Response to letter: non‐vitamin‐K oral anticoagulants may not significantly reduce the risk of fatal or disabling stroke compared with warfarin

Abstract

We agree with Zheng and colleagues that a statistical measure of heterogeneity is an important consideration in selecting a fixed or random effects model. However, it should not be considered the sole criterion. In meta‐analyses that include small numbers of trials, such as ours, results of tests for heterogeneity should be interpreted with caution, as they may be underpowered. A fixed effects model has been advocated when meta‐analysing small number of studies (less than five trials), given the concern about inaccurate estimation of between‐study variance (1), as the estimation of τ²(between‐study variance), and therefore of μ (estimate of underlying mean effect), is highly imprecise (2).