HIGH-DIMENSIONAL JOINT MODEL FOR LONGITUDINAL BINARY OUTCOME

Abstract

Binary outcomes are often collected in clinical and epidemiological studies to investigate the evolution of some outcomes over time. In studies with two or more binary outcomes, research questions often revolve around the joint evolution of the binary outcomes over time. However, independently modelling the evolution of each outcome variable ignores the correlation among the variables. Although generalized mixed models have been proposed to model the joint evolution of binary outcome variables over time, the estimation of the corresponding regression coefficients and covariance parameters may be computationally difficult as the number of outcome variables increases. In this study, we investigate the use of a pairwise generalized mixed models approach based pseudo-likelihood theory, in which all possible bivariate models are fitted, to estimate the parameters of a multivariate longitudinal binary data and compared it with univariate models.  This methodology is illustrated using data from a longitudinal study of the prevalence of four ailments in 200 children in the south-western part of Nigeria. This methodology is shown to be computationally easy and beneficial over the conventional multivariate generalized mixed-model methods. It is also advantageous over univariate generalized mixed-effects models as it incorporates the modeling. This research provides applied researchers with alternative tools to investigate the joint evolution of binary outcomes over time.