Application of generalized estimating equations to a dental randomized clinical trial

Longitudinal data present statistical problems of interest in clinical trials and epidemiologic studies. In this article, we consider a dental clinical trial in which the outcome measurements are taken on each subject at two follow-up times, and the primary interest is in the dependence of the outcome variable on covariates. The common data structure of these studies is the presence of an intraclass or serial correlation within primary sampling units or subjects. Recently generalized linear models have had extensions to methods for generalized estimating equations that take correlations within primary sampling units into account. We review and apply the Liang-Zeger methodology to a dental clinical trial. In this study, 109 adult male and female volunteers with pre-existing dental plaque were randomized to two mouth rinses (A and B) or a control mouth rinse with double blinding. The major research question in this analysis was: Are the two experimental mouth rinses more effective than the control mouth rinse in inhibiting the development of dental plaque? And if so, what is the effect of baseline plaque measurement?