Predicting personalized process-outcome associations in psychotherapy using machine learning approaches—A demonstration

Objective: Personalized treatment methods have shown great promise in efficacy studies across many fields of medicine and mental health. Little is known, however, about their utility in process-outcome research. This study is the first to apply personalized treatment methods in the field of process-outcome research, as demonstrated based on the alliance-outcome association. Method: Using a sample of 741 patients, individual regressions were fitted to estimate within-patient effects of the alliance-outcome association. The Boruta algorithm was used to identify patient intake characteristics that moderate the within-patient alliance-outcome association. The nearest neighbor approach was used to identify patients whose relevant pretreatment characteristics were similar to those of a target patient. The alliance-outcome associations of the most similar patients were subsequently used to predict the alliance-outcome association of the target patient. Results: Irrespective of the number of selected nearest neighbors, the correlation between the observed and predicted alliance-outcome associations was low and insignificant. According to the true error of the prediction, the demonstrated approach was unable to improve predictions made with a simple comparison model. Conclusion: The study demonstrated the application of personalized treatment methods in process-outcome research and opens many new paths for future research.