ABSTRACT
Until about 20 years ago reviews of the biomedical literature were generally ad hoc activities. A reviewer would typically search out a few of the relevant articles (o en from a personal collection of reprints), supplement these articles with some of those quoted in the reference list of the published papers, and produce a review. It was invariably idiosyncratic and incomplete; and it was an unreliable basis for drawing general conclusions. It is therefore hardly surprising that an analysis of 40 reviews published in the mid-1980s found major deciencies.1 Only one described the methods for identifying, selecting and validating the information included in the review; and just three of them attempted a quantitative synthesis (meta-analysis) of the results.
