Early Identification of Endometrial Malignancy in Postmenopausal Women with Asymptomatic Endometrial Thickening: A...

Early screening and management of asymptomatic postmenopausal women with endometrial thickening are essential to optimize diagnosis and treatment outcomes. However, no unified intervention standards or predictive models for high-risk subgroups exist. This study aimed to develop and validate a SHapley Additive exPlanations (SHAP)-based machine learning (ML) model to identify key risk factors for non-benign lesions in this population.

Enrolled in this retrospective cohort were 1031 asymptomatic postmenopausal women with endometrial thickening (≥ 5 mm) who underwent hysteroscopy at International Peace Maternity and Child Health Hospital from January 1, 2017, to July 31, 2025. This study comprehensively compiled 33 candidate predictors from accessible clinical datasets, covering demographic characteristics, disease attributes, transvaginal ultrasound results, and laboratory data. Least absolute shrinkage and selection operator (LASSO) regression was adopted for feature selection. Eight machine learning methods (Logistic Regression, Random Forest, Gradient Boosting, XGBoost, LightGBM, Naive Bayes, LDA, QDA) were leveraged to construct the model. Outcome interpretation was performed with the SHAP method, and a dynamic online nomogram was created to support clinical practice.

Parity, endometrial thickness (ET), mean platelet volume (MPV), platelet distribution width (PDW), and D-dimer were identified as independent risk factors. An online nomogram built upon these variables facilitated the real-time prediction of endometrial atypical hyperplasia (EAH)/ endometrial carcinoma (EC). Among eight machine learning models, the Gradient Boosting model achieved the superior performance, with an AUC of 0.763 (95% CI: 0.640-0.865), accuracy of 0.791, sensitivity of 0.667, and specificity of 0.805. Visualized interpretation at the individual patient level was achieved using the SHAP force plot.

We proposed a robust and interpretable ML-driven strategy for EAH/EC risk assessment in postmenopausal women with asymptomatic endometrial thickening. The model demonstrated superior predictive performance and feasibility for population-wide screening, serving as an efficient tool for the risk stratification of early endometrial malignancy prior to surgery and thus preventing overtreatment in low-risk individuals.