Background: While conducting two systematic reviews on preoperative fasting (adults and children) we became aware of a difficulty in including all the relevant trials within the meta-analyses. The primary outcomes of residual gastric volume and pH are known to have naturally skewed distributions. Of the 81 relevant comparisons, mean and standardised difference (SD) summary statistics were not available from 28 and thus could not be included in the meta-analyses using standard methods. In many cases these comparisons were from large well-conducted randomized controlled trials (RCTs) and to exclude them from the meta-analyses would be detrimental to the review.

Objectives:

- To develop a new approach to allow inclusion within the meta-analyses of trials that report summary data other than means and SD.

- To compare this approach to various simple imputation methods.


Methods: Many question the appropriateness of summarizing skewed data with means and SDs, but standard methods of metaanalysis require these quantities to check for consistency of effect estimates and (if appropriate) pooling. A simple, but not as yet validated method, would be to impute the means and SDs from the available summary data. Alternatively, a general mathematical approach (based on likelihood methods) can 'correctly' check for consistency between studies and pool studies extracting all information from the available summary data (O'Rourke 2002). The connection between the two approaches can be made direct by calculating 'pseudo-mean' and 'pseudo-SD' values that closely reproduce the same contrasts and poolings of the studies. These can then be compared with other imputations of means and SDs. These 'pseudo-values' can also be used to carry out the contrasts and poolings of the studies in RevMan.

Conclusions: We were able to contrast and pool all the relevant comparisons in the reviews using the 'pseudo-values' calculated from the likelihood based approach. We have yet to determine how closely other imputation methods match.