Modern Pharmacovigilance across Spontaneous Reports, Real-World Data, Machine Learning, and Active Surveillance: An Umbrella Review of Contemporary Evidence

Modern pharmacovigilance increasingly combines spontaneous reports with electronic health records, administrative claims, registries, distributed data networks, active monitoring, natural-language processing, and machine-learning methods. Yet the evidentiary roles of these approaches are often conflated: signal generation may be treated as causal assessment, computational performance as operational utility, and implementation activity as evidence of improved patient safety. This umbrella review evaluates contemporary review-level evidence across major surveillance approaches and examines methodological quality, primary-study overlap, convergence, contradiction, transferability, and implementation maturity. Systematic reviews, meta-analyses, scoping reviews, and mapping reviews were treated as the principal evidence units, while methodological and implementation papers were used only to define appraisal, overlap, reporting, or deployment boundaries. Review-level quality was assessed through critical domains appropriate to each review type rather than through unsupported numerical scores. Primary-study overlap was managed through planned citation-matrix and domain-specific interpretation, without assuming that agreement among overlapping reviews represented independent confirmation. The evidence indicates that spontaneous reports remain central to early signal generation and narrative characterization, whereas routinely collected healthcare data can provide denominators, longitudinal context, and comparative analyses when exposure, outcome, confounding, and data-provenance limitations are addressed. Machine learning and natural-language processing may improve extraction, prioritization, and evidence integration, but external validation, benchmark consistency, model maintenance, and demonstrated workflow benefit remain uneven. Active surveillance, registries, and distributed networks offer more structured ascertainment, although implementation maturity varies by setting, population, and governance infrastructure. Across domains, evidence converges on complementarity rather than replacement. Conflicts commonly reflect different surveillance tasks, data structures, validation standards, and decision contexts. Integrated pharmacovigilance should therefore preserve source provenance and maintain explicit boundaries among detection, validation, causal assessment, implementation, and decision use.