Nifedipine versus ritodrine for suppression of preterm labor; A meta‐analysis

Dear Sir, We read the article by Oei et al. (1) with great interest. There is growing attention in the literature on tocolytic therapy to calcium channel blockers, and many groups have performed randomized controlled studies to compare nifedipine to betamimetic treatment. A meta-analysis could have been very useful, since the power of the individual trials on this subject was insufficient to reveal any difference in such outcomes as prolongation of pregnancy or neonatal morbidity. The methodology used, however, had several flaws that may impair the validity of the author’s conclusions. – The data presented are not based on an intent-to-treat analysis, and no effort has been made to collect the additional data necessary for such an analysis from the authors of the original studies. Use of an intent-to-treat analysis is a major criterion of the quality of a meta-analysis (2). – All of the published randomized studies comparing nifedipine to ritodrine were considered, despite the obvious defects in some. One example is the trial by Kose et al. (3), which does not describe the randomization procedure and used two unbalanced groups (52 patients in the nifedipine group and 21 patients in the ritodrine group). Another example is the trial by Meyer et al. (4): it randomized patients only if they were still having regular uterine contractions after subcutaneous terbutaline. We consider that this introduces a major bias, because patients responding to beta-mimetics were systematically excluded from the trial. – The authors of the meta-analysis used the Mantel-Haenszel (5) method to calculate common odds ratio with 95% confidence interval. This method is inadequate in situations where outcomes are zero. For example, the odds ratio and confidence interval for perinatal mortality should not be calculated with this method since mortality did not occur in any group of several trials. In this meta-analysis, the method of Yussuf et al. (6) would have been more appropriate. Moreover, the authors state in the material and methods section that they have calculated the standard odds ratios with their 95% confidence intervals, but Fig. 2 reports relative risk instead. – Finally, the authors do not mention the number of patients for each outcome from each study. They do not provide the data necessary to calculate odd ratios and confidence intervals. Both of these requirements are part of the QUOROM statement checklist (2). We regret that possibly important therapeutic improvements