Separate calendar time from stepped-wedge exposure

Draw the actual rollout

In a stepped-wedge cluster randomised trial, clusters begin an intervention at different assigned times. Later observations are more likely to be exposed than earlier observations. A simple before-and-after comparison can therefore mix the intervention effect with background changes over calendar time.

The CONSORT stepped-wedge extension calls for transparent design diagrams, dates and methods that account for time. Start with a cluster-by-period table showing planned crossover, actual implementation and the observations collected. Distinguish repeated measurements of the same participants from repeated cross-sectional samples.

Specify the time and exposure variables

Define calendar periods consistently across clusters. Record transition or implementation periods and explain whether the intervention is expected to act immediately or after a delay. Separate calendar time from time since exposure: an intervention that strengthens with experience cannot always be represented adequately by a single on/off indicator.

Write the analysis rule for deviations from the randomised schedule. An analysis by assigned crossover date and an analysis by actual implementation answer different questions and may have different bias risks. Do not silently substitute actual dates because they make the data easier to fit.

Hypothetical example

Imagine six fictional clinics introducing a programme in three rollout steps. A national policy change occurs during the second step and affects every clinic. Outcomes improve after that date even in clinics that have not yet started the programme. Attributing the whole improvement to the programme would ignore a plausible calendar effect.

The useful diagnostic is a period-by-clinic display, including observations before intervention, rather than one pooled line labelled “before” and “after.” The example identifies a design issue; it does not establish that a particular adjustment model will eliminate all bias.

Check what the data can support

Review the number of clusters, observations per cluster-period and missing periods. Specify how within-cluster correlation and repeated participants are handled. A single design-effect formula from a parallel cluster trial does not automatically describe a stepped-wedge design.

Inspect sensitivity to plausible time specifications and exposure-effect patterns without selecting the preferred model by significance. Report the chosen assumptions, estimated effects with uncertainty and deviations from the schedule. Use the multilevel data and cluster planning worksheet as a starting record, then add the stepped-wedge timing details explicitly. Keep conclusions cautious when few clusters, sparse periods or closely aligned external changes limit the separation of time and intervention.

Source verification: 6 November 2026. Examples are hypothetical.