Promoting Rapid Uptake of Multilevel Latent Class Modeling via Best Practices: Investigating Heterogeneity in Daily Substance Use Patterns

PROJECT SUMMARY Substance use among young adult college students remains a major public health concern, with recent escalations in nonmedical Adderall use, vaping nicotine, and co-use of alcohol and cannabis. The landscape of substance use continues to evolve, with new modes of use (e.g., vaping) and rapid legislative changes. Daily use behaviors can take many forms, defined by number and type of substances used, mode of use, and quantity used. Further, daily use patterns may be related to both day-level psychosocial factors (e.g., motives, affect, stress) and person-level characteristics (e.g., sex, race/ethnicity, financial strain). Use of multiple substances within a day (“co-use”) is increasingly prevalent and confers high risk for acute and chronic consequences, but is insufficiently studied. The unprecedented amount of high-quality, intensive longitudinal data (ILD) on substance use and associated risk factors holds critical information that can help explain substance use etiology and inform the development of the next generation of interventions that are tailored to those who need it, when they need it. However, advanced methods for analyzing ILD—including multilevel latent class analysis (MLCA) and multilevel latent transition analysis (MLTA)—are needed to characterize the heterogeneity of patterns of use that unfold in daily life and to identify novel day-level intervention targets. These methods enable researchers to discover clinically relevant substance use behavior patterns and day-to- day transitions, dynamic risk factors embedded in daily life that are associated with problematic use patterns, and person-level characteristics that indicate for whom these risk factors are most salient. Yet MLCA and MLTA are underused for ILD analysis, in part due to the dearth of guidance for applied researchers. Guided by a socio-ecological framework, we will evaluate, apply, and disseminate MLCA and MLTA to analyze ILD, considering four ways to specify models: marginal modeling, a fully parametric random effects model, a common factor model, and a non-parametric approach for random effects. We will translate best practices for applying MLCA (Aim 1) and MLTA (Aim 2) to ILD on substance use and co-use. We will conduct coordinated MLCA and MLTA of daily data from two large studies of college students (3CAM: n=343, 15,798 person-days; Project SELF: n=2068, 33,722 person-days). Empirical analyses will disentangle person-level risk factors (e.g., trait anxiety) and dynamic risk factors (e.g., morning affect) to isolate potential novel, dynamic intervention targets. We will examine individual characteristics (e.g., race/ethnicity) as moderators to investigate disparities and identify group-specific mechanisms. To maximize our overall impact (Aim 3) and ensure broad, rigorous adoption of MLCA and MLTA, we propose a broad range of dissemination activities to include online instructional material, a code repository, workshops, and a conference. Findings from studies using these techniques will guide prevention scientists and clinicians in designing interventions targeting the most salient risk factors for specific individuals at the best time for intervention.