Multiplicity in systematic reviews and meta-analysis: Dealing with multiple source multiple outcomes

Abstract

Publication and reporting bias are well-documented in the scientific literature. Increased data and code sharing, and access to other sources of information such as Clinical Study Reports (CSRs), address concerns about the non-reproducibility of individual studies. Ironically, greater transparency has given rise to new problems. That is, systematic reviewers and meta-analysts can choose from among dozens of effect sizes that could be included in their analyses. Initiatives that increase validity and reproducibility in individual studies also create opportunities for bias in research synthesis and clinical guideline development. Scientists could adopt new methods to avoid cherry-picking at all stages of research and evidence synthesis.