Latent Variable Modeling for Principal Stratification Analyses of Longitudinal Studies

• Activity Number: 296
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Chen-Pin Wang*+
• Companies: The University of Texas Health Science Center at San Antonio
• Keywords: principal strartification ; causal effect ; direct effect ; indirect effect ; longitudinal study

Abstract

Lin, Ten Have, and Elliot (2007; 2008) proposed a principal strartification approach for a longitudinal randomized study to assess the treatment effect of a continuous outcome adjusting for repeated measures of a binary intermediate variable. Extending from Lin et al., we propose a principal strartification approach to compare the effects of two diabetes medications on cardiovascular diseases (a binary endpoint) directly and indirectly via glycemic control (a continuous intermediate variable with repeated measures) in a clinical cohort. The method involves a 2-step estimation procedure. Step 1 identifies posterior probabilities of principal strata, similar to that in Jo and Wang (2010). Step 2 uses the pseudoclass technique to derive the startum-specific treatment effects. Using the technique by Jo (2008) and Lin et al. (2007; 2008), we derive the direct and indirect treatment effects.