Predictors of live birth following fertility-sparing treatment in patients with atypical endometrial hyperplasia or early endometrial carcinoma

Introduction

Endometrial carcinoma (EC) ranks among the most common gynaecological malignancies globally. Approximately 7% of EC cases are diagnosed in women under 45 years of age, and epidemiological trends indicate a consistent rise in incidence within this population.1 Atypical endometrial hyperplasia (AEH), recognised as a premalignant condition, progresses to EC in approximately 29% of cases over several years.2 Given that up to 70% of premenopausal patients with endometrial cancer are nulliparous, fertility preservation represents a key clinical consideration.3

In recent years, high-dose progestin regimens have been increasingly recognised as safe and effective for treating AEH and EC.4 Combined with comprehensive metabolic management (lipid reduction, glycaemic control, blood pressure regulation and weight optimisation), this conservative therapy aims to induce endometrial lesion regression, thereby creating a favourable window for conception and fertility preservation,5 6 with contemporary clinical practice emphasising dual goals of optimal oncologic control and favourable reproductive outcomes.

Patients undergoing fertility-sparing treatment face unique challenges, including underlying endometrial pathology, prolonged therapeutic progesterone exposure, repeated intrauterine interventions (eg, diagnostic dilation and curettage with or without hysteroscopy) for disease surveillance or intrauterine adhesion assessment7 and frequent comorbidities (polycystic ovarian syndrome, obesity, metabolic disorders) that may impair endometrial receptivity and pregnancy maintenance. Although pregnancy outcomes after fertility-preserving interventions in EC and AEH have been reported,8 9 literature on determinants of successful live birth following complete remission (CR) remains limited, warranting further investigation.

This study aimed to evaluate live birth rates and identify factors associated with pregnancy outcomes in young AEH/EC patients who received fertility-sparing treatment. A predictive model was constructed by integrating demographic, clinical and therapeutic parameters to clarify perinatal factors influencing reproductive success, with the ultimate goal of safeguarding maternal fertility and improving pregnancy outcomes.

Methods

Results

Establishment and validation of a predictive model for live birth

This nomogram was constructed to assess the predictive utility of selected clinical variables in forecasting pregnancy outcomes. Based on initial univariate analysis identifying significant differences between live birth and pregnancy failure groups (table 2), four variables were incorporated into the nomogram (figure 1A): number of hysteroscopies, duration of oral progesterone therapy, presence of IR and administration of GnRH-a. ROC analysis demonstrated favourable discriminative capacity of the overall model, with an AUC of 0.780 (95% CI 0.656 to 0.903, p<0.001) (figure 1B). To further assess the stability of the model, bootstrap internal validation with 2000 resampling iterations was performed, yielding a bias-corrected AUC of 0.750 (95% CI 0.632 to 0.872) and a mean optimism of only 0.030, indicating minimal overfitting (online supplemental figure 3).

Figure 1

Request permissions

Establishment and predictive efficacy of disease correlation model. (A) The nomogram shows the times of hysteroscope, progesterone duration, IR and GnRH-a. (B) ROC analysis with an AUC of 0.780 (95% CI 0.656 to 0.903, p<0.001). (C) Hosmer-Lemeshow test p=0.098 (>0.05). (D) Decision curve analysis (DCA), with the red curve (Model) representing the net benefit predicted. AUC, area under the curve; GnRH-a, gonadotropin-releasing hormone agonist; IR, insulin resistance; ROC, receiver operating characteristic.

In the univariate analysis (online supplemental figure 2), the duration of oral progesterone showed relatively better predictive efficacy (AUC 0.695, 95% CI 0.570 to 0.821, p=0.006), followed by hysteroscopy frequency (AUC 0.673, 95% CI 0.539 to 0.809, p=0.015). GnRH-a usage demonstrated borderline significance (AUC 0.631, 95% CI 0.491 to 0.770, p=0.067), while IR showed limited predictive value (AUC 0.561, 95% CI 0.415 to 0.707, p=0.395). Calibration analysis using the Hosmer-Lemeshow test indicated good agreement between predicted and observed outcomes (p=0.098), as illustrated in figure 1C. DCA further demonstrated the clinical utility of the model, as shown in figure 1D.

Discussion

Fertility preservation for patients with AEH and EC has become an increasingly important focus in gynaecologic oncology.10 11 For the majority of young patients undergoing fertility-preserving treatments, the primary objective is to achieve the possibility of a successful live birth. Therefore, the optimisation of diagnostic and therapeutic strategies for pregnancy-related conditions, alongside comprehensive perinatal management, is of critical importance. Our previous research has explored determinants influencing pregnancy rates following fertility-preserving therapy.12 Building on extended follow-up data and a growing number of live births, the present study systematically examines potential factors associated with successful delivery in AEH and EC patients. Furthermore, we developed a predictive model based on key perinatal variables to elucidate their relative contributions to reproductive outcomes.

The study cohort comprised 89 patients with complete follow-up data and comparable baseline characteristics. Among the participants, 40 of 50 AEH patients (80.0%) and 27 of 39 EC patients (69.2%) achieved live birth after CR. These findings suggest that enhanced understanding of EC and AEH pathogenesis and treatment modalities may progressively improve post-preservation live birth outcomes in clinical practice.13–15 In this cohort, increased hysteroscopy frequency and prolonged progesterone therapy were associated with pregnancy failure, while concurrent IR and GnRH agonist use were also negatively associated with live birth outcomes. ROC curve analysis of selected clinical variables confirmed their prognostic relevance within our predictive model. The multivariate model demonstrated superior discriminative performance compared with individual predictors, underscoring the robustness of the selected variables. Model calibration, evaluated through the Hosmer-Lemeshow test and calibration plots, showed strong alignment between predicted probabilities and observed outcomes. DCA further supported the model’s clinical applicability by illustrating a favourable net benefit across a wide range of threshold probabilities.

The average duration of oral progestin use was 18 months in the live birth group, compared with 26.13 months in the pregnancy failure group, indicating that longer progestin exposure is correlated with poorer reproductive outcomes. While progesterone administration during the perinatal period has been shown to support pregnancy maintenance—potentially via promoting endometrial decidualisation, suppressing uterine contractility, and modulating immune responses.16 17 However, in fertility-preserving therapy, high-dose progestin is primarily to induce and sustain CR, thereby enabling conception.18 Some studies have further indicated that prolonged progestin administration (>9 months) does not negatively impact pregnancy success rates.19 Nevertheless, our findings reveal a negative association between oral progestin duration and live birth outcomes. This may be explained by two key considerations. First, a longer treatment duration may reflect delayed or suboptimal CR, indicating a more refractory disease course requiring extended progesterone therapy. Second, prolonged high-dose progestin exposure may induce adverse histopathological changes in the endometrium, impairing its receptivity. Existing evidence suggests that high-dose progestin may contribute to the development of progesterone resistance20 and decreased microvessel density,21 ultimately compromising implantation and pregnancy maintenance. Based on these findings, we recommend that fertility-preserving patients with pregnancy plans consider attempting conception as soon as possible after CR. It is necessary to use ART if possible, because ART may potentially improve pregnancy outcomes in patients with AEH and EC after CR.22 So, our future prospective study will prospectively record complete ART history to enable precise confounding adjustment.

Hysteroscopy serves as a crucial tool for evaluating endometrial pathology and administering localised treatment. In this study, data from AEH and EC patients undergoing fertility-preserving treatment revealed that the failure group underwent significantly more hysteroscopies (six times) compared with the success group (four times), suggesting a strong correlation between increased hysteroscopy frequency and a higher risk of pregnancy failure. Previous literature suggests that each hysteroscopic procedure may contribute to endometrial injury, which could affect implantation potential and early placental blood supply.23 The mechanism may involve mechanical damage that reduces endometrial receptivity,24 which constitutes a major focus of our team’s future investigations (figure 2). Our research suggests that five or more hysteroscopies may be closely associated with a decrease in the live birth rate, and similar literature has reported that reaching six hysteroscopes is associated with subsequent placenta accreta.25 Thus, the number of hysteroscopic procedures should be strictly controlled. Reducing unnecessary hysteroscopies can effectively lower the risk of endometrial damage, thereby enhancing clinical pregnancy rates. We recommend adopting more precise diagnostic methods, such as targeted hysteroscopic biopsies (already a routine practice in our institution’s fertility-preserving treatments), to minimise endometrial disruption. Furthermore, it is advisable to collect endometrial tissue samples via hysteroscopy only when necessary, typically at the end of each 12-week treatment cycle. Optimising hysteroscopy strategies with a focus on endometrial protection is a key measure to improve clinical pregnancy success rates in AEH and EC patients following fertility-preserving treatment.

In the present study, several well-established metabolic and endocrine risk factors—including obesity, PCOS and diabetes mellitus—did not demonstrate statistically significant associations with live birth outcomes.26–29 This finding contrasts with previous studies that have reported adverse effects of obesity and metabolic disorders on endometrial receptivity and pregnancy maintenance in the general population.26 30 31 Several factors may account for these null findings. First, our relatively small sample size (n=89) may have limited the statistical power to detect modest effect sizes, particularly given the low prevalence of certain comorbidities within our cohort. Second, the unique characteristics of the fertility-sparing patient population—comprising young women who have already achieved CR and demonstrated treatment responsiveness—may attenuate the impact of metabolic factors on reproductive outcomes.32 33 Third, all patients in this cohort received standardised comprehensive metabolic management (including lipid reduction, glycaemic control and weight optimisation) as part of the fertility-sparing protocol, which may have mitigated the potential adverse effects of these comorbidities. Future studies with larger, multicentre cohorts are warranted to further elucidate the role of metabolic and endocrine factors in live birth outcomes following fertility-sparing treatment.

Our study has several limitations. First, it is a retrospective analysis, which carries inherent retrospective bias. The conventional multivariable logistic regression model was established based on a small cohort with only 22 pregnancy failure events, which carries inherent risks of overfitting and unstable adjusted effect sizes. Second, the low prevalence of IR led to sparse cell counts and wide CIs, reducing estimation precision. Although we performed LASSO sensitivity analysis and bootstrap internal validation to confirm model stability and correct such bias (online supplemental figures 4, 5), our findings still require external validation in large independent cohorts for broader generalisability. In future research, we will expand the sample size and collect placental tissues from pregnant and postpartum patients to further explore the underlying mechanisms. Second, several potential confounding factors were not included in the analysis, such as embryo quality and endometrial thickness at embryo implantation. Third, unmeasured socioeconomic and lifestyle factors may also have influenced pregnancy outcomes. Additionally, as live birth was a relatively common outcome in the present study, OR might overestimate the magnitude of the associations. Nevertheless, our team has accumulated rich experience in investigating the mechanisms of endometrial cancer34 and in managing pregnancies after fertility-preserving treatment.35–38 We will further integrate clinical experience and research data to optimise diagnosis, treatment and postpartum management for fertility-sparing patients, and provide evidence-based references for clinical practice.

Conclusions

In conclusion, this study followed patients who attained CR after fertility-sparing therapy and categorised them based on subsequent live birth outcomes. Our results revealed that duration of oral progestin treatment and frequency of hysteroscopic procedures are independently associated with live birth outcomes. Moreover, the predictive model demonstrated acceptable discrimination and may provide clinical reference for live birth prediction among patients receiving fertility-sparing management, pending external validation.