Improving Causal Inference for National Adolescent Substance Use Datasets

Project Summary Prevention of adolescent substance use (SU) and interventions to reduce harmful use has long been a goal of NIDA in order to improve the health of individuals and lighten the societal burden of substance misuse. Research from longitudinal observational studies is critical for establishing sufficient evidence that the target of intervention could be a causal factor prior to developing and for fine-tuning interventions. This R21 responds to a call (RFA-DA-22-038) to leverage data from the Adolescent Brain and Cognitive Development (ABCD) study to develop new and advanced statistical methods, in this case, to improve causal inference and inform intervention target selection. We will develop and test a new model: a Discordant Sibling Random-Intercept Cross-Lagged Panel Model (DS RI-CLPM) using data on several (child- and parent-driven) facets of parental monitoring, a likely causal environmental factor for which there is strong extant evidence of an inverse association with adolescent SU (aim 1). We will also (aim 2) explore the comparability of estimates from the DS RI-CLPM model to standard RI-CLPM and DS models and other models to understand how the DS RI- CLPM compares to current models, and (aim 3) conduct a series of simulation studies to explore power and the minimum sample sizes needed to establish meaningful effects under various circumstances (i.e., effect sizes, degree of sibling discordance). The initial products of this R21 will include: 1) openly available code to fit the DS RI-CLPM model in tutorial form, 2) publications explaining the interpretation and utility of the model, and 3) publications of simulation studies explicating power, sample size, and other (i.e., model specification) considerations of the model. The expected end-users are twofold: This model can be used by any researcher interested in testing a stricter model of causality for their focal associations using large-scale, longitudinal data that includes twins and/or siblings, many of which are publicly available (e.g., ABCD, Add Health, harmonized data from the NIH Environment in Child Health Outcomes initiative, Twins Early Development Sample). Second, we have formed an advisory board of intervention experts to help guide our analyses, sensitivity analyses, presentation, and dissemination of findings to aid in translation of these findings to ongoing work in the intervention and prevention of adolescent SU. Through discussions with our advisory board, we expect that scientists who develop and test interventions will find results from this model of use. By subjecting multiple aspects of key malleable constructs to this model, we will provide intervention scientists with more detailed information about the likelihood of causal links and key targets for future interventions. A longer-term product of this R21 will be an R01 grant submission exploring the generalizability of this model (and the standard RI- CLPM and standard DS design) to intervention samples (extant data provided by members of our advisory board). Combined with results from this R21, the planned R01 is expected to yield critical information about how and when nationally representative longitudinal data are more and less informative for interventions.