Integrating multilevel, multidomain and multimodal neuroimaging factors to predict early alcohol exposure trajectories using explainable AI.

Various multilevel, multidomain factors at the individual-, family-, and environmental-level, and changes in neurobiology have been associated with the likelihood of developing alcohol use disorder (AUD) or binge drinking later in life. Prior studies have examined only limited subsets of these factors, typically focusing on cross-sectional associations with alcohol initiation, binge drinking, or AUD rather than exploring longitudinal alcohol use trajectories. Our study addresses these gaps by applying machine learning methods to a comprehensive set of multilevel, multidomain factors and multimodal brain imaging features (including brain structure and functional connectivity) to prospectively predict early alcohol sipping trajectories. Using data from the Adolescent Brain Cognitive Development Study, we identified functional connectivity features and multilevel factors that distinguish youth with an increasing alcohol sipping trajectory from those who initially experimented with alcohol but reduced their consumption over time. Moreover, structural and functional features predicted differences between youth who increasingly sipped over time and those who did not engage in alcohol experimentation. Interactions between age, socioeconomical status and positive attitudes towards drinking could predict a pattern of increasing alcohol sipping over time. These trends could inform how individual, family, environmental and neurobiological factors impact the development of different alcohol sipping trajectories over time.