Stepped wedge designs

Stepped wedge designs are a promising design option for implementation science, simultaneously satisfying ethical, logistical and causal inferential objectives for the investigation of the effectiveness of interventions at scale which have already been shown to be efficacious. We revised the manuscript using a maximum likelihood approach to calculate power for SWDs of binary outcomes with identity link. In the revision, we incorporate time effects in the framework. When time effects are included in the model, the power increases with increasing number of steps given cluster size and sample size. It is different with the model without time effect. When time effects are negligible, the model without time effects is much more powerful than the model with time effects. Though it is often the case that the time effects are believed to be small or negligible during the study period, particularly with studies of short duration, caution is needed. If we assume that there are no time effects at the design stage, and a time trend is found, the estimated intervention effect will be biased unless it is adjusted for time in the analysis. However, adjusting for this unanticipated time trend at the analysis phase will likely lead to an underpowered study.