Predicting the outcome of psychological treatments for borderline personality disorder and posttraumatic stress disorder: a machine learning approach to predict long-term outcome of Narrative Exposure Therapy vs. Dialectical Behavioral Therapy based

Background: A comorbidity between Borderline Personality Disorder (BPD) and Posttraumatic Stress Disorder (PTSD) is common, severely disabling, and hard to treat. The choice of an optimal psychotherapy based on patient characteristics remains challenging. Objective: This study develops models to predict the outcome of two psychotherapies for comorbid BPD and PTSD. Method: Data from two trials comparing Narrative Exposure Therapy (NET, N = 40) with Dialectical Behavior Therapy-based treatment (DBT-bt, N = 40) was analysed. A cross-validated genetic algorithm was used to detect baseline predictors of change in PTSD symptoms. Results: In the NET group higher education, more baseline PTSD symptoms, more traumatic experiences, fewer baseline BPD symptoms, and not taking antipsychotic medication predicted better treatment outcome. This model (RMSE = 8.98) outperformed the prediction of PTSD symptom reduction with baseline PTSD symptoms alone (RMSE = 10.07) or with all available predictor variables (RMSE = 12.97). Only more baseline PTSD symptoms were selected to predict a better treatment outcome after DBT-bt. This model (RMSE = 9.41) outperformed the prediction of change in PTSD symptoms with all available predictor variables (RMSE = 14.43). Conclusion: Differences in treatment outcome between NET and DBT-bt may be predictable at baseline, to identify which one of both treatments may be most beneficial for individual patients. The small sample size may restrict the generalizability of the results.