Non-Linear Association Between Serum Estradiol Decline and Clinical Outcomes in IVF/ICSI Cycles: A Retrospective Cohort Study

The assisted reproductive technology (ART) has been applied for decades, and the controlled ovarian hyperstimulation (COH) protocols have been improved significantly to obtain better clinical outcomes from in vitro fertilization (IVF) or intra-cytoplasmic sperm injection (ICSI). 1 The GnRH agonist long protocol remains the most widely used regimen. Unfortunately, these protocols often cause ovarian hyperstimulation (OHSS) in susceptible patients. In contrast to the established GnRH agonist protocol, the GnRH antagonist protocol has recently been proven to reduce OHSS rates while achieving comparable live birth rates. 2 , 3 Also, micro-stimulus protocol has been chosen by women with fertility impediments. Even though with modified protocols and advanced techniques, the success rates of IVF/ICSI are still not satisfying. Numerous parameters affecting IVF/ICSI outcomes, including the total number of MII oocytes retrieved, endometrial receptivity, quantity and quality of embryos and so on. 4–6 Some risk factors related to the failure of IVF/ICSI include advanced couple’s age, increased basic follicle-stimulating hormone (FSH) levels, decreased number of antral follicle and low estradiol (E2) levels during COH. 3 , 6 Besides, abnormal or dysplastic oocyte development often results in poor embryonic quality and unsatisfactory clinical outcomes. 7–9 Out of these factors, serum estradiol levels and follicle size are the necessary indicator for monitoring during COH procedure of IVF/ICSI cycles. 10 , 11 These two parameters were used to evaluate the ovarian response and the oocytes’ quality and maturation. A rational evaluation of ovarian reserve capacity and the ovary’s responsiveness to gonadotropins, followed by the selection of an appropriate COH protocol, is a critical factor determining the outcome of IVF/ICSI. The serum E2 level is considered to be one of the most important predictors for the success of IVF/ICSI. However, the prognostic value of change in serum E2 levels at different time points, particularly on or after the day of human chorionic gonadotropin (hCG) administration, remains inconsistent across studies. 12–14 Previous investigations have revealed a positive association between elevated E2 production and ovarian response to the stimulation, including maturation of oocytes and number of MII oocytes. 15 Meanwhile, various reports revealed that low serum E2 levels alone on the day of hCG administration predicted poor IVF/ICSI outcomes. 13 , 14 Kondapalli et al also found that a decline in serum E2 levels after hCG administration was associated with reduced clinical pregnancy and live birth rates. 16 However, other studies have assessed that there is no association between post-hCG E2 changes and live birth rates in either long or short GnRH agonist protocols. 14 These conflicting findings highlight the uncertainty surrounding the clinical significance of peri-trigger E2 dynamics. To date, prior research has predominantly focused on E2 levels at the time of hCG administration or on E2 changes after hCG triggering. A critical knowledge gap remains regarding the occurrence and clinical impact of transient E2 decline before hCG administration, the phenomenon of E2 decline before hCG administration remains poorly characterized. Only one recent study has specifically addressed pre-hCG E2 decline, 14 and its generalizability is limited by its restriction to patients without oral contraceptive pretreatment. Consequently, the prognostic significance of pre-hCG E2 dynamics across different COH protocols remains poorly characterized. Furthermore, understanding whether the relationship between E2 decline magnitude and adverse outcomes is linear or non-linear is clinically critical, as a threshold effect would identify a “safe range” of E2 fluctuation and help clinicians distinguish between physiological variation and pathological decline. Addressing this non-linearity could refine risk stratification and inform clinical decision-making. Accordingly, the present retrospective cohort study was designed to investigate the association between pre-hCG E2 decline and IVF/ICSI outcomes in patients undergoing long agonist, antagonist, and mild stimulation protocols, with a specific focus on characterizing the potential non-linear dose-response relationship.

In this retrospective cohort study, we analyzed clinical data from 1113 patients who received IVF/ICSI non-donor treatment with long, antagonist and micro stimulus protocols at the Reproductive Center in the First Affiliated Hospital of Anhui Medical University between January 2016 and December 2017. Patients’ characteristics and parameters of IVF/ICSI cycles, including COH protocol, Gn days, micromanipulation and other parameters, were obtained from the Reproductive Center in the First Affiliated Hospital of Anhui Medical University database. Hormone values originated from databases of the Endocrine Laboratories of the Reproductive Center in the First Affiliated Hospital of Anhui Medical University. All procedures followed the ethical standards of the Declaration of Helsinki and were approved by the Institutional Ethics Committee of the First Affiliated Hospital of Anhui Medical University (PJ20250788). A power calculation was performed using the formula for sample size estimation in logistic regression. \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}$$n = \left({Z\left({1 - \alpha /2} \right) + Z\left({1 - \beta } \right)} \right) / \left[{P\_exp \times \left({1 - P\_exp} \right) \times P\_event \times \left({1 - P\_event} \right) \times \left({ln OR} \right)} \right]$$\end{document} With a total of 1113 participants, an E2 decline prevalence of 19.6%, and observed clinical pregnancy and live birth rates of 66.9% and 60.8%, respectively, the study achieved >80% power to detect an odds ratio of 0.62 for the association between E2 decline and clinical pregnancy, and >90% power to detect an odds ratio of 0.49 for live birth, at a two-sided α of 0.05. The events-per-variable (EPV) ratios were 33.5 and 39.6 for pregnancy and live birth models, respectively, well above the recommended threshold of 10, indicating no risk of overfitting. Inclusion criteria for all enrolled cases were as follows: (1) First cycle of IVF-ET; (2) Given ultrasound measurement variability and potential misclassification of tiny antral follicles, only AFCs ≥5 mm were enrolled according to the inclusion criteria; (3) Complete clinical data availability. Exclusion criteria of infertility included: (1) Not the people who do the IVF with those three plans; (2) Those with familial genetic diseases; (3) Both spouses have chromosomal abnormalities; (4) People with incomplete clinical data. Patients with the long-term protocol received subcutaneous injection of triptorelin acetate (diphereline, Ipsen, France) at a dose of 0.9–1.2 mg starting in the mid-luteal phase of the preceding menstrual cycle, and began to be administered with recombinant follicle-stimulating hormone (rFSH) for ovarian stimulation at a dose of 150–225IU (Gonal-F, Merck Serono, Switzerland) or 100–200 IU (Puregon, N. V. Organon, Netherlands) after 14 days. Patients undergoing the antagonist protocol commenced gonadotropin administration on menstrual cycle day 2 or 3. The exogenous gonadotrophins were given daily to patients. They included rFSH at a dose of 150–225 IU (Gonal-F, Merck Serono, Switzerland) or 100–200 IU (Puregon, N.V. Organon, Netherlands). When the main ovarian follicles’ diameters were ≥12mm, antagonists at a dose of 0.25 mg (Orgalutran, N.V. Organon, Netherlands or Cetrotide, Pierre Fabre, France) were given. Patients undergoing the minimal stimulation protocol received daily oral clomiphene citrate (50–100 mg) combined with human menopausal gonadotropin (hMG; Menopur ® , 150–225 IU) starting on cycle day 2 or 3, continuing until hCG trigger. Final oocyte maturation was triggered with 10,000 IU hCG when at least one leading follicle reached ≥19 mm in mean diameter. Transvaginal ultrasound-guided oocyte retrieval was performed 34–38 hours post-trigger. Depending on the patients’ condition, embryos were freshly transferred or frozen. All transferred embryos were cultured to the blastocyst stage (day 5 or 6 post-fertilization). Blood samples were obtained at 7–8 a.m. and analyzed by radioimmunoassay. E2, progesterone (P) and luteinizing hormone (LH) levels were measured before and on the morning of hCG administration. Serum hormone measurement was performed with a chemiluminescence immunoassay kit (UniCel DxI 800, Beckman Coulter, USA). The variations <10%. During COH, serum E2 levels were serially monitored, and the transient decrease observed between consecutive measurements – despite the overall upward trend was defined as the “decline of E2”. During COH, serum E2 levels were expected to show an overall upward trend reflecting progressive follicular development. An “E2 decline” was operationally defined as any decrease in serum E2 concentration between two consecutive measurements, despite the overall rising trend. The magnitude of decline was calculated using both relative and absolute measures: Relative decline (%) = [(E2_previous − E2_current)/E2_previous] × 100%, used for the binary classification (decline vs no decline) and the quartile-based sensitivity analysis. Absolute decline (pg/mL) = E2_previous − E2_current, used as the continuous variable in the restricted cubic spline (RCS) dose-response analysis. For cycles in which more than one transient E2 decline was observed across consecutive measurements, the largest magnitude decline (maximum absolute reduction in pg/mL) was used as the index event for all analyses, to capture the most clinically significant perturbation. Patients with no observed decline at any consecutive measurement pair were classified as the “no decline” reference group. The primary outcomes were clinical pregnancy and live birth. The clinical pregnancy rate was identified with the presence of intrauterine gestational sac with detectable fetal heartbeat at 7 weeks confirmed by transvaginal sonography. The live birth represented the delivery of one or more viable infants (demonstrating spontaneous respiration or heartbeat) beyond the threshold of viability. Covariates were selected based on clinical relevance and prior literature. They include: female age, Body mass index (BMI), duration of infertility, type of infertility (primary vs secondary), endometrial thickness on hCG day, number of high-quality blastocysts, uterine factors, tubal factors, IVF/ICSI procedure, COH protocol, and E2 level on hCG day. Embryo quality is the single strongest predictor of implantation and live birth in IVF/ICSI. Uterine abnormalities (eg, fibroids distorting the cavity, adenomyosis, endometrial polyps, congenital uterine anomalies) are known to impair implantation and increase miscarriage risk. Primary and secondary infertility differ in underlying etiology, prior reproductive history, and prognosis. Secondary infertility may indicate prior proven fertility, which is associated with better IVF outcomes. Statistical analyses were conducted using the SPSS software package (Version 20.0; SPSS Inc., Chicago, USA). Normally distributed data were presented as mean ± standard deviation (SD), while non-normally distributed data were expressed as median (interquartile range, IQR). Categorical data were reported as frequency (percentage), as appropriate. Patients were stratified into four groups based on the magnitude of estradiol decline: 75 th percentile. A multivariate logistic regression model was employed to adjust for potential confounders and evaluate the association between pre-hCG E2 dynamics and clinical pregnancy and live birth outcomes. Restricted cubic spline (RCS) regression was performed to model the non-linear dose-response relationship between the magnitude of serum E2 reduction and adverse reproductive outcomes. All RCS models were adjusted for female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and fertilization method (IVF/ICSI). Non-linearity was formally tested using a likelihood ratio test comparing the full model against the reduced model with the linear term only. Inflection points were identified as the points where the second derivative of the fitted spline function equaled zero. Mediation analysis was conducted to decompose the total effect of E2 decline on reproductive outcomes (clinical pregnancy and live birth) into average causal mediation effect (ACME, indirect effect transmitted through mediator) and average direct effect (ADE, direct effect independent of the mediator). Two candidate mediators were separately tested: high quality blastocysts and endometrial thickness. A two-tailed P -value <0.05 was considered statistically significant.

Table 1 summarized demographics of 1113 cycles including baseline infertility parameters, ovarian stimulation characteristics and embryo laboratory manipulation parameters. Of these embryos, 73.1% were fertilized by IVF and 26.9% by ICSI. In brief, the age of patients was 31.07 ± 4.11 years, Body mass index (BMI) was 22.88 ± 4.26 kg/m 2 and mean (IQR) duration of infertility was 3.00 (2.00, 5.00) years. In addition, their mean (SD) pregnancies and miscarriages number were 0.83 (1.15) and 0.51 (0.89), and mean (IQR) basic E2, FSH and LH were 131.80 (80.0, 182.60), 6.81 (6.01, 7.72) and 4.24 (3.11, 5.96) respectively. The mean (IQR) serum E2 production on the day of hCG administration was 8986.0 (6623.0, 11,491.0) pg/mL. Table 1 Baseline Characteristic Analysis of Participants Variable Nonpregnancy Pregnancy P Non~Live Live P BMI 23.00 (20.90, 25.15) 22.70 (20.50, 25.10) 0.261 22.88 (20.80, 25.10) 22.80 (20.59, 25.06) 0.515 Age (y) Female 32.00 (28.00, 34.50) 29.00 (27.00, 33.00) <0.001 31.00 (28.00, 34.00) 29.00 (27.00, 33.00) <0.001 Age (y) Male 33.00 (30.00, 37.00) 31.00 (28.00, 35.00) <0.001 33.00 (29.00, 37.00) 31.00 (28.00, 34.50) <0.001 Duration of infertility (y) 3.00 (2.00, 5.00) 3.00 (2.00, 5.00) 0.064 3.00 (2.00, 5.00) 3.00 (2.00, 5.00) 0.157 Type of infertility 0.337 0.337  Primary infertility 195 (52.8) 414 (55.6) 235 (53.9) 374 (55.2)  Secondary infertility 174 (47.2) 330 (44.4) 201 (46.1) 303 (44.8) Tubal factor 0.334 0.227  No 148 (40.1) 321 (43.1) 174 (39.9) 295 (43.6)  Yes 221 (59.9) 423 (56.9) 262 (60.1) 382 (56.4) Endometriosis 0.750 0.852  No 349 (94.6) 707 (95.0) 413 (94.7) 643 (95.0)  Yes 20 (5.4) 37 (5.0) 23 (5.3) 34 (5.0) PCOS 0.086 0.242  No 318 (86.2) 611 (82.1) 371 (85.1) 558 (82.4)  Yes 51 (13.8) 133 (17.9) 65 (14.9) 119 (17.6) Uterine factor 0.005 0.096  No 316 (85.6) 678 (91.1) 381 (87.4) 613 (90.5)  Yes 53 (14.4) 66 (8.9) 55 (12.6) 64 (9.5) RSA 0.822 0.898  No 350 (94.9) 708 (95.2) 414 (95.0) 644 (95.1)  Yes 19 (5.1) 36 (4.8) 22 (5.0) 33 (4.9) Male factor 0.783 0.933  No 225 (61.0) 460 (61.8) 269 (61.7) 416 (61.4)  Yes 144 (39.0) 284 (38.2) 167 (38.3) 261 (38.6) Basic E2 (pmol/L) 135.00 (81.50, 183.20) 129.77 (82.10, 182.80) 0.539 135.80 (81.00, 181.50) 129.00 (82.70, 183.95) 0.513 Basic FSH (IU/L) 6.72 (5.90, 7.69) 6.85 (6.06, 7.73) 0.245 6.81 (6.00, 7.730 6.82 (6.02, 7.71) 0.790 Basic LH (mIU/mL) 4.15 (3.03, 5.75) 4.29 (3.15, 6.07) 0.237 4.17 (3.08, 5.76) 4.31 (3.15, 6.03) 0.358 Gn days 11.00 (9.00, 13.00) 11.00 (10.00, 13.00) 0.187 11.00 (9.00, 13.00) 11.00 (10.00, 13.00) 0.235 Total consumption of FSH 2100.00 (1600.00, 2725.00) 2069.00 (1600.00, 2700.00) 0.625 2100.00 (1575.00, 2700.00) 2100.00 (1650.00, 2700.00) 0.725 IVF/ICSI 0.003 0.056  IVF 217 (58.8) 505 (67.9) 268 (61.5) 454 (67.1)  ICSI 152 (41.2) 239 (32.1) 168 (38.5) 223 (32.9) Protocol <0.001 <0.001  Long-term 169 (45.8) 403 (54.2) 197 (45.2) 375 (55.4)  Antagonist 109 (29.5) 262 (35.2) 141 (32.3) 230 (34.0)  Micro stimulus 91 (24.7) 79 (10.6) 98 (22.5) 72 (10.6) Endometrium thickness 11.00 (9.00, 12.95) 11.50 (10.00, 13.90) <0.001 11.00 (9.00, 13.00) 11.50 (10.00, 13.95) <0.001 Day of hCG E2 8149.00 (5566.18, 10,951.00) 9450.50 (7109.25, 11,680.50) <0.001 8155.50 (5694.00, 10,904.00) 9503.00 (7261.00, 11,741.50) <0.001 Decline of E2 0.001 <0.001  No 281 (76.2) 626 (84.1) 328 (75.2) 579 (85.5)  Yes 88 (23.8) 118 (15.9) 108 (24.8) 98 (14.5) Day of hCG LH 1.79 (1.02, 4.08) 1.45 (0.83, 2.72) <0.001 1.75 (0.99, 3.73) 1.43 (0.83, 2.81) 0.001 Day of hCG P 3.68 (2.50, 5.39) 3.52 (2.43, 5.03) 0.186 3.52 (2.43, 5.19) 3.56 (2.46, 5.11) 0.950 Oocytes retrieved 8.00 (5.00, 11.00) 11.00 (8.00, 14.00) <0.001 8.00 (5.00, 11.00) 11.00 (8.00, 14.00) <0.001 Good quality blastocyst 1.00 (0, 3.00) 4.00 (3.00, 6.00) <0.001 1.00 (1.00, 3.00) 4.00 (3.00, 6.00) <0.001 Note : Oocyte yield reflects the ovarian response to stimulation and the efficiency of retrieval. Abbreviations : BMI, body mass index; y, year; PCOS, polycystic ovary syndrome; RSA, recurrent spontaneous abortion; E2, estradiol; FSH, follicle-stimulating hormone; LH, luteinizing hormone; Gn, Gonadotropin; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. Baseline Characteristic Analysis of Participants Note : Oocyte yield reflects the ovarian response to stimulation and the efficiency of retrieval. Abbreviations : BMI, body mass index; y, year; PCOS, polycystic ovary syndrome; RSA, recurrent spontaneous abortion; E2, estradiol; FSH, follicle-stimulating hormone; LH, luteinizing hormone; Gn, Gonadotropin; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. The proportion of nonpregnancy patients with decreased E2 levels was significantly higher than that of patients with successful pregnancies (23.8% vs 15.9%, P = 0.001). Similarly, the proportion of live births with reduced E2 levels was lower than that of non-live births (14.5% vs 24.8%, P < 0.001). Additionally, the number of retrieved oocytes and good-quality embryos was significantly higher in women who achieved a clinical pregnancy compared to those who did not ( P < 0.001). This trend was also consistent in birth outcomes, with live births associated with a higher oocyte yield and more good-quality embryos than in non-pregnant women ( P < 0.001). Other demographics are also displayed in Table 1 . Using clinical pregnancy outcome as the dependent variable (coded as 1 for pregnant and 0 for not pregnant), we conducted a multivariate logistic regression analysis to assess whether a decline in E2 levels (coded as 1 for reduction and 0 for no reduction) independently predicted pregnancy outcomes. The model was adjusted for potential confounders, including female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. The analysis revealed that a decline in E2 levels was an independently associated with reduced odds of clinical pregnancy (OR = 0.622, 95% CI: 0.424~0.910, P = 0.015). Conversely, greater endometrial thickness (OR = 1.085, 95% CI: 1.029~1.144, P = 0.002) and a higher number of high-quality blastocysts (OR = 1.936, 95% CI: 1.758~2.132, P < 0.001) emerged as independent positive predictors of pregnancy success. Other potential factors have no impact on the outcome of clinical pregnancy and detailed results were presented in Table 2 . Table 2 Multivariate Regression Analysis for the Association Between the Decline of Serum E2 Status and Pregnancy Outcome Variable Clinical Pregnancy OR 95% CI P Decline of E2 0.622 0.424~0.910 0.015 Age (y) Female 0.960 0.905~1.018 0.176 BMI 1.002 0.963~1.042 0.933 Duration of infertility (y) 1.035 0.978~1.095 0.234 Type of infertility Primary infertility 1 1 Secondary infertility 1.138 0.810~1.599 0.457 Endometrium thickness 1.085 1.029~1.144 0.002 Good quality blastocyst 1.936 1.758~2.132 <0.001 Uterine factor  No 1 1  Yes 0.691 0.421~1.133 0.143 Tubal factor  No 1 1  Yes 1.204 0.874~ 0.635 0.411 IVF/ICSI  ICSI 1 1  IVF 1.019 0.734~1.413 0.912 Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. Multivariate Regression Analysis for the Association Between the Decline of Serum E2 Status and Pregnancy Outcome Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. To further quantify the relationship between different levels of E2 reduction and pregnancy outcomes, we performed a sensitivity analysis restricted to patients with declining E2 levels, categorizing the magnitude of E2 reduction into quartiles to systematically assess its association with pregnancy outcomes. We adjusted for potential confounding factors using a multifactor model, such as female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. We found that a 50–75% decrease in E2 and a 75%−100% decrease were adverse influencing factors affecting pregnancy outcome (OR = 0.214, 95% CI: 0.068~0.672, P = 0.008; OR = 0.141, 95% CI: 0.043~0.463, P < 0.001). The higher the number of high-quality blastocysts, the more independent promoting factors that affect pregnancy (OR = 2.003, 95% CI: 1.589~2.524, P = 0.002). Furthermore, our sensitivity analysis revealed that maternal age represents a significant risk factor for adverse pregnancy outcomes among patients exhibiting declining E2 levels (OR = 0.854, 95% CI: 0.758~0.962, P = 0.009). The detailed results are presented in Table 3 . Multivariable logistic regression stratified by ovarian stimulation protocols was performed to assess the association between serum E2 decline and reproductive outcomes. In the long protocol subgroup, E2 decline was independently associated with reduced odds of clinical pregnancy (OR  =  0.529, 95% CI: 0.324–0.863, P   =  0.011). When E2 decline was further categorized into quartiles, increasing magnitude of E2 reduction corresponded to progressively lower odds of both endpoints in long-protocol cycles, with the 75–100% decline quartile showing the strongest negative effect (live birth: OR = 0.064, 95% CI: 0.022–0.187, P   < 0.001). For micro-stimulation cycles, although overall E2 decline showed significant negative associations with clinical pregnancy (OR = 0.201, 95% CI: 0.042–0.957, P  = 0.044), none of the quartile stratified analyses reached statistical significance ( Tables S1 and S2 ). Table 3 Multivariate Regression Analysis for the Association Between the Level of Serum E2 Reduction and Pregnancy Outcome Variable Clinical Pregnancy OR 95% CI P Decline of E2  0%~25% 1 1  25%~50% 0.525 0.172~1.601 0.257  50%~75% 0.214 0.068~0.672 0.008  75%~100% 0.141 0.043~0.463 0.001 Age (y) Female 0.854 0.758~0.962 0.009 BMI 1.018 0.918~1.129 0.732 Duration of infertility (y) 1.205 1.015~1.429 0.033 Type of infertility Primary infertility 1 1 Secondary infertility 1.452 0.624~3.382 0.387 Endometrium thickness 1.074 0.943~1.223 0.280 Good quality blastocyst 2.003 1.589~2.524 0.002 Uterine factor  No 1 1  Yes 0.298 0.064~1.392 0.124 Tubal factor No 1 1 Yes 1.392 0.625~ 3.101 0.418 IVF/ICSI ICSI 1 1 IVF 0.778 0.342~1.767 0.548 Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. Multivariate Regression Analysis for the Association Between the Level of Serum E2 Reduction and Pregnancy Outcome Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. Using live birth as the primary outcome (coded as 1 for live birth and 0 for no live birth), we performed multivariate binary logistic regression analysis with E2 reduction (coded as 1 for decrease and 0 for no decrease) as the independent variable. After adjusting for potential confounders including female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure, we found that E2 reduction was significantly associated with decreased likelihood of live birth (OR = 0.489, 95% CI: 0.339~0.707, P < 0.001). Advanced maternal age was significantly associated with reduced live birth rates in our study population (OR = 0.929, 95% CI: 0.891~0.969, P = 0.001). Both increased endometrial thickness (OR = 1.081, 95% CI: 1.028~1.137, P = 0.002) and higher numbers of high-quality blastocysts (OR = 1.751, 95% CI: 1.613~1.901, P < 0.001) emerged as independent positive predictors of live birth. Complete results are presented in Table 4 . Table 4 Multivariate Regression Analysis for the Association Between the Decline of Serum E2 Status and Live Birth Outcome Variable Live Birth OR 95% CI P Decline of E2 0.489 0.339~0.707 <0.001 Age (y) Female 0.929 0.891~0.969 0.001 BMI 1.006 0.969~1.046 0.744 Duration of infertility (y) 1.043 0.988~1.101 0.124 Type of infertility Primary infertility 1 1 Secondary infertility 1.249 0.905~1.723 0.176 Endometrium thickness 1.081 1.028~1.137 0.002 Good quality blastocyst 1.751 1.613~1.901 <0.001 Uterine factor No 1 1 Yes 0.958 0.590~1.556 0.861 Tubal factor No 1 1 Yes 0.840 0.620~ 1.138 0.260 IVF/ICSI ICSI 1 1 IVF 1.191 0.870~1.630 0.276 Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. Multivariate Regression Analysis for the Association Between the Decline of Serum E2 Status and Live Birth Outcome Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. To characterize the dose-response relationship between E2 reduction magnitude and live birth outcomes, we conducted a quartile-based sensitivity analysis in the subgroup of patients exhibiting E2 decline. After stratifying E2 reduction levels into quartiles and adjusting for all predefined covariates in our multivariate model, we identified a pronounced negative association between severe E2 reduction (75–100% decrease) and live birth success (OR = 0.158, 95% CI: 0.050~0.494, P = 0.001). This relationship demonstrated progressively stronger negative effects with increasing E2 reduction severity. Advanced maternal age demonstrated a significant inverse association with live birth rates in our cohort (OR = 0.814, 95% CI: 0.722~0.918, P = 0.001). In contrast, high-quality blastocyst number remained an independent positive predictor of live birth (OR = 1.799, 95% CI: 1.481~2.186, P < 0.001), exhibiting a robust association. The complete stratified analysis results are presented in Table 5 . Table 5 Multivariate Regression Analysis for the Association Between the Decline of Serum E2 and Live Birth Outcome Variable Live Birth OR 95% CI P Decline of E2  0%~25% 1 1  25%~50% 0.604 0.225~1.624 0.318  50%~75% 0.448 0.162~1.244 0.123  75%~100% 0.158 0.050~0.494 0.001 Age (y) Female 0.814 0.722~0.918 0.001 BMI 0.965 0.868~1.073 0.512 Duration of infertility (y) 1.215 1.026~1.439 0.024 Type of infertility Primary infertility 1 1 Secondary infertility 1.630 0.722~3.679 0.239 Endometrium thickness 1.085 0.957~1.230 0.203 Good quality blastocyst 1.799 1.481~2.186 < 0.001 Uterine factor No 1 1 Yes 0.213 0.034~1.322 0.097 Tubal factor No 1 1 Yes 1.132 0.523~ 2.450 0.753 IVF/ICSI ICSI 1 1 IVF 1.430 0.626~3.270 0.396 Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. Multivariate Regression Analysis for the Association Between the Decline of Serum E2 and Live Birth Outcome Note : Model has been adjusted female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. Abbreviations : BMI, body mass index; y, year; E2, estradiol; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection. Restricted cubic splines (RCSs) were used to analyze the Dose‒response relationship between the level of serum E2 reduction concentrations with pregnancy outcome and live birth outcome. Based on this relationship, a nonlinear model provided a better explanation of the relationship than a linear model after adjusting for female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. The pregnancy outcome and live birth outcome gradually decreased on the left side of the inflection point (2014.572 pg/mL), while on the right side, they marginally decreased and reached saturation ( P overall = 0.0002) ( Figure 1A ). Moreover, a similar non-linear relationship was observed between the level of serum E2 reduction concentrations (2426.718 pg/mL) and live birth outcome ( P nonline = 0.0096, Figure 1B ). In the long protocol subgroup, E2 decline was independently associated with reduced odds of live birth (OR = 0.419, 95% CI: 0.262–0.669, P  < 0.001). When E2 decline was further categorized into quartiles, increasing magnitude of E2 reduction corresponded to progressively lower odds of both endpoints in long protocol cycles, with the 75–100% decline quartile showing the strongest negative effect (live birth: OR = 0.064, 95% CI: 0.022–0.187, P  < 0.001) For micro-stimulation cycles, although overall E2 decline showed significant negative associations with live birth (OR = 0.097, 95% CI: 0.013–0.733, P  = 0.024), none of the quartile stratified analyses reached statistical significance ( Tables S1 and S2 ). Figure 1 Association of the level of serum E2 reduction with ( A ) pregnancy outcome and ( B ) live birth outcome among participants. Odds ratios (ORs) were adjusted for female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. The solid line represents ORs, and the shaded areas represent 95% confidence intervals. Two line graphs of odds ratio versus serum E2 reduction, showing a steep early drop then flattening. Association of the level of serum E2 reduction with ( A ) pregnancy outcome and ( B ) live birth outcome among participants. Odds ratios (ORs) were adjusted for female age, BMI, duration of infertility, type of infertility, endometrial thickness, number of high-quality blastocysts, uterine factors, tubal factors, and IVF/ICSI procedure. The solid line represents ORs, and the shaded areas represent 95% confidence intervals. Mediation analysis was performed to explore whether high quality blastocysts and endometrial thickness mediated the association between E2 decline and clinical pregnancy as well as live birth outcomes. For clinical pregnancy: The indirect effect mediated by high quality blastocysts was statistically significant (βACME = −0.09, βADE = 0.006, βtotal effect = −0.08), with the mediated proportion reaching 1.07% ( P < 0.001) ( Figure 2A ). By contrast, endometrial thickness did not exert a significant mediating effect on the E2 decline pregnancy pathway (βACME = −0.004, β ADE = −0.08, βtotal = −0.08, mediated proportion = 0.04%, P = 0.316) ( Figure 2B ). For live birth outcomes: A significant mediating effect of high‑quality blastocysts was observed (βACME = −0.10, βADE = 0.03, βtotal = −0.07, mediated proportion = 1.47%, P = 0.012) ( Figure 2C ). No statistically significant mediation was detected for endometrial thickness (βACME = −0.004, βADE = −0.006, βtotal = −0.06, mediated proportion = 0.05%, P = 0.340) ( Figure 2D ). Collectively, these findings indicated that the adverse impact of E2 decline on pregnancy and live birth rates was partially mediated through reduced formation of high quality blastocysts, while endometrial thickness was not a meaningful mediator in this causal pathway. Figure 2 Mediation analysis for the pathways from E2 decline to clinical pregnancy and live birth outcomes. Mediation pathways via high quality blastocysts ( A and C ) and endometrial thickness ( B and D ). βACME, average causal mediation effect (indirect effect); βADE, average direct effect; βtotal, total effect; Proportion, percentage of total effect explained by the mediator. A shows mediation analysis of E2 decline on pregnancy via high-quality blastocysts. betaACME equals minus 0.09, betaADE equals 0.006, betatotal equals minus 0.08, proportion equals 1.07 percent, P for proportion less than 0.001. B shows mediation analysis of E2 decline on pregnancy via endometrial thickness. betaACME equals minus 0.004, betaADE equals minus 0.08, betatotal equals minus 0.08, proportion equals 0.04 percent, P for proportion equals 0.316. C shows mediation analysis of E2 decline on live birth via high-quality blastocysts. betaACME equals minus 0.10, betaADE equals 0.03, betatotal equals minus 0.07, proportion equals 1.47 percent, P for proportion equals 0.012. D shows mediation analysis of E2 decline on live birth via endometrial thickness. betaACME equals minus 0.004, betaADE equals minus 0.006, betatotal equals minus 0.06, proportion equals 0.05 percent, P for proportion equals 0.340. Diagrams analyze E2 decline′s impact on pregnancy and birth through blastocysts and endometrial thickness. Mediation analysis for the pathways from E2 decline to clinical pregnancy and live birth outcomes. Mediation pathways via high quality blastocysts ( A and C ) and endometrial thickness ( B and D ). βACME, average causal mediation effect (indirect effect); βADE, average direct effect; βtotal, total effect; Proportion, percentage of total effect explained by the mediator.

Several researches have examined the relationship between E2 production at different time points during the COH and the success of IVF/ICSI. They included the absolute value of E2 on Day 5, Day 6, the day of hCG administration, and the change of E2 production on the day after hCG administration. 17 , 18 However, some patients may demonstrate transient decreases between consecutive measurements. Few studies have directly addressed the effect of the decrease in E2 production before the day of hCG administration. Thus, our study attempted to find if E2 decline affects the clinical outcomes of IVF/ICSI cycles. The hCG triggering timing is commonly decided by clinicians on the basis of the optimal leading follicle diameter and E2 level. In our study, the mean leading follicle diameter was 19 mm and E2 production per oocyte was 800–1500 pmol/L. Normally, the E2 production is a sign of oocytes’ maturation, and when it starts to decrease, it is considered that the oocytes’ quality starts to decline. 19 Meanwhile, poor oocyte quality often results in unsatisfactory clinical outcomes. 20 , 21 One study from Ireland showed that terminal E2 trends were directly correlated to follicular maturity and success of IVF. 22 And the serum E2 per oocyte ratio could be used as a predictor of reproductive outcome, such as clinical pregnancy rate. 23 This result was challenged by Zavy et al. They found that the serum E2 level on the triggering day did not predict the clinical outcome in IVF with either GnRH antagonist or long GnRH agonist protocols. 24 Another study also showed that high E2 level on the day of hCG administration had no detrimental effect on the implantation rate, clinical pregnancy rate, and live birth rate. 25 Our findings demonstrated a significant association between serum E2 levels and clinical outcomes, with both clinical pregnancy and live birth declining as serum E2 levels decreased prior to the hCG trigger day. E2 production serves not only as an indicator of oocyte maturation but also exhibits a positive correlation with oocyte quality, which may indirectly influence treatment outcomes. This relationship may provide a plausible explanation for our observed results. The key strength of our study is that the E2 rising and decreasing groups had comparable E2 levels on the day of hCG administration. This is particularly important because previous studies investigating the absolute value of E2 on the trigger day have reported inconsistent findings, making it difficult to isolate its independent effect. By ensuring similar baseline E2 levels between groups, our study design helps clarify the impact of E2 trends rather than absolute concentrations on clinical outcomes. 25 In predicting the reproductive outcome, the importance lies not in the E2 level on the day of hCG administration or on the day after, but rather in the E2 ratio. 26 However, Kyrou’s study showed that E2 concentrations on the day of hCG administration in GnRH antagonist cycles are not associated with pregnancy rates. 19 To prevent an interference of our results by the absolute E2 Level on the triggering day, we considered it as confounder in the multivariate regression analysis. The results displayed that the E2 level on the day of hCG administration had a significant impact on the clinical outcome of patients. Moreover, our analysis revealed a significant dose-dependent association between the magnitude of E2 reduction and adverse reproductive outcomes. Notably, when E2 reduction levels remained below the identified threshold (2014.572 pg/mL), the risk of poor prognosis demonstrated a progressive decline. This finding suggests that controlled E2 reduction within this range may represent the identified inflection point suggests a potential threshold effect that warrants prospective validation, and should not be interpreted as a clinically actionable safe range at this time. Existing evidence suggests that ovarian responsiveness and stimulation regimens may exert mediating effects on gonadal hormone dynamics, oocyte yields, embryo quality, and ultimately pregnancy and live birth. 27 Different gonadotropin preparations and adjuvant oral agents can modify follicular recruitment patterns and perifollicular steroid metabolism, which may further interact with serum estradiol fluctuations throughout the stimulation cycle. 28 Moreover, patients with distinct ovarian response profiles may exhibit divergent estradiol decline magnitudes and different susceptibility to its adverse reproductive consequences. 29 Meanwhile, progestin primed ovarian stimulation strategies also bring variable hormonal milieu and oocyte developmental potential across heterogeneous responder subgroups, implying that stimulation protocols may confound or mediate the association between estradiol drop and ART outcomes. Therefore, omission of ovarian responsiveness and ovulation inducing drug types constitutes one limitation of our current study, which may constrain causal interpretation of the observed relationships. Future work should integrate these potential mediating variables into stratified or mediation analysis models to clarify whether estradiol decline independently impairs reproductive performance or partly reflects individual variation in ovarian response and stimulation regimens, drawing insights from available relevant literature. The limitation of this study lies in its single-center design, which may introduce selection bias. Additional limitations are as follows. First, residual confounding may exist due to unmeasured covariates, such as progesterone concentration, total gonadotropin dosage, and patient psychological status. Second, as with all observational retrospective studies, causal conclusions cannot be drawn solely from the observed associations. Third, overadjustment bias may occur when conditioning on downstream intermediate variables (ie, high quality blastocysts). Fourth, the quantification of E2 decline was subject to uncertainty, given variable blood sampling intervals across patients and difficulties in handling multiple transient E 2 drops. Finally, our dataset was confined to cycles performed between 2016 and 2017. Future studies should incorporate larger, multi-center cohorts to validate our findings and enhance generalizability. In summary, our analysis on the clinical outcome of 1113 patients showed that the E2 decreased during the day of hCG administration had significant correlation with clinical outcomes, and the clinical pregnancy and live birth decreased with the decreasing level of serum E2 before hCG trigger day. Also, the patients with a greater decline in the E2 level more likely to have poor clinical outcomes, and decreased serum E2 levels exhibited a dose-response relationship with both pregnancy and live birth outcomes. Based on our findings, we offered clinicians an additional parameter to counsel patients about the likelihood of success and predicted patients’ clinical outcomes in a sense.