Freezing of all embryos in in vitro fertilization is beneficial in high responders, but not intermediate and low responders: an analysis of 82,935 cycles from the Society for Assisted Reproductive Technology registry

Using national database data, subdividing in vitro fertilization patients by ovarian response revealed that frozen embryo transfer was associated with improved pregnancy rates in high but not low/intermediate responders.

Volume 110, Issue 5, Pages 880–887


Kelly S. Acharya, M.D., Chaitanya R. Acharya, Ph.D., P.S.M., Katherine Bishop, M.D., Benjamin Harris, M.D., Douglas Raburn, Ph.D., Suheil J. Muasher, M.D.



To assess in vitro fertilization (IVF) and pregnancy outcomes in patients having their first frozen embryo transfer (FET) after a freeze-all cycle versus similar patients having their first fresh embryo transfer (ET).


Retrospective cohort study.




Registry data on 82,935 patient cycles from the Society for Assisted Reproductive Technology (SART).


All first fresh autologous IVF cycles were analyzed and compared to first FET cycles after a freeze-all first IVF stimulation. The cycles were subdivided into cohorts based upon the number of oocytes retrieved (OR): 1–5 (low), 6–14 (intermediate), and 15+ (high responders). Univariate analyses were performed on cycle characteristics, and multivariable regression analyses were performed on outcome data.

Main Outcome Measure(s)

Clinical pregnancy rate (CPR) and live-birth rate (LBR).


Of the 82,935 cycles analyzed, 69,102 patients had their first fresh transfer, and 13,833 had a first FET. High responders were found to have a higher CPR and LBR in the FET cycles compared with the fresh ET cycles (61.5 vs. 57.4%; 52.0 vs. 48.9%). In intermediate responders, both CPR and LBR were higher after fresh ET compared with FET (49.6% vs. 44.2%; 41.2 vs. 35.3%). Similarly, in low responders, CPR and LBR were higher after fresh compared with FET (33.2% vs. 15.9%; 25.9% vs. 11.5%).


A freeze-all strategy is beneficial in high responders but not in intermediate or low responders, thus refuting the idea that freeze-all cycles are preferable for all patients.

Read the full text here.

Please sign in or register for FREE

Your Fertility and Sterility Dialog login information is not the same as your ASRM or EES credentials. Users must create a separate account to comment or interact on the Dialog.

Go to the profile of Kevin Coetzee
about 4 years ago

We, the readers of and contributors to Fertility and Sterility, as part of the scientific research community wish to advance the understanding of human fertility and to improve the therapies and strategies used to treat infertility, based on rigorously validated evidence.


We, therefore read the recently published study by Acharya et al. (2018) with a certain measure of concern and dismay. In particular, the authors conclude their study with quite a provocative statement, ˝thus refuting the idea that freeze-all cycles are preferable for all patients˵. However, to the best of our understanding there is no valid scientific evidence from the Acharya et al. (2018) study, supporting this statement based on the data and results presented. On the contrary, readers reading the study might get the distinct impression that this statement corresponds to the authors a priori beliefs: ˝This expanding body of evidence suggests that freeze-all with FET cycles may be beneficial in women who are high responders to ovarian stimulation or are at increased risk of OHSS, but it appears to lose this beneficial effect when applied to women with a low or intermediate response˵.


While the numbers presented in this study ˝Registry data on 82,935 patient cycles from the Society for Assisted Reproductive Technology (SART)˵ are statistically pleasing, registry studies ultimately run the risk of being flawed/biased just as single-centre studies with limited sample sizes may do; because of the inclusion of heterogeneous methodologies, especially those affecting patient groups. Therefore, registry studies need to be vigilant of patient population differences and to be rigorous in preventing bias. Unfortunately, this is not evident from the study by Acharya et al. (2018), which seems to have been conducted with a certain measure of confirmation bias. While the authors purport to have performed multivariable regression analyses to adjust for the significant patient population differences, both in the Abstract and the Materials and Methods section, surprisingly, there is no evidence that multivariable regression analyses were performed in the Results section (including Tables). What may be of concern in this regard is that the outcomes of the regressions may have deliberately been excluded because the outcomes contradicted the authors priori beliefs or that the regressions were not performed because of the number of missing data points. On this matter, the authors state, ˝When demographic information was missing or unknown (for example, FSH and BMI frequently are not reported), missing values were omitted while computing the mean and variance˵; however, having up to 33% of data points missing may invalidate performing any regression. The authors, therefore, seem measured in their care not to report any statistics, with mean female age and mean number of embryos transferred probably significantly different, because of the number of cycles included in the analysis. These two variables are of particular concern, especially female age which has most often been selected in regression analyses to be the most predictive variable of reproductive outcomes such as live birth. Ignoring these significant differences, the authors report and use unadjusted reproductive outcomes to support their conclusions.


Ultimately the significant differences between the two patient populations compared, as the result of clinical decision-making, may completely invalidate any comparison, adjusted or unadjusted. The lack of attention to and understanding of patient data is no more clearly illustrated by the authors statement; ˝When the data set was initially filtered by all patients who had a freeze-all cycle (after excluding those with planned preimplantation genetic screening), we found 21,388 cycles. Of these, 4,105 cycles (19.2%) listed ‘‘endometrial receptivity concerns’’ as the reason for no transfer, 4,536 cycles (21.2%) listed ‘‘risk of OHSS,’’ and 4,119 (19.3%) listed ‘‘other.’’ The remainder were a combination of miscellaneous reasons˵. Not only because the proportion of other and miscellaneous was 60%, but also because the study population was significantly reduced to 13,833 through exclusions, with no proportional analysis performed thereafter.


We the scientific community, therefore, believe that while this study may have the potential to provide valuable evidence on freeze-all-IVF this will require the authors to perform the appropriate statistical analysis.


Kemal Ozgur MD, Murat Berkkanoglu MD, Hasan Bulut MD, Kevin Coetzee PhD.

Antalya IVF, Antalya, 07080, Turkey.

Go to the profile of Kelly Acharya
about 4 years ago

Drs. Ozgur, Berkkanoglu, Bulut, and Coetzee,

Thank you for your comments and for reading our article. As we are also a part of the scientific community that reads and contributes to Fertility and Sterility, we take pride in our work, methodology, results, and conclusions. We used rigorous and reproducible methodology and collaborated with a computational biologist to obtain our results, and the multivariable regresion was performed as was clearly stated in our manuscript. The results of the multivariable regression (in which, as stated, we adjusted for confounders including patient age, embryos transferred, etc) were reflected in the adjusted P values that were published. The results of the regression analysis included odds ratios and confidence intervals which were not included in the manuscript or tables only for clarity and simplicity of the manuscript, and not due to discrepancies between the findings of the regression analysis and the descriptive statistics listed. In fact, every result from the multivariable regression agreed with the findings as stated, and where the adjusted P values were listed as significant in our manuscript, these were also significant in the odds ratios and confidence intervals which were not explicitly provided. The
suggestion that we manipulated that data to present what fit with our prior beliefs is totally inappropriate and offensive.

As to the point about missing values for FSH and BMI, we agree that this is a limitation of the dataset we used and a limitation of our study. When the data was missing, it was not included in the analysis (as stated in the methods), but we believe that this does not invalidate the results. Large registry studies have limitations, but they also have strengths. The heterogeneous population and missing data are clearly limitations, including the missing data for the exact reason why the freeze-all was performed in some cases. We discuss this in our discussion
and conclusions, including the fact that this likely explains some of the difference between our findings and those of recent randomized controlled trials. The strengths of our analysis are in the fact that it included a large population (83K cycles) and used only first cycles. The heterogeneity of the population contributes to the generalizability, which is reflected in our main conclusion, that "one size fits all" is not necessarily the case for freeze-all versus fresh. In a heterogenous group of patients who had freeze-all cycles performed for heterogenous reasons, we found that frozen embryo transfer is not always clearly beneficial. We do not suggest that fresh embryo transfer should be performed on all patients either; we simply state that an individualized approach is likely beneficial. 

Kelly Acharya, MD

Suheil Muasher, MD

Go to the profile of Kevin Coetzee
almost 4 years ago

We wish to thank the authors for the willingness to defend their publication against our comments. The intention of our commentary was not to offend, only to provide a critical and accurate review, which any publication in Fertility and Sterility deserves. We are sure the authors agree that the competition for acceptance in a high impact factor journal such as Fertility and Sterility has become increasingly competitive and more difficult to achieve. The intention of the critical and rigorous peer-review procedure employed being that once published the evidence from a study can confidently be used in the treatment of patients. Moreover, after reviewing the authors’ study and their response to our comments, it is clear that applying their evidence in the treatment of patients would have to be dependent on trust; the trust that appropriate statistical analyses were performed. It is immediately apparent from reading the study's abstract that the authors did not attribute great importance to the adjusted statistical analyses, deciding neither to include the outcomes from the supposed multiple logistic regressions in the results nor the conclusion.We do not claim to be experts in statistical analysis; however, the advice we have sought and the response of the authors to our comments have confirmed our concerns and questions voiced in our original comment. Unfortunately, the authors decided for the sake of simplicity not to follow normal conventions in their statistical reporting; however, their argument for simplicity is not reassuring, as the onus is on the authors to provide a full statistical analysis that appropriately tests all assumptions and to provide the reader with an understandable interpretation. Moreover, in most large clinical datasets missing data is an inevitable outcome; however, in the authors' study, the concern is the proportion of missing data and the handling of missing data (Dong and Peng, 2013). While rates of 5% or less are inconsequential, rates of more than 10% may lead to biased estimates, loss of information, decreased statistical power, increased standard errors, and weakened generalizability of findings (Dong and Peng, 2013). The use of large datasets retrospectively extracted may introduce inherent biases that require appropriate and explicit analysis. In the authors' argument in defence of their study, they also misconstrue the words heterogeneity and generalizability. Of course, all large datasets will contain different groups (heterogeneous make-up); however, the make-up of their frozen (freeze-all) group definitely does not represent that of a general population. While standard-IVF treatment processes and procedures were most probably uniformly performed during the study period across clinics, those used in the freeze-all-IVF treatments were most likely not. The importance of adjusting for covariables in statistical analyses is no better illustrated than in the analysis of our study (Table: Ozgur et al., 2019). Live birth (LB) rates were found to increase with increasing numbers of oocytes retrieved, as was found by the authors in their own study; however, female age decreased with increasing numbers of oocytes retrieved. Performing a multiple logistic regression adjusting for female age, blastocyst number transferred, and oocyte number retrieved in the 1st frozen embryo transfers following oocyte retrieval, only female age (OR = 0.93 95% CI (0.912-0.954), p <0.001) and two blastocysts transferred (OR = 1.67 95% CI (1.312-2.127), p <0.001) were significant predictors of LB. The significance of female age in this analysis was further delineated by the finding that the per transfer LB rate of the 1-3 oocyte number subgroup with female age ≤35 years was 64.3%, which was similar to that of the >15 oocyte number group (62.3%), which had a median female age of 29.7 (26.4-33.6) years. It is our expressed hope, therefore, that all involved in the processes that lead have led to this discussion, have learned the importance of procedures, conventions, and accuracy. Moreover, in conclusion, we express the hope that all who participated in the processes that have led to the review of this study and posting of commentaries and responses have learned the importance of processes, conventions, and accuracy in the publication of evidence.



Oocyte number group



female age (yrs)







37.0 (33.9-39.5)

42.5 (34)

19.3 (34)



35.1 (31.5-38.6)

50.8 (168)

41.9 (168)



32.1 (28.2-35.1)

55.1 (177)

49.9 (177)



29.7 (26.4-33.6)

62.3 (369)

59.5 (369)


Dong Y, Peng CY. Principled missing data methods for researchers. Springerplus. 2013; 2: 222. (doi:10.1186/2193-1801-2-222).


Ozgur K, Bulut H, Berkkanoglu M, Donmez L, Coetzee K. Prediction of live birth and cumulative live birth rates in freeze-all-IVF treatment of a general population. J Assist Reprod Genet, accepted for publication in 2019 (doi:10.1007/s10815-019-01422-z).