Overlooking intention-to-treat results

Sir, We read with great interest the recently published randomized controlled trial (RCT) comparing clomifene citrate (CC) versus low-dose FSH for ovulation induction (OI) in women with polycystic ovary syndrome (Homburg et al., 2012). The study was well designed and conducted; however, we do not agree with the manner the results were reported and with authors’ conclusions. As stated by authors ‘the aim of this RCT was to test the hypothesis that the (cumulative) pregnancy rate and live birth rates (LBRs) in OI are higher with low-dose FSH than with CC as a first-line treatment’. Indeed, when evaluating the effectiveness of an intervention in subfertility the primary main outcome of an RCT should be live birth per allocated woman, respecting the intention-to-treat (ITT) principle (ASRM, 2008). In this trial comparing CC versus low-dose FSH, the ITT analysis was almost omitted, being limited to a few words in the middle of the article: ‘there was a similar proportion of women achieving a clinical pregnancy [80/159; 50.3 versus 59/143 or 41.3%, 95% confidence interval (CI): 22.1–20.3, P 1⁄4 0.1] or a live birth (72/159 or 45.3 versus 53/143 or 37%, 95% CI: 2.8–19.3, P 1⁄4 0.12)’. This information can be easily overlooked by a regular reader, since there is no mention to this analysis in the abstract, discussion or conclusion. The risk ratio (RR) between the two treatments respecting ITT was not significant: the LBR was higher in women allocated to use lowdose FSH comparing with CC, but the estimated effect was relatively imprecise. Considering an RR .1.2 or ,0.8 as the minimal clinically relevant difference (Martins et al., 2011), the 95% CI included both no effect and appreciable benefit favoring low-dose FSH (RR 1⁄4 1.22; 95% CI 1⁄4 0.93–1.61). Similar results were observed for clinical pregnancy: this rate was also higher in women allocated to use low-dose FSH, but the estimated was relatively imprecise with the 95% CI including both no effect and appreciable benefit (RR 1⁄4 1.22; 95% CI 1⁄4 0.95–1.56). The imprecise estimates occurred because authors used a very large effect (RR .1.5) as minimal clinically relevant when calculating the sample size. If authors want to demonstrate an RR .1.2 with an 80% power with a 1⁄4 0.05, they should evaluate 300 subjects per group. This sample size would also provide approximately 200–400 events (live births or clinical pregnancies), which is suggested as the minimum number of events to be observed before stopping a trial because of an early apparent benefit (Mueller et al., 2007). In summary, the conclusions of this study should be based on this analysis, otherwise the size of the effect of the intervention might be overestimated (ASRM, 2008). Therefore, authors should conclude that while it is unlikely that low-dose FSH will reduce live birth or clinical pregnancy chance when compared with CC, it is still premature to conclude the superiority of this method of OI. This trial should be continued or combined with the results of other trials in a meta-analysis before any further conclusion.