全文

Objective
To investigate the association between oocyte number and the rates of ovarian hyperstimulation syndrome (OHSS) and live birth (LB) in fresh autologous in vitro fertilization (IVF) cycles.

Design
Retrospective cohort study.

Setting
An academic reproductive medicine practice.

Patient(s)
We analyzed data from 256,381 IVF cycles using the 2008–2010 Society for Assisted Reproductive Technology national registry. Patients were divided into five groups based on retrieved oocyte number.

Main Outcome Measure(s)
Rates of OHSS and LB were calculated for each group. A generalized estimating equation (GEE) was used to assess differences in OHSS and LB between groups. Receiver operating characteristic (ROC) curves were used to evaluate oocyte number as a predictor of OHSS and LB.

Intervention(s)
None.

Result(s)
The LB rate increased up to 15 oocytes, then plateaued (0–5: 17%, 6–10: 31.7%; 11–15: 39.3%; 16–20: 42.7%; 21–25: 43.8%; and >25 oocytes: 41.8%). However, the rate of OHSS became much more clinically significant after 15 oocytes (0–5: 0.09%; 6–10: 0.37%; 11–15: 0.93%; 16–20: 1.67%; 21–25: 3.03%; and >25 oocytes: 6.34%). These trends remained after adjustment with the use of GEE. ROC curves revealed that although oocyte number is not useful in the prediction of LB, 15 retrieved oocytes is the number that best predicts OHSS risk.

Conclusion(s)
Retrieval of >15 oocytes significantly increases OHSS risk without improving LB rate in fresh autologous IVF cycles. In general, less aggressive stimulation protocols should be considered, especially in high-responders, to optimize outcomes.

Key Words
Oocyte number; ovarian hyperstimulation syndrome; live birth; pregnancy

Babies born as the result of assisted reproductive technology (ART) account for >1% of births each year in the United States (U.S.) (1). Other than the maternal and neonatal complications attributable to multiple gestation arising from the transfer of more than one embryo, no other associated risk accounts for more morbidity and mortality than ovarian hyperstimulation syndrome (OHSS). Despite strides to reduce the burden of this potentially fatal and completely iatrogenic complication, it remains a serious health concern for a significant percentage of patients undergoing in vitro fertilization (IVF).

OHSS is not uncommon. Although older studies report the presence of its moderate or severe form in up to 10% of all IVF cycles, more recent publications quote an incidence of ∼1%–5% 2, 3, 4, 5, 6, 7 and 8. According to the Society for Assisted Reproductive Technology (SART) database, this range would suggest that there were as many as 4,000 IVF cycles with moderate or severe hyperstimulation in the U.S. in 2011 alone. The true incidence of OHSS is probably much higher. Misreporting potentially results from the lack of a standardized definition, provider fear of punitive action, and the fact that SART data, including complication rates, are available to the public. It seems possible that the pressure on infertility clinicians to provide the highest possible pregnancy rates might sometimes result in underreporting of OHSS risk.

The often used terminology to classify OHSS dates back to that proposed by Rabau et al. in 1967 (9). Multiple revisions have been introduced since then, most notably by Schenker and Weinstein in 1978 (10) and Golen et al. in 1989 (4). Common to each description of severe OHSS is some combination of marked ovarian enlargement, ascites and/or hydrothorax, hemoconcentration, hypercoagulation, impaired renal perfusion, and electrolyte imbalance. However, a succinct consensus definition continues to be unclear from the literature of SART, the American Society for Reproductive Medicine, and the European Society for Human Reproduction and Embryology.

The hallmark of the pathophysiology of OHSS is an increase in vascular permeability resulting in extravasation of intravascular fluid into the third space 11 and 12. The triggering mechanism is the hCG-mediated increase in vascular endothelial growth factor, an angiogenic cytokine that stimulates vascular endothelium. This can result either from exogenous hCG or from hCG produced by a successfully implanted embryo. The former leads to so-called “early OHSS” and the latter to “late OHSS” (13).

In addition to the well recognized risk factors for OHSS, which include young age, lean body habitus, polycystic ovarian syndrome, high-dose gonadotropin treatment, high serum E2 levels, luteal-phase hCG supplementation, and pregnancy 14, 15, 16 and 17, multiple studies have reported its association with the number of developing follicles and retrieved oocytes in ART cycles. Because treatment is largely supportive, prevention is crucial. The need to identify patients at high risk is therefore imperative. Taken together, these studies suggest that: 1) Although no one parameter is sufficient to estimate risk, retrieved oocyte number is likely to be the best indicator, because it is the most direct measure of ovarian response; and 2) 10–20 (or more) retrieved oocytes is the range that best predicts moderate to severe OHSS 2, 3, 5, 6, 7, 18, 19 and 20.

The proposal of such a threshold raises the question of whether one also exists for live birth (LB) rate, the benchmark of ART success. Several studies have investigated this question 21, 22 and 23, including an analysis of more than 400,000 IVF cycles from the national ART registry of the United Kingdom (24). Also questioning whether retrieved oocyte number provides the most robust surrogate for clinical success, each of these studies attempted to correlate oocyte number and LB. They conclude that more oocytes are not always better, particularly in good-prognosis patients, because LB rate does not significantly increase after the retrieval of ∼6–15 oocytes in fresh autologous cycles 21, 22, 23 and 24. It should be noted that an optimal investigation of LB would include cumulative LB rate as an outcome. However, because the SART database does not currently provide for linkage between a patient's fresh and frozen cycles, these data are unavailable.

To our knowledge, no study has investigated the relationship between retrieved oocyte number and the rates of both OHSS and LB. There may exist a retrieved oocyte number that maximizes LB rate while minimizing OHSS rate. If discovered and implemented into common practice, it could improve the safety of ART without sacrificing clinical outcomes.

Materials and methods
Data Source
This study was approved by the Duke University Institutional Review Board. Anonymized data were obtained from the SART registry for all fresh nondonor IVF cycles performed in the U.S. from 2008 to 2010. SART is an independent organization promoting the advancement of ART practice, to which >90% of reproductive medicine clinics in the U.S. annually report cycle data. Collected data included patient demographics, infertility diagnoses and treatment history, ovarian reserve testing, cycle type and stimulation parameters, retrieved oocyte number, fertilization rate, transferred embryo number and quality, cryopreservation data, pregnancy outcomes, and complications. Of note, because SART does not precisely define OHSS, the criteria used for its diagnosis may have varied slightly from clinic to clinic. As mentioned previously, this may have contributed to underreporting.

Outcomes
The number of oocytes retrieved was categorized into six groups: 0–5, 6–10, 11–15, 16–20, 21–25, and >25. Primary outcomes included the rates of OHSS (moderate and severe) and LB per group. Mild OHSS is common to the majority of women undergoing controlled ovarian stimulation (COS) and is therefore of little clinical significance other than as a marker of the effectiveness of treatment (25). As such, it is not routinely reported to SART and was therefore not included as an outcome in the present study. Secondary outcomes included clinical pregnancy (CP), fertilization, and embryo cryopreservation rates per group. Clinical pregnancy was defined as visualization of an intrauterine gestational sac by transvaginal ultrasound coincident with a positive serum β-hCG. Live birth was defined as delivery of a live-born infant at ≥24 weeks gestation. The diagnosis of OHSS was without clear definition.

Statistical Analysis
The data set included fresh autologous cycles. The incidences of OHSS, LB, and CP were calculated per initiated cycle for each oocyte number group. Descriptive statistics are reported for baseline characteristics and secondary outcomes by oocyte number group. Means and standard deviations are reported for continuous outcomes, and frequencies and proportions are reported for categoric outcomes. The chi-square test was used to separately assess the association of each oocyte group with both OHSS and LB. Because the data set included patients with multiple IVF cycles, the generalized estimating equation (GEE) method, in the context of multivariate logistic regression for correlated data, was used to assess differences in OHSS and LB between the six groups while adjusting for patient baseline age, baseline body mass index (BMI), maximum basal follicle-stimulating hormone (FSH) level, and baseline smoking status 26 and 27. Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for OHSS and LB were estimated to evaluate the relative odds for OHSS and LB compared with the reference group of 11–15 retrieved oocytes. This group was chosen as the reference because it accounted for a high percentage of the total cycle number and was thought to be the likely range for optimal outcomes as based on similar large studies 7, 19 and 24. Finally, receiver operating characteristic (ROC) curves were constructed to describe the relationship between sensitivity and specificity for the retrieved oocyte number in the prediction of OHSS and LB. The SAS statistical package (release 9.2; SAS Institute) was used for analysis. With two primary end points and five group comparisons, the type 1 error was 0.005 with the use of the Bonferroni adjustment to reduce the possibility of spurious significant findings.

Results
A total of 257,971 cycles met the inclusion criteria. Of those, 1,590 cycles lacked the number of retrieved oocytes and were excluded, leaving 256,381 cycles for analysis. The overall rates of CP, LB, and OHSS were 45.1%, 36.8%, and 1.2%, respectively. Of the 3,104 total cycles with a diagnosis of OHSS, 69.3% were categorized as moderate and 30.7% as severe. Patients ≤35 years old accounted for 43.1% of the total cohort, and those age 35–37, 38–40, 41–42, and >42 years represented 14.5%, 21.8%, 11.4%, and 9.1%, respectively. The most common range of retrieved oocytes was 6–10, representing 29.5% of the cohort. The groups with 0–5 and 11–15 oocytes accounted for 18.4% and 24% of cycles, respectively. In total, 28.1% (72,089 cycles) yielded >15 oocytes, including 6.2% (16,005 cycles) in which >25 were harvested (Fig. 1).

Live birth rate increased steadily from 17% in the 0–5 retrieved oocytes group to 39.3% in the 11–15 oocytes group. Despite statistically significant differences in LB, owing to the large size of the cohort (P values of <.001 for overall group difference by chi-square test), clinically relevant differences diminished as the rate plateaued (42.7%, 43.8%, and 41.8% of cycles in the 16–20, 21–25, and >25 oocyte groups, respectively). This trend was in stark contrast to that for OHSS. A relatively inconsequential rise in OHSS rate was seen until ∼15 retrieved oocytes when the incidence was 0.93%. Above that threshold, a more dramatic increase in rate was seen, with 1.67%, 3.03%, and 6.34% of cycles in the 16–20, 21–25, and >25 oocyte groups, respectively, reporting moderate or severe OHSS ( Fig. 2).

Similar differences in estimated LB and OHSS rates remained after conducting GEE in a multivariate logistic regression while adjusting for age, BMI, basal FSH, and smoking status. Compared with the reference group of 11–15 retrieved oocytes, the groups with 0–5 and 6–10 oocytes demonstrated a lower chance of LB (ORs 0.46 [95% CI 0.44–0.48] and 0.83 [95% CI 0.80–0.85], respectively; both P values <.001) and OHSS (ORs 0.09 [95% CI [0.05–0.15] and 0.39 [95% CI [0.32–0.47], respectively; both P values <.001). However, with the retrieval of >15 oocytes, the odds of LB began to plateau and then decrease (ORs 1.06 [95% CI 1.03–1.1], 1.04 [95% CI 1–1.08], and 0.91 [95% CI 0.87–0.95] for the 16–20, 21–25, and >25 oocyte groups, respectively; all P values <.001), and the odds of OHSS significantly increased (ORs 1.62 [95% CI 1.41–1.86], 2.93 [95% CI 2.55–3.36], and 6.16 [95% CI 5.44–6.98] for the 16–20, 21–25, and >25 oocyte groups, respectively; all P values <.001). This increase in risk existed despite little variation in the number of embryos transferred per group (2.0–2.6 in all groups), a fact likely to control for difference in multiple gestation rate and any subsequent confounding of OHSS incidence due to multiples.

In addition, ROC curves were constructed to examine retrieved oocyte number as a predictor of OHSS and LB. The sensitivity and specificity of escalating oocyte thresholds (5, 10, 15, 20, and 25) were calculated and the resulting ROC curve produced (Fig. 3). The optimal cutoff point for OHSS was 15 oocytes, with the minimum distance to point (0, 1), and a sensitivity and specificity of 71.1% and 72.4%, respectively. The area under the ROC curve (AUC) was 0.784. A similar curve was produced for LB. With an AUC of 0.596, however, retrieved oocyte number did not accurately predict LB rate.

Secondary outcomes and baseline age are summarized descriptively by group in Table 1. Although the percentage of cycles with cryopreservation and the mean number of embryos cryopreserved increased linearly with the number of oocytes retrieved, the percentage of cycles with at least two embryos cryopreserved after fresh transfer (among those cycles with supernumerary embryos) was similar in all cycles with at least six retrieved oocytes (71% in the 0–5 oocyte group and >96% in all the other groups [data not shown]).

Discussion
This study demonstrates that when >15 oocytes are retrieved in fresh autologous IVF cycles, LB rate plateaus and OHSS rate significantly increases. This finding remains true after adjustment for age, BMI, basal FSH, and smoking status. Furthermore, ROC curve calculation revealed that 15 is the number of retrieved oocytes that maximizes sensitivity and specificity in the prediction of OHSS. However, oocyte number is less useful in the prediction of LB. To our knowledge, no previous publication has investigated both LB and OHSS, arguably the most significant outcomes in the practice of ART, using retrieved oocyte number as a potential surrogate. We analyzed more than 250,000 IVF cycles for this purpose.

Moderate and severe OHSS remain very real risks associated with COS in normal and high-responders. In Europe, the associated maternal mortality rate has been reported to be approximately three deaths per 100,000 IVF cycles 28 and 29. Yet no maternal deaths due to OHSS in fresh cycles were reported to SART from 2008 to 2010 (∼300,000 total IVF cycles). However, significant morbidity from OHSS is relatively common. According to SART figures, assuming an average incidence of 2% for moderate–severe OHSS, ∼2,000 women experience this complication in the U.S. each year.

As alluded to previously, the rates of OHSS calculated from this and other data sets are almost certainly underestimated. Several factors contribute to this troublesome realization. For one, none of the major infertility-related societies precisely define what moderate and severe OHSS are. Thus, standardized reporting is problematic. In addition, because SART data is available to the public, some physicians may be reluctant to report accurate rates of complications. Furthermore, for various reasons, ∼10% of ART clinics in the U.S. report nothing to SART. Therefore, we can only speculate as to their OHSS rates.

Existing literature regarding the use of similar parameters in the estimation of OHSS offers support to our findings. Peak serum E2 and high follicle number have been commonly investigated 3, 5, 7, 14 and 20. A frequently cited paper reports an 85.5% sensitivity and 69% specificity for OHSS risk when ≥13 follicles ≥11 mm in diameter are measured before retrieval (7). When ≥18 such follicles and/or a serum E2 level of ≥5,000 pg/mL were observed, sensitivity and specificity rose to 83% and 84%, respectively. Though seemingly useful, E2 level and follicle number are more indirect measures of COS, as well as more susceptible to intersonographer subjectivity, than number of retrieved oocytes. Those that have focused exclusively on oocyte number agree that 10–20 is likely to be the “magic range” above which OHSS risk becomes unacceptably high 18 and 19.

Similarly designed studies have sought to discover the relationship between retrieved oocytes and LB 21, 24 and 30. Interestingly, the same major conclusions are common among them. IVF appears to be an inefficient process with only a small minority of retrieved oocytes resulting in a LB, no matter the patient population. For the purpose of fresh autologous transfer, a higher number of oocytes is not always desirable and in fact can be detrimental. Particularly in the high-responder group, a high degree of oocyte wastage occurs. It is unclear exactly why, but proposed theories include a higher percentage of immature oocytes, direct oocyte harm from excessive hormone levels during COS, or perhaps, because many of these oocytes would naturally have been selected for atresia, an intrinsically decreased capacity to be fertilized 21, 22, 23 and 30. As with OHSS, published data suggest that a threshold exists with LB, likely in the range of 6–15 retrieved oocytes, above which only expense and complication rates increase.

A very large study performed using the United Kingdom's national ART registry was reported by Sunkara et al. in 2011 (24). They analyzed 400,135 cycles over a 17-year time span to establish a nomogram that predicts LB for a given number of retrieved oocytes and patient age. They concluded that oocyte number is a strong correlate for LB and that LB rate peaks at ∼15 and declines after 20 retrieved oocytes, regardless of age. In design and impact, that study is the most similar to the present one. However, it should be noted that although our analysis indicates an association between oocyte number and LB, including a peak LB rate around 15, oocyte number, in and of itself, does not seem to be a reliable predictor of LB. Moreover, the study by Sunkara et al. did not incorporate ROC curve analysis nor investigate OHSS risk.

Each of the above-mentioned studies has its limitations. With the exception of that by Sunkara et al., all were relatively small and composed of data from a single center. In addition, they either did not perform regression analyses to account for confounders 22, 23 and 30 or did so incompletely by omitting covariates known to influence oocyte yield, such as BMI and smoking status 21 and 24. The study by Patrizio and Sakkas included LB data based on embryos still in cryopreservation (22). They were thus forced to estimate LB rate based on historical in-house frozen embryo statistics. Furthermore, none of the studies investigated the combination of OHSS and LB. It is therefore more difficult to be confident in the extrapolation of the conclusions of these studies.

We used the largest IVF database in the U.S. to perform the present investigation. Cycle numbers as well as LB and OHSS rates were similar in 2011, the most recent year of SART database publication, which serves as an added measure of validation for our findings. Though some may consider the clinical heterogeneity of such a large cohort to be a weakness, it serves to bolster the generalizability of our conclusions. Another strength of our study design is the use of GEE modeling. This technique focuses on “population-averaged” effects and unifies standard error estimators to account for unmeasured dependence between outcomes 26 and 27.

We do, however, acknowledge several limitations of our study. As mentioned above, SART does not provide specific language to define OHSS. Therefore, each clinic reports its incidence with some degree of subjectivity. This is almost certain to result in variable reporting and/or underreporting. Additionally, we did not analyze data on stimulation protocol type or medication dosing, two variables known to affect oocyte yield and OHSS rate. Although doing so may have helped to elucidate the association between oocyte number and OHSS, particularly in high-responders, the subtle nuances inherent to various stimulation protocols would likely have made the significance of this information difficult to tease out. Finally, as mentioned earlier, owing to the fact that the current SART database does not allow linkage of fresh and frozen-thawed cycles for the same patient, only the fresh cycles were included in the analysis. As a result, we can not report on cumulative LB rate. It is possible that the plateau and subsequent decline in LB after 15 retrieved oocytes is obviated by an increase in cumulative LB rate as a result of pregnancies from frozen-thawed embryo transfers. That said, even when the objective is to cryopreserve supernumerary embryos for future transfer, existing data suggest no benefit to the retrieval of more than 18 oocytes (31). In addition, our data indicate that nearly all cycles in which six or more oocytes were retrieved had at least two embryos remaining to cryopreserve after the fresh transfer, enough for at least one frozen-thawed embryo transfer. Even among the low responders (0–5 retrieved oocytes), 71% of cycles had at least two extra embryos available for cryopreservation. These observations would seem to lessen the issue created by our inability to expound further on cumulative LB rate.

In our opinion, less aggressive ovarian stimulation should be the first-line mechanism to minimize the incidence of moderate–severe OHSS in high-responders. However, other preventative measures do exist, and we suggest their utilization as secondary strategies. GnRH agonists are now commonly used for final oocyte maturation to decrease the risk of OHSS in high-responders and oocyte donors. Their efficacy in doing so while providing for a high percentage of mature good-quality oocytes is well documented 32, 33, 34, 35 and 36. Less of a consensus exists, however, regarding their effect on pregnancy outcomes 33, 34 and 36. In fact, current recommendations from the Copenhagen GnRH Agonist Triggering Workshop Group suggest that the only two ways in which ongoing pregnancy rate after GnRHa trigger can approach that of hCG trigger are: 1) by adding one or more injections of low-dose hCG or recombinant LH at the time of trigger and/or in the luteal phase, in addition to “modified” E2 and progesterone supplementation; or 2) by freezing all embryos and transferring them after thaw in a subsequent programmed cycle (37). The latter strategy is well known to not only prevent OHSS but has recently been championed as a way to improve pregnancy outcomes by avoiding impaired endometrial receptivity in fresh cycles 38, 39, 40, 41, 42, 43, 44, 45 and 46. Moreover, dopamine agonists, such as cabergoline (Pfizer; New York, NY, USA), have been demonstrated to prevent moderate–severe OHSS in high-risk patients (47).

Perhaps more clearly than ever before, the findings of the present study challenge the classically held tenet of IVF that a higher number of oocytes and resulting embryos should be a primary aim of COS. Our data indicate that very aggressive ovarian stimulation regimens are no longer justified as a means to increase oocyte yield in fresh cycles, especially in good-prognosis patients. Doing so unacceptably skews the risk-benefit ratio, at least in terms of OHSS and LB. That is, harvesting more than 15 oocytes, regardless of patient age or other contributing factors, does not necessarily increase the number of live babies born yet considerably inflates the incidence of clinically significant OHSS and the extreme morbidity that comes with it. Furthermore, stimulation resulting in high oocyte numbers may affect other factors, including increased costs of medication and laboratory fees, the potential for more frequent ethical dilemmas associated with supernumerary embryos, and contribution to higher steroid hormone levels during COS, the latter of which, as noted above, may worsen pregnancy outcomes owing to poorer endometrial receptivity 38, 39, 40, 41, 42, 43, 44, 45 and 46.

Given the apparent relationship between high retrieved oocyte number and OHSS without concomitant increase in LB rate, it seems reasonable that the time has come to realize the more common implementation of less aggressive stimulation protocols, especially in high responders. Such practice, it seems, would provide for the proverbial “best of both worlds” in the setting of fresh autologous IVF, particularly when planned freeze-all/programmed transfer cycles are not possible or desirable. Further large prospective trials are needed to validate this association between retrieved oocyte number, LB, and OHSS.