全文

Objective
To determine the predictive attributes of antimüllerian hormone (AMH) in terms of oocyte yield, cycle cancellation, and pregnancy outcomes.

Design
Retrospective cohort.

Setting
Academic center.

Patient(s)
All patients initiating IVF at the Weill-Cornell Ronald O. Perelman and Claudia Cohen Center for Reproductive Medicine from April 2010 through January 2013.

Intervention(s)
In vitro fertilization without preimplantation genetic testing.

Main Outcome Measure(s)
Number of oocytes retrieved, cycle cancellation, clinical and ongoing pregnancy, implantation, and miscarriage rates.

Result(s)
Antimüllerian hormone was positively correlated with number of eggs retrieved. Number of oocytes retrieved increased with increasing AMH within each age group and diminished slightly within AMH groupings as age increased. Overall, AMH was significantly correlated with risk of cycle cancellation, with an area under the curve (AUC) of 0.74. Patients with undetectable AMH had a 13.3-fold increased risk of cancellation as compared with patients with an AMH >2.0 ng/mL. Antimüllerian hormone had an AUC of 0.83 for prediction of three or fewer oocytes; undetectable AMH exhibited sensitivity and specificity of 21.1% and 98.2%, respectively, for three or fewer oocytes retrieved. Antimüllerian hormone was less predictive of pregnancy, with AUCs ranging from 0.55 to 0.65. Even with undetectable AMH, 23.5% of patients <40 years old achieved live birth after transfer.

Conclusion(s)
Antimüllerian hormone is a fairly robust metric for the prediction of cancellation and how many oocytes may be retrieved after stimulation but is a relatively poor test for prediction of pregnancy after any given treatment cycle. Patients with extremely low levels of AMH still can achieve reasonable treatment outcomes and should not be precluded from attempting IVF solely on the basis of an AMH value.

Key Words
IVF; antimüllerian hormone; AMH; predictive value; cancellation

Antimüllerian hormone (AMH) has gained increasing traction in recent years as a meaningful marker of ovarian reserve. Some physicians believe so strongly regarding its predictive attributes, in fact, that they advocate to potentially abandon more historically conventional markers, such as day-3 FSH (1). Indeed, there are now several modalities to measure ovarian reserve, including AMH, day-3 FSH, and antral follicle count, each with its own merits and shortcomings. Although acceptance is fairly universal that AMH is correlated with response to stimulation, it remains more controversial how, if at all, AMH is correlated with IVF outcomes. Moreover, there is a dearth of literature regarding IVF cycles from patients with extremely low levels of AMH, with much of the data reported coming from patients with normal or high reserve (2). To address these questions, we set out to analyze our own experience with the second-generation AMH assay since we implemented it as a routine part of our center's infertility workup in April 2010.

Materials and methods

Cycle Inclusion Criteria

This study was approved by the Weill Cornell Medical College institutional review board. Our center began using the second-generation AMH assay as an element of the fertility workup in April 2010. Thus all cycles at the Ronald O. Perelman and Claudia Cohen Center for Reproductive Medicine of Weill Cornell Medical College from April 2010 through January 2013 were analyzed for potential inclusion. Inclusion criteria were all women who had an AMH level assessed within the previous 12 months before their IVF cycle start. None of the women included were using hormonal contraception immediately before AMH determination. When analyzing data for pregnancy outcomes, cases of preimplantation genetic diagnosis and screening were excluded, as were cases in which transfer, either fresh or frozen, was not anticipated (i.e., fertility preservation for cancer or other reasons).

Clinical and Laboratory Protocols

All AMH measurements were performed on site at the Weill Cornell Center for Reproductive Medicine. Serum samples were stored at −20°C until assayed as batched samples once or twice weekly using the GenII Beckman ELISA assay (Beckman Coulter). The lowest detectable level of AMH was 0.006 ng/mL. However, because of the infrequency of samples found in this range, all low levels that were <0.17 ng/mL were reported as such. Intra- and interassay variability were 5.82% and 13.39%, respectively, for an AMH level of 2.8 ng/mL. Protocols for ovarian stimulation, hCG triggering, egg retrieval, embryo culture, and ET were conducted per our standard protocols, as previously described (3).

Briefly, patients were either down-regulated using a GnRH agonist (Lupron; Abbott Pharmaceuticals) with subsequent stimulation via gonadotropins (Follistim [Merck]; Gonal-F [EMD-Serono]; and/or Menopur [Ferring]) or were stimulated with gonadotropins until criteria were met for pituitary suppression with a GnRH antagonist (Ganirelix Acetate, 0.25 mg [Organon]; or Cetrotide, 0.25 mg [EMD-Serono]). The GnRH agonists were used in either “long” or “short” protocols as deemed clinically appropriate. For GnRH antagonist–based cycles, patients were started on 0.25 mg of Ganirelix or Cetrotide according to a flexible dosing scheme as previously described (4). In all cases, ovarian stimulation was carried out to maximize follicular response while minimizing risk of ovarian hyperstimulation syndrome. Age, AMH, antral follicle count, patient weight, and prior response to controlled ovarian stimulation were used to calculate the starting dose of gonadotropins for each patient's stimulation.

Patients with three or fewer follicles were counseled regarding the risks and benefits of continuing their IVF cycle vs. cancellation. In general, patients with three or more follicles were encouraged to proceed with IVF, whereas patients with one follicle were encouraged to either cancel or undergo IUI, although each patient's situation was evaluated on an individual basis, and unifollicular aspirations were performed (5). Proceeding with IVF vs. canceling in the setting of only two follicles was a highly individualized clinical decision after the couple and their attending physician had conferred.

Oocyte maturation was triggered via hCG (Profasi [EMD-Serono]; Novarel [Ferring Pharmaceuticals]; or Pregnyl [Schering-Plough]), according to a sliding scale dosing scheme as previously described, ranging from 10,000 IU to 3,300 IU according to serum E2(4). The hCG trigger was administered in general when the two lead follicles attained a mean diameter of 17 mm, or adjusted to earlier or later time points according to prior IVF outcomes. For GnRH antagonist–based cycles in which E2 exceeded 3,000 pg/mL, an ovulatory trigger of 2 mg leuprolide acetate in conjunction with 1,500 IU hCG was administered, with appropriate E2 and P luteal support (6). In all other cases, luteal support was begun the day after retrieval with 25–50 mg of intramuscular P.

Oocyte retrieval was performed in the standard fashion under conscious sedation (propofol plus fentanyl) 35–37 hours after hCG administration. Fertilization was performed using conventional insemination or intracytoplasmic sperm injection according to the couple's history and male partner's semen analysis. Embryos were incubated in sequential in-house culture media. Embryo cohorts of good quality and sufficient number on day 3 were cultured to the blastocyst stage; transfer otherwise occurred on day 3. Embryo transfers were performed with Wallace catheters (Marlow/Cooper Surgical).

Study Variables

Total number of oocytes retrieved, mature oocytes retrieved, risk of cycle cancellation, clinical and ongoing pregnancy, implantation rate, and risk of miscarriage were the outcome variables assessed. Age was stratified into groups according to Society for Assisted Reproductive Technology age categories. Clinical pregnancy rates were defined as the number of cycles with at least one fetal heartbeat at 7 weeks' gestation divided by either number of cycles initiated, number of retrieved cycles, or number of cycles with transfer, as above. Ongoing pregnancy rate was defined as number of live births plus number of ongoing gestations at 24 or more weeks' gestation. Implantation rate was defined as the number of sacs visible on ultrasound at 6 weeks' gestation divided by the number of embryos transferred.

Statistical Analysis

SAS version 9.3 (SAS Institute) was used to perform all data analyses. Descriptive summaries were reported as mean ± SD for continuous variables or as percentages for categorical variables. Spearman or Pearson correlation coefficients (r) were calculated to evaluate the relationships between continuous variables (e.g., number of oocytes retrieved and AMH level), depending on whether data were normally distributed. Because categorical AMH levels had unequal variances with regard to number of oocytes retrieved for each age group, ANOVA analysis adjusting for unequal variance was used to evaluate the effects of AMH levels. A nonparametric Kruskal-Wallis test was used to assess the effect of AMH levels on implantation, given the skewed distribution of the data and the fact that there were discrete values for implantation rate. Logistic regression was performed to assess the effect of AMH levels on binary outcomes. The Cochran-Armitage trend test was used to test whether rates of binary outcome variables increased with AMH levels. Bonferroni-adjusted P values (adj-P) were reported where adjustments for multiple tests were necessary. Sensitivity and specificity were calculated to determine the diagnostic characteristics of the AMH assay, and receiver operating characteristic (ROC) curves were created and AUC (area under the ROC curve) determined for AMH's predictive attributes. Analyses were confined to the patient's first cycle at our center so as to control for repeated measures bias. All P values were based on two-sided tests, and P<.05 was considered to be statistically significant.

Results

From April 2010 through January 2013, 2,760 individual patients had an AMH level drawn at our center and subsequently initiated a total of 4,072 cycles of stimulation with the intent to undergo IVF. So as to minimize repeated measures bias, data presented henceforth will be limited to first cycles only. Supplemental Table 1 (available online) shows the AMH level distributions across ages. Supplemental Table 2 shows the number of oocytes retrieved in relation to serum AMH level, stratified by patient age according to Society for Assisted Reproductive Technology categories. Oocyte retrievals were performed on 88.7% of initiated cycles.

As shown in Figure 1A, a positive linear correlation exists between increasing AMH values and total number of oocytes retrieved (r = 0.57 overall, P<.0001). For every age group, number of total eggs retrieved increased significantly with increasing AMH value (r range, 0.40–0.62, all P<.0001). Within select AMH groups (0.71–1.00 and 1.01–2.00 ng/mL), number of oocytes retrieved decreased slightly across age groups, with a small decrease in total number of oocytes retrieved for the oldest patients as compared with the youngest. The maximum number of oocytes retrieved in a patient with undetectable AMH was 13, and a maximum of 16 oocytes were retrieved in the 0.17–0.30 AMH group. Maximum number of oocytes retrieved did not differ significantly across age groupings within AMH strata. Interestingly, AMH was more correlated with number of oocytes retrieved than antral follicle count both overall (r = 0.46 vs. 0.57) and for each individual age grouping ( Supplemental Fig. 1).

Cancellation rate was next analyzed with respect to serum AMH and age. Patients with an undetectable AMH had a 37.3% chance of cycle cancellation (see criteria for cancellation above). A strong inverse relationship between increasing AMH level and cycle cancellation was observed (adj-P<.0001); cancellation rates were 37.3%, 30.8%, 15.2%, 4.6%, 5.3%, and 4.3% for AMH groups <0.17, 0.17–0.30, 0.31–0.70, 0.71–1.00, 1.01–2.00, and >2.00 ng/mL, respectively. As shown in Supplemental Table 3, this inverse relationship was maintained for each individual age group (adj-P=.015 for age <35 years, and adj-P<.0001 for age groups >35 years); younger patients tended to have lower cancellation rates as compared with older patients with the same AMH level. Table 1 shows the quantified overall increased risk (via logistic regression) of cycle cancellation with decreasing AMH; patients with an undetectable AMH were 13.3 times more likely than patients with AMH >2.0 ng/mL to be cancelled (odds ratio [95% confidence interval] 13.3 [7.9–22.3]; adj-P<.0001). Even at undetectable levels of AMH, however, sensitivity was only 16.0% and specificity 96.6% for prediction of cancellation. Using multivariable logistic regression, adding additional demographic variables such as body mass index, number of previous IVF cycles, starting dose of gonadotropins, or antral follicle count did not significantly help to further predict cancellation beyond that which was predicted by age and AMH alone. Day-3 FSH, however, was a significant factor independent from AMH for the prediction of cycle cancellation (P=.0077).

As shown in Figure 2A's ROC curves, AMH exhibited an AUC of 0.830, 0.823, and 0.813 for the ability to predict three or fewer, four or fewer, or five or fewer eggs retrieved, respectively.

An AMH cutoff of <0.17 ng/mL had a sensitivity of 21.1% and specificity of 98.2% for predicting three or fewer oocytes retrieved. Sensitivity decreased to 13.2% for the prediction of five or fewer oocytes retrieved (only 9.3% when limited to mature oocytes). Alternatively, an AMH cutoff of 0.5 ng/mL had a sensitivity of 66.2% and a specificity of 80.8% for the prediction of three or fewer oocytes retrieved, with sensitivity decreasing to 56.8% when raising the oocyte threshold to five or fewer eggs (47.1% for mature eggs).

Day-3 FSH provided additional information to more finely discern ovarian reserve in combination with AMH. As Figure 1B shows, an increase in FSH was associated with decreasing number of oocytes retrieved for any given AMH group (r range −0.22 to −0.07, all adj-Ps<.02 with the exception of AMH <0.17 ng/mL, P=.13). Moreover, the correlation between AMH and number of total eggs retrieved was impacted by level of FSH (≤10 vs. >10 mIU/mL), with the highest correlation observed when FSH was >10 mIU/mL (Pearson correlation 0.47, adj-P<.0001) as opposed to ≤10 mIU/mL (Pearson correlation 0.35, adj-P<.0001).

Pregnancy rates were next analyzed with respect to serum AMH levels. Table 2 shows clinical pregnancy rate per retrieval according to AMH level, stratified by patient age. For each age group there was a statistically significant trend toward higher pregnancy rates with increasing AMH, with the exception of patients aged >40 years. As shown in Figure 2B, however, the ROC curves for AMH's prediction of clinical pregnancy per retrieval were uniformly low (AUC 0.57, 0.57, 0.59, 0.55, and 0.65 for ages <35, 35–37, 38–40, 41–42, and >42 years, respectively). Ongoing pregnancy per transfer is shown in Supplemental Table 4. Here a trend for increasing ongoing pregnancy with increasing AMH was less pronounced and was present only in patients aged <35 and 38–40 years (adj-P=.040 and .031, respectively). The ROC analysis again revealed low AUC values for this outcome according to age (AUC 0.56, 0.56, 0.57, 0.56, and 0.69, respectively; data not shown). Information on clinical pregnancies per initiated cycles and per transfer, and for ongoing pregnancies per initiated cycles and per retrieval can be found in Supplemental Tables 5–8. All pregnancy and delivery rates represent cycles without preimplantation genetic screening.

Finally, implantation was analyzed with respect to AMH level, stratifying for patient age. A trend toward higher implantation with increasing AMH was evident in patients aged <35 years (and corresponded with a higher incidence of blastocyst transfer), but no such relationship could be demonstrated for older patients (data with P values shown in Supplemental Table 9). In patients >40 years old, a statistically significant trend toward more embryos transferred with increasing AMH was evident, given that more embryos were available for transfer (r = 0.30, P<.0001 for age 41–42 years and r = 0.39, P<.0001 for age >42 years). No statistical differences were observed between miscarriage rates and AMH within any individual age group.

Discussion

Since its relationship to ovarian response was first noted, AMH has increasingly become a mainstay of the fertility workup (7). Given that it can be measured throughout the menstrual cycle, exhibits relatively low intercycle variability, and provides finer discrimination, it offers several benefits that augment day-3 FSH as a serum indicator of ovarian reserve 8, 9 and 10. A number of studies have demonstrated the usefulness of AMH in refining the starting dose of gonadotropins so as to maximize response while minimizing risk of ovarian hyperstimulation syndrome 11, 12 and 13. The scope of AMH's predictive attributes in terms of portending IVF success, however, remains to be further elucidated.

In the present study we analyzed the association between AMH and a number of outcomes associated with IVF, including number of oocytes retrieved, cycle cancellation, pregnancy, implantation, and risk of miscarriage. Our data corroborate the utility of AMH as a powerful predictor of cycle cancellation and oocyte yield but also lends support to the fact that AMH remains a relatively poor predictor of pregnancy, with no cutoff able to exclude the possibility of pregnancy. Moreover, our data contain a relatively large proportion of patients with low AMH, revealing reasonable pregnancy rates with autologous oocytes and without preimplantation genetic screening, even at the lower limits of the assay.

It has been well established that a strong association exists between serum AMH and number of oocytes retrieved after stimulation 7, 12, 14, 15 and 16. A significant linear relationship is observed between oocyte yield and live birth up to 15 oocytes, with number of oocytes retrieved acting as an independent predictor of pregnancy; retrieval of more than 15 oocytes is not associated with better outcomes 17, 18, 19 and 20. Interestingly, we found that older patients with the same AMH as younger patients had a marginally poorer response to stimulation in terms of number of oocytes retrieved, as has previously been demonstrated (21). This finding lends further credence to the notion that AMH should not be interpreted in isolation, but rather should be addressed in the context of patient age, day-3 FSH, antral follicle count, and prior response to stimulation.

Whereas some authors have claimed that AMH is associated with oocyte quality (22), numerous subsequent studies have demonstrated that no such correlation exists (23). Our findings reveal relatively poor ROC curves for pregnancy prediction, along with no change in miscarriage rates according to AMH, supporting this notion. Although a trend toward higher implantation with increasing AMH was observed for patients aged <40 years, this may have been due to having multiple blastocysts from which to select, because no such relationship was observed in patients aged >40 years.

Several studies have examined the predictive capacity of AMH for determining risk of poor response and/or cycle cancellation 24, 25, 26 and 27. Antimüllerian hormone has been touted as a metric that can correctly predict poor response, thus helping practitioners and patients alike avoid treatment in women “destined not to respond to controlled ovarian stimulation” (28). The cutoff that has been claimed to correctly predict futility has been a moving target, however, with studies citing AMH levels ranging anywhere from 0.1 to 1.26 ng/mL 1, 28, 29 and 30. Lee et al. (30), for instance, observed no clinical pregnancies and a 47.6% cancellation rate in patients with an AMH <0.48 ng/mL and age ≥40 years, concluding that this AMH level may be used “as a marker of futility for counseling [such] IVF/ICSI candidates” (30). Others have concluded that an AMH of <0.28 ng/mL predicts cycle cancellation or retrieval of three or fewer oocytes in 98% of cases (27), or that AMH <0.29 ng/mL over age 42 years is associated with no live births (31). Such studies have been limited by small sample sizes at the extremes of AMH values and have used disparate criteria for cycle cancellation, thus making direct comparison between studies challenging. Our data contradict these studies.

Still, AMH is useful in terms of counseling patients regarding their risk of cycle cancellation, as shown by our data revealing a 37.3% cancellation rate for poor response in patients with undetectable AMH. Nonetheless, it is important to note that younger patients with very low AMH have lower cancellation rates than their older counterparts and thus should potentially be counseled somewhat more optimistically regarding their ability to complete an IVF cycle. Whether knowledge of a patient's AMH level influenced practitioners' recommendations to cancel the cycle remains unclear, because data regarding such decisions was not available to be analyzed; we think it unlikely, on the basis of the relatively uniform understanding regarding the clinical utility of AMH across providers in our practice. It is possible, however, that knowledge of a patient's low AMH made it less likely they would be cancelled, given that such a patient is less likely to have a more robust response in subsequent stimulations.

Last, it is notable that day-3 FSH provided additional information beyond AMH that further refined the ability to predict number of oocytes retrieved as well as cycle cancellation. For any given AMH value, increasing FSH was associated with a poorer response in terms of oocyte yield, thus adding a finer level of discernment as opposed to AMH alone.

In conclusion, our findings support the idea that low AMH is strongly associated with poor response but is less robustly associated with ongoing pregnancy after IVF (32). As others have stated, patients with extremely low ovarian reserve still exhibit moderate but reasonable live birth rates, and we strongly believe that low AMH levels in isolation do not represent an appropriate marker for withholding fertility treatment 28, 33 and 34. Even further, there is no single AMH cutoff below which pregnancy does not occur, and patients need not be dissuaded from pursuing treatment with their own eggs, provided they have been adequately counseled regarding the risk of cycle cancellation.

Acknowledgments

The authors thank the Weill Cornell Clinical Translational Science Center for support in providing assistance with statistical analyses, and specifically the help of Ya-lin Chiu, M.S. in this regard.