Online Program

Joint GEEs for Multivariate Correlated Data with Incomplete Binary Outcomes

*Gul Inan, School of Statistics, University of Minnesota 
Recai M. Yucel, State University of New York at Albany 

Keywords: GEE, incomplete correlated binary responses, Kronecker product correlation matrix, multiple imputation, rounding

This study proposes a fully-parametric, but approximate multiple imputation (MI), method combined with a semiparametric generalized estimating equations (GEEs) to jointly analyze binary response variables in correlated data settings. Multivariate incomplete responses are multiply imputed using multivariate extensions of mixed-effects models, and then fitted using a joint GEE model in which covariates are associated with the marginal mean of responses with response-specific regression coefficients. Cluster-specific correlation structure for a given response variable and correlation structure between response variables are decomposed using Kronecker product, allowing a computational efficient estimation algorithm. Inferential combining rules of multiple imputation are used to combine these regression and correlation parameters. We assess the validity of the proposed MI-based GEE approach through a simulation study under different scenarios. Finally, the proposed approach is illustrated through Adolescent Alcohol Prevention Trial (AAPT) data.