Investigating Heterogeneity Across Studies - A Review of the Systematic Reviews in the Infectious Diseases Module of the Cochrane Library

Article type
Year
Authors
Hahn S, Gamer P, Williamson P
Abstract
Introduction: In a meta-analysis of a set of trials, an important issue is whether heterogeneity of treatment effect exists between trials . In the presence of heterogeneity, pooling estimates from a series of studies without accounting for such heterogeneity can result in a misleading conclusion. Therefore, reviewers should determine whether the trials can reasonably be described as sharing a common treatment effect before combining them statistically. In a systematic review, the reader should be made aware of significant heterogeneity and the analysis should attempt to investigate the reasons for it.

Objectives:

Methods: We have examined the 24 completed systematic reviews and their original protocols in the Infectious Diseases Module of the Cochrane Library in the issue 3, July 1998, with respect to this issue of heterogeneity. Each protocol was reviewed to see how reviewers have discussed the issue of heterogeneity, noting relevant terms used, sources of heterogeneity suggested and methods proposed for investigating any heterogeneity found. The review is then reviewed for those items above together with an assessment of whether heterogeneity is evident, whether it is noted by the reviewers and how it has been investigated. We focus only on the primary outcome for each review.

Results:

Discussion: Methods used to address heterogeneity in the systematic reviews examined included subgroup analysis, meta-regression, reference in text to estimates from random effect model, and sensitivity analysis. Our findings show that subgroup analysis is the most common method used for investigating heterogeneity in these systematic reviews. We also noted that the choice of subgroups is rarely justified in the protocol and the subgroup data are often not available for all trials included in meta-analysis, when subgroups defined by patient factors. A possible bias in meta-analysis due to selective reporting of subgroup analyses within studies was suggested as a potential problem that may arise when subgroup analyses were conducted with missing subgroup information for some trials included in the meta-analysis. These results have important implications for editors and referees of Cochrane reviews.