Suicide Risk in Borderline Personality Disorder: a Machine Learning Tool based on Clinical and MRI Data

Abstract

Borderline Personality Disorder (BPD) is a complex mental condition. Individuals with BPD have an average of three lifetime suicide attempts, and 10% of them die by suicide. Understanding risk factors linked to suicidal behaviors is crucial for effective intervention strategies. In recent years, machine learning (ML) approaches for predicting suicide risk in persons with mental disorders have been developed, but a reliable, BPD-specific tool is lacking. In this work, we developed DRAMA-BPD (Detecting Risk factors for suicide Attempts with Machine learning Approaches in Borderline Personality Disorder), a second-opinion tool to assess suicide risk in individuals with BPD. DRAMA-BPD, built upon a Support Vector Machine (SVM) classifier, is trained on the CLIMAMITHE (CLM) dataset, which encompasses sociodemographic, clinical, emotional assessments, and MRI data. Feature selection revealed that 6 out of the 7 most important features are MRI-derived, and a comprehensive review was conducted to ensure consistency with existing scientific literature. The classifier achieved an overall Area Under the Curve (AUC) of 0.73, Precision (P) of 0.75, Recall (R) of 0.70, and F1-score of 0.72. Tests were conducted on the independent SUDMEX_CONN dataset, yielding an AUC of 0.59, P of 0.46, R of 0.92, and F1 of 0.62. While there is a significant imbalance between Precision and Recall, these results demonstrate the potential utility of the proposed model.