Artificial Intelligence–Enabled Pharmacovigilance: A New Paradigm in Drug Safety Surveillance

Abstract

Pharmacovigilance (PV) has historically relied on manual intake and clinical review of spontaneous adverse event reports, constrained by under-reporting, variable data quality, duplicate cases, and operational latency. Over the last decade, the digitalization of Individual Case Safety Reports (ICSRs) and the maturation of machine learning (ML), natural language processing (NLP), and “augmented intelligence” workflows have enabled a shift from document-centric, manual processing toward data-centric, semi-automated surveillance. This review synthesizes the evolution from manual safety reporting to AI-enabled PV across the ICSR lifecycle: intake and triage, data extraction and normalization, coding (MedDRA/WHO Drug), case processing and medical assessment support, duplicate detection, signal detection and validation, and regulatory submission. We summarize the enabling regulatory/technical standards (e.g., ICH E2B (R3), EMA GVP), global surveillance ecosystems (FAERS, EudraVigilance, VigiBase), and contemporary guidance (e.g., CIOMS Working Group XIV) that shape responsible AI adoption. We present publication-ready tables describing (i) the PV technology timeline, (ii) AI methods mapped to PV tasks with validation metrics and risks, and (iii) an implementation governance checklist. A figure proposes an end-to-end AI-enabled PV operating model with human oversight and audit ability. We conclude that the most durable value arises from targeted automation (coding, extraction, prioritization) combined with robust governance data quality management, bias monitoring, explain ability, and continuous performance verification rather than full replacement of expert judgment.