Abstract:
|
The stepped wedge cluster randomized trial (SW-CRT) is a one-way crossover design, where the intervention is rolled out sequentially to the trial groups whose orders are randomly determined. We propose several novel stepped wedge cluster designs for data with two layers of clustering, focusing on cross-sectional designs in which different participants from each cluster are observed at each step. The proposed designs differ in the allocation of steps where units from the same cluster are exposed to the intervention. In design 1 (the existing design), all units in the same cluster transfer to the intervention in a single step. In designs 2 and 3, units from the same cluster complete the transition to the intervention within two adjacent and nonadjacent steps, respectively. In design 4, units in the same cluster are allocated to the intervention at every step. We also study the efficiency of the treatment parameter estimator at different intra-cluster correlation values under these designs via simulation. Under the linear mixed model, design 4 yields the most efficient estimator of the treatment parameter, since it maximizes the number of between-unit within-cluster comparisons.
|