Handling incomplete outcomes and covariates in cluster-randomized trials: doubly robust estimation, efficiency...

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Missing data is a common challenge in cluster-randomized trials (CRTs), affecting individual outcomes, covariates, and non-participant data.
• Existing methods often fail to address all types of missing data simultaneously in CRTs.
• Missing outcomes are a primary focus, but a comprehensive approach is lacking.

Purpose of the Study:

• To propose a novel doubly robust estimator for the average treatment effect in CRTs.
• To develop a method that simultaneously handles multiple types of missing data, including outcomes, covariates, and cluster sizes.
• To provide a robust framework for analyzing CRTs with complex missing data patterns.

Main Methods:

• A doubly robust estimation approach is proposed for various effect measure scales.
• The method accommodates missing outcomes under missingness at random and missing covariates without strict mechanism constraints.
• It also addresses missing cluster-population sizes using a uniform sampling mechanism.

Main Results:

• The proposed estimator offers a unified approach to handling diverse missing data scenarios in CRTs.
• Key considerations for enhancing precision include optimal weighting, machine learning integration, and modeling treatment assignment.
• A new sensitivity analysis framework is developed to assess the impact of violated missing data assumptions.

Conclusions:

• The developed doubly robust estimator provides a comprehensive solution for missing data in CRTs.
• The methods improve the precision and robustness of treatment effect estimation.
• The sensitivity analysis framework aids in evaluating the reliability of findings under potential data missingness violations.