Biosimilar stewardship at the interface of disproportionality methods and AI integration: current evidence and potential strategies

INTRODUCTION Biosimilar stewardship includes safety, appropriate use, and value. However, it faces challenges due to shortages in disproportionality methods. Artificial intelligence (AI) tools have been emerged as practical 'intelligence augmentation' strategies that strengthen, rather than replace, traditional pharmacovigilance (PV) signal management.

AREAS COVERED We conducted a narrative review with evidence mapping (Jan 2010-Nov 2025) across PubMed/MEDLINE, Embase, Scopus, Web of Science, Cochrane, citation-chasing, and targeted grey/first-party sources (regulators/HTA portals, PV centers, NHS dashboards, technical documentation). We prioritized biosimilar-specific pharmacovigilance, switching/implementation, and value-tracking evidence; general PV/AI literature was used primarily to describe analytic methods and governance where it directly informs biosimilar-specific problems A study-selection summary is provided for transparency. To improve interpretability for readers, we also include brief 'deployment vignettes' describing how key AI-enabled functions are operationally implemented in routine pharmacovigilance workflows.

EXPERT OPINION Medium-term priorities include federated, KPI-linked analytics and better external validation and transportability; near-term 'assist' deployments - such as narrative NLP, normalization, de-duplication, and multi-feature ranking under human-in-the-loop governance - can quantifiably improve case completeness, traceability, and triage. To achieve reliable production use, persistent bottlenecks (class imbalance, vocabulary mapping) require clear monitoring and standard assessment procedures.