Description
- Abstract:
- Adverse drug reactions (ADRs) pose a significant threat to medication safety, with risks varying across populations. Sex is a key factor influencing the heterogeneity of drug responses. However, existing pharmacovigilance research often focuses on risk prediction in the overall population, lacking systematic modeling and comparative analysis of sex-specific effects. This study developed sex-specific ADR prediction models among individuals with reported ADR cases. The goal is to identify risk factors for specific ADRs and quantify differences between males and females, thereby improving the interpretability of pharmacovigilance and providing clinically meaningful sex-specific safety insights for personalized risk prediction. Using the FAERS database (2015-2024), disproportionality analysis was employed to identify drugs significantly associated with each ADR. Two modeling strategies were subsequently applied for sex-stratified analysis: (i) Relaxed LASSO regression incorporating sex-drug interaction terms to estimate sex-specific effects within a linear framework; and (ii) sex-stratified Random Forest (RF) models combined with SHAP for interpretability, quantifying the predictive contribution of each drug in male and female models. Model performance was comprehensively evaluated using accuracy, ROC-AUC, PR-AUC, and F1 score. The two modeling approaches exhibited distinct predictive performance and risk identification characteristics across sex subgroups. The Relaxed LASSO model achieved superior overall predictive performance, capturing sex-drug interactions and identifying sex-biased drugs (e.g., Striant in males, Zolpidem in females). Conversely, the RF model revealed nonlinear structures, identifying cross-sex core drugs and sex-specific contributors (e.g., follitropin in females, influenza virus in males). These two methods are complementary in capturing sex heterogeneity. This study systematically revealed both shared and divergent ADR risk profiles across sexes via two interpretable models. Relaxed LASSO is well-suited for identifying and quantifying sex-drug interactions, while RF excels in uncovering complex risk structures. This multi-method comparison not only validates the necessity of sex-stratified modeling but also provides a methodological foundation for developing interpretable, robust, and clinically actionable sex-differentiated pharmacovigilance systems. The two methods identified distinct sets of risk-associated drugs, warranting further investigation. Future research should further integrate physiological, genomic, and other multidimensional data to advance the development of precision medication safety monitoring.
- Notes:
- Thesis (Sc. M.)--Brown University, 2026
Citation
Gao, Xurui,
"Sex-Specific Prediction of Adverse Drug Reactions Associated with Drug Combinations: A FAERS-Based Study"
(2026).
Biostatistics Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://repository.library.brown.edu/studio/item/bdr:yy4vk6ut/
Relations
Collection:
-
Biostatistics Theses and Dissertations
Theses and Dissertations for the Biostatistics department....