pharmacy-mcp / adr_analysis.py
Chris McMaster
Updates, improvements, new ADR features
f32824f
"""
Advanced Adverse Drug Reaction (ADR) Analysis Tools
This module provides comprehensive pharmacovigilance capabilities including:
- Enhanced FAERS database searches with filtering
- Naranjo probability scale calculator
- Disproportionality analysis (PRR, ROR, IC)
- Case similarity analysis
- Temporal pattern analysis
"""
import requests
import re
import math
import logging
from datetime import datetime, timedelta
from typing import Dict, List, Any, Optional, Tuple
from collections import defaultdict, Counter
from caching import with_caching
from utils import with_error_handling, make_api_request
logger = logging.getLogger(__name__)
@with_error_handling
@with_caching(ttl=1800)
def enhanced_faers_search(
drug_name: str,
adverse_event: str = None,
age_range: str = None,
gender: str = None,
serious_only: bool = False,
limit: int = 100
) -> Dict[str, Any]:
"""
Enhanced FAERS search with filtering capabilities for pharmacovigilance analysis.
Args:
drug_name: Drug name to search for
adverse_event: Specific adverse event/reaction to filter by (optional)
age_range: Age range filter like "18-65" or ">65" (optional)
gender: Gender filter "1" (male) or "2" (female) (optional)
serious_only: If True, only return serious adverse events
limit: Maximum number of results (default 100)
Returns:
Dict with enhanced case data including demographics, outcomes, and temporal info
"""
if not drug_name or not drug_name.strip():
raise ValueError("Drug name cannot be empty")
# Build search query
search_parts = [f'patient.drug.medicinalproduct:"{drug_name.strip()}"']
if adverse_event:
search_parts.append(f'patient.reaction.reactionmeddrapt:"{adverse_event.strip()}"')
if serious_only:
search_parts.append('serious:"1"')
if gender in ["1", "2"]:
search_parts.append(f'patient.patientsex:"{gender}"')
search_query = " AND ".join(search_parts)
base_url = "https://api.fda.gov/drug/event.json"
query_params = {
"search": search_query,
"limit": min(max(1, limit), 1000)
}
response = make_api_request(base_url, query_params, timeout=15)
if response.status_code != 200:
if response.status_code == 404:
return {
"cases": [],
"total_found": 0,
"query_info": {
"drug": drug_name,
"adverse_event": adverse_event,
"filters_applied": {
"age_range": age_range,
"gender": gender,
"serious_only": serious_only
}
},
"message": "No matching cases found"
}
raise requests.exceptions.RequestException(f"Enhanced FAERS search failed: {response.status_code}")
data = response.json()
cases = []
for rec in data.get("results", []):
case = extract_case_details(rec, age_range)
if case: # Only include if age filter passes
cases.append(case)
# Calculate summary statistics
summary_stats = calculate_case_statistics(cases)
return {
"cases": cases,
"total_found": data.get("meta", {}).get("results", {}).get("total", 0),
"filtered_count": len(cases),
"query_info": {
"drug": drug_name,
"adverse_event": adverse_event,
"filters_applied": {
"age_range": age_range,
"gender": gender,
"serious_only": serious_only
}
},
"summary_statistics": summary_stats
}
def extract_case_details(rec: Dict, age_range: str = None) -> Optional[Dict]:
"""Extract and structure case details from FAERS record."""
patient = rec.get("patient", {})
# Extract patient demographics
age = patient.get("patientagegroup")
age_years = patient.get("patientage")
gender = patient.get("patientsex")
# Apply age filter if specified
if age_range and age_years:
try:
age_num = float(age_years)
if not passes_age_filter(age_num, age_range):
return None
except (ValueError, TypeError):
pass
# Extract drug information
drugs = []
for drug in patient.get("drug", []):
drug_info = {
"name": drug.get("medicinalproduct", ""),
"characterization": drug.get("drugcharacterization"), # 1=suspect, 2=concomitant, 3=interacting
"indication": drug.get("drugindication", ""),
"start_date": drug.get("drugstartdate", ""),
"end_date": drug.get("drugenddate", ""),
"dosage": drug.get("drugdosagetext", ""),
"route": drug.get("drugadministrationroute", "")
}
drugs.append(drug_info)
# Extract reactions
reactions = []
for reaction in patient.get("reaction", []):
reaction_info = {
"term": reaction.get("reactionmeddrapt", ""),
"outcome": reaction.get("reactionoutcome") # 1=recovered, 2=recovering, 3=not recovered, 4=recovered with sequelae, 5=fatal, 6=unknown
}
reactions.append(reaction_info)
# Extract seriousness criteria
seriousness = {
"serious": bool(int(rec.get("serious", "0"))),
"death": bool(int(rec.get("seriousnessdeath", "0"))),
"life_threatening": bool(int(rec.get("seriousnesslifethreatening", "0"))),
"hospitalization": bool(int(rec.get("seriousnesshospitalization", "0"))),
"disability": bool(int(rec.get("seriousnessdisabling", "0"))),
"congenital_anomaly": bool(int(rec.get("seriousnesscongenitalanomali", "0"))),
"other_serious": bool(int(rec.get("seriousnessother", "0")))
}
return {
"safety_report_id": rec.get("safetyreportid"),
"receive_date": rec.get("receivedate"),
"patient": {
"age": age_years,
"age_group": age,
"gender": gender, # 1=male, 2=female
"weight": patient.get("patientweight")
},
"drugs": drugs,
"reactions": reactions,
"seriousness": seriousness,
"reporter_qualification": rec.get("primarysource", {}).get("qualification"), # 1=physician, 2=pharmacist, etc.
"country": rec.get("occurcountry")
}
def passes_age_filter(age: float, age_range: str) -> bool:
"""Check if age passes the specified filter."""
age_range = age_range.strip()
if age_range.startswith(">"):
threshold = float(age_range[1:])
return age > threshold
elif age_range.startswith("<"):
threshold = float(age_range[1:])
return age < threshold
elif age_range.startswith(">="):
threshold = float(age_range[2:])
return age >= threshold
elif age_range.startswith("<="):
threshold = float(age_range[2:])
return age <= threshold
elif "-" in age_range:
min_age, max_age = map(float, age_range.split("-"))
return min_age <= age <= max_age
return True
def calculate_case_statistics(cases: List[Dict]) -> Dict[str, Any]:
"""Calculate summary statistics from case data."""
if not cases:
return {}
# Demographics
ages = [float(case["patient"]["age"]) for case in cases if case["patient"]["age"]]
genders = [case["patient"]["gender"] for case in cases if case["patient"]["gender"]]
# Outcomes
serious_cases = sum(1 for case in cases if case["seriousness"]["serious"])
fatal_cases = sum(1 for case in cases if case["seriousness"]["death"])
# Reporter types
reporter_types = [case["reporter_qualification"] for case in cases if case["reporter_qualification"]]
# Most common reactions
all_reactions = []
for case in cases:
all_reactions.extend([r["term"] for r in case["reactions"]])
reaction_counts = Counter(all_reactions)
stats = {
"total_cases": len(cases),
"serious_cases": serious_cases,
"serious_percentage": round(serious_cases / len(cases) * 100, 1),
"fatal_cases": fatal_cases,
"fatal_percentage": round(fatal_cases / len(cases) * 100, 1) if len(cases) > 0 else 0,
"demographics": {
"age_stats": {
"mean": round(sum(ages) / len(ages), 1) if ages else None,
"median": sorted(ages)[len(ages)//2] if ages else None,
"range": [min(ages), max(ages)] if ages else None
},
"gender_distribution": dict(Counter(genders))
},
"top_reactions": dict(reaction_counts.most_common(10)),
"reporter_types": dict(Counter(reporter_types))
}
return stats
@with_error_handling
def calculate_naranjo_score(
adverse_reaction_after_drug: str, # "yes", "no", "unknown"
reaction_improved_after_stopping: str, # "yes", "no", "unknown"
reaction_reappeared_after_readministration: str, # "yes", "no", "unknown"
alternative_causes_exist: str, # "yes", "no", "unknown"
reaction_when_placebo_given: str, # "yes", "no", "unknown"
drug_detected_in_blood: str, # "yes", "no", "unknown"
reaction_worse_with_higher_dose: str, # "yes", "no", "unknown"
similar_reaction_to_drug_before: str, # "yes", "no", "unknown"
adverse_event_confirmed_objectively: str, # "yes", "no", "unknown"
reaction_appeared_after_suspected_drug_given: str # "yes", "no", "unknown"
) -> Dict[str, Any]:
"""
Calculate Naranjo Adverse Drug Reaction Probability Scale.
The Naranjo scale helps determine the likelihood that an adverse event
is related to drug therapy rather than other factors.
Args:
All parameters should be "yes", "no", or "unknown"
Returns:
Dict with score, probability category, and detailed breakdown
"""
# Naranjo scoring system
questions = [
{
"question": "Are there previous conclusive reports on this reaction?",
"answer": adverse_reaction_after_drug,
"scores": {"yes": 1, "no": 0, "unknown": 0}
},
{
"question": "Did the adverse event appear after the suspected drug was administered?",
"answer": reaction_appeared_after_suspected_drug_given,
"scores": {"yes": 2, "no": -1, "unknown": 0}
},
{
"question": "Did the adverse reaction improve when the drug was discontinued or a specific antagonist was administered?",
"answer": reaction_improved_after_stopping,
"scores": {"yes": 1, "no": 0, "unknown": 0}
},
{
"question": "Did the adverse reaction reappear when the drug was readministered?",
"answer": reaction_reappeared_after_readministration,
"scores": {"yes": 2, "no": -1, "unknown": 0}
},
{
"question": "Are there alternative causes (other than the drug) that could on their own have caused the reaction?",
"answer": alternative_causes_exist,
"scores": {"yes": -1, "no": 2, "unknown": 0}
},
{
"question": "Did the reaction reappear when a placebo was given?",
"answer": reaction_when_placebo_given,
"scores": {"yes": -1, "no": 1, "unknown": 0}
},
{
"question": "Was the drug detected in blood (or other fluids) in concentrations known to be toxic?",
"answer": drug_detected_in_blood,
"scores": {"yes": 1, "no": 0, "unknown": 0}
},
{
"question": "Was the reaction more severe when the dose was increased or less severe when the dose was decreased?",
"answer": reaction_worse_with_higher_dose,
"scores": {"yes": 1, "no": 0, "unknown": 0}
},
{
"question": "Did the patient have a similar reaction to the same or similar drugs in any previous exposure?",
"answer": similar_reaction_to_drug_before,
"scores": {"yes": 1, "no": 0, "unknown": 0}
},
{
"question": "Was the adverse event confirmed by any objective evidence?",
"answer": adverse_event_confirmed_objectively,
"scores": {"yes": 1, "no": 0, "unknown": 0}
}
]
total_score = 0
question_details = []
for q in questions:
answer = q["answer"].lower().strip()
if answer not in q["scores"]:
raise ValueError(f"Invalid answer '{answer}'. Must be 'yes', 'no', or 'unknown'")
score = q["scores"][answer]
total_score += score
question_details.append({
"question": q["question"],
"answer": answer,
"points": score
})
# Determine probability category
if total_score >= 9:
category = "Definite"
probability = "≥95%"
interpretation = "The adverse reaction is definitely related to the drug."
elif total_score >= 5:
category = "Probable"
probability = "75-95%"
interpretation = "The adverse reaction is probably related to the drug."
elif total_score >= 1:
category = "Possible"
probability = "25-75%"
interpretation = "The adverse reaction is possibly related to the drug."
else:
category = "Doubtful"
probability = "<25%"
interpretation = "The adverse reaction is doubtfully related to the drug."
return {
"total_score": total_score,
"category": category,
"probability": probability,
"interpretation": interpretation,
"question_breakdown": question_details,
"scale_info": {
"name": "Naranjo Adverse Drug Reaction Probability Scale",
"reference": "Naranjo CA, et al. Clin Pharmacol Ther. 1981;30(2):239-245",
"scoring": {
"Definite": "≥9 points",
"Probable": "5-8 points",
"Possible": "1-4 points",
"Doubtful": "≤0 points"
}
}
}
@with_error_handling
@with_caching(ttl=3600)
def disproportionality_analysis(
drug_name: str,
adverse_event: str,
background_limit: int = 10000
) -> Dict[str, Any]:
"""
Perform disproportionality analysis to detect potential drug-adverse event signals.
Calculates Proportional Reporting Ratio (PRR), Reporting Odds Ratio (ROR),
and Information Component (IC) with confidence intervals.
Args:
drug_name: Drug of interest
adverse_event: Adverse event of interest
background_limit: Number of background cases to sample for comparison
Returns:
Dict with PRR, ROR, IC values and statistical significance
"""
try:
base_url = "https://api.fda.gov/drug/event.json"
# Get cases for drug + adverse event (a)
drug_ae_query = {
"search": f'patient.drug.medicinalproduct:"{drug_name}" AND patient.reaction.reactionmeddrapt:"{adverse_event}"',
"limit": 1
}
drug_ae_response = make_api_request(base_url, drug_ae_query, timeout=10)
if drug_ae_response and drug_ae_response.status_code == 200:
drug_ae_data = drug_ae_response.json()
a = drug_ae_data.get("meta", {}).get("results", {}).get("total", 0)
else:
a = 0
if a == 0:
return {
"drug": drug_name,
"adverse_event": adverse_event,
"message": "No cases found for this drug-adverse event combination",
"signal_detected": False,
"case_count": 0
}
# Get total cases for drug (a + b)
drug_total_query = {
"search": f'patient.drug.medicinalproduct:"{drug_name}"',
"limit": 1
}
drug_total_response = make_api_request(base_url, drug_total_query, timeout=10)
if drug_total_response and drug_total_response.status_code == 200:
drug_total_data = drug_total_response.json()
total_drug_cases = drug_total_data.get("meta", {}).get("results", {}).get("total", 0)
b = max(total_drug_cases - a, 1) # Ensure b is at least 1
else:
b = max(a * 5, 10) # Conservative estimate
# Get total cases for adverse event (a + c)
ae_total_query = {
"search": f'patient.reaction.reactionmeddrapt:"{adverse_event}"',
"limit": 1
}
ae_total_response = make_api_request(base_url, ae_total_query, timeout=10)
if ae_total_response and ae_total_response.status_code == 200:
ae_total_data = ae_total_response.json()
total_ae_cases = ae_total_data.get("meta", {}).get("results", {}).get("total", 0)
c = max(total_ae_cases - a, 1) # Avoid zero
else:
c = max(a * 10, 100) # Conservative estimate
# Estimate total background cases (d)
# Use a reasonable estimate based on FAERS database size
total_cases_estimate = 15000000 # Approximate FAERS database size
d = max(total_cases_estimate - a - b - c, 1000)
# Calculate disproportionality measures
results = calculate_disproportionality_measures(a, b, c, d)
# Add metadata
results.update({
"drug": drug_name,
"adverse_event": adverse_event,
"contingency_table": {
"drug_ae": a,
"drug_other_ae": b,
"other_drug_ae": c,
"other_drug_other_ae": d,
"total": a + b + c + d
},
"data_sources": {
"drug_ae_cases": "FAERS API direct query",
"total_drug_cases": "FAERS API direct query",
"total_ae_cases": "FAERS API direct query",
"background_estimate": "Statistical approximation"
},
"data_notes": [
"This analysis uses FAERS data which has inherent limitations",
"Results should be interpreted by qualified pharmacovigilance professionals",
"Background estimates are approximations due to API limitations",
"Consider confounding factors and reporting biases"
]
})
return results
except Exception as e:
logger.error(f"Error in disproportionality analysis: {e}")
return {
"drug": drug_name,
"adverse_event": adverse_event,
"error": str(e),
"message": "Analysis failed due to data access issues",
"signal_detected": False,
"case_count": 0
}
def calculate_disproportionality_measures(a: int, b: int, c: int, d: int) -> Dict[str, Any]:
"""
Calculate PRR, ROR, and IC with confidence intervals.
2x2 contingency table:
AE of Interest Other AEs
Drug of Interest a b
Other Drugs c d
"""
# Proportional Reporting Ratio (PRR)
prr = (a / (a + b)) / (c / (c + d)) if (a + b) > 0 and (c + d) > 0 else 0
# PRR 95% CI (using log transformation)
if a > 0:
log_prr = math.log(prr)
se_log_prr = math.sqrt(1/a + 1/c - 1/(a+b) - 1/(c+d))
prr_ci_lower = math.exp(log_prr - 1.96 * se_log_prr)
prr_ci_upper = math.exp(log_prr + 1.96 * se_log_prr)
else:
prr_ci_lower = prr_ci_upper = 0
# Reporting Odds Ratio (ROR)
ror = (a * d) / (b * c) if b > 0 and c > 0 else 0
# ROR 95% CI
if a > 0 and b > 0 and c > 0 and d > 0:
log_ror = math.log(ror)
se_log_ror = math.sqrt(1/a + 1/b + 1/c + 1/d)
ror_ci_lower = math.exp(log_ror - 1.96 * se_log_ror)
ror_ci_upper = math.exp(log_ror + 1.96 * se_log_ror)
else:
ror_ci_lower = ror_ci_upper = 0
# Information Component (IC)
expected = ((a + b) * (a + c)) / (a + b + c + d)
ic = math.log2(a / expected) if expected > 0 and a > 0 else 0
# IC 95% CI (simplified approximation)
if a > 0:
ic_se = 1 / (math.log(2) * math.sqrt(a))
ic_ci_lower = ic - 1.96 * ic_se
ic_ci_upper = ic + 1.96 * ic_se
else:
ic_ci_lower = ic_ci_upper = 0
# Signal detection criteria
prr_signal = prr >= 2.0 and prr_ci_lower > 1.0 and a >= 3
ror_signal = ror >= 2.0 and ror_ci_lower > 1.0 and a >= 3
ic_signal = ic_ci_lower > 0 and a >= 3
signal_detected = prr_signal or ror_signal or ic_signal
return {
"proportional_reporting_ratio": {
"value": round(prr, 3),
"confidence_interval_95": [round(prr_ci_lower, 3), round(prr_ci_upper, 3)],
"signal_detected": prr_signal,
"interpretation": "PRR ≥2 with lower CI >1 suggests potential signal" if prr_signal else "No signal detected by PRR criteria"
},
"reporting_odds_ratio": {
"value": round(ror, 3),
"confidence_interval_95": [round(ror_ci_lower, 3), round(ror_ci_upper, 3)],
"signal_detected": ror_signal,
"interpretation": "ROR ≥2 with lower CI >1 suggests potential signal" if ror_signal else "No signal detected by ROR criteria"
},
"information_component": {
"value": round(ic, 3),
"confidence_interval_95": [round(ic_ci_lower, 3), round(ic_ci_upper, 3)],
"signal_detected": ic_signal,
"interpretation": "IC lower CI >0 suggests potential signal" if ic_signal else "No signal detected by IC criteria"
},
"overall_signal_detected": signal_detected,
"case_count": a,
"signal_strength": "Strong" if (prr_signal and ror_signal and ic_signal) else
"Moderate" if signal_detected else "Weak/None"
}
@with_error_handling
@with_caching(ttl=1800)
def find_similar_cases(
reference_case_id: str,
similarity_threshold: float = 0.7,
limit: int = 50
) -> Dict[str, Any]:
"""
Find cases similar to a reference case based on patient characteristics,
drugs, and adverse events.
Args:
reference_case_id: FAERS safety report ID to use as reference
similarity_threshold: Minimum similarity score (0-1)
limit: Maximum number of similar cases to return
Returns:
Dict with similar cases and similarity scores
"""
# First, get the reference case details
from drug_data_endpoints import fetch_event_details
try:
ref_case = fetch_event_details(reference_case_id)
except Exception as e:
raise ValueError(f"Could not fetch reference case {reference_case_id}: {e}")
ref_drugs = [drug.lower() for drug in ref_case["drugs"]]
ref_reactions = [reaction.lower() for reaction in ref_case["reactions"]]
if not ref_drugs:
raise ValueError("Reference case has no drug information")
# Search for cases with similar drugs
primary_drug = ref_drugs[0] if ref_drugs else ""
similar_cases_response = enhanced_faers_search(
drug_name=primary_drug,
limit=min(limit * 3, 500) # Get more cases to filter
)
similar_cases = []
for case in similar_cases_response["cases"]:
case_drugs = [drug["name"].lower() for drug in case["drugs"] if drug["name"]]
case_reactions = [reaction["term"].lower() for reaction in case["reactions"] if reaction["term"]]
# Skip the reference case itself
if case["safety_report_id"] == reference_case_id:
continue
# Calculate similarity score
similarity_score = calculate_case_similarity(
ref_drugs, ref_reactions,
case_drugs, case_reactions,
ref_case.get("full_record", {}).get("patient", {}),
case.get("patient", {})
)
if similarity_score >= similarity_threshold:
similar_cases.append({
"case": case,
"similarity_score": similarity_score,
"similarity_factors": get_similarity_factors(
ref_drugs, ref_reactions, case_drugs, case_reactions
)
})
# Sort by similarity score
similar_cases.sort(key=lambda x: x["similarity_score"], reverse=True)
return {
"reference_case_id": reference_case_id,
"reference_drugs": ref_drugs,
"reference_reactions": ref_reactions,
"similar_cases": similar_cases[:limit],
"total_similar_found": len(similar_cases),
"similarity_threshold": similarity_threshold,
"analysis_summary": {
"most_common_shared_drugs": get_most_common_shared_elements(
[case["similarity_factors"]["shared_drugs"] for case in similar_cases]
),
"most_common_shared_reactions": get_most_common_shared_elements(
[case["similarity_factors"]["shared_reactions"] for case in similar_cases]
)
}
}
def calculate_case_similarity(
ref_drugs: List[str], ref_reactions: List[str],
case_drugs: List[str], case_reactions: List[str],
ref_patient: Dict, case_patient: Dict
) -> float:
"""Calculate similarity score between two cases."""
# Drug similarity (Jaccard index)
ref_drugs_set = set(ref_drugs)
case_drugs_set = set(case_drugs)
drug_intersection = len(ref_drugs_set & case_drugs_set)
drug_union = len(ref_drugs_set | case_drugs_set)
drug_similarity = drug_intersection / drug_union if drug_union > 0 else 0
# Reaction similarity (Jaccard index)
ref_reactions_set = set(ref_reactions)
case_reactions_set = set(case_reactions)
reaction_intersection = len(ref_reactions_set & case_reactions_set)
reaction_union = len(ref_reactions_set | case_reactions_set)
reaction_similarity = reaction_intersection / reaction_union if reaction_union > 0 else 0
# Patient similarity (age and gender)
patient_similarity = 0
similarity_factors = 0
# Age similarity
ref_age = ref_patient.get("patientage")
case_age = case_patient.get("age")
if ref_age and case_age:
try:
age_diff = abs(float(ref_age) - float(case_age))
age_similarity = max(0, 1 - age_diff / 50) # Normalize by 50 years
patient_similarity += age_similarity
similarity_factors += 1
except (ValueError, TypeError):
pass
# Gender similarity
ref_gender = ref_patient.get("patientsex")
case_gender = case_patient.get("gender")
if ref_gender and case_gender and ref_gender == case_gender:
patient_similarity += 1
similarity_factors += 1
elif ref_gender and case_gender:
similarity_factors += 1
if similarity_factors > 0:
patient_similarity /= similarity_factors
# Weighted overall similarity
# Drugs and reactions are most important, patient characteristics less so
overall_similarity = (
0.5 * drug_similarity +
0.4 * reaction_similarity +
0.1 * patient_similarity
)
return round(overall_similarity, 3)
def get_similarity_factors(
ref_drugs: List[str], ref_reactions: List[str],
case_drugs: List[str], case_reactions: List[str]
) -> Dict[str, List[str]]:
"""Get the specific shared elements between cases."""
shared_drugs = list(set(ref_drugs) & set(case_drugs))
shared_reactions = list(set(ref_reactions) & set(case_reactions))
return {
"shared_drugs": shared_drugs,
"shared_reactions": shared_reactions,
"unique_to_reference_drugs": list(set(ref_drugs) - set(case_drugs)),
"unique_to_case_drugs": list(set(case_drugs) - set(ref_drugs)),
"unique_to_reference_reactions": list(set(ref_reactions) - set(case_reactions)),
"unique_to_case_reactions": list(set(case_reactions) - set(ref_reactions))
}
def get_most_common_shared_elements(element_lists: List[List[str]]) -> Dict[str, int]:
"""Get the most commonly shared elements across multiple cases."""
all_elements = []
for element_list in element_lists:
all_elements.extend(element_list)
return dict(Counter(all_elements).most_common(10))
@with_error_handling
@with_caching(ttl=3600)
def temporal_analysis(
drug_name: str,
adverse_event: str = None,
limit: int = 500
) -> Dict[str, Any]:
"""
Analyze temporal patterns of adverse events for a drug.
Args:
drug_name: Drug to analyze
adverse_event: Specific adverse event (optional)
limit: Maximum cases to analyze
Returns:
Dict with temporal patterns and time-to-onset analysis
"""
# Get cases with temporal information
cases_response = enhanced_faers_search(
drug_name=drug_name,
adverse_event=adverse_event,
limit=limit
)
cases = cases_response["cases"]
if not cases:
return {
"drug": drug_name,
"adverse_event": adverse_event,
"message": "No cases found for temporal analysis"
}
# Analyze time to onset patterns
onset_times = []
reporting_dates = []
for case in cases:
# Extract drug start dates and reaction onset
for drug in case["drugs"]:
if drug["name"].lower() == drug_name.lower() and drug["start_date"]:
try:
# Parse date (YYYYMMDD format)
start_date = datetime.strptime(drug["start_date"], "%Y%m%d")
# For now, we'll use receive date as proxy for reaction onset
# In practice, you'd want more sophisticated temporal extraction
if case["receive_date"]:
receive_date = datetime.strptime(case["receive_date"], "%Y%m%d")
onset_time = (receive_date - start_date).days
if 0 <= onset_time <= 365: # Filter reasonable onset times
onset_times.append(onset_time)
reporting_dates.append(receive_date)
except (ValueError, TypeError):
continue
# Calculate temporal statistics
temporal_stats = {}
if onset_times:
onset_times.sort()
temporal_stats["time_to_onset"] = {
"median_days": onset_times[len(onset_times)//2],
"mean_days": round(sum(onset_times) / len(onset_times), 1),
"range_days": [min(onset_times), max(onset_times)],
"percentiles": {
"25th": onset_times[len(onset_times)//4],
"75th": onset_times[3*len(onset_times)//4],
"90th": onset_times[9*len(onset_times)//10] if len(onset_times) >= 10 else max(onset_times)
},
"distribution": categorize_onset_times(onset_times)
}
if reporting_dates:
# Analyze reporting trends over time
reporting_dates.sort()
temporal_stats["reporting_trends"] = analyze_reporting_trends(reporting_dates)
return {
"drug": drug_name,
"adverse_event": adverse_event,
"total_cases_analyzed": len(cases),
"cases_with_temporal_data": len(onset_times),
"temporal_analysis": temporal_stats,
"interpretation": interpret_temporal_patterns(temporal_stats)
}
def categorize_onset_times(onset_times: List[int]) -> Dict[str, int]:
"""Categorize onset times into clinically relevant periods."""
categories = {
"immediate_0_1_day": 0,
"acute_1_7_days": 0,
"subacute_1_4_weeks": 0,
"delayed_1_3_months": 0,
"late_3_12_months": 0
}
for onset in onset_times:
if onset <= 1:
categories["immediate_0_1_day"] += 1
elif onset <= 7:
categories["acute_1_7_days"] += 1
elif onset <= 28:
categories["subacute_1_4_weeks"] += 1
elif onset <= 90:
categories["delayed_1_3_months"] += 1
elif onset <= 365:
categories["late_3_12_months"] += 1
return categories
def analyze_reporting_trends(reporting_dates: List[datetime]) -> Dict[str, Any]:
"""Analyze trends in adverse event reporting over time."""
# Group by year
year_counts = defaultdict(int)
for date in reporting_dates:
year_counts[date.year] += 1
# Calculate trend
years = sorted(year_counts.keys())
if len(years) >= 3:
recent_avg = sum(year_counts[year] for year in years[-3:]) / 3
early_avg = sum(year_counts[year] for year in years[:3]) / 3
trend = "increasing" if recent_avg > early_avg * 1.2 else "decreasing" if recent_avg < early_avg * 0.8 else "stable"
else:
trend = "insufficient_data"
return {
"yearly_counts": dict(year_counts),
"date_range": [min(reporting_dates).year, max(reporting_dates).year],
"trend": trend,
"peak_year": max(year_counts.keys(), key=lambda k: year_counts[k]) if year_counts else None
}
def interpret_temporal_patterns(temporal_stats: Dict) -> List[str]:
"""Provide clinical interpretation of temporal patterns."""
interpretations = []
if "time_to_onset" in temporal_stats:
onset_data = temporal_stats["time_to_onset"]
median_onset = onset_data["median_days"]
if median_onset <= 1:
interpretations.append("Immediate onset pattern suggests Type A (dose-dependent) reaction or acute hypersensitivity")
elif median_onset <= 7:
interpretations.append("Acute onset pattern typical of many drug allergies and dose-related effects")
elif median_onset <= 28:
interpretations.append("Subacute onset may suggest immune-mediated or cumulative toxicity")
elif median_onset <= 90:
interpretations.append("Delayed onset pattern may indicate idiosyncratic reactions or chronic toxicity")
else:
interpretations.append("Late onset suggests possible chronic effects or delayed hypersensitivity")
# Check distribution
distribution = onset_data.get("distribution", {})
immediate = distribution.get("immediate_0_1_day", 0)
total_with_onset = sum(distribution.values())
if total_with_onset > 0:
immediate_pct = immediate / total_with_onset * 100
if immediate_pct > 50:
interpretations.append(f"High proportion ({immediate_pct:.1f}%) of immediate reactions suggests acute mechanism")
if "reporting_trends" in temporal_stats:
trend = temporal_stats["reporting_trends"]["trend"]
if trend == "increasing":
interpretations.append("Increasing reporting trend may indicate growing awareness or emerging safety signal")
elif trend == "decreasing":
interpretations.append("Decreasing reporting trend may suggest improved safety monitoring or reduced use")
if not interpretations:
interpretations.append("Insufficient temporal data for meaningful interpretation")
return interpretations