Stratified randomisation: a hidden form of clustering?
© Kahan and Morris; licensee BioMed Central Ltd. 2011
Published: 13 December 2011
Many randomised trials use stratified permuted blocks or minimisation to balance key prognostic variables between treatment groups. It is widely argued in the statistical literature that any balancing variables should be adjusted for in the analysis, however a review of major medical journals shows that this is not commonly done. Our objective was to determine the effects of an unadjusted analysis after balancing.
The statistical properties of an unadjusted analysis after balancing are explored using theoretical results. A major simulation study using data from 5 trials is performed to determine the potential impact in real life situations.
We show that balancing on baseline covariates leads to correlation between the treatment groups (similarly, cluster randomised trials lead to correlation within treatment groups). If this correlation is ignored, and an unadjusted analysis is performed, the estimated variance of the treatment effect will be biased upwards, resulting in type I error rates that are too low, and a reduction in power. Conversely, an adjusted analysis results in nominal type I error rates, and optimal power.
Prognostic variables that have been balanced between treatment groups in the randomisation process should be adjusted for in the analysis. Unadjusted analyses lead to invalid results, whereas adjusted analyses maintain nominal properties.
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.