Data-Driven, Generalizable Prediction of Adolescent Sleep Disturbances in the Multisite ABCD Study.

Abstract

Sleep disturbances are common in adolescence and associated with a host of negative outcomes. Here we assess associations between multifaceted sleep disturbances and a broad set of psychological, cognitive, and demographic variables using a data-driven approach, canonical correlation analysis (CCA). Baseline data from 9,093 participants from the Adolescent Brain Cognitive Development℠ (ABCD) Study were examined using CCA, a multivariate statistical approach that identifies many-to-many associations between two sets of variables by finding combinations for each set of variables that maximize their correlation. We combined CCA with leave-one-site-out cross-validation across ABCD sites to examine the robustness of results and generalizability to new participants. The statistical significance of canonical correlations was determined by non-parametric permutation tests that accounted for twin, family, and site structure. To assess the stability of the associations identified at baseline, CCA was repeated using two-year follow-up data from 4,247 ABCD Study participants. Two significant sets of associations were identified: 1) difficulty initiating and maintaining sleep and excessive daytime somnolence were strongly linked to nearly all domains of psychopathology (r-squared = 0.36, p < .0001); 2) sleep breathing disorders were linked to BMI and African American/Black race (r-squared = 0.08, p < .0001). These associations generalized to unseen participants at all 22 ABCD sites and were replicated using two-year follow-up data. These findings underscore interwoven links between sleep disturbances in early adolescence and psychological, social, and demographic factors.