Using Meta-analysis to Answer Clinical Questions

To the Editor: In their meta-analysis, Dr Ioannidis and colleagues¹ found that randomized and nonrandomized studies often reach similar conclusions, although nonrandomized studies may yield larger estimates of the effect size. I agree with the authors that nonrandomized evidence can help explore clinical questions. However, analyzing sources of heterogeneity among randomized and nonrandomized trials might be more informative than simply pooling treatment results from only randomized clinical trials. Especially in the case of an unclear treatment effect, all available information should be used to resolve the uncertainty in outcome between studies. Important information from nonrandomized clinical trials should not be ignored; instead, heterogeneity in design, data collection, and analysis should be explored. Even in randomized clinical trials, the method of randomization might lead to differences in outcomes.