Meta-analysis of randomized clinical trials is gaining wide acceptance in the clinical and statistical literature. The application of meta- analytic methods has recently been extended to observational epidemiologic and uncontrolled studies. Because of the potential for bias to arise in the conduct of meta-analyses, several authors have recommended that readers evaluating or extracting data from original papers for a meta-analysis be masked (blinded) to certain information in the paper, including: the identity of the authors of those papers, the institutional affiliations of the authors, study results when evaluating the quality of the methods, and, in randomized trials, even the identity of the treatment and control groups. While this recommendation has high face validity, no study has yet evaluated whether blinding actually leads to the avoidance of bias. The principal goal of the proposed research is to conduct a randomized trial comparing blinded and unblinded conduct of a series of meta- analyses of randomized clinical trials. Specifically, the following hypotheses will be tested: 1. Meta-analyses conducted with readers blinded to identifying information produce systematically different results than those same meta-analyses conducted using unblinded readers. 2. The variability in the results of a given meta-analysis conducted independently by two pairs of readers will be higher between members of those pairs when the readers are blinded than when they are unblinded. 3. On the level of the individual study, both the decision to include or exclude particular studies from a meta-analysis and the exact values of data extracted from papers will depend on whether readers are blinded or not. Blinding will lead to systematic differences and increased variability between readers when compared with unblinded reading of studies.
