Collaborative worldwide overviews of randomized trials.

Abstract

Formal overviews (or meta-analyses) are now widely accepted as the most reliable way to evaluate the evidence from several randomized controlled trials that have all assessed a particular form of therapy. If a limited amount of trial data has accumulated, overviews can be undertaken at a relatively simple level, assembling just summary data extracted from published reports. Such overviews are likely to be incomplete and biased and may (because of the restricted number of analyses that are possible) not answer all the clinically important questions that might be addressed. Where a particularly large body of data from randomized trials has accumulated, more thorough and detailed overview analyses are needed. Such detailed reviews are greatly facilitated if a collaborative group of all the trialists is formed. Such groups have sought to collate individual patient data (a very few key items for every patient randomized). A central statistical secretariat then coordinates the process of data collection, checking, and analysis. Analyses are presented to the whole group for discussion and final reports are published in the name of the whole group. Experience from two very large groups that have followed this model, the Antiplatelet Trialists' Collaboration and the Early Breast Cancer Trialists' Collaborative Group, has shown that detailed collaborative overviews have many benefits: the results are particularly clear and therefore have substantial public health impact; the areas of statistical and medical agreement on the evidence can be defined; the areas of uncertainty (and hence future research priorities) are clarified; and the group can disseminate the results particularly widely and rapidly.