Metadbias

Metadbias, or meta-bias, refers to biases that affect the results of a meta-analysis either through biases in the body of evidence or through the methods used to synthesize that evidence. It encompasses biases that operate across multiple studies rather than within a single study, and it can influence the overall conclusions drawn from a systematic review.

Common forms of metabias include publication bias (studies with significant or favorable results are more likely to be published), language bias (non-English studies are underrepresented), outcome reporting bias (studies report only selected favorable outcomes), time-lag bias (older negative results are published more slowly), and citation bias (widely cited studies receive greater attention regardless of quality). Other sources include selective inclusion criteria, outcome-switching within trials, and industry sponsorship affecting study availability or reporting.

Metadbias can distort pooled effect estimates, inflate precision, and lead to misleading conclusions about the magnitude or existence of an effect. It can undermine the reliability of evidence syntheses and the decisions based on them.

Detection and adjustment methods include funnel plots, and statistical tests such as Egger’s and Begg’s tests, as well as more advanced approaches like trim-and-fill methods and selection models. P-curve and related methods, sensitivity analyses, and meta-regression exploring study size or funding can provide additional insight. Transparent reporting, preregistered protocols, and comprehensive search strategies help mitigate metabias.

Best practices to reduce metabias involve prospectively registering review protocols, including grey literature, translating non-English studies when possible, exhaustively documenting study inclusion and data extraction decisions, and using robust statistical methods to assess potential biases and their impact on conclusions.