|
Abstract:
|
Meta-analyses have become increasingly popular to conduct, especially in public health and medicine where multiple, independent clinical trials can be combined to produce one overall conclusion. Meta-analyses are especially useful when small clinical trials lack sufficient power in themselves to detect a treatment effect or when events are rare or adverse. Estimating heterogeneity, which manifests when treatment effects differ across studies for reasons not due to chance, is a crucial step in a meta-analysis to ensure the accuracy of results. Many heterogeneity estimators exist, and their performances have been analyzed under a variety of settings, but little research has been conducted on how these estimators behave in the rare event setting. We explore, via a simulation study, the performance of several commonly used heterogeneity estimators with rare event meta-analysis data. We find that while some estimators outperform others, they all consistently, and rather shockingly, severely fail to detect non-negligible heterogeneity. Accordingly, we propose a new heterogeneity estimator for rare event meta-analyses based on a permutation approach.
|
