Risk calculation circuit abnormalities plus psychosocial risk variables predict problematic substance use in youth with externalizing disorders

Abstract Prior research has identified brain regions associated with problematic substance use in youth, yet it remains unclear how neural processes during decision-making contribute to later drug use. Moreover, few studies have integrated psychosocial and environmental risk factors into predictive frameworks. This study investigated whether brain activation during risky decision-making in drug-naïve, high-risk children predicts problematic substance use during adolescence. Youth ( n = 95; 64 male, mean baseline age=11.7 years) with externalizing disorders completed the Balloon Analogue Risk Task (BART) during functional MRI. Activation contrasts from six regions of interest, identified using a regularization-based feature selection method, were incorporated into cost-sensitive logistic regression models along with psychosocial and environmental variables, including family history of substance use, parental monitoring, and violence exposure. Models were adjusted for age at conversion to drug use, sex assigned at birth, and maternal education. Psychosocial-only factors showed fair predictive accuracy (AUC = 0.76; accuracy= 0.74) with good specificity and fair sensitivity. Neural activation–only models showed poor predictive accuracy (AUCs = 0.60–0.67; accuracy = 55–78%) with good specificity but limited sensitivity. Incorporating both psychosocial and neural factors substantially improved model performance (AUCs = 0.83–0.86; accuracy up to 82%), with fair sensitivity and good specificity in the adjusted models. These findings suggest that neural activity in regions involving risk evaluation, reward response, and sensory integration, together with relevant psychosocial factors predicts later problematic substance use, emphasizing the value of multidimensional models for early identification youth at elevated risk.