Model misspecification in stepped wedge trials: Random effects for time or treatment

Area of Science:

• Biostatistics
• Clinical Trial Design

Background:

• Mixed models are standard for analyzing stepped wedge trials (SWTs), accounting for clustered and repeated measures.
• Model misspecification, specifically regarding random effects (time vs. treatment), can invalidate treatment effect inference.

Purpose of the Study:

• To investigate the impact of model misspecification on variance component estimates and treatment effect variance in SWTs.
• To provide analytical insights into the convergence of misspecified model estimates and their bias.

Main Methods:

• Utilized analytical solutions for asymptotic results to examine variance component convergence under misspecification.
• Assessed the influence of study design and true variance components on bias in model-based standard errors.

Main Results:

• Found that bias in standard errors is dependent on study design and variance component magnitudes.
• Identified specific scenarios where incorrect random effect selection significantly impacts inference.
• Demonstrated that trends are influenced by trial design and correlation structure assumptions.

Conclusions:

• Model misspecification in SWTs can substantially affect treatment effect inference.
• Provided tools for researchers to assess the impact of misspecification in specific trial designs.
• Emphasized the importance of sensitivity analyses to confirm the robustness of SWT findings.