Background: Longitudinal, within-subject assessments are essential for tracking the illness course of borderline personality disorder (BPD), marked by clinical instability. Quantitative language analysis has proven valuable for in-situ psychopathology studies. Here, we examine lexical features in repeated clinical interviews in relation to BPD treatment response.
Methods: Ten adolescents BPD-spectrum patients in a residential treatment program, with at least 3 recorded and transcribed clinical interviews, are included in this report. The Borderline Evaluation of Severity over Time (BEST) scale served as the outcome measure. 82 lexicosemantic features were extracted via Linguistic Inquiry and Word Count (LIWC). Within-subjects associations of these features to BEST outcomes were assessed via repeated measures correlation. To counter feature collinearity, the features retained were those demonstrating the strongest associations within each hierarchical cluster.
Results: Subjects averaged 4.5 visits. The BEST total score demonstrated a repeated measures correlation r=-0.569 (95%CI: -0.76, -0.29) with visit count, suggesting improvements in overall BPD symptom severity during treatment. A two-cluster solution was identified based on the Silhouette Score. The top feature in one hierarchical cluster, the LIWC “Work” feature, reflecting the frequency of words relating to personal concerns about labor and school, demonstrated a negative correlation r=-0.509 (95%CI: -0.76, -0.12) against the BEST total. In the other cluster, the LIWC “Sad” feature, reflecting the frequency of words with a sad theme, demonstrated a positive correlation r=0.471 (95%CI: 0.07, 0.74).
Conclusions: The findings suggest that quantitative lexical features captured during a naturalistic, clinical encounter can be useful, objective markers of BPD symptom improvement.