[Skip to Content]
[Skip to Content Landing]
December 5, 2001

Using Meta-analysis to Answer Clinical Questions

Author Affiliations

Stephen J.LurieMD, PhD, Senior EditorIndividualAuthorJody W.ZylkeMD, Contributing EditorIndividualAuthor

JAMA. 2001;286(21):2669-2670. doi:10.1001/jama.286.21.2665

To the Editor: Dr Ioannidis and colleagues performed a meta-analysis of previously published meta-analyses to address the results of randomized vs nonrandomized studies.1 For clinicians who are not familiar with this method, a number of questions are likely to arise.

First, do meta-analyses yield reliable results? If so, this implies that biases of various studies cancel each other out through meta-analysis. Is this assumption generally accepted? Second, how does meta-analysis avoid compounding previous errors, especially in an environment in which peer influence is considerable? It appears that a meta-analysis may serve only to make scientists aware of their colleagues' research and may not substitute for original work. In this case, what does a meta-analysis of meta-analyses stand for? The meta-analyses presented by the authors originate from a small spectrum of medical specialties and institutions. They are certainly neither random in their inception nor cross-sectional in their representation. Is it possible that meta-analyses ad infinitum only perpetuate or even amplify faults of the original studies? Finally, was all this analysis really worth the effort just to reach the conclusion that there are discrepancies between randomized and nonrandomized studies?

First Page Preview View Large
First page PDF preview
First page PDF preview