Publication and the meta-analysis of clinical trials

Publication bias is the tendency on the part of investigators, referees, and editors to favor publication of results of studies rejecting the null hypothesis. Although it has not been studied extensively in the field of medicine, the bias is generally believed to exist. Data from a survey of authors who had published clinical trials have provided evidence for its existence: authors claimed that 55% of their published trials and 14% of their unpublished trials had results favoring the test treatment. On the other hand, a recent survey of obstetricians and pediatricians did not identify a large number of unpublished trials.; One solution to the problem of publication bias is to register trials at the outset (prospective registration) so that information regarding all trials undertaken is available, regardless of subsequent publication status. There are a number of subject matter-based trials registers in existence, the Oxford Database of Perinatal Trials being one of the best developed. It has been used as a basis for numerous meta-analyses, including one on the pharmacological control of pain in labour, presented herewith.; Recently, there has been interest in meta-analyses of clinical trials, the combining of data from similar but separate studies in order to reach a conclusion regarding treatment efficacy. If the results of meta-analyses are being used to set policy with regard to treatment, then publication bias could have a serious impact on patient care.; A further problem is that it may be difficult to identify all published trial results in a given area; this would result in a meta-analysis based on incomplete data. For example, MEDLINE searches typically retrieve fewer than half of the available articles describing clinical trials. A related, though less common, problem is that data that were in error or that were retracted may be included in meta-analyses. In fact, the latter problem has recently been addressed: MEDLINE indexing now identifies both retracted articles and correction notices.; Meta-analyses themselves must be identifiable if they are to be of use to those making treatment decisions. A number of different terms are used synonymously with meta-analysis--a fact that hampers MEDLINE searches. Agreement by investigators on standard terminology is necessary to facilitate both communication and retrieval of the published meta-analyses.