AI for Causality Assessment in Pharmacovigilance: Protocol for a Scoping Review

Abstract

Background Pharmacovigilance aims to protect patient safety by identifying and managing adverse events associated with pharmaceuticals. Determining the causality of these adverse events is central at both the individual case and population levels; however, it is increasingly challenging as the volume and complexity of safety data grow. Although AI and related technologies have been proposed to support causality assessment, limited research has examined how these methods are used, their information and quality requirements, or how associated risks are addressed. Objective This scoping review aims to determine the available evidence on AI-based methods for causality assessment in pharmacovigilance. The primary objective is to characterize how these methods are applied or proposed with a focus on their functional roles, reported data inputs and information needs, and associated risks. Secondary objectives include comparing applications at the individual case and population levels; describing the types of AI-based techniques and automation tools used in causality assessment workflows; and summarizing reported data quality considerations and governance mechanisms, including risk management approaches. Methods Sources describing or proposing AI-based approaches, including data-driven models (machine learning, natural language processing, knowledge graphs, and causal inference) and knowledge- or rule-based systems implementing causal assessment logic, will be eligible. Searches will be conducted in PubMed, Web of Science Core Collection, ProQuest, EBSCOhost, and Ichushi Web and will be restricted to English- and Japanese-language sources. Two reviewers will independently screen records and full-text articles, with disagreements resolved by a third reviewer. Data will be charted on use cases, information inputs, data quality dimensions, model characteristics, governance mechanisms, and identified risks. Synthesis will follow a reflexive thematic analysis approach and be reported in accordance with PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) guidelines informed by applicable PRISMA-S (Preferred Reporting Items for Systematic reviews and Meta-Analyses literature search extension) elements. Results This protocol was registered in the Open Science Framework platform on December 23, 2025. The registration was subsequently updated on May 19, 2026, to reflect an extension to the data collection period. A preliminary database search was conducted in December 2025, retrieving a total of 760 records, of which the preliminary title and abstract screening identified 196 (25.8%) articles for full-text review. Database searches are scheduled for July 2026. Data charting is scheduled for August 2026, and synthesis is scheduled for September 2026. Findings are expected to be submitted for publication by the end of December 2026. Conclusions This review is expected to provide a structured map of AI-based applications for causality assessment in pharmacovigilance, clarify reported information inputs and data quality dimensions, and synthesize risk management and governance approaches. The findings are expected to inform methodological development, practical implementation, and the governance of AI-supported causality assessment. Trial Registration Open Science Framework 10.17605/OSF.IO/QVF5C; https://osf.io/qvf5c/overview International Registered Report Identifier (IRRID) DERR1-10.2196/101691