Advancing pharmacovigilance through artificial intelligence: A review of applications and ethical considerations

Artificial intelligence (AI) is increasingly applied in pharmacovigilance (PV) to identify, prioritize, and interpret adverse drug reactions (ADRs) across real-world data sources. This narrative review synthesizes recent peer-reviewed studies (2015 – 2024) and maps AI use across four domains: extraction of ADRs from social and clinical text, supervised and ensemble signal detection in spontaneous reporting systems and electronic health records (EHRs), knowledge-graph-based discovery of drug–event associations, and prediction of outcome seriousness to support triage. Across domains, implementations most consistently enhance intake, coding, prioritization, and the timeliness of safety assessment, while graph-based methods surface plausible associations for follow-up and seriousness models aid risk stratification. Cross-cutting challenges include heterogeneous and shifting data, annotation burden, class imbalance (especially for rare events), and concerns around transparency, privacy, and fairness. Evidence remains predominantly retrospective, with uneven external validation, underscoring the need for prospective studies, standardized reporting and calibration, fairness audits, and closer alignment with regulatory signal-management workflows spanning detection, validation, analysis, prioritization, and assessment. By clarifying where AI is already dependable and where methodological and ethical gaps persist, this review offers practical directions for integrating AI into routine PV with auditable thresholds, monitoring, and human oversight.