & Tosh, 2014). To our knowledge, this case
study shows the most rapid and pronounced gut microbiome
changes after acute exercise in the human literature. These
extraordinary microbial dynamics highlight the importance
of physical activity in determining human microbiome as-
sembly and emphasize yet another way in which human
movement can be one of the most powerful modulators of
human health.
Taxonomic richness (i.e., alpha diversity) is often con-
sidered a key indicator of gut microbiome health that is gen-
erally thought to increase with exercise training (Estaki et
TABLE 1
Regional bone mineral density (g/cm2) in a world-
class ultramarathon runner preparing for the Western States 100-mile
Endurance Race
Baseline
Pre-race
%Δ
Head
1.931
1.910
−1.088
Arms
0.647
0.653
+0.927
Legs
1.264
1.273
+0.712
Trunk
0.890
0.907
+1.910
Ribs
0.739
0.742
+0.406
Spine
0.970
0.992
+2.268
Pelvis
0.984
1.014
+3.048
Total
1.074
1.086
+1.117
|
5 of 7
GROSICKI et al.
al., 2016), as was recently observed throughout an ultra-en-
durance rowing race (Keohane et al., 2019). Paradoxically,
in this study alpha diversity, depicted as Shannon Diversity
Index, decreased following 19 weeks of highly specific
race preparation (2.83–2.73). This observation may be
attributed to the proliferation of select advantageous bac-
terial taxa, such as those involved in butyrate production
(e.g., Faecalibacterium, +40% baseline to pre-race), con-
comitant with the decline of less relevant microorganisms.
In contrast, unpublished observations from our laboratory
in a cohort of 28 young recreationally active individuals
(~30 years) demonstrated alpha diversity changes of plus or
minus 0.05 over a 3-week time period. Post-race, alpha di-
versity increased by a similar extent (+0.07) nearly reaching
baseline levels (2.80). Of the post-race changes in relative
genus abundance, a likely beneficial 143-fold increase in
Veillonella (~10% relative abundance post-WSER) was the
most pronounced. Indeed, Scheiman et al. (2019) recently
proposed an ergogenic role for Veillonella involving lactate
recycling after observing a similar, albeit less profound, el-
evation (~3% relative abundance) in Veillonella of runners
1–5 days after the Boston Marathon.
While an increase in Veillonella abundance was likely
a highly favorable adaptation to the 163 km race, other
microbial dynamics such as post-race insurgences of
Haemophilus, a bacterial genus composed of many sig-
nificant pathogenic species (e.g., H. Influenzae) and
Streptococcus (genus-level taxon of Group A Streptococcus
pyogenes) were also observed. It may be speculated that
intestinal proliferation of pathogenic bacterial species
plays a role in the increased incidence of infectious epi-
sodes observed in endurance athletes following prolonged
endurance exercise (Nieman, Johanssen, Lee, & Arabatzis,
1990). Moreover, reductions in butyrate-producing bacteria
FIGURE 2
Tracking the microbiome of a world-class ultramarathon runner. (a) Alpha diversity, represented as Shannon Diversity Index
(H), of the gut microbiome in a world-class ultramarathon runner measured at Baseline (21 week pre-Western States Endurance Run [WSER;
163 km mountain footrace]), Pre-Race (2 week pre-WSER), Post-Race (2 hr post-WSER), and Recovery (10 days post-WSER) measured via 16S
rRNA sequencing. (b) Relative phylum-level gut microbiome composition changes at the same time-points listed above. (c) Relative abundance
of bacteria genera over the course of the investigation. Genera comprising ≥5% of total fractional abundance for at least one time-point are
individually represented while genera of lower abundance were condensed (Other)
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Relative Phylum Abundance (%)
Actinobacteria
Bacteroidetes
Firmicutes
Fusobacteria
Proteobacteria
2.60
2.65
2.70
2.75
2.80
2.85
2.90
Baseline
Pre-Race
Post-Race
Recovery
α-Diversity (Shannon Diversity Index)
a
b
c
Pre-Race
Post-Race
100
90
80
70
60
50
40
30
20
10
0
Baseline
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100
Relative Genus Abundance (%)
Alloprevotella
Anaerostipes
Blautia
Faecalibacterium
Haemophilus
Pseudobutyrivibrio
Roseburia
Streptococcus
Subdolingranulum
Veillonella
Other
Recovery
6 of 7 |
GROSICKI et al.
(e.g., Subdolingranulum) associated with mucosal integrity
likely exacerbate this state of hypervulnerability to infec-
tion as well as contributing to elevated levels of circulat-
ing inflammation (Gill et al., 2015; Grosicki, Fielding, &
Lustgarten, 2018) and gastrointestinal distress (Jeukendrup
et al., 2000). Surprisingly however, the athlete did not re-
port any significant gastrointestinal complaints during or
following the event. Interpreted together, genus-level shifts
in gut microbiome composition post-WSER highlight the
complexity of interpreting microbiome data and reinforce
the importance of avoiding oversimplifying macro-level
observations (e.g., F/B ratio and/or alpha diversity) of mi-
crobial community structure (Shade, 2017).
In conclusion, these data add to a growing body of litera-
ture demonstrating the potency of acute exercise in shaping
human microbiome composition. Though other factors (e.g.,
diet, travel, etc.) may have influenced our findings, no purpose-
ful changes in diet or feelings of malaise were reported by the
ultramarathon runner over the course of the data collection pe-
riod. Nonetheless, more research involving carefully structured
training studies in both healthy and clinical populations and
interdisciplinary research teams is needed to fully understand
the complex interaction between physical activity and the gut
microbiome.
ACKNOWLEDGMENTS
We would like to thank the ultramarathon runner for his en-
thusiastic participation as well as his coaches, Roxanne Vogel
and Magdalena Boulet, for their contributions to his success.
We would also like to thank Amanda Fernandez, B.S. (Georgia
Southern University), Brett Cross, B.S. (Georgia Southern
University), and Gwenaelle Begue, Ph.D. (Sacramento State
University) for technical assistance, and Michael Lustgarten,
Ph.D. (Tufts University) for analytical insight.
CONFLICT OF INTEREST
The authors have no conflict to disclose.
AUTHOR CONTRIBUTIONS
GJG and JRB conceived and designed the research. GJG,
RPD, and RJB performed the experiments, analyzed data,
interpreted results of experiments, and prepared the figures.
GJG drafted the manuscript. GJG, RPD, and JRB edited, re-
vised, and approved the final version of the manuscript.
ORCID
Gregory J. Grosicki
https://orcid.org/0000-0001-8929-4903
REFERENCES
Ahn, J., Sinha, R., Pei, Z., Dominianni, C., Wu, J., Shi, J., … Yang,
L. (2013). Human gut microbiome and risk for colorectal cancer.
Journal of the National Cancer Institute, 105, 1907–1911. https ://
doi.org/10.1093/jnci/djt300
Allen, J. M., Mailing, L. J., Cohrs, J., Salmonson, C., Fryer, J. D., Nehra,
V., … Woods, J. A. (2018). Exercise training-induced modification
of the gut microbiota persists after microbiota colonization and at-
tenuates the response to chemically-induced colitis in gnotobiotic
mice. Gut Microbes, 9, 115–130. https ://doi.org/10.1080/19490
976.2017.1372077
Bell, P. G., Furber, M. J., Someren, K., Anton-Solanas, A., & Swart,
J. (2017). The physiological profile of a multiple Tour de France
winning cyclist. Medicine and Science in Sports and Exercise, 49,
115–123. https ://doi.org/10.1249/MSS.00000 00000 001068
Bik, E. M., Ugalde, J. A., Cousins, J., Goddard, A. D., Richman, J., &
Apte, Z. S. (2018). Microbial biotransformations in the human distal
gut. British Journal of Pharmacology, 175, 4404–4414. https ://doi.
org/10.1111/bph.14085
Borg, G. (1970). Perceived exertion as an indicator of somatic stress.
Scandinavian Journal of Rehabilitation Medicine, 2, 92–98.
Caporaso, J. G., Lauber, C. L., Walters, W. A., Berg-Lyons, D.,
Lozupone, C. A., Turnbaugh, P. J., … Knight, R. (2011). Global
patterns of 16S rRNA diversity at a depth of millions of sequences
per sample. Proceedings of the National Academy of Sciences, 108,
4516–4522. https ://doi.org/10.1073/pnas.10000 80107
Conley, D. L., & Krahenbuhl, G. S. (1980). Running economy and
distance running performance of highly trained athletes. Medicine
and Science in Sports and Exercise, 12, 357–360. https ://doi.
org/10.1249/00005 768-19802 5000-00010
Costill, D. L., Thomason, H., & Roberts, E. (1973). Fractional utili-
zation of the aerobic capacity during distance running. Medicine
and Science in Sports and Exercise, 5, 248–252. https ://doi.
org/10.1249/00005 768-19730 0540-00007
Cuddy, J., Slivka, D., Hailes, W., & Dumke, C. (2009). Total energy ex-
penditure, body water turnover, hydration status, and blood compo-
sition during the Western States 100. Medicine and Science in Sports
and Exercise, 41, S336. https ://doi.org/10.1249/01.MSS.00003
55133.78976.4a
Durk, R. P., Castillo, E., Marquez-Magana, L., Grosicki, G. J., Bolter,
N. D., Lee, C. M., & Bagley, J. R. (2018). Gut microbiota compo-
sition is related to cardiorespiratory fitness in healthy young adults.
International Journal of Sport Nutrition and Exercise Metabolism,
29, 249–253.
Estaki, M., Pither, J., Baumeister, P., Little, J. P., Gill, S. K., Ghosh, S.,
… Gibson, D. L. (2016). Cardiorespiratory fitness as a predictor of
intestinal microbial diversity and distinct metagenomic functions.
Microbiome, 4, 42. https ://doi.org/10.1186/s40168-016-0189-7
Gill, S. K., Hankey, J., Wright, A., Marczak, S., Hemming, K., Allerton, D.
M., … Costa, R. J. (2015). The impact of a 24-h ultra-marathon on cir-
culatory endotoxin and cytokine profile. International Journal of Sports
Medicine, 36, 688–695. https ://doi.org/10.1055/s-0034-1398535
Gomez, A., Sharma, A. K., Mallott, E. K., Petrzelkova, K. J., Jost
Robinson, C. A., Yeoman, C. J., … Leigh, S. R. (2019). Plasticity
in the human gut microbiome defies evolutionary constraints.
mSphere, 4, e00219–e00271. https ://doi.org/10.1128/mSphe
re.00271-19
Grosicki, G. J., Fielding, R. A., & Lustgarten, M. S. (2018). Gut mi-
crobiota contribute to age-related changes in skeletal muscle size,
composition, and function: Biological basis for a gut-muscle axis.
Calcified Tissue International, 102, 433–442.
Hammer, Ø., Harper, D. A., & Ryan, P. P. A. S. T. (2001). Paleontological
statistics software package for education and data analysis.
Paleontologica Electronica, 4, 9.
|
7 of 7
GROSICKI et al.
Jeukendrup, A. E., Vet-Joop, K., Sturk, A., Stegen, J. H., Senden, J.,
Saris, W. H., & Wagenmakers, A. J. (2000). Relationship between
gastro-intestinal complaints and endotoxaemia, cytokine release
and the acute-phase reaction during and after a long-distance triath-
lon in highly trained men. Clinical Science, 98, 47–55. https ://doi.
org/10.1042/cs098 0047
Keohane, D. M., Woods, T., O'Connor, P., Underwood, S., Cronin, O.,
Whiston, R., … Molloy, M. G. M. (2019). Four men in a boat: Ultra-
endurance exercise alters the gut microbiome. Journal of Science
and Medicine in Sport, 22, 1059–1064.
Khanna, S., & Tosh, P. K. (2014). A clinician's primer on the role of the
microbiome in human health and disease. Mayo Clinic Proceedings,
89, 107–114. https ://doi.org/10.1016/j.mayocp.2013.10.011
Larsen, N., Vogensen, F. K., van den Berg, F. W. J., Nielsen, D. S.,
Andreasen, A. S., Pedersen, B. K., … Jakobsen, M. (2010). Gut mi-
crobiota in human adults with type 2 diabetes differs from non-di-
abetic adults. PLoS One, 5, e9085. https ://doi.org/10.1371/journ
al.pone.0009085
Mailing, L. J., Allen, J. M., Buford, T. W., Fields, C. J., & Woods, J. A.
(2019). Exercise and the gut microbiome: A review of the evidence,
potential mechanisms, and implications for human health. Exercise
and Sport Sciences Reviews, 47, 75–85.
McInnes, P. (2010). Manual of procedures for human microbiome proj-
ect: Core sampling Protocol A (HMP Protocol #07-001). edited by
Health NIo2010.
McMiken, D. F., & Daniels, J. T. (1976). Aerobic requirements
and maximum aerobic power in treadmill and track running.
Medicine and Science in Sports and Exercise, 8, 14–17. https ://doi.
org/10.1249/00005 768-19762 1000-00015
Minalla, A. R., Dubrow, R., & Bousse, L. J. (2001). Feasibility of
high-resolution oligonucleotide separation on a microchip. In Proc.
SPIE 4560, microfluidics and BioMEMS (September 28, 2001).
https ://doi.org/10.1117/12.443044
Nieman, D. C., Johanssen, L. M., Lee, J. W., & Arabatzis, K. (1990).
Infectious episodes in runners before and after the Los Angeles
Marathon. Journal of Sports Medicine and Physical Fitness, 30,
316–328.
Relman, D. A. (2015). The human microbiome and the future prac-
tice of medicine. JAMA, 314, 1127–1128. https ://doi.org/10.1001/
jama.2015.10700
Scheiman, J., Luber, J. M., Chavkin, T. A., MacDonald, T., Tung, A.,
Pham, L. D., … Kostic, A. D. (2019). Meta-omics analysis of elite
athletes identifies a performance-enhancing microbe that functions
via lactate metabolism. Nature Medicine, 25, 1104–1109. https ://
doi.org/10.1038/s41591-019-0485-4
Shade, A. (2017). Diversity is the question, not the answer. ISME
Journal, 11, 1–6. https ://doi.org/10.1038/ismej.2016.118
Turnbaugh, P. J., Ley, R. E., Hamady, M., Fraser-Liggett, C. M., Knight,
R., & Gordon, J. I. (2007). The human microbiome project. Nature,
449, 804–810. https ://doi.org/10.1038/natur e06244
Wang, Z., Klipfell, E., Bennett, B. J., Koeth, R., Levison, B. S., Dugar,
B., … Hazen, S. L. (2011). Gut flora metabolism of phosphatidyl-
choline promotes cardiovascular disease. Nature, 472, 57–63. https
://doi.org/10.1038/natur e09922
WHO Study Group. (1994). Assessment of fracture risk and its appli-
cation to screening for postmenopausal osteoporosis. Report of a
WHO Study Group. World Health Organization Technical Report
Series, 843, 1–129.
How to cite this article: Grosicki GJ, Durk RP,
Bagley JR. Rapid gut microbiome changes in a
world-class ultramarathon runner. Physiol Rep.
2019;7:e14313. https ://doi.org/10.14814/ phy2.14313
| Rapid gut microbiome changes in a world-class ultramarathon runner. | [] | Grosicki, Gregory J,Durk, Ryan P,Bagley, James R | eng |
PMC10600198 | 1
Vol.:(0123456789)
Scientific Reports | (2023) 13:18222
| https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports
Anthropometric profiles and body
composition of male runners
at different distances
Aleksandra Stachoń , Jadwiga Pietraszewska & Anna Burdukiewicz *
Anthropometric parameters are crucial prerequisite to achieve success in professional running
sports. However, it is not clear how these parameters are relevant for athletes performing on a less
demanding sport level as academic competitions. To help coaches and selectors working on this
level, we have explored anthropometric variables and body composition in 68 academic athletes:
26 sprinters, 22 middle distance runners, and 20 long distance runners. Sprinters have a more
massive body shape, shorter lower legs in relation to the length of the thigh, broader shoulders and
narrower hips, greater musculature and cellular mass. A slender figure, a longer shin, and the greatest
subcutaneous fat and extracellular mass characterize long-distance runners. Middle-distance runners
are the slimmest, and have a narrow trunk and little subcutaneous fat. Sprinters and long-distance
runners are mesomorphic, while middle-distance runners present more mixed mesomorph-ectomorph
type. The principal component analysis highlighted the importance of the overall size of the body,
limbs musculature and the length of the lower limb together with its segments, and also body fatness.
This approach emphasized the morphological distinctiveness of runners at particular distances and
allows the use of somatic features as predictors of running performance.
Running is the most popular and, at the same time, the simplest form of movement that brings multi-directional
benefits to the body, including improving the functioning of the heart, nervous and digestive systems1. The
popularization of this sport among amateurs is facilitated by the fact that it can be practiced at various distances,
and various technical solutions are available to support training control2,3. In recent years, the popularization of
running as a form of physical recreation has resulted in this issue being reflected in numerous scientific works4.
Physiological aspects of running performance were studied, and issues related to running economy and energy
costs were analyzed5,6. In addition, biomechanical aspects affecting the running economy have been studied7–9.
The issue of morphological diversity of runners focused on the characteristics of body proportions, body com-
position and somatic structure of competitors of specific distances and the impact of anthropometric features on
the results achieved by athletes10. It has been shown that professional athletes (Olympic champions, finalists, and
running event participants) are clearly differentiated in age, height and body weight11,12. Sprint champions tend
to be heavier than lower-ranked competitors, while distance runners show the opposite trend11,13. A cohort study
including elite and leisure runners shows that body composition is a better predictor of running performance
than body mass index. Furthermore, fat mass was found to be negatively associated with running speed. High val-
ues of the fat-free index had a positive effect on the performance of women, while no such relationship was found
for men14. In turn body composition studies of University level male track and field athlete of India have shown
that sprinters have the lowest body fat. However, there was no significant difference between middle and long
distance runners15. The Sheldon’ typology modified by Heath and Carter16 is often used to assess the physique
in sports. That typology allows the assessment of body shape in the form of a somatotype that is understood as a
description of the current morphological state of an individual. The somatotype is expressed by three numbers,
each of which represents one of the basic components: endomorphy, mesomorphy and ectomorphy17. Endomor-
phy is relative fatness, mesomorphy is characterized by relative musculo-skeletal robustness and ectomorphy
is relative linearity or slenderness of a physique. Highly trained athletes also differ from the general population
of athletes by having less endomorphy and greater mesomorphy. Long distance runners who do mostly aerobic
training are less endomorphic and mesomorphic but have more ectomorphy than other athletes16. Similar trends
have been reported in the young elite middle and long distance runners18 and Iranian cross-country runners19.
However, the cited studies had limitations related to, e.g., the number of studied athletes and their sports level.
OPEN
Faculty of Physical Education and Sport Sciences, Wroclaw University of Health and Sport Sciences, al. Ignacego
Jana Paderewskiego 35, 51-612 Wrocław, Poland. *email: [email protected]
2
Vol:.(1234567890)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
As mentioned above, the body morphology of athletes is the result of selection and adaptation of the body
to training loads, which differ in individual disciplines20. Running at different distances makes it necessary to
use various training methods21. The volume and intensity of training are adapted to the distance the athlete will
face. Optimizing selection processes and training methods based on the somatic factors seem to be essential to
success in running at particular distances22. Our study aims to determine anthropometric profiles, including
body size and proportions, somatotype and body composition of long-, middle- and short-distance runners.
Our hypothesis is: the long-, middle- and short-distance runners are diversified in body size and proportions,
somatotype as well as in body composition, which may be called as ‘various anthropometric profiles’.
Materials and methods
Participants
Sixty eight male college athletes (age 20.7 ± 2.05 yrs old) participated in the study. This sample included 26 sprint-
ers (S), 22 middle distance runners (M) and 20 long distance runners (L). The runners were classified into the
S, M or L groups according to their declaration of participation in sprints (200 m and 400 m), middle distances
(800 m and 1500 m) and long distances (3000 m, 5000 m, 10,000 m). The athletes were involved in regional and
national level competitions and trained at least 4 times a week for 2 h per day. Age and training experience did
not significantly differ among the surveyed athletes: sprinters (20.37 ± 1.71 yrs old, 5.16 ± 2.21 years of experi-
ence), middle-distance runners (20.31 ± 1.55 yrs old, 5.50 ± 2.46 years of experience), long-distance runners
(21.39 ± 2.74 yrs old, 6.75 ± 2.34 years of experience).
Academic athletes practice sports in the clubs of the Academic Sports Associations functioning at Wrocław
universities. The athletes who fulfilled these criteria inquired for the study. The criterion for inclusion in the
study was at least a 3 years of practice and competed at national level, no injuries and no special diets in the
period preceding the study. The conditions for exclusion were a break in training and injuries or diseases that
prevented the measurements.
The research was approved by the Senate’s Research Bioethics Commission of the Wroclaw University of
Health and Sport Sciences, Poland [consent number 2/2020], and conducted according to the requirements
stipulated in the Declaration of Helsinki. Participants were fully informed about all experimental procedures
and written informed consent was obtained from all of them.
Measurements and calculations
The measurements were carried out taking into account the training periodization, at the beginning of the
preparation period and in the Central Research Laboratory of Wroclaw University of Health and Sport Sciences,
Poland (Quality Management System Certificate: PN-EN ISO 9001:2015—Certificate Reg. No.: PW-15105-22X).
All participants visited the laboratory once and underwent a series of measurements.
Measurements were made by experienced anthropometrists in the morning at room temperature about
22-24ºC. Anthropometric measurements were performed following measurement protocols established by the
International Society for the Advancement of Kinanthropometry (ISAK). Each anthropometrist took a series of
measurements assigned to them and was accompanied by a person recording them. Measurements were taken
on the right side of the participant’s body. Anthropometric equipment from GPM Siber Hegner Machinery
Ltd. was used. (Zurich, Switzerland): an anthropometer, a sliding caliper—Martin type, a spreading caliper, a
skinfold caliper, an anthropometric tape. Body weight was measured using an electronic scale with an accuracy
of 0.1 kg (Fawag, Lublin, Poland). Each measure was taken two times by the same investigator. Technical error
of measurement was < 3% for skinfolds, and < 1% for breadths, lengths and girths. The mean values were used
in the statistical analysis.
The results of measurements recommended for monitoring athletes23 were included in the study. Heights,
lengths, widths and circumferences were measured to the nearest 0.1 cm: body height, lower limb height to
trochanterion point, thigh length between trochanterion and tibiale laterale points, tibia length between tibiale
mediale and sphyrion tibiale points, foot length, upper limb length between acromiale and dactylion points,
arm length between acromiale and radiale points, forearm length between radiale and stylion points, biacro-
mial breadth, biiliocristal breadth, biepicondylar humerus breadth, biepicondylar femur breadth, ankle breadth
between malleolare tibiale and malleolare fibulare points. The following girths were measured: chest at the level
of the mesosternale point, gluteal at the level of the greatest posterior protuberance of the buttocks, arm flexed,
forearm, thigh taken 1 cm below the level of the gluteal fold and calf. Skinfold sites were landmarked at the sub-
scapular, abdominal, supraspinale, triceps, forearm, front thigh and medial calf. All sites were then measured
using caliper with 10 g × mm−2 constant pressure.
The measured features were used to calculate the following indices: body mass index BMI (body mass/body
height2 [kg/m2]), lower limb length index (lower limb length/body height), upper limb index (upper limb length/
body height), crural index (tibia length/thigh length), biacromial index (biacromial breadth/body height), biili-
ocristal index (biiliocristal breadth/body height), biiliocristal-acromiale index (biiliocristal breadth/biacromial
breadth), bone massiveness index (biepicondylar femur breadth + ankle breadth/body height), flexed arm girth
index (flexed arm circumference/arm length), forearm girth index (forearm circumference/forearm length), thigh
girth index (thigh circumference/thigh length), calf girth index (calf circumference/tibia length).
In addition, two indices characterizing subcutaneous fat were calculated: fat distribution index and subcu-
taneous fat index. The fat distribution index is the quotient of the sum of skinfolds on limb segments (∑ tri-
ceps + forearm + front thigh + medial calf) and trunk (∑ subscapular + supraspinale + abdominal). Subcutaneous
fat index (SFI) takes into account trunk and extremity skinfolds and body height (∑ trunk skinfolds + ∑ limbs
skinfolds/body height)24.
3
Vol.:(0123456789)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
The somatotype of each subject was also determined according to Sheldon’s method, modified by Heath and
Carter16. Somatotype Calculation and Analysis software classified the average somatotype of each group and
illustrated the outcome in a somatotype chart25.
The non-invasive bioelectrical impedance method assessed body composition with tetrapolar version hand-
to-foot electrodes (BIA 101 analyzer, Akern, Bodygram 1.31 software). Measurements were made considering
the manufacturer’s rules for obtaining correct results. The following features were used in the analysis: fat mass
(FM) [kg, %], body cell mass (BCM) [kg, %] and extracellular mass (ECM) [kg, %].
An interview was conducted with the respondents to collect information on the date of birth and training
experience, diets and supplements, and the occurrence of injuries.
Statistical analysis
Statistical analysis was performed using the Statistica 13.3 package (TIBCO Software Inc.). Descriptive statistics
were used for the quantitative analysis of the collected data. The Shapiro–Wilk test was used to examine the dis-
tributions of the analyzed features. Differentiation in the level of development of the analyzed features between
the groups was assessed using the analysis of variance and post-hoc Tukey HSD test for unequal samples. The
results in the text and tables are presented in the form of mean and standard deviation. The significance level for
all tests and statistical procedures was set at an p value of 0.05.
Differences in somatic composition were examined using the SANOVA—Somatotype Analysis of Variance
procedure25. A ternary plot was used to examine the relationship between the three components of body com-
position (FM, BCM, ECM) in groups of runners. The distribution of individual competitors’ points in the three
body composition variables system was assessed using the χ2 test. Principal component (PC) analysis was per-
formed, expressing a linear combination of the morphological variables. The analysis was preceded by a Box-Cox
transformation, which enabled the principal components to be based on correlations. The number of factors was
defined using the Kaiser criterion26.
Institutional review board
The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethics
Committee of the University School of Physical Education in Wrocław, Poland (2/2020).
Informed consent
Written informed consent has been obtained from the patient(s) to publish this paper.
Results
Short-distance runners are characterized by significantly higher body weight than competitors from other groups
(Table 1). Body height, length of the lower and upper limbs as well as their segments did not show any significant
differences between the groups of runners. Among the width features, only the biacromial diameter is signifi-
cantly wider in the group of sprinters compared to middle-distance runners. The massiveness of the skeleton
assessed by the width of the elbow, knee and inner ankle reaches the highest values among examined sprinters.
All the analyzed circumferences of the trunk and limb segments are the largest in the group of short-distance
runners. The significantly thickest supraspinale and triceps skinfolds are characteristic of long-distance runners.
Also, the sum of skinfolds on the trunk and limb segments is significantly greater in long-distance runners.
BMI is significantly higher in the group of sprinters. The smallest massiveness is characteristic of middle-
distance runners (Table 2). The general proportions of the length of the upper and lower limbs are similar in all
groups of runners, but the proportions of functional segments differ. Sprinters are characterized by a significantly
shorter tibia in relation to the length of the thigh compared to other groups of runners. They also have broader
shoulders in relation to body height. The highest values of the hip width index characterize long-distance runners.
The lowest values of the discussed indicator occurred among sprinters. The relative massiveness of the epiphysis
of the lower limb is significantly lower among short- and medium-distance runners. Significantly larger girths
of limb segments in relation to their length occurred among sprinters compared to other groups.
The distribution of subcutaneous fat on the limbs and trunk expressed by the fat distribution index does not
significantly differ between the groups of runners (Table 2). However, among the study participants, there was a
tendency to increase the fatness of the limbs in relation to the trunk with the lengthening of the distance. The low-
est adiposity of the limbs in relation to the adiposity of the trunk occurs in sprinters, while the highest—applies
to the group of long-distance runners. The content of subcutaneous fat in relation to body height, expressed by
the subcutaneous fat index, is significantly lower in the groups of short- and medium-distance runners.
Endomorphy is significantly lower in groups of short and middle distance runners than in long distance
runners (Table 2). On the other hand, the mesomorphic component reaches significantly higher values in the
group of short and long distance runners. However, significantly greater ectomorphy is characteristic of middle
distance runners compared to sprinters. The average somatotype of middle runners are mesomorph-ectomorph
(1.60–3.82–3.81), while sprinters (1.68–4.94–2.90) and long distance runners (2.11–4.72–3.36) are ectomorphic
mesomorphs (Fig. 1). The somatotype variance analysis showed a statistically significant difference in the soma-
totypes of runners (F = 7.41; p = 0.001).
The average values of the fat and extracellular mass percentages do not show statistically significant differences
between the groups. In contrast, cell mass is significantly higher in sprinters compared to other groups. Distribu-
tions of individual competitors’ points in the system of three body composition variables (Fig. 2) assessed with
the χ2 test also show statistically significant differences between groups (χ2 = 31.49, p < 0.05).
As a result of the principal components analysis, three principal components were identified, explaining the
problem in approximately 74% (Table 3). The first principal component (PC1), which has the greatest part in total
4
Vol:.(1234567890)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
variability, is highly correlated with body weight and height, shoulder width and hip width, length of the lower
limb and its segments, length of the upper limb, as well as with the muscle circumferences of the arm, forearm,
thighs and thighs. Thus, the mentioned variable characterizes the overall size of the body. The second principal
component (PC2) is positively correlated with muscle circumferences and negatively correlated with the length
of the lower limb, the length of the thigh and the tibia. It can therefore be concluded that PC2 characterizes the
massiveness and proportions of the limbs. This principal component divides the participants into two groups:
(1) athletes with well-developed musculature, especially of the upper limb, and (2) athletes with a tendency to
have longer lower limbs. In the case of the third principal component (PC3), the most diagnostic features are the
sum of skinfolds on the trunk and the sum of skinfolds on the limbs. This component characterizes body fatness.
Table 4 summarizes the mean principal component scores of the athletes from different distances. It can
be concluded that the PC2 and PC3 components are significantly differentiated. The sprinter group’s overall
body size (including massiveness and muscularity) is slightly higher. The second principal component PC2
substantially differs between the studied groups. Intergroup differentiation in the light of PC3 is also statistically
significant. There is clearly a higher level of development of the analyzed muscle girths in short-distance athletes.
Attention is drawn to the distinctiveness of long-distance runners, characterized by greater subcutaneous fat
compared to competitors from other groups.
Discussion
In accordance with the reviewer’s comment, we have introduced in the manuscript the following sentence:
The results obtained make it possible to present the detailed somatic characteristics of runners at different dis-
tances, confirming the hypothesis that their anthropometric profiles are differentiated. The surveyed men were
of similar age and had similar training experience. It is well known that running performance is determined by
power output and running efficiency6. The first factor is related to the athlete’s physiological profile, while the
second factor characterizes the efficiency in the conversion of power to translocation. It is directly related to the
Table 1. Statistical characteristics and inter-group differences of the anthropometric features in short
(S), middle (M) and long (L) distance runners (SD—standard deviation; asignificantly different from M;
bsignificantly different from L).
Group of runners
Variable
S
M
L
p
Mean (SD)
Mean (SD)
Mean (SD)
Body mass [kg]
74.3 (6.56)b
69.2 (7.77)
67.6 (8.99)
0.011
Body height [cm]
180.6 (6.16)
181.3 (6.33)
177.3 (6.91)
0.109
Lower limb length [cm]
95.3 (3.99)
95.9 (4.34)
94.0 (3.89)
0.297
Thigh length [cm]
47.6 (2.17)
47.3 (2.22)
46.7 (2.07)
0.183
Tibia length [cm]
40.2 (1.89)
40.6 (2.13)
40.2 (1.77)
0.219
Foot length [cm]
27.0 (1.35)
26.7 (1.41)
26.6 (1.30)
0.603
Upper limb length [cm]
78.7 (3.02)
79.1 (3.21)
78.1 (3.13)
0.598
Arm length [cm]
33.8 (1.62)
34.0 (1.42)
33.6 (1.44)
0.695
Forearm length [cm]
26.1 (1.13)
26.4 (1.49)
25.8 (1.32)
0.259
Biacromial breadth [cm]
41.5 (1.86)a
40.0 (1.72)
40.7 (2.05)
0.024
Biiliocristal breadth [cm]
28.0 (1.50)
27.3 (1.55)
28.2 (1.92)
0.206
Humerus breadth [cm]
7.1 (0.48)
6.9 (0.37)
7.0 (0.67)
0.361
Femur breadth [cm]
9.8 (0.44)
9.6 (0.36)
9.8 (0.52)
0.185
Ankle breadth [cm]
7.7 (0.37)
7.4 (0.42)
7.5 (0.40)
0.130
Chest girth [cm]
95.3 (3.88)
93.2 (4.41)
93.6 (6.64)
0.333
Gluteal girth [cm]
96.2 (3.83)
94.5 (4.48)
93.6 (4.15)
0.111
Arm flexed girth [cm]
31.2 (1.75)ab
28.9 (1.63)
29.4 (2.37)
0.000
Forearm girth [cm]
25.9 (1.71)
24.8 (1.53)
25.1(1.55)
0.058
Thigh girth [cm]
55.5 (3.01)ab
53.1 (3.10)
51.8 (3.14)
0.001
Calf girth [cm]
37.1 (2.46)a
35.3 (2.65)
35.5 (2.28)
0.022
Subscapular skinfold [mm]
7.8(1.32)
7.1 (1.10)
8.1 (2.19)
0.095
Supraspinale skinfold [mm]
5.7 (1.44)b
5.6 (1.51)b
7.6 (3.47)
0.009
Abdominal skinfold [mm]
6.1(1.75)
6.2 (1.88)
7.2 (2.88)
0.195
Triceps skinfold [mm]
4.4 (1.36)b
4.6 (1.28)b
6.0 (2.56)
0.007
Forearm skinfold [mm]
3.1 (0.59)
3.0 (0.38)
3.3 (0.65)
0.119
Front thigh skinfold [mm]
7.9 (1.06)
7.4 (1.12)
8.2 (1.51)
0.107
Medial calf skinfold [mm]
3.8 (0.83)
3.7 (0.97)
4.4 (1.23)
0.101
∑ trunk skinfolds
19.6 (3.57
18.9 (3.76)b
22.9 (7.87)
0.038
∑ limbs skinfolds
17.2 (3.32)b
17.0 (3.21)b
20.3 (5.11)
0.010
5
Vol.:(0123456789)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
Table 2. Statistical characteristics and inter-group differences of the anthropometric indices, somatotype’s
components and body composition in short (S), middle (M) and long (L) distance runners (SD—standard
deviation; asignificantly different from M; bsignificantly different from L).
Group
S
M
L
p
Mean (SD)
Mean (SD)
Mean (SD)
Body proportions
Body mass index
22.76 (1.42)ab
20.99 (1.54)
21.44 (1.89)
0.001
Lower limb length index
52.75 (1.31)
52.88 (1.17)
53.00 (1.37)
0.801
Upper limb length index
43.59 (1.02)
43.61 (0.91)
44.06 (1.12)
0.244
Crural index
84.52 (2.12)ab
85.91 (1.89)
86.61 (1.90)
0.002
Biacromial index
22.98 (1.05)a
22.05 (1.02)b
22.97 (1.20)
0.007
Biiliocristal index
15.50 (0.90)
15.06 (0.79)b
15.87 (0.86)
0.012
Biiliocristal-acromiale index
67.48 (3.34)
68.37 (3.95)
69.19 (3.71)
0.294
Bone massiveness index
9.66 (0.39)a
9.37 (0.20)b
9.75 (0.47)
0.002
Flexed arm girth index
92.69 (6.94)ab
85.14 (5.14)
87.69 (6.97)
0.000
Forearm girth index
99.33 (6.62)a
94.13 (7.40)
97.71 (6.75)
0.038
Thigh girth index
116.62 (7.04)b
112.44 (6.67)
111.57 (6.26)
0.026
Calf girth index
92.34 (6.67)a
86.02 (6.01)
87.96 (5.74)
0.002
Fat distribution index
0.89 (0.18)
0.91 (0.18)
0.92 (0.20)
0.823
Subcutaneous fat index
20.44 (3.52)b
19.81 (3.25)b
24.33 (6.23)
0.003
Somatotype components
Endomorphy
1.68 (0.38)b
1.60 (0.31)b
2.11 (0.76)
0.004
Mesomorphy
4.94 (1.00)a
3.82 (0.75)b
4.72 (1.12)
0.000
Ectomorphy
2.90 (0.77)a
3.81 (0.80)
3.36 (0.84)
0.001
Body composition
% Fat mass
17.06 (3.31)
18.71 (3.57)
18.91 (2.52)
0.095
% Extracellular mass
31.53 (7.27)
32.89 (5.82)
33.22 (3.56)
0.583
% Body cell mass
51.34 (6.56)ab
48.40 (7.45)
48.03 (3.13)
0.035
Figure 1. Mean of somatotypes in groups of male runners (S—short distance, M—middle distance, L—long
distance).
6
Vol:.(1234567890)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
Figure 2. Distribution of runners in the system of three variables of body composition (○—short distance, Δ—
middle distance, □—long distance).
Table 3. Principal components loadings and correlations between the components and original variables.
Variable
PC1
PC2
PC3
Eigenvalues
7.02
2.52
1.59
% total variance
46.82
16.77
10.63
Cumulative eigenvalues
7.02
9.54
11.13
% cumulative
46.82
63.59
74.21
Factor loadings
Body mass
0.91
0.27
0.05
Body height
0.84
− 0.35
0.03
Leg length
0.84
− 0.49
− 0.04
Thigh length
0.82
− 0.43
− 0.03
Tibia length
0.73
− 0.54
0.02
Foot length
0.77
− 0.13
0.14
Arm length
0.79
− 0.39
0.06
Biacromial breadth
0.54
0.35
− 0.05
Biiliocristal breadth
0.65
0.06
0.06
Arm flexed girth [cm]
0.66
0.66
− 0.18
Forearm girth [cm]
0.64
0.64
− 0.08
Thigh girth [cm]
0.73
0.44
− 0.10
Calf girth [cm]
0.65
0.42
− 0.23
∑ trunk skinfolds
0.17
0.30
0.81
∑ extremities skinfolds
0.03
0.11
0.90
Table 4. Mean principal component scores of the runners from different groups (S—short distance; M—
middle distance; L—long distance; asignificantly different from M; bsignificantly different from L).
Group
S
M
L
p
PC1
0.86
− 0.23
− 0.69
0.100
PC2
0.74a
− 0.91
0.03
0.001
PC3
− 0.41b
− 0.19
0.60
0.010
7
Vol.:(0123456789)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
athlete’s biomechanical profile and depends on anthropometric dimensions, limb morphology, and learned and
developed movement patterns7,9.
As shown in other studies, height and length dimensions were slightly larger in the group of middle-distance
runners27. The body height of the athletes from the study groups was similar and oscillated around the value of
the 50th percentile (178.7 cm) for the nationwide population28. This suggests that this parameter is not important
in the selection process for specific running distances. Body shape assessed by BMI turned out to be slimmer in
the middle- and long-distance runners groups than in sprinters, which was also indicated by other researchers29.
In previously published studies, differences in body weight were noted depending on the results obtained by
runners. The medalists of the Olympic Games in the distances of 100, 200, and 400 m were heavier compared to
the finalists and other participants. In turn, the medalists at distances of 5,000 m and 10,000 m and the marathon
were lighter than the finalists and other competitors participating in these distances11. High body weight and
BMI values in sprinters result from the fact that muscle mass plays a significantly role in their efforts13, which
was shown in the currently presented research in the form of a high level of sprinters’ mesomorphy. As previ-
ous studies have shown, effective sprinting requires strong deep muscles of the trunk (psoas major, transversus
abdominis, and multifidus muscle), which, acting earlier than other muscles, are the basis of limb strength and
thus affect the athlete’s sports results30. Also, Tottori et al.31, comparing the cross-sectional areas of the trunk
and lower limb muscles in sprinters, found significantly larger psoas major and gluteus maximus muscles than
non-runners. In addition, they not only correlated significantly with 100 m sprint performance, but they turned
out to be good predictors of top performance over that distance.
The upper body and arms also play an important role in running, providing balance and promoting efficient
movement. In our research, the transverse dimensions of the trunk (shoulder width, hip width), skeletal mass
and limb circumferences were significantly larger in sprinters, which is confirmed in other studies31,32. No such
differentiation was found between long- and middle-distance runners, confirming previous observations33,34. The
shape of the torso contributes to the locomotor efficiency and energetics of running, influencing the respiratory
mechanics and biomechanics of the limbs with its morphology33. Studies of variations in trunk morphology in
the context of locomotor ability have shown a relationship between trunk shape and running performance33.
According to these results, people with a narrower torso can achieve higher speeds33, which was confirmed by
our research. In addition, as other studies have shown34 less thoracic kyphosis and a flattened chest positively
affect the mechanics of breathing (chest mobility). On the other hand, increased lordosis has a beneficial effect
on the biomechanics of the lower limbs, playing an important role in mitigating the effects of shocks transmitted
through the human spine during dynamic activities such as running34. The width of the pelvis also affects the
work of the psoas major muscle, affecting its rotation capacity and hip flexors35.
The length proportions do not significantly differ between the examined groups of athletes. Slightly higher
values of the lower limb length index and upper limb length index characterize long-distance runners, which
was confirmed in other studies9,36. In the presented work, it was shown that sprinters are characterized by slightly
shorter lower limbs compared to other groups of athletes. Research by other authors shows that long legs are
beneficial in sprinters, but only to the optimal level correlating with their height37. If the lower limbs are above this
optimal length, they can generate problems producing the high stride frequency that is a prerequisite for good38.
In the group of middle-distance runners, slightly longer lower limbs and their segments were demonstrated, as
noted by other authors39. Researchers also note that the main difference between long- and short-distance run-
ners is stride length, not running pace40. Studies have shown that shorter distance runs require longer strides41,42,
which may be reflected in the proportions between the lower limb segments to some extent.
Bereket43 noted that the body’s size and proportions affect the energy of locomotion and the speed of move-
ment. It was found that taller people with a wider pelvis, having a longer lower leg move at a much higher optimal
walking speed at a lower energy cost, which was justified, among others, by the importance of the length of the
distal section in heat dissipation44. The analysis of intergroup differences of proportions within the lower limb
of runners at various distances carried out in the current study provided interesting results. The relationship
between the proximal and distal segments of the lower limb is noteworthy. It was found that long-distance run-
ners have a relatively long shin (relative to the thigh), suggesting that long-term intense exercise may promote
proportions that favore more efficient heat loss in the lower limb. On the other hand, sprinters have a significantly
shorter shin in relation to the length of the thigh (crural index) compared to other groups of competitors. Also,
Tomita et al.45 showed that the ratio of tibia length to femur length significantly correlated with running perfor-
mance in sprinters, suggesting that this particular morphological factor may play an important role in achieving
better running performance in specialized 400 m sprinters.
Likewise, the relative slenderness of the thigh and lower leg are significant factors in running economy46. In
the presented studies, a significantly more massive skeleton, assessed by the width of the epiphyses, is charac-
teristic of long-distance runners, which can be justified by the influence of varied effort47,48. The musculature
of the limbs is shaped differently. Significantly larger circumferences of limb segments were found in sprinters
compared to other groups of runners. Korhonen et al.49 showed that muscle thickness was a strong predictor of
the braking forces generated during sprinting. Similarly, other studies have found that sprinters with higher lean
body mass in the lower limbs showed higher mean power in the Wingate test50. Current research has shown that
long- and middle-distance runners are characterized by slimmer limb segments compared to sprinters, which
has been confirmed in the literature39.
The diversified energy cost of running over particular distances affects muscle mass development and body
fatness32. In addition, the amount of subcutaneous adipose tissue in different body regions may be of practical
importance, as changes in subcutaneous adipose tissue distribution are also associated with changes in running
performance51. This enables skinfolds to be used as useful predictors of running performance. As noted by other
authors52, also in the presented study, it was found that little subcutaneous fat characterizes groups of short-
and middle-distance runners. Thicker skinfolds are characteristic of long-distance runners, which may reflect
8
Vol:.(1234567890)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
differences in their metabolism52,53. Subcutaneous adipose tissue is an important and most natural reservoir of
energy necessary for athletes to perform long-term efforts54. It also performs important endocrine functions55.
The distribution of subcutaneous fat is slightly different in the studied groups of athletes. Limb fatness relative
to the trunk is low in sprinters and middle-distance runners, which can be explained by muscle group-specific
adipose tissue loss due to systematic training52.
Body composition is also an important characteristic of runners. Its basic components, adipose tissue and
lean mass with cellular and acellular fractions, strongly correlate with the ability to increase muscle strength,
contributing to improved performance and running economy56. Using anthropometric and DXA methods to
analyze body composition, it was shown that the low cost of locomotor energy was associated only with param-
eters indicating relative slenderness of the body46. In a current BIA study, middle- and long-distance runners have
been shown to have a slightly higher percentage of fat and extracellular mass compared to sprinters, reflecting
their slimmer physique27. Extracellular mass is known to include connective tissues such as collagen, elastin,
skin, tendons and bones57,58. In turn, the body cell mass responsible for metabolism is significantly higher in
short-distance runners13. Also, the distribution of individual athletes’ points in the system of three variables of
body composition assessed with the χ2 test showed statistically significant differences between groups. Studies
by other authors confirm that the differences in the body composition of runners are a consequence of different
workloads resulting from the length of the distance covered18,59.
The morphological characteristics of runners are supplemented by the somatotypological assessment, which
allows for determining the size of endomorphy, mesomorphy and ectomorphy in the body structure16. It should
be noted that the athletes are slender with moderate musculature and low body fat, which has also been noted
in other studies. SANOVA showed a statistically significant difference in the somatotypes of the tested runners.
Middle-distance runners are mesomorph-ectomorph (1.60–3.82–3.81), and sprinters (1.68–4.94–2.90) and long-
distance runners (2.11–4.72–3.36) are an ectomorphic mesomorph. The somatotypes of the examined athletes are
similar to the somatotypes of Croatian runners over various distances27 and participants of the Olympic Games
in 198416. Runners from Croatia presenting a higher sports level (were in the top 15 on the Croatian Athletic
Association rank list for the specific event) than the academic runners surveyed in actual research are charac-
terized by higher values of endomorphy and ectomorphy (S: 2.0–4.2–3.0; M: 2.1–3.8–3.3; L: 2.6–3.5–3.7), while
the mesomorphy in the groups of short- and long-distance runners is slightly lower. In turn, the participants
of the 1984 Olympics clearly dominate the size of the mesomorphic component (S: 1.7–5.2–2.8; 400 m runners
1.5–4.6–3.4; M: 1.5–4.3–3.6; L: 1.4–4.2–3.7), which is a consequence of their higher sports level.
Using principal components analysis made it possible to isolate three principal components explaining the
problem in approximately 74%. The first principal component characterizes the overall size and development of
musculature and does not significantly differ between the examined groups of runners. The second component
differs significantly between short-distance and middle-distance competitors. It is positively correlated with
muscle circumferences and negatively correlated with lower limb length, thigh length and lower leg length, con-
firming the observations of other authors indicating the relationship between a more linear body and running
economy46,60. The third principal component characterizes the body’s subcutaneous fat in terms of distribution.
As previous studies have shown, the anthropometric assessment of athletes should include the assessment of all
skinfolds, as the reduction of subcutaneous fat is not the same in all segments. In the case of runners, it concerns
mainly the lower limbs52.
The morphological distinctiveness of distance runners is expressed in overall body size, muscularity and fat-
ness as well as lower limb proportions. These characteristics should be particularly controlled by coaches during
the selection process for individual competitions, as well as throughout the training period. The optimization of
selection processes and training methods based on these characteristics seems to be an important element for
achieving success in running at different distances.
Conclusions
The conducted analyses indicate the diversification of anthropometric profiles in runners of different distances.
Sprinters dominate a more massive body shape, shorter lower legs in relation to the length of the thigh, broader
shoulders and narrower hips, greater musculature and cellular mass. Long-distance runners are characterized
by a slim figure, a high crural index, a slightly wider pelvis in relation to the width of the shoulders, and the
greatest adiposity and extracellular mass. Middle-distance runners are the slimmest, and have a narrow trunk
and little subcutaneous fat. The physique of sprinters and long-distance runners is dominated by mesomor-
phy, while middle-distance runners are mesomorph-ectomorph. The principal component analysis reduced the
multidimensional structure to three variables: overall body size, limbs musculature and the length of the lower
limb together with its segments, and body fatness. This approach emphasizes the morphological distinctiveness
of runners at particular distances and allows the use of somatic features as predictors of running performance.
Our study revealed essential differences between the student athletes in the three running disciplines. How-
ever, in the future we plan to link this anthropometric diversity to the other relevant fields as biomechanics to
find general success predictors in running. Moreover, we would like to confront our results with a similar study
conducted on the cohort of the elite-level runners from the abovementioned disciplines. Therefore, we would be
able to provide coaches and selectors with the robust set of success predictors both in the running performed on
the elite and non-elite level. This way, future research of the morphological diversity of runners should focus on
these somatic features indicated in this manuscript. The future study should be conducted in more homogene-
ous groups in terms of sports level, especially groups of high-level athletes and should analyze the importance
of indicated features for better performance.
9
Vol.:(0123456789)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
Study limitations
The small size of the analyzed groups of athletes may limit the interpretation of the results. While the analysis
was successful, increasing the sample size would have given a clearer picture of intergroup variability. In addi-
tion to that, the competitors were qualified to groups on the basis of a survey concerning the competitions and
distances they competed on. However, this division may not be completed, as it happens that athletes change the
distance during their career, which may affect the picture of morphological diversity. Interpretation may also be
limited by the varied sports’ level of the respondents.
Data availability
The datasets used and/or analyzed during the current study are available from the corresponding author on
reasonable request.
Received: 14 May 2023; Accepted: 15 October 2023
References
1. Pedersen, B. K. & Saltin, B. Exercise as medicine: Evidence for prescribing exercise as therapy in 26 different chronic diseases.
Scand. J. Med. Sci. 25, 1–72 (2015).
2. Janssen, M., Scheerder, J., Thibaut, E., Brombacher, A. & Vos, S. Who uses running apps and sports watches? Determinants and
consumer profiles of event runners’ usage of running-related smartphone applications and sports watches. PLoS One 12, e0181167
(2017).
3. Janssen, M. et al. Understanding different types of recreational runners and how they use running-related technology. IJERPH 17,
2276 (2020).
4. Eime, R. et al. Integrating public health and sport management: SPORT participation trends 2001–2010. Sport Manag. Rev. 18,
207–217 (2015).
5. Barnes, K. R. & Kilding, A. E. Running economy: Measurement, norms, and determining factors. Sports Med. Open 1, 1–15 (2015).
6. Maldonado-Martin, S., Mujika, I. & Padilla, S. Influence of body mass and height on the energy cost of running in highly trained
middle- and long-distance runners. Int. J. Sports Med. 23, 268–272 (2002).
7. Folland, J., Allen, S., Black, M., Handsaker, J. & Forrester, S. Running technique is an important component of running economy
and performance. Med. Sci. Sports Exerc. 49, 1412–1423 (2017).
8. Kyröläinen, H., Belli, A. & Komi, P. V. Biomechanical factors affecting running economy. Med. Sci. Sports Exerc 33, 1330–1337
(2001).
9. Laumets, R. et al. Lower leg length is associated with running economy in high level Caucasian distance runners. J. Hum. Kinet.
15, 229–239 (2017).
10. Pavlović, R. & Kozina, Z. Differences in anthropometric characteristics and body composition of athletes in cyclic endurance-type
activities: A case study. J. Adv. Sport Phys. Educ. 5, 225–234 (2022).
11. Khosla, T. Standards on age, height and weight in Olympic running events for men. Br. J. Sports Med. 12, 97–101 (1978).
12. Longo, A. F., Siffredi, C. R., Cardey, M. L., Aquilino, G. D. & Lentini, N. A. Age of peak performance in Olympic sports: A com-
parative research among disciplines. JHSE 11, 31–41 (2016).
13. Barbieri, D. et al. Body composition and size in sprint athletes. J. Sports Med. Phys. Fit. 57, 1142–1146 (2017).
14. Herrmann, F. R. et al. Running performance in a timed city run and body composition: A cross-sectional study in more than 3000
runners. Nutrition 61, 1–7 (2019).
15. Shafeeq, V. A., Abraham, G. & Raphel, S. Evaluation of body composition and somatotype characteristics of male track and field
athletes in India. J. Exp. Sci. 1, 7–10 (2010).
16. Carter, J. E. L. & Heath, B. H. Somatotyping: Development and Applications (Cambridge University Press, 1990).
17. Carter, L. Somatotyping. in Anthropometrica: A Textbook of Body Measurement for Sports and Health Courses (eds: Kevin Norton
& Tim Olds) 147–170 (University of New South Wales Press, 1996).
18. Muñoz, C., Muros, J., López-Belmonte, Ó. & Zabala, M. Anthropometric characteristics, body composition and somatotype of
elite male young runners. IJERPH 17, 674 (2020).
19. Arazi, H., Mirzaei, B. & Nobari, H. Anthropometric profile, body composition and somatotyping of national Iranian cross-country
runners. Turk. J. Sport Exerc. 17, 35–41 (2015).
20. O’Connor, H., Olds, T. & Maughan, R. Physique and performance for track and field. J. Sports Sci. 25(Suppl 1), S49-60 (2007).
21. Jung, A. P. The impact of resistance training on distance running performance. Sports Med. 33, 539–552 (2003).
22. Blagrove, R. C., Howatson, G. & Hayes, P. R. Effects of strength training on the physiological determinants of middle- and long-
distance running performance: A systematic review. Sports Med. 48, 1117–1149 (2018).
23. Norton, K., Olds, T., Olive, S. & Craig, N. Anthropometry and sports performance. in Anthropometrica: A Textbook of Body Meas-
urement for Sports and Health Courses (eds: Kevin Norton & Tim Olds) 287–364 (University of New South Wales Press, 1996).
24. Stachoń, A. & Pietraszewska, J. Body composition in male physical education university students in view of their physical activity
level. Hum. Mov. 14, 205–209 (2013).
25. Goulding, M. Somatotype—Calculation and Analysis (Sweat Technologies, 2002).
26. Kaiser, H. F. The application of electronic computers to factor analysis. Educ. Psychol. Meas. 20, 141–151 (1960).
27. Vucetic, V., Matkovic, B. & Sentija, D. Morphological differences of elite Croatian track-and-field athletes. Coll. Antropol. 32,
863–868 (2008).
28. Kułaga, Z. et al. Polish 2010 growth references for school-aged children and adolescents. Eur. J. Pediatr. 170, 599–609 (2011).
29. Sedeaud, A. et al. BMI, a performance parameter for speed improvement. PLoS One 9, e90183 (2014).
30. Fujita, S. et al. A 100-m sprint time is associated with deep trunk muscle thickness in collegiate male sprinters. Front. Sports Act.
Living 24, 32 (2019).
31. Tottori, N. et al. Trunk and lower limb muscularity in sprinters: What are the specific muscles for superior sprint performance?.
BMC Res. Notes 14, 74 (2021).
32. Weyand, P. G. & Davis, J. A. Running performance has a structural basis. J. Exp. Biol. 208, 2625–2631 (2005).
33. Bastir, M. et al. Variation in human 3D trunk shape and its functional implications in hominin evolution. Sci. Rep. 12, 11762 (2022).
34. Castillo, E. R. & Lieberman, D. E. Shock attenuation in the human lumbar spine during walking and running. J. Exp. Biol. 221,
jeb177949 (2018).
35. Copaver, K., Hertogh, C. & Hue, O. The effects of psoas major and lumbar lordosis on hip flexion and sprint performance. Res. Q.
Exerc. Sport. 83, 160–167 (2012).
36. Ueno, H. et al. The potential relationship between leg bone length and running performance in well-trained endurance runners.
J. Hum. Kinet. 70, 165–172 (2019).
10
Vol:.(1234567890)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
37. Majumdar, A. S. & Robergs, R. A. The science of speed: Determinants of performance in the 100 m sprint. Int. J. Sports Sci. Coach
6, 479–493 (2011).
38. Kriswanto, E. S. et al. Effect of leg length on running speed of sports and health sciences students in Indonesia: A meta-analysis
study. JPES 21, 2697–2705 (2021).
39. Mooses, M. et al. Anthropometric and physiological determinants of running performance in middle-and long-distance runners.
Kinesiology 45, 154–162 (2013).
40. Bushnell, T. & Hunter, I. Differences in technique between sprinters and distance runners at equal and maximal speeds. Sports
Biomech. 6, 261–268 (2007).
41. Mackala, K. Optimisation of performance through kinematic analysis of the different phases of the 100 metres. New Stud. Athl.
22, 7–16 (2007).
42. Mackala, K. & Mero, A. A kinematics analysis of three best 100 m performances ever. J. Hum. Kinet. 36, 149–160 (2013).
43. Bereket, S. Effects of anthropometric parameters and stride frequency on estimation of energy cost of walking. J. Sports Med. Phys.
Fit. 45, 152–161 (2005).
44. Wall-Scheffler, C. M. Size and shape: Morphology’s impact on human speed and mobility. J. Anthropol. https:// doi. org/ 10. 1155/
2012/ 340493 (2012).
45. Tomita, D. et al. A pilot study on a potential relationship between leg bone length and sprint performance in sprinters; Are there
any event-related differences in 100-m and 400-m sprints?. BMC Res. Notes 13, 1–6 (2020).
46. Black, M. I., Allen, S. J., Forrester, S. E. & Folland, J. P. The anthropometry of economical running. Med. Sci. Sports Exerc. 52,
762–770 (2020).
47. Nikander, R., Sievänen, H., Uusi-Rasi, K., Heinonen, A. & Kannus, P. Loading modalities and bone structures at nonweight-bearing
upper extremity and weight-bearing lower extremity: A pQCT study of adult female athletes. Bone 39, 886–894 (2006).
48. Weidauer, L., Eilers, M., Binkley, T., Vukovich, M. & Specker, B. Effect of different collegiate sports on cortical bone in the tibia. J.
Musculoskelet. Neuronal Interact. 12, 68–73 (2012).
49. Korhonen, M. et al. Biomechanical and skeletal muscle determinants of maximum running speed with aging. Med. Sci. Sports
Exerc. 41, 844–856 (2009).
50. Perez-Gomez, J. et al. Role of muscle mass on sprint performance: Gender differences?. Eur. J. Appl. Physiol. 102, 685–694 (2008).
51. Abe, T. et al. Longitudinal associations between changes in body composition and changes in sprint performance in elite female
sprinters. Eur. J. Sport Sci. 20, 100–105 (2020).
52. Arrese, A. L. & Ostáriz, E. S. Skinfold thicknesses associated with distance running performance in highly trained runners. J. Sports
Sci. 24, 69–76 (2006).
53. Arrese, A. L., Badillo, J. J. G. & Ostáriz, E. S. Differences in skinfold thicknesses and fat distribution among top-class runners.
Sports Med. Phys. Fit. 45, 512–517 (2005).
54. Kolnes, K. J., Petersen, M. H., Lien-Iversen, T., Højlund, K. & Jensen, J. Effect of exercise training on fat loss—Energetic perspec-
tives and the role of improved adipose tissue function and body fat distribution. Front. Physiol. 12, 1634 (2021).
55. Kershaw, E. E. & Jeffrey, S. F. Adipose tissue as an endocrine organ. J. Clin. Endocrinol. Metab. 89, 2548–2556 (2004).
56. Allen, S. J., Black, M. I., Forrester, S. E., Handsaker, J. C. & Folland, J. P. The relationship of anthropometry and body composition
with running economy. in 35th Conference of the International Society of Biomechanics in Sports vol. 35 1080–1083 (2017).
57. Heymsfield, S., Wang, Z., Baumgartner, R. & Ross, R. Human body composition: Advances in models and methods. Annu. Rev.
Nutr. 17, 527–558 (1997).
58. Talluri, T., Lietdke, R. J., Evangelisti, A., Talluri, J. & Maggia, G. Fat-free mass qualitative assessment with bioelectric impedance
analysis (BIA). Ann. N. Y. Acad. Sci. 873, 94–98. https:// doi. org/ 10. 1111/j. 1749- 6632. 1999. tb094 54.x (1999).
59. Aikawa, Y., Murata, M. & Omi, N. Relationship of height, body mass, muscle mass, fat mass, and the percentage of fat with athletic
performance in male Japanese college sprinters, distance athletes, jumpers, throwers, and decathletes. J. Phys. Fit. Sports Med. 9,
7–14 (2020).
60. Watts, A. S., Coleman, I. & Nevill, A. The changing shape characteristics associated with success in world-class sprinters. J. Sports
Sci. 30, 1085–1095 (2012).
Acknowledgements
The authors would like to thank Justyna Andrzejewska for support regarding data collection. Also the authors
thank all the participants of the survey for their understanding and the time devoted and anonymous reviewers
for their helpful comments.
Author contributions
A.B. designed and conducted the research, obtained the data, statistically analyzed the whole data set, wrote the
paper and prepared figures, A.S. and J.P. obtained the data and reviewed the paper. All authors have read and
agreed to the published version of the manuscript.
Funding
This research was funded by Wrocław University of Health and Sport Sciences, Poland, grant number PN/
BK/2020/08.
Competing interests
The authors declare no competing interests.
Additional information
Correspondence and requests for materials should be addressed to A.B.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
11
Vol.:(0123456789)
Scientific Reports | (2023) 13:18222 |
https://doi.org/10.1038/s41598-023-45064-9
www.nature.com/scientificreports/
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.
© The Author(s) 2023
| Anthropometric profiles and body composition of male runners at different distances. | 10-25-2023 | Stachoń, Aleksandra,Pietraszewska, Jadwiga,Burdukiewicz, Anna | eng |
PMC9266034 | Citation: Yang, S.-J.; Yang, F.; Gao, Y.;
Su, Y.-F.; Sun, W.; Jia, S.-W.; Wang, Y.;
Lam, W.-K. Gender and Age
Differences in Performance of Over
70,000 Chinese Finishers in the Half-
and Full-Marathon Events. Int. J.
Environ. Res. Public Health 2022, 19,
7802. https://doi.org/10.3390/
ijerph19137802
Academic Editors: Paul B.
Tchounwou and Ukadike
Chris Ugbolue
Received: 11 May 2022
Accepted: 23 June 2022
Published: 25 June 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright:
© 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed
under
the
terms
and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
International Journal of
Environmental Research
and Public Health
Article
Gender and Age Differences in Performance of Over
70,000 Chinese Finishers in the Half- and Full-Marathon Events
San-Jun Yang 1, Fan Yang 1,2,*
, Yuan Gao 3, Yan-Feng Su 4, Wei Sun 4, Sheng-Wei Jia 2, Yu Wang 5,*
and Wing-Kai Lam 6,*
1
Department of Physical Education and Research, China University of Mining and Technology—Beijing,
Beijing 100083, China; [email protected]
2
Li Ning Sports Science Research Center, Li Ning (China) Sports Goods Company Limited,
Beijing 101111, China; [email protected]
3
School of Physical Education, Yanshan University, Qinhuangdao 066004, China; [email protected]
4
School of Physical Education and Coaching, Shanghai University of Sport, Shanghai 200438, China;
[email protected] (Y.-F.S.); [email protected] (W.S.)
5
School of Kinesiology and Health, Capital University of Physical Education and Sports, Beijing 100091, China
6
Sports Information and External Affairs Centre, Hong Kong Sports Institute, Sha Tin, Hong Kong
*
Correspondence: [email protected] (F.Y.); [email protected] (Y.W.);
[email protected] (W.-K.L.); Tel.: +86-189-11326682 (F.Y.); +86-135-01126242 (Y.W.);
+86-186-11783188 (W.-K.L.)
Abstract: (1) Background: The aim of the present study was to examine the characteristics of over
70,000 long-distance finishers over the last four years in Chinese half- and full-marathon events;
(2) Methods: The available data of all finishers (n = 73,485; women, n = 17,134; men, n = 56,351) who
performed half- and full-marathon events in Hangzhou from 2016 to 2019 were further analyzed for
the characteristics of gender, age and average running speed; (3) Results: The total men-to-women
ratio was the lowest in the half-marathon event (1.86) and the highest in the full-marathon event
(17.42). Faster running performance in males than in females and faster average running speed
in short-distance runners were shown. Gender and race distance were observed to have the most
significant effects on average running speed (p < 0.01). For both male and female finishers, the slowest
running speed was shown in older age groups (p < 0.01) during the full marathon. Our results
indicated that the gender difference in performance was attenuated in the longer race distances and
older age groups; (4) Conclusions: Understanding the participation and performances across different
running distances would provide insights into physiological and biomechanical characteristics for
training protocols and sports gear development in different groups.
Keywords: gender; men-to-women ratio; marathon; age; running speed
1. Introduction
The health benefits of endurance exercise might partially explain the increase in
participation in marathon races during the last decades [1]. In recent years, marathon
running has been considered a globally popular physical activity that can cater to the
various healthy lifestyle needs of urban residents [2,3]. This running boom has gradually
spread around the world. Well-known New York, London, Paris, and Berlin marathon
events all had between 30,000 and 40,000 finishers [4]. While distance running used to
be a male-dominated sport, today females account for 43% of marathon runners in the
USA [5]. Marathon events have developed later in China than in Western countries. The
number of marathon events held in China increased from 12 to 53 between 2010 and
2014 [6] to approximately 1100 in 2017 [7], which involved nearly 5 million participants
and increased to over 2.2 million participants from 2016 to 2017 [7]. The number further
increased to a total of 1900 events in 2019 [1]. The growing popularity of running has
inspired a large amount of research on running biomechanics, performance and sport
Int. J. Environ. Res. Public Health 2022, 19, 7802. https://doi.org/10.3390/ijerph19137802
https://www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2022, 19, 7802
2 of 9
gears in the past decades [8]. Studies have found that proper pace can effectively reduce
the risk of musculoskeletal injury [9], and different running strategies should be used in
long-distance running according to gender, age and the event the runner is training for [10].
Moreover, differences in running biomechanics between Chinese men and women were
observed, with female runners showing greater range of motion in the hip and knee joints,
and a smaller shoe-to-ground angle during the heel touch-down phase. This is believed to
be a form of self-regulation that women use to reduce the impact of landing, and men rely
more on the performance of their shoes to achieve the purpose of buffering [11].
Following the increase in female participation in distance running [12], investigations
into gender differences in running mechanics were intensified in the Western world [12–14].
The gender difference studies indicated clear differences in females’ body fat and running
speed [3], resulting in distinct movement characteristics and injury etiology. Nikolaidis
et al. studied the performance and age composition of different genders during marathons,
where they found that women achieved their best marathon race time ~5 years earlier
in life compared to men. Women’s participation increased disproportionately to men’s
participation, leading to an increase in the ratio of men to women [15]. In addition, more
and more seniors are joining in marathon races [16]. The sex gap between elite female
marathon racers and elite male runners may have reached its limit [17]. The age structure
of most male marathon runners is larger and older than that of females (male: 40–44 years;
female: 30–34 years) [18].
Although studies have been conducted on anthropometry, physiology and training
characteristics have improved our understanding of the predictors of race time [19], as well
as age- and gender-related differences in pacing during endurance running [20–23]. While
Western counterparts have been extensively analyzed regarding their running characteris-
tics, little attention has been paid to the Chinese population. Several articles have shown
differences in running between Chinese and Western populations; one study found that,
compared with Western women, Chinese women use the medial forefoot more during the
push-off phase of running [24]. Western female runners have greater ankle valgus angles
than males, while there is no significant difference between Chinese females and males [25].
Thus, research findings obtained from Western runners may not be directly applicable to
the Chinese population because of racial differences. The present study examined trends of
the men-to-women ratio, number of finishers and performances by gender and age groups
across four years in half- and full-marathon events, respectively. It was expected that
there would be a different men-to-women ratio and different performances across different
running events. The comparison could help to better understand potential training require-
ments for both elite and recreational runners of different age groups, as well as to effectively
estimate the demand of running shoes for different gender and age groups. To optimize
female performance and health in sport, we need to include women in our analyses in
order to better understand peculiarities that may exist in physiology. Therefore, we are
happy to enrich the existing pool of knowledge with more data on female participation
and performance in marathon racing. Understanding the participation and performance
across different running distances (half- and full-marathon events) would provide insights
into the physiological and biomechanical characteristics for training protocols for different
gender and age groups.
2. Materials and Methods
2.1. Participants and Data Acquisition
The complete marathon event data for this study were officially obtained from
Hangzhou Marathon Organizing Committee (https://www.hzim.org) [26]. The records
were collected from all half marathons and full marathons between 2016 and 2019, and
were officially certified by the World Athletics Organization. The Hangzhou marathon
event included both full and half marathons and runs. Regretfully, the Hangzhou marathon
event has been suspended due to the spread of COVID-19 in 2019. The study data included
participants who completed the race in the appropriate amount of time. Age and gender
Int. J. Environ. Res. Public Health 2022, 19, 7802
3 of 9
information was provided for the period of 4 years. Ultimately, the study included a total
of 73,485 participants (male, n = 56,351; female, n = 17,134).
2.2. Procedures
Age intervals of five years were selected to represent age groups among younger
and older finishers in their categories. All runners over 71 years old were placed in one
category, as there were only a few male runners in the oldest age group, while the oldest
male runner was 74 years old. In total, the finishers were classified into 11 age groups;
21–25, 26–30, 31–35, 36–40, 41–45, 45–50, 51–55, 56–60, 61–65, 66–70 and 71+ years. Changes
in gender participation are described by the men-to-women ratio (MWR, the quotient of
males divided by female completers) [27].
2.3. Statistical Analysis
The official race time (i.e., accurate in seconds) was obtained for all finishers in both
races. The average running speed in km/h was calculated using the final race time (h)
divided by race distance (km) to allow comparison of performances between two long-
distance races. All descriptive statistics were reported as mean and standard deviation.
Prior to statistical analyses, data distribution normality was verified by visual inspection
of histograms and QQ plots [18]. To assess age and gender distribution among finishers
in the half- and full-marathon events, a chi-square test (χ2) was performed. Statistical
differences in marathon performance between 11 age groups and two events were observed.
Meanwhile, their interactions were calculated using a two-way ANOVA, post hoc with
Bonferroni-corrected tests, and the significance level was controlled at 0.05. All data were
organized and summarized using Microsoft Office Excel 2019 (Microsoft Corporation,
Redmond, WA, USA) and statistical testing was performed using SPSS 20.0 (IBM, Armonk,
NY, USA).
3. Results
3.1. Participation by Gender, Race Distance, and Age Group
The MWR as well as the total number of male and female finishers in each age group
and race distance are presented in Table 1.
Table 1. Distribution of male and female finishers in each age group and race distance.
Half-Marathon
Full-Marathon
Age Groups
Males
Females
Total
MWR
Males
Females
Total
MWR
21–25
940
505
1445
1.86
639
158
797
4.04
26–30
3667
2166
5833
1.69
3046
714
3760
4.27
31–35
4381
2063
6444
2.12
5110
1073
6183
4.76
36–40
4360
1712
6072
2.55
6274
1249
7523
5.02
41–45
3514
1538
5052
2.28
6733
1325
8058
5.08
46–50
2820
1404
4224
2.01
6573
1386
7959
4.74
51–55
1428
633
2061
2.26
3569
678
4247
5.26
56–60
618
179
797
3.45
1540
193
1733
7.98
61–65
208
47
255
4.43
502
64
566
7.84
66–70
111
20
131
5.55
209
12
221
17.42
71+
28
0
28
-
18
0
18
-
Total
22,075
10,267
32,342
2.15
34,213
6852
41,065
4.99
MWR = men-to-women ratio.
The total MWR was 2.15 and 4.99 in the half and full marathon, respectively. A
gender × race distance association in participation was shown (χ2 = 2294.505, p < 0.01,
ϕ = 0.177). A gender × age group association in participation was observed in the half
marathon (χ2 = 89.091, p < 0.01, ϕ = 0.081) and in the full marathon (χ2 = 53.431, p < 0.01,
ϕ = 0.050). Furthermore, a race distance × age group association in participation was
Int. J. Environ. Res. Public Health 2022, 19, 7802
4 of 9
shown for male finishers (χ2 = 1687.054, p < 0.01, ϕ = 0.187) and for female finishers
(χ2 = 600.388, p < 0.01, ϕ = 0.197) of different age groups. In the half marathon, the lowest
MWR was observed in the age group of 26–30 years (1.69), whereas the highest MWR was
observed in the age group of 36–40 years (6.12). In the full marathon, the lowest MWR was
observed in the youngest age group (4.04), whereas the highest MWR was observed in the
age group of 66–70 years (17.41).
3.2. Performance (Average Running Speed)
3.2.1. Overall Effects
The two-way ANOVA showed significant effects of gender [F (1, 73,365) = 612.757,
p < 0.001] and race distance [F (1, 73,365) = 7.914, p < 0.005], as well as age group
[F (10, 73,365) = 119.550, p < 0.001]. Moreover, we found significant interactions for age
group × race distance [F (10, 73,365) = 3.763, p < 0.001], while no interaction was observed
for gender × race distance [F (1, 73,365) = 3.270, p = 0.071] and for gender × age group
[F (9, 73,365) = 1.329, p = 0.216].
3.2.2. Performance by Gender and Race Distance
A significant effect of gender on average running speed is shown (p < 0.001) in Figure 1,
where male finishers (with the performance of 10.03 ± 1.67 km/h) were faster than female
finishers (with the performance of 9.08 ± 1.27 km/h). In addition, a significant effect of
race distance on average running speed was observed (p < 0.001). Figure 1 also shows
that performance in the full-marathon event (9.95 ± 1.71 km/h) was faster than that in the
half-marathon event (9.63 ± 1.53 km/h).
0.177). A gender × age group association in participation was observed in the half mara
thon (χ2 = 89.091, p < 0.01, φ = 0.081) and in the full marathon (χ2 = 53.431, p < 0.01, φ =
0.050). Furthermore, a race distance × age group association in participation was shown
for male finishers (χ2 = 1687.054, p < 0.01, φ = 0.187) and for female finishers (χ2 = 600.388,
p < 0.01, φ = 0.197) of different age groups. In the half marathon, the lowest MWR was
observed in the age group of 26–30 years (1.69), whereas the highest MWR was observed
in the age group of 36–40 years (6.12). In the full marathon, the lowest MWR was observed
in the youngest age group (4.04), whereas the highest MWR was observed in the age group
of 66–70 years (17.41).
3.2. Performance (Average Running Speed)
3.2.1. Overall Effects
The two-way ANOVA showed significant effects of gender [F (1, 73,365) = 612.757, p <
0.001] and race distance [F (1, 73,365) = 7.914, p < 0.005], as well as age group [F (10, 73,365) =
119.550, p < 0.001]. Moreover, we found significant interactions for age group × race dis-
tance [F (10, 73,365) = 3.763, p < 0.001], while no interaction was observed for gender × race
distance [F (1, 73,365) = 3.270, p = 0.071] and for gender × age group [F (9, 73,365) = 1.329, p = 0.216].
3.2.2. Performance by Gender and Race Distance
A significant effect of gender on average running speed is shown (p < 0.001) in Figure
1, where male finishers (with the performance of 10.03 ± 1.67 km/h) were faster than fe-
male finishers (with the performance of 9.08 ± 1.27 km/h). In addition, a significant effect
of race distance on average running speed was observed (p < 0.001). Figure 1 also shows
that performance in the full-marathon event (9.95 ± 1.71 km/h) was faster than that in the
half-marathon event (9.63 ± 1.53 km/h).
No gender × race distance interaction on average running speed was found (p > 0.05),
while the gender difference was lower in the half-marathon event (+9.98%) than in the
full-marathon event (+10.25%). Half-marathon finishers were slower than full-marathon
finishers among females (9.02 ± 1.22 versus 9.17 ± 1.35 km/h, respectively, p < 0.001), as
well as among males (9.92 ± 1.57 versus 10.11 ± 1.73 km/h, respectively, p < 0.001).
Figure 1.
Race speed by race distance and gender.
Error bars represent standard deviations.
ˆˆ p < 0.001; ** p < 0.001.
No gender × race distance interaction on average running speed was found (p > 0.05),
while the gender difference was lower in the half-marathon event (+9.98%) than in the
full-marathon event (+10.25%). Half-marathon finishers were slower than full-marathon
finishers among females (9.02 ± 1.22 versus 9.17 ± 1.35 km/h, respectively, p < 0.001), as
well as among males (9.92 ± 1.57 versus 10.11 ± 1.73 km/h, respectively, p < 0.001).
Int. J. Environ. Res. Public Health 2022, 19, 7802
5 of 9
3.2.3. Performance by Age Group and Race Distance
The age group × race distance interaction had a significant effect on average running
speed (p < 0.001). Under closer examination of the performance, male finishers had the
fastest average running speed of 10.33 ± 1.60 km/h, while female finishers had the slowest
average running speed of 8.54 ± 1.03 km/h, regardless of the type of event.
In the half-marathon event, the fastest male age group was 61–65 years (average
running speed of 10.32 ± 1.45 km/h), while the slowest male age group was 26–30 years
(average running speed of 9.61 ± 1.57 km/h). In the full-marathon event, the male finishers
had the fastest speed of 10.39 ± 1.62 km/h in the 46–50 age interval and the slowest speed
of 9.16 ± 1.18 km/h in the 71+ age interval (Figure 2a).
Figure 1. Race speed by race distance and gender. Error bars represent standard deviations. ^^ p <
0.001; ** p < 0.001.
3.2.3. Performance by Age Group and Race Distance
The age group × race distance interaction had a significant effect on average running
speed (p < 0.001). Under closer examination of the performance, male finishers had the
fastest average running speed of 10.33 ± 1.60 km/h, while female finishers had the slowest
average running speed of 8.54 ± 1.03 km/h, regardless of the type of event.
In the half-marathon event, the fastest male age group was 61–65 years (average run-
ning speed of 10.32 ± 1.45 km/h), while the slowest male age group was 26–30 years (av-
erage running speed of 9.61 ± 1.57 km/h). In the full-marathon event, the male finishers
had the fastest speed of 10.39 ± 1.62 km/h in the 46–50 age interval and the slowest speed
of 9.16 ± 1.18 km/h in the 71+ age interval (Figure 2a).
In the half-marathon event, female finishers had the fastest average running speed of
9.61 ± 1.21 km/h in the 66–70 age interval, but the slowest average running speed of 8.65 ±
1.13 km/h in the 26–30 age group. In the full-marathon event, the fastest female finisher
was observed in the 46–50 age group (9.43 ± 1.28 km/h), while the slowest female was in
the 66–70 age group (8.48 ± 0.91 km/h) (Figure 2b).
Figure 2. Race speed by age group and gender in the half-marathon event (a) and in the full-mara-
thon event (b). Error bars represent standard deviations. ## p < 0.01; # p < 0.05. ^^ p < 0.01; ^ p < 0.05.
3.2.4. Performance by Gender and Age Group
Although the main effect of gender × age group interaction on the average running
speed (p = 0.216) was obtained, Bonferroni post hoc comparisons revealed a significant
difference (Table 2).
In both male and female finishers, there was a significant difference between the age
group of 21–25 years and the age group of 36–60 years (p < 0.01), as well as the age group
of 61–55 years (p < 0.05). The same difference was found in the age groups of 26–30 years
and 31–65 years (p < 0.01).
There was also a difference in the average running speed between age groups of 31–
35 years, 36–60 years (p < 0.01) and 61–55 years (p < 0.05). Another significant difference
was observed between the 36–40 age interval and the 41–55 age interval (p < 0.01). The
performance of the 46–50 age interval showed a significant difference with the 41–45 age
interval (p < 0.01) and the 51–55 age interval (p < 0.01).
Figure 2. Race speed by age group and gender in the half-marathon event (a) and in the full-marathon
event (b). Error bars represent standard deviations. ˆˆ p < 0.01; ˆ p < 0.05.
In the half-marathon event, female finishers had the fastest average running speed
of 9.61 ± 1.21 km/h in the 66–70 age interval, but the slowest average running speed of
8.65 ± 1.13 km/h in the 26–30 age group. In the full-marathon event, the fastest female
finisher was observed in the 46–50 age group (9.43 ± 1.28 km/h), while the slowest female
was in the 66–70 age group (8.48 ± 0.91 km/h) (Figure 2b).
3.2.4. Performance by Gender and Age Group
Although the main effect of gender × age group interaction on the average running
speed (p = 0.216) was obtained, Bonferroni post hoc comparisons revealed a significant
difference (Table 2).
In both male and female finishers, there was a significant difference between the age
group of 21–25 years and the age group of 36–60 years (p < 0.01), as well as the age group
of 61–55 years (p < 0.05). The same difference was found in the age groups of 26–30 years
and 31–65 years (p < 0.01).
There was also a difference in the average running speed between age groups of
31–35 years, 36–60 years (p < 0.01) and 61–55 years (p < 0.05). Another significant difference
was observed between the 36–40 age interval and the 41–55 age interval (p < 0.01). The
performance of the 46–50 age interval showed a significant difference with the 41–45 age
interval (p < 0.01) and the 51–55 age interval (p < 0.01).
Int. J. Environ. Res. Public Health 2022, 19, 7802
6 of 9
Table 2. Bonferroni post hoc tests of age groups.
Age Groups
21–25
26–30
31–35
36–40
41–45
46–50
51–55
56–60
61–65
66–70
71+
21–25
-
-
-
**
**
**
**
**
*
-
-
26–30
-
-
**
**
**
**
**
**
**
-
-
31–35
-
##
-
**
**
**
**
**
*
-
-
36–40
##
##
##
-
**
**
**
-
-
-
-
41–45
##
##
##
##
-
**
-
-
-
-
-
46–50
##
##
##
##
##
-
**
-
-
-
-
51–55
##
##
##
##
##
-
-
-
-
-
56–60
##
##
##
-
-
-
-
-
-
-
-
61–65
#
##
#
-
-
-
-
-
-
-
-
66–70
-
-
-
-
-
-
-
-
-
-
-
71+
-
-
-
-
-
-
-
-
-
-
-
* represent male, # represent female, ** or ## p < 0.01; * or # p < 0.05.
4. Discussion
The aim of this study was to assess the age, gender and average speed characteristics
of over 70,000 long-distance finishers over the last four years.
We found that the 31–35 and 36–40 age groups had the largest number of male
finishers in the half-marathon event, while the 26–30 and 36–40 age groups had the largest
number of female finishers. These findings are partly in agreement with previous studies,
which reported that, regardless of gender, the largest number of finishers was found
in the 24–34 age group [28]. In addition, a study conducted in Switzerland found that
the two largest numbers of full-marathon finishers were in the age ranges 40–44 and
45–49 in men, but 35–39 and 40–44 in women; this study also reported that most half-
marathon participants were 30–34 years old and 45–49 years old in men, but 25–29 years
old and 30–34 years old in women [29]. Although there are differences among age groups
in Asian and European populations, over 30 years of age appeared to be the dominant
age in participants in long-distance running, which may be related to the relationship
with physical needs, social influences and disposable time [30]. The study showed that
the largest age group participating in full marathon events was usually older than the
half-marathon runners. Such results may be due to older runners being more emotionally
stable and responsible than younger people [31].
In terms of the number of finishers by race distance, there were ~1.3 times more
marathon runners than half-marathon runners, which is not in agreement with the previous
findings. The reason for this phenomenon may be that the marathon organizing committee
limited the number of half-marathon runners during the registration period. In Switzerland,
the total number of half-marathon runners was about 2.6 times higher than the marathon
runners between 2000 and 2010 [32]. In Greece, the total number of half-marathon runners
was about 3.5 times higher than that in the Oslo Marathon between 2008 and 2018 [28].
This could be explained by the fact that the registered number of half-marathon events is
substantially lower than the full-marathon events held in China [33].
It is not difficult to understand the phenomenon that the number of people who finish
the half marathon is higher than the number of people who finish the full marathon. The
demand for physical fitness in the full marathon is undoubtedly higher. Running long
distances can lead to dehydration, damage to muscle tissue and an increase in body tem-
perature, which must be more difficult to overcome during a full marathon [34]. Moreover,
the depletion of glycogen stores in the body causes athletes to “hit the wall (HTW)”; the
frequency of HTW among elite and non-elite runners was 51%, with the greatest proportion
being found in non-elite runners. Thus, HTW also increases the risk of withdrawal as the
distance of the competition increases [35].
Our data confirmed that a higher number of male finishers than females was also
observed in a longer-distance event, which is similar to the “Marathon des Sables” (7-day
competition) with a men-to-women ratio (MWR) of 6.76 [3], the “Western States 100-Mile
Endurance Run” (161 km) with an MWR of 5.28 [36], or Double Iron Ultra-Triathlon
Int. J. Environ. Res. Public Health 2022, 19, 7802
7 of 9
(MWR: 8.96) to Deca Iron Ultra-Triathlon (MWR: 6.94) [37]. However, over the past few
years, an increase in women finishers was observed in half- and full-marathon events, as
indicated by the decreased trend of MWR. A plausible explanation for the MWR variation
by race distances/intensities might be the consideration of females as relatively “novice”
runners compared to males. One study reported that the MWR was decreased from 10.2
in the 1970s to 1.5 in the 2010s in the “New York City Marathon” [38]. Women complete
shorter-distance races first and then longer-distance races according to their ability and
physiological features. The majority opted for the half marathon, resulting in a higher
number of female finishers in the half marathon than in the full marathon. Regarding
changes in finishers by gender and age group, similar trends were shown across all race
distances, with older groups having a higher MWR than younger age groups.
The current findings showed faster running performance in males than in females
and faster average running speed in shorter race distances. This may be attributable to the
innate physiological advantage of male runners and gender differences in training habits,
with males having greater body weights and lower body fat percentages in terms of physiol-
ogy [3]. The research pointed out that men and women are born with different muscle fiber
properties, and the advantage of men is that the circumference of muscle fibers is larger
than that of women, so they are more powerful [39]. This is the difference between men
and women in gene expression in human skeletal muscle [40]. Furthermore, male athletes
exhibit higher maximal oxygen uptake (VO2max) and anaerobic thresholds than females in
long-distance running [41]; there is a positive correlation between VO2max and thermoreg-
ulatory ability [42]. Some articles have pointed out that in the non-competition training
phase, men have longer training distances than women and have more weekly training
sessions, so the running experience of male runners is higher than that of women [43];
the same goes for training differences for a half marathon [3]. Some researchers have also
found that the larger male-to-female ratio in the older group is due to the lower number of
female finishers [44]. Our data also found that the average pace of women in half-marathon
events increased with age. In previous research, the slowest pace occurred in the older age
groups [22]. One of the reasons for this could be that most young female participants in the
half marathon were attempting long-distance running for the first time, which makes the
pace slower in younger age groups. At the same time, the running speeds of older Chinese
women groups were better than Western women. Eastern women are morphologically
thinner than Western women [38], thus saving more running economy in long-distance
running and are, therefore, faster on average.
There were several limitations in the present study. First of all, the role of environ-
mental conditions, such as detailed data on temperature, humidity, and wind, was not
considered. It is reasonable to involve the effect of these environmental parameters on
the endurance performance [45]. Secondly, the number of participants was higher than
other studies, as four-year data were accessible. Long-term research should be conducted
in future. Thirdly, other endurance events such as 10 km races should be included.
5. Conclusions
This study found that the number of female finishers in the half- and full-marathon
races has increased, but, overall, there are still more males than females. A higher number
of younger female finishers participated in both race distances. Moreover, the overall
performance of male’s running is better than female’s running, but as the race distance and
the age of participants increased, the difference in performance caused by gender gradually
weakened. It should be emphasized that the analysis of performance trends is related to
changes in MWR based on age group and race distance.
Int. J. Environ. Res. Public Health 2022, 19, 7802
8 of 9
Author Contributions: Conceptualization, Y.-F.S., F.Y. and S.-J.Y.; methodology, F.Y. and S.-J.Y. and
W.-K.L.; software, F.Y., Y.G. and S.-W.J.; validation, Y.G., S.-J.Y. and W.-K.L.; formal analysis, F.Y.;
investigation, S.-J.Y., W.S. and Y.-F.S.; resources, S.-J.Y.; data curation, S.-W.J., S.-J.Y., Y.W. and W.S.;
writing—original draft preparation, S.-J.Y.; writing—review and editing, W.-K.L. and F.Y.; visualiza-
tion, S.-J.Y. and W.-K.L.; supervision, S.-J.Y. and Y.W.; project administration, F.Y.; funding acquisition,
F.Y. and W.-K.L. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by Beijing Technology and Innovation Service Development
Research Fund, grant number 154218001.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or
in the decision to publish the results.
References
1.
Ren, Z.; Zuo, Y.; Ma, Y.; Zhang, M.; Smith, L.; Yang, L.; Loprinzi, P.D.; Yu, Q.; Zou, L. The Natural Environmental Factors
Influencing the Spatial Distribution of Marathon Event: A Case Study from China. Int. J. Environ. Res. Public Health 2020, 17, 2238.
[CrossRef] [PubMed]
2.
Guicciardi, M.; Carta, M.; Pau, M.; Cocco, E. The Relationships between Physical Activity, Self-Efficacy, and Quality of Life in
People with Multiple Sclerosis. Behav. Sci. 2019, 9, 121. [CrossRef] [PubMed]
3.
Knechtle, B.; Knechtle, P.; Rosemann, T.; Senn, O. Sex differences in association of race performance, skin-fold thicknesses, and
training variables for recreational half-marathon runners. Percept. Mot. Ski. 2010, 111, 653–668. [CrossRef] [PubMed]
4.
Hennig, E.M. Eighteen years of running shoe testing in Germany—A series of biomechanical studies. Footwear Sci. 2011, 3,
71–81. [CrossRef]
5.
Statistics, R.M. Marathon Statistics 2019 Worldwide. Available online: https://runrepeat.com/research-marathon-performance-
across-nations (accessed on 7 March 2020).
6.
Zuo, Y.; Zou, L.; Zhang, M.; Smith, L.; Yang, L.; Loprinzi, P.D.; Ren, Z. The temporal and spatial evolution of marathons in China
from 2010 to 2018. Int. J. Environ. Res. Public Health 2019, 16, 5046. [CrossRef]
7.
Xing, X.; Zhang, R.; Taks, M. The effects of health, social, and consumption capital on running-related expenditures in China. Eur.
Sport Manag. Q. 2020, 22, 1–21. [CrossRef]
8.
Trudeau, M.B.; Willwacher, S.; Weir, G.; Rohr, E.; Ertel, C.; Bruggemann, G.-P.; Hamill, J. A novel method for estimating an
individual’s deviation from their habitual motion path when running. Footwear Sci. 2019, 11, 135–145. [CrossRef]
9.
De Koning, J.J.; Foster, C.; Bakkum, A.; Kloppenburg, S.; Thiel, C.; Joseph, T.; Cohen, J.; Porcari, J.P. Regulation of pacing strategy
during athletic competition. PLoS ONE 2011, 6, e15863. [CrossRef]
10.
Cuk, I.; Nikolaidis, P.T.; Villiger, E.; Knechtle, B. Pacing in Long-Distance Running: Sex and Age Differences in 10-km Race and
Marathon. Medicina 2021, 57, 389. [CrossRef]
11.
Isherwood, J.; Wang, H.; Sterzing, T. Running biomechanics and running shoe perception of Chinese men and women. Footwear
Sci. 2020, 13, 1–13. [CrossRef]
12.
Lilley, K.; Dixon, S.; Stiles, V. A biomechanical comparison of the running gait of mature and young females. Gait Posture 2011, 33,
496–500. [CrossRef] [PubMed]
13.
Sinclair, J.; Selfe, J. Sex differences in knee loading in recreational runners. J. Biomech. 2015, 48, 2171–2175. [CrossRef] [PubMed]
14.
Davis, I.S.; Bowser, B.J.; Mullineaux, D.R. Greater vertical impact loading in female runners with medically diagnosed injuries: A
prospective investigation. Br. J. Sports Med. 2016, 50, 887–892. [CrossRef] [PubMed]
15.
Nikolaidis, P.T.; Rosemann, T.; Knechtle, B. Sex differences in the age of peak marathon race time. Chin. J. Physiol. 2018, 61, 85–91.
[CrossRef] [PubMed]
16.
Jokl, P.; Sethi, P.; Cooper, A. Master’s performance in the New York City marathon 1983–1999. Br. J. Sports Med. 2004, 38, 408–412.
[CrossRef] [PubMed]
17.
Schneider, A.L.; Nikolaidis, P.T.; Knechtle, B. Improved performance in master runners competing in the european championships
between 1978 and 2014. J. Strength Cond. Res. 2019, 33, 2559–2569. [CrossRef]
18.
Cuk, I.; Nikolaidis, P.T.; Knechtle, B. Sex differences in pacing during half-marathon and marathon race. Res. Sports Med. 2020, 28,
111–120. [CrossRef]
19.
Knechtle, B.; Barandun, U.; Knechtle, P.; Zingg, M.A.; Rosemann, T.; Rüst, C.A. Prediction of half-marathon race time in
recreational female and male runners. Springerplus 2014, 3, 248. [CrossRef]
Int. J. Environ. Res. Public Health 2022, 19, 7802
9 of 9
20.
March, D.S.; Vanderburgh, P.M.; Titlebaum, P.J.; Hoops, M.L. Age, sex, and finish time as determinants of pacing in the marathon.
J. Strength Cond. Res. 2011, 25, 386–391. [CrossRef]
21.
Breen, D.; Norris, M.; Healy, R.; Anderson, R. Marathon pace control in masters athletes. Int. J. Sports Physiol. Perform. 2018, 13,
332–338. [CrossRef]
22.
Nikolaidis, P.T.; Knechtle, B. Pacing in age group marathoners in the “New York City Marathon”. Res. Sports Med. 2018, 26, 86–99.
[CrossRef] [PubMed]
23.
Nikolaidis, P.T.; Knechtle, B. Pacing strategies in the ‘Athens classic marathon’: Physiological and psychological aspects. Front.
Physiol. 2018, 9, 1539. [CrossRef]
24.
Hennig, E. Gender differences for running in athletic footwear. In Proceedings of the 5th Symposium on Footwear Biomechanics,
Zurich, Switzerland, 5–7 July 2001.
25.
Sinclair, J.; Greenhalgh, A.; Edmundson, C.J.; Brooks, D.; Hobbs, S.J. Gender differences in the kinetics and kinematics of distance
running: Implications for footwear design. Int. J. Sports Sci. Eng. 2012, 6, 118–128.
26.
Hangzhou Marathon Result Inquiry. Available online: http://en.hzim.org/p/hzim/match/queryresult.jsf (accessed on 4 May 2020).
27.
Nikolaidis, P.T.; Cuk, I.; Rosemann, T.; Knechtle, B. Performance and Pacing of Age Groups in Half-Marathon and Marathon. Int.
J. Environ. Res. Public Health 2019, 16, 1777. [CrossRef] [PubMed]
28.
Nikolaidis, P.T.; Cuk, I.; Clemente-Suarez, V.J.; Villiger, E.; Knechtle, B. Number of finishers and performance of age group women
and men in long-distance running: Comparison among 10 km, half-marathon and marathon races in Oslo. Res. Sports Med. 2021,
29, 56–66. [CrossRef] [PubMed]
29.
Cuk, I.; Nikolaidis, P.T.; Markovic, S.; Knechtle, B. Age Differences in Pacing in Endurance Running: Comparison between
Marathon and Half-MarathonMen and Women. Medicina 2019, 55, 479. [CrossRef] [PubMed]
30.
Tian, H.B.; Qiu, Y.J.; Lin, Y.Q.; Zhou, W.T.; Fan, C.Y. The Role of Leisure Satisfaction in Serious Leisure and Subjective Well-Being:
Evidence from Chinese Marathon Runners. Front. Psychol. 2020, 11, 581908. [CrossRef]
31.
García-Naveira, A.; Barquín, R.R.; Pujals, C. Personality differences according to participation in or abstention from sport,
competition level and age category in football players based on Costa and McCraes´ s Model. Rev. De Psicol. Del Deporte 2011,
20, 29–44.
32.
Anthony, D.; Rüst, C.A.; Cribari, M.; Rosemann, T.; Lepers, R.; Knechtle, B. Differences in participation and performance trends in
age group half and full marathoners. Chin. J. Physiol. 2014, 57, 209–219. [CrossRef]
33.
Bairan, L. Research on Current Situation and Development Path of Marathon Events in Hubei Province. J. Sports Sci. 2019, 7,
65–70. [CrossRef]
34.
Del Coso, J.; Fernández, D.; Abián-Vicen, J.; Salinero, J.J.; González-Millán, C.; Areces, F.; Ruiz, D.; Gallo, C.; Calleja-González, J.;
Pérez-González, B. Running pace decrease during a marathon is positively related to blood markers of muscle damage. PLoS
ONE 2013, 8, e57602. [CrossRef]
35.
Buman, M.P.; Omli, J.W.; Giacobbi, P.R., Jr.; Brewer, B.W. Experiences and coping responses of “hitting the wall” for recreational
marathon runners. J. Appl. Sport Psychol. 2008, 20, 282–300. [CrossRef]
36.
Hoffman, M.D.; Wegelin, J.A. The Western States 100-mile endurance run: Participation and performance trends. Med. Sci. Sports
Exerc. 2009, 41, 2191. [CrossRef] [PubMed]
37.
Knechtle, B.; Knechtle, P.; Lepers, R. Participation and performance trends in ultra-triathlons from 1985 to 2009. Scand. J. Med. Sci.
Sports 2011, 21, e82–e90. [CrossRef] [PubMed]
38.
Vitti, A.; Nikolaidis, P.T.; Villiger, E.; Onywera, V.; Knechtle, B. The “New York City Marathon”: Participation and performance
trends of 1.2 M runners during half-century. Res. Sports Med. 2020, 28, 121–137. [CrossRef] [PubMed]
39.
Miller, A.E.; MacDougall, J.D.; Tarnopolsky, M.A.; Sale, D.G. Gender diferences in strength and muscle fber characteristics. Eur. J.
Appl. Physiol. Occup. Physiol. 1993, 66, 254–262. [CrossRef]
40.
Welle, S.; Tawil, R.; Thornton, C.A. Sex-related diferences in gene expression in human skeletal muscle. PLoS ONE 2008,
3, e1385. [CrossRef]
41.
Maldonado-Martin, S.; Mujika, I.; Padilla, S. Physiological variables to use in the gender comparison in highly trained runners.
J. Sports Med. Phys. Fit. 2004, 44, 8–14.
42.
Besson, T.; Macchi, R.; Rossi, J.; Morio, C.; Kunimasa, Y.; Nicol, C.; Vercruyssen, F.; Millet, G.Y. Sex Differences in Endurance
Running. Sports Med. 2022, 52, 1235–1257. [CrossRef]
43.
Leyk, D.; Erley, O.; Ridder, D.; Leurs, M.; Rüther, T.; Wunderlich, M.; Sievert, A.; Baum, K.; Essfeld, D. Age-related changes in
marathon and half-marathon performances. Int. J. Sports Med. 2007, 28, 513–517. [CrossRef]
44.
Hunter, S.K.; Stevens, A.A. Sex differences in marathon running with advanced age: Physiology or participation? Med. Sci. Sports
Exerc. 2013, 51, 1–9. [CrossRef] [PubMed]
45.
Knechtle, B.; Di Gangi, S.; Rüst, C.A.; Villiger, E.; Rosemann, T.; Nikolaidis, P.T. The role of weather conditions on running
performance in the Boston Marathon from 1972 to 2018. PLoS ONE 2019, 14, e0212797. [CrossRef] [PubMed]
| Gender and Age Differences in Performance of Over 70,000 Chinese Finishers in the Half- and Full-Marathon Events. | 06-25-2022 | Yang, San-Jun,Yang, Fan,Gao, Yuan,Su, Yan-Feng,Sun, Wei,Jia, Sheng-Wei,Wang, Yu,Lam, Wing-Kai | eng |
PMC9209328 | ARTICLE
OPEN
Rethinking aerobic exercise intensity prescription in adults with
spinal cord injury: time to end the use of “moderate to
vigorous” intensity?
Michael J. Hutchinson
1 and Victoria L. Goosey-Tolfrey
1✉
© The Author(s) 2021
STUDY DESIGN: Cohort study.
OBJECTIVES: To investigate and critique different methods for aerobic exercise intensity prescription in adults with spinal cord
injury (SCI).
SETTING: University laboratory in Loughborough, UK.
METHODS: Trained athletes were split into those with paraplegia (PARA; n = 47), tetraplegia (TETRA; n = 20) or alternate health
condition (NON-SCI; n = 67). Participants completed a submaximal step test with 3 min stages, followed by graded exercise test to
exhaustion. Handcycling, arm crank ergometry or wheelchair propulsion were performed depending on the sport of the participant.
Oxygen uptake (V̇O2), heart rate (HR), blood lactate concentration ([BLa]) and ratings of perceived exertion (RPE) on Borg’s RPE scale
were measured throughout. Lactate thresholds were identified according to log-V̇O2 plotted against log-[BLa] (LT1) and 1.5 mmol
L−1 greater than LT1 (LT2). These were used to demarcate moderate (<LT1), heavy (>LT1, < LT2) and severe (>LT2) exercise intensity
domains.
RESULTS: Associations between percentage of peak V̇O2 (%V̇O2peak) and HR (%HRpeak) with RPE differed between PARA and TETRA.
At LT1 and LT2, %V̇O2peak and %HRpeak were significantly greater in TETRA compared to PARA and NON-SCI (P < 0.05). The variation
in %V̇O2peak and %HRpeak at lactate thresholds resulted in large variability in the domain distribution at fixed %V̇O2peak and %HRpeak.
CONCLUSIONS: Fixed %V̇O2peak and %HRpeak should not be used for aerobic exercise intensity prescription in adults with SCI as the
method does not lead to uniform exercise intensity domain distribution.
Spinal Cord (2022) 60:484–490; https://doi.org/10.1038/s41393-021-00733-2
INTRODUCTION
For adults with spinal cord injury (SCI), aerobic exercise is
beneficial for improving indices of physical [1] and mental [2]
health. On this theme, scientific guidelines published in 2018
describe the dose of aerobic exercise required to improve
cardiorespiratory fitness and cardiometabolic health in adults
with SCI [3]. Central to the guidelines is information on the
frequency (e.g., 3 times per week) and duration (e.g., 30 min) of
the exercise, both of which are simple to define and monitor. The
final important aspect of the guidelines is the exercise intensity. If
aerobic exercise is performed at too low an intensity, without
sufficient exercise volume, it will not lead to beneficial physiolo-
gical adaptations [4]. Despite this, the guidelines provide no clear
prescription of the exercise intensity, other than to say that
aerobic exercise should be of a “moderate to vigorous” intensity
[3]. The lack of clarity with the exercise intensity terminology is a
hindrance to adults with SCI using the guidelines to inform their
exercise habits; practitioners actively prescribing exercise training;
and researchers investigating the effects of exercise training
interventions on markers of health in adults with SCI. There is,
therefore, an urgent need to better understand aerobic exercise
intensity prescription in adults with SCI.
Guidelines for non-disabled adults define thresholds for five
intensity zones (very light, light, moderate, vigorous, near-
maximal/maximal) according to many physiological variables [5].
These variables include percentage maximum oxygen uptake (%
V̇O2max) and heart rate (%HRmax), oxygen uptake and heart rate
reserve (%V̇O2R, %HRR), and ratings of perceived exertion (RPE)
[5]. However, despite the SCI guidelines adopting the “moderate”
and “vigorous” descriptives, there is no equivalent resource
published for adults with SCI regarding the physiological thresh-
olds coinciding with these descriptors. Furthermore, given the
physiological consequences of SCI on cardiovascular and respira-
tory responses to exercise [6], there is no justification for simply
adopting the percentage thresholds utilised for non-disabled
adults.
An alternative approach to exercise intensity prescription is to
consider whether different methods result in participants exercis-
ing in the same of three exercise domains (moderate, heavy,
severe) [7]. This is because of the similar V̇O2 and blood lactate
responses between individuals exercising in these domains [7].
Specifically, the moderate intensity domain (below lactate thresh-
old (LT)) is characterised by steady state responses for V̇O2 and
blood lactate concentration ([BLa]) [8]. In the heavy intensity
Received: 26 July 2021 Revised: 19 November 2021 Accepted: 23 November 2021
Published online: 8 December 2021
1Peter Harrison Centre for Disability Sport, School of Sport, Exercise and Health Sciences, Loughborough University, Loughborough, UK. ✉email: [email protected]
www.nature.com/sc
1234567890();,:
domain (between LT and critical power/speed (CP/CS)) there is a
delayed steady state response due to the V̇O2 “slow component”,
whilst in the severe domain (above CP/CS) no steady state
response is observed [8].
To satisfy the aim of producing a homogenous exercise
intensity, the fixed percentage approach is only valid if it is
demonstrated that equal relative intensities result in individuals
exercising in the same intensity domain [7]. However, in a recent
study of non-disabled participants, no fixed %V̇O2max or %HRmax,
typically used for exercise prescription, resulted in all participants
being in the same intensity domain [9]. This has led to assertions
that using fixed %V̇O2max or %HRmax for prescribing exercise
intensity is inaccurate and will lead to significant inter-individual
physiological responses, precluding homogenous exercise inten-
sity prescription [7, 9]. Furthermore, with evidence that individual
participant %V̇O2R:%HRR relationships diverges from the assumed
linear trajectory, there are also questions over how appropriate %
V̇O2R and %HRR are for prescribing exercise intensity at the
individual level [10].
For adults with SCI there is currently nothing more to inform
aerobic exercise intensity prescription than the arbitrary use of
“moderate to vigorous” intensity [3]. Furthermore, evidence in
non-disabled adults would suggest a need to rethink the
traditional use of fixed percentages [7, 9, 10]. Therefore, this
study aimed to investigate and critique potential methods for
prescribing aerobic exercise intensity in adults with SCI.
METHODS
This study was performed via a retrospective analysis of athlete data
collected in the author’s laboratory. All procedures were approved by the
Human participants ethical sub-committee at Loughborough University,
and participants provided written, informed consent.
Participants
Data were available for 134 individuals (male: 98; female: 36). Participants
were split into those with paraplegia (PARA), tetraplegia (TETRA), or
alternate health condition (NON-SCI), see Table 1. Examples of health
conditions for NON-SCI included spina bifida, limb deficiency, cerebral
palsy, and arthrogryposis. Participants were competitive athletes, compet-
ing at a national or international level, from one of the following sports:
handcycling, para-alpine ski, paratriathlon, wheelchair basketball, wheel-
chair rugby or wheelchair tennis.
Exercise testing
Participants completed a submaximal step test followed by graded
exercise test (GXT) to exhaustion. Handcycle (HC) tests were performed
in the participants own handcycle attached to a Cyclus 2 ergometer
(Avantronic Richter, Leipzig, Germany). For some Paratriathlon, and all
para-alpine ski athletes, arm crank ergometry (ACE) was used (Lode Angio,
Lode B. V., Groningen, the Netherlands). The ergometer was positioned
vertically so the crank axis centre was level with the shoulder, and
horizontally to allow slight elbow flexion at the furthest point of the crank
cycle. Wheelchair basketball, rugby and tennis players performed a
wheelchair propulsion (WCP) test using a motorised treadmill (HP Cosmos,
Traunstein, Germany) and their own custom sports wheelchair.
Submaximal tests were individualised based on the sport, sex, training
status and level of impairment of the participant, with the goal of
completing 6-8 stages (average: 6; range: 4–10). HC and ACE tests started
at 15–60 W, with 10–20 W increments every 3 min. WCP tests started at
0.7–2.8 m s−1 and were increased by 0.2-0–4 m s−1 every 3 min. V̇O2
(Metalyzer 3B, Cortex, Leipzig, Germany) and HR (RS400, Polar, Kempele,
Finland) were continually monitored throughout. The Metalyzer was
calibrated before each participant against ambient air and a mix of 15% O2,
5% CO2, with the volume calibrated using a 3 L syringe. RPE was verbally
reported in the final minute of each stage using Borg’s 6–20 RPE scale [11].
A capillary blood sample from the ear lobe was collected at the end of
each stage for measurement of [BLa] (Biosen C-line, EKF Diagnostics,
Barleben, Germany). HC and ACE tests were continuous, however,
WCP tests were discontinuous as the treadmill needed to be slowed
between stages to facilitate blood sampling. For discontinuous tests, the
typical interval between stages was 45–60 s. Submaximal tests continued
until [BLa] exceeded 4 mmol L−1 or RPE was rated as 17. The RPE criteria
was used in TETRA where there may have been blunted lactate responses
[12].
Following the submaximal test, participants received 15 min of active
recovery or rest before performing a GXT to exhaustion. The starting
workload was set to that from the preceding test when [BLa] increased by
0.5 mmol L−1 above rest. Participants performed 1 min at this load, before
the exercise intensity were increased in a stepwise manner by 10–20 W
min−1 (HC/ACE) or 0.1 m s−1 min−1 (WCP) until participants reached
volitional exhaustion. This was defined as an inability to maintain
their preferred cadence at the required PO for HC/ACE, or the required
speed of the treadmill, despite verbal encouragement. V̇O2 and HR were
again monitored throughout, with RPE and [BLa] measured at the end of
the test.
Data processing
V̇O2 and HR data were subjected to a 30 s rolling average, with the greatest
of these from the GXT recorded as peak values (V̇O2peak, HRpeak). V̇O2 and
HR in the final 30 s of each submaximal stage were extracted and
calculated as percentages of peak (%V̇O2peak, %HRpeak). Using the
submaximal data, the lactate thresholds were identified as the intersection
of the horizontal and ascending sections of the plot of log-[BLa] against
log-V̇O2 (LT1) [13], and at [BLa] equal to LT1 plus 1.5 mmol L−1 (LT2) [14].
The inverse of the log-V̇O2 at these points were calculated to give the V̇O2
at LT1 and LT2. HR at LT1 and LT2 was identified by interpolation of the
linear V̇O2:HR relationship for each participant. RPE was modelled against
[BLa] using a quadratic function for each participant, with the resultant
coefficients used to calculate the RPE at LT1 and LT2 [15]. Exercise intensity
domains were defined as moderate (<LT1), heavy (between LT1 and LT2)
and severe (>LT2).
Statistical analyses
Analyses were performed using IBM SPSS Statistics Version 23.0 (IBM
Corp., Armonk, NY) and MLWiN Version 3.05 [16]. Data are presented
as mean (standard deviation) with statistical significance accepted at P <
0.05. Data were checked for normal distribution using the Shapiro Wilk
statistic.
All individual RPE data points were modelled against the corresponding
%V̇O2peak and %HRpeak using dynamic multilevel models with lagged
independent variable, whilst accounting for the initial condition. Separate
models were created for %V̇O2peak and %HRpeak, which served as the
independent variable, with RPE as the dependent variable. Models were
multilevel to adjust for the repeated stages performed by each participant
and were used due to their ability to characterise group- and individual-
level effects [17]. Stage was defined as the first, and participant as the
second level. Models accounted for the initial condition (e.g., stage (i) = 1)
as it was thought that RPE would depend on the %V̇O2peak and %HRpeak
when i = 1. The need for them to be dynamic and incorporate a lagged
independent variable was required as it was thought RPE at subsequent
measurement occasions (when i > 1) would be dependent on %V̇O2peak
and %HRpeak for that, as well as the previous, measurement occasion (i.e.,
i–1). Potential confounding variables were added to the models to assess
whether they improved the model fit with fixed effects, or random effects
for between- and within-individual variation. Confounding variables were
sex (male/female), group (PARA/TETRA/NON-SCI) and exercise mode (ACE/
HC/WCP). The resultant models were used to calculate the %V̇O2peak and %
HRpeak corresponding to each value on Borg’s RPE scale.
Differences in V̇O2 (L min−1, ml kg−1 min−1, %V̇O2peak), HR (beats·min−1,
%HRpeak) and RPE between groups at LT1 and LT2 were assessed via one-
way analysis of variance with Bonferroni post-hoc correction for multiple
comparisons. Standardised effect sizes (ES) were calculated to describe the
magnitude of differences and categorised as trivial (< 0.2), small (0.2–0.6),
moderate (0.6–1.2), large (1.2–2.0) and very large (> 2.0) [18]. For each
group, the percentage of participants in each intensity domain (moderate,
heavy, severe) were calculated at 5% intervals from 35 to 95% V̇O2peak and
%HRpeak.
RESULTS
The associations between RPE and both %V̇O2peak and %HRpeak
were not significantly affected by sex or exercise mode, so
stratification based on these variables was not needed. The %
M.J. Hutchinson and V.L. Goosey-Tolfrey
485
Spinal Cord (2022) 60:484 – 490
V̇O2peak and %HRpeak coinciding with each rating on Borg’s RPE
scale for PARA and TETRA can be found in Table 2. The full RPE
models
against
%V̇O2peak
and
%HRpeak
can
be
found
in
the Supplementary Material.
RPE and %V̇O2peak model
RPE was significantly affected by the initial %V̇O2peak (i = 1) (P <
0.01), by %V̇O2peak at subsequent occasions when i > 1 (P < 0.01),
as well as by the lagged %V̇O2peak (i.e., i–1) (P < 0.01). Each of these
variables also showed significant between-individual variation,
which was incorporated into the model. There was also an effect
of Group at occasions when i > 1 (P = 0.01). TETRA showing
significantly greater within-individual variation for the effect of %
V̇O2peak on RPE compared to PARA and NON-SCI. As such, PARA
and NON-SCI remained grouped, as there was no difference
between these groups.
RPE and %HRpeak model
RPE was significantly affected by the initial %HRpeak (i = 1) (P <
0.01), by %HRpeak at subsequent occasions when i > 1 (P < 0.01), as
well as by the lagged %HRpeak (i.e., i–1) (P < 0.01). These effects
were fixed and showed no significant between- or within-
individual variation. There was a fixed effect for Group, with the
association between RPE and %HRpeak being significantly different
for PARA (P = 0.03). There was no difference between TETRA and
NON-SCI, so these remained grouped in this model.
Responses at LT1 and LT2
The V̇O2, HR and RPE at LT1 and LT2 are shown in Fig. 1. At LT1
there was a significant group effect for absolute (F2 = 7.11, P <
0.01; Fig. 1a) and relative V̇O2 (F2 = 17.65, P < 0.01; Fig. 1c), %
V̇O2peak (F2 = 9.86, P < 0.01; Fig. 1e), HR (F2 = 42.79, P < 0.01; Fig. 1i)
and %HRpeak (F2 = 7.94, P < 0.01; Fig. 1k). LT1 occurred at a
significantly smaller absolute and relative V̇O2 in TETRA compared
to PARA (ES = 0.86, 1.00) and NON-SCI (ES = 1.00, 1.62). However,
%V̇O2peak at LT1 was significantly greater in TETRA compared to
PARA (ES = 0.96) and NON-SCI (ES = 0.94). Similarly, HR at LT1 was
smaller in TETRA compared to PARA (ES = 2.33) and NON-SCI
(ES = 2.74), whereas %HRpeak was greater in TETRA compared to
PARA
(ES = 1.00)
and
NON-SCI
(ES = 0.69).
There
was
no
significant difference between groups for RPE (F2 = 0.48, P =
0.62) at LT1 (Fig. 1g).
There was also a significant group effect at LT2 for absolute
(F2 = 9.96, P < 0.01; Fig. 1b) and relative V̇O2 (F2 = 19.75, P < 0.01;
Table 1.
Participant characteristics by group.
PARA
TETRA
NON-SCI
Sample size (n)
47
20
67
Sex (M/F)
30/17
18/2
50/17
Age (years)
33 ± 8a
32 ± 7a
27 ± 7
Body mass (kg)
70.9 ± 14.1a
70.8 ± 12.9
64.5 ± 12.8
Neurological level of injury
T4-L2
C3-C7
-
Injury completeness
Complete: 21
Complete: 8
-
Incomplete: 23
Incomplete: 4
Unavailable: 3
Unavailable: 8
Time since injury (years)
12 ± 9
12 ± 6
-
Peak oxygen uptake
(L·min−1)
2.5 ± 0.6b
1.7 ± 0.5
2.6 ± 0.7b
(ml·kg−1·min−1)
35.1 ± 8.2b
23.4 ± 5.8
39.9 ± 8.3b
Peak heart rate
(beats·min−1)
188 ± 9b
134 ± 20
187 ± 10 b
Sport (n)
Handcycling
8
0
3
Paratriathlon
11
0
8
Para alpine ski
2
1
2
Wheelchair basketball
20
1
31
Wheelchair rugby
0
15
13
Wheelchair tennis
6
3
10
Test mode (n)
Arm crank ergometry
11
1
7
Handcycling
10
0
6
Wheelchair propulsion
26
19
54
a: significantly greater than NON-SCI; b: significantly greater than TETRA, P < 0.05.
Table 2.
Resultant calculations of percentage peak oxygen uptake
and heart rate by group based on multilevel modelling.
RPE
%V̇O2peak
%HRpeak
PARA
TETRA
PARA
TETRA
6
16
22
40
43
7
22
28
44
48
8
28
34
49
52
9
34
40
54
57
10
40
46
58
62
11
46
52
63
66
12
52
58
68
71
13
58
64
73
76
14
64
70
77
81
15
70
76
82
85
16
76
82
87
90
17
82
88
91
95
18
88
94
96
99
19
94
100
100
20
100
M.J. Hutchinson and V.L. Goosey-Tolfrey
486
Spinal Cord (2022) 60:484 – 490
Fig. 1d), %V̇O2peak (F2 = 14.80, P < 0.01; Fig. 1f), HR (F2 = 58.99, P <
0.01; Fig. 1i) and %HRpeak (F2 = 6.10, P < 0.01; Fig. 1l). Absolute and
relative V̇O2 at LT2 were significantly smaller in TETRA compared
to PARA (ES = 0.83, 0.97) and NON-SCI (ES = 1.03, 1.72). However,
%V̇O2peak at LT2 was significantly greater in TETRA compared to
PARA (ES = 1.26) and NON-SCI (ES = 1.06). Furthermore, HR at LT2
was significantly smaller in TETRA than in PARA (ES = 3.00) and
NON-SCI (ES = 3.55), while %HRpeak was significantly greater in
TETRA compared to PARA (ES = 0.93) and NON-SCI (ES = 0.62).
There was no significant difference between groups in RPE (F2 =
2.18, P = 0.19) at LT2 (Fig. 1h).
Intensity classification
Thresholds for %V̇O2peak and %HRpeak corresponding with
intensity classifications used in non-disabled exercise guidelines
are shown in Table 3. These data suggest there are differences
between non-disabled individuals, PARA and TETRA in the
thresholds for intensity classifications. Frequency distribution of
individuals
within
moderate,
heavy
and
severe
intensity
domains for discrete percentages of %V̇O2peak and %HRpeak
are shown in Figs. 2 and 3, respectively. These show that no %
V̇O2peak or %HRpeak typically used for exercise prescription
purposes leads to all participants being in the same domain,
with many %V̇O2peak including participants spread across all
three domains.
DISCUSSION
This study aimed to investigate potential methods of aerobic
exercise intensity
prescription in adults with SCI.
Findings
demonstrate that there are differences between PARA and TETRA
for
the
%V̇O2peak
and
%HRpeak
corresponding
with
the
Fig. 1
Group physiological responses at lactate thresholds. a Absolute V̇O2 at LT1. b Absolute V̇O2 at LT2. c Relative V̇O2 at LT1. d Relative
V̇O2 at LT2. e %V̇O2peak at LT1. f %V̇O2peak at LT2. g RPE at LT1. h RPE at LT2. i HR at LT1. j HR at LT2. k %HRpeak at LT1. l %HRpeak at LT2.
Data are presented at mean (SD) with individual points overlaid. Within each group, numbers refer to the same participant in each figure.
Asterisk (*) indicates significantly greater than the identified group, P < 0.05.
Table 3.
Classification of exercise intensity for individuals with paraplegia and tetraplegia, compared to non-disabled guidelines.
Very light (RPE ≤
8)a
Light (RPE 9–11)a
Moderate (RPE
12–13)a
Vigorous (RPE
14–17)a
Near maximal-maximal (RPE ≥
18)a
%V̇O2peak
Non-disableda
≤36% V̇O2peak
37–45% V̇O2peak
46–63% V̇O2peak
64–90% V̇O2peak
≥91% V̇O2peak
Paraplegia
≤31% V̇O2peak
32–49% V̇O2peak
50–61% V̇O2peak
62–85% V̇O2peak
≥86% V̇O2peak
Tetraplegia
≤37% V̇O2peak
38-55% V̇O2peak
56–67% V̇O2peak
68–91% V̇O2peak
≥92% V̇O2peak
%HRpeak
Non-disableda
≤56% HRpeak
57–63% HRpeak
64-76% HRpeak
77–95% HRpeak
≥96% HRpeak
Paraplegia
≤51% HRpeak
52–65% HRpeak
66-75% HRpeak
76–93% HRpeak
≥94% HRpeak
Tetraplegia
≤54% HRpeak
55–68% HRpeak
69–78% HRpeak
79–97% HRpeak
≥98% HRpeak
a data from Riebe et al. [5].
M.J. Hutchinson and V.L. Goosey-Tolfrey
487
Spinal Cord (2022) 60:484 – 490
descriptions of “moderate” and “vigorous” exercise intensity, as
used by the exercise guidelines for adults with SCI [1, 3]. However,
the findings also show that using fixed %V̇O2peak or %HRpeak
cannot guarantee a homogenous domain-specific exercise inten-
sity prescription for adults with SCI.
Fixed percentages and intensity domains for exercise
intensity prescription
The finding of adults with SCI being in different intensity domains,
as defined in this study according to LT1 and LT2, despite being at
the same %V̇O2peak or %HRpeak supports similar evidence in non-
disabled participants [9]. The domain-specific distribution is also
arguably more variable in adults with SCI. For V̇O2peak, Iannetta
et al. [9]. report participants in moderate, heavy, and severe
domains only at 70% V̇O2peak, whereas in this study this was
shown at several fixed percentages, 55-70% V̇O2peak in PARA and
60-70% V̇O2peak in TETRA. It would, therefore, given recent calls to
stop using fixed percentages for exercise intensity prescription in
non-disabled participants [7, 9, 10], seem appropriate that this is
expanded to apply to adults with SCI performing aerobic exercise.
This would apply to all adults with SCI, and not just the athletic
population utilised in this study. Inter-individual variation will exist
in sedentary or low-active participants as much, if not more, than
in athletic groups, further limiting the use of fixed percentages for
exercise intensity prescription.
Instead, more attention should be given to methods that can
lead to participants exercising within the same exercise intensity
domain, each of which are characterised by distinct V̇O2 kinetic
and blood lactate profiles [19, 20]. This study utilised LT1 to
identify the transition between moderate and heavy intensity
domains, in accordance with the literature [19, 20]. However, a
limitation within the current study was the use of LT2 to identify
the heavy-severe domain transition, due to a lack of evidence
supporting this, as well as the number of different methods used
to measure and identify LT2 [7, 21]. As such, firm conclusions
cannot be made regarding the heavy-severe domain transition
from this study. It would have been more appropriate to measure
CP/CS for this purpose [19–21], however, only data from a GXT
were available in this study.
This highlights an important limitation to the widespread
implementation of intensity domain-related exercise prescription.
Specifically, the suitability of different testing protocols for
identifying different threshold concepts, as well as data collection
and threshold identification methods used [21]. This poses the
challenge of how to simply prescribe domain-specific exercise
intensity. Data from the present study may support the use of RPE
for this purpose, as no difference in RPE was found between
groups at LT1 (Fig. 1g) and LT2 (Fig. 1h), in support of previous
findings [15]. Mean RPE at LT1 and LT2 in this study were found to
be 11 and 15, respectively, suggesting that these values could be
Fig. 2
Distribution of individuals in moderate, heavy, and severe
intensity domains at fixed %V̇O2peak. a PARA group. b TETRA
group. c NON-SCI group.
Fig. 3
Distribution of individuals in moderate, heavy and severe
intensity domains at fixed %HRpeak. a PARA group. b TETRA
group. c NON-SCI group.
M.J. Hutchinson and V.L. Goosey-Tolfrey
488
Spinal Cord (2022) 60:484 – 490
used to guide exercise prescription using a simple and easy to
implement method. However, the SD for these values ranged from
1 to 2 units dependent on group. This inter-individual variation
serves, therefore, as a limitation to the use of fixed RPE values for
exercise intensity prescription and highlights the importance of
individualisation in this context. Individualisation, though, poses a
further challenge, due to the trade-off between the precision
required for research purposes versus the simple messaging for
population-level exercise guidelines.
Implications for research
Our results highlight the need to prescribe exercise intensity in a
way that ensures a homogenous intensity domain distribution
between participants within both acute and longitudinal study
designs. Previously, studies have conducted training interventions
with an intensity of either a target range, or fixed value, using
variables such as %V̇O2peak, %HRpeak and RPE, e.g. [22–26]., which
will have led to significant domain heterogeneity. This means the
dose of exercise stimulus would not have been controlled
between
participants.
Moving
forwards,
researchers
should
identify the domain transitions and prescribe intensity in relation
to these. That being said, the authors are not aware of any
investigations into CP/CS using participants with SCI, so initial
studies
need
to
investigate
the
validity
and
reliability
of
identifying domain transitions in participants in SCI. Furthermore,
V̇O2 kinetic responses to exercise in each domain should be
investigated in participants with SCI, due to potential differences
in V̇O2 kinetics between non-disabled participants and those with
SCI [27]. Finally, these investigations must also account for any
differences based on the mode of exercise [28].
Our findings also emphasise the need to individualise the
exercise
intensity
prescription.
In
non-disabled
participants,
standardised intensity prescription (e.g., 55–75% V̇O2peak), has
been shown to lead to significant heterogeneity in responsiveness
[29] leading to participants being described as either “responders”
or “non-responders” to the intervention. However, an individua-
lised exercise prescription (relative to ventilatory threshold) in
non-disabled
participants
resulted
in
100%
responsiveness,
compared to 60% in the standardised intervention [30]. Subse-
quently, future exercise research in participants with SCI should
individualise the intensity prescription according to intensity
domains, whilst also report individual responsiveness to an
intervention. This will improve methodological control while also
increasing confidence in conclusions made based on the data.
Implications for exercise guidelines using “moderate to
vigorous” exercise intensity
Exercise guidelines must balance the precision required for a
specific intensity prescription, against the need for a simple
population-level recommendation. Furthermore, scientific guide-
lines must undergo a knowledge translation process to ensure
that the scientific integrity of the guidelines are maintained, whilst
also incorporating the varied needs of all potential end-users [31].
In adults with SCI, the scientific guidelines [1, 3] recommend
performing aerobic exercise at a “moderate to vigorous” intensity,
without providing any specific details on what this means. The same
intensity prescription is also used in a community and clinical-
practice version of the guidelines [31]. Results from the current study
would
suggest
the
scientific
integrity
of
such
an
intensity
prescription is questionable. Table 3 shows equivalent thresholds
for PARA and TETRA for “moderate” and “vigorous” intensity, based
on the guidelines for non-disabled adults [5]. Combining these with
the intensity domain distributions in Fig. 2, shows participants were
spread across moderate, heavy, and severe domains at both
“moderate” and “vigorous” intensities. It should also be noted that
participants in the present study were competitive athletes, and that
responses would likely be even more variable for sedentary or low-
active populations. This shows how the use of “moderate” to
“vigorous” intensity will not lead to anything close to resembling a
uniform exercise intensity prescription between individuals. It also
shows how someone expecting to perform “moderate” intensity
exercise may actually be much closer to their maximum capacity.
This will likely decrease the pleasure the person feels during the
exercise, which could ultimately impact on whether they decide to
continue doing it [32].
As it is operationally more difficult to define compared to
frequency (e.g., 3 times a week) and duration (e.g., 30 min), it is
possible that exercise intensity becomes an ignored piece of
exercise guidelines. Perhaps research could seek to understand
end-user
perceptions
of
“exercise
intensity”,
or
needs
for
interpreting and monitoring intensity, before using that informa-
tion alongside physiological principles to underpin an evidence-
informed intensity prescription. Alternatively, maybe the focus for
exercise guidelines should shift from exercise intensity to also
acknowledge factors that might help individuals become and stay
active, such as their pleasure when performing exercise [32].
CONCLUSION
Prescribing a “moderate to vigorous” exercise intensity will not
lead to a uniform intensity domain distribution in adults with SCI.
Neither will the use of fixed percentages of V̇O2peak or HRpeak, or
generic values of RPE, due to inter-individual variation. Such
methods of exercise intensity prescription should not be used in
this population. Future research should individualise the intensity
prescription to ensure a homogenous inter-individual domain
distribution. However, the accurate testing required to ensure an
individualised intensity prescription poses a challenge to exercise
guidelines aimed at informing behaviour at the population-level.
Data archiving
The datasets generated and analysed during the current study are
available from the corresponding author on reasonable request.
REFERENCES
1. van der Scheer JW, Martin Ginis KA, Ditor DS, Goosey-Tolfrey VL, Hicks AL, West
CR, et al. Effects of exercise on fitness and health of adults with spinal cord injury.
Neurology 2017;89:736–45.
2. Kim DI, Lee J, Park H, Jeon JY. The relationship between physical activity levels
and mental health in individuals with spinal cord injury in South Korea. Int J
Environ Res Public Health. 2020;17:1–9.
3. Martin Ginis KA, van der Scheer JW, Latimer-Cheung AE, Barrow A, Bourne C,
Carruthers P, et al. Evidence-based scientific exercise guidelines for adults with
spinal cord injury: an update and a new guideline. Spinal Cord. 2018;56:308–21.
4. van der Scheer JW, de Groot S, Tepper M, Faber W, ALLRISC Group, Veeger DH,
et al. Low-intensity wheelchair training in inactive people with long-term spinal
cord injury: a randomized controlled trial on fitness, wheelchair skill performance
and physical activity levels. J Rehabil Med. 2016;48:33–42.
5. Riebe D, Ehrman JK, Liguori G, Magal M ACSM’s guidelines for exercise testing
and prescription. Tenth edit. (Wolters Kluwer: Philadelphia, PA, 2018).
6. Theisen D. Cardiovascular determinants of exercise capacity in the Paralympic
athlete with spinal cord injury. Exp Physiol. 2012;97:319–24.
7. Jamnick NA, Pettitt RW, Granata C, Pyne DB, Bishop DJ. An examination and
critique of current methods to determine exercise intensity. Sports Med.
2020;50:1729–56.
8. Burnley M, Jones AM. Power–duration relationship: physiology, fatigue, and the
limits of human performance. Eur J Sport Sci. 2018;18:1–12.
9. Iannetta D, Inglis EC, Mattu AT, Fontana FY, Pogliaghi S, Keir DA, et al. A critical
evaluation of current methods for exercise prescription in women and men. Med
Sci Sport Exerc. 2020;52:466–73.
10. Ferri Marini C, Sisti D, Leon AS, Skinner JS, Sarzynski MA, Bouchard C, et al. HRR
and VO2R fractions are not equivalent: is it time to rethink aerobic exercise
prescription methods? Med Sci Sport Exerc. 2021;53:174–82.
11. Borg GA Borg’s Perceived Exertion and Pain Scales. (Human Kinetics: Champaign,
IL, 1998).
12. Hutchinson MJ, Kilgallon JW, Leicht CA, Goosey-Tolfrey VL. Perceived exertion
responses to wheelchair propulsion differ between novice able-bodied and
trained wheelchair sportspeople. J Sci Med Sport. 2019;23:403–7.
M.J. Hutchinson and V.L. Goosey-Tolfrey
489
Spinal Cord (2022) 60:484 – 490
13. Beaver WL, Wasserman K, Whipp BJ. Improved detection of lactate threshold
during exercise using a log-log transformation. J Appl Physiol. 1985;59:1936–40.
14. Dickhuth H-H, Huonker M, Münzel T, Drexler H, Berg A, Keul J Individual anae-
robic threshold for evaluation of competitive athletes and patients with left
ventricular dysfunction. In: Bachl N, Graham TE, Lollgen H (eds). Advances in
Ergometry. (Springer: Berlin Heidelberg, 1991), pp 173-9.
15. Hutchinson MJ, Kouwijzer I, de Groot S, Goosey-Tolfrey VL Comparison of two
Borg exertion scales for monitoring exercise intensity in able-bodied participants,
and those with paraplegia and tetraplegia. Spinal Cord. 2021. https://doi.org/
10.1038/s41393-021-00642-4.
16. Charlton C, Rasbash J, Browne WJ, Healy M, Cameron B MLwiN Version 3.05. 2020.
17. Kristjansson SD, Kircher JC, Webb AK. Multilevel models for repeated measures
research designs in psychophysiology: an introduction to growth curve model-
ing. Psychophysiology 2007;44:728–36.
18. Batterham AM, Hopkins WG. Making meaningful inferences about magnitudes.
Int J Sports Physiol Perform. 2006;1:50–57.
19. Poole DC, Jones AM. Oxygen uptake kinetics. Compr Physiol. 2012;2:933–96.
20. Poole DC, Burnley M, Vanhatalo A, Rossiter HB, Jones AM. Critical power: an
important fatigue threshold in exercise physiology. Med Sci Sport Exerc.
2016;48:2320–34.
21. Poole DC, Rossiter HB, Brooks GA, Gladden LB. The anaerobic threshold: 50+
years of controversy. J Physiol. 2021;599:737–67.
22. Bakkum AJ, de Groot S, Stolwijk-Swuste JM, van Kuppevelt DJ, ALLRISC, van der
Woude LH, et al. Effects of hybrid cycling versus handcycling on wheelchair-
specific fitness and physical activity in people with long-term spinal cord injury: a
16-week randomized controlled trial. Spinal Cord. 2015;53:395–401.
23. Hicks AL, Martin KA, Ditor DS, Latimer AE, Craven C, Bugaresti J, et al. Long-term
exercise training in persons with spinal cord injury: effects on strength, arm
ergometry
performance
and
psychological
well-being.
Spinal
Cord.
2003;41:34–43.
24. Kim D-I, Lee H, Lee B-S, Kim J, Jeon JY. Effects of a 6-week indoor hand-bike
exercise program on health and fitness levels in people with spinal cord injury: a
randomized controlled trial study. Arch Phys Med Rehabil. 2015;96:2033–40.
25. Pelletier CA, Totosy de Zepetnek JO, MacDonald MJ, Hicks AL. A 16-week ran-
domized controlled trial evaluating the physical activity guidelines for adults with
spinal cord injury. Spinal Cord. 2015;53:363–7.
26. Valent LJM, Dallmeijer AJ, Houdijk H, Slootman HJ, Janssen TW, Post MWM, et al.
Effects of hand cycle training on physical capacity in individuals with tetraplegia:
a clinical trial. Phys Ther. 2009;89:1051–60.
27. Gollie JM, Herrick JE, Keyser RE, Chin LMK, Collins JP, Shields RK, et al. Fatigability,
oxygen uptake kinetics and muscle deoxygenation in incomplete spinal cord
injury during treadmill walking. Eur J Appl Physiol. 2017;117:1989–2000.
28. Barstow TJ, Scremin AME, Mutton DL, Kunkel CF, Cagle TG, Whipp BJ. Peak and
kinetic cardiorespiratory responses during arm and leg exercise in patients with
spinal cord injury. Spinal Cord. 2000;38:340–5.
29. Bouchard C, An P, Rice T, Skinner JS, Wilmore JH, Gagnon J, et al. Familial
aggregation of V̇O(2max) response to exercise training: Results from the HERI-
TAGE family study. J Appl Physiol. 1999;87:1003–8.
30. Weathermax RM, Harris NK, Kilding AE, Dalleck LC. Incidence of V˙O2max
responders to personalized versus standardized exercise prescription. Med Sci
Sport Exerc. 2019;51:681–91.
31. Hoekstra F, McBride CB, Borisoff J, Fetterly MJ, Ginis S, Latimer-Cheung AE, et al.
Translating the international scientific spinal cord injury exercise guidelines into
community and clinical practice guidelines: a Canadian evidence-informed
resource. Spinal Cord. 2020;58:647–57.
32. Ekkekakis P, Parfitt G, Petruzzello SJ. The pleasure and displeasure people feel
when they exercise at different intensities: decennial update and progress
towards a tripartite rationale for exercise intensity prescription. Sports Med.
2011;41:641–71.
ACKNOWLEDGEMENTS
We would like to thank the various members of the Peter Harrison Centre for
Disability Sport who were involved in the collection of data used in this study. These
include Dr John Lenton, Dr Christof Leicht, Dr Tom Paulson, Dr Terri Paulson, Dr Katy
Griggs, Dr Ben Stephenson, Dr Ben Stone, Tom O’Brien, and Conor Murphy.
AUTHOR CONTRIBUTIONS
MJH and VLGT were both responsible for developing the concept of the study and
were both involved in the data collection. MJH performed the data analysis, with MJH
and VLGT involved in the interpretation of the data. MJH drafted the original version
of the manuscript, with VLGT providing comments. MJH and VLGT were both
responsible for reviewing and approving the final version of the manuscript.
FUNDING
This work was supported by funding from the Peter Harrison Foundation.
STATEMENT OF ETHICS
We certify that all applicable institutional regulations concerning the ethical use of
human volunteers were followed during the course of this research.
COMPETING INTERESTS
The authors declare no competing interests.
ADDITIONAL INFORMATION
Supplementary information The online version contains supplementary material
available at https://doi.org/10.1038/s41393-021-00733-2.
Correspondence and requests for materials should be addressed to Victoria L.
Goosey-Tolfrey.
Reprints and permission information is available at http://www.nature.com/
reprints
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims
in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly
from the copyright holder. To view a copy of this license, visit http://creativecommons.
org/licenses/by/4.0/.
© The Author(s) 2021
M.J. Hutchinson and V.L. Goosey-Tolfrey
490
Spinal Cord (2022) 60:484 – 490
| Rethinking aerobic exercise intensity prescription in adults with spinal cord injury: time to end the use of "moderate to vigorous" intensity? | 12-08-2021 | Hutchinson, Michael J,Goosey-Tolfrey, Victoria L | eng |
PMC7005585 | ORIGINAL RESEARCH
published: 31 January 2020
doi: 10.3389/fphys.2020.00030
Edited by:
Hassane Zouhal,
University of Rennes 2, France
Reviewed by:
Ajmol Ali,
Massey University, New Zealand
Martin Burtscher,
University of Innsbruck, Austria
*Correspondence:
Yifan Lu
[email protected]
Specialty section:
This article was submitted to
Exercise Physiology,
a section of the journal
Frontiers in Physiology
Received: 23 October 2019
Accepted: 15 January 2020
Published: 31 January 2020
Citation:
Li F, Nie J, Zhang H, Fu F, Yi L,
Hopkins W, Liu Y and Lu Y (2020)
Effects of Matched Intermittent
and Continuous Exercise on Changes
of Cardiac Biomarkers in Endurance
Runners. Front. Physiol. 11:30.
doi: 10.3389/fphys.2020.00030
Effects of Matched Intermittent and
Continuous Exercise on Changes of
Cardiac Biomarkers in Endurance
Runners
Feifei Li1,2, Jinlei Nie3, Haifeng Zhang1,4, Frank Fu5, Longyan Yi6, Will Hopkins7,
Yang Liu1,4 and Yifan Lu2*
1 College of Physical Education, Hebei Normal University, Shijiazhuang, China, 2 College of Sports Medicine
and Rehabilitation, Beijing Sport University, Beijing, China, 3 School of Health Sciences and Sports, Macao Polytechnic
Institute, Macao, China, 4 Provincial Key Lab of Measurement and Evaluation in Human Movement and Bio-information,
Hebei Normal University, Shijiazhuang, China, 5 Dr Stephen Hui Research Centre for Physical Recreation and Wellness,
Hong Kong Baptist University, Hong Kong, China, 6 Institute of Sport and Health Sciences, Beijing Sport University, Beijing,
China, 7 College of Sport and Exercise Science, Victoria University, Melbourne, VIC, Australia
Purpose: Endurance runners training with high-intensity intermittent exercise might
experience damage to cardiac muscle. We have therefore compared changes of cardiac
biomarkers after workload-matched intermittent and continuous exercise.
Methods: Twelve endurance runners [11 males, 1 female; means ± SD ˙VO2max,
62.4 ± 5.4 ml kg−1 min−1; velocity of ˙VO2max (v ˙VO2max), 17.1 ± 1.4 km h−1] completed
an intermittent and continuous exercise trial in random order. Intermittent exercise
consisted of running at 90% v˙VO2max for 2 min followed by 50% v˙VO2max for 2 min,
repeated for 92 min. Continuous exercise was performed at 70% v˙VO2max for 92 min.
Blood samples were drawn before and 0, 2, 4, 24, and 48 h after exercise for assay
of various cardiac biomarkers. Changes in concentration of biomarkers were averaged
for the comparison of intermittent with continuous exercise after adjustment for baseline
concentration and exercise intensity expressed as percent of heart-rate reserve (%HRR);
magnitudes were assessed by standardization.
Results: There were moderate and large increases in high-sensitivity cardiac troponin-I
and -T respectively following exercise. The differences between the increases adjusted
to the mean intensity of 78 %HRR were trivial, but at 85 %HRR the increases for cardiac
troponin-I and -T were moderately higher for intermittent compared with continuous
exercise (factor difference, ×/÷90% confidence limits: 3.4, ×/÷1.9 and 2.1, ×/÷1.8
respectively). Differences in the changes in other cardiac biomarkers were trivial.
Conclusion: Prolonged intermittent exercise is potentially more damaging to cardiac
muscle than continuous exercise of the same average running speed at higher average
heart rates in endurance runners.
Keywords: high-intensity intermittent exercise, intensity, biomarkers, cardiac troponin, marathon runners
Frontiers in Physiology | www.frontiersin.org
1
January 2020 | Volume 11 | Article 30
Li et al.
Exercise-Induced Cardiac Biomarkers Elevation
INTRODUCTION
There is increasing evidence that an acute bout of intense
exercise can induce a minor and temporary elevation of cardiac-
specific biomarkers, including various kinds of cardiac troponin,
markers diagnostic of myocardial infarction (Thygesen et al.,
2012; Eijsvogels et al., 2016). The elevation of cardiac troponin
is affected by exercise intensity (Fu et al., 2009), exercise duration
(Eijsvogels et al., 2010), baseline level (Legaz-Arrese et al., 2011),
training experience of the participants and age (Tian et al., 2012),
gender (Kong et al., 2017), cardiovascular risk factors (Vilela
et al., 2014), and environment (Li et al., 2016a). A transient
increase in myocardial membrane permeability appears to be
responsible, but the mechanism is still under debate (Nie et al.,
2010, 2016; Li et al., 2016b). Eijsvogels et al. (2016) concluded that
exercise intensity was the strongest single predictor of exercise-
induced cardiac troponin elevation, and Richardson et al. (2018)
found that cardiac troponin was associated with both mean and
peak heart rate during exercise. Gresslien and Agewall (2016)
hypothesized that there might be a threshold exercise intensity
where cardiac troponin release became more marked.
Athletes experience transiently high intensities of exercise
when they perform interval training to enhance performance but
the impact of such exercise on the heart in comparison with
continuous exercise is unclear (Billat, 2001; George et al., 2012;
Tschakert and Hofmann, 2013). Continuous exercise resulted
in a greater cardiac troponin concentration than intermittent
in sedentary men in a study by Ranjbar et al. (2017) whereas
Nie et al. (2018) reported no difference between the effects
of intermittent and continuous exercise on cardiac troponin
changes in sedentary women. In two studies of endurance
athletes, intermittent running produced greater increases in
cardiac troponin than continuous running, but the exercise
intensity and duration were not carefully manipulated and
matched (Stewart et al., 2016; Weippert et al., 2016). Other
markers of cardiac muscle damage, including N-terminal pro
brain natriuretic peptide (NT-pro-BNP) could also provide
information on the relative effects of intermittent and continuous
exercise (Scherr et al., 2011). The purpose of this study was
therefore to compare the changes of cardiac biomarkers after
workload-matched high-intensity intermittent and continuous
exercise on experienced endurance runners. It was hypothesized
that prolonged intermittent exercise potentially induced more
perturbations of cardiac biomarkers in endurance runners.
MATERIALS AND METHODS
Participants
After approval of this study by the local Ethical Committee, 12
(11 males and 1 female) endurance runners [means ± standard
deviation (SD): age, 23.5 ± 5.5 years; body mass, 63.3 ± 3.9 kg;
height, 170.5 ± 5.5 cm; %body fat, 12.6 ± 3.9%; ˙VO2max,
62.4 ± 5.4 ml kg−1 min−1; velocity of ˙VO2max (v ˙VO2max),
17.1 ± 1.4 km h−1; peak heart rate, 192 ± 9 min−1] were
recruited from the Department of Physical Education at a local
university. They had no history of disease or cardiac symptoms,
none were smokers, none were vegetarians or had any special
dietary habits and none had taken any drugs or antioxidant
supplements in the month before the study. Training history
(5.4 ± 3.4 years), training volume (44 ± 25 km week−1), and
personal best time in a recent marathon race (186 ± 16 min)
were self-reported. An initial medical screening and examination
were performed by a team of medical doctors and technicians.
All of the participants had normal resting blood pressures
and electrocardiographic results. All participants provided their
written consent and were fully informed about the purposes,
procedures, and potential cardiovascular risks of this study. The
study took place in the local sports science research center
between October and December in the afternoon. Conditions
were similar for each test, with small variations in temperature
(20.7 ± 2.3◦C), humidity (49 ± 13%). All of the participants
refrained from intense exercise and alcohol 48 h before
and after each trial and were allowed to ingest pure water
freely during tests.
Study Design
The design was a randomized-order crossover. On their first visit
to the laboratory, the runners performed a test for estimation of
˙VO2max and corresponding v ˙VO2max. On two subsequent visits,
the runners completed either intermittent or continuous exercise.
The order in which each participant completed the two trials was
selected at random and separated by at least 7 days, during which
they trained normally. Blood samples were drawn pre-exercise
(−1.5 h), immediately post-exercise (0 h), and at four later time
points (2, 4, 24, and 48 h) for measurement of biomarkers.
Determination of the ˙VO2max and v˙VO2max
A treadmill (H/p/cosmos Pulsar 4.0, H/p/cosmos Sports and
Medical gmbh, Nussdorf-Traunstein, Germany) with a 2% slope
was used in the determination of ˙VO2max (Max-1, Physio-Dyne
Instrument, NY, United States) and in the exercise trials. After a
general 5- to 10-min warm-up with self-set speed, the participants
ran at an initial speed of 12 km h−1, which was increased by
1 km h−1 every 3 min without any pause between stages. When
the respiratory exchange ratio reached 1.00, the stages were
shortened to 2 min. The test stopped either when the increase
in oxygen consumption ( ˙VO2) was less than 2.1 ml kg−1 min−1
while the respiratory exchange ratio was greater than or equal
to 1.15, or when exhaustion was reached. ˙VO2max was recorded
as the highest 30-s average value of the recorded ˙VO2. The
corresponding v ˙VO2max was recorded at the minimal speed at
which ˙VO2max was reached, as long as this speed was sustained
for at least 1 min.
Exercise Trials
Each participant completed a 1000-m warm-up at their own pace
prior to each trial. In the intermittent trial, each bout consisted
of a hard run of 90% v ˙VO2max for 2 min, followed by an easy
run of 50% v ˙VO2max for 2 min, 23 bouts and 92 min in total. In
the continuous trial, participants ran at 70% v ˙VO2max for 92 min.
Running velocities for intermittent trial were 15.4 ± 1.3 and
8.4 ± 0.8 km h−1, hard and easy rung respectively; for continuous
trial was 12.0 ± 0.9 km h−1.
Frontiers in Physiology | www.frontiersin.org
2
January 2020 | Volume 11 | Article 30
Li et al.
Exercise-Induced Cardiac Biomarkers Elevation
Measurements
Heart rate (S810, Polar, Finland) were recorded every 2 min
during each exercise trial. Rating of perceived exertion on Borg’s
scale (Borg, 1982) was recorded at 30, 32, 58, 60, 90, and 92 min
during each trial. Venous blood samples of 5 ml were dropped
from the antecubital vein by venous punctures, clotted at room
temperature and centrifuged at 3000 × g for 15 min. The
separated serum was then drawn and stored at −80◦C for further
analysis. These laboratory methods are the standard methods
used by the local Hospital Clinical Chemistry Laboratory and
have been validated.
The high-sensitivity cardiac troponin-I was analyzed using a
commercially available high-sensitive immunochemistry STAT
assay from Abbott Diagnostics on an Architect i2000SR (Abbott
Diagnostics, Chicago, IL, United States) with the detection
limit of 1.6 pg ml−1 and 99th percentile of the assay was
26 pg ml−1. The high-sensitivity cardiac troponin-T analysis
method was based on the new electrochemiluminescence
technology and using Elecsys 2010 automated batch analysers
(Roche Diagnostics, Mannheim, Germany). The measurement
range was 3–1000 pg ml−1. The 99th percentile cutoff
concentration was 14 pg ml−1. NT-pro-BNP was determined
by an Elecsys pro-BNP ECLIA on the Modular Analytics E170
analyzer (Roche Diagnostic, Mannheim, Germany) with the
analytical range 5–35,000 pg ml−1. The corresponding upper
reference limit was 125 pg ml−1. C-reactive protein (CRP)
was measured by immunoturbidometric assay from Sekisui
Diagnostics (Tokyo, Japan) on an AU2700 analyzer (Olympus
Germany, Beckman Coulter, Krefeld, Germany). The cutoff point
was set as 3 mg L−1. Serum creatine kinase-MB (CK-MB) and
creatine kinase (CK) were detected by DC800 analyzer (Beckman
Coulter, Krefeld, Germany) using commercial kit according to
electrochemiluminescence technology with cutoff values 25 and
195 U L−1, respectively.
Statistical Analysis
Raw data were presented as means mean ± SD unless otherwise
stated. All variables were log-transformed for analysis then
back-transformed to express effects as factors, after adjustment
for the modifying effects of the pre-exercise concentration
and either exercise intensity expressed as percent of heart-
rate reserve (%HRR) or training volume, using a spreadsheet
(Hopkins, 2017). Changes between the two modes of exercise
were compared for the average of post-exercise 0, 1, 4, and 24 h
for all biomarkers except for CRP, which was the average of
24 and 48 h. In the absence of thresholds for acute changes in
cardiac biomarkers associated with substantial increased risk of
morbidity or mortality in endurance athletes, the magnitudes of
the changes were assessed using threshold standardized changes
of 0.20, 0.60, 1.20, 2.0, and 4.0 for small, moderate, large, very
large, and extremely large, respectively (Hopkins et al., 2009;
Hopkins, 2019a).
Uncertainty in the estimates of effects are presented as
90% compatibility limits. Probabilistic decisions about true
(large-sample) magnitudes accounting for the uncertainty were
based on one-sided hypothesis tests of substantial (at least
small) effects followed by Bayesian inference. The p-value for
rejecting an hypothesis of a substantial effect magnitude of
a given sign was the area of the sampling distribution of
the effect with substantial values of that sign (Lakens et al.,
2018), evaluated via log transformation. Effects were considered
decisive with a p-value threshold of <0.05. If an hypothesis
of a substantial magnitude of a given sign was rejected, the
p-value for the hypothesis of the other sign was interpreted as
evidence for that hypothesis, since the p-value corresponds to
the posterior probability of the magnitude of the true effect
in a reference Bayesian analysis with a minimally informative
prior (Hopkins and Batterham, 2016; Hopkins, 2019b). This
p-value is reported qualitatively using the following scale:
0.25–0.75, possibly; 0.75–0.95, likely; 0.95–0.995, very likely;
>0.995, most likely (Hopkins et al., 2009). If neither hypothesis
was rejected, the magnitude of the effect was considered to be
unclear, and the magnitude of the effect is shown without a
probabilistic qualifier.
RESULTS
No runners reported any cardiac symptoms during or after
the exercises. Heart rate during the intermittent trial was
160 ± 12 min−1 (176 ± 12 min−1 for the hard running and
145 ± 13 min−1 for the easy running); during the continuous trial
heart rate was 162 ± 11 min−1. The intensities were 78 ± 6 and
79 ± 5 %HRR for intermittent and continuous trials, respectively,
or 83 ± 4 and 84 ± 4 %peak heart rate. Rating of perceived
exertion during intermittent and continuous trials was 14 ± 3
and 12 ± 3, respectively. Values for pre-exercise concentration of
biomarkers are shown in Table 1, and values for high-sensitivity
cardiac troponin-I pre- and post-exercise are shown in Figure 1.
The baseline level of high-sensitivity cardiac troponin-I in three
runners exceeded the upper reference limit of 26 pg ml−1,
and after intermittent and continuous exercises this limit was
exceeded by 11 and 10 runners, respectively.
The mean and factor SD for the individual averaged changes in
concentrations of biomarkers for the intermittent and continuous
trials are presented in Table 2. These changes were large for
high-sensitivity cardiac troponin-T, moderate for high-sensitivity
TABLE 1 | The pre-exercise concentrations of biomarkers.
Raw
Back-trans.
(mean ± SD)
(mean ×/÷ SD)
High-sensitivity cardiac troponin I
(hs-cTnI, pg ml−1)
70 ± 180
8.6 ×/÷ 6.2
High-sensitivity cardiac troponin T
(hs-cTnT, pg ml−1)
8 ± 10
5.3 ×/÷ 2.0
N-terminal pro brain natriuretic peptide
(NT-pro-BNP, pg ml−1)
28 ± 26
18 ×/÷ 2.6
C-reactive protein (CRP, mg L−1)
1.1 ± 1.1
0.7 ×/÷ 2.9
Creatine kinase-MB (CK-MB, U L−1)
3.9 ± 2.8
3.3 ×/÷ 1.8
Creatine kinase (CK, U L−1)
350 ± 250
289 ×/÷ 1.9
Back-trans.:
back-transformed
mean
and
factor
SD
of
the
log-transformed concentrations.
Frontiers in Physiology | www.frontiersin.org
3
January 2020 | Volume 11 | Article 30
Li et al.
Exercise-Induced Cardiac Biomarkers Elevation
FIGURE 1 | Back-transformed means of log-transformed concentrations of
high-sensitivity cardiac troponin I (hs-cTnI) 1.5 h before and 0–48 h after
continuous and intermittent exercise. Error bars are standard deviations.
cardiac troponin-I and CK-MB, and small for NT-Pro-BNP,
CRP, and CK. The individual changes following intermittent and
continuous exercise, with regression lines showing prediction by
baseline concentration, training volume and exercise intensity,
are shown in Figure 2 for one of the biomarkers, high-
sensitivity cardiac troponin-I. The figure shows apparently
similar modifying effects of baseline concentration and training
volume on high-sensitivity cardiac troponin-I, but a marked
difference in the modifying effect of exercise intensity.
Comparisons of the changes in high-sensitivity cardiac
troponin-I and in the other biomarkers adjusted to the mean
values of the modifiers and to approximately 1 SD above the
mean values are shown in Table 2. In summary, there were
trivial differences between the two modes of exercise at the mean
values of the modifiers, although the differences were not clear
for some measures (high-sensitivity cardiac troponin-T, CK-MB,
CK). Effects of 1 SD of baseline concentration of the marker
and of training volume had at most small observed effects,
and true values for the clear effects were at most only possibly
substantial. However, intensity approximately 1 SD above the
mean of heart-rate reserve (85 %HRR) had clear moderate
effects on high-sensitivity cardiac troponin-I and -T, with high
likelihoods that the true change with intermittent was greater
than that with continuous exercise. With the other biomarkers
only CRP showed a clear difference at 85 %HRR (a small,
possibly substantial increase with intermittent compared with
continuous), effects on the other biomarkers being trivial or
small and unclear.
DISCUSSION
To our knowledge, this is the first study to explore the effect
of intermittent exercise on cardiac biomarker concentrations by
comparing workload-matched continuous exercise in endurance
runners. Our novel findings suggest that there was little difference
in the effect of exercise mode on cardiac-troponin elevation
at typical training intensity, but for runners exercising at
higher relative heart rates, prolonged intermittent exercise is
potentially more damaging to cardiac muscle than comparable
continuous exercise.
Referring to the mean values of high sensitivity cardiac
troponin-I in Figure 1 and Table 1, there was a similar effect
of two forms of exercise. Of course, as markers for diagnosis of
acute myocardial injury in clinical settings, the actual magnitude
of individual cardiac troponin level itself should be considered.
The peak level of high-sensitivity cardiac troponin-I exceeded
the upper reference limit of 26 pg ml−1 in most runners after
both forms of exercise. Others have noted that cardiac troponin
TABLE 2 | Changes in concentration of biomarkers following continuous and intermittent exercise, with magnitude-based decisions for the comparison of the changes
adjusted to mean values of potential modifiers and to 1 SD above baseline concentration, to approximately 1 SD above mean Tvol (67 km week−1), and to
approximately 1 SD above mean exercise intensity (85% HRR).
Factor change scoresa
Effect for intermittent/continuousb
(mean ×/÷ SD)
(mean, ×/÷90CL) and magnitude-based decisionc
Continuous
Intermittent
At mean valuesd
At baseline + 1 SDd
At 67 km week−1
At 85 %HRR
hs-cTnI
4.4 ×/÷ 2.4
4.4 ×/÷ 1.8
1.0, ×/÷1.4 trivial, ↔00
1.7, ×/÷1.7 small, ↑*
1.8, ×/÷2.0 small,↑*
3.4, ×/÷1.9 moderate,↑***
hs-cTnT
3.0 ×/÷ 2.1
3.2 ×/÷ 1.4
1.0, ×/÷1.4 trivial
1.1, ×/÷1.7 trivial, ↔0 ↑*
1.3, ×/÷1.6 trivial, ↔0 ↑*
2.1, ×/÷1.8 moderate,↑**
NT-pro-BNP
1.8 ×/÷ 1.3
1.7 ×/÷ 1.2
1.0, ×/÷1.2 trivial, ↔00
1.1, ×/÷1.3 trivial, ↔0 ↑*
0.9, ×/÷1.4 trivial
1.1, ×/÷1.4 trivial
CRP
1.3 ×/÷ 1.7
1.6 ×/÷ 1.4
1.2, ×/÷1.3 trivial, ↔0 ↑*
1.1, ×/÷1.4 trivial, ↔0 ↑*
1.1, ×/÷1.3 trivial, ↔00
1.3, ×/÷1.4 small, ↑*
CK-MB
1.6 ×/÷ 1.5
1.7 ×/÷ 1.6
1.1, ×/÷1.5 trivial
1.0, ×/÷1.9 trivial
1.0, ×/÷2.0 trivial
1.0, ×/÷1.9 trivial
CK
1.4 ×/÷ 1.4
1.5 ×/÷ 1.5
1.0, ×/÷1.4 trivial
1.2, ×/÷1.6 small, ↑
1.0, ×/÷1.5 trivial
1.1, ×/÷1.7 small, ↑
SD, standard deviation; 90CL, 90% confidence limits as factors; Tvol, training volume, km week−1; %HRR, percent of heart-rate reserve; hs-cTnI, high-sensitivity cardiac
troponin I, pg ml−1; hs-cTnT, high-sensitivity cardiac troponin T, pg ml−1; NT-pro-BNP, N-terminal pro brain natriuretic peptide, pg ml−1; CRP, C-reactive protein, mg L−1;
CK-MB, creatine kinase-MB, U L−1; CK, creatine kinase, U L−1. aChanges, adjusted to overall mean values of baseline concentration and intensity, are for the mean
of 0, 1, 4, and 24 h post-exercise, except for CRP, which is the change for the mean of 24 and 48 h post-exercise. bAll effects are adjusted for baseline concentration.
cThe qualitative magnitude shown is derived by standardization for the observed effect. For clear effects, likelihood of a true substantial increase (↑) is indicated as
*possible; **likely; ***very likely. Likelihood of a true trivial change (↔) is indicated as follows: 0possible; 00likely. dMeans and SD are shown in Table 1. These effects are
adjusted for intensity.
Frontiers in Physiology | www.frontiersin.org
4
January 2020 | Volume 11 | Article 30
Li et al.
Exercise-Induced Cardiac Biomarkers Elevation
FIGURE 2 | Individual factor changes of high-sensitivity cardiac troponin-I (hs-cTnI) averaged over 0–24 h following intermittent and continuous exercise, with
regression lines showing prediction by baseline concentration, training volume and exercise intensity. Horizontal dotted lines indicate no change in concentration;
vertical dotted lines indicate mean values of the predictors.
often exceeds the upper reference limit following exercise (Vilela
et al., 2014; Gresslien and Agewall, 2016). Although strongly
needed, there are no agreed thresholds for acute changes in
cardiac troponin associated with substantial increased risk of
morbidity or mortality in athletes, as the clinical relevance of
exercise-induced cardiac troponin is still under debate (Skadberg
et al., 2017; Nie, 2018). It has been argued previously that cardiac
troponin is not randomly elevated after vigorous exercise but
increases in certain “susceptible” individuals (Nie et al., 2011;
Tian et al., 2014; Legaz-Arrese et al., 2015). We discuss below
the extent to which resting concentration of cardiac troponin
and level of training are two characteristics that modify release
of cardiac troponin following exercise in our study.
Our findings for the effect of intensity of intermittent
exercise are consistent with other studies, reviewed in the
Introduction, showing greater elevation of cardiac biomarkers
with higher intensity
exercise.
In
particular,
it obtained
pronounced high-sensitivity cardiac troponin-T elevation after
intermittent exercise with average heart rate of 160 min−1, which
was similar to the mean heart rate of our runners (Richardson
et al., 2018). Stewart et al. (2016) suggested that a heart rate
of 145 min−1 was a threshold for the elevation of cardiac
troponin. Gresslien and Agewall (2016) and Nie et al. (2018)
also presented evidence for a threshold exercise intensity. With
the prolonged exercise of trained runners in our study, there is
evidence of a threshold heart rate for intermittent exercise, but
not for continuous exercise (Figure 2, right-hand scatterplot).
The greater increase in cardiac troponin concentrations with
intermittent compared with continuous exercise at high relative
intensities is obviously due to the higher peak heart rates
during the on-phase of each interval; the lower heart rates
during the off-phase do not compensate for the on-phase for
high relative intensities, but they more than compensate for
low relative intensities, on average. Clearly, however, there are
individual differences in the response to both types of exercise
that are not explained by relative exercise intensity across the
range of intensities.
As shown in Figure 2, the baseline level of cardiac troponin
and training volume can explain some of the individual
differences in response to the intermittent and continuous
exercise. The three participants with the highest baseline
concentrations of cardiac troponin-I experienced absolutely no
increase with either form of exercise, so any cardiac pathology
represented by the concentration of this marker is apparently not
exacerbated by exercise in these participants. In the analysis that
included all participants with adjustment for intensity, there was
only a small or trivial mean increase with intermittent compared
with continuous exercise at high baseline concentrations of
either form of cardiac troponin, and the true increases were
only possibly substantial. We therefore do not consider that
intermittent exercise is a concern for runners with high baseline
values of cardiac troponin, at least with this sample: if anything,
baseline concentration had a protective effect for exercise-
induced increases. There was a similar protective effect for
training volume, again with only a possibly small relative mean
increase for intermittent compared with continuous exercise for
participants with high training volumes.
We found small increases in NT-pro-BNP after both forms
of exercise, with little difference between two exercise modes.
NT-pro-BNP is a marker of cardiac stress elicited from volume
or pressure overload in clinical settings of myocardial injury
(Kociol et al., 2010). The magnitude of increase in NT-pro-BNP
is primarily dependent on exercise duration as well as basal
concentration, but not exercise intensity (Scharhag et al., 2008;
Legaz-Arrese et al., 2011; Serrano-Ostariz et al., 2011). There
seems a “ceiling effect” that NT-pro-BNP is maximized at a low
exercise intensity so that further increase requires accumulation
over time (Legaz-Arrese et al., 2011). We also observed small
increases in CRP after intermittent and continuous exercise and
possible small effect of exercise mode at 85 %HRR. CRP and
other inflammatory markers have been reported to increase 24–
74 h after endurance exercise (Scherr et al., 2011). Stewart et al.
(2016) also observed CRP increased to a greater extent with
intermittent exercise accompanying cardiac troponin elevations.
Frontiers in Physiology | www.frontiersin.org
5
January 2020 | Volume 11 | Article 30
Li et al.
Exercise-Induced Cardiac Biomarkers Elevation
Although this marker is not cardiac-specific, inflammatory
cytokines related to cardiac injury have been reported to affect the
release of cardiac troponin after exercise (La Gerche et al., 2015).
Inflammation may therefore have contributed to the exercise-
induced cardiac troponin elevation via a transient increase in
myocardial membrane permeability. Further research is needed
to address this issue.
LIMITATIONS AND PERSPECTIVE
The unclear comparisons of exercise mode and some of the
unclear moderating effects for the markers other than cardiac
troponin need resolving with a larger sample size, which
would also improve the precision of the estimates of the effect
with cardiac troponin. Other factors that may contribute to
the individual variation of exercise-induced cardiac troponin
elevation have not been verified for the small sample size.
Competing at similar intensity, younger age predicts cardiac
troponin elevation among endurance runners (Eijsvogels et al.,
2015). Sex effect may also influence the individual variation as
one female runner was included in the present study. Exercise-
induced cardiac troponin elevation occurred in both male
and female runners, lower in female (Kong et al., 2017), but
not exactly the case in a meta-analysis (Shave et al., 2007).
Furthermore, core temperature, diet and fluid intake, which may
have impact on the performance of endurance runners were not
monitored between the two trial.
Cardiac troponin, a biomarker to diagnosis acute myocardial
infarction is moderately to largely increased after typical long
hard training in endurance runners. Some runners training with
high-intensity intermittent exercise for improved performance
may be at increased risk of cardiac stress. Baseline concentration
of cardiac biomarkers and training volume have protective effects
for cardiac stress of endurance runners participating with high
training volumes. Training can be a potentially precondition to
better cardiomyocyte tolerance to intense exercise.
CONCLUSION
Intermittent and continuous exercise at similar mean heart rates
have similar mean effects on elevation of cardiac troponins in
marathon runners performing long hard training. Elevation of
cardiac troponins is greater at higher intensities and less at lower
intensities with intermittent but not continuous exercise. Baseline
level of cardiac troponin and training volume can explain some
of the individual differences in response to intermittent and
continuous exercise.
DATA AVAILABILITY STATEMENT
All datasets generated for this study are included in the
article/supplementary material.
ETHICS STATEMENT
The studies involving human participants were reviewed and
approved by the Beijing Sport University (BSUIRB, 2019003H).
The patients/participants provided their written informed
consent to participate in this study.
AUTHOR CONTRIBUTIONS
FL, JN, HZ, FF, LY, and YLu conceived and designed the study. FL,
LY, WH, YLi, and YLu performed the experiments and analyzed
the data. All authors drafted the manuscript.
FUNDING
Financial support was from a research grant received from the
Hebei Normal University (No. L062018B14) and Fundamental
Research
Funds
for
the
Central
Universities
of
China
(No. 2018GJ018).
ACKNOWLEDGMENTS
We acknowledge the Institute of Sport and Health Sciences and
the Key Laboratory of Exercise Stress and Adaptation of Beijing
Sport University for the data collection.
REFERENCES
Billat, L. V. (2001). Interval training for performance: a scientific and empirical
practice. Special recommendations for middle- and long-distance running. Part
I: aerobic interval training. Sports Med. 31, 13–31. doi: 10.2165/00007256-
200131010-00002
Borg, G. A. (1982). Psychophysical bases of perceived exertion. Med. Sci. Sports
Exerc. 14, 377–381.
Eijsvogels, T., George, K., Shave, R., Gaze, D., Levine, B. D., Hopman, M. T. E., et al.
(2010). Effect of prolonged walking on cardiac troponin levels. Am. J. Cardiol.
105, 267–272. doi: 10.1016/j.amjcard.2009.08.679
Eijsvogels, T. M., Hoogerwerf, M. D., Maessen, M. F., Seeger, J. P., George,
K. P., Hopman, M. T., et al. (2015). Predictors of cardiac troponin release
after a marathon. J. Sci. Med. Sport 18, 88–92. doi: 10.1016/j.jsams.2013.
12.002
Eijsvogels, T. M. H., Fernandez, A. B., and Thompson, P. D. (2016). Are there
deleterious cardiac effects of acute and chronic endurance exercise? Physiol. Rev.
96, 99–125. doi: 10.1152/physrev.00029.2014
Fu, F., Nie, J., and Tong, T. K. (2009). Serum cardiac troponin T in adolescent
runners: effects of exercise intensity and duration. Int. J. Sports Med. 30,
168–172. doi: 10.1055/s-0028-1104586
George, K., Whyte, G. P., Green, D. J., Oxborough, D., Shave, R. E., Gaze, D., et al.
(2012). The endurance athletes heart: acute stress and chronic adaptation. Br. J.
Sports Med. 46(Suppl. 1), i29–i36. doi: 10.1136/bjsports-2012-091141
Gresslien, T., and Agewall, S. (2016). Troponin and exercise. Int. J. Cardiol. 221,
609–621. doi: 10.1016/j.ijcard.2016.06.243
Hopkins, W. G. (2017). Spreadsheets for analysis of controlled trials, crossovers
and time series. Sportscience 21, 1–4.
Hopkins, W. G. (2019a). Compatibility intervals and magnitude-based decisions
for standardized differences and changes in means. Sportscience 23, 1–4.
Frontiers in Physiology | www.frontiersin.org
6
January 2020 | Volume 11 | Article 30
Li et al.
Exercise-Induced Cardiac Biomarkers Elevation
Hopkins, W. G. (2019b). Magnitude-based decision. Sportscience 23, 5–7.
Hopkins, W. G., and Batterham, A. M. (2016). Error rates, decisive outcomes and
publication bias with several inferential methods. Sports Med. 46, 1563–1573.
doi: 10.1007/s40279-016-0517-x
Hopkins, W. G., Marshall, S. W., Batterham, A. M., and Hanin, J. (2009).
Progressive statistics for studies in sports medicine and exercise science. Med.
Sci. Sports Exerc. 41, 3–13. doi: 10.1249/MSS.0b013e31818cb278
Kociol, R. D., Pang, P. S., Gheorghiade, M., Fonarow, G. C., O’Connor, C. M., and
Felker, M. (2010). Troponin elevation in heart failure prevalence, mechanisms,
and clinical implications. J. Am. Coll. Cardiol. 56, 1071–1078. doi: 10.1016/j.
jacc.2010.06.016
Kong, Z. W., Nie, J. L., Lin, H., George, K., Zhao, G., Zhang, H. F., et al. (2017). Sex
differences in release of cardiac troponin T after endurance exercise. Biomarkers
22, 345–350. doi: 10.1080/1354750X.2016.1265007
La Gerche, A., Inder, W. J., Roberts, T. J., Brosnan, M. J., Heidbuchel, H.,
and Prior, D. L. (2015). Relationship between inflammatory cytokines and
indices of cardiac dysfunction following intense endurance exercise. PLoS One
10:e0130031. doi: 10.1371/journal.pone.0130031
Lakens, D., Scheel, A. M., and Isager, P. M. (2018). Equivalence testing for
psychological research: a tutorial. Adv. Methods Pract. Psychol. Sci. 1, 259–269.
doi: 10.1177/2515245918770963
Legaz-Arrese, A., George, K., Carranza-Garcia, L. E., Munguia-Izquierdo, D.,
Moros-Garcia, T., and Serrano-Ostariz, E. (2011). The impact of exercise
intensity on the release of cardiac biomarkers in marathon runners. Eur. J. Appl.
Physiol. 111, 2961–2967. doi: 10.1007/s00421-011-1922-3
Legaz-Arrese, A., Lopez-Laval, I., George, K., Jose Puente-Lanzarote, J., Castellar-
Otin, C., Reverter-Masia, J., et al. (2015). Individual variability of high-
sensitivity cardiac troponin levels after aerobic exercise is not mediated by
exercise mode. Biomarkers 20, 219–224. doi: 10.3109/1354750x.2015.1068851
Li, F., Hu, Y., Nie, J., and Fu, F. H. (2016a). Effects of acute, intermittent exercise
in hypoxic environments on the release of cardiac troponin. Scand. J. Med. Sci.
Sports 26, 397–403. doi: 10.1111/sms.12463
Li, F., Nie, J., Lu, Y., Tong, T. K., Yi, L., Yan, H., et al. (2016b). The impact of
intermittent exercise in a hypoxic environment on redox status and cardiac
troponin release in the serum of well-trained marathon runners. Eur. J. Appl.
Physiol. 116, 2045–2051. doi: 10.1007/s00421-016-3460-5
Nie, J., George, K., Duan, F., Tong, T. K., and Tian, Y. (2016). Histological evidence
for reversible cardiomyocyte changes and serum cardiac troponin T elevation
after exercise in rats. Physiol. Rep. 4:e13083. doi: 10.14814/phy2.13083
Nie, J. L. (2018). Cardiac troponins: potential biomarkers for the detection of
subclinical coronary artery disease in athletes participating in endurance sports.
Eur. J. Prev. Cardiol. 25, 110–111. doi: 10.1177/2047487317711744
Nie, J. L., Close, G., George, K. P., Tong, T. K., and Shi, Q. D. (2010). Temporal
association of elevations in serum cardiac troponin T and myocardial oxidative
stress after prolonged exercise in rats. Eur. J. Appl. Physiol. 110, 1299–1303.
doi: 10.1007/s00421-010-1604-6
Nie, J. L., George, K. P., Tong, T. K., Tian, Y., and Shi, Q. D. (2011). Effect of
repeated endurance runs on cardiac biomarkers and function in adolescents.
Med. Sci. Sports Exerc. 43, 2081–2088. doi: 10.1249/MSS.0b013e31821d
4a82
Nie, J. L., Zhang, H. F., Kong, Z. W., George, K., Little, J. P., Tong, T. K.,
et al. (2018). Impact of high-intensity interval training and moderate-
intensity continuous training on resting and postexercise cardiac troponin T
concentration. Exp. Physiol. 103, 370–380. doi: 10.1113/EP086767
Ranjbar, R., Ahmadi, M. A., Zar, A., and Krustrup, P. (2017). Acute effect of
intermittent and continuous aerobic exercise on release of cardiac troponin
T in sedentary men. Int. J. Cardiol. 236, 493–497. doi: 10.1016/j.ijcard.2017.
01.065
Richardson, A. J., Leckie, T., Watkins, E. R., Fitzpatrick, D., Galloway, R., Grimaldi,
R., et al. (2018). Post marathon cardiac troponin T is associated with relative
exercise intensity. J. Sci. Med. Sport 21, 880–884. doi: 10.1016/j.jsams.2018.
02.005
Scharhag, J., George, K., Shave, R., Urhausen, A., and Kindermann, W. (2008).
Exercise-associated increases in cardiac biomarkers. Med. Sci. Sports Exerc. 40,
1408–1415. doi: 10.1249/MSS.0b013e318172cf22
Scherr, J., Braun, S., Schuster, T., Hartmann, C., Moehlenkamp, S., Wolfarth, B.,
et al. (2011). 72-h kinetics of high-sensitive troponin T and inflammatory
markers after marathon. Med. Sci. Sports Exerc. 43, 1819–1827. doi: 10.1249/
MSS.0b013e31821b12eb
Serrano-Ostariz, E., Terreros-Blanco, J. L., Legaz-Arrese, A., George, K., Shave, R.,
Bocos-Terraz, P., et al. (2011). The impact of exercise duration and intensity
on the release of cardiac biomarkers. Scand. J. Med. Sci. Sports 21, 244–249.
doi: 10.1111/j.1600-0838.2009.01042.x
Shave, R., George, K. P., Atkinson, G., Hart, E., Middleton, N., Whyte, G., et al.
(2007). Exercise-induced cardiac troponin T release: a meta-analysis. Med. Sci.
Sports Exerc. 39, 2099–2106. doi: 10.1249/mss.0b013e318153ff78
Skadberg, O., Kleiven, O., Bjorkavoll-Bergseth, M., Melberg, T., Bergseth, R.,
Selvag, J., et al. (2017). Highly increased Troponin I levels following high-
intensity endurance cycling may detect subclinical coronary artery disease
in presumably healthy leisure sport cyclists: the North Sea Race Endurance
Exercise Study (NEEDED) 2013. Eur. J. Prev. Cardiol. 24, 885–894. doi: 10.1177/
2047487317693130
Stewart, G. M., Yamada, A., Haseler, L. J., Kavanagh, J. J., Chan, J., Koerbin, G.,
et al. (2016). Influence of exercise intensity and duration on functional and
biochemical perturbations in the human heart. J. Physiol. 594, 3031–3044. doi:
10.1113/JP271889
Thygesen, K., Alpert, J. S., Jaffe, A. S., Simoons, M. L., Chaitman, B. R., White,
H. D., et al. (2012). Third universal definition of myocardial infarction. Nat.
Rev. Cardiol. 9, 620–633. doi: 10.1038/nrcardio.2012.122
Tian, Y., Nie, J. L., George, K. P., and Huang, C. Y. (2014). Reproducibility of
cardiac biomarkers response to prolonged treadmill exercise. Biomarkers 19,
114–120. doi: 10.3109/1354750X.2014.880855
Tian, Y., Nie, J. L., Huang, C. Y., and George, K. P. (2012). The kinetics of
highly sensitive cardiac troponin T release after prolonged treadmill exercise
in adolescent and adult athletes. J. Appl. Physiol. 113, 418–425. doi: 10.1152/
japplphysiol.00247.2012
Tschakert, G., and Hofmann, P. (2013). High-intensity intermittent exercise:
methodological and physiological aspects. Int. J. Sports Physiol. Perform. 8,
600–610. doi: 10.1123/ijspp.8.6.600
Vilela, E. M., Bastos, J. C. C., Rodrigues, R. P., and Nunes, J. P. L. (2014). High-
sensitivity troponin after running - a systematic review. Netherl. J. Med. 72,
5–9.
Weippert, M., Divchev, D., Schmidt, P., Gettel, H., Neugebauer, A., Behrens,
K., et al. (2016). Cardiac troponin T and echocardiographic dimensions after
repeated sprint vs. moderate intensity continuous exercise in healthy young
males. Sci. Rep. 6:24614. doi: 10.1038/srep24614
Conflict of Interest: The authors declare that the research was conducted in the
absence of any commercial or financial relationships that could be construed as a
potential conflict of interest.
The handling Editor and reviewer AA declared their involvement as co-editors in
the Research Topic, and confirm the absence of any other collaboration.
Copyright © 2020 Li, Nie, Zhang, Fu, Yi, Hopkins, Liu and Lu. This is an open-access
article distributed under the terms of the Creative Commons Attribution License
(CC BY). The use, distribution or reproduction in other forums is permitted, provided
the original author(s) and the copyright owner(s) are credited and that the original
publication in this journal is cited, in accordance with accepted academic practice. No
use, distribution or reproduction is permitted which does not comply with these terms.
Frontiers in Physiology | www.frontiersin.org
7
January 2020 | Volume 11 | Article 30
| Effects of Matched Intermittent and Continuous Exercise on Changes of Cardiac Biomarkers in Endurance Runners. | 01-31-2020 | Li, Feifei,Nie, Jinlei,Zhang, Haifeng,Fu, Frank,Yi, Longyan,Hopkins, Will,Liu, Yang,Lu, Yifan | eng |
PMC7037282 | International Journal of
Environmental Research
and Public Health
Article
Amateur Runners’ Commitment: An Analysis of
Sociodemographic and Sports Habit Profiles
David Parra-Camacho 1
, Manuel Alonso Dos Santos 2
and
María Huertas González-Serrano 3,*
1
Department of Physical Education and Sports, Faculty of Physical Activity and Sport Sciences,
Universitat de València, 46010 Valencia, Spain; [email protected]
2
Department of Management, Faculty of Economics and Business Administration, Universidad Católica de la
Santísima Concepción, Concepción 4090541, Chile; [email protected]
3
Department of Teaching and Learning of Physical Education, Plastic and Music Education,
Universidad Católica de Valencia, 46110 Valencia, Spain
*
Correspondence: [email protected]; Tel.: +34-963-983-707
Received: 14 January 2020; Accepted: 30 January 2020; Published: 2 February 2020
Abstract: The aim of this work is to analyse the commitment to running among urban runners by
identifying groups regarding commitment to this sport and by defining their sociodemographic profile
and their sports habits. A sample of 1806 participants in popular urban races in the city of Valencia
was interviewed using an 11-item questionnaire on commitment to running, sociodemographic
characteristics, and sports habits. The psychometric properties of the running-commitment scale
allowed for the identification of two factors in commitment to running: enthusiasm for running
(6 items) and affliction from running (5 items). Subsequently, a cluster analysis combining hierarchical
and non-hierarchical methods was performed, identifying three groups of runners: highly committed
(n = 650), moderately committed (n = 749), and slightly committed (n = 407). Highly committed
runners positively rate all aspects of running enthusiasm (M = 4.15), while moderately committed
runners show a more neutral attitude (M = 3.41) and slightly committed runners disagree on these
aspects (M = 2.41). Both highly (M = 1.32) and moderately (M = 2.04) committed runners disagree
on the affliction-related aspects of running, while slightly committed runners show a trend towards
neutrality on some affliction indicators. The variables referring to age, level of studies, sports habits,
and running addiction contributed to differentiating the identified groups.
Keywords: runners; commitment to running; clusters; sports habits; running addiction; urban
runners; amateur runners
1. Introduction
Today, running has become an immensely popular pastime practised in the public sphere by
millions of recreational participants around the world. However, until the 1960s, recreational jogging
in streets, parks, or forests was considered an odd activity [1]. An example of the boom in running
that has taken place in some Spanish cities can be found in Valencia (Spain), where currently more
than 30 popular races are organized every year [2]. Valencia is a city with a strong running tradition,
with runners participating in this activity since the beginning of the 19th century [2]. Also, the number of
nonprofessional sports clubs offering different sports activities has been increasing recently in the main
European cities [3,4]. However, both nonprofessional and professional clubs have limited information
about the profiles of athletes based on their commitment to the activities they offer. The groups of
runners are heterogeneous and have different subcultures; they could be differentiated in the frequency
of the practice of the sport, in the motives, or in the participation in competitions [5]. For each group,
Int. J. Environ. Res. Public Health 2020, 17, 925; doi:10.3390/ijerph17030925
www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2020, 17, 925
2 of 19
sports associations and clubs and governmental or nonprofit organizations can design communications,
activities, and services specifically targeted at them. Commitment could be a segmentation variable to
obtain different groups of subjects with different motivations, activities, and interests.
Commitment is a psychological construct that, from a sporting point of view, represents the desire
and willingness to continue practising a sport [6,7]. When associated with positive factors, such as the
intrinsic enjoyment of the activity itself; opportunities to participate successfully; personal investments
of time, money, and experience in the sport; and social pressures from parents, coaches, peers, etc.,
commitment increases in tandem with increases in these factors [8]. However, when associated with
negative factors, such as alternatives for successful participation in other attractive sports, commitment
decreases [8].
Another concept that has been associated with or related to this construct in previous research
on commitment to running is the notion of a positive addiction to running, defined by Glasser [9]
as enjoyment that increases mental strength and, when lost, produces some type of physical or
psychological discomfort [10]. This concept is opposed to that of a negative addiction to continuous
running, which can dominate the runner’s life and can produce the unwanted effects of overtraining
syndrome [11]: fatigue, decreased performance, and mood disorders.
Some studies have observed among endurance sports participants a relationship between
commitment to and dependence on exercise [12]. Endurance athletes usually train a considerable
number of hours, which some studies have found to have a positive correlation with addiction to
running among marathon runners [13,14]. However, most research indicates that commitment to
running and negative addiction to running are two different concepts predicted by different variables
and that, although there is a high correlation between them, the two phenomena do not necessarily
appear together [15–19].
In this regard, Pargman [20] built on these concepts (positive and negative addiction to running)
to describe the existence of two types of runners: addict-dependent runners, who have a positive
addiction towards continuous running and perceive joy and happiness in running (if they do not
do it, they feel bad), and committed-dedicated runners, who have a negative addiction towards
continuous running and a broader, rational, and pragmatic intellectual component that makes them
give continuous running a very high priority in their lives without necessarily liking it.
Among the first contributions on commitment to running, Carmack and Martens [21],
who developed and validated the unidimensional commitment to running (CR) scale, equate
commitment to running with the concept of positive addiction to continuous running. Subsequently,
several studies have used this scale to analyse the characteristics of runners and to learn about their
sociodemographic profile and sports habits. Carmack and Martens [21], and Joseph and Robbins [22]
explained that the variables that contributed to increasing the commitment to running were related to
the time spent on training (number of days of training/week and minutes/training), while Thornton
and Scott [23] also highlighted distance covered in kilometres as a predictor variable. Improved
personal performance among marathon runners was an important factor in increasing the commitment
to running for this profile of runners, while amateur runners emphasized the intrinsic enjoyment of
the activity or improved health [24].
In the Spanish context, several investigations have also been carried out among Spanish runners on
the psychological variables related to motivation, commitment, and addiction to running [5,19,25–27].
In this regard, it is important to highlight the validation of the Carmack and Martens scale [21]
in Spanish marathon runners by Ruiz-Juan and Zarauz [25], who retained 11 indicators from the
12 proposed on the original unidimensional scale that allowed for evaluation of the commitment of
the runners. Several studies have analysed the relationships between training variables and sports
habits and the commitment and addiction to running in marathon runners [18,19,28]. In addition,
the motivation of amateur Spanish runners when participating in this physical activity has also been
analysed in various studies [5,26,27].
Int. J. Environ. Res. Public Health 2020, 17, 925
3 of 19
However, studies that segment runners into groups with different behaviours towards running
are not abundant, and most of the contributions in this regard have focused on identifying groups or
on classifying runners according to their motivation to practice this physical activity [5,29,30] rather
than according to their commitment to running.
Therefore, numerous contributions that analyse diverse psychological variables such as runners’
motivation [17,20,26,27,30–35], addiction [15,19,28,36], or commitment [12,19,21–23,25,37] to running
have been made in the study of the running phenomenon. However, many of these studies have
focused on long-distance runners, such as half-marathon and marathon runners [19,28,31]. There are
few contributions that focus on Spanish urban runners who participate in short-distance races [5,26,27]
and that classify runners into clusters according to their commitment to running, identifying
groups with different sociodemographic characteristics and sports habits. Research on short-distance
(versus long-distance) races allows researchers to study a larger universe because more and more
people practice sports for health and social reasons but are less concerned about winning long-distance
races [38].
This paper analyses the commitment to running among runners who participated in urban races
of less than 10 km. These data provide information on the behaviour of this type of urban sportspeople
in the context of Spanish short-distance runners. Spain is a country with an established running
practice: the percentage of the population who practice running is 30.4% [39].
The aim of this paper is to analyse the commitment to running among urban runners by identifying
groups with a greater or lesser commitment to this sport. In addition, the profile of these groups is
defined to determine their sociodemographic characteristics and to be able to analyse their sports habits.
2. Materials and Methods
2.1. Sample, Procedures, and Questionnaire
In this study, a sample of 1806 runners who participated in the city of Valencia’s urban circuit of
popular races during 2015 were interviewed. The circuit has 10 races. The total number of runners in
the circuit was 50,038 recurring runners (M = 5004). The sample size used is at least 95% confidence
(5% margin of error). The questionnaires were collected after the end of this race circuit during
the month of January 2016 by using an online format. This research used a self-supplied online
questionnaire because of the difficulty of obtaining valid responses among active or tired athletes
immediately after the race.
The questionnaire was made up of three sections of questions. The first included questions of a
sociodemographic nature referring to age, gender, occupation, level of studies, and income.
The second section contained questions referring to the sports habits of the participants: number
of years they have been running, weekly frequency with which they run, preferred distance when
participating in a popular race, affiliation with a club, whether the club is federated, with whom they
usually run, the distance they usually run weekly, the athletic level they consider themselves to have as
runners, the number of long-distance races finished (half marathons and marathons), future intentions
regarding participation in urban races, and three items on addiction to running. The three indicators on
the intentions of amateur runners were adapted from the scale on behavioural intentions by Zeithaml,
Berry, and Parasuraman [40], whereas the three indicators on running addiction were extracted from
the Spanish validation by Zarauz and Ruiz-Juan [36] of the scale on running addiction. The adaptation
of the scale consisted in translating and adapting the scales to the race under research. The indicators
on behavioural intention of amateur runners showed adequate reliability (Cronbach alpha = 0.91,
Composite Reliability = 0.92; Average Variance Extracted = 0.79). The indicators of both future
intentions and running addiction were assessed with a five-point Likert scale with the following
response options: 1 = strongly disagree, 2 = disagree, 3 = neither disagree nor agree, 4 = agree,
and 5 = strongly agree.
Int. J. Environ. Res. Public Health 2020, 17, 925
4 of 19
The third section includes indicators related to the participants’ commitment to running.
The Ruiz-Juan and Zarauz scale [25] validated in the Spanish context in marathon runners was
used to collect participants’ perceptions of their own commitment to running. This scale is composed
of 11 indicators evaluated by the interviewees through a Likert scale with the same response options as
those of future intentions and running addiction.
For this type of research, it was not necessary to get approval from the Committee of Ethics of the
university where this study was carried out (University of Valencia). According to the Committee of
Ethics and Human Research from this university, it is not necessary to get approval to carry out an
opinion survey about a topic or issue, professional status, or satisfaction with certain matters.
However, it is obligatory to include a preamble in the survey with presented information about
the project (theme and purpose), the benefits that the information collected by the survey may provide,
the willingness of the participation, and the anonymous treatment of data (Data Protection Law). It is
also compulsory to indicate a contact person to ask for further information and to put one paragraph
in which the survey respondent voluntarily accepts participation in the research and gives consent
tacitly when responding to the survey. Thus, following these guidelines indicated by the Committee of
Ethics and Human Research from the University of Valencia to develop this sort of research, all this
information was added at the beginning of the questionnaire.
2.2. Statistical Analysis
First, the psychometric properties of the running commitment scale were tested on the sample
under study by performing an exploratory factor analysis (EFA) and a confirmatory factor analysis
(CFA). The EFA was performed with the FACTOR program following the recommendations of
Lloret-Segura et al. [41] using the maximum likelihood (ML) method and oblimin direct rotation.
To determine the number of factors, the optimal implementation of the parallel analysis procedure was
used [42]. The model fit was observed using the root mean square of the residuals (RMSR) coefficient
as well as the goodness-of-fit index (GFI) proposed by Tanaka and Huba [43]. The RMSR should be less
than 0.05 [44], and the GFI value should be less than 0.95 [45]. Other indicators that were taken into
account were the generalized G-H index (>0.80) to analyse the replicability of the factors derived from
the EFA [46]. The measures for sample adequation of Kaiser Meyer Olkin (KMO) were also observed,
as was Bartlett’s sphericity test. Items with factorial loads below 0.40 or above this value on two or
more factors were eliminated before the next EFA was carried out.
After applying the EFA, a CFA was performed on the factorial solution derived from the EFA
using robust maximum likelihood estimation (MLR) with the aim of correcting the possible absence of
multivariant normality using statistics such as the χ2 of Satorra-Bentler [47]. Thus, for the evaluation of
global fit, different goodness-of-fit indices recommended in the literature [48], such as the significance
of the chi-squared value and its robust correction offered by Satorra-Bentler (S-B χ2) [49], were used.
In addition, other coefficients that allowed for testing the adequacy of the proposed models, such
as the ratio of χ2 and its degrees of freedom (χ2/df) [50], were calculated, with the acceptable values
being less than five [51]. However, these indices are affected by the sample size, so the standardized
root mean square residual (SRMR) index was used, where values of 0.06 or less indicate an excellent
fit and values of 0.08 or less indicate a good fit [51]. In the same way, the coefficients of the indices
of robust goodness-of-fit of the proposed model, the compared fit index (CFI) and the incremental
fit index (IFI), were tested. For these indicators, a fit with values above 0.90 is considered good [52].
To finalize, the root mean square error of approximation (RMSEA) is shown, with a score below
0.08 being considered a good fit [53].
In the evaluation of the reliability of the scale, three measurements were taken into account:
the Cronbach’s alpha, composite reliability (CR), and average variance extracted (AVE) values for each
factor [54]. In addition, convergent validity was tested through the significance of the factorial loads in
their respective dimensions and the values of the associated t-tests. Additionally, discriminant validity,
which has to do with seeing the clear distinction between any pair of constructs, was evaluated using
Int. J. Environ. Res. Public Health 2020, 17, 925
5 of 19
the method suggested by Fornell and Larcker [55]. This method confirms discriminant validity if the
square root of the AVE value of a determined factor is greater than the correlation coefficients between
the factor and any other in the proposed scale. The other criterion to assure discriminant validity is
that the correlations between the different pairs of factors must be less than 0.85 [48].
After checking the psychometric properties of the running commitment scale, a cluster analysis
was performed to identify groups of participants with different characteristics according to their
running commitment. This analysis was performed using the statistical program SPSS version 24.0 for
Windows (IBM, Armonk, NY, USA), with the 11 items of the running commitment scale. Two estimation
methods (hierarchical and nonhierarchical) of the cluster solution were combined to optimize the
results. The hierarchical cluster analysis was performed using the Ward clustering process and, as a
measure of similarity, the Euclidean distance squared. A nonhierarchical analysis was applied to
the groups proposed in the previous analysis through the K-means method using as initial centres
the means of the variables obtained for each cluster solution of the hierarchical analysis. To define
the characteristics of the group profiles and to evaluate predictive validity, ANOVAs and chi-square
tests were performed with variables that were not included in the initial analysis. The value of the
contingency coefficient (C) was also used to check the intensity of the association and the size of the
effect of the related variables.
3. Results
3.1. Sociodemographic Characteristics
Table 1 shows the sociodemographic characteristics. The respondents had an average age of 39.48
(SD = 9.21), 74.4% being men and 25.6% being women. In terms of occupation, most are employed
(88.4%). The majority was university studies (59.2%) and had an income level of less than 18,000 euros
per year (51.3%).
Table 1. Sociodemographic characteristics of the sample.
Variable
Response Option
Mean and Percentage
Age
39.48
(SD 1 = 9.21)
Gender
Male
74.4%
Female
25.6%
Occupation
Employee
88.4%
Unemployed
6.1%
Student
1.2%
Other (retired, pensioner, domestic tasks, etc.)
4.3%
Level of Studies
Primary
6.1%
Secondary
6.2%
Baccalaureate/ Professional training
28.5%
University
59.2%
Income Level
Less than 12,000 euros
27.0%
12,001–18,000 euros per year
24.3%
18,001–24,000 euros per year
18.2%
24,001–30,000 euros per year
14.2%
30,001–36,000 euros per year
6.9%
More than 36,001 euros per year
9.5%
1 SD = Standard Deviation.
3.2. Psychometric Properties of the Commitment to Running Scale
First, the properties of the items on the commitment-to-running scale were analysed by checking
the corrected correlation item–total values, as well as the mean, standard deviation, asymmetry,
and kurtosis values. Table 2 shows the statistics. The mean scores of five indicators were inverted as
indicated by Ruiz-Juan and Zarauz [25]. The values of the item–total corrected correlation coefficients
Int. J. Environ. Res. Public Health 2020, 17, 925
6 of 19
were higher than the cutoff point recommended by the literature (≥0.30) [56]. Additionally, the values
of asymmetry and kurtosis are acceptable for most variables because they are lower than 3.0 [57]
except for item CR7, of which the value exceeded the recommended cutoff point for ensuring a normal
distribution of the data. On the other hand, indicators on the intentions of amateur runners showed
adequate reliability with a Cronbach alpha of 0.91, CR = 0.92, and AVE = 0.79.
Table 2. Mean, standard deviation, corrected item–total correlation, alpha if the item is removed,
asymmetry, and kurtosis values of the indicators of the running-commitment scale.
Number
Items
Means (SD) 1
R IT-c 2
α without
Item
Asymmetry
Kurtosis
CR1
I am looking forward to running
3.79 (0.93)
0.62
0.79
−0.58
0.21
CR2
Running is drudgery (R)
3.83 (1.12)
0.53
0.80
−0.68
−0.46
CR3
I do not enjoy running (R)
4.34 (0.91)
0.48
0.80
−1.54
2.21
CR4
Running is vitally important to me
3.23 (1.10)
0.59
0.79
−0.23
−0.51
CR5
Life is so much richer as a result of running
3.41 (1.07)
0.56
0.80
−0.47
−0.26
CR6
Running is pleasant
4.09 (0.75)
0.56
0.80
−0.80
1.39
CR7
I dread the thought of running (R)
4.72 (0.68)
0.30
0.82
−3.01
10.20
CR8
I would arrange or change my schedule to
meet my need to run
3.29 (1.19)
0.46
0.81
−0.37
−0.68
CR9
I have to force myself to run (R)
3.66 (1.12)
0.42
0.81
−0.46
−0.63
CR10
To miss a day’s run is a sheer relief (R)
4.41 (0.83)
0.39
0.81
−1.45
1.89
CR11
Running is the high point of my day
2.93 (1.15)
0.47
0.81
−0.09
−0.68
1 SD = Standard Deviation; 2 Corrected item–total correlation; (R) = Reverse scoring; 1 = Totally disagree,
2 = Disagree, 3 = Neither agree nor disagree, 4 = Agree, and 5 = Totally agree.
After checking the properties of the items, an EFA was carried out for the 11 items related to
the running commitment of the interviewed participants. The parallel analysis procedure suggested
grouping the indicators into two factors. It was not necessary to eliminate any indicators because all
of them had factor loads above 0.40 and no cross-factor loads exceeded this value in the two factors.
Table 3 shows the grouping of the indicators into the two factors: enthusiasm for running (6 items) and
affliction from running (5 items).
Table 3. Rotated factor structure of the commitment-to-running scale of runners participating in
popular endurance races, communalities, eigenvalues, and explained variance.
Number
Items
F1
F2
Com. 1
Factor 1: Enthusiasm for running
CR1
I am looking forward to running
0.56
0.48
CR4
Running is vitally important to me
0.81
0.64
CR5
Life is so much richer as a result of running
0.78
0.59
CR6
Running is pleasant
0.40
0.37
CR8
I would arrange or change my schedule to meet my need to run
0.62
0.37
CR11
Running is the high point of my day
0.65
0.41
Factor 2: Affliction from running
CR2
Running is drudgery
0.57
0.43
CR3
I do not enjoy running
0.67
0.48
CR7
I dread the thought of running
0.67
0.40
CR9
I have to force myself to run
0.56
0.34
CR10
To miss a day’s run is a sheer relief
0.67
0.43
G-H Index
0.85
0.80
Eigenvalue
4.01
2.01
Variance Explained (%)
36.51
18.28
Items
6
5
1 Com. = Communality.
Int. J. Environ. Res. Public Health 2020, 17, 925
7 of 19
To check the fit of the model, the RMSR and gamma index or GFI coefficients were analysed,
which showed values within the recommended cutoff points: RMSR = 0.03 (<0.05) and GFI = 0.99
(>0.95). In addition, the generalized G-H index showed values higher than 0.80 in the two factors
detected by the EFA (0.85 for the running enthusiasm factor and 0.80 for the running affliction factor),
indicating the possibility of good replicability of the dimensions in other studies [46]. The variance
explained by the 11 items grouped in the two factors was 54.78%.
On the other hand, a CFA was derived from the two-factor solution of the EFA and from the
unidimensional proposal of the scale presented by the authors of the original scale [21]. The CFA
considering the scale as unidimensional did not show a good fit, as seen in the goodness-of-fit indices
(χ2 = 2005.81 (df = 44); p < 0.01; SRMR = 0.125; RMSEA = 0.122 (confidence interval (CI) = 0.116–0.128);
CFI = 0.73; IFI = 0.73)).
However, the factorial solution derived from the EFA by grouping the indicators into two factors
did show optimal model fit values. The value of χ2 was significant (χ2 = 653.95; df = 43; p < 0.05),
although the relationship between the value of χ2 and the degrees of freedom (normed chi-square)
was quite high since values below five are considered acceptable [51]. This is because the value of this
statistic is very sensitive to the size of the sample, which could erroneously indicate a poor adjustment
of the data [48,58]. For this reason, it is recommended that other goodness-of-fit indices be used [48]:
SRMR, RMSEA, CFI, and IFI. These indicators showed a good fit: (SRMR = 0.077; RMSEA = 0.076
(IC = 0.070–0.082); CFI = 0.90; IFI = 0.90)).
To analyse reliability, the Cronbach’s alpha, CR, and AVE measurements were observed (see Table 4).
The Cronbach’s alpha values were higher than the 0.70 recommended by the literature [54]. This criterion
was also fulfilled for the CR values [55], with values of 0.83 for the running enthusiasm factor and
0.77 for the running affliction factor. Finally, for the AVE values, it was found that the two factors did
not have values higher than the 0.50 recommended by the literature [59]. According to Hatcher [60],
when the reliability of the construct is acceptable, a marginally low AVE value can be accepted
(see Table 4). Therefore, the decision was made to retain these two factors without combining them
into a single factor since the unidimensional solution did not offer a good fit.
Table 4. Factorial loads, Cronbach’s alpha, composite reliability, and average variance extracted values
from commitment to running scale indicators.
Number
Items
λ
α
CR 1
AVE 2
Factor 1: Enthusiasm for running
0.82
0.83
0.45
CR1
I am looking forward to running
0.68
CR4
Running is vitally important to me
0.79
CR5
Life is so much richer as a result of running
0.76
CR6
Running is pleasant
0.54
CR8
I would arrange or change my schedule to meet my need to run
0.61
CR11
Running is the high point of my day
0.62
Factor 2: Affliction from running
0.76
0.77
0.40
CR2
Running is drudgery
0.66
CR3
I do not enjoy running
0.69
CR7
I dread the thought of running
0.57
CR9
I have to force myself to run
0.6
CR10
To miss a day’s run is a sheer relief
0.64
1 CR = Composite Reliability; 2 AVE = Average Variance Extracted.
To analyse convergent validity, it was found that the values of the t-tests associated with the
factorial loads of the items were higher than 1.96 (p < 0.05), ranging from 15.06 to 27.57. Regarding
discriminant validity, on the one hand, the correlation between the two factors was less than 0.85
(r = 0.35; p < 0.01). On the other hand, it was found that the square root of the AVE value was higher
than the correlations between pairs of factors. This criterion was also complied because the values of
the square root of the AVE were √AVE = 0.67 for factor 1 and √AVE = 0.63 for factor 2.
Int. J. Environ. Res. Public Health 2020, 17, 925
8 of 19
3.3. Identification and Description of Clusters
A cluster analysis was conducted to identify groups with different characteristics according to
their commitment to running among the interviewed runners participating in urban short-distance
races. Following the procedure recommended by Hair et al. [54], a hierarchical cluster analysis was first
performed using Ward’s method of observing the increase in the agglomeration coefficients between
clusters two and three, three and four, and four and five. After observing these coefficients, the solutions
of two, three, four, and five groups were used to apply the second analysis of the k-median clusters
by using the initial centres from the hierarchical cluster analysis. It was considered appropriate to
contrast all the solutions because there are not many previous studies that identify clusters of amateur
runners according to their commitment to running, as mentioned in the theoretical framework.
This study used the solution of three clusters because it identified three groups of urban runners
with different levels of commitment to running, helping to interpret and identify their sociodemographic
profiles and sports habits. It is important to note that the choice of an ideal cluster solution depends on
the theoretical foundations, common sense, and practical judgement of the researcher [54]. Table 5
shows the mean values (centroids) for each of the 11 variables of the commitment-to-running scale
introduced in the analysis and the results of the ANOVA test carried out to confirm the significance
of the differences between groups (F statistic and significance level). In this table, it can be seen that
all the mean scores of the identified groups present statistically significant differences (p < 0.001).
The variables that distinguish the clusters most correspond to the running enthusiasm factor: “Running
is vitally important to me” (F = 896.45) and “Life is so much richer as a result of running” (F = 768.53).
The item that differentiates the clusters least is “I dread the thought of running” (F = 47.61).
Table 5. Average scores for each variable in the three clusters (obtained through the k-averages method).
Number
Items
1 = Highly
Committed
(n = 650) (SD) 1
2 = Moderately
Committed
(n = 749) (SD) 1
3 = Lowly
Committed
(n = 407) (SD) 1
F
p-Value
Factor 1: Enthusiasm for running
4.15 (0.46)
3.41 (0.35)
2.43 (0.41)
2263.89
<0.001 *
CR1
I am looking forward to running
4.42 (0.64)
3.73 (0.73)
2.89 (0.87)
543.02
<0.001 *
CR4
Running is vitally important to me
4.08 (0.78)
3.16 (0.77)
2.00 (0.80)
896.45
<0.001 *
CR5
Life is so much richer as a result
of running
4.12 (0.78)
3.47 (0.74)
2.18 (0.87)
768.53
<0.001 *
CR6
Running is pleasant
4.51 (0.58)
4.01 (0.59)
3.58 (0.86)
263.28
<0.001 *
CR8
I would arrange or change my
schedule to meet my need to run
4.09 (0.91)
3.26 (0.93)
2.09 (0.97)
577.43
<0.001 *
CR11
Running is the high point of my day
3.71 (0.96)
2.83 (0.90)
1.87 (0.88)
516.87
<0.001 *
Factor 2: Affliction from running
1.32 (0.32)
2.04 (0.65)
2.16 (0.68)
388.42
<0.001 *
CR2
Running is drudgery
1.46 (0.75)
2.39 (1.01)
2.91 (1.14)
320.47
<0.001 *
CR3
I do not enjoy running
1.16 (0.54)
1.89 (0.96)
2.04 (0.93)
192.12
<0.001 *
CR7
I dread the thought of running
1.06 (0.33)
1.46 (0.85)
1.31 (0.65)
66.28
<0.001 *
CR9
I have to force myself to run
1.73 (0.89)
2.65 (1.03)
2.77 (1.16)
190.17
<0.001 *
CR10
To miss a day’s run is a sheer relief
1.20 (0.52)
1.82 (0.89)
1.79 (1.90)
126.35
<0.001 *
1 SD = Standard Deviation; * Statistically significant mean differences p < 0.001.
Cluster 1 was labelled “highly committed” (n = 650; 35.99%) because all indicators related to
Factor 1 about running enthusiasm present averages close to or above the value of four on the Likert
scale, which indicates a high degree of agreement with these statements. This group of runners presents
a high commitment to this physical activity because they consider running to be of vital importance
for their lives (M = 4.08), they would reorganize or change their schedule to satisfy the need to run
(M = 4.09), and they consider their lives to be much richer because they practice this physical activity
(M = 4.12). In addition, they show a high degree of agreement about the desire to run (M = 4.42) and
enjoy the experience of running (M = 4.51). However, they disagree with the aspects related to factor 2
about a sense of affliction from running. For example, they tend to disagree with items related to a
sense of drudgery over the activity (M = 1.46), an absence of enjoyment when running (M = 1.16),
Int. J. Environ. Res. Public Health 2020, 17, 925
9 of 19
a feeling of dread over the practice of running continuously (M = 1.06), and a sense of relief when not
running for one day (M = 1.20).
Cluster 2 was identified as “moderately committed” (n = 749; 41.47%) and includes a higher
proportion of the interviewed runners; this group of runners is characterized by presenting average
scores with a tendency towards agreement on some indicators of the dimension of enthusiasm for
running but with a more moderate tendency than in cluster 1. Thus, the runners in this group consider
the practice of running to be a pleasant activity (M = 4.01) and are willing to practise it (M = 3.73).
They also show a positive tendency in considering that running enriches their lives (M = 3.47). However,
they do not consider the practice of this physical activity to be essential to their lives (M = 3.16) or
to be the high point of their day (M = 2.83). On the other hand, this group also disagrees with most
of the negative aspects (affliction for running) associated with a commitment to running. However,
the indicator related to the need to force themselves to run shows a score close to the value of three on
the scale (M = 2.65), which would indicate some neutrality in the assessments of this group.
Cluster 3 was labelled “slightly committed” (n = 407; 22.54%) because they show a tendency
to disagree with many aspects related to enthusiasm for running. In this regard, they disagree with
the items stating that the practice of this physical activity is of vital importance for them (M = 2.00),
that their life is much richer because they practice running (M = 2.18), that the practice of this activity is
the high point of their day (M = 1.87), and that they reorganize their schedules to satisfy the need to run
(M = 2.09). This group only shows a tendency to agree that running is a pleasant activity (M = 3.58).
In the same vein as the other groups, these runners show disagreement with the aspects related to a sense
of affliction from running, although there is a tendency towards neutrality in the aspects referring to a sense
of drudgery over the practice of this activity (M = 2.91) and a need to force themselves to run (M = 2.77).
3.4. Profile of the Groups
Table 6 describes the profile of the runners that make up each group using other independent
variables (sociodemographic and sports habits variables) that allow us to ensure the predictive validity
of the groups, the percentages for each sociodemographic, and sports habits variable according to
the cluster.
Statistically significant differences were observed among the groups in the case of the
sociodemographic variables related to age (F(2, 1803) = 5.73, p < 0.05) and level of studies (χ2(6) = 18.94,
p < 0.01), although the size of the effects (contingency coefficients) presented reduced values (see Table 6).
It was also found that there were statistically significant differences in the mean scores among the
groups on the indicators related to running addiction: “Some days, even if I do not feel like running,
I do it anyway” (F(2, 1803) = 104.78, p ≤ 0.001), “I feel like I need to run at least once every day”
(F(2, 1803) = 150.05, p ≤ 0.001), and “I’ve stopped running for at least a week for some other reason
than injury” (F(2, 1803) = 46.84, p ≤ 0.001). Additionally, on the indicator related to future intentions
to recommend participation in popular running races to others, statistically significant differences in
mean group scores were observed (F(2.1803) = 4.72, p < 0.01).
The highly committed group is characterized by an average age of 39.51 (SD = 9.33), with a
majority percentage of men (76.62%), which is higher than the other groups. It is verified that there is a
greater degree of agreement in this group with indicators related to running addiction, as they show a
tendency towards agreement on, for instance, the item stating that, some days, even if they do not feel
like running, they go running anyway (M = 3.71), with a significantly higher average compared to
the other two groups. Significantly higher differences, albeit with a more neutral trend in the average
scores, are also observed in this group for the indicator related to the need to run at least once every
day (M = 3.04) and lower differences for the indicator related to having stopped running for at least
one week for a reason that was not injury related (M = 3.04). Finally, the runners in this group strongly
agree with the items on their future intentions regarding participation, recommendation, and positive
comments about running popular races, with slightly higher scores on all these indicators than those
of the other two groups.
Int. J. Environ. Res. Public Health 2020, 17, 925
10 of 19
Table 6. Characteristics of the different groups (clusters).
Variable
Response Option
1 = Highly Committed
(n = 650)
2 = Moderately Committed
(n = 749)
3 = Lowly Committed
(n = 407)
Age *
F(2, 1803) = 5.73, p = 0.03
39.51
(SD 1 = 9.33)
40.13
(SD 1 = 9.26)
38.22
(SD 1 = 8.79)
Gender
χ2(2) = 2.71, p = 0.26
Male
76.62%
73.03%
73.22%
Female
23.38%
26.97%
26.78%
Occupation
χ2(6) = 2.13, p = 0.91
Employee
88.15%
88.79%
87.96%
Unemployed
6.92%
5.47%
5.90%
Student
1.08%
1.34%
1.23%
Other (retired, pensioner, domestic
tasks, etc.)
3.85%
4.41%
4.91%
Level of studies **
χ2(6) =18.94, p = 0.004
C 2 = 0.10
Primary
8.46% (3)
5.47%
3.44%
Secondary
7.23%
6.14%
4.67%
Baccalaureate/Professional training
29.85%
27.64%
27.76%
University
54.46%
60.75%
64.13% (1)
Income level
χ2(10) = 13.71, p = 0.19
Less than 12,000 euros
29.38%
25.90%
25.06%
12,001–18,000 euros per year
25.85%
22.56%
24.82%
18,001–24,000 euros per year
15.54%
19.49%
20.15%
24,001–30,000 euros per year
14.00%
14.69%
13.51%
30,001–36,000 euros per year
7.23%
6.14%
7.86%
More than 36,001 euros per year
8.00%
11.21%
8.60%
How often you run during the week? ***
χ2(6) = 290.78, p ≤ 0.001
C 2 = 0.37
Five or more times a week
14.31% (2) (3)
4.01%
1.97%
Three to five times a week
62.46% (2) (3)
51.27% (3)
29.73%
Once or twice a week
21.69%
40.32% (1)
51.84% (1) (2)
Less frequently
1.54%
4.41% (1)
16.46% (1) (2)
Int. J. Environ. Res. Public Health 2020, 17, 925
11 of 19
Table 6. Cont.
Variable
Response Option
1 = Highly Committed
(n = 650)
2 = Moderately Committed
(n = 749)
3 = Lowly Committed
(n = 407)
Preferred distance in popular races ***
χ2(10) =119.49, p ≤ 0.001
C 2 = 0.25
Less than 7.5 km
25.23%
36.85% (1)
52.58% (1) (2)
Between 7.5 km and 10 km
28.15%
30.31%
27.27%
Between 10 km and 15 km
27.69% (3)
22.83% (3)
13.51%
Between 15 km and 20 km
10.15% (2) (3)
6.14%
4.91%
Between 20 km and 30 km
4.62% (3)
2.54%
1.72%
More than 30 km
4.15% (2)
1.34%
0.00%
How do you usually run?
χ2(2) = 0.34, p = 0.84
Alone
62.46%
61.15%
60.93%
Accompanied
37.54%
38.85%
39.07%
Level you consider you are as a runner ***
χ2(4) = 133.61, p ≤ 0.001
C 2 = 0.26
High level
8.31% (2) (3)
2.67%
1.47%
Intermediate
64.92% (2) (3)
53.14% (3)
39.07%
Low level
26.77%
44.19% (1)
59.46% (2) (3)
Are you a member of a sports club?
χ2(2) = 53.84, p ≤ 0.001
C 2 = 0.17
Yes
48.77% (2) (3)
40.05% (3)
26.04%
No
51.23%
59.95% (1)
73.96% (2) (3)
Are you sports federated?
χ2(2) = 10.47, p = 0.005
C 2 = 0.08
Yes
6.92% (2) (3)
3.60%
3.44%
No
93.08%
96.40% (1)
96.56% (1)
Distance usually run weekly (kilometres) ***
F(2, 1803) = 106.76, p ≤ 0.001
35.11
(SD 1 = 18.54)
26.16
(SD 1 = 14.65)
20.58
(SD 1 = 16.01)
Years running **
F(2, 1803) = 7.20, p = 0.01
8.04
(SD 1 = 8.95)
7.32
(SD 1 = 8.20)
6.08
(SD 1 = 6.68)
Participation in half marathons ***
F(2, 1803) = 20.70, p ≤ 0.001
7.99
(SD 1 = 22.17)
4.05
(SD 1 = 10.06)
2.30
(SD 1 = 6.33)
Int. J. Environ. Res. Public Health 2020, 17, 925
12 of 19
Table 6. Cont.
Variable
Response Option
1 = Highly Committed
(n = 650)
2 = Moderately Committed
(n = 749)
3 = Lowly Committed
(n = 407)
Participation in marathons ***
F(2, 1803) = 17.84, p ≤ 0.001
1.38
(SD 1 = 3.77)
0.74
(SD 1 = 2.74)
0.32
(SD 1 = 1.26)
Some days, even if I do not feel like running, I do it anyway ***
F(2, 1803) = 104.78, p ≤ 0.001
3.71
(SD 1 = 1.02)
3.35
(SD 1 = 0.99)
2.76
(SD 1 = 1.13)
I feel like I need to run at least once every day ***
F(2, 1803) = 150.05, p ≤ 0.001
3.04
(SD 1 = 1.20)
2.49
(SD 1 = 1.07)
1.84
(SD 1 = 0.97)
I have stopped running for at least a week for another reason that was not an injury ***
F(2, 1803) = 46.84, p ≤ 0.001
3.04
(SD 1 = 1.44)
3.58
(SD 1 = 1.17)
3.75
(SD 1 = 1.31)
Future intentions regarding participation in
urban popular races
I am willing to continue
participating in popular urban races
F(2, 1803) = 0.85, p = 0.45
4.42
(SD 1 = 1.03)
4.38
(SD 1 = 0.99)
4.35
(SD 1 = 0.91)
I will recommend participation in
popular urban races to others **
F(2, 1803) = 4.72, p = 0.009
4.48
(SD 1 = 3.77)
4.37
(SD 1 = 3.77)
4.32
(SD 1 = 3.77)
I will speak well of popular urban
races to others
F(2, 1803) = 2.04, p = 0.13
4.46
(SD 1 = 0.89)
4.39
(SD 1 = 0.86)
4.37
(SD 1 = 0.80)
1 SD = Standard Deviation; 2 C = Contingency Coefficient; indications of statistically significant relationship or statistically significant mean differences: * p < 0.05; ** p ≤ 0 0.01; *** p ≤ 0.001;
(1) (2) (3) results are based on bilateral tests with a level of significance 0.05. The results table shows for each significant pair the key of the group of runners with the proportion of the
smallest column below the group of runners with the largest proportion of the column.
Int. J. Environ. Res. Public Health 2020, 17, 925
13 of 19
The moderately committed group is characterized by a higher average age of 40.13 (SD = 9.26),
significantly higher than the slightly committed group, and a majority percentage of men (73.03%).
These runners tend towards agreement on the indicator related to having stopped running for at least
one week for a reason other than an injury (M = 3.58), with a significantly higher average compared to
the higher commitment group level. In the same way as the other groups, they strongly agree with the
items on their future intentions regarding participation, recommendation, and positive comments on
popular races.
The third, slightly committed group is characterized by the lowest average age with 38.22
(SD = 8.79) and a majority percentage of men (73.22%). They tend towards agreement on the indicator
related to having stopped running for at least one week for a reason that was not injury related
(M = 3.75) with a significantly higher average compared to the group with greater commitment to the
race. Following the trend observed in the other groups, they strongly agree on the item on their future
intentions regarding participation, recommendation, and positive comments on popular races.
4. Discussion
This study analyses the running commitment of runners participating in urban popular races
from the commitment-to-running scale validated with Spanish marathon runners. The analysis of
the psychometric properties of the scale showed that the grouping of the indicators into two factors
presented a better adjustment to the data collected in this study, identifying two dimensions related
to positive aspects (enthusiasm for running) and negative aspects (affliction from running) of the
psychological construct of commitment to running. In previous works on the validation of the
running-commitment scale, the possibility of grouping the indicators under two factors was assessed
but was rejected due to the lack of interpretability and the high degree of uncertainty of the second
factor identified [25]. For this reason, it was decided to test the fit of the scale on a single factor, although
the various goodness-of-fit indices showed values distant from those recommended in the literature
as acceptable.
This study identified three groups of participants in urban popular races with different
commitments to running: “highly committed”, “moderately committed”, and “slightly committed”.
The denomination of these groups was made from the interpretation of the average scores of the
variables on the commitment-to-running scale that allowed us to identify three groups with clearly
different levels of commitment towards the practice of this physical activity.
The sociodemographic variables that contributed to significantly differentiating the identified
groups were age and educational level. However, previous studies that have analysed the commitment
to running without segmenting respondents into groups did not find significant differences in the
items of the running-commitment scale according to age [15,25]. In other studies identifying groups
according to their motivations for running, age was found to be a variable that differentiated the
identified clusters [27,29,30]. The results varied according to gender, with some studies showing
differences in the commitment to running according to this variable [10,19,21,25,37], while other
studies, in the same vein as the results of this study, detected no significant differences by gender [15].
It is important to note that the profile of the runners in these studies was runners of long-distance races
such as half-marathons and marathons, different from the sample analysed in this study. Regarding the
level of studies, in the work of Zarauz and Ruiz-Juan [19] on marathon runners, it was observed that a
lower level of studies significantly predicted the commitment to running. In this study, it was found
that there were significant disproportions of runners with a higher level of studies among runners with
a lower level of commitment to running. In any case, more studies on amateur short-distance runners
are needed to determine which sociodemographic variables contribute to differentiating subgroups of
runners according to their commitment to running.
On the other hand, variables related to sports habits (training frequency, preferred distance,
years of running, number of kilometres run weekly, level at which runners considered themselves to be,
participation in long-distance races, and membership of clubs and sports federations) also contributed
Int. J. Environ. Res. Public Health 2020, 17, 925
14 of 19
to significantly differentiating the groups identified, confirming the need to consider runners from a
heterogeneous point of view, as observed in other studies [27,29,30,61]. In previous studies, it was
found that weekly training frequency, number of years of running, and number of kilometres run were
positively related to commitment to running [10,21–23,25]. Similarly, a preference for participating in
long-distance running was associated with higher scores on commitment to running [25,37,62].
From the point of view of the variables related to the running-addiction scale, it is worth noting
that statistically significant differences were found between the groups, proving the existence of a
higher level of involvement in the activity in runners with a high commitment to running. In the
theoretical framework, it has been emphasized that previous studies have proved that there is a high
correlation between higher values of commitment to running and addiction to this physical activity,
although both constructs may be influenced by different variables. In a previous study, it was observed
that, in half-marathon runners, the variable that seemed to have most importance with respect to why a
half-marathon runner goes from being healthily committed to his sport practice to being pathologically
addicted to it is the number of kilometres he runs each week: if it is low, a healthy commitment
increases, while if it is high, in men, negative addiction increases [10]. In the present work, although
this finding was not empirically proven, it is observed that the runners in the highly committed group
train significantly more kilometres per week than the other groups and, in addition, they show a
significantly higher tendency to agree on statements related to running addiction, such as the item
stating that, some days, even if they do not feel like running, they practice the activity anyway. In this
sense, the points made by Ruiz-Juan, Zarauz, and Flores-Allende [10] could be confirmed but in
runners who participated in urban popular races of reduced distance. In any case, more empirical
evidence is needed to confirm the hypothesized relationship between the number of kilometres run
and running addiction in amateur short-distance runners.
On the other hand, analysis of the groups’ profile verified that the group denominated as highly
committed presents a tendency to agree on most of the indicators related to the factor of enthusiasm to
run, indicating a clear commitment through the practice of this activity. In contrast, on the aspects
related to the affliction factor from running, they show the inverse tendency, which also corroborates the
high commitment of these runners to this physical activity. This is the group with the highest proportion
of people with the lowest level of studies, in contrast to the group with the lowest commitment to
running, indicating a similar trend to that observed by Zarauz and Ruiz-Juan [19] on the influence of
this variable on the commitment to running. This is a group with sporting habits defined by the constant
and regular practice of this physical activity: they tend to run alone, have a training frequency of three
or more times per week, display a preference for distances in popular races between 7.5 km and 15 km,
are members of sports clubs, have the highest average number of kilometres run weekly, have the
greatest number of years practising running, and have the highest average number of long-distance
races run. This type of runner has been identified by Leedy [37] as “committed runners”. Additionally,
Carmack and Martens [21] related the high commitment to running to running addiction; these runners
were later classed by Pargman [20] as addict-dependent. However, this concept should be considered
in a positive sense, that is, as defined by this author, from the perspective that they are participants
who show pleasure, enjoyment, and joy in the practice of this sport, as identified through the average
scores observed on the indicators of the running enthusiasm factor.
With respect to the group known as moderately committed, a more neutral and moderate attitude
is observed in the indicators referring to the enthusiasm for running, highlighting aspects such as the
desire to run or pleasure in running while showing a neutral evaluation on items such as the statement
that the practice of this physical activity is the high point of the day or of vital importance for their lives.
In the case of items on the affliction from running factor, they present a tendency towards disagreement,
albeit less pronounced than that of the group with greater commitment. Their sociodemographic profile
differs in that they are older than the rest of the groups and have a higher level of education than the
most committed group. The sports habits of this group present intermediate levels between the groups
with lesser commitment and greater commitment, with the highlights being that they tend to run alone,
Int. J. Environ. Res. Public Health 2020, 17, 925
15 of 19
their training frequency is three to five times per week, they display a preference for short-distance
races (less than 10 km), they run a shorter distance each weekly than the highly committed group,
and they have fewer years practising this physical activity. On the other hand, regarding the indicators
on the running addiction scale, they do not agree that they need to run at least once every day and
show a slight tendency towards agreement on the item regarding having stopped running for reasons
other than an injury. This runner profile is characterized by a recreational type of orientation and less
commitment to the practice of running despite considering it an enjoyable and positive experience
for their lives. Previous work on motivation and commitment to running has identified profiles of
recreational runners [37], relational runners [5], socializing hedonists [27], and social competitors [29],
who may have similarities in terms of their sporting habits and sociodemographic profile. This group
could be identified with a healthier profile from a psychological point of view, since they do not
show symptoms of possible running addiction, as could be deduced from the scores of the indicators
extracted from the running addiction scale.
Finally, the group dubbed slightly committed is the group with the lowest level of commitment
to running because it is the group that disagrees most with the indicators related to enthusiasm for
running (running is of vital importance to their life; the practice of enriches their life; they reorganize
schedules in their daily life; and running is the highlight of the day). Unlike for the other two groups,
the only indicator with a positive trend for this group is the fact that they consider running a pleasant
activity. In the same way, the other groups disagree with the aspects related to affliction from running.
However, the slightly committed group has a certain tendency towards low motivation and considers
the activity to be drudgery, as the tendency towards neutrality observed in some scores of the running
affliction factor points out. From the point of view of sociodemographic characteristics, this is the
youngest group and has the highest level of studies. They are runners who train less frequently (once or
twice a week), prefer short-distance races (less than 7.5 km), are less likely to have affiliations with
sports clubs, run fewer kilometres weekly, and have fewer years practising this sport. There is a lower
degree of dependence on the activity, as seen from the scores on the running-addiction indicators,
giving a secondary role to the practice of running.
4.1. Practical Implications
The objective of this research was to analyze the commitment of runners among urban runners
and to segment and classify the runners. Research in this area and on commitment to running is
limited, especially among urban short-distance runners. The groups of runners were identified and
described sociodemographic profile and sports habits: highly committed, moderately committed,
and slightly committed. The conclusions of the work can be useful for the administration, management,
and organization of local sports in terms of knowing the profiles of participants in popular races
and of developing measures to improve the quality of life of the participants. In this regard, a high
commitment to running can imply risks for health and can devolve into addictive behaviours that
must be prevented to ensure that a runner is healthily committed to their sport practice rather than
pathologically addicted to it. Running is an activity that has experienced an important boom in
Spanish cities such as Valencia, requiring the implementation of sports policies and strategic plans that
respond to the social demands of this collective. For this reason, it is advisable to offer information
and public services (sports, health, psychology, and education) to runners through the organizations
and communication initiatives associated with this popular activity. Additionally, the identification
of different profiles, characteristics, and habits can benefit sport entities and public organizations to
encourage sport activity.
4.2. Limitations and Future Lines of Research
This work presents some limitations that should be taken into account when generalizing and
extrapolating the results to other populations; for example, the type of sampling used does not allow
for generalization of the results to the set of urban amateur runners. In carrying out f more studies
Int. J. Environ. Res. Public Health 2020, 17, 925
16 of 19
of different populations of amateur runners, efforts should be made to identify subgroups of the
population of urban runners in different countries and regions with the aim of elaborating typologies of
runners at the international level as well as identifying possible differences according to sociocultural
and socioeconomic characteristics. Similarly, the relationship between commitment to running and
possible addictions to this physical activity in runners who are starting out in the sport or who
participate in shorter-distance races should be explored in greater depth with the aim of preventing
future behaviour that could be dangerous to health.
5. Conclusions
The aim of this paper is to analyse the commitment to running among urban runners by identifying
groups with a greater or lesser commitment to this sport. Three groups of urban amateur runners with
different levels of commitment to running were identified: highly committed, moderately committed,
and slightly committed. Also, the study identifies two factors within the psychological construct of
commitment to running: enthusiasm for running and affliction from running.
Sociodemographic variables related to age and educational level contribute to differentiating the
groups of amateur runners according to their commitment to running. It is also found that most of the
variables related to sports habits contribute to differentiating the groups: the frequency with which
they go out to run, their preferred distance in popular races, the number of years they have practised
running, the number of kilometres run weekly, the level they consider themselves to be at as runners,
participation in long-distance races, and membership in clubs and sports federations.
Highly committed runners positively value the aspects related to enthusiasm for running and
show the reverse trend in the aspects related to the affliction from running.
Moderately committed runners are characterized by a moderate attitude towards running
enthusiasm and a tendency to disagree on aspects related to running affliction.
Slightly committed runners show a low commitment to the practice of running, with a tendency
towards disagreement on the indicators related to enthusiasm for running except the one related to
considering it a pleasant activity. They disagree with the aspects related to the affliction from running,
although with a tendency to neutrality when considering running to be a drudgery and an activity for
which they have low motivation.
Author Contributions: Conceptualization, D.P.-C. and M.H.G.-S.; methodology, D.P.-C. and M.A.D.S.; formal
analysis, M.H.G.-S. and M.A.D.S.; writing—original draft preparation, D.P.-C., M.H.G.-S. and M.A.D.S.;
writing—review and editing, D.P.-C. and M.A.D.S. All authors have read and agree to the published version of
the manuscript.
Funding: This research received no external funding.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Scheerder, J.; Breedveld, K.; Borgers, J. Who Is Doing a Run with the Running Boom? In Running across
Europe: The Rise and Size of One of the Largest Sport Markets; Scheerder, J., Breedveld, K., Borgers, J., Eds.;
Palgrave Macmillan: London, UK, 2015; pp. 1–27.
2.
Llopis-Goig, L.; Vilanova, A. A Sociological Analysis of the Evolution and Characteristics of Running.
In Running across Europe: The Rise and Size of One of the Largest Sport Markets; Scheerder, J., Breedveld, K.,
Borgers, J., Eds.; Palgrave Macmillan: London, UK, 2015; pp. 225–239.
3.
Wicker, P.; Breuer, C. Understanding the Importance of Organizational Resources to Explain Organizational
Problems: Evidence from Nonprofit Sport Clubs in Germany. Volunt. Int. J. Volunt. Nonprofit Organ. 2013,
24, 461–484. [CrossRef]
4.
Guaita, J.M.; Roig-Tierno, N.; Mas-Tur, A. Impact of non-professional runner clubs: An AHP approach to the
economic, environmental and sociocultural factors. J. Sports Econ. Manag. 2018, 8, 63–72.
5.
Llopis Goig, R.; Llopis Goig, D. Una tipología sociocultural de los corredores populares en España. Apunt. Educ.
Física Deportes 2012, 108, 9–16. [CrossRef]
Int. J. Environ. Res. Public Health 2020, 17, 925
17 of 19
6.
Scanlan, T.K.; Carpenter, P.J.; Simons, J.P.; Schmidt, G.W.; Keeler, B. An Introduction to the Sport Commitment
Model. J. Sport Exerc. Psychol. 1993, 15, 1–15. [CrossRef]
7.
Scanlan, T.K.; Carpenter, P.J.; Simons, J.P.; Schmidt, G.W.; Keeler, B. The Sport Commitment Model:
Measurement Development for the Youth-Sport Domain. J. Sport Exerc. Psychol. 1993, 15, 16–38. [CrossRef]
8.
Scanlan, T.K.; Russell, D.G.; Beals, K.P.; Scanlan, L.A. Project on Elite Athlete Commitment (PEAK): II. A Direct
Test and Expansion of the Sport Commitment Model with Elite Amateur Sportsmen. J. Sport Exerc. Psychol.
2003, 25, 377–401. [CrossRef]
9.
Glasser, W. Positive Addiction; Harper & Row: Oxford, UK, 1976.
10.
Ruiz-Juan, F.; Zarauz, A.; Flores-Allende, G. Dependencia al entrenamiento y competición en corredores de
fondo en ruta. Cult. Cienc. Deporte 2016, 11, 149–155.
11.
Meeusen, R.; Duclos, M.; Gleeson, M.; Rietjens, G.; Steinacker, J.; Urhausen, A. Prevention, diagnosis and
treatment of the Overtraining Syndrome. Eur. J. Sport Sci. 2006, 6, 1–14. [CrossRef]
12.
Lu, F.J.-H.; Eva, Y.-W.; Wang, J.-M.; Chang, J.-N.; Huang, M.-Y.; Wang, C.-H. Exercisers’ Identities and
Exercise Dependence: The Mediating Effect of Exercise Commitment. Percept. Mot. Ski. 2012, 115, 618–631.
[CrossRef]
13.
Zarauz-Sancho, A.; Ruiz-Juan, F.; Flores-Allende, G.; Giráldez, V.A. Variables predictoras de la percepción
del éxito: Aspectos diferenciales en corredores de ruta/Predictor Variables of the Perception of Success:
Differential Aspects in Route Runners. Rev. Int. Med. Cienc. Act. Física Deporte 2016, 16, 601–615. [CrossRef]
14.
Karr, T.M.; Zunker, C.; Thompson, R.A.; Sherman, R.T.; Erickson, A.; Cao, L.; Crosby, R.D.; Mitchell, J.E.
Moderators of the association between exercise identity and obligatory exercise among participants of an
athletic event. Body Image 2013, 10, 70–77. [CrossRef] [PubMed]
15.
Chapman, C.L.; De Castro, J.M. Running addiction: Measurement and associated psychological characteristics.
J. Sports Med. Phys. Fit. 1990, 30, 283–290.
16.
Horton, R.S.; Mack, D.E. Athletic identity in marathon runners: Functional focus or dysfunctional
commitment? J. Sport Behav. Mob. 2000, 23, 101–119.
17.
Masters, K.S.; Ogles, B.M.; Jolton, J.A. The Development of an Instrument to Measure Motivation for
Marathon Running: The Motivations of Marathoners Scales (MOMS). Res. Q. Exerc. Sport 1993, 64, 134–143.
[CrossRef] [PubMed]
18.
Zarauz, A.; Ruiz-Juan, F. Super-adherencia del maratoniano: Variables predictoras y diferencias de género.
Univ. Psychol. 2012, 11, 895–907.
19.
Zarauz, A.; Ruiz-Juan, D.F. Compromiso y adicción negtiva al entrenamiento y competición de los
maratonianos. Rev. Int. Med. Cienc. Act. Física Deporte 2011, 11, 817–834.
20.
Pargman, D. The way of the runner: An examination of motives for running. In Psychology in Sports: Methods
and Applications; Suinn, R.M., Ed.; Burguess: Minneapolis, MN, USA, 1980; pp. 90–98.
21.
Carmack, M.A.; Martens, R. Measuring Commitment to Running: A Survey of Runners’ Attitudes and
Mental States. J. Sport Exerc. Psychol. 1979, 1, 25–42. [CrossRef]
22.
Joseph, P.; Robbins, J.M. Worker of runner? The impact of commitment to running and work on self
identification. In Psychology of Running; Sacks, M.H., Sacks, M.L., Eds.; Human Kinnetics: Champaign,
IL, USA, 1981; pp. 131–146.
23.
Thornton, E.W.; Scott, S.E. Motivation in the committed runner: Correlations between self-report scales and
behaviour. Health Promot. Int. 1995, 10, 177–184. [CrossRef]
24.
Summers, J.J.; Machin, V.J.; Sargent, G.I. Psychosocial Factors Related to Marathon Running. J. Sport Exerc.
Psychol. 1983, 5, 314–331. [CrossRef]
25.
Ruiz-Juan, F.; Zarauz, A. Validation of the Spanish version of the Commitment to Running Scale (CR).
Estud. Psicol. 2011, 32, 195–207. [CrossRef]
26.
Llopis Goig, D.; Llopis Goig, R. Razones para participar en carreras de resistencia. Un estudio con corredores
aficionados. Cult. Cienc. Deporte 2006, 2, 33–40.
27.
Parra-Camacho, D.; González-Serrano, M.H.; González-García, R.J.; Calabuig, F. Sporting Habits of Urban
Runners: Classification According to Their Motivation. Int. J. Environ. Res. Public Health 2019, 16, 4990.
[CrossRef]
28.
Ruiz-Juan, F.; Zarauz, A. Variables que hacen adicto negativamente a correr al maratoniano español.
Retos Nuevas Tend. Educ. Física Deporte Recreación 2015, 21, 38–42.
Int. J. Environ. Res. Public Health 2020, 17, 925
18 of 19
29.
Rohm, A.J. A Mixed-Method Approach for Developing Market Segmentation Typologies in the Sports
Industry. Sport Mark. Q. 2006, 15, 29–39.
30.
Ogles, B.M.; Masters, K.S. A typology of marathon runners based on cluster analysis of motivations. J. Sport
Behav. 2003, 26, 69–85.
31.
Ruiz-Juan, F.; Zarauz, A. Análisis de la motivación en corredores de maratón españoles. Rev. Latinoam. Psicol.
2014, 46, 1–11. [CrossRef]
32.
Clough, P.; Shepherd, J.; Maughan, R. Motives for Participation in Recreational Running. J. Leis. Res. 1989,
21, 297–309. [CrossRef]
33.
Ogles, B.M.; Masters, K.S.; Richardson, S.A. Obligatory running and gender: An analysis of participative
motives and training habits. Int. J. Sport Psychol. 1995, 26, 233–248.
34.
Besomi, M.; Leppe, J.; Martínez, M.J.; Enríquez, M.I.; Mauri-Stecca, M.V.; Sizer, P.S. Running motivations
within different populations of Chilean urban runners. Eur. J. Physiother. 2017, 19, 8–11. [CrossRef]
35.
Nikolaidis, P.T.; Chalabaev, A.; Rosemann, T.; Knechtle, B. Motivation in the Athens Classic Marathon:
The Role of Sex, Age, and Performance Level in Greek Recreational Marathon Runners. Int. J. Environ. Res.
Public. Health 2019, 16, 2549. [CrossRef]
36.
Zarauz, A.; Ruiz-Juan, F. Psychometric properties of the Spanish version of the Running Addiction Scale
(RAS). Span. J. Psychol. 2011, 14, 967–976.
37.
Leedy, M.G. Commitment to distance running: Coping mechanism or addiction? J. Sport Behav. Mob. 2000,
23, 255–270.
38.
Andersen, J.J. The State of Running 2019 | RunRepeat. Available online: https://runrepeat.com/state-of-
running (accessed on 24 January 2020).
39.
Consejo Superior de Deportes. Encuesta de Hábitos Deportivos en España 2015; Ministerio de Educación,
Cultura y Deporte: Madrid, Spain, 2015.
40.
Zeithaml, V.A.; Berry, L.L.; Parasuraman, A. The Behavioral Consequences of Service Quality. J. Mark. 1996,
60, 31–46. [CrossRef]
41.
Lloret-Segura, S.; Ferreres-Traver, A.; Hernández-Baeza, A.; Tomás-Marco, I. El análisis factorial exploratorio
de los ítems: Una guía práctica, revisada y actualizada. An. Psicol. Ann. Psychol. 2014, 30, 1151–1169.
[CrossRef]
42.
Timmerman, M.E.; Lorenzo-Seva, U. Dimensionality assessment of ordered polytomous items with parallel
analysis. Psychol. Methods 2011, 16, 209–220. [CrossRef]
43.
Tanaka, J.S.; Huba, G.J. A general coefficient of determination for covariance structure models under arbitrary
GLS estimation. Br. J. Math. Stat. Psychol. 1989, 42, 233–239. [CrossRef]
44.
Harman, H.H. Análisis Factorial Moderno; Saltés: Madrid, Spain, 1980.
45.
Ruiz, M.Á.; Pardo, A.; San Martín, R. Modelos de ecuaciones estructurales. Papeles Psicólogo 2010, 31, 34–45.
46.
Ferrando, P.J.; Lorenzo-Seva, U. Assessing the Quality and Appropriateness of Factor Solutions and Factor
Score Estimates in Exploratory Item Factor Analysis. Educ. Psychol. Meas. 2018, 78, 762–780. [CrossRef]
47.
Chou, C.-P.; Bentler, P.M.; Satorra, A. Scaled test statistics and robust standard errors for non-normal data in
covariance structure analysis: A Monte Carlo study. Br. J. Math. Stat. Psychol. 1991, 44, 347–357. [CrossRef]
48.
Kline, R.B. Principles and Practice of Structural Equation Modeling, 4th ed.; Guilford Publications: New York,
NY, USA, 2015.
49.
Satorra, A.; Bentler, P.M. Corrections to test statistics and standard errors in covariance structure analysis.
In Latent Variables Analysis:
Applications for Developmental Research; von Eye, A., Clogg, C.C., Eds.;
Sage Publications, Inc.: Thousand Oaks, CA, USA, 1994; pp. 399–419.
50.
Wheaton, B.; Muthén, B.; Alwin, D.F.; Summers, G.F. Assessing Reliability and Stability in Panel Models.
Sociol. Methodol. 1977, 8, 84–136. [CrossRef]
51.
Hu, L.; Bentler, P.M. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria
versus new alternatives. Struct. Equ. Model. Multidiscip. J. 1999, 6, 1–55. [CrossRef]
52.
MacCallum, R.C.; Austin, J.T. Applications of Structural Equation Modeling in Psychological Research.
Annu. Rev. Psychol. 2000, 51, 201–226. [CrossRef] [PubMed]
53.
Browne, M.W.; Cudeck, R. Alternative ways of assessing model fit. In Testing Structural Equation Models;
Bollen, K.A., Long, J.S., Eds.; Sage: Newbury Park, CA, USA, 1993; pp. 136–162.
54.
Hair, J.F.; Black, W.C.; Babin, B.J.; Anderson, R.E. Multivariate Data Analysis, 7th ed.; Pearson: London,
UK, 2014.
Int. J. Environ. Res. Public Health 2020, 17, 925
19 of 19
55.
Fornell, C.; Larcker, D.F. Evaluating Structural Equation Models with Unobservable Variables and
Measurement Error. J. Mark. Res. 1981, 18, 39–50. [CrossRef]
56.
Nunnally, J.C. Teoría Psicométrica; McGraw-Hill: Madrid, Spain, 1995.
57.
Chou, C.-P.; Bentler, P.M. Estimates and tests in structural equation modeling. In Structural Equation Modeling:
Concepts, Issues, and Applications; Sage Publications, Inc.: Thousand Oaks, CA, USA, 1995; pp. 37–55.
58.
Schumacker, R.E.; Lomax, R.G.; Lomax, R.G. A Beginner’s Guide to Structural Equation Modeling; Routledge:
Mahwah, NJ, USA, 2012.
59.
Bagozzi, R.P.; Yi, Y. On the evaluation of structural equation models. J. Acad. Mark. Sci. 1988, 16, 74–94.
[CrossRef]
60.
Hatcher, L. A Step-by-Step Approach to Using the SAS System for Factor Analysis and Structural Equation Modelling;
SAS Institute: Cary, NC, USA, 1994.
61.
Vos, S.; Scheerder, J.; Boen, F.; Feys, J. A Typology of Runners: Implications for Marketing Strategies.
In Proceedings of the European Sport Management Conference, Heidelberg, Germany, 10–13 September 2008;
pp. 321–323.
62.
Funk, D.; Jordan, J.; Ridinger, L.; Kaplanidou, K. Capacity of Mass Participant Sport Events for the
Development of Activity Commitment and Future Exercise Intention. Leis. Sci. 2011, 33, 250–268. [CrossRef]
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
| Amateur Runners' Commitment: An Analysis of Sociodemographic and Sports Habit Profiles. | 02-02-2020 | Parra-Camacho, David,Alonso Dos Santos, Manuel,González-Serrano, María Huertas | eng |