The State of the Art in Meta-Analysis Software: Evolution, Shortcomings, and Future Directions

Meta-analysis is central to evidence-based medicine, yet much of the biomedical literature continues to rely on software with outdated statistical defaults. Legacy programs such as RevMan 5.4 and MetaDiSc 1.4 persist in practice despite their reliance on the DerSimonian–Laird estimator and the obsolete Moses–Littenberg model for diagnostic test accuracy. Their modern successors—RevMan 7 and MetaDiSc 2.0—have incorporated robust methods including Restricted Maximum Likelihood estimation, Hartung–Knapp–Sidik–Jonkman intervals, prediction intervals, and hierarchical bivariate models. However, their dissemination has been limited, and both remain constrained in handling more advanced approaches such as Bayesian modelling or network meta-analysis. Comprehensive Meta-Analysis, although widely used, raises further concerns about transparency and reproducibility due to undisclosed algorithms and unclear defaults. Reliance on fragile defaults embeds preventable bias and exaggerated precision into the evidence base. By contrasting legacy practices with current methodological standards, this critique provides practical recommendations for researchers, reviewers, and editors to promote transparent, reproducible, and methodologically sound meta-analytic practice.