Determining the key childhood and adolescent risk factors for future BPD symptoms using regularized regression: comparison to depression and conduct disorder

Objective Research has yielded factors considered critical to risk for borderline personality disorder (BPD). Yet, these factors overlap and are relevant to other disorders, like depression and conduct disorder (CD). Regularized regression, a machine learning approach, was developed to allow identification of the most important variables in large datasets with correlated predictors. We aimed to identify critical predictors of BPD symptoms in late adolescence (ages 16-18) and determine the specificity of factors to BPD versus disorders with putatively similar etiology. Method We used a prospective longitudinal dataset (n = 2,450) of adolescent girls assessed on a range of clinical, psychosocial, and demographic factors, highlighted by previous research on BPD. Predictors were grouped by developmental periods: late childhood (8-10) and early (11-13) and mid-adolescence (14-15), yielding 128 variables from 41 constructs. The same variables were used in models predicting depression and CD symptoms. Results The best-fitting model for BPD symptoms included 19 predictors and explained 33.2% of the variance. Five constructs - depressive and anxiety symptoms, self-control, harsh punishment, and poor social and school functioning - accounted for most of the variance explained. BPD was differentiated from CD by greater problems with mood and anxiety in BPD and differences in parenting risk factors. Whereas the biggest parenting risk for BPD was a punitive style of parenting, CD was predicted by both punitive and disengaged styles. BPD was differentiated from MDD by greater social problems and poor behavioral control in BPD. Conclusions The best predictors of BPD symptoms in adolescence are features suggesting complex comorbidity, affective activation, and problems with self-control. Though some risk factors were non-specific (e.g., inattention), the disorders were distinguished in clinically significant ways.

Automated summary

Research used machine learning to explore key predictors of borderline personality disorder (BPD) in adolescents, finding associations with other disorders like depression and conduct disorder. The model highlighted complex comorbidity, affective activation, and self-control issues as the best predictors of BPD symptoms in adolescence.