Infertility is a major global health problem that threatens female reproductive health. Assisted reproductive technology (ART) is considered one of the most effective treatments for infertility (1). Ovarian hyperstimulation syndrome (OHSS) is a common and severe iatrogenic complication of in vitro fertilization (IVF) in which clinical symptoms are ovarian enlargement, abnormal capillary permeability, excessive estradiol, ascites, and pleural effusion (2). The incidence of mild OHSS is 20%–33%, and the incidence of moderate to severe OHSS is 3%–8% (3). In addition to causing miscarriage, prolonged time to pregnancy, and pregnancy complications, severe OHSS can even cause acute renal insufficiency, acute respiratory distress, venous thrombosis, and even death of patients, which brings both economic and psychological burdens to patients (4). The main risk factor for OHSS is high ovarian response (HOR) which is defined as an abnormal sensitivity to exogenous gonadotropin, resulting in multiple follicle recruitment, development, and steroid abnormalities (5, 6).

GnRH antagonist protocol is gaining popularity among fertility centers worldwide because of its simplified procedures, reduced drug dosages, shorter treatment cycles, safety, and efficacy (7). Several studies have shown that the GnRH antagonist protocol is beneficial for HOR patients, as it can significantly reduce the risk of OHSS (8, 9). However, OHSS can’t be avoided completely, so it is essential to predict HOR to reduced it whenever possible.

Age, anti-mullerian hormone (AMH), antral follicle count (AFC), and basal sex hormones are commonly used to predict ovarian response (10), but most studies predict the risk of HOR with only one or two factors. To our knowledge, only a small amount of literatures construct the multi-variable models without the calibration curve and the decision curve analysis. In our study, we identified the independent risk factors affecting HOR and constructed a nomogram to predict the occurrence of HOR based on the results of the multivariate logistic regression. In addition, we drew the calibration curve and the decision curve to assure the accuracy and utility of the model. We aimed to create a practical algorithm to assist clinicians in implementing personalized reproductive strategies.

Adopting individualized treatment strategies has been one of the most important topics in ART. The selection of an individualized controlled ovarian stimulation protocol is based on the ovarian response so that patients can obtain the appropriate number of follicles, embryos, and optimal pregnancy outcomes while avoiding severe adverse effects and complications, such as OHSS. Although the GnRH antagonist protocol significantly reduces the risk of OHSS, controlling and reducing it in HOR patients is a major challenge for clinicians. In our study, multivariate logistic regression analysis revealed that age, BMI, FSH, AMH, and AFC were independent risk factors for HOR in the GnRH antagonist protocol, which is consistent with the results of previous studies.

In previous studies, age and FSH often were used to assess ovarian response. Broer et al. conducted a meta-analysis showing that age was the strongest single predictor of high response (OR = 0.89, 95% CI: 0.85–0.93) (12). After the age of 35, the number of follicles decreases sharply, probably due to the incidence of chromosomal defects increases, and the sensitivity of granulosa cells to gonadotropin decreases. Some studies suggested that serum FSH increases significantly when ovarian reserve is severely declined and fluctuates significantly during the menstrual cycle; therefore, FSH has no optimal sensitivity or specificity in predicting ovarian response (16). Our data indicated that FSH was associated with ovarian response, but it shouldn’t be single-used as the predictor due to its defects.

AFC and AMH have recently become the most widely used markers for assessing ovarian reserve and ovarian response in ART. AFC is the most intuitive indicator of ovarian reserve function in patients, as it reflects the number of ovarian follicles that can be stimulated by exogenous gonadotropin in a given cycle. A review reported that AFC > 14 was a predictor of HOR (17). Another study demonstrated that the optimal threshold for predicting HOR was AFC > 16 or 18 (sensitivity: 82%, specificity: 80%) (10). However, AFC requires careful counting by skilled ultrasound operators. Of all the ovarian markers, AMH showed the best ability to predict ovarian response. AMH is mainly secreted by granulosa cells of the ovary and regulates folliculogenesis. Serum AMH reflects the number of antral follicles in the ovary, and its concentration is not affected by the menstrual cycle. Rong Li et al. demonstrated that the optimal AMH cut-off value predicting high ovarian response was 2.6 ng/mL (sensitivity: 81.28%, specificity: 59.51%) (18). Another study with 373 cycles suggested that an AMH cut-off value greater than 4.385 ng/mL may predict HOR, and the AUC was 0.845 (95% CI: 0.778–0.912) (19). Studies involving AMH have reported that the optimal threshold for predicting HOR is AMH > 3.18, 3.50, or 4.5 ng/mL (10, 20, 21). Although AMH is a reliable biomarker of ovarian response, there is no standard cut-off value because of differences in reagent kits and assays.

High BMI is associated with impaired ovarian response and has a negative effect on IVF outcomes (22). Qiu et al. noted that high BMI in PCOS patients had a negative effect on ovarian response (23). In our study, low BMI is associated with the development of HOR. BMI ≥ 28 kg/m2 is defined as obesity in China. Obesity alters the function of the hypothalamic-pituitary axis, which is associated with high levels of insulin, androgen, and estrogen and is involved in the impairment of ovarian folliculogenesis. Although obese women receive higher doses of gonadotropin and longer stimulation periods than normal weight women, they may still have poor ovarian response because of decreased drug bioavailability (24). Chalumeau et al. developed a predictive model of ovarian response, including BMI and other factors, that could explain 60% of the variance in ovarian response to stimulation (25).

Recent studies have used multi-variable models which are more credible and stable. A meta-analysis with complete available data from 1023 patients revealed that FSH, AFC, and AMH could all be combined with age for HOR prediction, where age+AMH/AFC was more accurate compared with age alone for prediction, and age+FSH exhibited a smaller increase in accuracy. The combination of age+AFC+AMH (AUC=0.85) demonstrated good accuracy for predicting HOR (12). Tan et al. reported that AFC, AMH, and P levels on the human chorionic gonadotropin (HCG) day were identified as independent predictors of HOR. The nomogram was established with the data of 480 eligible patients (26). In our study, greater number of cycles were included and a combined prediction model with age, BMI, FSH, AMH, and AFC was developed with good performance. Simultaneously, we plotted a nomogram to visualize our model. The AUC value of the combined prediction model reached 0.884, indicating the excellent discrimination of the model, and the validation confirmed the accuracy and feasibility of the model.

In the present study, risk factors of HOR were investigated and a well-calibrated prediction model was successfully proposed in patients undergoing the GnRH antagonist protocol. Our model possesses the following strengths: The variables enrolled in the nomogram can be easily measured, making it convenient to use in clinical practice. Moreover, the model was validated to ensure its reproducibility in a wider population. It was not only the discrimination and calibration but also the utility that evaluated the performance of clinical prediction models in this study. The utility of models can be assessed by Decision curve analysis (DCA), which plots net benefit (NB) at a range of clinically reasonable risk thresholds. We constructed decision models for the groups respectively, of which the results demonstrated favourable net benefits. In IVF cycles, the use of the GnRH antagonist protocol, coasting protocol, low initial gonadotropin dose, aspirin, calcium, the GnRH agonist trigger, and the whole embryo freezing strategy should be considered for patients with HOR to reduce the likelihood of OHSS, venous thrombosis, and other hazards in patients (27).

Our study also has some limitations. Environmental exposure, genetic elements, and unhealthy lifestyle habits also affect HOR, which were not included in the study because of the limited data sources. In the future, prospective, large-scale, multi-center clinical studies should be conducted to develop a more systematic and comprehensive clinical prediction model.

In conclusion, we aimed at infertile women using GnRH antagonist protocol in ART and successfully developed a prediction model that enables us to predict HOR patients in a simple, effective and visual way. We hope the model can help clinicians select personalized treatments to improve ART outcomes.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by the Ethics Committee of Reproductive Medicine of Henan Provincial People’s Hospital with the number SYSZ-LL-2021091501. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and institutional requirements.

YJ conceived study design. TW and CC performed the data collection and statistical analysis. JG edited the manuscript. CZ was responsible for providing data and guiding study. All authors contributed to the article and approved the submitted version.