AI Agents Enhance Mental Health Medication Information Seeking.

Patients frequently seek medication information online, but psychiatric drug safety data is fragmented between authoritative yet abstract regulatory records and experience-rich but unvalidated patient narratives. This paper introduces a multi-agent framework, augmented with a knowledge graph, to integrate these disparate sources while preserving the provenance of each claim. The system unifies data from 466,525 Reddit posts, 60,782 WebMD reviews, and two decades of U.S. FDA Adverse Event Reporting System records for nine antidepressants. A large language model (LLM) entity recognition pipeline achieved high F1 scores for identifying medications and conditions. The study found that community platforms often show high concordance with each other but less with regulatory reports, suggesting patient-generated data offers an independent safety signal. Notably, many adverse events for sertraline appeared in community sources hundreds of days before their corresponding FDA dates. By grounding the Neo4j knowledge graph in standard medical vocabularies, the framework ensures every claim is traceable and distinguishes regulatory facts from patient experiences, paving the way for more auditable psychiatric medication information.