Stepped wedge cluster randomized trials (SW-CRTs) represent a broad class of study designs that are increasingly used in a wide range of settings to study the impact or implementation of new health care, public health and social interventions. They are unique to the conventional individually randomized trial in that they randomize groups of individuals and unique among cluster randomized trials as they allow each cluster or site to serve as its own control. SW-CRTs often collect a rich set of baseline information, including baseline measurements of the patient-centered outcomes they target to improve and the characteristics of the participating patients and clusters, which can provide insight into how interventions work and for whom. However, such baseline information has not been commonly utilized in existing methods or recommendations for designing and analyzing SW-CRTs, resulting in a missed opportunity for more precise evidence generation through SW-CRTs. In addition, the target population and treatment effect of interest (e.g., estimands) require substantial clarifications in SW-CRTs so that researchers can answer the right scientific question about the impact of health care delivery approaches. This research aims to overcome these current shortcomings and improve the conduct of SW-CRTs in four distinct ways.

First, the research team will develop new methods for planning SW-CRTs by adjusting for baseline measurements of the patient-centered outcome of interest, focusing on a new model that can characterize the temporal trend of the baseline and endline outcomes and complex correlations between all outcomes in each hospital or clinic. They will develop mathematical formulas to help researchers identify the number of individuals and sites required for future SW-CRTs when baseline outcome measurements are explicitly accounted for. These formulas will be developed and tested using statistical simulations (i.e., assessing the methods using hypothetical trials that look like actual trials that the team has worked on or has access to) and will help researchers select the optimal approach to incorporate baseline information in different settings.

Second, the team will define a new class of weighted average causal effect estimands to quantify the treatment effect evidence in SW-CRTs by recognizing that the unequal cluster sizes can contribute to variations of treatment effect in each cluster-period of an SW-CRT. This contribution is needed as conventional methods relying on restrictive assumptions can unintentionally bias trial results or change the interpretation of results if certain assumptions are violated. The new methods will not be susceptible to such issues and are hence robust to model misspecification. Additionally, the newly developed methods will be a principled approach to adjust for baseline covariates that are measured for patients and clusters and will achieve an important gain in statistical power for identifying treatment effect signals in SW-CRTs without relying on strong model assumptions.

Third, researchers will develop a class of statistical methods called flexible Bayesian tree-based mixed models to estimate the treatment effects in the overall population and stakeholder-prioritized subpopulation(s) in SW-CRTs. The new models will improve over conventional parametric models for SW-CRTs to flexibly accommodate unknown covariate effects through Bayesian nonparametric methods and will maximally extract all information from data to improve the statistical power. In addition, the study team will provide statistical formulations to summarize the Bayesian model output and ensure that the target of causal inference is clear. As before, statistical simulations will be used to validate the proposed new methods and to compare them with the existing methods to generate best practice recommendations.

Finally, the team will apply the new methods to recently completed SW-CRTs and create open-access statistical code and tutorials to facilitate practical implementation and improve patient-centered outcomes research via SW-CRT designs. The research team will work with stakeholders, including principal investigators of SW-CRTs and biostatistical leaders in pragmatic trials, to make study results accessible to the broader research community.