Adverse Drug Reactions (ADR) Prediction Using Random Forest Algorithm and Neural Networks

Abstract

This study develops an adverse drug reaction (ADR) prediction model using machine learning based on Indonesian pharmacovigilance data. Comparing the performance of Random Forest (RF) and Deep Neural Network (DNN) on a dataset of 12,543 cases from five referral hospitals (2019-2023), results show RF achieved superior AUC of 0.891 (95%CI 0.876-0.906) compared to DNN (AUC 0.863, p=0.013). SHAP analysis identified five key predictors: drug interactions (SHAP=0.42), age >65 years (0.38), bioavailability <30% (0.35), polypharmacy ≥5 drugs (0.33), and renal impairment (0.29). Prospective validation demonstrated 83.6% accuracy with 0.47-second response time. Implementation in Clinical Decision Support Systems (CDSS) could potentially reduce 29-34% of severe ADR incidents and save IDR 1.8 billion annually. These findings support the development of precision pharmacovigilance in Indonesia aligned with national digital health transformation.