|
Abstract:
|
In a stepped wedge design (SWD),an intervention is rolled out in a staggered manner over time, in groups of experimental units, so that by the end, all units receive the intervention. For example,in the MaxART study, the date at which to offer universal antiretroviral therapy to otherwise ineligible clients was randomly assigned in nine “steps” of four months duration so that after three years, all 14 facilities offered early treatment. In the common alternative, the cluster randomized trial (CRT), experimental units are randomly allocated on a single common start date to the interventions to be compared. Often, the SWD is more feasible than the CRT, both for practical and ethical reasons, but takes longer to complete. The SWD permits both within- and between- unit comparisons, while the CRT only allows between-unit comparisons. Thus, confounding bias by time-invariant factors tends to be lower in a SWD than a CRT, but the SWD cannot as readily control for confounding by time-varying factors. Generally, SWDs have greater statistical power than CRTs, especially as the ICC and within-cluster sample size increases. Software availability and controversies will be discussed.
|
