New Prediction Tool Flags IVF Patients at Risk of Low Egg Retrieval

One of the most disheartening moments in fertility treatment arrives before it truly begins: a young woman undergoes weeks of hormonal stimulation, only for clinicians to retrieve fewer eggs than hoped at the end of it. Now, a large multicenter study from China suggests that this outcome can be predicted with surprising accuracy before a single injection is given. Researchers report that a simple scoring tool, built from five routinely measured clinical parameters, reliably identifies women aged 35 and younger who are likely to experience low oocyte retrieval during their first cycle of in vitro fertilization or intracytoplasmic sperm injection.

The study, published in the Journal of Ovarian Research, is notable less for introducing a new model than for testing one rigorously. Prediction models in reproductive medicine are abundant, but very few are subjected to what statisticians call external validation, the process of confirming that a tool works not just in the population where it was developed but in entirely new groups of patients treated at different hospitals. Without such validation, a model’s apparent accuracy may be nothing more than a statistical artifact of its original dataset, a phenomenon researchers describe as overfitting.

To address this, a team led by Chang Liu, Huan Zhang and Yili Teng of the Reproductive Medicine Center at the First Affiliated Hospital of Wenzhou Medical University assembled an exceptionally large cohort: 14,702 women aged 35 or younger undergoing their first IVF or ICSI cycle between January 2023 and December 2025. The data came from six reproductive centers across several Chinese provinces, including sites in Shandong, Zhejiang, Guangxi and elsewhere. The researchers validated the existing nomogram, a graphic scoring instrument that converts patient characteristics into a probability, using two complementary strategies. Temporal validation tested the model on later patients at the original development center, while geographic validation tested it across five independent hospitals.

All of the predictors in the model are collected before ovarian stimulation begins, which is precisely what makes the tool clinically actionable. They include the woman’s age, her anti-Müllerian hormone level, her antral follicle count as seen on ultrasound, her baseline follicle-stimulating hormone level, and the ratio of follicle-stimulating hormone to luteinizing hormone. Each of these measures reflects a different facet of what reproductive biologists call the ovarian reserve, the pool of remaining follicles in the ovaries. Anti-Müllerian hormone, secreted by small growing follicles, serves as a chemical census of that pool. The antral follicle count provides a direct visual count of recruitable follicles, while FSH and its ratio to LH capture how hard the pituitary gland must work to drive the ovaries, a signal that rises as ovarian responsiveness declines.

The scale of the problem the model addresses is underscored by the study’s headline finding: across all six centers, the rate of low oocyte retrieval among these young women reached 47.2 percent. In other words, nearly one in two patients who might assume that youth guarantees a bountiful egg harvest in fact ended their first stimulation cycle with a disappointing yield. Low oocyte retrieval matters because the number of eggs retrieved is tightly linked to the number of embryos available for transfer, the chances of success per cycle, and the emotional and financial burden placed on patients who must repeat stimulation or accumulate eggs over multiple attempts.

The validation results were strikingly consistent. Across the six centers, the model’s area under the receiver operating characteristic curve, a standard measure of discrimination that reflects how well a tool separates those who will experience an outcome from those who will not, ranged from 0.800 to 0.888, with an overall value of 0.836 and a 95 percent confidence interval of 0.829 to 0.842. In practical terms, an AUC above 0.8 is generally regarded as good discrimination, meaning that a randomly chosen patient who experiences low oocyte retrieval receives a higher risk score than a randomly chosen patient who does not roughly four times out of five. That level of performance, sustained across hospitals with different patient populations and laboratory practices, is the strongest evidence yet that the underlying biology captured by the five parameters is genuinely generalizable.

Discrimination alone, however, is not enough for a clinical tool. A model must also be well calibrated, meaning that when it predicts, for example, a 30 percent risk of low retrieval, roughly three in ten comparable patients actually experience it. The researchers assessed this using calibration plots and the mean absolute error, a metric that quantifies the average gap between predicted and observed probabilities. The mean absolute error values ranged from just 0.004 to 0.040 across centers, indicating that the nomogram’s predicted probabilities track observed outcomes closely. Such calibration is what allows clinicians to communicate a risk estimate to a patient and trust that the number carries real meaning rather than systematic distortion.

The team also evaluated clinical usefulness through decision curve analysis, a technique that weighs the benefits of correctly identifying high-risk patients against the costs of unnecessary interventions in low-risk ones. Compared with the extreme strategies of treating everyone or treating no one, the nomogram delivered a higher net benefit across a broad range of threshold probabilities, the risk levels at which a clinician and patient would choose to act. The researchers found no statistically significant association between center-level characteristics and model discrimination, though they noted that variation in AUC across sites may still reflect differences in patient characteristics, a reminder that even validated tools perform within the texture of local populations.

For patients and clinicians, the practical implications are considerable. A woman identified as high risk before stimulation could, in principle, be counseled more realistically about her first cycle, offered an individualized stimulation protocol with adjusted gonadotropin dosing, or prepared for the possibility of a dual-ovulation or multiple-cycle approach to accumulate eggs. Low oocyte retrieval in young patients can signal an occult diminished ovarian reserve or unexpected hyporesponsiveness that standard screening sometimes misses, and early identification turns a cycle-ending surprise into a manageable planning problem. Because every input to the nomogram is already part of routine pretreatment workup in IVF clinics worldwide, implementing the tool requires no new tests, only a calculation.

The study was approved by the institutional review board of the First Affiliated Hospital of Wenzhou Medical University and conducted in accordance with the Declaration of Helsinki, with informed consent waived given its retrospective design. Funded by the Wenzhou Municipal Science and Technology Bureau, the work arrives amid a broader movement in reproductive medicine toward pretreatment risk stratification, in which statistical and machine-learning models compress thousands of patient records into instruments usable at the bedside. This nomogram, now tested across six hospitals and nearly fifteen thousand patients, offers one of the clearest demonstrations that such tools can travel well beyond their birthplace. For the roughly half of young IVF patients who would otherwise be blindsided by a poor egg retrieval, that portability may prove the study’s most valuable result, turning a moment of disappointment into an opportunity for foresight.

Subject of Research: External validation of a pretreatment prediction nomogram for low oocyte retrieval in young women undergoing IVF/ICSI

Article Title: Prediction of low oocyte retrieval in young women undergoing IVF/ICSI: a multicenter external validation study

Article References: Liu, C., Zhang, H., Yang, H., Sun, Z., Zhu, Q., Zhang, S., Lv, Y., Fu, X., Dai, C., & Teng, Y. (2026). Prediction of low oocyte retrieval in young women undergoing IVF/ICSI: a multicenter external validation study. Journal of Ovarian Research. https://doi.org/10.1186/s13048-026-02266-9

Image Credits: AI Generated

DOI: 10.1186/s13048-026-02266-9

Keywords: IVF, ICSI, low oocyte retrieval, ovarian reserve, anti-Müllerian hormone, antral follicle count, nomogram, predictive model, multicenter validation, reproductive medicine, fertility treatment, ovarian stimulation