A learning pharmacovigilance system for frail patients: integrating real-world data, artificial intelligence, clinical pharmacology and medicines optimization to prevent medication-related harm

INTRODUCTION: Frail patients face changing medication-related risks that require repeated clinical reassessment. We propose a learning pharmacovigilance framework that connects safety evidence with medication decisions and outcomes. AREAS COVERED: We searched PubMed/MEDLINE and Scopus on 8 September 2026 for publications from 2015-2026; we searched ClinicalTrials.gov on the same date. We purposively selected articles for this narrative review. Six layers connect pharmacological knowledge, longitudinal data, artificial intelligence, clinical translation, outcome capture, and governed learning. We distinguish prognosis, clinical causal attribution, and treatment-effect estimation, and examine frailty assessment, prescribing-optimization trials, large language models, and regulatory requirements for updates. Clinical translation assigns responsibility for medication review, decisions, and follow-up. Outcome assessment includes withdrawal events, while implementation requires workload assessment and silent prospective testing. EXPERT OPINION: Four randomized trials of prescribing optimization did not demonstrate significant benefit on their primary clinical outcomes; the integrated framework remains clinically unvalidated. Comparative evaluation should separate the contribution of structured clinical review from that of outcome-linked updating; over the next five years, we expect local model control and version traceability to become priorities for regulated medication-safety applications.