UAB-led project targets repeat mental health emergency visits with explainable AI

Abdulaziz Ahmed, Ph.D., at UAB will lead a five-year, nearly $4 million National Institute of Mental Health project with Vanderbilt University Medical Center to develop AI-MERRA, an explainable AI system for mental health-related emergency department returns. Nationwide, about one in eight emergency department visits involves a mental health or substance use diagnosis, and at UAB these visits rank fourth by volume but have the highest 30-day return rate. In Ahmed’s preliminary cohort, 27 percent of visits were followed by another ED visit within 30 days.

The tool will combine LLMs that extract risk factors from clinical notes with structured electronic health record data and explainable machine-learning models. It is intended to identify high-risk patients before discharge, forecast return risk from 24 hours to 90 days, and give clinicians brief, patient-specific explanations tied to factors such as symptoms, medication adherence, prior utilization, follow-up care, housing and transportation.

The grant builds on work published in JAMIA Open in June 2026 using 42,464 de-identified ED visits from 27,904 patients. A review of 100 explanations tested for hallucination. Ninety-nine were fully aligned with expert review; the remaining explanation contained a one-point numerical discrepancy, reporting 92 instead of 93. The NIH-funded project will expand validation across UAB and Vanderbilt while assessing hallucination, bias, fairness and workflow usability.