Connectome-Based Predictive Modeling of Concurrent and Prospective Substance Use in Adolescence

Abstract

Understanding the neural mechanisms of adolescent substance use is a critical public health issue, with direct implications for bolstering prevention and treatment strategies. Yet this effort is challenging because substance use is multi-faceted, commonly used brain network features are not optimized to capture both local and global aspects of intrinsic connectivity, and because substance use facets change over time. In this study, we operationalized adolescent substance use along three dimensions--intent, access, and family-developmental history--and trained predictive models of each facet using resting-state connectivity. Trait impulsivity, a known risk factor, was also examined. Using Baseline and 2 Year Follow-Up data from the ABCD Bids Community Collection (ABCC), we found that prediction was more successful at follow-up than baseline. At baseline, predictive accuracy was modest and intent to use substances was the most accurately predicted facet. Prediction accuracies at follow-up were much higher, with access and family-developmental history being better predicted, signaling a developmental shift in the brain-behavior mapping of substance use vulnerabilities. These findings suggest that the neurobiological correlates of substance use are dynamic across adolescence, possibly reflecting changing phenotype. More broadly, these results underscore the importance of modeling distinct substance use facets and accounting for developmental timing to understand risk trajectories, while contributing to a growing literature that shows early-developing individual differences are predictive of later outcomes.