Meta-analysis on recombinant versus urinary follicle stimulating hormone

A recent meta-analysis (Daya and Gunby, 1999) on recombin-ant versus urinary follicle stimulating hormone (rFSH versus uFSH) deserves methodological comment. The a priori definition of eligibility criteria for studies to be included in a meta-analysis are central and must be systematic. It was up to the authors to decide whether their search would be extended to include all unpublished studies, but selective inclusion of some studies, performed with only one of the two recombinant compounds considered and, as it happens, amongst those investigations favouring this compound , is unacceptable. Ironically, the validity score of both unpublished studies had the highest rank order (Table II of the paper), making it difficult to understand the delay in publishing studies of such quality. In any case, this paradox confirms that scoring the quality of clinical trials for meta-analyses is a tricky job (Jüni et al., 1999). Overall, this selection of two unpublished studies favouring follitropin alpha accounted for 40% of the patients receiving this drug; added to the fact that four of the six other studies on this compound came from simple abstracts, this means that 79% of the patients treated with follitropin alpha came from investigations which were not peer-reviewed (as compared to 13% in the follitropin beta series). Table V of the paper suggests a comparison between follitropin alpha and follitropin beta. As everybody knows, this suggestion of a direct comparison between two agents which were not tested within a common investigation is illegitimate from a methodological standpoint. Secondly, the test of homogeneity made by the authors hid a striking heterogeneity between the series included, as the success rate ranged from 18 to 51% for rFSH (16-40% for uFSH) in those trials on follitropin alpha versus only 30 to 35% (27–28% for uFSH) in the studies on follitropin beta. In addition, the sole studies favouring uFSH as compared to rFSH came from the follitropin alpha's series, one of them having been double-blind! Such difference in heterogeneity should have precluded any attempt to discriminate between both series – if not the meta-analysis itself. Thirdly, a simple inspection of data shows that the statistical significance favouring follitropin alpha and unduly put forward in Table V is due to the results of two studies which should not have been included in the meta-analysis, the first because it was one of the positive unpublished studies (Schats et al.), the second because it was not even …