Proposed variations of the stepped-wedge design can be used to accommodate multiple interventions

Abstract

Objectives: Stepped-wedge design (SWD) cluster-randomized trials have traditionally been used for evaluating a single intervention. We aimed to explore design variants suitable for evaluating multiple interventions in an SWD trial.

Study design and setting: We identified four specific variants of the traditional SWD that would allow two interventions to be conducted within a single cluster-randomized trial: concurrent, replacement, supplementation, and factorial SWDs. These variants were chosen to flexibly accommodate study characteristics that limit a one-size-fits-all approach for multiple interventions.

Results: In the concurrent SWD, each cluster receives only one intervention, unlike the other variants. The replacement SWD supports two interventions that will not or cannot be used at the same time. The supplementation SWD is appropriate when the second intervention requires the presence of the first intervention, and the factorial SWD supports the evaluation of intervention interactions. The precision for estimating intervention effects varies across the four variants.

Conclusion: Selection of the appropriate design variant should be driven by the research question while considering the trade-off between the number of steps, number of clusters, restrictions for concurrent implementation of the interventions, lingering effects of each intervention, and precision of the intervention effect estimates.

Keywords: Cluster-randomized trial; Efficiency; Multiple interventions; Pragmatic trials; Stepped-wedge trial design; Study design.

Figure 1

The traditional stepped wedge study 0=Control period; 1=Intervention period

Figure 2

Stepped Wedge Design Variations* 0=Control period; 1=Intervention 1 period; 2=Intervention 2 period; 1+2=Period with both Intervention 1 and Intervention 2. *To aid in comparisons, each design is constructed to have 5 time intervals and 24 clusters (i.e. each group contains 3, 8, 8, and 6 clusters for the Concurrent, Replacement, Supplementation and Factorial design, respectively). Variants of each design are possible.

Figure 3

Variances of the estimates of intervention effects by design* * Each design was analyzed with 24 clusters, 5 time steps and 100 individuals sampled cross-sectionally per time step per cluster. Note that the variance of the Intervention 1 estimate is identical for the Supplementation and Replacement designs.