Artificial Intelligence in Depression Medication Enhancement (AIDME): A Cluster Randomized Trial of a Deep Learning Enabled Clinical Decision Support System for Personalized Depression Treatment Selection and Management
David Benrimoh,Kate Whitmore,Maud Richard,Grace Golden,Kelly Perlman,Sara Jalali,Timothy Friesen,Youcef Barkat,Joseph Mehltretter,Robert Fratila,Caitrin Armstrong,Sonia Israel,Christina Popescu,Jordan F. Karp,Sagar V. Parikh,Shirin Golchi,Erica Moodie,Junwei Shen,Anthony Gifuni,Manuela Ferrari,Mamta Sapra,Stefan Kloiber,George F Pinard,Boadie W. Dunlop,Karl Looper,Mohini Ranganathan,Martin Enault,Serge Beaulieu,Soham Rej,Fanny Hersson-Edery,Warren Steiner,Alexandra Anacleto,Sabrina Qassim,Rebecca McGuire-Snieckus,Howard C. Margolese
DOI: https://doi.org/10.1101/2024.06.13.24308884
2024-09-26
Abstract:Background: There has been increasing interest in the use of Artificial Intelligence (AI)-enabled clinical decision support systems (CDSS) for the personalization of major depressive disorder (MDD) treatment selection and management, but clinical studies are lacking. We tested whether a CDSS that combines an AI which predicts remission probabilities for individual antidepressants and a clinical algorithm based on treatment can improve MDD outcomes.
Methods: This was a multicenter, cluster randomized, patient-and-rater blinded and clinician-partially-blinded, active-controlled trial that recruited outpatient adults with moderate or greater severity MDD. All patients had access to a patient portal to complete questionnaires. Clinicians in the active group had access to the CDSS; clinicians in the active-control group received patient questionnaires; both groups received guideline training. Primary outcome was remission (<11 points on the Montgomery Asberg Depression Rating Scale (MADRS)) at study exit.
Results: 47 clinicians were recruited at 9 sites. Of 74 eligible patients, 61 patients completed a post-baseline MADRS and were analyzed. There were no differences in baseline MADRS (p = 0.153). There were more remitters in the active (n= 12, 28.6%) than in the active-control (0%) group (p = 0.012, Fisher exact). Of three serious adverse events, none were caused by the CDSS. Speed of improvement was higher in the Active than the Control group (1.26 vs. 0.37, p = 0.03).
Conclusions: While limited by sample size and the lack of primary care clinicians, these results demonstrate preliminary evidence that longitudinal use of an AI-CDSS can improve outcomes in moderate and greater severity MDD.