gms | German Medical Science

Stepped wedge cluster randomized trials (SWCRT) are a versatile alternative to parallel cluster designs and are becoming an increasingly popular tool in health services research for evaluation of complex interventions. The power of tests of treatment effects in stepped wedge designs is known to be less sensitive to misspecification of the intra cluster correlation (ICC), but – unlike parallel cluster randomised trials - SWCRT need to be adjusted for potential time trends.

Analytical asymptotic formulae for the power of a Wald test of an intervention effect in an SWCRT can be derived for given ICC (and other random effects) as done in Hussey, Hughes [1] and Baio et al. [2]. The latter also demonstrate with some simulations that these formula based power calculations are feasible for continuous, binary, and count data.

However, in mixed models often used for analysis of SWCRT, the proposed procedures tend to be anti-conservative. Hussey and Hughes [1] mention this issue and propose a fix by jackknife variance estimation. We therefore investigate, if and when standard errors have to be adjusted in order to control type I error and how the power is affected.

For a set of SWCRT designs and parameter settings and analysis methods, we compare the performance of HH analytical power formulae, and simulations based power calculation methods. We consider models with and without a random cluster by treatment interaction (random slope) and fitting methods for generalized linear mixed models with treatment effect estimates with and without resampling adjustments of standard errors.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.