Advances in Finite Mixture Modeling for Substance Use and HIV Research

PROJECT SUMMARY (See instructions): Tailoring programs for substance use and HIV inverventions is critical because the greatest effects will be realized if the right program is offered to the right individuals at the optimal time. At the same time, tailoring programs to each individual in a population may not be feasible. It may be possible, however, to detect a finite set of subgroups of individuals based on common risk exposure, dominant risk effects, treatment response, or mediational pathways; if so, we can leverage preventive intervention resources more effectively by matching appropriate program components to the different subgroups. The overall goals ofthe project are to advance a framework based on finite mixture models and bring it to the forefront in intervention science so that it can be used to address new questions in research on substance use and HIV. The Specific Aims of this Project are: 1. To enable substance use and HIV scientists to simultaneously (a) identify latent classes of risk and (b) use membership in these latent classes to predict a dependent variable; 2. To enable substance use and HIV scientists to identify latent classes characterized by the relations between multiple risks and key dependent variables; 3. To enable substance use and HIV scientists to simultaneously (a) identify latent classes characterized by different sizes of treatment effects and (b) use covariates to predict membership in these treatment-response latent classes; 4. To enable substance use and HIV scientists to identify latent classes of individuals based on shared mediational pathways; and 5. Dissemination and software development. For each method developed in this project, empirical results will be published in journals focused on prevention/substance use/HIV, and methodological innovations and results of simulation studies will be published in quantitative journals. In addition, by working with the Software Development and Technology Transfer Gore, user-friendly software will be made available for download as SAS procedures and STATA plug-ins for conducting LCA with a dependent variable, LC regression, and LC mediation. Adoption ofthe proposed methods by substance use and HIV scientists in the analysis of empirical diata will inform the development of programs tailored for and targeted to different subgroups of individuals.