Researchers are often interested in analyzing data that arise from longitudinal, repeated measures or clustered design, and there exists correlation between observations on a given subject. If the outcomes are approximately multivariate normal, then well established likelihood based methods of analysis are available. But if the outcomes are binary or counts, general likelihood based approaches are less tractable.

The responses in a longitudinal study are usually positively correlated. In analyzing longitudinal data, this dependence must be taken into account to avoid misleading inference. Logistic regression may be used to analyze and identify the risk and prognostic factors for the disease status. But it ignores the possible correlation among the repeated observations. Bahadur's representation (1961) is an alternative approach for analyzing correlated binary data. But it becomes complicated when the number of repeated observations for each individual increase.

Generalized estimating equation (GEE) has become an important strategy in the analysis of repeated observations. It takes the working correlation among the repeated outcomes into account. GEE is non-likelihood based and does not require the complete specification of the joint distribution of the repeated measurements. We applied standard GEE methodology on diabetes mellitus data collected by BIRDEM during 1984-1996.

People with diabetes mellitus are vulnerable to variety of complications over time. Complications associated with diabetes are eye, kidney, skin, dental, heart, large and small vessels etc. We consider three complications namely eye, dental and skin in order to apply GEE approach.

Longitudinal or clustered studies often have missing data either by design or happenstance. Although GEE is a popular strategy analyzing longitudinal data, but unbalanced or incomplete data can complicate the GEE analysis. If the missingness can be thought of as being missing completely at random (MCAR) then the consistency of results obtained by GEE holds. But if the missingness mechanism is missing at random (MAR) or not missing at random (NMAR) then consistency of GEE does not hold.

Deleting the cases with missing data and run analysis on the remaining observations are popular among the practitioners. Another attractive approach is to imputing the missing data with plausible values. All the naïve methods are not without pitfalls. However Xie and Paik (1997) developed SA imputation method to handle missing data in the GEE texture when missingness is MAR. SA method attempts to eliminate the bias originated from naïve methods. In this study we tried to explore the relationship between the repeated outcomes and covariates applying on diabetes mellitus data considering missing data issues.