A simple method to correct for the design effect in systematic reviews of trials using paired dichotomous data

Abstract

Background and objectives: In systematic reviews of interventions of studies where randomization was done by individual but data are paired (such as eyes, ears), it is necessary to account for the natural clustering present. The Cochrane Handbook suggests treating these as examples of cluster randomized trials. An incorrect analysis (without adjustment) would usually overestimate the precision of the estimate. We discuss a simple method of adjustment that deals with this problem.

Methods: From a cross-tabulation of the event being present on the "left" and "right" body part, we estimate the design effect which is a measure of the inflation on the variance due to clustering. This estimate is then used to obtain an adjusted effect size per trial by reducing the number of events and the sample size in each intervention group.

Results: In a systematic review on Auto-inflation for Glue Ear, data on improvement were obtained for pairs of ears. The design effect obtained from these data was 1.25. In a meta-analysis, the weights given to the trials changed after adjustment from 33% to 11% in one case.

Conclusion: In a systematic review, when dealing with paired data, it is possible to give adequate weighting to each trial using a simple adjusting method.