Abstract / Summary
Abstract Background Cervical cancer is one of the most common cancers among women worldwide. Although curative radiotherapy can significantly improve the prognosis of patients with locally advanced disease, Radiation-induced Late Rectal Injury(RLRI) severely compromises patients’ long-term quality of life. Objective To investigate risk factors for Grade ≥ 2 RLRI in patients with locally advanced cervical cancer following curative radiotherapy, and to construct and validate a nomogram model. Methods This study included 287 patients who underwent radical radiotherapy for locally advanced cervical cancer at two Grade A tertiary hospitals in Qinghai between May 2020 and May 2025. Demographic data, underlying medical conditions, clinical and pathological information, and rectal dose-volume histogram (DVH) parameters were collected.According to Section 7.3, the data was randomly divided into a training cohort of 204 cases and a validation cohort of 83 cases.Screen for variables with P < 0.1 using univariate logistic regression,Lasso regression reduces dimensionality and addresses multicollinearity,Multi-factor Logistic Regression (backward method) to identify independent predictors and build a nomogram model;Use the receiver operating characteristic (ROC) curve, area under the curve (AUC), the Hosmer-Lemeshow test, calibration curves, and decision curve analysis (DCA) to evaluate the model. Results The training set included 47 cases with ≥ Grade 2 RLRI (23.04%), and the validation set included 17 cases (20.48%).After filtering using Lasso regression, five variables were identified: hypertension, type 2 diabetes, Dmax, D2cc, and D1cc. These variables were included in a multivariate logistic regression analysis, which ultimately identified hypertension, Dmax, and D2cc as three independent risk factors. A corresponding nomogram model was then constructed based on these findings.The AUC for the training cohort was 0.846 (95% CI: 0.782–0.910), and the AUC for the validation cohort was 0.772 (95% CI: 0.648–0.895), indicating good discriminatory power.The Hosmer-Lemeshow tests for the training and validation cohorts were χ² = 11.26, P = 0.258, and χ² = 8.28, P = 0.505, respectively, indicating that the models fit the data well;The calibration curves for the training and validation cohorts align with the ideal diagonal, indicating that the model is well calibrated.The DCA results show that within the 0–0.7 risk threshold range, the model demonstrates a significant net profit advantage. Conclusion A nomogram model based on hypertension, Dmax, and D2cc can intuitively predict the risk of severe RLRIfollowing radiation therapy for locally advanced cervical cancer, and may provide guidance for dose constraints on the rectum as a critical organ and for early intestinal protection.