Construction of a Predictive Model for Late Bleeding After Endoscopic Submucosal Dissection in Patients with Benign Colorectal Tumors

doi:10.12104/j.issn.1674-4748.2025.19.016

Abstract：Objective: To explore the predictive value of late bleeding after endoscopic submucosal dissection (ESD) in patients with benign colorectal tumors. Methods: A retrospective study was conducted on the clinical data of 200 patients who underwent ESD for benign colorectal tumors between January 2021 and May 2024. Patients were divided into two groups based on whether they experienced late bleeding: the non-occurrence group (n=184) and the occurrence group (n=16). Univariate analysis was used to identify potential risk factors, and multivariate logistic regression analysis was performed to determine independent influencing factors and construct a predictive model for late bleeding after ESD. The receiver operating characteristic (ROC) curve was used to evaluate the model's discrimination ability. Results: Univariate analysis showed that tumor size, lesion location, tumor depth, intraoperative bleeding, and preoperative use of antithrombotic drugs were significant factors affecting late bleeding after ESD (all P < 0.05). Multivariate analysis confirmed that tumor size, lesion location, tumor depth, intraoperative bleeding, and preoperative use of antithrombotic drugs were independent risk factors for late bleeding after ESD (all P < 0.05). Based on these factors, a predictive model was established, with the formula: Logit(P) = -7.325 + 0.069 × tumor size - 0.830 × lesion location + 1.898 × tumor depth + 5.047 × presence of intraoperative bleeding + 4.048 × preoperative use of antithrombotic drugs. The area under the ROC curve (AUC) was 0.940 [95% CI (0.870, 1.000)], with a maximum Youden index of 0.835, sensitivity of 0.938, and specificity of 0.897, indicating that the predictive model has high predictive value. Conclusion: The constructed predictive model for late bleeding after ESD in patients with benign colorectal tumors demonstrates good performance. Medical staff can use this model to identify high-risk patients and implement targeted interventions.