Mary,

To be sure I understand: Each of the 6 clusters receives the intervention at a different point in time. So clusters 2-6 are controls for cluster 1 in the first 6 months, and it gets more complicated from there. Is that correct?

Analysis strategies depend on other information. My first question would be how often are your outcomes measured? Second, are the outcomes meaningful at a within-cluster level, or are they cluster-level outcomes?

Pat

Hi Patrick,

You are right; during the first 6 months of the program cluster 1 receives the intervention, while clusters 2-6 are control clusters. In the second 6 months, clusters 1 and 2 are receiving the intervention, while clusters 3-6 are controls. And so on.

There will be monthly site-level data collection of routine indicators in intervention areas, and quarterly cluster-/district-level data collection of routine indicators in non-intervention areas. This is the plan thus far.

The clusters are in fact districts of (unequal sizes), so the terms cluster and district can be used interchangeably. By the term site, I mean health facility.

Thanks,

Mary

a regression model can be used with a random effect for the cluster, a covariate for the

calendar time period, and a variable to indicate in which cluster and time periods the

intervention was applied. There is a description of this and other statistical approaches in the paper:

MA Hussey and JP Hughes (2007) Design and analysis of stepped wedge cluster randomized trials. Contemporary Clinical Trials 28 (2007) 182–191

This review published recently in BMJ is also useful: The stepped wedge cluster randomised trial: rationale,design, analysis, and reporting, at:

http://www.bmj.com/content/bmj/350/bmj.h391.full.pdf

An extension to the CONSORT guidelines to cover stepped-wedge designs, is being developed:

http://www.equator-network.org/wp-content/uploads/2009/02/Consort-SW-Protocol-V1.pdf