From black box to clear box: explainable AI for next-gen pharmacovigilance

Pharmacovigilance (PV) plays a vital role in post-marketing surveillance of drug safety; however, traditional methods are hindered by underreporting, data heterogeneity, and delayed signal detection. Artificial Intelligence (AI) has been increasingly employed to overcome these shortcomings through automation, advanced pattern recognition, and multimodal fusion of patient data.

The regulatory acceptance of these systems rests on explainability, transparency, and auditability, which necessitate the inclusion of explainable AI (XAI) methods. This review thoroughly examines the international regulatory environment for AI in PV and suggests using Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) to build regulatory trust. Research materials, including official guidelines and publications, were systematically sourced from authoritative regulatory websites in the US, Europe, and India, as well as from major academic databases such as Google Scholar, ScienceDirect, and PubMed.

The EU AI Act classifies PV algorithms as 'high-risk,' imposing strict requirements for human oversight and transparency, aligning with the FDA's GMLP and India's Responsible AI principles. The review emphasizes traceability and governance, recommending convergence with human oversight into a regulatory-fit XAI framework to maximize safety signal detection and foster international trust in AI-based decision-making.