Sole reliance on I2 may mislead

We read the recommendations by Kiran and colleagues1 on graphical displays to present meta-analysis results with great interest. Like the authors, we think that graphics play a major role to concisely summarise meta-analysis results and to evaluate heterogeneity between studies and small-study effects. We agree with most recommendations1 and actually a forest plot (see figure 1 and online ) generated with our R package meta (https://cran.r-project.org/web/packages/meta/) almost looks like a blueprint of figure 2 in ref 1. However, in contrast to the authors, we would like to argue that solely relying on I2 2 to investigate heterogeneity² might be misleading.

Figure 1

Forest plot generated with forest.meta() from R package meta.

Almost 10 years ago, we demonstrated analytically and using simulations that I² does not depend on the number of...