The Causal Effect of Substance Use Latent Class Membership On a Distal Outcome

Introduction. Substance use, abuse, and dependence are known to be associated with elevated levels of depression (e.g., Brook, et al., 2002), yet it is unclear whether substance use behavior profiles have a long-term causal impact on depression. Recent advances in causal inference methods have been proposed to account for possible confounders (i.e., variables that jointly influence substance use and subsequent depression) so that causation may be inferred. Modern causal inference techniques are rapidly being adopted by prevention scientists, but methods for drawing causal inferences about the consequences of behaviors measured as latent classes have not been available. This talk will demonstrate a new approach to estimating causal effects of latent class membership (substance use) on a distal outcome (depression).

Methods. The sample comprised 703 adolescents (53.3% female; mean age=14.9) from the 9th-grade cohort of the National Longitudinal Study of Adolescent Health. Potential confounders, including individual, family, peer, school, and neighborhood characteristics, were assessed at Grade 9. The latent exposure, substance use behavior latent class membership, was assessed at Grade 10 using indicators of alcohol, tobacco, and marijuana use. The outcome, depression, was assessed via a modified version of the CES-D during emerging adulthood. We estimated the causal effect of adolescent substance use behavior profile on depression in emerging adulthood using the R package LCCA.

Results. We identified 4 latent classes: Nonusers (51% of sample) were unlikely to report any substance use; Smokers (18%) experimented with cigarettes and in many cases were regular smokers; Drinkers (11%) had recent heavy episodic drinking but no cigarette use; Heavy Users (20%) engaged in regular smoking, heavy drinking, and marijuana use. Prior to adjusting for confounding, using a model-based approach to LCA with a distal outcome, membership in Smokers, Drinkers, and Heavy Users was associated with elevated levels of depression compared to Nonusers. Preliminary results based on LCCA, which adjusts for confounders, suggest that elevated depression is a consequence of membership in the Drinkers class but not the other substance use classes.

Conclusion. Modern approaches for LCA with a distal outcome can provide insight into the potential risk conferred by class membership, but as with standard regression-based approaches, coefficients describing the association between latent class membership and the distal outcome are correlational. We demonstrate a new methodological approach for estimating the causal effect of latent class membership on a distal outcome in the context of a brief software demonstration so that prevention scientists may adopt this approach in their own work.