Power analysis for concurrent balanced or imbalanced multiple-intervention stepped wedge design: a simulation-based approach

Abstract

Background

The concurrent multiple-intervention stepped wedge design (M-SWD) is one of the most widely used variants of the SWD. We aimed to conduct power analysis for concurrent balanced (equal number of clusters in intervention groups) and imbalanced (unequal number of clusters in intervention groups) M-SWDs.

Methods

We conducted power analysis using a simulation-based approach with cross-sectional or closed-cohort designs and examined impact of design parameters (cluster size and number of clusters) and correlation parameters (total random effects variance (TRE), cluster autocorrelation coefficient (CAC), and individual autocorrelation coefficient (IAC)) on the powers of statistical tests for treatment effects.

Results

With a fixed total sample size, increasing the number of clusters improves statistical power. When two treatment effects differ greatly, the concurrent imbalanced M-SWD saves sample size compared to the balanced design and powers could achieve the target value when the ratio of clusters approximates the inverse ratio of two effects. However, the allocation ratio should be no greater than 4:1. Additionally, statistical powers increased with decreasing TRE and increasing CAC and IAC. The impact of autocorrelation coefficients on powers is more pronounced when these parameters are large.

Conclusion

When two treatment effects differ greatly, the concurrent imbalanced M-SWD, with an allocation ratio no larger than 4:1, is a preferred design over the balanced one. For both concurrent balanced and imbalanced M-SWD, it is recommended to set large number of clusters with small cluster sizes and to carefully consider estimates of correlation parameters when designing the trial.