GEE with Gaussian estimation of the correlations when data are incomplete

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Missing data in longitudinal studies pose challenges for standard statistical methods.
• Generalized Estimating Equations (GEE) are commonly used for correlated binary outcomes.
• Handling missing binary response data requires robust estimation techniques.

Purpose of the Study:

• To propose and evaluate a modified Generalized Estimating Equation (GEE) approach for analyzing longitudinal binary data with missing values.
• To assess the performance of the modified GEE under missing at random (MAR) assumptions.
• To compare the modified GEE with existing methods like standard GEE, multiple imputation, and weighted estimating equations.

Main Methods:

• A modified GEE approach utilizing Gaussian estimation for correlation parameters.
• Simulation studies were conducted using repeated binary outcomes with MAR data.
• Comparison with standard GEE, multiple imputation, and weighted estimating equations.

Main Results:

• The proposed modified GEE yields consistent regression parameter estimates under missing completely at random (MCAR).
• When data are missing at random (MAR) and the working correlation matrix is correctly specified, the bias in the modified GEE is negligible.
• The modified GEE performed favorably compared to other methods in simulation studies.

Conclusions:

• The modified GEE offers a viable and robust method for handling missing binary response data in longitudinal studies.
• The method demonstrates good performance, particularly when the correlation structure is accurately specified.
• The approach is illustrated effectively in a real-world clinical trial setting.