Why negative meta-analyses may be false?

摘要

Results of meta-analyses are regarded as the highest level of evidence. A statistically non-significant effect size from a meta-analysis is typically considered true negative even in the presence of a statistically significant signal in individual studies, presumed to be false positive. Here we provide examples from neuroimaging, genetics and psychopharmacology of why meta-analyses may frequently yield false negative results from true positive findings. This may happen in situations when individual studies report findings in opposing directions, the sum of which yields a non-significant overall effect size. Such non-significant meta-analyses, which show statistical heterogeneity and include studies with opposing effect sizes do not provide an accurate estimate of the overall effect and may have lower heuristic value than individual studies. Over reliance on such meta-analyses may falsely identify certain potentially fruitful research avenues as blind alleys.