Analysis time-varying treatment effects in Stepped-wedge cluster randomized trials

Most current analysis methods for Stepped-wedge cluster randomized trials (SW-CRTs) assume a constant treatment effect, even though emerging evidence suggests this assumption is often violated in practice. Given this background, we collected a large number of de-identified datasets from stepped wedge trials, then we analyzed each dataset by methods that do and do not rely on the assumption of a constant treatment effect and compared the results. We developed an R package to compute robust variances for generalized linear mixed models fitted using the ‘glmer’ function. I also led a large-scale simulation study to evaluate the package’s performance in handling time-varying treatment effects. The package is currently in the process of publication to CRAN.