Controlled ovarian stimulation (COS) is a fundamental component of assisted reproductive technologies (ART) (1). It maximizes the potential for successful outcomes by stimulating multiple follicles to mature simultaneously, thereby increasing the availability of viable embryos for transfer. However, multiple follicle growth also raises concerns regarding the excessive ovarian response. During the process, follicular development and serum E₂ are closely monitored to justify the decision to trigger oocyte maturation and in the prevention of ovarian hyperstimulation syndrome (OHSS) (2, 3).

The importance of E₂ in follicular development in both natural and stimulated cycles has been well-established, as it is secreted by the granulosa cells (GC) in response to endogenous or exogenous follicle-stimulating hormone (FSH) stimulation to support the follicular growth and maturation (4). During monitoring of stimulated cycles, serum levels of E₂ are expected to increase steadily with follicle growth until final maturation (5). However, unexpected E₂ decline during COS monitoring is also reported in monitored COS cycles, yet its clinical significance remains controversial. An early study suggested that a decline in serum E₂ before triggering is associated with a dramatically decreased pregnancy rate (6). However, milder effects or no effect of E₂ decline during COS monitoring are also reported in other studies (7). The conflicting conclusions surrounding the clinical significance of E₂ decline during COS can be attributed to various factors, such as limited sample size, heterogeneity of patients, or failure to adjust for important confounders. Importantly, previous research has predominantly focused on the outcomes of the fresh transfer, where patients with excessive ovarian response and patients with very poor response may both be excluded from the fresh transfer cycle. Selection bias may occur when paitents being included in the study basing of their exposure or outcomes status (8). In addition, the role of chance may also be a consideration when only fresh transfer is evaluated. Due to the morphology-based embryo selection having only limited discriminatory power (9), the “correct” embryos may be selected following multiple transfers.

Evaluating the cumulative birth rates taking into account all transfer attempts in a complete COS cycle (10) may minimize the bias associated with fresh transfer and the random effect of embryo selection. Moreover, the majority of the previous studies have associated the E₂ decline with decreased fertilization or reduced embryo yield, suggesting fewer chances of transfer attempts (11–13). We hypothesize that an E₂ decline observed during COS monitoring may compromise the cumulative live birth via the mediation of reduced embryo yield, even if the fresh transfer outcomes are not affected. The present study aims to evaluate the impact of E₂ decline during COS monitoring on the cumulative live birth rate in a large COS cohort, exploring the mediation effect of embryo yield and the contribution of patient heterogeneity.

In this study, 30947 ART cycles were reviewed for potential inclusion, and 27487 cycles were finally included. The number of cycles that underwent at least one E₂ decline was 2863 (10.3%). The characteristics of the patients are shown in Table 1 . The patients encountering E₂ decline during ovarian stimulation monitoring were associated with older female age, higher BMI, poorer AFC, and a lower proportion of agonist cycle. However, they also have a higher proportion of PCOS diagnoses and a history of delivering live births. Some of the characteristics of male counterparts, including age and Total motile sperm count (TMC), were also significantly different between patients with and without E₂ decline. Following PS-matching, the standardized differences (D) for all the covariates were lower than 0.1, and the distribution of propensity scores (distance) was identical between comparison groups ( Supplementary Figure S2 ).

Table 2 shows the outcomes of patients with or without E₂ decreased. The decrease in E₂ levels during monitoring was associated with poorer ovarian response, which led to a prolonged duration of stimulation, increased total gonadotropin dosage, and a lower oocyte yield in both matched and unmatched cohorts. The number of mature oocytes, embryos, and high-quality embryos also decreased accordingly. These changes ultimately resulted in approximately a 4% difference (59% VS 55%, P = 0.002) in the cumulative live birth rate in the matched cohort. In the multivariate GLM analyses, the adjusted odds ratios (OR) for cumulative live birth were similar in the unmatched and the matched cohort (OR 0.83, 95% CI: 0.76,0.91 for the unmatched cohort; OR 0.84, 95%CI: 0.75,0.94). On the other hand, the live birth following fresh transfer did not significantly differ between patients with and without an E₂ decline after PS-matching.

The mediation analyses suggested that the decreased cumulative live birth rate following E₂ decline was largely mediated by the reduced number of mature oocytes or decreased embryo availability. The mature oocyte yield mediated 75.3% (P<0.001) of CLBR decline in the unmatched cohort and 78.3% (P = 0.006) in the matched cohort. Embryo number mediated 76.5% (P = 0.002) and 72.5% (P = 0.01) in the unmatched and matched cohorts, respectively.

The GAM model suggested a U-shaped association between E₂ decline and cumulative live birth (edf=2, P = 0.01). The cumulative live birth rates were negatively associated with the degree of E₂ decline when the differences between the two visits were less than 108.9 pg/ml (natural log-transformed 4.68) ( Figure 2 ). However, the negative association diminished in cycles with greater E₂ decline. The OR for the negative association was 0.862(95%CI: 0.765-0.972). When patients were stratified according to COS protocols, the pattern of association between E₂ decline and CLBR was similar between the agonist and antagonist protocols ( Supplementary Figure S3 ). However, the dose-response lacked statistical significance in patients with the antagonist protocol (p=0.38). We further visualized the difference between the dose-response curves of agonist and antagonist protocols in the GAM model. It shows that the CLBR was significantly lower in the antagonist protocol than the agonist protocol when the E₂ decline is higher than 4.28 pg/ml (log transformed 1.455), and the difference further extends when the degree of E2 decline further increases. Considering a much lower CLBR in the antagonist protocol than that in the agonist protocol (37.3% versus 71.1%), the lack of association in the antagonist protocol would be due to a lack of power.

Table 3 demonstrates the characteristics of the E₂ decline during monitoring. Most of the E₂ decline occurred between the 8th and 13^th day of stimulation, with a median E₂ decline of 244 pg/ml. The E₂ decline only had a limited association with follicle count and mean diameter at the visit. As shown in a Pearson correlation matrix ( Supplementary Figure S4 ), the degree of E₂ decline did not correlate with follicle diameter changes in comparison with the previous visit, and only a low correlation with follicle count changes. A modest correlation (r=0.181) between the degree of E₂ decline and the Gn dosage change in the previous visit was also observed. However, only a small proportion of patients with E₂ decline (5.1%, n=146) were associated with a Gn dosage decrease in the previous visit. On the other hand, 18.8% (n=538) of the patients underwent Gn dosage increase following the observation of decline. In addition, about 11.4% of the patients underwent consecutive E₂ decline at two visits. Since the descriptive data suggested that there was considerable heterogeneity in patients with E₂ decline, we further investigated the effects of heterogeneity on cumulative outcomes.

Because the Venn diagram ( Supplementary Figure S5 ) suggests that there was limited overlapping between patients who underwent Gn dosage decrease before E₂ decline, patients with Gn dosage increase following E₂ decline, and patients with consecutive E₂ decline in two consecutive visits, we analyzed them independently in multivariate models. In multivariate glm models, we compared the aforementioned E₂ decline subgroups (patients with and without Gn increase following E₂ decrease, patients with and without consecutive decline at two visits, and patients with and without Gn decrease before E₂ decline) with patients without E₂ decline ( Supplementary Figure S6 ). The size of association (ORs) did not significantly differ between the patients with and without Gn increase following E₂ decrease or patients with and without Gn decrease before E₂ decline ( Supplementary Figure S6 ). However, the OR in patients with two consecutive E₂ declines (0.58, 95% CI: 0.45,0.74) was significantly lower than that in patients without (OR 0.87, 95%CI: 0.79,0.96).

To confirm the effect of heterogeneity, we also carried out PS-matching in the subgroups (detailed in Supplementary Table S1 - 6 , Supplementary Figures S7 - 9 ), and the summarized results are shown in Table 4 . Increasing the dosage of Gn following E₂ decline appeared to improve cumulative live birth, but the difference diminished following matching or multivariate analyses. On the other hand, the patients with consecutive E₂ decline in two visits tended to have lower cumulative live birth rates than patients without in both the unmatched and matched cohorts. Although the P values became marginal in the matched cohort due to reduced sample size, the size of the association was robust in both the unmatched and matched cohorts, with or without multivariate adjustment.

E-values of these associations ( Supplementary Table S8 ) suggested that the minimal strength of the association that an unknown confounding needs to explain away the association between E₂ decline and CLBR was 1.43(95%CI: 1.27,1.56), which was greater than that of the well-known predictor, female age (1.25, 95%CI: 1.23,1.27). For patients with a consecutive estradiol decline, the E-value for E₂ decline was even higher (1.95, 95%CI: 1.6,2.35).

In our study, we observed a decline in E₂ levels in 2,863 out of 27,487 patients undergoing COS, corresponding to an incidence of 10.63%. This incidence indicates that E₂ decline is a moderate occurrence in the context of COS, highlighting the need for further investigation into its clinical significance. However, the data also shows a modest association between the E₂ decline and decreased CLBR following complete cycles. Compared with propensity score-matched controls, a 4% difference in CLBR was observed. Therefore, E₂ decline may be a potential detriment in COS treatment as a part of ART.

Historically, the impact of E₂ decline on the ART outcomes is debatable and the majority of the studies focused on the fresh transfer cycles. Some studies reported a dramatic decrease in pregnancy rates following E₂ decline. For instance, Kulshrestha, et al. reported a 14% pregnancy rate in 23 patients undergoing pre-triggering E₂ decline (6). In a cohort of 78 patients with spontaneous E₂ decline before hCG administration, Fisher et al. found that clinical pregnancy rates were significantly reduced (13% vs. 39%, p = 0.012) (11). In contrast, Styer et al (7) reported no significant differences in live birth or pregnancy loss rates between cycles with E₂ decline and controls in GnRH agonist-downregulated protocols (n=65). Other studies demonstrated a modest decrease in pregnancy and live births following E₂ decline (12, 25). This discrepancy may stem from variations in study populations or protocols. Many of the studies have only limited sample sizes (dozens of patients per group), which could be easily affected by random variation and thus give opposite conclusions (26). In addition, important determinators of embryo transfer outcomes, such as the number of embryos transferred and the stage of the transferred embryo, are adjusted in none of the above-mentioned studies. Our study contributes to the ongoing debate by adding evidence in a patient cohort with a large sample size and confounders adjusted, showing a neutral effect of E₂ decline on fresh transfer.

Despite conflicting conclusions on the outcomes following the transfer, the majority of the previous studies agree that E₂ decline during COS may decrease embryo yield for subsequent treatment (11–13). It also raises concerns regarding embryo developmental competence. In a donor-recipient cohort, Cobo et al’s study indicates that a fall of ≥30% in serum E2 concentration during ovarian stimulation in donors negatively affects pregnancy rates and embryo quality in recipients (27). The authors warn against a decreased embryo quality following E₂ decline and suggest considering a cycle cancellation. Corroborating these studies, our study demonstrates that E₂ decline significantly reduced the total number of oocytes, embryos, and high-quality embryos, mediating a decrease in CLBR following complete cycles. However, considering the limited effect size, we do not recommend cycle cancellation due to the E₂ decline.

The degree of E₂ decline is another potential concern. Previous studies may use arbitrary thresholds such as 10% or 30% to define E₂ decline (7, 11–13, 25, 28). However, since E₂ is secreted in granulosa cells, the magnitude of both absolute and relative E₂ decline is largely dependent on the size of the patients’ growing follicle cohort. Excessive E₂ decline could only be expected in hyperresponders. On the other hand, the marginal benefit of CLBR per embryo number increase would also be narrowed in hyperresponders ( Supplementary Figure S10 ). Considering the decrease in CLBR was primarily mediated by embryo yield, it may explain the U-shaped curve observed in GAM analyses. Nevertheless, our dose-response curves still suggested that patients with a suboptimal response (eg. embryo <=4) are more vulnerable to the E₂ decline, and an E₂ decrease of around 100 pg/ml might be a threshold of concern.

We also explored the heterogeneity of patients encountering E₂ decline. Several previous studies have distinguished between spontaneous E₂ decline and intended E₂ reductions via Gn dose adjustment (7, 13), showing that suboptimal outcomes are only found in patients with spontaneous E₂ decline. Echoing the studies, we also found patients having a Gn adjustment prior to E₂ decline tended to have an optimal outcome, though the small number of such patients hampered drawing a firm conclusion. On the other hand, we noted that a significant part of the patients underwent consecutive E₂ decline in two visits, which led to a poorer prognosis. Although the mechanism is not known, the decline may reflect compromised granulosa cell function or accelerated follicular atresia in follicles (27). While COS recruits a cohort of antral follicles by administration of exogenous Gn, follicles in the cohort may have different degrees of atresia which is probably due to unsynchronized growth (29). The follicles destinated to atresia before triggering produce E₂ in a gradually reduced production rate but resulted in no oocyte. The consecutive decline would suggest a high degree of atresia in the cohort and impaired follicular dynamics.

When encountering a decline in E₂ levels during COS, clinicians may face management decisions, such as maintaining the scheduled dose of Gn or Gn dosage adjustment during COS. Increasing Gn dosage during COS is a commonly used strategy to maximize the ovarian response and oocyte yield (30, 31). However, its usefulness in patients with E₂ decline is less clear. Styer et al. reported no significant differences in live birth rates in patients with E₂ decline, regardless of whether the dose of gonadotropins was adjusted (7), suggesting increasing the Gn dosage following the occurrence of E₂ decline may not be effective management. Our subgroup analyses also support the point in a larger, multivariate-adjusted cohort. Nevertheless, the evidence so far is based on retrospective studies, and further study is warranted to investigate to optimize the management of these patients.

The strengths of the study may include a larger sample size than previous studies (7, 11–13, 25, 28), with multivariate adjustment and reporting of CLBR following the complete cycle. By focusing on CLBR, our study bridges a gap in existing literature, statistically linking repeated E₂ declines to diminished reproductive success.

The limitations of this study lie in its retrospective research design, which inherently carries the risk of biases related to the selection and evaluation of clinical cases. Although we employed multivariate analysis to adjust for confounding variables, there is still a possibility that unknown or unmeasured factors may have influenced the research results. For instance, although the culture system, equipment, and staff remained stable during the study period, unmeasured variations in the laboratory environment, such as fluctuation in air quality, may contribute to confounding. Nevertheless, the E-values suggest that the minimal strength of the association that an unknown confounding needs to explain away the association between E₂ decline and CLBR was greater than the well-known predictor, female age. These limitations emphasize the need for caution when interpreting the research findings, and also highlight the importance of using prospective research methods in the future to validate our findings and further elucidate the relationship between E₂ levels and reproductive outcomes during the COS process.

In addition, due to the study being observational, the study cohort was also involved in interventions such as Gn dose adjustment during COS. Such interventions are based on the clinicians` decisions on individual cases and inevitably introduce bias. Although we also carried out subgroup analyses, they may be underpowered and heterogeneous.

Finally, the reasons for the occurrence of continuous decreases in E₂ are still unclear. It is possible that patients who underwent E₂ decline for different reasons are misclassified in our study and previous ones, which resulted in a skewed conclusion.

Although E₂ is frequently monitored during COS (32), the value of routine E₂ monitoring during COS has already been questioned (33). Our data suggest that the decline in E₂ during COS monitoring is associated with the CLBR following a complete cycle, indicating it remains a critical biomarker in predicting the outcomes during COS. However, the overall size of the association is modest, with an E-value (1.43, 95%CI: 1.27,1.56) comparable to that of female age. Further attention should be paid to specific subgroups of patients, such as patients with consecutive E₂ decline.