The comparison focused exclusively on women classified as POSEIDON groups 3 and 4, a framework for low-prognosis patients that uses age, antral follicle count and anti-Müllerian hormone concentration to characterize expected ovarian response. Group 3 comprises women younger than 35 and group 4 those aged 35 or older, with both groups required to have an antral follicle count below 5 and/or an anti-Müllerian hormone concentration below 1.2 nanograms per milliliter. The single-center cohort included 203 cycles using GnRH antagonist, 243 using mild stimulation and 146 using progestin-primed ovarian stimulation. Cycles from January 2017 through June 2023 were analyzed, and the primary outcome was the probability of at least one live birth from all fresh and frozen embryos derived from one initiated stimulation cycle.

Because this was an observational study, the protocols were not assigned randomly, and the groups initially differed in important characteristics. The GnRH-antagonist group was younger and had higher anti-Müllerian hormone and antral follicle count values, while the mild-stimulation group had lower anti-Müllerian hormone and longer median infertility duration. To address these imbalances, the researchers used entropy balancing, a weighting method that optimizes balance simultaneously across three treatment groups. The prespecified variables were age, anti-Müllerian hormone, antral follicle count, baseline follicle-stimulating hormone, luteinizing hormone and estradiol, body-mass index, infertility duration and infertility type. After weighting, the maximum absolute standardized mean difference between any pair of groups for all nine variables was below 0.001, while the overall effective sample size was 448, or 76% of the original 592 observations.

The three regimens differed substantially in their laboratory profiles. Mild stimulation used a median total gonadotropin dose of 1,400 IU, compared with 2,089 IU for GnRH antagonist and 2,700 IU for progestin-primed stimulation. Trigger-day estradiol was highest with GnRH antagonist at a weighted median of 4,108 pmol/L, followed by mild stimulation at 3,919 pmol/L and progestin-primed treatment at 2,941 pmol/L. The GnRH-antagonist approach also produced higher weighted median numbers of total follicles, 6 versus 4 with each alternative; dominant follicles, 4 versus 3; retrieved oocytes, 4 versus 3 and 4; metaphase-II oocytes, 3 versus 2 and 3; and usable embryos, 2 versus 1 and 2. These embryological differences were statistically significant, but they did not produce significant differences in the two cumulative clinical endpoints.

Weighted multivariable logistic regression, with additional adjustment for age and anti-Müllerian hormone, likewise found no independent protocol association with cumulative live birth. Relative to GnRH antagonist, the adjusted odds ratio was 0.83 for mild stimulation and 1.15 for progestin-primed stimulation; neither was significant, nor was the direct comparison between progestin-primed stimulation and mild stimulation. The corresponding odds ratios for clinical pregnancy were 0.86, 1.08 and 1.26, again with no significant differences. All confidence intervals crossed the null value of 1. The live-birth model showed adequate calibration, but the clinical-pregnancy and interaction models had suboptimal calibration according to the Hosmer–Lemeshow test, adding uncertainty to the subgroup estimates. No significant interaction with POSEIDON group was detected for either cumulative live birth or clinical pregnancy.

A notable pattern concerned the threshold for producing any transferable embryo. The weighted proportion of cycles yielding at least one usable embryo was 83.5% with GnRH antagonist, 80.7% with mild stimulation and 83.8% with progestin-primed stimulation. Rates of no oocytes retrieved were 6.0%, 6.4% and 7.6%, respectively, while cancellation because no usable embryo was available occurred in 16.5%, 19.3% and 16.2%. None of these comparisons was statistically significant. The authors described this as a possible threshold effect: a protocol might influence how many embryos are produced after the first usable embryo becomes available without materially changing the probability of crossing that threshold. They explicitly treated the pattern as exploratory, and the observational design cannot establish that protocol choice causes embryo availability to remain similar.

One safety-related laboratory difference was detected. Premature luteinizing-hormone surges occurred in 1.5% of progestin-primed cycles, 4.1% of GnRH-antagonist cycles and 8.6% of mild-stimulation cycles, with a global P value of 0.047. After Holm correction, the significant pairwise difference was between mild stimulation and progestin-primed stimulation; the other pairwise comparisons were not significant. No cases of moderate or severe ovarian hyperstimulation syndrome occurred in any group. Miscarriage occurred in 9 of 94 clinical pregnancies with GnRH antagonist, 12 of 79 with mild stimulation and 8 of 55 with progestin-primed stimulation. These secondary event counts were not presented as statistically tested protocol comparisons, and the study was not large enough to draw strong conclusions about uncommon safety outcomes.

The findings illustrate why embryo production and clinical success must be evaluated separately. In this low-reserve population, GnRH-antagonist stimulation recruited more follicles and yielded more usable and high-quality embryos, yet cumulative live birth remained statistically indistinguishable from the two less intensive approaches. The results therefore do not establish that the protocols are equivalent in every respect, nor do they show that additional embryos have no value. They show that the study detected no protocol-related difference in live birth or clinical pregnancy, with uncertainty still present in the estimates. The clinical-pregnancy models’ poor calibration, modest effective sample size and exploratory threshold analysis also caution against interpreting the null comparisons as proof of exact equivalence.

Several limitations constrain how far the results can be generalized. The study was retrospective, single-center and based on first IVF cycles, which reduced confounding from prior treatment outcomes but may limit broader applicability. The authors excluded several major causes of infertility, including severe male-factor infertility requiring intracytoplasmic sperm injection, untreated hydrosalpinx, moderate-to-severe endometriosis, major fibroids and other specified conditions. This made the population unusually focused on diminished ovarian reserve but also means the findings do not apply automatically to patients with additional infertility diagnoses. Preimplantation genetic testing was not routinely performed, so the chromosomal status of embryos was unknown, and endometrial factors affecting receptivity were not systematically assessed. Changes in laboratory and transfer practices over the 2017–2023 study period, the possibility of residual confounding, underpowering for small differences and the absence of systematic neonatal data also limit interpretation. Prospective multicenter studies would be needed to test the proposed threshold pattern and evaluate safety, costs and patient-centered outcomes.

Subject of Research: Comparative effectiveness of three ovarian-stimulation protocols in patients with diminished ovarian reserve

Article Title: Comparative effectiveness of GnRH antagonist, mild stimulation, and PPOS protocols in POSEIDON groups 3 and 4: a retrospective cohort study with entropy balancing

Article References: Zhao, H., Ni, Y., Wang, A., Liu, J., Lu, X., & Lu, S. (2026). Comparative effectiveness of GnRH antagonist, mild stimulation, and PPOS protocols in POSEIDON groups 3 and 4: a retrospective cohort study with entropy balancing. Frontiers in Endocrinology, 17, Article 1969038. https://doi.org/10.3389/fendo.2026.1969038

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1969038

Keywords: IVF, diminished ovarian reserve, POSEIDON groups 3 and 4, cumulative live birth, GnRH antagonist, mild stimulation, progestin-primed stimulation, entropy balancing, embryo yield, assisted reproduction, Comparative, effectiveness