A Population-Averaged Model For Analyzing Longitudinal Binary Data

摘要

This paper applies the Generalized Estimating Equations (GEE) technique in a study that analyzes a set of longitudinal data on 81 type-2 diabetic patients at a healthcare centre of a higher-learning institution. The study aims to model the association between a binary response variable, which is the glycated haemoglobin (HbAlc) level of the diabetic patients and a set of covariates which are gender, race, working status, counseling, body mass index (BMI), high density lipid (HDL) level, triglycerides (TG) level, creatinine level and blood pressure of the patients. Although the population-averaged approach of the GEE is known to be robust to the misspecification of the correlation structure of the repeated measurements (responses), a correct specification will however, produce more efficient parameter estimates. Hence, four different working correlation structures were fitted to model the interdependence of the repeated measurements (responses) and a comparison of the results shows that an exchangeable correlation structure produces the best-fitting model. The result also shows that race, working status, counseling, HDL level and TG level of patients are significantly associated with their HbAlc level.