Electronic Letters to:
|
|
eLetters: Electronic letters published:
-
![[Read eLetter]](/icons/misc/sectionDOWN.gif)
A commentary on the study by Veleva et al.
- Manisha Desai (6 April 2009)
|
|
|||
|
Manisha Desai, Associate Professor Columbia University Department of Biostatistics
Send letter to journal:
manisha.desai{at}columbia.edu Manisha Desai
|
The authors conducted an important study to investigate whether elective single embryo transfer (eSET) produces more favorable clinical outcomes than double embryo transfer (DET) in women trying to achieve a pregnancy that results in a healthy baby. They concluded that “eSET with cryopreservation is more effective and less expensive than DET and that it should be adopted as the treatment of choice. “ There are several fundamental flaws in the design and analysis that significantly weaken the authors’ conclusions. Study Design Issues A comparison of eSET and DET across two different time periods is problematic. The authors use time period as a proxy for procedure, making it unclear if the differences observed are actually due to procedure rather than time period. The DET period spans 1995-1999, where 96% of the procedures were DET, and the eSET period spans 2000-2004, where 54% were DET. While procedure varied between the two time periods, there are other important factors that also varied, and these may contribute to the differences in outcome. Most importantly, laboratory and transfer protocols for growing, selecting and transferring the embryos change over time as more is learned about how to optimize a positive outcome. These protocols have a large impact on successful outcome (Eytan et al., 2007; Meldrum et al., 1987). Although the authors state that all patients received the same clinical protocol – the long protocol – it is difficult to believe that this protocol was not improved over the 10-year study period. It is important to note that eSET is a procedure for which not all patients in this study are eligible. From 1996-2000, a woman had to produce at least 4 good embryos to qualify for eSET. After 2000, a woman had to have at least 1. This makes the DET and eSET patients inherently different with respect to embryo quality and therefore outcome. In this study, though, the authors compare DET and eSET time periods, not patients. One may feel more reassured, therefore, if the patient populations of the two time periods were comparable with respect to relevant factors such as embryo quality. But this is not the case. In fact the patient populations differ in potentially critical ways. For example, the main diagnoses of the women in the two time periods differed significantly (P<0.001). Specifically, there were more tubal factor- related diagnoses in the DET era and fewer women with unexplained infertility. Additionally, there were significantly more top quality embryos in the eSET period and more oocytes collected. Both of these differences could be indicative of a more fertile patient population in the eSET period, improvements in the technical expertise and/or protocol of extracting oocytes, laboratory improvements, or revisions of the clinical protocol. Issues with the Clinical Data Analysis The outcomes present analytic challenges that were not addressed. First, the outcomes that involved rates per ovum pick up contained multiple responses per woman, which induced a correlation. Thus, methods that can handle correlated data should have been used when estimating standard errors. The standard errors upon which the test statistics were based were likely underestimated. This is typical when the feature of interest, in this case – procedure period, is constant within a woman, and can have serious implications for inference, making it more likely to falsely conclude there is a difference when there is not. Second, the number of cycles each woman engaged in and the rate at which they entered cycles varied. This could have several consequences. Women entering their 5th cycle are different from those entering their 1st cycle. The former are likely a less fertile group. Those that go through cycles at faster rates may also differ from those who go through cycles slower. It may be that the latter group is younger and feels less pressured by time. This should be accounted for in the analysis. Length of follow-up time was considered as a potential confounder. While correlated with number of cycles, years of follow-up includes periods of time, however, when a woman is not engaged in a cycle and therefore not at risk of getting pregnant through one of the procedures. Cycle number, therefore, is a more relevant confounder that should be adjusted for. The logistic regression analysis used to estimate the odds ratios neither accounts for the varying number of cycles nor addresses the correlation of responses from the same woman. In fact, it is unclear which confounders (such as age) if any were adjusted for in the clinical analysis. Study Design and Analytic Suggestions Although ideal, it is not always possible to conduct a large randomized clinical trial for both reasons of cost and willingness of patients to accept treatment assignment. To enhance the current non- randomized design, I suggest excluding those subjects who would have been ineligible for eSET. Furthermore, instead of using time period as a proxy for period, I suggest using the actual procedure used and comparing the procedures over the same time period, as this is critical in order to adjust for lab, transfer, or clinical protocol changes over time. eSET was well established at this clinic in 2000, so a reasonable study period might span 2000 to 2004, where roughly half the subjects received DET. After excluding those DET patients who would have been ineligible for eSET, however, the ratio of sample sizes for the two groups would likely change. Women from the two arms could be matched on confounders such as age at study entry, average number of high quality embryos over all cycles, and total number of cycles completed by the end of the study period. Whether or not a pregnancy resulting in a healthy baby was achieved for a given ovum pick up is a reasonable outcome. Then a more appropriate analysis that accounts for the correlated nature of the data would be a logistic regression model employing a generalized estimating equations (GEE) approach. It might make sense to assume an exchangeable or auto- regressive correlation structure. Use of robust standard error estimates, however, would provide statistical validity of the results even if this were not a reasonable assumption. Additionally, one should adjust for potential confounders such as cycle type (fresh or frozen), cycle number, number of high quality embryos, and age. Whether or not a pregnancy resulting in a healthy baby was achieved for a woman over her follow-up period is another sensible outcome. For this a standard logistic regression model could be applied, where the following confounders should be adjusted for: total number of cycles, average number of high quality embryos, and age at start of study. An ideal study investigating whether eSET results in superior outcomes relative to DET would be a multi-center randomized clinical trial where women who are eligible for both procedures (i.e., have at least 1 embryo of decent quality) are randomized to one of the two treatment arms and followed over the same study period. Such a study would have comparable patient populations and comparable clinical and laboratory protocols between treatment arms, making any differences observed more likely to be due to the procedure. A multi-center trial would help in accruing eligible subjects and generalizing the results to a larger population. One may also want to re-think the goal of such a study. Given the increase in risk incurred by a multiple pregnancy, a change in clinical practice might be warranted simply if eSET were found to be not worse than DET in its rate of giving birth to a healthy baby. To that end, one should reformulate the hypotheses and conduct a non-inferiority randomized clinical trial. Conclusions The authors conclude that eSET is more effective than DET and should be adopted as a treatment of choice. I believe the conclusions are both premature and possibly misleading. I would advocate a re-analysis of the current study. If strong differences in the procedures persist, this would provide justification for a larger randomized clinical trial before such strong clinical recommendations can be made. References: Eytan O, Elad D, Jaffa A (2007) Evaluation of the embryo transfer protocol by a laboratory model of the uterus. Fertility and Sterility. Volume 88. Issue 2. pp. 485-493. Meldrum DR, De Ziegler D, Chetkowski R, et al. (1987) Evoluation of a highly successfully in vitro fertilization-embryo transfer program. Fertility and Sterility. Volume 48, Issue 1, pp. 86-93. Conflict of Interest:None declared |
|||