Background
Brain characteristics, environmental risks, and their interactions are believed to contribute to the development of psychopathology. However, it is unclear whether these factors differ in their utility for predictive versus explanatory goals; few predictive studies have combined and compared neuroimaging and environmental features, and Brain × Environment studies of explanatory processes have not adopted comprehensive, data-driven methods. Here, we integrated neuroimaging and environmental data to predict youths’ psychopathology and to explore brain-based measures of sensitivity to environmental influence.
Methods
Leveraging the Adolescent Brain Cognitive Development (ABCD) Study (release 5.1), we conducted analyses with up to 9,370 participants with complete baseline environmental, neuroimaging, and psychopathology data. Symptoms were assessed using the Child Behavior Checklist. Environmental exposures measured from parent-report, youth-report, and geocoded data were used to create a general exposome factor and six specific subfactors. Anatomical, resting-state functional connectivity, and diffusion MRI data were included as brain-based predictors. Nested cross-validated models assessed the predictive utility of environmental data, neuroimaging data, and their interactions. Elastic net models identified Brain × Environment interactions in nonpredictive analyses.
Results
Environmental data far outperformed neuroimaging in predicting psychopathology in unseen data. Although Brain × Environment interactions did not improve prediction accuracy, they revealed neural moderators of environmental effects in nonpredictive models. Measures involving the hippocampus, amygdala, salience network, and attentional networks were overrepresented among neural moderators. Certain brain features conferred increased or decreased sensitivity to supportive and adverse environments.
Conclusions
Our findings not only highlight the value of environmental data for predicting preadolescents’ psychopathology but also show that neuroimaging data play a role in supporting explanatory models of neural sensitivity to context.