Towards Outcome-Driven Patient Subgroups: A Machine Learning Analysis Across Six Depression Treatment Studies

•What is the primary question addressed by this study?Can machine learning models of depression treatment response be trained to generate treatment-relevant subgroups from a pool of patients with major depression?•What is the main finding of this study?Using the Differential Prototypes Neural Network to analyze six studies of antidepressant medication treatment (n = 5,438), we trained a model with the potential to improve population remission rates by 6.5% (15.6% relative improvement). The model generated three novel patient subgroups. These subgroups differed from each other in terms of symptoms (such as psychomotor agitation) and demographic characteristics.•What is the meaning of the finding?Machine learning models can be trained to generate treatment-relevant patient subgroups, thereby potentially improving the ability to personalize treatment for depression.