Classification and discrimination of emotion dysregulation disorders using machine learning.

Attention-deficit/hyperactivity (ADHD), bipolar (BD) and borderline personality (BPD) disorders are severe psychiatric illnesses often presenting with overlapping emotion dysregulation symptoms. To date, it is unknown whether these disorders share a continuum on emotion dysregulation. To address this, we employed machine learning (ML) algorithms on psychological and environmental features from 232 adults to predict and discriminate these disorders. We recruited 92 patients diagnosed with either ADHD, BD, or BPD. Considering heritability, we recruited 67 genetically unrelated ADHD, BD and BPD offspring to determine how offspring overlap with or discriminate from their parent disorder. Seventy-three healthy age-matched healthy controls (HC) were also recruited. Features included clinical dimensions associated with emotion dysregulation (e.g., impulsivity, mania, rumination) as well as childhood trauma and parental bonding. BPD showed the greatest independence, discriminating very strongly from HC, strongly from ADHD and moderately from BD and BPD offspring, with mania as most predictive overall. ADHD discriminated very strongly from HC, strongly from BPD, moderately from BD, and weakly from ADHD offspring, with impulsivity as most predictive. BD discriminated moderately from ADHD, BPD and BD offspring, and weakly from HC, with maternal bonding most predictive. HC discriminated weakly from all offspring groups, with perseverance as most predictive. Results suggest ADHD, BD and BPD are independent psychiatric constructs reliably classified by ML via emotion regulation traits and environmental factors, whereas their offspring more closely resemble HC. We thus show ML may allow for cost-effective classification and prediction of emotion dysregulation disorders (ADHD, BD, BPD) relative to HC.