Stepped-wedge design

A stepped-wedge design is a cluster-randomised trial in which all clusters start in the control condition and cross over to the intervention one at a time, or in groups, at randomly assigned times, until every cluster has switched. Drawn as a chart it looks like a staircase, hence the name. It suits interventions that everyone expects to adopt eventually and that cannot practically be rolled out everywhere at once: a new care pathway, a surveillance policy, a change in surgical practice, a training programme. Because every cluster gets the intervention, it is easier to recruit sites and to win ethical approval than a design that withholds the intervention from half of them for good, and the comparison of before-and-after within each cluster is combined with the comparison between clusters at each time point.

SANO is the worked example and a landmark. Dutch hospitals treating oesophageal cancer with chemoradiation followed by surgery switched, in randomised order, to a policy of active surveillance for patients whose tumour had disappeared completely on reassessment, operating only if the cancer regrew. Two-year overall survival was non-inferior (hazard ratio 1.14), about a third of surveillance patients had avoided oesophagectomy at two years, short-term quality of life was better, and regrowth was usually still resectable. The design let a national network test organ preservation without asking any hospital to refuse it to its patients indefinitely.

The design's weakness is time. Because intervention periods are, on average, later than control periods, anything else that changes over the trial (new drugs, better imaging, a pandemic) is partly confounded with the intervention, and the analysis must model the time trend explicitly. Stepped-wedge trials also take longer than a parallel cluster trial of the same size and are statistically less efficient when the intervention effect is large. They are trials of implementation as much as trials of efficacy, and their results should be read that way.