No evidence-based therapy for borderline personality disorder (BPD) exhibits a clear superiority. However, BPD is highly heterogeneous, and different patients may specifically benefit from the interventions of a particular treatment.
From a randomized trial comparing a year of dialectical behavior therapy (DBT) to general psychiatric management (GPM) for BPD, long-term (2-year-post) outcome data and patient baseline variables (
n= 156) were used to examine individual and combined patient-level moderators of differential treatment response. A two-step bootstrapped and partially cross-validated moderator identification process was employed for 20 baseline variables. For identified moderators, 10-fold bootstrapped cross-validated models estimated response to each therapy, and long-term outcomes were compared for patients randomized to their model-predicted optimal v.non-optimal treatment. Results
Significant moderators surviving the two-step process included psychiatric symptom severity, BPD impulsivity symptoms (both GPM > DBT), dependent personality traits, childhood emotional abuse, and social adjustment (all DBT > GPM). Patients randomized to their model-predicted optimal treatment had significantly better long-term outcomes (
d= 0.36, p= 0.028), especially if the model had a relatively stronger (top 60%) prediction for that patient ( d= 0.61, p= 0.004). Among patients with a stronger prediction, this advantage held even when applying a conservative statistical check ( d= 0.46, p= 0.043). Conclusions
Patient characteristics influence the degree to which they respond to two treatments for BPD. Combining information from multiple moderators may help inform providers and patients as to which treatment is the most likely to lead to long-term symptom relief. Further research on personalized medicine in BPD is needed.