Novel methods for the analysis of stepped wedge cluster randomized trials

Abstract

Stepped wedge cluster randomized trials (SW‐CRTs) have become increasingly popular and are used for a variety of interventions and outcomes, often chosen for their feasibility advantages. SW‐CRTs must account for time trends in the outcome because of the staggered rollout of the intervention. Robust inference procedures and nonparametric analysis methods have recently been proposed to handle such trends without requiring strong parametric modeling assumptions, but these are less powerful than model‐based approaches. We propose several novel analysis methods that reduce reliance on modeling assumptions while preserving some of the increased power provided by the use of mixed effects models. In one method, we use the synthetic control approach to find the best matching clusters for a given intervention cluster. Another method makes use of within‐cluster crossover information to construct an overall estimator. We also consider methods that combine these approaches to further improve power. We test these methods on simulated SW‐CRTs, describing scenarios in which these methods have increased power compared with existing nonparametric methods while preserving nominal validity when mixed effects models are misspecified. We also demonstrate theoretical properties of these estimators with less restrictive assumptions than mixed effects models. Finally, we propose avenues for future research on the use of these methods; motivation for such research arises from their flexibility, which allows the identification of specific causal contrasts of interest, their robustness, and the potential for incorporating covariates to further increase power. Investigators conducting SW‐CRTs might well consider such methods when common modeling assumptions may not hold.

Statistical Properties of Stepped Wedge Cluster-Randomized Trials in Infectious Disease Outbreaks

“…for asymptotic inference for the IRT, CRT, MEM, and CPI analyses and permutation inference for the NPWP and SC analyses. As seen in other settings, the asymptotic inference for MEM and CPI leads to greatly inflated Type I Error (14,20,22). The permutation inference for NPWP and SC has greatly reduced power compared to the IRT and CRT methods, with less than 50% power to detect a true direct vaccine effect of 0.6 compared to over 90% for the CRT and nearly 100% for the IRT.…”

“…The results from the IRT and CRT designs are analyzed as described by We do not consider the horizontal crossover method since it is not well suited to capture indirect effects and is highly sensitive to time trends (22). To account for the delay in symptom onset (and thus the delayed effect of the intervention), we remove the first period on intervention for each cluster from all SWT analyses.…”

“…Periods with zero cases in both arms do not contribute to the effect estimate. For hypothesis testing, we use asymptotic inference for MEM and CPI and permutation inference for NPWP and SC (20)(21)(22). Permutation inference was not done for the MEM and CPI methods because of the high computational burden of these methods.…”

“…SWTs in particular are highly susceptible to misspecification and can produce biased results when time trends and time-intervention interactions are not modeled correctly (18,19). Type I Error of hypothesis tests can be preserved by using permutation-based inference, but this generally results in reduced power (20)(21)(22). New methods for analyzing SWTs have recently been proposed which preserve Type I Error but may have more precision and higher power than permutation tests based on mixed effects models when applied to data-generating settings for which mixed effects models are misspecified.…”

“…was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. that compare within-cluster differences between two time points across clusters (22).…”