Model-Robust Standardization in Cluster-Randomized Trials

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Generalized linear mixed models and generalized estimating equations are standard for cluster-randomized trials.
• These conventional methods can yield ambiguous treatment effect estimates with model misspecification or informative cluster sizes.

Purpose of the Study:

• To present a unified, model-robust approach for estimand-aligned inference in cluster-randomized trials.
• To develop consistent estimators for cluster-average and individual-average treatment effects.

Main Methods:

• A novel standardization approach to align regression model output with estimands.
• Introduction of always-consistent estimators for marginal treatment effects.
• Exploration of a deletion-based jackknife variance estimator.
• Development of a test for informative cluster size.

Main Results:

• The proposed estimators ensure consistent inference for treatment effects, irrespective of model specification accuracy.
• The approach provides a reliable method for handling informative cluster sizes.
• Simulation studies confirm the advantages of the proposed estimators across various scenarios.

Conclusions:

• The developed model-robust standardization methods offer reliable and consistent estimation of treatment effects in cluster-randomized trials.
• The MRStdCRT R package implements these novel statistical approaches for practical application.