In Reply Dr Mills and colleagues wish to deemphasize I2 statistics. We agree that I2 statistics can be problematic and that the naive pooled model (ie, the model assuming no study-to-study heterogeneity) is often inappropriate for meta-analyses. This is why, in our study, I2 statistics were only reported for completeness and never factored into our analysis.
First, Mills and colleagues assume that I2 statistics played a prominent role in our analysis. They did not. They were reported in accordance with PRISMA1 guidelines but served no other function. We quantified the extent of study-to-study heterogeneity with a component of variance from the best-fit random-effects logistic regression model.
Blume JD, Deppen SA, Grogan EL. Heterogeneity in Meta-analysis of FDG-PET Studies to Diagnose Lung Cancer—Reply. JAMA. 2015;313(4):419-420. doi:10.1001/jama.2014.16485