On the 6 March 2014, the RSS group meeting was hosted at the University of Warwick. The speaker for the session was Alan Girling from the School of Health and Population Sciences at the University of Birmingham. Alan’s talk was titled: 'Statistical considerations for the design of Stepped Cluster Trials'.
The talk began with an overview of some of the popular designs for cluster trials. Alan presented the common design structures in an easy to interpret fashion using a number of graphics. Stepped wedge designs were then introduced as the focus of the rest of the talk. The stepped wedge design is a balanced alternative to the ANCOVA design, comprised of a before and after study in every cluster with staggered implementation. A general benefit of the stepped wedge design over a parallel cluster design is that the power of a stepped wedge study can be increased without limit by taking larger samples within clusters, without increasing the number of clusters.
Alan went on to talk about the optimal structure of a stepped wedge design. Optimality in this sense requires maximising the precision of the treatment effect estimate via selection of the experimental intervention uptake points for each of the clusters.
The best balanced design was presented as a stepped wedge design in which some clusters have the uptake point at the very beginning and some clusters have no uptake point. This effectively creates some parallel treatment and control clusters, with the remainder following a standard stepped wedge design. The exact number of parallel and stepped wedge clusters is dependent upon the number of experimental units in each cluster and the intra-cluster correlation. Designs of this form can dominate the standard stepped wedge design in terms of relative efficiency and produce greater relative efficiency than the standard parallel design for particular values of the intra-cluster correlation and number of experimental units in each cluster.
Alan also presented a second optimised stepped wedge design under general time-dependent, cluster-specific sampling rates. In this case it is optimal to sample just before and just after implementation and also the two extreme cases, sampling the cluster which receives the intervention first at the last time point and sampling the cluster which receives the intervention last at the first time point.
As a particular example, Alan introduced the EPOCH (Enhanced Peri-Operative Care for High-risk patients) trial. This trial aims to investigate quality improvement for emergency laparotomy surgery using a stepped wedge design. Each cluster within the trial is comprised of a group of hospitals in the same geographical region and the intervention is the introduction of a newly unified treatment approach.
Alan concluded by summarising the two kinds of optimal design, describing the hybrid parallel and stepped wedge design as offering a practical alternative to standard designs and stating that the second optimal design offers mostly insight. In both cases the best design depends upon both the cluster size and the intra-cluster correlation, both of which need to be focused on when designing a cluster study.