Statistics notes: Sample size in cluster randomisation
BMJ 1998; 316 doi: https://doi.org/10.1136/bmj.316.7130.549 (Published 14 February 1998) Cite this as: BMJ 1998;316:549
All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
Sir,
I welcome the contribution of Kerry and Bland on cluster randomization (BMJ 1998; 316: 549). As long as our world contains more patients than doctors, this issue should be taken into account when planning a study.
The example they present, however, may be somewhat misleading. The proposed study addresses the effect of two interventions on cholesterol level. One is an intensive dietary intervention by practice nurses, the other is usual general practice care.
The data they use to estimate clustering are on cholesterol levels from a thrombosis prevention trial. It is unclear whether these are pre- or post-intervention data, but in both cases I doubt whether the magnitude of clustering of cholesterol concentrations in this study is a reliable estimate for the proposed study. I would not be surprised when the clustering in the practice nurse intervention would turn out to be much higher, as this may be much more affected by personal factors.
Johannes C van der Wouden
Department of General Practice
Erasmus University Rotterdam
PO Box 3000
1738 DR Rotterdam The Netherlands
email: vanderwouden@hag.fgg.eur.nl
fax + 31 10 436 07 17
tel + 31 10 408 76 11
Competing interests: No competing interests
Re: Intracluster correlation underestimated?
Dr van der Wouden makes a good point. In any sample size calculation, care should be taken to choose appropriate values for the variances. Such calculations inevitably involve some guess-work, however, and are always approximate. It is a good idea to consider the implications of greater variability than anticipated. We should also err on the side of caution. Researchers are often criticised for taking a sample which is too small, seldom for taking one which is too large.
It may well be that in this case variation between clusters, and hence the intracluster correlation coefficient, would be increased by the presence of a different treatment provider in each cluster. In fact, in a study with a similar intervention, as yet unpublished, we obtained values very like those cited.
We need more information about the variances and intracluster correlations found in different types of cluster randomised trials, to help future trials to be better designed.
Competing interests: No competing interests