ABSTRACT
The development of hierarchical models may help ease recent skepticism in the literature about the usefulness of meta-analysis. LeLorier et al. (1) study the relationship between meta-analyses and subsequent large randomized clinical trials that address similar issues. They summarize a meta-analysis by the mean of the population of trials, and a confidence interval. They conclude that some meta-analyses are faulty because the subsequent clinical trials don’t “agree” with the meta-analysis interval. In this paper, I address their definition of “agree” and show that it ignores the possibility of an important form of heterogeneity. When variability across trials is explicitly modeled, and the distribution of trials is addressed, then the future clinical trials do in fact “agree” with the meta-analysis.
