Meta-analysis of observational epidemiological studies: a review.

Meta-analyses (integration of findings by quantitative analyses of results of individual studies) are already widely used in the psychological and educational sciences and in the pooling of clinical trial results. Examples of the application of such techniques to the results of observational epidemiological studies are now proliferating. In this paper meta-analysis of the results of observational epidemiological studies is reviewed. Uncritical adoption of techniques used in pooled analyses of clinical trial results may be inappropriate. Some alternative approaches, including meta-regression techniques, are discussed, and illustrated with reference to examples of meta-analyses of studies of breast cancer risk following oral contraceptive use and of patterns of post-bereavement mortality. Although substantial difficulties beset the use of meta-analysis in epidemiology, many of these problems are also implicit in the execution of traditional, narrative reviews. Foremost among these difficulties are those associated with publication bias and with making due allowance for the quality of the studies being combined. Unlike traditional narrative reviews, meta-analyses require explicit statement of the criteria for the review and hence highlight these difficulties. Nonetheless, careful and critical application of appropriate meta-analytical techniques facilitates quantitative exploration of inhomogeneities in, and (where appropriate) synthesis of study results.