Development and Internal Validation of a Novel Risk Prediction Score for Endometrial Malignancy in Premenopausal Women with Abnormal Uterine Bleeding

1. Introduction

Abnormal uterine bleeding (AUB) is one of the most frequent reasons for consultation in gynaecology.¹ It can be caused by structural or non-structural disorders of the uterus. According to the PALM-COEIN classification system of the International Federation of Gynecology and Obstetrics (FIGO), the causes include polyps, adenomyosis, leiomyomas, malignancy and hyperplasia, coagulopathy, ovulatory dysfunction, endometrial disorders, and iatrogenic or not yet classified causes.²

Although most cases of AUB are unrelated to pre-cancerous or cancerous endometrial pathologies, their seriousness should not be underestimated.³ It is established that in postmenopausal women with AUB, the risk of endometrial cancer rises to 10%.^4,5 However, this risk drops to less than 1% when transvaginal ultrasound shows an endometrial thickness (ET) of less than 4 mm.⁶

For premenopausal women, such risk stratification is difficult, as the predictive value of ET assessment has yielded controversial results in the literature.^7–10 In this group, other clinical factors are considered to assess the risk of endometrial hyperplasia (EH) or cancer: obesity, nulliparity, age, infertility, intermenstrual bleeding, anovulation, and diabetes.¹¹

Based on these factors, guidelines recommend endometrial biopsy for women over 40 years, and for those under 40 with comorbidities.^12,13

Despite these recommendations, many studies have not provided conclusive results regarding the impact of the mentioned risk factors.^6,14,15 Consequently, the optimal management of premenopausal patients suffering from AUB remains unclear, hence the interest of our work to develop a predictive model based on clinical variables to assess the risk of pre-cancerous or cancerous endometrial pathologies in cases of AUB in premenopausal women.

2. Methods

2.2 Study population

Premenopausal women presenting with AUB who were referred to our department were identified. An endometrial biopsy was performed during a diagnostic hysteroscopy or a fractional curettage and hemostasis procedure in cases of heavy AUB. All these women subsequently underwent surgical treatment (hysterectomy).

AUB was defined by the presence of bleeding from the uterine corpus that was abnormal in volume, regularity, and/or timing, according to what was reported by women.²

Patients meeting the following criteria were included:

Inclusion criteria:

• - Endometrial pathology revealed by endometrial biopsy: atypical endometrial hyperplasia (AEH), endometrial cancer … with a definitive histological diagnosis on the hysterectomy specimen (considered the reference standard).

• - Complete and available medical and surgical records.

Exclusion criteria:

• - Postmenopausal women (absence of menstruation for at least 12 months after the age of 40⁵).

• - Hysterectomy not performed (No definitive histological diagnosis).

• - Incomplete or missing data.

A study flowchart detailing case selection and exclusions has been developed ( Figure 1).

Figure 1. Flowchart for the study.

3. Results

During the study period, a total of 209 patients were included; 13 cases (6.22%) had endometrial malignancy.

3.3 Receiver operating characteristic (ROC) curve analysis

The predictive performance of several clinical and ultrasound variables for endometrial malignancy was evaluated using ROC analysis: ( Figure 2) ( Table 6).

• ▪ Endometrial thickness demonstrated the highest discriminative ability, with an AUC of 0.842 (p < 0.001). A thickness >9 mm was identified as the optimal threshold, yielding a sensitivity of 69%, specificity of 87%, a positive predictive value of 26.5%, and a negative predictive value of 97.7%. This indicates that increased endometrial thickness is a strong predictor of malignancy, particularly useful for ruling out disease when the measurement is below the threshold.

• ▪ Gravidity showed a moderate predictive value with an AUC of 0.690 (p = 0.022). A gravidity of ≤1 was associated with high specificity (96%) but low sensitivity (23%), suggesting that low gravidity may indicate increased risk of malignancy, though many cases could be missed if this criterion is used alone.

• ▪ The number of hemorrhage episodes had limited predictive value, with an AUC of 0.576 (p = 0.360). Using ≤2 episodes as a threshold, the sensitivity was 62% and specificity 69%, indicating that this variable alone is a weak predictor of malignancy.

Figure 2. Predictive performance of gravidity, number of hemorrhage episodes and endometrial thickness for endometrial malignancy evaluated using ROC analysis.

Table 6. Predictive performance of gravidity, number of hemorrhage episodes and endometrial thickness for endometrial malignancy was evaluated using ROC analysis.

Variable	AUC	p-value	95% CI	Criterion	Sensitivity (%)	Specificity (%)	PPV* (%)	NPV** (%)
Gravidity	0.690	0.022	0.542–0.838	≤1	23.08	96.43	30	95
Number of hemorrhage episodes	0.576	0.360	0.398–0.754	≤2	61.54	68.88	11.6	96.4
Endometrial thickness	0.842	<0.001	0.715–0.968	>9	69.23	87.11	26.5	97.7

* PPV: Predictive positive value.

** NPV: Negative predictive value.

Overall, endometrial thickness is the most significant individual predictor of malignancy, while low gravidity adds specificity but is limited by poor sensitivity. The number of hemorrhage episodes does not reliably predict malignancy.

3.5 Predictive score for endometrial malignancy

A risk score for predicting endometrial malignancy was developed using the variables found to be statistically significant in multivariate analysis.

The included predictors were:

• - Endometrial thickness greater than 9 mm,

• - Oral contraceptive use,

• - Presence of vascularization,

• - Low number of hemorrhage episodes (≤1),

• - And lower bleeding abundance.

Each predictor was weighted according to its odds ratio:

• - Endometrial thickness >9 mm and oral contraceptive use were assigned 3 points each,

• - Vascularization was assigned 4 points,

• - while a low number of hemorrhage episodes and lower bleeding abundance were considered protective and assigned −2 and −1 points, respectively.

The total score allows stratification of patients into risk categories: scores of 7 or higher indicate a high risk of malignancy.

The predictive score for endometrial malignancy demonstrated excellent discriminative ability ( Figure 3):

• - The area under the ROC curve (AUC) was 0.901 (95% CI: 0.825–0.976, p < 0.001), indicating high accuracy in distinguishing malignant from benign cases.

• - Using a threshold of >7, the score showed a sensitivity of 77% and a specificity of 90%, with a positive predictive value of 34% and a negative predictive value of 98%, suggesting strong effectiveness in ruling out malignancy among low-risk patients.

Figure 3. ROC curve associated with the prediction model of endometrial cancer in premonopausal women with abnormal uterine bleeding.

4. Discussion

The evaluation of women presenting with AUB to exclude endometrial malignancy relies on a combination of clinical risk assessment, imaging, and histopathological sampling.

The presented results, when compared to the broader literature, reveal a consistent emphasis on endometrial thickness as a pivotal triage tool. They also highlight the development and variable utility of integrated risk prediction models.

Evidence consistently highlights the crucial role of transvaginal ultrasound–derived endometrial thickness, which demonstrates robust diagnostic accuracy in predicting endometrial malignancy across studies.

Endometrial thickness among postmenopausal women with AUB below 4 mm seems to be associated with a very low risk of endometrial cancer.¹⁹ Unfortunately, there is no established consensus on the threshold for endometrial thickness for premenopausal women. In accordance with the results of other studies, our data indicated that patients with thicker endometrium exhibited a higher risk of endometrial malignancy.²⁰

Our findings, which identified endometrial thickness >9 mm as an independent risk factor for AUB in perimenopausal women, are consistent with previously published evidence. Tian et al.²¹ similarly reported that an endometrial thickness ≥10 mm was an independent risk factor for AUB, while Sahu et al.²² observed that the majority of perimenopausal women with AUB had an endometrial thickness of 10–12 mm (35.7%), followed by 7–9 mm (27.1%).

In contrast, Getpook et al.²³ demonstrated that an endometrial thickness ≤8 mm was unlikely to be associated with malignant pathology in premenopausal AUB.

Collectively, these concordant findings reinforce the clinical significance of transvaginal ultrasonography in assessing endometrial thickness as a key predictor of endometrial pathology in premenopausal women with AUB.

This principle holds strong in clinical guidelines, confirming endometrial thickness as the most robust initial screening parameter.

Several studies have developed integrated risk prediction models to address the limitations of using endometrial thickness alone.

Giannella et al.³ proposed a model including BMI ≥30 kg/m², diabetes, and endometrial thickness >11 mm, which achieved robust accuracy (AUC 0.854; sensitivity 75.0%, specificity 90.8%; PPV 30.0%, NPV 98.6%).

Similarly, the PAD30 score developed by Bagepalli Srinivas et al.,²⁴ based on anovulatory bleeding pattern, age ≥45 years, BMI ≥30 kg/m², and diabetes mellitus, showed good diagnostic performance (AUC 0.84; sensitivity 85.7%, specificity 87.6%).

Ruan et al.²⁰ incorporated metabolic diseases, family history, age ≥40 years, resistance index of endometrial vasculature ≤0.5, and endometrial thickness ≥10 mm into a nomogram, which demonstrated good discrimination with an AUC 0.837 (95% CI 0.800-0.874) and calibration in both the development and validation cohorts.

Compared with these models, our predictive score demonstrated even stronger discriminative ability, with an AUC of 0.901 (95% CI 0.825–0.976, p < 0.001). Using a cutoff >7, the model achieved a sensitivity of 77% and specificity of 90%, with a particularly high negative predictive value of 98%, supporting its effectiveness in reliably excluding malignancy in low-risk patients. These results suggest that our score may provide an accurate and clinically useful tool, comparable or superior to previously published models, for guiding the selective use of invasive diagnostic procedures.

The evaluation of patient-specific risk factors shows both consistency and variation.

The divergence underscores that while certain risk factors are epidemiologically important, their utility in a specific predictive algorithm can vary based on the study population and the other variables in the model.

In our study, oral contraceptive use emerged as an independent predictor of endometrial malignancy (OR 29.87, 95% CI 1.52–587.08, p = 0.025). This result contrasts with extensive evidence from large-scale epidemiological studies and meta-analyses, which consistently report a protective effect of oral contraceptives against endometrial cancer. The Collaborative Group on Epidemiological Studies on Endometrial Cancer, through an individual participant meta-analysis including over 27,000 women with endometrial cancer across 36 studies, demonstrated a substantial and sustained risk reduction associated with oral contraceptive use.²⁵ Similarly, Michels et al.²⁶ and Harajka et al.²⁷ confirmed these findings in systematic reviews and meta-analyses, while Karlsson et al.²⁸ further highlighted the time-dependent protective effect, with longer duration of use conferring greater benefit.

The discrepancy with our findings may be explained by the short duration of oral contraceptive use among women in our cohort, which might have been insufficient to exert a protective effect. Additionally, the small sample size and wide confidence interval suggest caution in interpretation. Nevertheless, this unexpected association highlights the importance of considering duration and patterns of oral contraceptive exposure when assessing their relationship with endometrial cancer risk.

Our findings demonstrated that vascularization on transvaginal ultrasound was found to be one of the strongest predictors of endometrial malignancy. Logistic regression analysis showed that the presence of abnormal vascularization significantly increased the risk, with an odds ratio of 98.34 (95% CI: 3.73–2594.79, p = 0.006). Notably, patients with a vascular score of 2 had a malignancy rate of 71%, whereas normal vascularization or a score of 1 was associated with substantially lower rates.

These findings are in line with recent evidence supporting the role of Doppler ultrasound vascular scoring in differentiating benign from malignant endometrial lesions. Tirnovanu et al.²⁹ reported that a vascular score of 1 typically excludes endometrial cancer, with high sensitivity (87.5%) and specificity (79%). Conversely, a cutoff score of 2 provided excellent discriminative performance, yielding 100% sensitivity and 86.3% specificity, which reflects the increased neovascularization commonly observed in malignant tumors. Such results are also consistent with the International Endometrial Tumor Analysis (IETA) consensus,³⁰ which emphasizes abnormal vascular patterns as a key ultrasound feature suggestive of malignancy.

Taken together, our findings reinforce the clinical value of color Doppler vascular assessment in women with abnormal uterine bleeding. The strong association between vascular score ≥2 and malignancy suggests that integrating vascularization into risk prediction models may substantially improve diagnostic accuracy. This parameter could therefore serve as a non-invasive adjunct to guide clinical decision-making and reduce unnecessary invasive procedures.

In our cohort, two clinical characteristics of bleeding were independently protective: having ≤1 hemorrhagic episode (OR 0.20, 95% CI 0.08–0.52, p = 0.001) and lower bleeding abundance (OR 0.30, 95% CI 0.13–0.65, p = 0.002). Clinically, this pattern is plausible: malignant endometrial lesions often generate recurrent and profuse bleeding due to fragile neovascularization and disordered repair, whereas isolated or low-volume events are more typical of benign etiologies (e.g., anovulatory dysfunction, polyps, simple hyperplasia).

Most of the available studies have not addressed the predictive value of bleeding frequency or volume. Specifically, several key works—including Giannella et al.,³ Timmermans et al.,¹⁹ Sahu et al.,²² and Getpook et al.²³—focused primarily on endometrial thickness or metabolic and clinical factors, without evaluating bleeding characteristics as independent predictors of malignancy. Similarly, Tian et al.²¹ did not report hemorrhage count or abundance in relation to cancer risk. Bleeding burden was not included as a variable into the nomogram of Ruan et al.²⁰

The closest conceptual overlap is PAD30’s anovulatory pattern,²⁴ which acknowledges that bleeding characteristics carry predictive signal, indirectly supporting our observation that clinical bleeding features can refine risk.

Our findings highlight the potential value of bleeding characteristics as predictive markers in premenopausal AUB. Specifically, the number of hemorrhagic episodes and the abundance of bleeding represent simple, clinically accessible variables that can be systematically collected during patient history. As our data suggest, fewer episodes and lower bleeding abundance are associated with a reduced likelihood of malignancy, underscoring the importance of detailed bleeding history in refining risk stratification.

5. Conclusions

This study developed and internally validated a novel risk prediction score for endometrial malignancy in premenopausal women presenting with AUB. The score based on five key predictors: endometrial thickness greater than 9 mm, oral contraceptive use, and the presence of vascularization as risk factors, while a low number of hemorrhage episodes and lower bleeding abundance were protective.

A cutoff value of 7 or higher identified high risk of endometrial malignancy, with excellent diagnostic accuracy (AUC 0.901). At this threshold, the score achieved a sensitivity of 77% and specificity of 90%, with a particularly high negative predictive value of 98%, making it especially useful for ruling out malignancy among low-risk patients.

These findings suggest that the proposed score could serve as a valuable non-invasive decision-support tool to guide the selective use of endometrial biopsy. While external validation in larger and more diverse populations remains necessary, the model has the potential to improve patient-centered care by reducing unnecessary invasive procedures and ensuring timely diagnosis in women at elevated risk.

Ethical considerations

We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

The study protocol was approved on 6 Mars 2025 by the institutional ethics committee of Charles Nicolle Hospital, Tunis, Tunisia before conducting the study (approval number: FWA 00032748- IORG0011243).

Consent to participate

As this was a retrospective study using anonymized data, informed consent was waived.