Speaker:
Study designs that repeatedly measure an outcome on a subject over time are common in the biomedical sciences. These study designs require longitudinal data analysis methods that can account for within subject association between the repeated points. However, the most common methods for accounting for this correlation introduce important interpretation differences when working with non-linear link functions, which are commonly used for discrete outcomes such as binary or count data. This talk will discuss differences in interpretation of regression coefficients when fitting marginal models with GEE or subject specific generalized linear mixed models, as well as strategic tips to help diagnosis and correct some common issues when fitting models of either type. An example will be provided using hospitalization counts ascertained through Medicare claims linked to a dialysis population tracked by the United States Renal Data System (USRDS).