Multiply robust generalized estimating equations for cluster randomized trials with missing outcomes

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Generalized Estimating Equations (GEEs) are standard for analyzing cluster randomized trials (CRTs).
• Informatively missing outcomes in CRTs can lead to biased parameter estimates using standard GEEs.
• Existing methods for missing data in CRTs often require strong assumptions about propensity score or outcome models.

Purpose of the Study:

• To develop a novel weighted GEE approach for CRTs with informatively missing outcomes.
• To provide robust estimation of marginal mean, scale, and correlation parameters.
• To relax strong model assumptions required by existing methods.

Main Methods:

• Developed new weighted GEEs allowing for multiple propensity score and covariate-conditional mean models.
• Proposed an iterative algorithm for implementation of the multiply robust estimator.
• Evaluated performance via Monte Carlo simulations.

Main Results:

• The new weighted GEEs provide consistent estimators if at least one specified model is correct.
• Simulations demonstrated the robustness and improved accuracy of the proposed method.
• The method was successfully applied to the Botswana Combination Prevention Project data.

Conclusions:

• The proposed multiply robust weighted GEEs offer a more flexible and reliable approach for CRTs with missing outcomes.
• This method enhances the ability to draw accurate conclusions from complex public health intervention studies.
• It addresses a critical limitation in the statistical analysis of cluster randomized trials.