Introduction: Endometrial cancer (EC) is one of the most common gynecological cancers worldwide, with a rising incidence that highlights the urgent need for effective screening methods. Despite early-stage diagnosis and favorable survival rates, current screening methods such as transvaginal ultrasonography lack specificity, often necessitating invasive procedures and revealing a significant gap in EC detection.
Methods: We conducted a comprehensive analysis of DNA methylation in cytological specimens as a biomarker for EC screening. Using a literature review and the UALCAN and Wanderer databases, we identified 7 hypermethylated genes associated with EC. Endometrial samples were collected from 300 women, and endometrial cytology testing (ECT) and quantitative methylation-specific PCR (qMSP) were used to evaluate these genes. An XGBoost algorithm-based model was developed to predict EC using DNA methylation data, with performance assessed through sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC).
Results: The methylation levels of HTR1B, CELF4, and TBX5 were significantly elevated in EC and atypical hyperplasia compared to benign samples. The diagnostic model combining these genes demonstrated superior performance, achieving 97% sensitivity and an accuracy of 90%. SHAP value analysis indicated that TBX5 (1.079), HTR1B (0.990), and CELF4 (0.712) positively influenced the model's predictive power, with weights for TBX5 and HTR1B being similar but higher than that of CELF4.
Conclusion: Integrating DNA methylation markers into ECT offers a non-invasive and highly accurate approach to EC screening. This model's high diagnostic accuracy and reliability have the potential to transform EC diagnosis, reducing reliance on invasive procedures and improving clinical management.
Keywords: DNA methylation; XGBoost algorithm; diagnostic model; endometrial cancer; endometrial cytology testing (ECT).
© 2026 Wiley Periodicals LLC.