.
VigiBase, the WHO global database of adverse event reports for medicines and vaccines, receives information on suspected adverse effects of medicinal products from countries, regions and territories that are members of the WHO Programme for International Drug Monitoring. The database serves as a global resource for pharmacovigilance signal management and scientific development. Its initial establishment through the programme in 1968 has also contributed to the international harmonisation of adverse event report data. As of 31 December 2024, the database includes over 40 million individual case reports coded in standard terminologies (WHODrug and MedDRA®). These reports, of which 80
Many high-stakes artificial intelligence (AI) applications target low-prevalence events, where apparent accuracy can conceal limited real-world value. Relevant AI models range from expert-defined rules and traditional machine learning to generative large language models (LLMs) constrained for classification. As the effort and expertise required to develop modern AI decrease, there is a risk that organizations devote too little time to understanding their limitations and sources of error. We outline key dimensions for critical appraisal of AI in rare-event recognition, including problem framing and test set design, prevalence-aware statistical evaluation, robustness assessment, and integration into human workflows. In addition, we propose an approach to structured case-level examination (SCLE), to complement statistical performance evaluation, and a set of considerations to guide procurement or development of AI models for rare-event recognition. We instantiate the framework in pharmacovigilance, drawing on three studies: rule-based retrieval of pregnancy-related reports, duplicate detection combining machine learning with probabilistic record linkage, and automated redaction of person names using an LLM. We highlight pitfalls specific to the rare-event setting including optimism from unrealistic class balance and lack of difficult positive controls in test sets—and show how cost-sensitive targets align model performance with operational value. While grounded in pharmacovigilance practice, the principles generalize to domains where positives are scarce, and error costs may be asymmetric.
Artificial intelligence (AI) is increasingly being explored to support pharmacovigilance activities including processes involving individual case safety reports. These technologies may improve efficiency and enable new analytical capabilities, but they also raise important questions regarding the responsible use of these systems in contexts that directly support regulatory decision-making and patient safety. In response to these opportunities and challenges, the Council for International Organizations of Medical Sciences (CIOMS) set up Working Group XIV on Artificial Intelligence in Pharmacovigilance to establish and promote principles and guidance for this area. The resulting CIOMS report was published in December 2025 and presents seven guiding principles: risk-based approach, human oversight, validity and robustness, transparency, data privacy, fairness and equity, and governance and accountability. To maximise stakeholder awareness and understanding, this article provides an overview of the CIOMS XIV guidance and illustrates how these principles may be applied in practice through selected use cases. The report aims to support stakeholders in developing and deploying AI systems in a manner that maintains scientific rigor, transparency, and public trust while enabling innovation to support safe use of medicines worldwide.
Medicinal product information documents (PIDs) detail clinical characteristics and instructions for monitoring, preventing, and mitigating adverse drug reactions (ADRs). They vary across countries, but there have been no recent international comparisons. We have therefore quantified and compared the completeness of the information given in drug labelling from different countries. From the websites of 35 regulatory agencies, we retrieved the PIDs of medicinal products that were involved in signals communicated by regulators in 2014–2019. We developed a data extraction framework based on the Dose-relatedness, Time course, Susceptibility (DoTS) clinical classification of ADRs and used its implications for prevention and mitigation to score the completeness of related instructions in PIDs. To extract and classify monitoring instructions, we used a modified Systematic Instructions for Monitoring (SIM) method. PIDs had sufficiently complete instructions for prevention when the DoTS score was ≥ 5/12, and sufficiently complete monitoring instructions when the SIM score was ≥ 3/6. We used proportions of PIDs having a score ≥ 1 to determine the relative availability of clinical characteristics or instructions in a country, compared with all other countries. We quantified their pairwise disagreements using Jaccard’s distance and identified clusters with similar patterns of completeness using agglomerative hierarchical clustering. PIDs were available on the websites of 18 of 35 regulatory agencies. They concerned 364 distinct medicinal products, which were involved in 627 signals. Across all countries, the instructions for prevention or mitigation met sufficient completeness for a median of 30
Adverse event reports of glomerular diseases (GDs) following coronavirus disease 2019 (COVID-19) vaccination collected nationally have been shared to VigiBase, the global database of adverse event reports for medicines and vaccines. From literature reports, different GDs following COVID-19 vaccination have different time-to-onsets (TTOs), highlighting the potential value of TTO in suggesting underlying disease mechanisms and supporting its relevance in signal validation prior to in-depth signal assessment. The aim of this study was to characterise the TTO in reports of GDs in association with COVID-19 vaccines inVigiBase. The analysis focuses on exploring the value of TTO for signal validation, an early step preceding signal assessment. Reports for four different GDs in association with COVID-19 vaccines were extracted from VigiBase on 3 April, 2023. The reports were de-duplicated and TTO information was manually verified. Demographics and clinical characteristics were summarised per GD. The median TTO was compared between GDs and by disease status (relapse and new onset cases) within each GD. In the extracted data, 245 unique cases shared to VigiBase and containing information about both TTO and disease status for the four selected GDs in association with COVID-19 vaccines were identified. Reports for IgA nephropathy represented the majority (49