A Machine Learning Based Causal Interface for Time Varying Environmental Predictors of Substance Use Initiation in the ABCD Study
Year: 2026
Type: preprint
Abstract: Background The Adolescent Brain Cognitive Development (ABCD) Study ® offers rich longitudinal data on environmental, genetic, and other factors related to substance use initiation. Classical marginal structural models (MSMs) require choosing which covariates to include in propensity models, but this choice is difficult in the presence of hundreds of correlated predictors. Methods We analyzed longitudinal data from the Adolescent Brain Cognitive Development Study using a European-ancestry unrelated cohort, in which each individual contributed repeated observations over time. Interval-level binary outcomes were defined for initiation of alcohol, nicotine, cannabis, and any substance, restricting analyses to participants at risk prior to initiation. All predictors were constructed as lagged variables to preserve temporal ordering. We implemented a two-stage machine learning–based causal framework. First, we performed graph discovery using a “Granger-inspired lagged predictive modeling” approach, applying elastic-net logistic regression to identify predictive relationships between lagged environmental variables and future initiation outcomes. Robust candidate edges were selected using subject-level bootstrap stability selection. Second, we estimated adjusted effect sizes for stable edges using double machine learning (DML)-style partialling-out with cross-fitting. For each candidate predictor, we defined the treatment as the lagged variable of interest and adjusted for high-dimensional lagged covariates. Cross-fitting with group-based splitting accounted for within-subject dependence, and nuisance functions were estimated using random forest models. Cluster-robust standard errors were used for inference. Results Across analyses, behavioral and environmental predictors showed stronger and more consistent associations with substance-use initiation than polygenic risk scores (PRSs). In the graph-discovery analysis, stable edges were mainly related to rule-breaking behavior, sensation seeking, resiliency, sleep, screen-related measures, and family- or parent-related variables. PRS variables were less consistently selected and did not appear as dominant predictors. Cannabis initiation showed strong stable links with parent-reported rule-breaking behavior and behavioral symptoms. Nicotine initiation was linked to sensation seeking, behavioral symptoms, and screen-related measures. DML-style effect estimates were modest in magnitude but supported positive associations for several stable predictors, including rule-breaking behavior for cannabis initiation and sensation seeking, Brief Problem Monitor (BPM) behavioral items, and screen-related measures for nicotine initiation. Conclusions In this European-ancestry unrelated cohort, adolescent substance-use initiation was more consistently associated with modifiable behavioral and environmental factors than with PRS variables. The findings suggest that cannabis and nicotine initiation share some risk factors but also have distinct predictor profiles. These results highlight the importance of longitudinal, multi-method approaches for identifying early risk patterns and suggest that prevention strategies may benefit from focusing on behavioral regulation, sensation seeking, family context, sleep, and screen-related behaviors.
Source: medRxiv
Authors: Mengman Wei, Lasya Yadlapati, Qian Peng
Language: English
Subfield: Epidemiology
Field: Medicine
Domain: Health Sciences
Open Access status: green
Funder: National Institutes of Health
OpenAlex ID: W7154697742