Considerations Before Selecting a Stepped-Wedge Cluster Randomized Trial Design for a Practice Improvement Study

Ann M. Nguyen; Charles M. Cleland; L. Miriam Dickinson; Michael P. Barry; Samuel Cykert; F. Daniel Duffy; Anton J. Kuzel; Stephan R. Lindner; Michael L. Parchman; Donna R. Shelley; Theresa L. Walunas

doi:10.1370/afm.2810

Article Figures & Data

Figures

Tables
Additional Files

Download figure
Open in new tab
Download powerpoint
Figure 1.
Sample SW-CRT scheme.
Q=quarter; SW-CRT = stepped-wedge cluster randomized trial.

View popup

Table 1.

Overview of Cooperatives

Cooperative	Region Served	No. of Participating Practices	Study Design
Midwest	Wisconsin, Illinois, Indiana	226	Parallel cluster randomized trial^a
New York City	New York City	276	SW-CRT
North Carolina	North Carolina	228	SW-CRT
Northwest	Washington, Oregon, Idaho	209	2×2 parallel randomized factorial trial^b
Oklahoma	Oklahoma	226	SW-CRT
Southwest	Colorado, New Mexico	202	Parallel cluster randomized trial^c
Virginia	Virginia	208	SW-CRT

SW-CRT = stepped-wedge cluster randomized trial.
↵a This parallel cluster randomized trial had 4 waves with 2 intervention arms in each wave (intervention and intervention plus), which included additional support. Practices were recruited in 4 separate waves, and waves began at staggered times.
↵b The 2×2 randomized factorial trial had 2 factors (factor A and factor B), each with an on and off level potentially added to the standard level, leading to 4 study conditions. Practices were recruited up front, and all arms followed the same timeline.
↵c This parallel cluster randomized trial had 2 intervention arms (1 standard and 1 enhanced), which included additional support. Practices were recruited up front, and both arms followed the same timeline.

View popup

Table 2.

Advantages and Challenges of Using SW-CRTs

	Description	Representative Quote
Advantage
Incentivized recruitment	All sites receive the intervention and serve as their own control.	“To ask a site to engage in research for a year without receiving resources doesn’t seem possible.” (New York City)
Staggered resource allocation	Resources (eg, personnel) can be allocated over a longer period.	“You can shift resources from one sequence to another, which eases workforce logistical concerns.” (ESCALATES)
Statistical power	Power can be higher than parallel cluster randomized controlled trials under certain conditions.	“Stepped-wedge designs potentially have the power advantage over alternative designs, though the trade-off is the time issue.” (New York City)
Challenge
Time-sensitive recruitment	Recruitment must occur at all sites up front.	“Recruitment challenges were the main reason we chose the parallel cluster randomized trial and not a stepped-wedge design.” (Midwest)
Retention	Sites might drop out owing to a long lag time between recruitment and the start of the intervention.	“By the time all partnership agreements were signed and sites were randomized, 47 had dropped out of the study.” (Northwest)
Randomization requirements and practice preferences	It might be difficult to randomize sites into sequences, given that real-world practice priorities are often changing.	“The real-world environment does not really respect the randomization. Stepped-wedge designs are complicated by the fact that they have defined start and stop dates. That’s not how quality improvement works.” (North Carolina)
Achieving treatment schedule fidelity	It might be difficult to deliver the intervention as prescribed (eg, sites might cross-talk across sequences).	“We held weekly meetings with all facilitators to deliberately talk about cross-contamination and staying with the timeline.” (Oklahoma)
Intensive data collection	Sites might have difficulty contributing data for specified outcome measures for every time block of the implementation timeline.	“Practice burden is usually greater for SW-CRTs (compared to other designs) in terms of measurement because every practice has to report every measure for every time block.” (Southwest)
Hawthorne effect	Sites might modify their behavior before the intervention begins.	“We anticipated that the sites would start preparing (before the intervention started).” (Midwest)
Temporal trends	Effect of intervention might be confounded by underlying temporal trends.	“Ideally, we would use the stepped-wedge design in a scenario where there aren’t significantly different covariates across clusters” (Southwest) and “when an outcome isn’t already expected to be improving.” (New York City)