Table 4 Key considerations for designing a stepped wedge trial
Consideration | Why this is important |
|---|---|
Total number of measurements acceptable | Requires either a design with fewer steps, or an incomplete design • Both reduce power (especially designs with fewer steps) • Seasonality increases variability Incomplete designs: • can be more complicated e.g. calculating power Seasonality can introduce bias so important to consider the timing of measurements |
Duration of measurement period | • Very short measurement periods can be logistically difficult to ensure all data collection occurs within the required timeframe • Larger measurement periods (for the same study duration) will reduce the maximum number of steps possible, which reduces power Very wide measurement periods will introduce more variability and possible bias when seasonality is present |
Implementation period | Requires an incomplete design: • can be more complicated e.g. calculating power • Seasonality can introduce bias so important to consider the timing of measurements May also affect the duration of measurement period (see above) as the implementation time will need be one or more measurement periods |
Other restrictions on timings | • Constraints mean that the most powerful configurations may not be feasible • When seasonality is present, may introduce variability and/or bias due to gaps at key times which affect coverage and control/intervention balance (see below) |
Coverage of full study duration | For incomplete designs, measurements should cover the full study duration (i.e. avoid staggered designs) to accurately estimate any time trends or seasonality |
Control-intervention balance within measurement periods | Where possible, incomplete designs should have a balance of measurements under both control and intervention conditions to be able to separate time trends/seasonality from intervention effects |
Modelling of time | • All stepped wedge analyses should include a time effect, but how this is modelled becomes more important in the presence of time trends • Seasonality and/or time trends increase variability, which will reduce power– better modelling of time will reduce this extra variability Seasonality and time trends can introduce bias if not modelled correctly which can be substantial e.g. larger than effect sizes |
Class sizes | • restricts the maximum number of pupils possible to recruit, which reduces maximum power than can be achieved • smaller class sizes may therefore require more schools, and/or more pupils with repeated measures and so higher retention rates are needed (see below) |
Repeat measurements: Cross-sectional, open or closed cohort | • repeated pupil measures (open or closed cohort) will increase power • may be harder to retain existing pupils than recruit new ones, especially over longer study durations closed cohort may not be possible in some situations e.g. longer studies when pupils move up a year/to a different school |
Recruitment and retention rates | • dictates the number of pupils per school in each measurement period • more pupils increases power |