[Meta-analyses. A tool for maximizing therapy study evidence?]

Medical decisions have to rely on evidence-based medicine. However, evidence is not a clearly defined term. Several evidence criteria have been proposed, yet statistical meta-analyses are considered to be core criterion of empirical studies in that they combine single evidence from different studies into an encompassing conclusion. We argue however that meta-analyses have major limitations in this context. Five arguments are presented: (1) the nature of a study design affects the results and thus comprehensive meta-analyses covering studies with different designs are not informative; (2) combining different studies into a meta-analysis is methodologically not informative; (3) meta-analyses consider the variance between studies not as informative but as random noise; (4) the strategy to identify informative studies is a decisive determinant of meta-analyses; and (5) the value of conclusions from meta-analyses depends on the choice of statistics chosen as combined results from the various single studies. Instead, meta-analyses are useful tools for generating hypotheses in a posteriori analysis.