What Should Be Included in Meta-Analyses?

OBJECTIVE: To explore the impact of methodologic issues on the results of meta-analyses. The following issues were examined: the type of literature search strategy used; inclusion or exclusion of non-peer-reviewed studies; the inclusion or exclusion of non-English language publications; the effect of trial quality; and the inclusion or exclusion of non-placebo-controlled studies. METHODS: The International Study of Perioperative Transfusion (ISPOT) meta-analyses were used to evaluate each of the methodologic issues. The 10 meta-analyses consisted of technologies to reduce the need for perioperative red blood cell transfusion. The number of trials for each of the meta-analyses varied from 2 to 45. Both EMBASE and MEDLINE searches were conducted, including the use of systematic search strategies. RESULTS: MEDLINE identified the vast majority of trials. Alone, MEDLINE would have missed 8 studies compared to 10 for EMBASE. Use of the systematic search strategies greatly reduced the number of articles to be reviewed compared to open searches. Type of publication, country of study origin, inclusion of non-English publications, and trial quality had very little impact on the estimates of effect. The use of placebo versus open-label control affected the magnitude of the odds ratio for two of the meta-analyses. The results of the two meta-analyses were not statistically significant if only placebo-controlled trials were included. CONCLUSIONS: While methodologic issues had very little impact on the ISPOT meta-analyses, further studies are needed in a variety of other clinical settings. Because MEDLINE, coupled with a review of the references in the identified trials, identified the vast majority of trials, one needs to consider the costs and benefits of searching EMBASE and the pursuance of unpublished and unindexed trials.