Machine learning-driven pharmacovigilance of antidiabetic drugs: comparative reporting patterns and classification of serious adverse events.

PURPOSE: This study aims to evaluate the comparative adverse drug event (ADE) reporting patterns associated with antidiabetic drugs and develop machine learning (ML)-based classification models for the seriousness of reported ADEs. METHODS: We performed a retrospective analysis of 28,633 antidiabetic-related ADEs reported to the Korea Institute of Drug Safety and Management- Korea Adverse Event Reporting System database (KAERS DB 2505A0010) from 2015 to 2024. Disproportionality analyses identified safety signals using reporting odds ratios (RORs) with 95% confidence intervals (CIs). Factors associated with serious adverse event (SAE) classification were assessed using multivariate logistic regression, and three ML-based classification models were developed. RESULTS: Older adults accounted for 64.1% of the reported ADEs, with 2.35% classified as SAEs. Sulfonylureas demonstrated the highest SAE reporting signal (ROR 2.60, 95% CI 2.22-3.05). Male sex, older age, and sulfonylurea exposure were associated with higher odds of reports being classified as serious. Across the ML models, diabetic neuropathy treatment consistently emerged as the most influential feature contributing to SAE classification. CONCLUSION: ML-based classification complemented disproportionality analyses by identifying features associated with classification of reported ADEs as serious. Further validation using integrated longitudinal real-world data is warranted to confirm the robustness of these findings.