1 - (SYM 127) Personalized Mechanistic Network Models of Borderline Personality Disorder Symptoms

Background: Borderline personality disorder (BPD) is diagnosed in the DSM-5 based on the presence of at least five of nine symptoms (American Psychiatric Association, 2013). Dimensional approaches to BPD diagnosis instead assess for BPD based on elevations in several higher-order personality traits: neuroticism, antagonism, and disinhibition, considered putative mechanisms of BPD. Previous work using each of these diagnostic approaches has found significant heterogeneity in both DSM-5 symptom presentations and dimensional personality profiles among BPD patients, suggesting optimal treatment may differ across individuals. We explored whether BPD patients display unique patterns of associations between personality domains and symptoms using idiographic network models, and examined whether these networks can be organized into distinct subgroups based on shared associations.

Methods: Participants (N = 98; Mage = 28.13, 76% female, 80% White) were treatment-seeking adults diagnosed with BPD based on DSM-5 criteria who received BPD Compass (Sauer-Zavala et al., 2023), an 18-session cognitive-behavioral intervention, as part of a randomized controlled trial. Participants completed self-report measures of BPD symptom severity and BPD-relevant personality dimensions prior to each session, and following treatment completion. We generated personalized longitudinal network models of BPD symptoms and personality traits containing lagged and contemporaneous associations using GIMME, an analytic framework which yields individual networks (Gates & Molenaar, 2012). We tested whether any paths were shared across more than half the sample to assess heterogeneity, and whether any subgroups of networks defined by shared associations emerged.

Results: 53 networks converged normally with acceptable to good fit (Χ² = 39.27 - 76.40; RMSEA = 0 - .16; CFI = .95 – 1), although 7 networks had poor RMSEA values (i.e., > .08). Across models, no paths were shared by a majority of networks. We identified 8 subgroups with >1 participant in them; however, in only one subgroup was a significant association present among more than half of its members.

Discussion: Individual networks and subgroups demonstrated sweeping heterogeneity, suggesting that grouping diverse individuals within the same diagnostic category may undermine treatment effectiveness. It may, therefore, be necessary to move away from nomothetically tested, one-size-fits-all treatment approaches in favor of treatment selection based on idiographic, functional assessment.