JMIR Publications

ABSTRACT

Background:

Pharmacovigilance aims to protect patient safety by identifying and managing adverse events associated with medicinal products. Assessing the causality of adverse events in relation to drug exposure is central to pharmacovigilance at both the individual case and population levels, but it has become increasingly challenging due to the growing volume and complexity of safety data. Artificial intelligence (AI) and related automation technologies are increasingly proposed to support causality assessment. However, there is limited research on how these methods are used, what information and quality requirements they entail, or how associated risks are addressed.

Objective:

This scoping review examines the use of AI-based methods for assessing causal relationships between medicinal products and adverse events in pharmacovigilance, at both individual case and population levels. It maps current applications and proposals, and synthesizes associated data and information quality requirements, alongside risks and their management. The review also compares these requirements and risks across levels and explores complementarities between AI-supported processes. Additionally, it characterizes the types of AI techniques and automation tools used within causality assessment workflows, and summarizes reported governance mechanisms, including human oversight, explainability, monitoring, and model maintenance.

Methods:

Eligible sources will describe or propose AI-based approaches, including data-driven models (e.g., machine learning, natural language processing, knowledge graphs, and causal inference) and knowledge- or rule-based systems that implement causal assessment logic. These approaches must be applied to relationships between medicinal products and adverse events at either the individual or population level. Automation tools, such as robotic process automation, will be included only when they are explicitly integrated with AI-based methods for causality assessment. Searches will be conducted in PubMed, the Web of Science Core Collection, ProQuest, EBSCOhost databases, and Ichushi-Web, with inclusion restricted to English and Japanese sources. Records and full texts will be screened independently by two reviewers, and disagreements will be resolved by a third reviewer. Data will be charted according to use cases, workflow steps, information inputs, data quality indicators, model characteristics, performance metrics, human oversight, governance mechanisms, and identified risks or failure modes. Findings will be synthesized descriptively and mapped across individual- and population-level applications.

Results:

This protocol was registered on the Open Science Framework platform on December 23, 2025. This review will provide a structured overview of the current use, proposals, and discussions surrounding the adoption of AI-based methods for causality assessment in pharmacovigilance. It will also clarify the information requirements, quality considerations, and risk management approaches reported in the literature.

Conclusions:

The results of this review are expected to inform methodological development, practical implementation, and the governance of AI-supported causality assessment.