Abstract
Treatments for major depressive disorder (MDD) include antidepressant medications and evidence-based psychotherapies, which are approximately equally efficacious. Individual response to treatment, however, is variable, implying individual differences that could allow for prospective differential prediction of treatment response and personalized treatment recommendation. We used machine learning to develop predictor variables that combined demographic and clinical items from a randomized clinical trial. The variables predicted a meaningful proportion of variance in end-of-treatment depression severity for cognitive behavioral therapy (39.7%), escitalopram (32.1%), and duloxetine (67.7%), leading to a high accuracy in predicting remission (71%). Further, we used these variables to simulate treatment recommendation and found that patients who received their recommended treatment had significantly improved depression severity and remission likelihood. Finally, the prediction algorithms and treatment recommendation tool were externally validated in an independent sample. These results represent a highly promising, easily implemented, potential advance for personalized medicine in MDD treatment.