AI-Empowered Multimodal Data Integration and Knowledge Discovery for Pharmacovigilance

Authors Li, Zhou, Yeh and others reported on medRxiv an evidence-aware framework for integrating pharmacovigilance data from the FDA Adverse Event Reporting System (FAERS), biomedical literature, clinical trials, and Mass General Brigham electronic health records. The framework combines large language model information extraction, ontology-driven semantic normalization, cross-source evidence assessment, statistical and machine learning ranking, and component-based adverse drug event relation inference. The study states that the constructed knowledge system contains 11,588 drug nodes, 44,843 component nodes, 3,275 adverse event nodes, 993 indication nodes, and more than 2.47 million relations, with each drug-adverse event relation assigned a confidence score. The authors state that, after incorporating this knowledge resource, the comprehensiveness, relevance, and overall quality of large language model drug-specific adverse event synthesis improved, and the precision and F1 score of clinical adverse event relation extraction also increased.