> ## Content Index
> Fetch the complete content index at: https://www.fdaweb.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Researchers Urge FDA to Use AI for Psychiatric Drug Safety Monitoring
- URL: https://www.fdaweb.com/researchers-urge-fda-to-use-ai-for-psychiatric-drug-safety-monitoring/
- Published: 2026-11-17T22:58:50.000Z
- Updated: 2026-11-17T22:58:50.000Z
- Author: David McFarland
- Tags: Drugs

FDA's existing postmarketing surveillance systems may systematically miss important adverse effects of psychiatric drugs, and artificial intelligence could help identify safety signals that conventional reporting methods fail to capture, according to a new scientific review.

The [10/13 article in *Cureus*](https://doi.org/10.7759/cureus.117858?ref=fdaweb.com) recommends that FDA expand the Sentinel System's use of electronic health records to identify psychiatric adverse events that insurance claims and spontaneous reports frequently miss. The authors also call for applying FDA's risk-based AI credibility framework to pharmacovigilance tools.

The review examines FDA's Adverse Event Reporting System (FAERS), MedWatch, Sentinel and Risk Evaluation and Mitigation Strategies (REMS), focusing on antidepressants, antipsychotics, mood stabilizers and newer therapies such as esketamine.

The researchers say psychiatric adverse effects are particularly difficult to detect because symptoms such as suicidal ideation, anxiety, cognitive changes and sleep disturbances can overlap with the conditions being treated. Other effects, including tardive dyskinesia and metabolic disorders, may develop years after treatment begins.

Citing a systematic review spanning multiple therapeutic areas, the authors note a median adverse-event underreporting rate of 94%, although they caution that the estimate is not specific to psychiatric drugs.

The review highlights an analysis of 962 esketamine cases reported to FAERS during the drug's first year on the market. Disproportionality signals included dissociation, sedation and suicidal ideation, but the authors caution that REMS-related reporting, publicity and patients' underlying depression complicate interpretation.

Artificial intelligence, particularly natural language processing, could help identify adverse events recorded in clinical notes but never submitted to FDA. One cited study involving approximately 264,000 psychiatric patients achieved an average F-measure and accuracy of 0.83 in identifying medication-related adverse effects.

However, the authors warn that most AI pharmacovigilance studies remain retrospective proofs of concept, with limited validation in actual regulatory workflows.

They recommend prospective testing of AI tools, better integration of FAERS with electronic health records and claims data, and greater attention to reporting biases.

The review also points to FDA's 1/2025 draft guidance on AI supporting regulatory decisions, which proposes evaluating model credibility according to its intended use and the consequences of errors.

The researchers emphasize that AI should assist in identifying and prioritizing potential safety signals, while causality assessments and regulatory decisions remain the responsibility of clinicians and regulators.