How to achieve model-robust inference in stepped wedge trials with model-based methods?

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Stepped wedge designs are increasingly used in cluster randomized trials.
• Model-based analysis is standard for evaluating treatment effects in these designs.
• Properties of these analyses under model misspecification are not well understood.

Purpose of the Study:

• To investigate the conditions under which model-based methods for stepped wedge designs provide consistent estimation of marginal treatment effects.
• To determine the impact of working model misspecification on the validity of these analyses.
• To identify requirements for robust inference.

Main Methods:

• Focus on linear mixed models and generalized estimating equations with various working correlation structures.
• Theoretical analysis of consistency for nonparametric marginal treatment effect estimands.
• Use of sandwich variance estimators and g-computation for robust inference.

Main Results:

• Consistency for nonparametric estimands generally requires a correctly specified treatment effect structure.
• Other aspects of the working model (covariates, random effects, error distribution) can be misspecified.
• Sandwich variance estimators provide valid inference; g-computation is needed for non-identity link functions or ratio estimands.

Conclusions:

• Model-based analysis of stepped wedge designs can be robust to certain types of model misspecification.
• Correct specification of the treatment effect is crucial for valid estimation.
• The findings offer guidance for analyzing stepped wedge trials and ensuring reliable treatment effect estimation.