Addressing Lagged Effects and Interval Censoring in the Stepped Wedge Design of Cluster Randomized Clinical Trials
MetadataShow full item record
The stepped wedge design (SWD) of cluster-randomized trials has been growing in popularity and is increasingly being used to efficiently evaluate the rollout of interventions in a community setting. The design is especially useful in resource limited settings where it may not be feasible to introduce interventions all at once and where there are substantial clusters/communities/groups to provide answers regarding intervention effects. This paper reviews recent work on, and applications of, the SWD. Issues unique to the design are raised and two such issues, lagged effects and interval censoring in the context of the SWD, are addressed. The effects of frailty on hazard ratio estimation are also briefly analyzed. If an intervention is not fully implemented or not fully effective in all study units in the time step in which it is assigned, i.e., before the next time step, there are implications for effect estimation and power with the SWD and these implications have not been well studied. Additionally, methods for estimating and modeling lags in effect in this context are needed. A two-step method is proposed to estimate lagged effects and the estimator is evaluated in simulations. Suggestions are made for calculating power when a lagged effect is anticipated or when there are delays in intervention rollout. The method is applied to the stepped wedge trial of expedited partner treatment in Washington State. There has been little work on issues and methods of analysis in stepped wedge trials with time-to-event endpoints. In this paper, the finite sample behavior of the hazard function and hazard ratio estimate in the presence of frailty are studied and analytic methods to address interval censoring of time-to-event endpoints in the context of the SWD are developed. An EM Algorithm approach with random cluster effects and true failure times as joint latent variables is proposed to address interval censoring in discrete time. A method for computing the power of stepped wedge trials with time-to-event endpoints is proposed and the effect of interval censoring on power is investigated. Finally, the implications of this work on applied research and plans for future work are discussed.
- Biostatistics