Drawing the borderline: Predicting treatment outcomes in patients with borderline personality disorder

Background: A routinely collected big data set was analyzed to determine the effectiveness of naturalistic inpatient treatment and to identify predictors of treatment outcome and discontinuation. Methods: The sample included 878 patients with borderline personality disorder who received non-manualized dialectic behavioral therapy in a psychosomatic clinic. Effect sizes (Hedge's g) were calculated to determine effectiveness. A bootstrap-enhanced regularized regression with 91 potential predictors was used to identify stable predictors of residualized symptomand functional change and treatment discontinuation. Results were validated in a holdout sample and repeated cross validation. Results: Effect sizes were small to medium (g = 0.28-0.51). Positive symptom-related outcome was predicted by low affect regulation skills and no previous outpatient psychotherapy. Lower age, absence of work disability, high emotional and physical role limitations and low bodily pain were associated with greater improvement in functional outcome. Higher education and comorbid recurrent depressive disorder were the main predictors of treatment completion. The predictive quality of the models varied, with the best being found for symptom related outcome (R-2 = 18%). Conclusion: While the exploratory process of variable selection replicates previous findings, the validation results suggest that tailoring treatment to the individual patient might not be based solely on sociodemographic, clinical and psychological baseline data.