Douglas Mental Health University Institute (McGill University) — multicenter trial across 9 sites

Treatment selection for major depressive disorder (MDD) is highly variable, with many patients requiring multiple medication trials before achieving remission. Clinicians lacked personalized, data-driven tools to predict which antidepressant would most likely lead to remission for an individual patient. Clinical studies evaluating AI-enabled decision support systems for depression treatment were largely absent.

A deep-learning-enabled clinical decision support system (CDSS) developed by Aifred Health was deployed across 9 outpatient psychiatric sites in a cluster randomized trial. The CDSS predicted individual remission probabilities for specific antidepressants and incorporated a clinical management algorithm. Patients had access to a portal to complete questionnaires, and active-group clinicians received CDSS recommendations alongside guideline training.

Remission (<11 on MADRS) was achieved by 28.6% of patients in the active CDSS group (n=12) versus 0% in the active-control group (P=.012, Fisher's exact). Speed of improvement was significantly higher in the active group (1.26 vs 0.37, P=.03). No serious adverse events were attributable to the CDSS, demonstrating preliminary evidence that longitudinal AI-CDSS use can improve moderate-to-severe MDD outcomes.

• Deep learning–based treatment prediction can meaningfully shift remission rates when integrated into clinical workflows, not just used as a one-time recommendation tool.
• Active-control design (both arms received guideline training and patient questionnaires) isolates the incremental benefit of AI personalization beyond standard-of-care improvement.
• Multicenter cluster randomization is feasible for AI-CDSS trials, though sample size remains a key limitation for definitive conclusions.