Model-robust standardization in stepped wedge cluster randomized trials

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Stepped-wedge cluster-randomized trials (SW-CRTs) are common in healthcare and implementation science.
• Traditional analysis methods for SW-CRTs rely on models with implicit weights and assumptions, potentially leading to ambiguous estimates if the model is incorrect.

Purpose of the Study:

• To propose a model-robust standardization framework for SW-CRTs.
• To generalize existing methods for parallel-arm cluster-randomized trials to address informative cluster sizes in SW-CRTs.
• To define and estimate various causal average treatment effects under a super population framework.

Main Methods:

• Developed a standardization framework that generalizes methods from parallel-arm cluster-randomized trials.
• Defined horizontal and vertical average treatment effects for individuals and clusters.
• Introduced a procedure to standardize parametric and semiparametric working models for analysis aligned with specific causal estimands.

Main Results:

• The proposed estimators are consistent for their target causal estimands, even if the working regression model is misspecified.
• Estimator efficiency increases as the working model better reflects the true data-generating process.
• Extensive simulations confirmed the finite-sample properties of the new estimators.

Conclusions:

• The novel framework provides robust and efficient estimation of causal effects in SW-CRTs.
• The methods are applicable across various healthcare and implementation science research settings.
• An R package (MRStdLCRT) is available for practical implementation.