This hospital-based cohort study recruited a total of 1156 patients who underwent IVF/ICSI cycles using the GnRH-ant protocol at Women’s Hospital of Nanjing Medical University between January 2016 and December 2022. The patients were diagnosed according to the POSEIDON criteria of Groups 3 or 4 (AFC ≤ 5 or AMH < 1.2 ng/ml). The exclusion criteria included: (1) The number of oocytes retrieved in this IVF/ICSI cycle was less than 3, (2) polycystic ovarian syndrome, endometriosis, history of ovarian surgery, metabolic or endocrine abnormalities, (3) Abnormal parental karyotypes, (4) preimplantation genetic diagnosis (PGT) cycle, (5) recurrent implant failure or spontaneous abortions, congenital or acquired uterine malformations, (6) missing cycle data or follow-up. This study adhered to the Declaration of Helsinki and was approved by the ethics committee of Nanjing Maternity and Child Health Care Hospital (NJFY-2023KY-018). The study was retrospective and analyzed patient data anonymously, eliminating the need for informed patient consent. The study flowchart was shown in Figure S1.

During the days 2 to 4 of natural menstrual cycle, ovarian reserve assessments were meticulously conducted, occurring 1 to 3 months preceding the commencement of ovarian stimulation procedures.

The AFC, defined as the cumulative number of follicles measuring 2 to 10 mm in diameter within the ovary, was meticulously measured using two-dimensional transvaginal ultrasound. This assessment was performed by a team of highly skilled reproductive medicine experts at our reproductive center. Each member of the team has undergone rigorous training in ultrasonography and reproductive medicine, boasting a minimum of 5 years of professional expertise. This ensures the utmost precision and reproducibility of the AFC measurements.

The serum concentration of AMH was accurately measured utilizing the Beckman DX1800 chemiluminescence analyzer (serial no. 607564). The assay employed the Beckman AMH reagent (batch no. 971017) and calibrator (batch no. 989302) to ensure precision. For quality control, Preci Control AMH (batch no. 42628901) was utilized to safeguard the accuracy and reproducibility of the results. Blood specimens were obtained from the patient in the morning, during the early follicular phase of the menstrual cycle, to capture the most representative AMH levels.

DFP was defined as the ratio of the number of follicles measuring ≥ 18 mm to the number of follicles measuring ≥ 12 mm on the trigger day. Our study exclusively focused on patients with a poor ovarian response, with a median follicle count of 5 on the trigger day. Meanwhile, existing GnRH-ant protocol guidelines recommend triggering when there are ≥ 2 follicles measuring ≥ 18 mm. Therefore, we adopted a DFP threshold of 40% (2/5) for patient stratification, dividing them into DFP ≤ 40% and DFP > 40% groups.

All patients participating in the study underwent a flexible GnRH-ant protocol. On the second or third day of their menstrual cycle, blood samples were collected to assess baseline serum levels of follicle-stimulating hormone (FSH), LH, and estradiol (E2), progesterone. Considering age, body mass index (BMI), AFC and AMH levels, the initial dose of Gn was tailored for each patient and was injected daily from the second or third day of the menstrual cycle. The Gn category encompasses recombinant FSH for injection (r-FSH, GONAL-f, Merck Serono, Italy; PUREGON, Merck Sharp & Dohme, Germany), as well as human menopausal gonadotropin (HMG, Menotropins for Injection, Lizhu Pharmaceutical Group, China).Once the diameter of the dominant follicle reached 12–14 mm or the E2 levels surpassed 300 ng/L, subcutaneous administration of GnRH antagonists (Cetrorelix, Merck Serono, Darmstadt, Germany) commenced, with dosages ranging from 0.125 to 0.25 mg/day. These dosages were tailored to each patient’s weight and serum LH levels on the initial day of GnRH-ant protocol, and were maintained until the trigger day. Follicular growth was closely monitored by ultrasound and sex hormone levels every 2–3 days, enabling precise gonadotropin dosing adjustments.

The trigger time was determined according to the diameter and number of dominant follicles, the time of using Gn and the level of hormone. Final oocyte maturation was triggered by either HCG (Lizhu Pharmaceutical Factory, China) alone or with a dual trigger comprising 2000 IU HCG and GnRH agonist (0.2 mg Decapeptyl, Ferring International Center SA). Oocyte retrieval was scheduled 36 h later, ensuring optimal conditions for successful fertilization and subsequent embryo development. Oocytes were inseminated approximately 4–6 h after follicular aspiration by IVF or ICSI, depending on sperm quality. Morphologic criteria were used for embryo scoring. According to our previously published article [19], the embryos were cultured in vitro for 3 to 6 days for fresh embryo transfer (ET) cycle or cryopreservation.

For fresh ET, the following criteria must be met: endometrial thickness should be at least 8 mm with a uniform echo pattern, progesterone levels should remain below 1.5 µg/L, and without any relevant medical history. On day 3, one to two available cleavage embryos with high score are selected for ET. For frozen-thawed embryo transfer (FET), patients with embryo freeze-all strategy or patients who did not reach live delivery after fresh ET performed with endometrial preparation protocol for FET, including the natural/stimulated cycle and the artificial cycle, depending on the characteristics and preferences of each woman. One or two thawed embryos were transferred depending on the age, BMI, embryo quality, and personal will of each subject. For luteal phase support (LPS), intramuscular progesterone at a dose of (40 mg, Xianju Pharmaceutical Factory) and oral Duphaston (30 mg, Abbott Healthcare Products B.V.) were administered once daily. If a positive pregnancy test was obtained two weeks after ET, progesterone therapy was maintained until the 8th to 10th week of gestation.

The serum β-hCG test was performed 2 weeks post-FET. The implantation rate was calculated as the number of gestational sacs divided by the number of embryos transferred. Clinical pregnancy was defined as the presence of an intrauterine gestational sac with or without a fetal heartbeat, observed through transvaginal ultrasound after 6 weeks of gestation. Early miscarriage was defined as pregnancy loss before 12 weeks of gestation, whereas late miscarriage was defined as pregnancy loss after 12 weeks but before 28 weeks of gestation. Live birth was defined as a fetus born alive after 28 weeks of pregnancy. The main pregnancy outcomes were cumulative clinical pregnancy rate (CCPR) and cumulative live birth rate (CLBR). CCPR was calculated as the number of clinical pregnancy cycles divided by the number of first oocyte retrieval cycles, and CLBR was calculated as the number of live birth cycles divided by the number of first oocyte retrieval cycles. Secondary pregnancy outcomes were chemical pregnancy, clinical pregnancy, miscarriage, and live birth rates. Laboratory outcomes measured included the No. of oocytes retrieved, 2 pronuclei (PN), cleavages, available embryos, blastocysts, and high-quality blastocysts and the ratio of available embryos, blastocysts, and high-quality blastocysts.

The sample size estimation was conducted using PASS software, which based on the two primary outcomes, No. of oocytes retrieval and CLBR. It was estimated that the No. of oocytes retrieval in DFP ≤ 40% was about 5, while the group of DFP > 40% was 4. With a power of 0.8, the alpha of 0.05, the sample size ratio of 0.6, the mean difference of 1, and the standard deviation of 1.5, the estimated sample size for each group was 6 vs. 4. Regarding the CLBR, it was assumed to be around 60% and 45% in the DFP ≤ 40% and DFP > 40% group in Poseidon Group 3, and the rate was expected to be around 35% and 20% in the DFP ≤ 40% and DFP > 40% group in Poseidon Group 4. The sample size required was 257 vs. 129 (Poseidon Group 3) and 191 vs. 96 (Poseidon Group 4), with a power of 0.8, the alpha of 0.05, and the sample size ratio of 0.5. The sample size is basically enough to detect the main results difference between the two groups.

Statistical analyses were conducted using SPSS 27.0 software and R 4.2.1 statistical software. Continuous variables are presented as medians with interquartile ranges, and categorical variables as numbers/total numbers (percentages). Independent samples t-test was conducted to compare the arithmetic means of the two groups, while the χ2-test was applied to analyze the frequencies of attributive features. Restricted cubic splines (RCS) were used to visualize dose-response associations between DFP and reproductive outcomes, with continuous confounders (female age, male age, infertility type, infertility duration, BMI, basal FSH, basal LH, AMH, total Gn dose, total GnRH-ant dose, trigger drugs, sperm density, and insemination method) included. The RCS model incorporated three knots positioned at the 5th, 50th, and 95th percentiles. Multivariate logistic regression analysis was performed to examine the independent effects of clinical characteristics on CLBR, with adjusted OR (aOR) and 95% confidence intervals (CIs) calculated. All tests were two-tailed, and a P-value of < 0.05 was considered statistically significant.