Potential dangers in the customary methods of conducting meta-analyses

In a meta-analysis of randomized trials comparing recombinant and urinary FSH for ovarian stimulation in infertility treatment cycles, the clinical pregnancy rate per cycle started was significantly higher with recombinant FSH. Before the results of the different trials can be pooled, it is important to determine whether the results are homogeneous so that one can obtain an overall treatment estimate that is without bias. There are several methods to ascertain lack of homogeneity, including logistic regression analysis, sensitivity analysis comparing fixed-effect and random-effect models, testing based in the chi2 distribution, and graphical methods comparing event rates in experimental and treatment groups. These methods all confirmed that the treatment effect was homogeneous across all trials thereby providing assurance that the overall conclusion of the superior efficacy of recombinant FSH compared with urinary FSH is based on an unbiased analysis.