Does blinding affect the results of meta-analyses?

Berlin J., Miles C., Cirigliano M., Conill A., Goldmann D., Horowitz D., Jones F., Scott E., Hanchak N. and Williams S. University of Pennsylvania School of Medicine, Philadelphia, PA, 19104, USA.

Introduction/Objective: To determine whether blinding readers of meta-analyses to identifying information in original papers has an effect on the results of the meta-analyses.
Methods: Readers were randomly assigned to perform meta-analyses under either blinded or unblinded conditions. Papers given to blinded readers were electronically scanned from published papers, reprinted in a common type face, and stripped of identifying information about authors, institutions, journals of publication, and treatment assignments. Five previously published meta-analyses were chosen at random, and we applied the original inclusion and exclusion criteria for the published studies. The Mantel-Haenszel summary odds ratio was calculated for each meta-analysis separately for blinded and unblinded conditions and compared using inverse-variance-weighted, least-squares regression, taking the pairing of observations into account.
Results: The summary odds ratio for the five, blinded meta-analyses was 0.63 (95% Confidence Interval (CI): 0.57, 0.69), compared with 0.64 (0.57, 0.72) for the unblinded meta-analyses (p = 0.39 for the blinded versus unblinded comparision). The blinding process was extremely time-consuming.
Discussion: Blinding readers to identifying information in original papers appears to have little effect on the summary odds ratio calculated for the meta-analysis, and it is time-consuming. Therefore, we recommend that blinding not be used when meta-analyses are performed.

Amsterdam 1997 016.01