Pignon J P, Arriagada R
Department of Biostatistics, Institut Gustave-Roussy, Villejuif, France.
Lung Cancer. 1994 Mar;10 Suppl 1:S135-41. doi: 10.1016/0169-5002(94)91675-6.
If meta-analyses of randomized trials were better in quality and were correctly interpreted, then there would be far less reasons for criticisms. The large number of meta-analyses to date are ample evidence of the usefulness of this technique in giving a synoptic appraisal of apparently discordant trials, but their shortcomings should be known. The first step needed to optimize the quality of meta-analyses is to improve the quality of randomized clinical trials. A registry of randomized trials is the best way to decrease publication bias. Meta-analyses based on individual data provided by each trialist allow a better appreciation of the quality of a trial, an increase of the statistical power, and for some covariates to be taken into account; such is not generally the case for those based on published data. Trialists should collect and conserve relevant data on ongoing trials and ensure that these trials are registered. These data would serve for the conduct of future unbiased meta-analyses, the duration of which would be reduced considerably, and unnecessary trials would be avoided.
