AI-Clinician Treatment Concordance Linked to Better Psychiatry Outcomes
Key Clinical Summary
- A retrospective cohort study included 39,630 adult psychiatric outpatients treated across Mayo Clinic sites from January 2019 through December 2025.
- Greater agreement between artificial intelligence recommendations and clinician decisions was associated with lower mortality, suicidality, hospitalization, nonadherence, and healthcare costs.
- Because the observational findings establish associations rather than causation, investigators called for prospective testing and implementation research.
Greater concordance between clinical decisions and recommendations from an artificial intelligence clinical decision support system (AI-CDSS) was associated with better 12-month outcomes among psychiatric outpatients, according to a poster presented at the 2026 Psych Congress in New Orleans, Louisiana.
Study Findings
The cohort study randomly sampled 39,630 adults receiving outpatient care for at least 1 psychiatric diagnosis at Mayo Clinic locations in Minnesota, Arizona, Florida, Wisconsin, and Iowa. The AI-CDSS system independently generated treatment recommendations at clinical decision points using retrieval-augmented generation and was blinded to clinician decisions from the current visit.
The primary concordance measure, the F1 score, was calculable for 30,673 patients. Among these patients, 61.1% were female, the mean age was 47.4 years, 69.7% had at least 2 psychiatric diagnoses, and 80.5% had at least 1 medical comorbidity.
Higher concordance was associated with lower odds of all-cause mortality at 12 months (odds ratio [OR], 0.59; 95% confidence interval [CI], 0.42-0.82; p = .002). It was also associated with lower odds of new suicidal thoughts or behaviors (ORs =0.37-0.40 [95%CIs=0.25-0.54], p < 0.001) psychiatric hospitalizations or emergency department visits (ORs=0.49-0.65 [95%CI=0.38-0.82], p < 0.001), appointment nonattendance (OR=0.60 [95%CI=0.48-0.74], p = 0.002), and treatment nonadherence (OR=0.74 [95%CI=0.61-0.90], p = 0.004).
Higher concordance was additionally associated with $1593.36 lower psychiatric costs and $11,735.69 lower medical costs. Findings using accuracy and prevalence- and bias-adjusted kappa were largely consistent.
Clinical Implications
The results suggest that agreement between artificial intelligence recommendations and clinician decisions may identify care patterns associated with meaningful clinical and economic outcomes. However, the study does not establish that following the system’s recommendations caused those outcomes. Patient characteristics, clinician judgment, and other unmeasured factors may have contributed to the observed associations.
Expert Commentary
The investigators concluded that the findings “suggest clinical value of this AI-CDSS, warranting prospective testing and implementation.”
The findings support prospective evaluation of artificial intelligence-assisted psychiatric decision-making. Controlled studies are needed to determine whether implementing the system improves outcomes across diverse patients and clinical settings.
Reference
Adam T. 12-Month outcomes of treatment concordance between artificial intelligence-based recommendations and real-world clinical decisions: Mortality, suicidality, hospitalization, nonadherence, and cost. Poster presented at: Psych Congress, 15-21, 2026, New Orleans, Louisiana.
