Clinical Validation and Application of an Artificial Intelligence-Based Model for Detection and Prediction of Adverse Drug Reactions: A Literature Review

Abstract

Background : Adverse drug events (ADEs) account for nearly 50% of preventable medical harm globally and place a burden on health care systems. Conventional pharmacovigilance based on spontaneous reporting faces challenges such as underreporting, low data quality, and delays in detecting safety signals. Artificial intelligence (AI) and machine learning (ML) are increasingly being applied to address these limitations, but evidence of their effectiveness and barriers to implementation remain scattered. Objective : To analyze the progress, effectiveness, benefits, and challenges of applying AI in pharmacovigilance, including signal detection, ADE prediction, clinical information extraction, causality assessment, and regulatory integration. Methodology : This study employed a narrative review of articles on AI and ML in pharmacovigilance and drug toxicity prediction. The literature included systematic reviews, scoping reviews, predictive modeling studies, and WHO policy documents. Results: Statistical methods such as PRR, ROR, BCPNN, and MGPS are still used because they are simple and auditable, despite their limited sensitivity. ML models such as Random Forest, XGBoost, logistic regression, and transformers like BERT and GPT show promising performance in the detection and prediction of ADEs, with accuracies of up to 94%. However, external validation, class imbalance, and interpretability remain challenges. Conclusion : AI has the potential to improve the efficiency and accuracy of pharmacovigilance, but requires improved data quality, real-world validation, model transparency, methodological standardization, and ongoing integration into clinical workflows.