Online Program

Estimating Causal Effects of Latent Treatment Classes: Natural clusters of drug treatment services for adolescents

Beth Ann Griffin, RAND Corporation 
Elizabeth Letourneau , Johns Hopkins Bloomberg School of Public Health 
*Megan Schuler, Johns Hopkins Bloomberg School of Public Health 
Elizabeth Stuart, Johns Hopkins University 

Keywords: causal inference, adolescent, substance use, latent variables, latent class analysis

Overview: In some public health and public policy studies, variables of interest are not directly observed and thus are more appropriately modeled as latent variables. However, standard causal inference methods, such as propensity scores, assume direct observation. This talk will present an overview of the latent class causal analysis method proposed by Schafer and Kang (2010), which extends propensity score methods to settings in which treatment is modeled as a latent variable. Additionally, we apply this method to investigate the effectiveness of various groupings of adolescent substance abuse treatment services. There are many treatment options for adolescent substance abuse, including adolescent-centered therapies, family-based therapy, case-management, and biological drug testing. Typically, effectiveness studies assess outcomes for a specific intervention protocol, which is typically comprised of several discrete services. However, in practice, adolescents may receive fewer or more services than those officially associated with a particular program due to concurrent program enrollment, noncompliance, or lack of program fidelity. Thus, a relevant policy objective is to determine the effectiveness of common “clusters” of substance abuse treatment components adolescents actually receive.

Methods: Data are from a multisite, longitudinal observational study of adolescents (ages 12-18) enrolled in substance abuse treatment programs as part of SAMHSA Center for Substance Abuse Treatment funding. The current study is restricted those who report receiving only outpatient drug treatment services between baseline and 3-month study visits (N = 4854). The Global Assessment of Individual Needs was the primary study instrument: the Treatment Received Scale was used to assess which drug treatment services youth received and the Substance Frequency Scale and Substance Problems Scale were used to assess substance use outcomes. Propensity scores were used to adjust for observed imbalances across treatment classes on numerous variables including demographics, baseline substance use, and juvenile justice system involvement. Latent class causal analysis was done to identify latent treatment classes and estimate the effects of those classes on substance use outcomes

Results: Results indicate that a 4 class model is optimal. Treatment classes are described as follows: High Adolescent, Family, and Case Management services (12%); High Adolescent and Family services (12%); High Adolescent services (39%); and Low Adolescent services (37%). These groups represent varying levels of treatment intensity and comprehensiveness, ranging from only adolescent-centered services to the combination of adolescent-centered, family-based and case management services. Differences in substance abuse outcomes across latent treatment classes will be discussed.