Table 1 Key general characteristics of the stepped wedge design and their implications
Characteristic | Implication |
|---|---|
Randomization is usually at the cluster level | Statistical analyses need to take into account that measurements of subjects within a cluster may be correlated |
| Â | Concealment of allocation will not always be possible. Blinding of outcome assessment is therefore more difficult to achieve |
Cross-over element: each cluster will switch from control to experimental intervention | The cross-over allows for a within-cluster comparison which may increase statistical power |
| Â | Sample size calculations as well as analyses become more complex |
Two subtypes: | Â |
 - switch involves the same patients (cohort-type) | Cohort-type SWD allow for within-patient comparison, which may further increase efficiency, but critical evaluation whether carry-over effects may compromise the results of the study is necessary |
 - switch involves different patients (cross-sectional type) |  |
Switch from control to experimental intervention is spread over calendar time | A research team can plan and execute the switch in treatment in a dedicated way as not all clusters switch at the same point in time |
| Â | Interim analyses need to take into account that the number of measurements in the control and intervention groups are very imbalanced at early stages and will only be comparable at the end of the study |
| Â | It offers the possibility to assess changes in cost-effectiveness over time when the uptake of interventions is difficult or slow due to implementation barriers that need to be overcome |
| Â | A study with an SWD may need a relatively long time to complete |
All clusters will experience the experimental intervention | This feature may enhance participation of clusters in the study |
| Â | The switch in each cluster allows investigation and monitoring of implementation problems |
Fixed design in which all clusters start at the same point in time and all steps have the same time span | Preparations for data collection need to be finished in each hospital which can easily delay the start of the study |
| Â | Lower than anticipated inclusion rates increase the risk for an underpowered study as solutions like adding more clusters or extending the length of the remaining steps seriously affect the design and are not recommended |