Significance of serum AMH and antral follicle count discrepancy for the prediction of ovarian stimulation response in Poseidon criteria patients

Abstract

Purpose

To determine the risk of not being a poor responder in ovarian stimulation (OS) for in vitro fertilization (IVF) when ovarian reserve markers are discordant—one falling within Poseidon’s criteria normal range (e.g., anti-Müllerian hormone (AMH) ≥ 1.2 ng/mL or antral follicle count (AFC) ≥ 5), and the other in the poor ovarian reserve range.

Methods

A tri-center retrospective cohort study (2015–2017) involving women with discordant AMH and AFC values undergoing their first IVF/ICSI cycle using conventional OS (cOS, ≥ 150 IU/day of follicle-stimulating hormone). Discordant serum AMH and AFC values were defined according to Poseidon’s criteria (AMH < 1.2 ng/mL and AFC ≥ 5 or AMH ≥ 1.2 ng/mL and AFC < 5). Poor ovarian response (POR) was < 4 retrieved oocytes. Receiver operating characteristic (ROC) curves were used to determine AMH and AFC cut-offs for non-POR. Logistic regression analysis evaluated factors associated with non-POR.

Results

Out of 8797 patients who underwent assessment with both AMH and AFC, 1172 (13.3%) exhibited discordant values. Of these, 854 (72.9%) had ≥ 4 oocytes retrieved. Within this group, 726 (85.0%) had “low” AMH values, whereas 128 (15.0%) had “low” AFCs. An AFC of 6 had 77% sensitivity and 52% specificity (AUC = 0.700), while AMH of 1.19 ng/mL had 31% sensitivity and 85% specificity (AUC = 0.492) for non-POR. AFC and the use of recombinant gonadotropins were positive predictors of non-POR.

Conclusions

When serum AMH is < 1.19 ng/mL, but AFC is ≥ 6, there is a moderate likelihood of a non-POR during stimulation. Conversely, if AFC is < 5 but serum AMH is ≥ 1.19 ng/mL, the chances of non-POR are low. Among patients with discordant markers, AFC emerges as the primary predictor of oocyte yield.

Introduction

Ovarian reserve markers hold diagnostic and prognostic value in the infertile couple’s work-up [1]. Diminished ovarian reserve (DOR) is a contributor to infertility and may impact assisted reproductive technology (ART) success rates [1, 2]. Precise evaluation of ovarian reserve, particularly with anti-Müllerian hormone (AMH) levels and the antral follicle count (AFC), aids in predicting ovarian stimulation response during in vitro fertilization [3, 4]. These markers enable personalized gonadotropin dose selection and counseling concerning ovarian stimulation outcomes [5, 6]. Patients with DOR tend to exhibit poor ovarian response (POR) to gonadotropin stimulation [7, 8], resulting in fewer retrieved oocytes [9] and diminished reproductive outcomes, including decreased cumulative live birth rates (cLBRs) [10].

AMH is synthesized by granulosa cells in pre-antral and antral follicles, playing a role in preventing premature follicular depletion [11]. Since it mirrors the ovarian follicular pool, diminishing follicle count leads to decreased serum AMH levels. AMH and AFC both positively reflect ovarian reserve and are anticipated to align [12, 13]. These parameters are easily measurable in patients’ serum and by transvaginal ultrasound, respectively, and are often used by clinicians in routine practice [3, 4]. However, as demonstrated in prior research [14–17], incongruence in these biomarkers’ values complicates predicting ovarian response to gonadotropin stimulation and expected oocyte yield, pivotal markers of ART success.

Over the past decade, diagnostic standardization of POR has advanced through Bologna criteria [9] and Poseidon criteria [18, 19]. The latter categorize patients based on age, oocyte/embryo aneuploidy risk, ovarian reserve biomarkers (AFC and/or AMH), and response to ovarian stimulation [18, 20]. The Poseidon criteria define “expected” (groups 3 and 4, < 35 and ≥ 35 years) and “unexpected” PORs (groups 1a and 2a, < 35 and ≥ 35 years), with < 4 retrieved oocytes after conventional ovarian stimulation (COS) [1, 19]. Therefore, the Poseidon system assigns patients to groups depending on age and ovarian reserve markers, with an AFC ≥ 5 and/or an AMH ≥ 1.2 ng/mL considered normal ovarian reserve markers. Yet, addressing patients with discordant ovarian marker values (i.e., AMH < 1.2 ng/mL and an AFC ≥ 5, or an AMH ≥ 1.2 ng/mL and an AFC < 5) is challenging. Identifying the more precise marker for non-poor response (> 4 retrieved oocytes) in discordant cases remains unresolved. Likewise, optimal cut-offs for discordant AMH and AFC values, ensuring high sensitivity and specificity in non-POR prediction, need to be determined.

Hence, our study had three primary objectives. First, we aimed to report the prevalence of non-POR (defined as obtaining ≥ 4 oocytes) in IVF conventional ovarian stimulation when discordant ovarian reserve markers were present. These markers included one indicating normal ovarian reserve according to Poseidon criteria (e.g., AMH ≥ 1.2 ng/mL or AFC ≥ 5), while the other indicated DOR. Secondly, we sought to identify the threshold values of AMH and AFC within the discordant group that could distinguish a higher likelihood of obtaining ≥ 4 oocytes. Lastly, we investigated the factors associated with non-POR in patients exhibiting discordant ovarian reserve markers.

Materials and methods

This retrospective cohort study involved consecutive infertile women aged 22–45 undergoing their first IVF/ICSI cycle at three centers (ANDROFERT, Andrology and Human Reproduction Clinic, in Campinas, Brazil; Anatolia IVF, in Ankara, Turkey; and IVFMD, My Duc Hospital, in Ho Chi Minh City, Vietnam) during 2015–2017.

Inclusion criteria encompassed (i) assessment of ovarian reserve by both AMH and AFC within 3 months of IVF cycle initiation, (ii) discordant serum AMH and AFC values—one normal per Poseidon criteria, the other indicating DOR (i.e., AMH ≥ 1.2 ng/mL and AFC < 5, or AFC ≥ 5 and AMH < 1.2 ng/mL), (iii) standard OS with ≥ 150 IU/day of FSH, and (iv) a cycle that ended in oocyte collection. Each patient contributed only one IVF/ICSI cycle. Notably, none of the participating patients was treated with oral contraceptive pills or any other hormonal contraception prior to ovarian reserve measurements.

Exclusion criteria were (i) patients lacking both ovarian reserve markers assessment, (ii) having concordant AMH and AFC values (both normal or both low per Poseidon), (iii) undergoing IVF/ICSI for reasons other than infertility (e.g., donor cycles, preimplantation genetic testing for monogenic diseases, and fertility preservation cycles), and (iv) undergoing natural cycle IVF or mild stimulation protocols, as per Poseidon criteria [21], standard stimulation is required for patient classification.

Eligible patients were assessed and treated according to each institution’s protocols, as previously detailed [22]. Ovarian reserve evaluations were conducted 1–3 months before stimulation in a natural menstrual cycle, following standardized procedures. Early follicular phase AFC measurements were carried out using two-dimensional transvaginal ultrasonography, performed by physicians at each study center, adhering to practice AFC guidelines [3]. AMH serum values were measured using the modified Beckman Coulter generation II enzyme-linked immunosorbent assay [23], as previously described [24]. Both markers were evaluated within the same patient’s menstrual cycle, though AMH results were unavailable during AFC assessments. Additionally, both AFC and AMH results were accessible before any chosen OS regimen.

OS regimen and gonadotropin dosage were determined according to each center’s policies, considering ovarian reserve and the patient’s age. The standard OS included two protocols: (i) long gonadotropin-releasing hormone (GnRH) agonist protocol and (ii) GnRH antagonist protocol. Patients received daily subcutaneous injections of (i) highly purified human menopausal gonadotropins (hMG), (ii) recombinant-FSH, (iii) a combination of recombinant-FSH and hMG, or (iv) a combination of recombinant-FSH and recombinant-LH (in a 2:1 ratio).

Daily gonadotropin doses ranged from 150 to 450 IU. Ovarian response was monitored through sequential transvaginal ultrasonography and estradiol measurements, with dose adjustments as needed. Both fixed and flexible GnRH antagonist protocols were employed. To trigger final oocyte maturation, subcutaneous administration of either (i) recombinant-human chorionic gonadotropin (hCG) at 250 mcg or (ii) GnRH agonist at 0.2 mg triptorelin was used. Oocyte retrieval was performed by transvaginal ultrasound-guided puncture of follicles 35–37 h after the trigger injection. Collected follicular fluid was assessed in the IVF laboratory, and the total retrieved oocyte count was recorded.

We collected data on patient demographics, infertility workup, and cycle characteristics. PORs were those with < 4 oocytes retrieved, whereas non-PORs were patients with ≥ 4 oocytes retrieved. The primary outcome was the identification of threshold serum AMH and AFC values distinguishing patients likely to retrieve ≥ 4 oocytes (non-POR). Secondary outcomes included (i) non-POR prevalence (i.e., obtaining ≥ 4 oocytes) and (ii) factors associated with POR in patients exhibiting discordant markers.

Data were modeled using receiver operating characteristic (ROC) curves and Youden J statistics. Multivariable nominal logistic regression analyses examined the association between patient clinical and treatment parameters and non-POR (i.e., collecting four or more oocytes). The study center was considered a fixed factor. We conducted two-way (study center vs. predictor) interaction tests to investigate whether the “study center” had any relation between the predictors and the response variable. Clinical predictors included female age, AFC, AMH, body mass index (BMI), infertility duration, and infertility etiology. Treatment predictors included gonadotropin type (HMG, rec-FSH + HMG, rec-FSH alone, rec-FSH + rec-LH), trigger type (hCG trigger, GnRH-agonist trigger), duration of stimulation, total gonadotropin dose, and daily gonadotropin dose. Categorical data, expressed as counts and percentages, were analyzed by Pearson chi-square. Continuous data are presented as medians and interquartile ranges (IQR) and were analyzed by Kruskal–Wallis and Wilcoxon tests, as appropriate. Computations were performed using JMP PRO 16 (SAS Institute, Cary, NC, USA).

The study received ethical approval from the institutions’ ethics committees: Instituto Investiga, Brazil (CAAE 26429219.0.0000.5599); Hacettepe University, Turkey (KA-180070); and My Duc Hospital, Vietnam (05/18/DD-BVMD).

Results

During the study duration, a total of 8797 patients underwent assessments of both AMH and AFC. Among them, 1172 individuals (13.3%) displayed discordant serum AMH and AFC values. The demographics and treatment characteristics of this group with discordant ovarian reserve parameters are detailed in Table 1. Specifically, within this group, 943 patients (80.5%) exhibited “low” AMH values (< 1.2 ng/mL) and adequate AFC (≥ 5), per Poseidon criteria, while 229 patients (19.5%) had “low” AFC values (< 5) coupled with adequate AMH values (≥ 1.2 ng/mL). Out of the 1172 patients with discordant ovarian biomarkers values, 854 (72.9%) attained 4 or more retrieved oocytes. Among the subset of 854 individuals with discordant biomarker values and ≥ 4 oocytes retrieved, 726 patients (85.0%) had “low” AMH values by Poseidon criteria, whereas 128 patients (15.0%) displayed “low” AFC values (Fig. 1).

Table 1.

Baseline characteristics of the study cohort (patients with discrepant ovarian reserve parameters) (N = 1172)

Variable
Age (years)	36 (33–39)
BMIa (kg/m2)	21.37 (19.91–23.34)
AMHb	0.94 (0.65–1.15)
AFCc	6 (5–8)
Ovarian stimulation protocol
GnRHd antagonist	1070 (91.3)
Long GnRH agonist	102 (8.7)
Gonadotropins utilized
HMGe	10 (0.85)
Rec-FSHf	211 (18)
Rec-FSH + HMG	776 (66.21)
Rec-FSH + rec-LHg	175 (14.93)
Trigger type
GnRH agonist trigger	62 (5.29)
hCGh	1110 (94.71)
Total gonadotropin dose (IU)	3000 (2550–3600)
Daily gonadotropin dose (IU)	358.33 (300–383.33)
Duration of stimulation (days)	9 (8–10)
Duration of infertility (months)	48 (24–84)
Primary infertility cause
Two or more female factors	50 (4.27)
Advanced maternal age (≥ 39 years)	189 (16.13)
Anatomic (e.g., uterine factor)	8 (0.68)
Endocrine (e.g., PCOSi)	15 (1.28)
Endometriosis	40 (3.41)
Male factor	194 (16.55)
Tubal factor	117 (9.98)
Unexplained	530 (45.22)
Other	29 (2.47)
Number of collected oocytes	5 (3–8)
MIIj oocytes	4 (2–6)

Continuous variables are presented as median (interquartile range (IQR)) and categorical variables are presented as n (%)

^aBMI, body mass index; ^bAMH, anti-Müllerian hormone; ^cAFC, antral follicle count; ^dGnRH, gonadotropin-releasing hormone; ^eHMG, human menopausal gonadotropins; ^frec-FSH, recombinant follicle-stimulating hormone; ^grec-LH, recombinant luteinizing hormone; ^hhCG, human chorionic gonadotropin; ⁱPCOS, polycystic ovary syndrome; ^jMII, metaphase II

Fig. 1.

Patient breakdown

Table 2 compares patients with discordant ovarian reserve biomarker values and < 4 oocytes retrieved versus those with discordant values and ≥ 4 oocytes retrieved. Notably, those with < 4 oocytes were older (37 vs. 36 years, p < 0.0001), had a lower median BMI (21.1 vs. 21.63 kg/m², p = 0.04), exhibited a lower median AFC (5 vs. 7, p < 0.0001), had a longer median duration of infertility (60 vs. 48 months, p = 0.023), were more likely to have undergone IVF due to advanced maternal age (22.33% vs. 13.82%, p = 0.013), and showed a lower median number of collected oocytes (2 vs. 6, p < 0.0001) and MII oocytes (2 vs. 5, p < 0.0001).

Table 2.

Baseline characteristics of patients with discordant ovarian reserve parameters with < 4 oocytes retrieved versus 4 or more oocytes retrieved

Variable	< 4 oocytes retrieved (N = 318, 27.13%)	≥ 4 oocytes retrieved (N = 854, 72.87%)	p-value
Age (years)	37 (33.75–40)	36 (32–39)	< .0001
BMIa (kg/m2)	21.1 (19.65–23.31)	21.63 (19.98–23.4)	0.04
AMHb	0.9 (0.52–1.32)	0.96 (0.7–1.14)	0.658
AFCc	5 (4–6)	7 (5–9)	< .0001
Ovarian stimulation protocol			0.081
GnRHd antagonist	298 (93.71)	772 (90.4)
Long GnRH agonist	20 (6.29)	82 (9.6)
Gonadotropins utilized			0.0002
HMGe	7 (2.2)	3 (0.35)
Rec-FSHf	44 (13.84)	167 (19.56)
Rec-FSH + HMG	205 (64.47)	571 (66.86)
Rec-FSH + rec-LHg	62 (19.5)	113 (13.23)
Trigger type			0.557
GnRH agonist trigger	19 (5.97)	43 (5.04)
hCGh	299 (94.03)	811 (94.96)
Total gonadotropin dose (IU)	3000 (2400–3750)	3000 (2550–3525)	0.515
Daily gonadotropin dose (IU)	364.29 (300–383.33)	347.73 (300–383.33)	0.505
Duration of stimulation (days)	9 (8–10)	9 (8–10)	0.054
Duration of infertility (months)	60 (33–96)	48 (24–84)	0.023
Infertility cause			0.013
Two or more female factors	12 (3.77)	38 (4.45)
Advanced maternal age	71 (22.33)	118 (13.82)
Anatomic (e.g., uterine factor)	1 (0.31)	7 (0.82)
Endocrine (e.g., PCOSi)	1 (0.31)	14 (1.64)
Endometriosis	12 (3.77)	28 (3.28)
Male factor	42 (13.21)	152 (17.8)
Other	9 (2.83)	20 (2.34)
Tubal factor	26 (8.18)	91 (10.66)
Unexplained	144 (45.28)	386 (45.2)
Number of collected oocytes	2 (1–3)	6 (5–8)	< .0001
MIIj oocytes	2 (1–2)	5 (4–7)	< .0001

Continuous variables are presented as median (interquartile range (IQR)) and categorical variables are presented as n (%)

^aBMI, body mass index; ^bAMH, anti-Müllerian hormone; ^cAFC, antral follicle count; ^dGnRH, gonadotropin-releasing hormone; ^eHMG, human menopausal gonadotropins; ^frec-FSH, recombinant follicle-stimulating hormone; ^grec-LH, recombinant luteinizing hormone; ^hhCG, human chorionic gonadotropin; ⁱPCOS, polycystic ovary syndrome; ^jMII, metaphase II

Figure 2 illustrates the AMH and AFC ROC curves, evaluating the cut-offs for ≥ 4 oocytes retrieved in the discordant group. An AMH value of 1.19 ng/mL yielded a sensitivity of 31% and a specificity of 85% (AUC = 0.492) in identifying patients more likely to have ≥ 4 oocytes retrieved (Fig. 2A). Additionally, an AFC value of 6 exhibited a sensitivity of 77% and specificity of 52% (AUC = 0.700) in distinguishing patients with an increased likelihood of ≥ 4 oocytes retrieved (Fig. 2B).

Fig. 2.

Receiver operating characteristic curves for predicting non-poor response (≥ 4 oocytes retrieved) in patients with discordant biomarkers’ values per the Poseidon criteria. A AMH; B AFC

In the multivariable logistic regression analysis exploring the association between clinical factors and retrieval of ≥ 4 oocytes (non-POR by Poseidon criteria), AFC and gonadotropin type were significant predictors. Specifically, a positive relationship was observed between AFC (P < 0.0001) and non-POR and between OS with recombinant gonadotropins (i.e., FSH alone or combined with LH activity provided by HMG or rec-LH; P = 0.02) and non-POR (P = 0.02). When exploring the interactions between the variable “study center” and predictors, we found no significant interactions between study centers and predictors concerning the binary outcome “non-POR by Poseidon criteria.”

Discussion

Our main goal was to determine the AMH and AFC threshold values for accurate prediction of non-POR during IVF ovarian stimulation, specifically when AMH and AFC were discordant, with one marker within Poseidon criteria’s normal range (e.g., anti-Müllerian hormone (AMH) ≥ 1.2 ng/mL or antral follicle count (AFC) ≥ 5), and the other indicating poor ovarian reserve. Our findings can be summarized as follows: (1) We identified a prevalence of 12.3% for discordant ovarian reserve markers within our cohort, in which 85.0% had “low” AMH (< 1.2 ng/mL) per Poseidon criteria, while 15.0% had “low” AFC values. (2) In the discordant group, ROC analysis for achieving ≥ 4 oocytes retrieved showed that an AFC of 6 had 77% sensitivity and 52% specificity (AUC = 0.700), while an AMH threshold of 1.19 ng/mL had 31% sensitivity and 85% specificity (AUC = 0.492); (3) AFC and the use of recombinant gonadotropins for OS were positive predictors of non-POR.

The inconsistency between AFC and AMH values can be attributed to various technical, physiological, and external factors related to these biomarkers, as noted in prior studies [3, 25–28]. Both markers exhibit fluctuations within and across cycles [29], and AFC is susceptible to marked inter- and intra-operator variability [3]. In a previous retrospective study by Guo et al. [30] involving 19,239 patients in their first fresh IVF/ ICSI cycle, around 10% of infertile women experienced a discrepancy between measured AMH and AFC, a prevalence closely resembling our findings. Moreover, another retrospective study [17] reported an AMH and AFC discordance rate of 20% among their infertile population undergoing IVF/ICSI.

We also found that AFC was more accurate than AMH in predicting non-POR in patients with discordant biomarkers under the Poseidon classification. These findings may be justified by the fact that an AFC threshold of 6 is more restrictive and conservative than the correspondent AMH of 1.19 ng/mL. Additionally, a possible explanation for the superiority of AFC over AMH for predicting oocyte yield may be related to the entities each ovarian reserve parameter represents. AFC refers to the number of 2–9 mm follicles responsive to gonadotropins that may be selected for growth and development during OS and, therefore, represents the recruitable follicular pool [31]. By contrast, AMH, produced by granulosa cells of pre-antral and small antral follicles of less than 4 mm [32], is an indirect measure of early-growing follicles [33] that may not all be able to enter follicular recruitment during OS.

Only a limited number of studies have investigated the predictive value of discordant AMH and AFC on ovarian stimulation success, yielding conflicting results. Similar to our findings, a retrospective study [17] evaluating 203 patients with discordant AFC and AMH levels found AFC to be a more effective predictor than AMH for POR. This study concluded that AFC should be the preferred indicator for predicting ovarian response, thus aiding in developing optimal individualized OS protocols. The authors of the study, as mentioned above, divided participants into four groups based on Bologna criteria [9] (AFC < 7 and an AMH < 1.1 indicating DOR). Among these groups, Group A had both AFC and AMH levels within the normal range, Group B had normal AFC but low AMH levels, Group C had low AFC but normal AMH levels, and Group D had both low AFC and AMH levels. In the two groups with discordant AFC and AMH levels, Group B demonstrated significantly higher oocyte yield, good-quality embryo rate, and clinical pregnancy rate than Group C. The incidence of non-POR was notably higher in Group B than in Group C. Stratifying by female age for all age categories (including ≤ 30 years and those above), Group B consistently exhibited higher oocyte yield and clinical pregnancy rate than Group C, indicating superior predictive accuracy of outcome by AFC over AMH. However, unlike our study, this study employed the threshold values of AMH and AFC as per the Bologna criteria.

In another report, Aslan et al. [34] aimed to determine which ovarian reserve marker was more predictive when encountering discordant AMH and AFC among patients with DOR (n = 662 cycles). Patients undergoing ICSI due to DOR were divided into three groups: Group 1 (n = 418 cycles)—patients with low AMH (< 1.1 ng/mL) and low AFC (< 7); Group 2 (n = 167 cycles)—low AMH (< 1.1 ng/mL) and normal AFC (≥ 7); and Group 3 (n = 77 cycles)—normal AMH (≥ 1.1 ng/dL) and low AFC (< 7). Demographics, follicle output rate (FORT) score, and follicle to oocyte index (FOI) score of the groups were analyzed. FORT and FOI serve as reliable qualitative markers of ovarian response, gauging the alignment between the pool of antral follicles at the beginning of OS and the number of pre-ovulatory follicles at the end (33–35). A low FORT or FOI usually translates into fewer pre-ovulatory follicles/oocytes following gonadotropin stimulation than anticipated based on AFC/AMH levels.

In the study mentioned above, as the primary outcome, FORT and FOI scores were higher in Group 3 than in the other two groups. Thus, the authors concluded that serum AMH level was more predictive of stimulation success when discordant with AFC. This contrasts with our findings, which revealed a higher precision of AFC in predicting the number of oocytes retrieved (an AFC < 5 with normal AMH indicating a low likelihood of retrieving more than 3 oocytes). Notably, unlike our study, the authors did not utilize the Poseidon-defined thresholds for DOR, nor did they determine stimulation success based on the total number of retrieved oocytes and the categorization of patients as POR or non-POR. Moreover, they did not explore the AMH and AFC threshold values for discordant results predicting a non-POR with optimal sensitivity and specificity.

In another retrospective study [30], 19,239 patients undergoing their first fresh IVF/ICSI cycle with various GnRH protocols (antagonist, GnRH-a long, and GnRH-a ultra-long) were evaluated. The authors grouped their patients into four categories based on the AFC and serum AMH thresholds established by the Poseidon criteria to predict POR: Group 1 (normal AMH and normal AFC); Group 2 (low AMH and normal AFC); Group 3 (normal AMH and low AFC); and Group 4 (low AMH and low AFC). They found that Group 3 patients exhibited significantly higher oocyte yield and number of available embryos than Group 2 (6.94 ± 3.67 vs. 6.23 ± 3.63, p < 0.001; and 4.26 ± 2.76 vs. 3.75 ± 2.66, p < 0.001, respectively).

Furthermore, their results indicated a gradual decrease in the number of retrieved oocytes (p < 0.001) and MII oocytes (p < 0.001), accompanied by a progressive increase in the incidence of POR (p < 0.001) from Group 1 to Group 3 to Group 2 and finally to Group 4. This trend suggested that ovarian responsiveness in patients with discordant AMH and AFC falls between the responsiveness seen when both markers are either concordantly low or concordantly normal. This finding aligns with prior research [35]. However, their conclusions differed from our study’s findings, where we identified AFC as a more accurate determinant for predicting the number of retrieved oocytes, displaying a higher AUC than AMH. Noteworthy, the study above examined the number of retrieved oocytes and embryos available as continuous variables rather than examining the AMH and AFC cut-offs for predicting a non-POR according to Poseidon’s classification, as we did. This aspect potentially limits the practical application of their results in clinical practice.

Lastly, in another retrospective study [15], the correlation between AMH and AFC and their influence on oocyte yield in OS was investigated across 37 different fertility clinics, involving a total of 519 infertile patients with a favorable prognosis. The study evaluated both a GnRH-agonist-treated cohort and a GnRH-antagonist. The results indicated that, as a continuous variable, AMH exhibited a stronger correlation with oocyte yield than AFC in both GnRH-agonist and GnRH-antagonist protocols. As a result, the researchers concluded that AMH served as a more robust predictor of ovarian response to gonadotropin stimulation, regardless of whether GnRH-agonist or GnRH-antagonist protocols were applied. Nevertheless, this study had certain limitations as it excluded POR from previous stimulation cycles, which might have impacted the generalizability of the findings. Additionally, the study did not assess the true discordance between AMH and AFC in terms of “normal” versus “poor” ovarian reserve thresholds as defined by Poseidon’s criteria. Moreover, the primary outcome measure did not involve categorizing responses as poor or non-poor but instead focused on the continuous count of retrieved oocytes. This aspect restricted the clinical relevance and applicability of the study’s outcomes.

We found that AFC and the use of recombinant gonadotropins were positively associated with the likelihood of experiencing non-POR. These findings are consistent with those of Zhang et al. [17], who also found that AFC was a more effective predictor than AMH. Regarding the association between recombinant gonadotropins and ovarian response, our results align with previous studies showing a higher oocyte yield when using recombinant FSH instead of highly purified hMG. This trend was observed in studies using the long GnRH agonist protocol [36] and in a meta-analysis of sixteen randomized controlled trials [37]. Similarly, a retrospective study including 30,630 IVF cycles from four European countries [38] demonstrated that IVF treatment cycles using rFSH yielded statistically more oocytes (and more mature oocytes) while requiring lower total gonadotropin dosages per cycle compared to highly purified hMG (10.8 ± 6.02 vs. 9.77 ± 5.53; and 8.58 ± 5.27 vs. 7.72 ± 4.59, respectively).

Our findings are also consistent with the Merit study, a randomized controlled trial [39], which reported 1.8 oocytes less in the highly purified hMG group, favoring rFSH. A proposed explanation for this phenomenon [40] is that LH activity (provided by hCG in hMG preparations) may trigger atresia of some of the follicles during the follicular phase. This observation was also noted when comparing hMG and rFSH in ovulation induction in World Health Organization Group II anovulatory infertility patients [41], where patients receiving highly purified hMG had fewer intermediate follicles. However, the reasons why recombinant FSH, even with added LH, was found to be associated with non-POR in our study remains to be clarified. It is possible that the recombinant properties of the product render it more efficient for follicular recruitment than urinary FSH or hMG, though this would need to be explored in further pharmacologic studies. Also, rec-LH supplementation may overcome the impairment in the function of the LH system seen in some patients with DOR that might be exacerbated by the pituitary suppression regimen used for OS [42].

The strength of our study primarily lies in its large sample size, which facilitated the creation of ROC curves, enabling the derivation of AMH and AFC thresholds for predicting a non-POR to stimulation within the cohort. Another strength was the use of a multi-center database to increase the applicability of the findings across three distinct populations: South American (predominantly Brazilian), Southeast Asian, and Turkish European, which includes a significant number of Middle Eastern subjects.

To our knowledge, this is the largest and only study to date that evaluated the AMH and AFC threshold values for predicting a non-POR to OS, all based on Poseidon’s criteria, in cases where AMH and AFC values are discordant. Identifying these threshold values can enhance the accuracy of predicting ovarian response to stimulation and facilitate the customization of treatment protocols and gonadotropin dosages. Furthermore, our finding of the superior accuracy of AFC in prognosticating oocyte yield may suggest cautious reliance on AFC for treatment planning, particularly when confronted with discordant AMH results.

Our study’s limitations include its retrospective design, which might introduce unnoticed bias. Additionally, AFC measurements were conducted by different operators using varying machines, potentially affecting accuracy despite the generally low inter-observer variability reported for AFC [43]. Similarly, AMH measurements were obtained from different laboratories, although using the same manual assay. While automated assays tend to exhibit lower inter-laboratory variability than manual assays [44], the latter was the only available method during our study period. Nonetheless, efforts were made to decrease this potential source of bias in the design phase of the study by selecting study centers that shared the same assay for AMH measurements and used similar standard operating procedures for AFC determination. Therefore, though we cannot entirely rule out that assay-related issues may have contributed to the discordancy between AFC and AMH, this would likely have been similar across the study centers. Lastly, genetic variations in FSH receptors could lead to varying ovarian responses across populations, a factor not accounted for in our model. A prospective study on a large database would provide more robust insights.

In conclusion, under the Poseidon stratification, if serum AMH level is below 1.19 ng/mL, but AFC is at least 6, there is a moderate likelihood of a non-POR to stimulation. We cautiously suggest these patients be treated a priori with the maximal FSH dosage during ovarian stimulation, in accordance with the European Society of Human Reproduction and Embryology guidelines [45] and be counseled accordingly about the potential risk of a poor response to stimulation. Although these patients are moderately likely to have a non-POR to stimulation, such treatment tailoring may lead to improved treatment outcomes, with an increased oocyte yield for these specific patients. Conversely, if the AFC is less than 5 and the serum AMH is at least 1.19 ng/mL, the chances of obtaining more than 3 oocytes are relatively low. In cases of discordant ovarian reserve markers according to the Poseidon criteria and when the risk of a poor response to ovarian stimulation is apparent, AFC emerges as the key predictor of the oocyte yield. Patients should be counseled accordingly based on these findings.

Author contribution

AH, MHD, and SCE designed the study and helped with data acquisition, analysis, and interpretation. HY and LV participated in data acquisition, interpretation, and article revision for critical intellectual content. All authors contributed intellectually to the writing or revision of the manuscript, approved the final version, and are accountable for all aspects of the work.

Data availability

All data supporting the findings of this study are available within the paper.

Declarations

Ethics approval

This study was performed in line with the principles of the Declaration of Helsinki. The study received ethical approval from the institutions’ ethics committees: Instituto Investiga, Brazil (CAAE 26429219.0.0000.5599); Hacettepe University, Turkey (KA-180070); and My Duc Hospital, Vietnam (05/18/DD-BVMD).

Informed consent

Informed consent was waived due to the retrospective design of the study.

Competing interests

MHD received unrestricted research grants from Merck Serono and Ferring. HY declares receipt of payment for lectures from Merck and Ferring. LNV receives speaker fees and conferences from Merck, Merck Sharp and Dohme (MSD), and Ferring and research grants from MSD and Ferring. SCE declares receipt of unrestricted research grants from Merck and lecture fees from Merck and MedEA. AH has no conflict of interest to declare.