Abstract
Background
In clinical practice, scheduling oocyte retrieval is a challenging issue that requires comprehensive consideration of factors such as the woman's age, ovarian response, hormone levels and other variables. Moreover, there is currently no consensus on how to effectively consider these factors and their weights in order to optimize the scheduling of oocyte retrieval for obtaining more mature oocytes.
Objective
To effectively identify the key determinants of oocyte retrieval time through an extensive analysis of retrospective clinical data, and to develop a valuable predictive model for optimizing the timing of oocyte retrieval in assisted reproductive technology (ART).
Study design
This retrospective study included 49,961 oocyte pickup (OPU) cycles, as well as 5567 subsequent ET cycles and 45,198 FET cycles between January 2010 and August 2024. Multiple linear regression (MLR) and directed acyclic graphs (DAG) were employed to identify the key determinants associated with oocyte retrieval time. Oocyte pickup (OPU) cycles achieving a minimum of 70% oocyte retrieval rate and 80% oocyte maturation rate, indicating well-organized timing of oocyte retrieval, were assigned to Group 1, while the remaining cycles were assigned to Group 2. The data from Group 1 was randomly divided into training and validation sets using the "sample" function in R software. The training set data was utilized to develop a predictive model for oocyte retrieval time based on former identified determinants using the "lm" function in R software. Subsequently, the performance of this model was evaluated and visualised using the "performance" and "plot" function, and further validated with the validation set from Group 1 as well as the data from Group 2.
Results
Female age, AFC, COH protocol, number of follicles > 14 mm in diameter on the day of trigger, and hormone levels on the day of trigger (including E2, P, and LH) were key determinants for the timing of oocyte retrieval. A valuable predictive formula for determining the optimal timing of oocyte retrieval has been formulated and validated: 37.43–0.02219*Female age + 0.01383*AFC + 0.00006* E2 level on the trigger day-0.00939*P level on the trigger day-0.05194*LH level on the trigger day + 0.01497*Number of follicles > 14 mm in diameter on the trigger day + β (β = 0.0000 in Short agonist protocol, β = -0.3320 in PPOS protocol, β = 0.8361 in GnRH antagonist protocol, β = -1.2280 in Mild stimulation protocol, β = 0.4160 in Long agonist protocol).
Conclusion
The female age, antral follicle count (AFC), controlled ovarian hyperstimulation (COH) protocol, number of follicles measuring > 14 mm in diameter on the trigger day, as well as hormone levels including E2, P, and LH on the trigger day are crucial factors influencing oocyte retrieval time. A robust predictive model for oocyte retrieval time was successfully developed from these factors and validated within a well-organized timing group for oocyte retrieval (oocyte retrieval rate ≥ 70% and oocyte maturation rate ≥ 80%).
Keywords: The timing of oocyte retrieval, Key determinants, Oocyte pickup (OPU), Oocyte retrieval rate, Mature oocyte rate, Assisted reproductive technology (ART)
Introduction
In controlled ovarian hyperstimulation (COH) practice, the Luteinizing Hormone (LH) surge is typically induced by hCG, a GnRH agonist, or a combination of both to trigger the final stage of oocyte maturation. Oocyte pickup (OPU) is usually performed 32–36 h after triggering in most IVF cycles, based on studies conducted on patients who underwent ovulation trigger using Clomiphene Citrate (CC) and/or human menopausal gonadotropin (hMG) [1, 2]. This time interval from trigger to OPU plays a crucial role as it sets off significant biological processes following the LH surge including decoupling of gap junctions between granulosa cells and oocytes, resumption of meiosis, expansion of cumulus cells, and initiation of luteinization [3, 4]. Moreover, this interval has close associations with assisted reproductive technology (ART) outcomes. Several studies have reported that different intervals—either longer or shorter—can lead to varying results; however, no consensus has been reached yet. Research suggests that in individualized COH protocols retrieving oocytes too early or too late can result in lower oocyte yield and reduced rate of mature oocytes. Additionally, this optimal interval may vary across different COH protocols [5]. Other studies has indicated that a longer lag time is linked with higher yields of mature oocytes [3, 6] and increased fertilization rate [7]. Therefore, the interval between LH trigger and oocyte retrieval is an essential factor that should not be overlooked when considering ART outcomes.
The timing of oocyte retrieval is highly flexible and necessitates careful consideration of all factors. Several studies have indicated that the impact of the interval between human chorionic gonadotropin injection and oocyte retrieval varies with age [8]. Furthermore, it has been demonstrated that the number of follicles on the day of retrieval plays a crucial role in determining the optimal timing of oocyte retrieval [9, 10]. Close monitoring of serum E2 levels, particularly estrogen levels on the day of retrieval, is also essential for evaluating ovulation [11–16]. The study illustrates the ranking of variable importance and reveals that hormone levels (E2) are key influential variables in predicting ovulation timing [17]. Additionally, attention has been given to determining individualized optimal intervals between trigger and oocyte retrieval under different COH regimens. For instance, a time interval ranging from 32 to 36 h is commonly employed for patients receiving clomiphene citrate (CC) and/or human menopausal gonadotropin (hMG) [2, 18, 19]. In contrast, patients receiving gonadotropin-releasing hormone (GnRH) agonists or antagonists are most recommended to have an interval between 36 and 39 h [20–22], while for most patients using the PPOS protocol, the optimal ovulation trigger–OPU interval ranges from 36.4 to 37.8 h [23]. Moreover, a conference abstract reported distinct individualized trigger-OPU intervals based on COH regimens: prolonged regimens require a range of 35–36 h; burst regimens necessitate a range of 35–37 h; antagonist regimens call for an interval between 36 and 37 h [24].
The timing of oocyte retrieval significantly influences the rates of oocyte retrieval rates and mature oocyte [3, 5, 25, 26] and is influenced by multiple factors; however, a consensus on the optimal time interval has not yet been reached. In current clinical practice, there is a tendency to use a fixed time for oocyte retrieval or consider only one or two influencing factors, which lacks comprehensive consideration and requires improvement. Therefore, in this study, we conducted a retrospective analysis on a large cohort of patients with diverse clinical characteristics and controlled ovarian hyperstimulation (COH) regimens. We identified key influencing factors associated with oocyte retrieval time and derived a valuable reference model for determining the optimal timing of oocyte retrieval based on the high rate of oocyte retrieval and maturation in a specific patient group. This predictive model provides valuable insights into determining the optimal timing of oocyte retrieval.
Materials and methods
Patients
This retrospective study was conducted at the Department of Assisted Reproduction, Ninth People’s Hospital affiliated with Shanghai Jiao Tong University School of Medicine, between January 2010 and August 2024. The study included a total of 49,961 cycles involving ovarian stimulation and oocyte pick-up (OPU) following selection and exclusion criteria. Inclusion criteria were applied as shown in Fig. 1: patients with oocyte retrieval time greater than or equal to 31 h and less than or equal to 41 h. Exclusion criteria were as follows: (1) E2 level on the day after trigger decreased by more than 10% compared to the trigger day which could increase the likelihood of spontaneous premature ovulation and decrease oocyte retrieval rate; (2) LH level greater than or equal to10 mIU/ml on the trigger day; (3) Patients undergoing ovarian stimulation with letrozole due to its impact on estrogen levels during treatment. These comprised 2,329 cases of short agonist protocol, 34,167 cases of PPOS (progestin-primed ovarian stimulation) protocol, 6,284 cases of GnRH antagonist protocol, 4,574 cases of mild stimulation protocol, and 2,607 cases of long agonist protocol. All embryo transfer cycles resulting from these 49,961 oocyte retrieval cycles were included in the analysis, encompassing a total of 50,765 transfer cycles consisting of both fresh embryo transfers (5,567 cycles) and frozen embryo transfers (45,198 cycles).
Fig. 1.
A flow chart illustrating the study population and study design. The training and validation sets were obtained through random sampling using R software (version 4.3.2)
Ovarian stimulation and retrieval of oocytes
The five COH regimens are briefly described as follows. In the short agonist protocol, patients were administered a daily subcutaneous injection of 0.1 mg of triptorelin (Ferrin Pharmaceuticals, China), a short-acting gonadotropin-releasing agonist, starting on day 2 or 3 of the natural cycle to stimulate the ovaries. Additionally, they received a daily injection of 150–225 IU of hMG starting on day 3 of the cycle [27]. The PPOS protocols encompass various ovarian stimulation cycles, including ovarian stimulation cycles utilizing dydrogesterone (DYG) + hMG, medroxyprogesterone acetate (MPA) + hMG, micronized progesterone + hMG treatment and luteal-phase (LPS) ovarian stimulation protocol [28]. In the DYG + hMG treatment [25], patients received daily ovarian stimulation from menstrual cycle day 3 (MC3) to the trigger day with injections of hMG (150–225 IU; Anhui Fengyuan Pharmaceutical Co., Anhui, China) along with oral administration of DYG (Duphaston; 20 mg/day; Abbott Biologicals B.V., Hoofddorp, Netherlands). The MPA + hMG treatment [23] involved oral intake of MPA (10 mg/day; Shanghai Xinyi Pharmaceutical Co., Shanghai, China), while in the micronized progesterone + hMG treatment [26], patients were prescribed oral Utrogestan (100 mg/day; Laboratories Besins International, Paris, France). For LPS treatment, 225 IU of hMG was injected 1–3 days after ovulation alongside administration of letrozole (2.5 mg/day; Jiangsu Hengrui Medicine Co. Ltd., Jiangsu, China) [29].
In the GnRH-antagonist protocol, hMG at doses ranging from 150 to 225 IU per day was administered along with intravenous administration of cetrorelix (Pierre Fabre Medicament Production-Aquitaine Pharm International, France), a GnRH antagonist at doses between 0.125 and 0.25 mg per day when one follicle reached a diameter of 14 mm every day. The dosage was adjusted based on transvaginal ultrasound examination [5]. The mild ovarian stimulation protocols offered dosing flexibility and included letrozole (Jiangsu Hengrui Medicine Co., Ltd., Lianyungang, China) at a dosage of 2.5 mg, clomiphene (Codal Synto Limited, Limassol, France) at a dosage of 50 mg/day, and hMG at doses ranging from 75 to 150 IU [5, 30]. In the long agonist protocol, a long-acting gonadotropin-releasing agonist (leuprolide acetate, 3.75 mg, Lizhu Pharmaceutical Trading Co., China) was administered from days 2 to 5 of the menstrual cycle. If downregulation was quantified 35 days later, human menopausal gonadotropin (hMG) (150–225 IU/day) would be given until the trigger day [5, 31].
When at least three follicles reached a diameter of 18 mm or one dominant follicle reached 20 mm, final oocyte maturation was triggered using 5000 IU of human chorionic gonadotropin (hCG) (Lizhu Pharmaceutical Trading Co., China) for the short agonist protocol and the long agonist protocol. For the other three protocols, triptorelin stimulation (0.1–0.2 mg; decapeptyl, Ferring Pharmaceuticals, Guangdong, China) or intramuscular injections of hCG (1000–5000 IU; Lizhu Pharmaceutical Trading Co., Zhuhai, China), or cotriggered with subcutaneous triptorelin (0.1–0.2 mg) and intramuscular injections of hCG (1000–5000 IU) were applied [28, 32].
Laboratory protocols
The fertilization of the aspirated oocytes was performed in vitro, either through conventional insemination or ICSI, based on semen parameters. On day 3 of fertilization, the embryos were assessed and classified using the Cummins criteria [33] to evaluate their quality. In our center, if patients had fewer than six good-quality cleavage-stage embryos, all such embryos were cryopreserved; conversely, if there were more than six, the surplus good-quality embryos underwent blastocyst culture. Embryos of inferior quality were cultured until reaching the blastocyst stage before being cryopreserved. Subsequently, the Gardner and Schoolcraft scoring system [34] was employed to select morphologically good blastocysts (grade ≥ 4BC) for vitrification on day 5 or 6.
Embryo transfer
Fresh embryo transfer (ET) was performed on the third day post oocyte retrieval, wherein 1–2 high-quality embryos were meticulously selected for uterine cavity transfer. Luteal support commenced with oral progesterone (Duphaston) at a dosage of 40 mg/day and vaginal utrogestan at a dosage of 0.4 g/day on the same day as oocyte retrieval. The dosages were adjusted based on the pregnancy test results obtained 14 days after embryo transfer [31].
Frozen embryo transfer (FET) was conducted following previously established protocols [27, 28, 35]. Briefly, patients with regular menstrual cycles underwent natural cycle treatment, while those with irregular menstrual cycles received letrozole in combination with hMG if necessary to stimulate the development of a single follicle. Hormone replacement therapy (HRT) was recommended for patients with thin endometrium during natural cycles. Once the endometrial thickness reached ≥ 8 mm, day 3 or day 5/6 embryos were scheduled for transfer based on ovulation timing in natural and letrozole cycles or the time of progesterone administration for hormone replacement therapy. A maximum of two embryos per patient per FET cycle were transferred. Luteal support was continued until gestation reached 10 weeks once pregnancy was achieved [32]. All patients who underwent the PPOS protocol opted for the freeze-all strategy due to alterations in the endometrial environment during the oocyte retrieval cycle. [28].
Women undergoing GnRH antagonist protocol, mild stimulation protocol, short agonist protocol, or long agonist protocol are provided with the option to choose between fresh embryo transfer and frozen embryo transfer. The final decision regarding the initial attempt of frozen embryo transfer or fresh embryo transfer is collaboratively made by the patient and the clinician in clinical practice [36].
Main outcomes
The primary outcome measured in this study was the timing of oocyte retrieval. Secondary outcomes included oocyte retrieval rate and mature oocyte rate. Oocyte retrieval rate was calculated as the number of retrieved oocytes or cumulus-oocyte complexes (COCs) divided by the number of follicles measuring ≥ 14 mm on the day of retrieval. Mature oocyte rate referred to the ratio between mature oocytes and total retrieved oocytes.
To assess the efficacy of oocyte retrieval timing, we employed two indicators reflecting the extremes of oocyte retrieval outcomes within a single COH protocol. Ideally, when timing is optimal, a high number of follicles should be retrieved without complications such as premature ovulation or surgical difficulties, which may indicate either too early or too late timing. To avoid artificially inflated retrieval rates due to follicular flushing, we used the mature oocyte rate as a secondary indicator; a low mature oocyte rate would typically accompany a high retrieval rate. Since oocyte retrieval timing often requires balancing retrieval and maturity rates in clinical practice, we defined specific criteria to classify cycles into two subgroups: if both the oocyte retrieval rate and the mature oocyte rate reach 70% and 80%, respectively, during a given cycle, it is categorized as Group 1, indicating the well-organized retrieval timing. Conversely, if either the oocyte retrieval rate or the mature oocyte rate does not meet its respective threshold, it is categorized as Group 2, indicating the insufficiently planned retrieval timing.
Statistical analysis
The statistical analyses were performed using R software (version 4.3.2), IBM SPSS Statistics (version 25.0), and GraphPad Prism (version 9.5.0). We conducted multiple linear regression analysis using IBM SPSS Statistics (version 25.0) to examine various potential factors that significantly impact the timing of oocyte retrieval. Additionally, we employed DAG analysis to elucidate the causal relationships among different variables, eliminating intermediate factors and confounding variables. Consequently, we accurately identified women's age, AFC, COH protocol, number of follicles larger than 14 mm in diameter on the trigger day, as well as hormone levels including E2, P, and LH on the trigger day—all of which are crucial determinants influencing the timing of oocyte retrieval. Directed Acyclic Graph (DAG) drawn in www.dagitty.net.
The normality of continuous variables was assessed by visually examining histograms and Q-Q plots, as well as conducting the Shapiro–Wilk test. Continuous variables were presented as mean ± standard deviation (SD) and analysed using either a two independent-sample t-test or Mann–Whitney U-test, depending on appropriateness. Categorical variables were expressed as numbers (%) and compared using the chi-square test. All P-values were based on two-sided tests with statistical significance set at P < 0.05. GraphPad Prism (version 9.5.0) was utilized to generate violin plots illustrating the oocyte recovery rate and mature oocyte rate in Group 1 and Group 2, along with scatter plots depicting the distribution of oocyte retrieval time within the five protocols.
In order to develop a predictive formula for determining the optimal timing of oocyte retrieval, we utilized the data from Group 1, which represents a well-organized retrieval timing group achieving an oocyte retrieval rate of at least 70% and an oocyte maturation rate of at least 80%. The data from Group I was randomly divided into two parts using R through random sampling. Subsequently, 70% of the data was allocated as the training set to construct the prediction model, while the remaining 30% served as the validation set for validating this model (Fig. 1). The training set data were utilized to establish a predictive model for oocyte retrieval time, which underwent assessment for linearity, homogeneity of variance, and collinearity using the"plot"function followed by posterior predictive checks. The"lm"function in R software (version 4.3.2) was employed to develop this predictive model for the timing of oocyte retrieval, with"performance"function to calculate adjusted R-squared value to evaluate model performance. Standardized coefficients beta (β), along with their corresponding 95% confidence intervals (CI), were used to quantify the impact of determinants on the timing of oocyte retrieval. Subsequently, the validation set data were inputted into the model using the"predict"function, and student's t-test monitoring revealed no statistical difference between observed and predicted values for oocyte retrieval time. GraphPad Prism (version 9.5.0) was used to generate box plots to illustrate the statistical differences between the actual and predicted oocyte retrieval times in Group 2, as well as violin plots to show the distribution differences between the actual and predicted oocyte retrieval times.
Ethical approval
This retrospective study used an anonymized database to keep patients'personally identifiable information absolutely confidential. This retrospective study was conducted in accordance with the Declaration of Helsinki and was approved by the ethics committee of Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine (approval number: SH9H-2024-T134-1).
Results
The MLR and DAG analysis of key determinants associated with the timing of oocyte retrieval
A flowchart illustrating the methodology employed in this study is presented in Fig. 1. Multiple linear regression analysis was conducted to investigate the primary determinants influencing oocyte retrieval time. The significant influencers and insignificant influential factors of oocyte retrieval time are depicted in Table 1. Female age, hMG duration, P level on the trigger day and LH level on the trigger day emerged as a significant negative determinant for oocyte retrieval time. Conversely, BMI, Basal LH, AFC, hMG dose, number of follicles > 14 mm in diameter on the trigger day, and E2 level on the trigger day were found to be positive determinants (Table 1). The study revealed that, compared to the short agonist protocol (reference), the oocyte retrieval time was relatively shorter in the mild stimulation protocol, GnRH antagonist protocol and PPOS protocol, whereas it was relatively longer in the long agonist protocol. Compared to the dual trigger method, the oocyte retrieval time was significantly shorter in the GnRH agonist and hCG trigger methods.
Table 1.
The significant determinants and insignificant influencers of oocyte retrieval time were identified through Multiple Linear Regression analysis (MLA)
| Parameter | β | 95%CL | P value |
|---|---|---|---|
| Female age (year) | −0.160 | (−0.030, −0.027) | < 0.001 |
| BMI (kg/m2) | 0.022 | (0.004, 0.009) | < 0.001 |
| Infertility duration (year) | 0.002 | (−0.002, 0.003) | 0.684 |
| Basal FSH (mIU/ml) | 0.002 | (0.000, 0.000) | 0.536 |
| Basal LH (mIU/ml) | 0.032 | (0.008, 0.014) | < 0.001 |
| AFC (n) | 0.071 | (0.009, 0.013) | < 0.001 |
| hMG duration (day) | −0.050 | (−0.025, −0.013) | < 0.001 |
| hMG dose (IU) | 0.098 | (0.000, 0.000) | < 0.001 |
| Number of follicles > 14 mm in diameter on the trigger day (n) | 0.067 | (0.010, 0.014) | < 0.001 |
| E2 level on the trigger day (pg/ml) | 0.060 | (0.000, 0.000) | < 0.001 |
| P level on the trigger day (ng/ml) | −0.014 | (−0.006, −0.002) | < 0.001 |
| FSH level on the trigger day (mIU/ml) | 0.000 | (−0.002, 0.002) | 0.935 |
| LH level on the trigger day (mIU/ml) | −0.081 | (−0.050, −0.040) | < 0.001 |
| COH protocols | |||
| Short agonist protocol | Reference | ||
| PPOS protocol | −0.209 | (−0.499, −0.407) | < 0.001 |
| GnRH antagonist protocol | −0.313 | (−1.002, −0.905) | < 0.001 |
| Mild stimulation protocol | −0.406 | (−1.478, −1.378) | < 0.001 |
| Long agonist protocol | 0.107 | (0.436, 0.537) | < 0.001 |
| Ovulation trigger methods | |||
| Dual trigger (GnRH agonist + hCG) | Reference | ||
| GnRH agonist | −0.016 | (−0.324, −0.108) | < 0.001 |
| hCG | −0.055 | (−0.173, −0.116) | < 0.001 |
Multiple linear regression analysis was conducted to explore the significant influencers on oocyte retrieval time. β Standardized Regression coefficient, 95%CL 95% confidence internal, AFC antral follicle count, E2 estrogen, P progesterone, LH luteinizing hormone, BMI body mass index
All potential determinants of oocyte retrieval time were further analysed using Directed Acyclic Graph (DAG) analysis to eliminate confounding and mediating variables, thereby identifying crucial determinants essential for robust research modeling across multiple correlations. The minimal adjustment sets required to estimate the total effect of the number of follicles (> 14 mm in diameter on the trigger day) on oocyte retrieval time included AFC, COH protocol, E2 level on the trigger day, female age, P level on the trigger day, and LH level on the trigger day (Fig. 2).
Fig. 2.
Directed acyclic graph (DAG) analysis was employed to identify the key determinants for oocyte retrieval time. AFC, antral follicle count; E2: estrogen; P: progesterone, LH: luteinizing hormone; BMI: body mass index
The basic clinical characters and embryo outcomes within Group 1 (Well-organized retrieval timing group) and Gorup 2 (insufficiently planned retrieval timing group)
The baseline characteristics of two timing groups across five COH protocols are summarized in Table 2. A total of 49, 961 oocyte pick-up cycles were included, with PPOS being the most common COH protocol (34,167 cycles), followed by GnRH antagonist protocol (6,284 cycles), mild stimulation protocol (4,574 cycles), long GnRH agonist protocol (2,607 cycles), and short GnRH agonist protocol (2,329 cycles). Demographic and basic characteristics, including female age, BMI, duration of infertility, basal FSH, LH levels, and hormone levels on the trigger day, were generally comparable between patients with well-organized retrieval timing and those with insufficiently planned retrieval timing within each COH protocol. However, a consistently lower percentage of primary infertility was observed in patients with well-organized retrieval timing across all COH protocols. Among all causes of infertility, tubal factors were found to be the most prevalent. Furthermore, a higher percentage of tubal factors was observed in the group with good timing groups for PPOS, short GnRH-a and GnRH antagonist protocols.
Table 2.
The basic clinical characteristics and embryo outcomes within Group 1 (Well-organized retrieval timing group) and Gorup 2 (insufficiently planned retrieval timing group)
| PPOS | Long GnRH agonist | Short GnRH agonist | Mild stimulation | GnRH Antagonist | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Group 1 | Group 2 | Group 1 | Group 2 | Group 1 | Group 2 | Group 1 | Group 2 | Group 1 | Group 2 | |
| Number of women (n) | 13,493 | 20,674 | 1077 | 1530 | 872 | 1457 | 1501 | 3073 | 2548 | 3736 |
| Female age (year) | 38.68 ± 5.87a | 39.02 ± 5.73b | 37.82 ± 4.859 | 37.58 ± 4.910 | 38.65 ± 5.04 | 39.04 ± 4.97 | 41.28 ± 6.02 | 41.08 ± 5.78 | 36.73 ± 5.51 | 36.76 ± 5.45 |
| BMI (kg/m2) | 21.92 ± 3.41a | 22.01 ± 3.56b | 21.78 ± 3.37 | 21.84 ± 3.59 | 21.74 ± 3.42 | 22.03 ± 3.67 | 22.42 ± 3.63a | 22.15 ± 4.06b | 22.22 ± 3.40 | 22.07 ± 3.42 |
| Infertility duration (year) | 3.34 ± 3.12a | 3.65 ± 3.33b | 3.58 ± 3.28 | 3.69 ± 3.02 | 3.73 ± 3.21 | 3.86 ± 3.32 | 3.71 ± 3.59 | 3.86 ± 3.74 | 3.20 ± 3.07a | 3.39 ± 3.19b |
| Primary infertility (n, %) | 6647(49.26%)a | 11,040(53.40%)b | 542(50.32%)a | 879(57.45%)b | 398(45.64%)a | 780(53.53%)b | 682(45.44%)a | 1536(49.98%)b | 1334(52.35%)a | 2116(56.64%)b |
| Basal FSH (mIU/ml) | 6.81 ± 3.14a | 6.64 ± 2.97b | 3.48 ± 2.02 | 3.41 ± 1.84 | 5.47 ± 2.33 | 5.45 ± 2.28 | 6.54 ± 2.80a | 6.34 ± 2.62b | 6.66 ± 2.74a | 6.48 ± 2.53b |
| Basal LH (mIU/ml) | 3.81 ± 2.54 | 3.76 ± 2.64 | 0.79 ± 1.14 | 0.87 ± 1.25 | 3.60 ± 2.12 | 3.69 ± 2.38 | 3.85 ± 5.44 | 3.70 ± 4.67 | 3.47 ± 2.43 | 3.36 ± 2.15 |
| Causes of infertility (n, %) | ||||||||||
| Tubal factor | 6409(47.50%)a | 9106(44.05%)b | 502(46.61%) | 684(44.71%) | 519(59.52%)a | 797(54.70%)b | 692(46.10%)a | 1538(50.05%)b | 1035(40.62%)a | 1379(36.91%)b |
| Male factor | 1477(10.95%) | 2394(11.58%) | 95(8.82%) | 163(10.65%) | 74(8.49%) | 144(9.88%) | 148(9.86%) | 308(10.02%) | 342(13.42%)a | 570(15.26%)b |
| Endometriosis | 536(3.97%) | 855(4.14%) | 87(8.08%)a | 85(5.56%)b | 23(2.64%) | 46(3.16%) | 74(4.93%) | 125(4.07%) | 82(3.22%) | 125(3.35%) |
| PCOS | 298(2.21%)a | 551(2.67%)b | 8(0.74%) | 23(1.50%) | 7(0.80%) | 24(1.65%) | 15(1.00%)a | 61(1.99%)b | 65(2.55%) | 118(3.16%) |
| Combined factor | 2462(18.25%)a | 4303(20.81%)b | 217(20.15%) | 345(22.55%) | 145(16.63%) | 287(19.70%) | 285(18.99%)a | 572(18.61%)a | 355(13.93%)a | 613(16.41%)b |
| Unexplained factor | 2311(17.13%) | 3465(16.76%) | 168(15.60%) | 230(15.03%) | 104(11.93%) | 159(10.91%) | 287(19.12%)a | 469(15.26%)b | 669(26.3%) | 931(24.9%) |
| AFC (n) | 9.26 ± 6.37a | 9.63 ± 6.79b | 7.13 ± 6.37a | 7.93 ± 6.76b | 8.28 ± 5.45a | 8.97 ± 6.33b | 6.07 ± 5.53a | 7.34 ± 6.50b | 8.98 ± 5.88a | 9.59 ± 6.56b |
| hMG duration (d) | 8.67 ± 2.14a | 8.91 ± 2.26b | 11.49 ± 3.10 | 11.46 ± 3.12 | 9.07 ± 3.24 | 9.23 ± 3.54 | 6.81 ± 4.55a | 7.62 ± 4.00b | 8.63 ± 2.42a | 8.87 ± 2.35b |
| hMG doses (IU) |
1821.03 ± 605.74a |
1863.78 ± 621.13b |
2562.39 ± 840.82 |
2545.82 ± 856.93 |
1761.12 ± 754.69a |
1828.64 ± 848.92b |
1287.89 ± 1020.99a |
1460.05 ± 931.34b |
1929.26 ± 710.82a |
1965.56 ± 696.48b |
| Number of follicles > 14 mm on the trigger day (n) | 7.00 ± 5.54a | 7.39 ± 5.97b | 7.66 ± 5.27a | 8.60 ± 5.53b | 7.48 ± 4.72 | 7.72 ± 4.88 | 4.28 ± 4.33a | 5.57 ± 5.07b | 6.25 ± 4.80a | 6.62 ± 4.91b |
| E2 level on the trigger day (pg/ml) |
2458.07 ± 1595.32 |
2476.54 ± 1569.68 |
2465.59 ± 1487.52a |
2625.35 ± 1468.83b |
2664.37 ± 1485.68 |
2748.35 ± 1446.24 |
1488.08 ± 1348.22a |
1919.59 ± 1480.29b |
2195.01 ± 1490.30 |
2230.10 ± 1457.53 |
| P level on the trigger day (ng/ml) | 1.10 ± 5.77a | 1.27 ± 3.44b | 0.75 ± 1.03a | 0.88 ± 1.64b | 1.03 ± 1.66 | 1.19 ± 2.66 | 1.44 ± 3.12 | 1.37 ± 3.23 | 0.58 ± 0.61 | 0.62 ± 0.80 |
| FSH level on the trigger day (mIU/ml) | 14.51 ± 4.10a | 14.26 ± 4.08b | 14.57 ± 4.18 | 14.30 ± 4.31 | 12.44 ± 4.20 | 12.23 ± 4.39 | 12.75 ± 4.60 | 12.55 ± 4.41 | 13.95 ± 3.52 | 13.83 ± 3.53 |
| LH level on the trigger day (mIU/ml) | 2.41 ± 1.70a | 2.30 ± 1.62b | 0.92 ± 1.01 | 0.97 ± 1.05 | 3.58 ± 2.06 | 3.64 ± 2.11 | 4.03 ± 2.45 | 3.69 ± 2.32 | 2.23 ± 1.63a | 2.06 ± 1.55b |
| Oocyte retrieval time (h) | 36.49 ± 0.85a | 36.45 ± 0.88b | 37.35 ± 1.00 | 37.41 ± 1.00 | 36.76 ± 0.96a | 36.68 ± 1.05b | 35.29 ± 1.25a | 35.18 ± 1.20b | 36.02 ± 0.72 | 36.03 ± 0.75 |
| Oocytes retrieved (n) | 9.48 ± 7.47a | 7.70 ± 6.93b | 9.78 ± 6.37a | 8.75 ± 6.38b | 8.88 ± 5.97a | 7.08 ± 5.16b | 5.37 ± 5.28 | 5.53 ± 5.65 | 8.89 ± 6.82a | 7.59 ± 6.55b |
| Mature oocytes (n) | 8.72 ± 6.69a | 5.86 ± 5.28b | 9.06 ± 5.75a | 6.56 ± 4.63b | 8.29 ± 5.48a | 5.54 ± 4.02b | 5.05 ± 4.82a | 4.09 ± 4.27b | 8.19 ± 6.11a | 5.40 ± 4.66b |
| Oocyte retrieval rate (%) | 89.59 ± 10.85a | 61.58 ± 22.40b | 88.64 ± 10.62a | 63.82 ± 20.88b | 88.14 ± 10.93a | 58.79 ± 19.65b | 92.18 ± 10.75a | 56.43 ± 23.08b | 90.31 ± 10.57a | 65.94 ± 22.97b |
| Mature oocyte rate (%) | 94.39 ± 7.04a | 76.55 ± 26.71b | 94.25 ± 6.99a | 78.22 ± 22.35b | 94.78 ± 6.96a | 80.53 ± 22.49b | 96.93 ± 6.13a | 72.81 ± 31.85b | 94.25 ± 7.29a | 74.04 ± 26.00b |
| Ovulation trigger methods | ||||||||||
| IVF (n, %) | 7987(59.19%)a | 10,921(53.40%)b | 577(53.57%)a | 657(43.00%)b | 532(61.01%)a | 777(53.55%)b | 942(62.80%)a | 1800(59.27%)b | 1285(50.43%)a | 1409(38.04%)b |
| ICSI (n, %) | 4074(30.19%)a | 7589(37.10%)b | 431(40.02%)a | 783(51.24%)b | 306(35.09%)a | 605(41.70%)b | 526(35.07%) | 1125(37.04%) | 976(38.30%)a | 1951(52.67%)b |
| IVF + ICSI (n, %) | 1432(10.61%)a | 1943(9.50%)b | 69(6.41%) | 88(5.76%) | 34(3.90%) | 69(4.76%) | 32(2.13%)a | 112(3.69%)b | 287(11.26%)a | 344(9.29%)b |
| Viable embryos (n) | 3.73 ± 3.12a | 2.57 ± 2.60b | 3.85 ± 2.69a | 2.95 ± 2.37b | 3.45 ± 2.67a | 2.45 ± 2.29b | 2.19 ± 2.30a | 1.89 ± 2.28b | 3.74 ± 2.85a | 2.60 ± 2.22b |
| Top-quality embryos (n) | 3.32 ± 3.34a | 2.25 ± 2.64b | 3.17 ± 2.85a | 2.34 ± 2.44b | 3.05 ± 2.89a | 2.11 ± 2.34b | 1.86 ± 2.28a | 1.65 ± 2.30b | 3.07 ± 2.94a | 1.99 ± 2.25b |
| Clinical pregnancy rate (%) |
48.56 (7164/14754) |
49.60 (8596/17329) |
34.10 (636/1865) |
31.52 (720/2284) |
45.95 (465/1012) |
44.30 (595/1343) |
41.35 (502/1214) |
43.80 (967/2208) |
34.61 (1368/3953)a |
32.44 (1558/4803)b |
| Miscarriage rate (%) |
14.00 (1003/7164) |
14.86 (1277/8596) |
23.11 (147/636)a |
18.06 (130/720)b |
16.77 (78/465) |
15.63 (93/595) |
17.93 (90/502) |
16.65 (161/967) |
19.66 (269/1368) |
20.09 (313/1558) |
| Live birth rate (%) |
40.37 (5956/14754) |
41.11 (7124/17329) |
24.99 (466/1865) |
24.82 (567/2284) |
37.06 (375/1012) |
35.44 (476/1343) |
32.54 (395/1214) |
35.33 (780/2208) |
26.61 (1052/3953) |
24.92 (1197/4803) |
Continuous variables were presented as mean ± standard deviation (SD) and analysed using either a two independent-sample t-test or Mann–Whitney U-test, depending on appropriateness. Categorical variables were expressed as numbers (%) and were compared using the chi-square test. Group 1 (Well-organized retrieval timing group) included cycles where oocyte retrieval rate ≥ 70% and mature oocyte rate ≥ 80%, whereas the remaining cycles were classified as Gorup 2 (insufficiently planned retrieval timing group). Different letters as a or b represent significant differences between groups: P < 0.05. PPOS progestin-primed ovarian stimulation, hMG human menopausal gonadotropin, BMI body mass index, AFC antral follicle count
However, significant difference was observed in AFC, hMG duration and dosages, and the number of follicles > 14 mm in diameter on the trigger day between two timing groups across most COH protocols. Nevertheless, a greater number of oocytes were retrieved, with a higher proportion of mature oocytes, viable embryos, and top-quality embryos in the good timing groups. Except for the long GnRH-a and GnRH-ant protocols where no significant difference was found in oocyte retrieval time, the oocyte retrieval time was longer in the good timing groups for PPOS, short GnRH-a protocols and mild stimulation protocol. Regarding pregnancy outcomes, all embryo transfer cycles resulting from oocyte retrieval in these individuals were included, encompassing a total of 50,765 transfer cycles comprising 5,567 fresh embryo transfers and 45,198 frozen embryo transfers. The clinical pregnancy rate per transfer cycle, as well as the miscarriage rate and live birth rate per transfer cycle, showed comparable results between the well-organized timing and insufficient planned timing groups across most COH protocols.
The distribution of oocyte retrieval rate, oocyte mature rate and oocyte retrieval time within two timing groups
Figure 3 plotted the distribution of oocyte retrieval rate, oocyte mature rate and oocyte retrieval time within Group 1 (well-organized retrieval timing group: achieving an oocyte retrieval rate of at least 70% and an oocyte maturation rate of at least 80%) and Group 2 (insufficiently planned retrieval timing group: having an oocyte retrieval rate below 70% or an oocyte maturation rate below 80%). In comparison to Group 2, Group 1 demonstrated significantly higher and more concentrated distribution of oocyte retrieval rates and mature oocyte rates across various protocols, including the short agonist protocol (Fig. 3A1-A2), PPOS protocol (Fig. 3B1-B2), GnRH antagonist protocol (Fig. 3C1-C2), mild stimulation protocol (Fig. 3D1-D2), and long agonist protocol (Fig. 3E1-E2). The distributions of oocyte retrieval time exhibited significant differences in the short agonist protocol (Fig. 3A3), PPOS protocol (Fig. 3B3), and mild stimulation protocol (Fig. 3D3). However, there were minimal differences observed in the long agonist protocol (Fig. 3E3) and the distributions were similar in the GnRH antagonist protocol (Fig. 3C3).
Fig. 3.
The violin diagram illustrates the oocyte retrieval rate, oocyte mature rate, and scatter plots of oocyte retrieval time for two timing groups within five protocols. A1-E1, A2-E2, and A3-E3 demonstrate the oocyte retrieval rate, oocyte mature rate, and oocyte retrieval time separately in the short agonist protocol, PPOS protocol, GnRH antagonist protocol, mild stimulation protocol, and long agonist protocol. Group 1: Well-organized retrieval timing group which achieving an oocyte retrieval rate of at least 70% and an oocyte maturation rate of at least 80%; Group 2: Insufficiently planned retrieval timing group which having an oocyte retrieval rate below 70% and an oocyte maturation rate below 80%
The development, evaluation and validation of a predictive formula for determining the optimal timing of oocyte retrieval
The training set data (70% of Group 1) was utilized to develop a multivariate linear regression (MLR) model for predicting the optimal oocyte retrieval time formula (Fig. 4A). This model incorporates key determinants, including antral follicle count (AFC), controlled ovarian hyperstimulation (COH) protocol, estradiol (E2) level on the trigger day, female age, number of follicles > 14 mm in diameter on the trigger day, and progesterone (P) and luteinizing hormone (LH) levels on the trigger day as identified in previous analyses (Fig. 2). The β coefficients within this predictive model of oocyte retrieval time varied across different COH protocols (Fig. 4A). The predictive model demonstrated linearity (Fig. 4B) and homogeneity of variance (Fig. 4C), as evidenced by a consistent and horizontal reference line. Moreover, it exhibited no outliers or discrete values in the influential observations (Fig. 4D), and the collinearity analysis revealed no evidence of multicollinearity among the variables (Fig. 4E, VIF < 4).
Fig. 4.
A predictive model for the timing of oocyte retrieval was developed, evaluated, and validated. A The predictive formula incorporated multiple linear regression covariates to optimize the timing of oocyte retrieval. Evaluation indices including residual fit plot (B), location scale plot (C), residual leveraged plot (D), Variance inflation factor plot (E), and model evaluation analysis (F) were employed for assessing the performance of this predictive model. R.2: R-Squared; MAE: mean absolute error; MSE: mean squared error; RMSE: root mean squared error. Posterior Predictive Check analysis (G) and test set validation (H) were performed to evaluate the predictive performance of the model. Furthermore, data from Group 2 were employed to assess the model's performance, involving comparisons between actual and predicted oocyte retrieval timing (I) as well as their distribution patterns (J). n.s., not significant; ****p < 0.0001, determined by a two independent-sample t-test (two-sided)
Furthermore, the predictive model for oocyte retrieval time demonstrated excellent performance when comparing the training set and validation set, as evidenced by similar measures of adjusted R2 (0.311 for the training set, 0.313 for the validation set), R2 (0.312 for the training set, 0.314 for the validation set), MAE (mean absolute error; 0.611 for the training set, 0.616 for the validation set), MSE (mean squared error; 0.653 for the training set and 0.675 for the validation set), and RMSE (root mean squared error; 0.808 for the training set and 0.822 for the validation set) in both datasets (Fig. 4F). The validity of this predictive model was further confirmed by a similar distribution between true values (represented by green thick line) and predicted values from the model (represented by blue line) in the training dataset. Most lines overlapped or exhibited minimal differences, indicating negligible disparities between the true and predicted values, thereby establishing the reliability of the predictive model (Fig. 4G). Similarly, no significant distinction was observed between the true and predicted oocyte retrieval time in the validation set data (Fig. 4H).
Additionally, we utilized Group 2, defined as cycles with oocyte retrieval rates below 70% or maturation rates below 80%, as an additional validation dataset. A substantial discrepancy was identified between the actual and predicted oocyte retrieval times (Fig. 4I, p < 0.0001, two-tailed t-test). Furthermore, the violin plot illustrating the distribution of actual versus model-predicted oocyte retrieval times for Group 2 reveals a marked difference in distribution patterns, with the actual timing displaying more discrete values (Fig. 4J). Furthermore, we utilized this group to evaluate the generalizability of our model. Specifically, cycles in Group 2 where the discrepancy between actual and model-predicted oocyte retrieval times did not exceed 30 min (MAE ~ 30 min in our model) were classified as approved cycles; the remaining cycles were categorized as rejected cycles. The approved group constituted 48.99% of the total, while the rejected group accounted for 51.01% (n = 14,928 for the approved group, n = 15,542 for the rejected group) (Supplemental Fig. 1A). Furthermore, the oocyte retrieval rate of the approved group was significantly higher than that of the rejected group (Supplemental Fig. 1B), whereas the mature oocyte rate did not exhibit a significant improvement (Supplemental Fig. 1C).
Discussion
Principal findings
This study presents the first comprehensive and objective analysis of key factors influencing the timing of oocyte retrieval, including age, AFC, COH protocol, number of follicles > 14 mm in diameter on the trigger day, as well as E2, P and LH levels on the trigger day. Furthermore, we calculated the weights of these determinants to establish a predictive model for oocyte retrieval timing. The model was further validated in a well-organized retrieval timing group with a minimum oocyte retrieval rate of 70% and oocyte maturation rate of 80%. These findings provide valuable insights into optimizing the timing of oocyte retrieval in patients undergoing assisted reproductive technology to maximize mature oocyte yield.
Results in the context of what is known
In the context of current study, there is a prevailing inclination towards employing a standardized time frame for oocyte retrieval. During the initial stages of in vitro fertilization (IVF), the timing of oocyte retrieval was predominantly restricted. A report from 1982 indicated that hCG administration occurred 36–38 h prior to laparoscopic oocyte retrieval [37]. However, other reports suggested that it was common practice to administer hCG 32–36 h before oocyte retrieval in order to prevent cycle cancellation due to spontaneous LH surge [38, 39]. Subsequently, several studies explored the correlation between oocyte retrieval time and mature oocyte rate. Raziel et al. investigated the impact of extending the interval from 35.3 ± 0.7 h to 38.6 ± 1.2 h in patients with ≥ 47% immature oocytes in their previous cycle and observed a significant increase in maturation rate within this prolonged interval [40]. Furthermore, a meta-analysis conducted in 2011 demonstrated that longer exposure duration to hCG before oocyte retrieval (> 36 h) resulted in higher rates of oocyte maturation (RR, 0.67; 95%CI, 0.62–0 0.73) compared to shorter intervals (< 36 h) [41]. These findings suggest that rather than adhering strictly to uniformity, determining an optimal duration for hCG exposure prior to oocyte retrieval may be contingent upon individual factors.
Despite considering the influencing factors for oocyte retrieval time, the current understanding is limited to only one or two factors, lacking comprehensive consideration and requiring further improvement. Previous findings have suggested that extending hCG exposure by 36.5 h may be advantageous for patients aged ≥ 40 years, indicating a potential age-dependent optimal duration of hCG exposure before oocyte retrieval [8]. Another aspect explored in previous research is the optimal ovulation trigger–oocyte pickup interval within a specific COH protocol. Studies have primarily focused on antagonist protocol cycles triggered exclusively by GnRH [37]. Several studies have reported that for patients who use clomiphene citrate (CC) and/or human menopausal gonadotropin (hMG) for ovulation trigger, the commonly practised interval is 32–36 h [2, 18, 19]. For patients who use gonadotropin-releasing hormone (GnRH) agonists or antagonists, the most recommended time interval is 36–39 h [20–22], which yields more ideal ART outcomes. Additionally, a retrospective study investigated the PPOS protocol and found that the optimal ovulation trigger–oocyte pickup interval ranged from 36.4 to 37.8 h for most patients using this protocol [23]. Furthermore, a comparative study involving four different ovarian stimulation protocols revealed that in order to retrieve more than 60% of oocytes and over 80% mature oocytes, the trigger-to-retrieval interval should be delayed according to the type of stimulation: mild stimulation protocol < GnRH antagonist protocol < short agonist protocol < long agonist protocol [5].
Clinical implications
In our study, we innovatively apply directed acyclic graphs (DAGs) to identify key determinants of oocyte retrieval time among a number of potential influencers. DAGs, as an intuitive data structure, integrate a priori knowledge and highlight key factors among multiple possibilities while eliminating confounding and mediating variables. The largest and most comprehensive review of the use of DAGs revealed their increasing popularity in applied health research as a transparent approach for identifying confounding variables to estimate causal effects. Furthermore, DAGs demonstrated exceptional proficiency in identifying crucial determinants necessary for robust research modeling across numerous correlations [42]. This approach allows for comprehensive interpretation of oocyte retrieval time by incorporating variables into the multiple linear regression model [43]. Therefore, our research employs MLR and DAGs analysis to clearly identify and quantify the contributions of fundamental clinical characteristics such as age, AFC, E2 levels, P levels, LH levels, follicular diameter and number on the day of trigger, as well as different COH protocols. All these factors collectively play pivotal roles in determining oocyte retrieval time.
Research implications
Previous studies have investigated various approaches to determine the optimal timing for oocyte retrieval. In 2012, Kawachiya, Satoshi et al. developed a strategy that combined NSAIDs with LH and progesterone levels on trigger day to estimate oocyte retrieval time [44]. Additionally, Kato et al. proposed a model in 2022 based on pre-trigger LH levels, adjusted by identifying the ascending or descending phase of the LH surge, to determine the appropriate timing [45]. Conversely, Li et al. emphasized that preovulatory serum progesterone levels are more accurate predictors of ovulation compared to LH levels [46]. Furthermore, Lu et al. reported that besides LH or P4 levels, decreases in preovulatory serum estradiol levels can serve as markers for premature ovulation [47]. However, it is important to note that these studies were conducted in natural or unstimulated cycles and previous research has demonstrated variations in oocyte retrieval time across different COH protocols [5]. Moreover, these studies relied solely on one or two hormonal profiles for determining oocyte retrieval timing; while informative and practical approaches, they often lack the precision required for accurately predicting optimal oocyte retrieval timing.
Strengths and limitations
In contrast to previous approaches for determining oocyte retrieval time, our predictive model not only incorporates a more comprehensive evaluation by considering key determinants identified through MLR and DAG analysis but also elucidates the varying weights of influences on oocyte retrieval timing. The weights of influences on oocyte retrieval timing decrease in the following order: COH regimen, luteinizing hormone level on the day of trigger, female age, number of follicles with a diameter greater than 14 mm on the day of trigger, AFC, P and E2 levels on the trigger day. From mild stimulation protocol to GnRH antagonist protocol, PPOS protocol, short agonist protocol, and finally long agonist protocol; there is a trend towards progressively extending oocyte retrieval time as supported by a retrospective study [5]. The inconsistent weights of influences related to oocyte retrieval timing were also reported in an ovulation prediction study that constructed two machine learning models; however, it only explored follicle diameter and pre-ovulatory serum LH, E2, and P4 levels [17]. Our comprehensive and objective analysis, considering the varying weights of influence on oocyte retrieval timing among determinants, has the potential to enhance the prediction of optimal timing for oocyte retrieval in clinical practice.
This valuable predictive model for determining the optimal timing of oocyte retrieval is derived from data obtained from Group 1, which represents a well-organized cohort with a minimum oocyte retrieval rate of 70% and an oocyte maturation rate of 80%. The dataset from Group 1 was randomly divided into two subsets: the training set (comprising 70% of the dataset) and the validation set (comprising 30% of the dataset). The predictive model exhibits excellent linearity (Fig. 4B) and homogeneity of variance (Fig. 4C). Moreover, no outliers or influential observations were detected in the model, and analysis of covariance revealed no multicollinearity among variables. To assess its performance, we calculated R-squared (R2), mean absolute error (MAE), mean squared error (MSE), and root mean square error (RMSE) between actual data and predictions generated by this model [48, 49]. Evaluation parameters demonstrated minimal discrepancies between the training and validation sets, indicating reliable random grouping. Importantly, there was significant overlap or slight deviation observed between real data and predictions made by this model in both sets; these findings were further confirmed during validation analysis. The findings suggest that our predictive model for oocyte retrieval time can effectively represent the timing of a well-coordinated oocyte retrieval group, thereby providing valuable guidance to optimize the yield of mature oocytes in clinical practice.
Additionally, we utilized Group 2 data as a secondary validation dataset. The results demonstrated a significant discrepancy between the model-predicted and actual oocyte retrieval times, suggesting that Group 2, with less favorable oocyte retrieval outcomes, had insufficient planning of retrieval timing. When assessing the model's generalizability in Group 2, more than half of the cycles were deemed unsuitable by the model. The approved group exhibited a significantly higher oocyte retrieval rate compared to the rejected group, while the mature oocyte rate did not show a significant improvement (Supplemental Fig. 1C). This relatively modest improvement can likely be attributed to a considerable proportion of cycles in Group 2 resulting in either zero oocyte retrieval or zero mature oocyte retrieval (new Supplemental Fig. 1B and C). These patients may generally exhibit characteristics such as older age, longer durations of infertility, and elevated FSH levels, which are indicative of diminished ovarian reserve and a higher incidence of infertility due to endometriosis [50]. Such conditions cannot be fully addressed by optimizing oocyte retrieval timing alone.
However, it is important to acknowledge the limitations of our study. Firstly, we would like to emphasize that this retrospective study solely relied on clinical data obtained from a single fertility center. Though we did not investigate the causal relationship between retrieval timing and oocyte outcomes, our modeling approach—similar to prior studies such as MammaPrint [51] or Alife’s trigger day prediction [52]—focused on learning from successful cycle patterns rather than testing mechanistic hypotheses. This translates historical clinical success into practical, data-driven tools for real-world IVF decision-making. Secondly, although the model’s R2 value of 0.31 and MAE of approximately 37 min may appear relatively modest, they still provide clinically meaningful guidance [53, 54]. In real-world IVF settings, factors such as scheduling variability, procedural constraints, and patient heterogeneity limit the precision of any predictive tool. Within this context, our model is not intended to entirely replace fixed scheduling strategies but rather to serve as an early warning system for atypical cases. It facilitates risk stratification and individualized adjustments while maintaining model simplicity and avoiding the inclusion of excessive input variables. Nonetheless, future work is needed to develop more precise predictive tools that incorporate additional physiological and cycle-specific parameters. Furthermore, the validation process was conducted internally without external verification, necessitating further confirmation in future research. The inclusion of a large number of ovarian stimulation cycles (49,961 in total) with diverse clinical characteristics, including 2,329 cases of short agonist protocol, 34,167 cases of PPOS protocol, 6,284 cases of GnRH antagonist protocol, 4,574 cases of mild stimulation protocol and 2,607 cases of long agonist protocol partially overcomes this limitation and provides comprehensive reference information.
Conclusion
This study demonstrates that female age, AFC, E2 levels, P levels, LH levels, number of follicles > 14 mm in diameter on the trigger day, as well as COH protocols are crucial factors for accurately predicting the timing of oocyte retrieval. By employing generalized linear regression analysis to synthesize these variables, we have developed a reliable and universally applicable reference formula for determining the optimal oocyte retrieval time in women with varying COH protocols and basic clinic characteristics. This not only enhances the acquisition of mature oocytes but also provides clinicians with a valuable tool for making informed decisions regarding oocyte retrieval timing.
Acknowledgements
We truly appreciate all of the medical professionals working in the Assisted Reproduction Department of Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine, including the doctors, nurses and lab technicians.
Clinical trial number
Not applicable.
Authors' contributions
YTH and ZK were responsible for data acquisition and analysis, as well as manuscript drafting. XS contributed to the data analysis. YHN, YQZ, and YLL participated in clinical procedures and data collection. LW supervised the study design and execution while critically revising it for important intellectual content. All authors have given their approval to the final version of the manuscript and agree to be accountable for all aspects of the work.
Funding
This study was financially supported by grants from the National Natural Science Foundation of China (grant number: 82071603, 82471738 to L.W., 82271732 to Y.K., 82201888 to X. Shen) and grants from the Natural Science Foundation of Shanghai (grant number: 18ZR1422600 to LW).
National Natural Science Foundation of China,82201888,82071603
Data availability
The data underlying this article will be shared by the corresponding author upon reasonable request. Please note that the cross-border sharing of original clinical data is subject to compliance with Chinese law.
Declarations
Ethics approval and consent to participate
This retrospective study used an anonymized database to keep patients'personally identifiable information absolutely confidential. This retrospective study was conducted in accordance with the Declaration of Helsinki and was approved by the ethics committee of Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine (approval number: SH9H-2024-T134-1). This is a retrospective and non-interventional study, and written informed consent was waived by the Ethics Committee of Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Yuting Huang and Zhe Kuang contributed equally to this work.
References
- 1.RG E, S PC. Induction of follicular growth, ovulation and luteinization in the human ovary. J Reprod Fertil Suppl. 1975;22:121–163. [PubMed]
- 2.Testart J, Frydman R. Minimum time lapse between luteinizing hormone surge or human chorionic gonadotropin administration and follicular rupture. Fertil Steril. 1982;37(1):50–3. [DOI] [PubMed] [Google Scholar]
- 3.Gudmundsson J, et al. Luteinization to oocyte retrieval delay in women in whom multiple follicular growth was induced as part of an in vitro fertilization/gamete intrafallopian transfer program. Fertil Steril. 1990;53(4):735–7. [DOI] [PubMed] [Google Scholar]
- 4.Larsen WJ, Wert SE, Brunner GD. A dramatic loss of cumulus cell gap junctions is correlated with germinal vesicle breakdown in rat oocytes. Dev Biol. 1986;113(2):517–21. [DOI] [PubMed] [Google Scholar]
- 5.Shen X, et al. The ovulation trigger-OPU time interval of different ovarian protocols in ART: a retrospective study. Arch Gynecol Obstet. 2020;302(2):519–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Weiss A, et al. Lag time from ovulation trigger to oocyte aspiration and oocyte maturity in assisted reproductive technology cycles: a retrospective study. Fertil Steril. 2014;102(2):419–23. [DOI] [PubMed] [Google Scholar]
- 7.Jamieson ME, et al. In vivo and in vitro maturation of human oocytes: effects on embryo development and polyspermic fertilization. Fertil Steril. 1991;56(1):93–7. [DOI] [PubMed] [Google Scholar]
- 8.Reichman DE, et al. Effect of time between human chorionic gonadotropin injection and egg retrieval is age dependent. Fertil Steril. 2011;95(6):1990–5. [DOI] [PubMed] [Google Scholar]
- 9.Helmer A, et al. Optimal Timing of Ovulation Triggering to Achieve Highest Success Rates in Natural Cycles—An Analysis Based on Follicle Size and Oestradiol Concentration in Natural Cycle IVF. Front Endocrinol. 2022;13. [DOI] [PMC free article] [PubMed]
- 10.Abbara A, et al. Follicle Size on Day of Trigger Most likely to Yield a Mature Oocyte. Front Endocrinol. 2018;9. [DOI] [PMC free article] [PubMed]
- 11.Laufer N, et al. The association between preovulatory serum 17 beta-estradiol pattern and conception in human menopausal gonadotropin-human chorionic gonadotropin stimulation. Fertil Steril. 1986;46(1):73–6. [DOI] [PubMed] [Google Scholar]
- 12.Meyer WR, et al. Value of estradiol response after human chorionic gonadotropin administration in predicting in vitro fertilization success. Fertil Steril. 1999;72(3):542–5. [DOI] [PubMed] [Google Scholar]
- 13.Chiasson MD, et al. Measuring estradiol levels after human chorionic gonadotropin administration for in vitro fertilization is not clinically useful. Fertil Steril. 2007;87(2):448–50. [DOI] [PubMed] [Google Scholar]
- 14.Kondapalli LA, et al. A decrease in serum estradiol levels after human chorionic gonadotrophin administration predicts significantly lower clinical pregnancy and live birth rates in in vitro fertilization cycles. Hum Reprod. 2012;27(9):2690–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Huang R, et al. Serum estradiol level change after human chorionic gonadotropin administration had no correlation with live birth rate in IVF cycles. Eur J Obstet Gynecol Reprod Biol. 2014;178:177–82. [DOI] [PubMed] [Google Scholar]
- 16.Lu X, et al. Decrease in preovulatory serum estradiol is a valuable marker for predicting premature ovulation in natural/unstimulated in vitro fertilization cycle. J Ovarian Res. 2018;11(1). [DOI] [PMC free article] [PubMed]
- 17.Li Y, H Zeng, J Fu. Preovulatory progesterone levels are the top indicator for ovulation prediction based on machine learning model evaluation: a retrospective study. J Ovarian Res. 2024;17(1). [DOI] [PMC free article] [PubMed]
- 18.Edwards RG, Steptoe PC. Induction of follicular growth, ovulation and luteinization in the human ovary. J Reprod Fertil Suppl. 1975;22:121–63. [PubMed] [Google Scholar]
- 19.Trounson A, Leeton JF, Wood C. In vitro fertilization and embryo transfer in the human. Follicular Maturation Ovulation. 1982:313–322.
- 20.Fleming R, Coutts JR. Induction of multiple follicular development for IVF. Br Med Bull. 1990;46(3):596–615. [DOI] [PubMed] [Google Scholar]
- 21.Mansour RT, Aboulghar MA, Serour GI. Study of the optimum time for human chorionic gonadotropin-ovum pickup interval in in vitro fertilization. J Assist Reprod Genet. 1994;11(9):478–81. [DOI] [PubMed] [Google Scholar]
- 22.Vits A, Gerris J, Joostens M. Comparison between two HCG-to-oocyte aspiration intervals (36 versus 38) on the outcome of in-vitro fertilization. Hum Reprod. 1994;9:12–5.7962457 [Google Scholar]
- 23.Shen X, et al. Optimal Ovulation Trigger–Oocyte Pickup Interval in Progestin-Primed Ovarian Stimulation Protocol: A Retrospective Study Using Propensity Score Matching. Front Endocrinol. 2019;10. [DOI] [PMC free article] [PubMed]
- 24.Tien NV, Hoi NX. Optimizing the time interval between hCG triggering and occyte retrieval in different ovarian protocols in art: a prospective study in vietnam. Fertil Steril. 2014;102(3).
- 25.Deng M, et al. A moderately extended time interval between hCG administration and oocyte retrieval is good for most patients with oocyte retrieval scheduled on the same day: a retrospective cohort study. J Obstet Gynaecol. 2019;40(7):1006–11. [DOI] [PubMed] [Google Scholar]
- 26.Bokal EV, et al. Prolonged HCG action affects angiogenic substances and improves follicular maturation, oocyte quality and fertilization competence in patients with polycystic ovarian syndrome. Hum Reprod. 2005;20(6):1562–8. [DOI] [PubMed] [Google Scholar]
- 27.Kuang Y, et al. Medroxyprogesterone acetate is an effective oral alternative for preventing premature luteinizing hormone surges in women undergoing controlled ovarian hyperstimulation for in vitro fertilization. Fertil Steril. 2015;104(1):62-70.e3. [DOI] [PubMed] [Google Scholar]
- 28.Nie Y, et al. The cumulative live birth rates of 18 593 women with progestin-primed ovarian stimulation-related protocols and frozen-thawed transfer cycles. Hum Reprod Open. 2024;2024(1). [DOI] [PMC free article] [PubMed]
- 29.Chen H, et al. Comparison of live-birth defects after luteal-phase ovarian stimulation vs. conventional ovarian stimulation for in vitro fertilization and vitrified embryo transfer cycles. Fertil Steril. 2015;103(5):1194–1201.e2. [DOI] [PubMed]
- 30.Wang N, et al. Luteal-phase ovarian stimulation vs conventional ovarian stimulation in patients with normal ovarian reserve treated for IVF: a large retrospective cohort study. Clin Endocrinol. 2015;84(5):720–8. [DOI] [PubMed] [Google Scholar]
- 31.Hu Z, et al. Differences in Ectopic Pregnancy Rates between Fresh and Frozen Embryo Transfer after In Vitro Fertilization: A Large Retrospective Study. J Clin Med. 2022;11(12). [DOI] [PMC free article] [PubMed]
- 32.Huang J, et al. A Higher Estradiol Rise After Dual Trigger in Progestin-Primed Ovarian Stimulation Is Associated With a Lower Oocyte and Mature Oocyte Yield in Normal Responders. Front Endocrinol (Lausanne). 2019;10:696. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Cummins JM, et al. A formula for scoring human embryo growth rates in in vitro fertilization: its value in predicting pregnancy and in comparison with visual estimates of embryo quality. J In Vitro Fert Embryo Transf. 1986;3(5):284–95. [DOI] [PubMed] [Google Scholar]
- 34.Gardner DK, Schoolcraft WB. Culture and transfer of human blastocysts. Curr Opin Obstet Gynecol. 1999;11(3):307–11. [DOI] [PubMed] [Google Scholar]
- 35.Yu S, et al. New application of dydrogesterone as a part of a progestin-primed ovarian stimulation protocol for IVF: a randomized controlled trial including 516 first IVF/ICSI cycles. Hum Reprod. 2018;33(2):229–37. [DOI] [PubMed] [Google Scholar]
- 36.Wu J, et al. Fertility and Neonatal Outcomes of Freeze-All vs. Fresh Embryo Transfer in Women With Advanced Endometriosis. Front Endocrinol. 2019;10. [DOI] [PMC free article] [PubMed]
- 37.Ranit H, et al. Is there a preferred time interval between gonadotropin-releasing hormone (GnRH) agonist trigger and oocyte retrieval in GnRH antagonist cycles? A retrospective cohort of planned fertility preservation cycles. J Assist Reprod Genet. 2024;41(6):1531–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Marinho AO, et al. Real time pelvic ultrasonography during the periovulatory period of patients attending an artificial insemination clinic. Fertil Steril. 1982;37(5):633–8. [DOI] [PubMed] [Google Scholar]
- 39.Tuwatananurak JP, et al. Machine Learning Can Improve Estimation of Surgical Case Duration: A Pilot Study. Journal of Medical Systems, 2019. 43(3). [DOI] [PubMed]
- 40.Raziel A, et al. In vivo maturation of oocytes by extending the interval between human chorionic gonadotropin administration and oocyte retrieval. Fertil Steril. 2006;86(3):583–7. [DOI] [PubMed] [Google Scholar]
- 41.Wang W, et al. The time interval between hCG priming and oocyte retrieval in ART program: a meta-analysis. J Assist Reprod Genet. 2011;28(10):901–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Tennant PWG, et al. Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations. Int J Epidemiol. 2021;50(2):620–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Lipsky AM, Greenland S. Causal Directed Acyclic Graphs. JAMA. 2022;327(11). [DOI] [PubMed]
- 44.Kawachiya S, et al. Short-term, low-dose, non-steroidal anti-inflammatory drug application diminishes premature ovulation in natural-cycle IVF. Reprod Biomed Online. 2012;24(3):308–13. [DOI] [PubMed] [Google Scholar]
- 45.Kato K, Ueno S. Japan—Kato Ladies Clinic. In: von Wolff M, editor. Natural Cycle and Minimal Stimulation IVF : From Physiology to Clinical Practice, 2022. Springer International Publishing. pp. 261–270.
- 46.Li Y, Zeng H, Fu J. Preovulatory progesterone levels are the top indicator for ovulation prediction based on machine learning model evaluation: a retrospective study. J Ovarian Res. 2024;17(1):169. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Lu X, et al. Decrease in preovulatory serum estradiol is a valuable marker for predicting premature ovulation in natural/unstimulated in vitro fertilization cycle. J Ovarian Res. 2018;11(1):96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Garcia JMV, Bahloul MA, Laleg-Kirati T-M. A Multiple Linear Regression Model for Carotid-to-Femoral Pulse Wave Velocity Estimation Based on Schrodinger Spectrum Characterization. In: 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2022. pp. 143–147. [DOI] [PubMed]
- 49.Kouadri S, et al. Prediction of irrigation groundwater quality parameters using ANN, LSTM, and MLR models. Environ Sci Pollut Res. 2021;29(14):21067–91. [DOI] [PubMed] [Google Scholar]
- 50.Wang Y, et al. Causes and Effects of Oocyte Retrieval Difficulties: A Retrospective Study of 10,624 Cycles. Front Endocrinol. 2022;12:564344. [DOI] [PMC free article] [PubMed]
- 51.van 't Veer LJ, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature. 2002;415(6871):530–536. [DOI] [PubMed]
- 52.Fanton M, et al. An interpretable machine learning model for predicting the optimal day of trigger during ovarian stimulation. Fertil Steril. 2022;118(1):101–8. [DOI] [PubMed] [Google Scholar]
- 53.Vasilevskis EE, et al. Mortality Probability Model III and Simplified Acute Physiology Score II: Assessing Their Value in Predicting Length of Stay and Comparison to APACHE IV. Chest. 2009;136(1):89–101. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Gupta A, Stead TS, Ganti L. Determining a Meaningful R-squared Value in Clinical Medicine. Acad Med Surg. 2024.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data underlying this article will be shared by the corresponding author upon reasonable request. Please note that the cross-border sharing of original clinical data is subject to compliance with Chinese law.