Abstract
BACKGROUND: Despite the rapid growth in gaming consumption and associated harms in adolescents, data-driven research to identify brain networks underlying problematic gaming remains limited. This study aimed to identify neural networks predictive of problematic-gaming severity in youth using connectome-based predictive modelling (CPM), a machine-learning approach that employs whole-brain functional connectivity data. METHODS: From the Adolescent Brain Cognitive Development study at the two-year follow-up, 1,036 participants (Mage = 12.0, 60.7% male) were studied. CPM with 10-fold cross-validation was applied to problematic-gaming scores and functional magnetic resonance imaging (fMRI) data collected during the performance of a reward-processing task. To determine generalizability, additional CPM analyses were performed using other task-based (e.g., those relevant to response inhibition, emotion regulation, and working memory) and resting-state fMRI data. RESULTS: CPM successfully predicted problematic-gaming scores (r = 0.12, p = 0.002). Predictive networks involved several connections within and between canonical networks implicated in visual processing (visual area 2 and visual association networks), cognitive control and executive functioning (frontoparietal and medial frontal networks), and relevance and motor response (salience and sensorimotor networks). CPM predicted problematic-gaming scores across all analyzed brain states and found shared predictive canonical networks, indicating generalizability. Applying the final reward-processing model to other task-based and resting-state fMRI data also successfully predicted problematic-gaming severity. CONCLUSIONS: The identified large-scale networks predictive of problematic-gaming severity in adolescents may serve as promising targets for personalized and novel interventions. Before using these results to guide clinical advances, future research should use external samples to evaluate replicability of the identified network.