Background: Standardised mean differences (SMD) are often used in statistical meta-analyses to combine outcomes measured using different scales. They are calculated by dividing the difference in post-intervention means between two groups by the pooled standard deviation of the outcome. Methods for calculating this statistic from studies which allocate participants individually are well-known, but extensions to trials which allocate participants by cluster are less well known. Use of such trials in meta-analysis is often hampered by incomplete reporting and analysis carried out without taking clustering into account.
There are few published sources to assist reviewers in calculating SMDs of these trials. This was highlighted when SMDs were being calculated for a recent systematic review and explicit formulae were developed to facilitate its meta analysis.

Objectives: To develop methods for calculating the SMD from a variety of different published data and outline the potential pitfalls.

Methods: Formulae for calculating the SMD and its associated standard error from published sources were examined and adapted to take account of cluster randomisation.

Results: When the standard deviation between individuals is not reported, we show how to compute the SMD using either the standard deviation between cluster means, the standard error on the difference between two means, or the standard error on the intervention effect computed by an analysis of covariance which adjusts for baseline values of one or more covariates. We also show how to use an assumed intra-class correlation to adjust a reported analysis that entirely ignores the clustering. Examples from a systematic review of children and healthy eating are presented showing the correct application of the above methods.

Conclusion: When calculating SMDs from trials which allocate participants by cluster, the clustering must be allowed for and the correct standard deviation must be used. The calculation can be done using a wide range of published results.