Beyond pooled analysis: sex-specific cardiovascular adverse event reporting signals with antiarrhythmic drugs in...

Sex differences in drug metabolism and cardiac electrophysiology may influence the cardiovascular safety profiles of antiarrhythmic drugs (AADs). However, pharmacovigilance studies often pool data, potentially obscuring sex-specific cardiovascular adverse event (CV-AE) reporting patterns. We aim to investigate sex-specific CV-AE reporting profiles associated with commonly used antiarrhythmic drugs using pharmacovigilance data. We analyzed 122,853 medication-role records representing 38,430 unique FAERS reports involving eight AADs. Disproportionality analyses defined index-drug exposure by the primary-suspect role, whereas the report-level XGBoost pipeline separately encoded primary-suspect drugs and secondary-suspect or concomitant medications. Analyses included pooled and sex-stratified disproportionality, drug-adjusted common odds ratios, and a gradient-boosting machine learning model to assess ventricular arrhythmia (VA) reporting patterns. Sex-stratified analysis uncovered "hidden" associations undetected in pooled data, including propafenone-QT prolongation in men and amiodarone-QT prolongation in women. Adverse event profiles were often drug-specific and opposing: cardiogenic shock showed higher reporting odds among males for sotalol, whereas flecainide showed higher reporting odds in females. After adjustment, women had higher odds of torsade de pointes and men of ventricular arrhythmia. However, a machine learning model for VA demonstrated that sex contributed modestly (1.4% SHAP contribution) compared with co-medications and the primary drug, though drug-specific associations differed. Sex-specific CV-AE profiles reporting patterns are often masked in pooled analyses, underscoring the need for sex-stratified pharmacovigilance to guide personalized antiarrhythmic therapy. The identified associations should be considered hypothesis-generating and require validation in datasets with richer clinical information.