Meta-analysis in evidence-based healthcare: a paradigm shift away from random effects is overdue

Each year up to 20 000 systematic reviews and meta-analyses are published whose results influence healthcare decisions, thus making the robustness and reliability of meta-analytic methods one of the world's top clinical and public health priorities. The evidence synthesis makes use of either fixed-effect or random-effects statistical methods. The fixed-effect method has largely been replaced by the random-effects method as heterogeneity of study effects led to poor error estimation. However, despite the widespread use and acceptance of the random-effects method to correct this, it too remains unsatisfactory and continues to suffer from defective error estimation, posing a serious threat to decision-making in evidence-based clinical and public health practice. We discuss here the problem with the random-effects approach and demonstrate that there exist better estimators under the fixed-effect model framework that can achieve optimal error estimation. We argue for an urgent return to the earlier framework with updates that address these problems and conclude that doing so can markedly improve the reliability of meta-analytical findings and thus decision-making in healthcare.