Nomogram for endometrial cancer/atypical hyperplasia risk prediction.

Points assigned per predictor (top) yield total points (middle) corresponding to risk probability (bottom). Risk stratification: low (< 40), medium (40–70), high (> 70).

Model Validation

The nomogram demonstrated robust discrimination and calibration performance during validation. ROC analysis yielded AUCs of 0.82 (95% CI: 0.80–0.85) in the training cohort and 0.83 (95% CI: 0.73–0.93) in the validation cohort (both P < 0.05; Fig. 3a). Calibration plots showed excellent agreement between predicted and observed outcomes, with a calibration slope of 1.000 and intercept of − 0.177 (Fig. 3b). The model achieved a Brier score of 0.172, indicating low overall prediction error. Decision curve analysis further confirmed the model's clinical utility across threshold probabilities of 10% to 50%, particularly showing significantly greater standardized net benefit compared to the "intervene-all" or "intervene-none" strategies within the critical 20%-40% high-risk threshold range(Fig. 3c).

ROC curves: training (navy; AUC = 0.82), validation (red; AUC = 0.83);

Calibration: logistic fit (solid), nonparametric (dashed) vs. ideal (grey); Brier = 0.172. (c) Decision curve: model

Decision curve: model (thick solid) vs. 'All' (dashed) and 'None' (thin solid) strategies; net benefit superior at 10–50% thresholds (optimal 20–40%).

DISCUSSION

This study developed a nomogram demonstrating significant advantages in predicting AEH/EC risk among women who are over 40 years old and undergoing endometrial biopsy. Key findings identified Cumulative Menstrual Years (CMY > 40 years) as a strong independent risk factor (increasing risk by 7.28-fold) and confirmed the critical predictive value of Postmenopausal Bleeding (PMB) timing, particularly PMB occurring ≥ 5 years postmenopause (OR = 14.55). By menstrual irregularity, integrating CMY, BMI status, parity, diabetes status, AUB, stratified PMB, endometrial thickness, and duration of progesterone administration, the model achieved exceptional predictive performance (AUC 0.82–0.83), significantly outperforming existing models based on conventional indicators.

Regarding risk factors, the key findings of this study show both consensus and differences with previous research. Points of consensus include: CMY > 40 years, as a quantifiable indicator of cumulative estrogen exposure, significantly increasing AEH/EC risk (OR = 7.28) aligns with numerous epidemiological studies (24, 25); the strongest risk associated with late PMB (≥ 5 years) confirmed by stratified analysis (OR = 14.55) is consistent with clinical guidelines and the majority of academic viewpoints(35); significantly increased risk associated with thickened endometrium (≥ 13mm premenopause, ≥5mm postmenopause; OR = 2.92) also matches classical research and guideline recommendations(31, 32); and menstrual irregularity (OR = 3.93) as an independent risk factor likely reflects anovulation or endocrine dysfunction(36).

The primary difference lies in this study’s observed inverse association between diabetes and endometrial cancer risk with an odds ratio of 0.40, contrasting with most studies reporting positive associations. (37, 38) This paradox may stem from insufficient control of BMI confounding, as adjustment occurred only via dichotomization at the Asian standard of 23 kg/m², lacking fine stratification or continuous variable analysis; whereas literature identifies BMI as a key shared risk factor and potent confounder—evidenced by Luo et al(39).and Lucenteforte et al(40). showing attenuated associations after BMI adjustment. Other factors include surveillance bias, where increased gynecological examinations in diabetics may enable earlier precancerous intervention; metformin’s potential influence despite the WHI finding no protective effect with a hazard ratio of 1.00, while lab studies suggest antitumor mechanisms; and population heterogeneity, where differing Asian BMI thresholds of 23 kg/m² versus Western standards of ≥ 25 kg/m² may introduce bias. Thus, the true diabetes-endometrial cancer relationship requires validation through larger prospective studies such as WHI-style extensions; rigorous confounder control including BMI modeled as a continuous variable or finely stratified into six tiers, with waist-hip ratio inclusion; obesity-diabetes interaction assessment; and stratification by diabetes duration and treatment per the WHI’s time-dependent analysis showing residual risk in new-onset diabetes.

Compared to models reported in previous literature, the nomogram constructed in this study possesses significant advantages and unique value. The primary advantage is a markedly improved prediction accuracy (AUC 0.82–0.83), significantly better than prior models relying on conventional indicators (age, BMI, bleeding history; AUC 0.64–0.77). This is mainly attributed to the innovative integration of CMY as a quantifiable indicator of estrogen exposure and the critical stratification of PMB timing (< 5 years vs ≥ 5 years). Secondly, this model addresses the limitation of "dynamic risk assessment" in existing models. Existing models often use static stratification based on cross-sectional data, whereas this model, based on a retrospective cohort (≥ 2 biopsies, median follow-up 3 years), inherently allows its predictors (such as continuously accumulating CMY or newly occurring late PMB) to be suitable for dynamically monitoring changes in individual risk over time (e.g., CMY extension, new onset of late PMB). This enables assessment of risk evolution from benign/hyperplastic conditions to AEH/EC, thereby achieving more precise timing for follow-up and intervention. Thirdly, the model boasts high clinical practicality and applicability in primary care settings. All included variables (CMY, PMB timing, ET, menstrual regularity, diabetes status) can be easily obtained at the primary care level through history taking, basic physical examination, and transvaginal ultrasound. This overcomes the major obstacle faced by models relying on advanced imaging (e.g., MRI texture analysis) or expensive molecular markers, which are difficult to implement in resource-limited areas. Its intuitive risk score and stratification (low/medium/high risk) facilitate rapid clinical decision-making.

The limitations of this study include its single-center retrospective design with inherent selection bias from including only a biopsy cohort; a limited number of AEH/EC cases (n = 66), affecting precision despite synthetic oversampling; a highly homogeneous cohort (99.4% Han Chinese), reducing generalizability; incomplete adjustment for BMI (a key confounder in the diabetes-EC association); and lack of data on genetic syndromes, molecular markers, and diabetes-specific variables (such as type, duration, and treatment). Future directions include: external validation in multi-ethnic, geographically diverse cohorts; prospective evaluation of biopsy reduction rates and impact on early detection; model refinement through incorporating BMI as a continuous variable or finer strata (e.g., WHO Asian BMI categories); development of a digital calculator for point-of-care use; and larger studies to clarify diabetes’ independent role through rigorous confounder adjustment. The model is developed based on data from a Chinese population (99.4% Han Chinese cohort) and employs BMI cutoffs suitable for Asian populations. This partially addresses the issue of insufficient generalizability of existing mainstream models (mostly built on European and American population data) to the Chinese population, providing a practical tool for primary healthcare in China that is more aligned with local epidemiological characteristics. Its clear potential for clinical benefit lies in effectively identifying low-risk individuals (score < 40 points), potentially substantially reducing (> 90%) unnecessary invasive endometrial biopsies and their associated physical and psychological burdens and economic costs, while maintaining a high detection rate (sensitivity ≥ 80%) for AEH/EC. This ability to reduce unnecessary biopsies and optimize resource utilization in primary care and resource-limited settings lacking advanced diagnostic equipment is one of its most significant value propositions.

Conclusion

This study developed and validated a practical nomogram that significantly improves prediction of AEH/EC risk in women undergoing endometrial biopsy. Its superior performance (AUC 0.82–0.83) stems from integrating two key innovations: cumulative menstrual years (CMY > 40 years; OR = 7.28) as a novel measure of estrogen exposure, and stratified PMB timing (≥ 5 years postmenopause; OR = 14.55) as a critical high-risk indicator. Relying solely on accessible clinical/ultrasound parameters, the model enables dynamic risk monitoring and effectively identifies low-risk women (score < 40 points), while maintaining high sensitivity. Optimized for the Han Chinese population, this tool offers substantial clinical utility for resource-limited settings.Future work must: (1) validate the model in multi-ethnic cohorts, (2) refine BMI adjustment using WHO Asian classifications, and (3) prospectively quantify biopsy reduction rates and cost savings. Development of an open-access digital calculator is prioritized for point-of-care use.