Application of GEE and MRM in Evaluation of the Efficacy of an HIV Prevention Intervention

• Activity Number: 409
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Xinjian Zhang*+ and Brenda Chen and Tobey Sapiano and Gary Uhl
• Companies: CDC/DID/NCHHSTD/DHAP/CSDMB/SST and CDC and CDC and CDC
• Keywords: Generalized estimating equation (GEE) ; Mixed-effects Regression Models (MRM) ; Sisters Informing Sisters about Topics on AIDS (SISTA) ; HIV intervention

Abstract

Generalized estimating equations (GEE) and mixed/random effects regression models (MRM) are two approaches commonly used to evaluate longitudinal designs that involve repeated measurements. The merits of these two kinds of models are poorly understood by non-statisticians. In particular, this presentation will focus on appropriate uses of GEE and mixed models. Sisters Informing Sisters about Topics on AIDS (SISTA) is an HIV prevention intervention designed for African American women. Using SISTA as an example, this presentation will illustrate procedures for model selection and interpretation of results. GEE logistic regression models and random intercept models (mixed effects models) are applied to evaluate changes in HIV risk behaviors from baseline to two follow-up time points. Population-level and subject- level parameter estimates, and differences in parameter interpretation (e.g., evaluation of intervention efficacy at the population level vs the exploration of HIV risk change variability within and among subjects) from the two types of models are compared. Recommendations for model selection and parameter interpretation for the evaluation of HIV intervention preventi