A Conditional Mixed Model for Matched Case-Crossover Studies

• Activity Number: 651
• Type: Topic Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Inyoung Kim*+ and Feng Guo and Chun Gun Park
• Companies: Virginia Polytechnic Institute and State University and Virginia Tech Transportation Institute and Kyonggi University
• Keywords: Matched case-crossover study ; Mixed model ; Semiparametric Bayesian approach

Abstract

We propose a conditional mixed model to analyze the matched case-crossover study. A unified approach is developed to study covariates associated with binary outcome and to test for variations over stratum effects. A conditional logistic model is commonly used to eliminate the stratum effects but it assumes that that the variations over stratum effects are considered to be constant. Furthermore, the observations with identical covariates values do not contribute to the inference. The proposed conditional mixed model overcomes these limitations by treating the stratum variable as a random effect which depends on subjects in each stratum. We show theoretically that the estimates of parameters of interest in an unconditional generalized linear mixed model are biased, while the estimates in a conditional mixed model are unbiased. We develop both frequentist and Bayesian approaches to fit a conditional mixed model. A simulation study was performed to compare these approaches. The result shows that the semiparametric Bayesian approach is more efficient than other methods.