In Reply Dr Hoaglin raises several points regarding the statistical methods used in our study. However, his concerns are general criticisms of usual statistical methods in meta-analyses and do not directly address our primary objective—assessing the association between analytic strategy and estimates of treatment outcomes—nor do they provide any counteracting scientific interpretation of our message.
Hoaglin comments on 3 choices for conducting meta-analyses: the use of the DerSimonian-Laird estimator of the between-trial variance, the use of a continuity correction to deal with zero-cell trials, and the use of the Q statistic. Because these choices are the same for all analytic strategies we compared, it is unlikely that they explain the differences observed between these analytic strategies or bias our conclusions in any way.
Dechartres A, Trinquart L, Ravaud P. Assessment of Heterogeneity in Meta-analyses—Reply. JAMA. 2014;312(21):2287. doi:10.1001/jama.2014.14349