Abstract
BACKGROUND: The incidence of esophageal cancer is high, and its prognosis is poor. Endoscopic submucosal dissection (ESD) is an important, minimally invasive treatment for early esophageal cancer, but the risk of postoperative bleeding affects its efficacy. AIM: To explore risk factors of bleeding after ESD and evaluate the predictive value of a gradient boosting machine (GBM) model for postoperative bleeding. METHODS: The clinical data of 178 early esophageal cancer patients who underwent ESD at the Affiliated Hospital of Xuzhou Medical University from October 2019 to October 2024 were analyzed retrospectively. Patients were divided into two groups (bleeding and non-bleeding). Univariate and multivariate logistic regression analyses identified risk factors for postoperative bleeding, leading to the construction of the GBM prediction model. The receiver operating characteristic (ROC) curve evaluated the predictive efficacy of the GBM model and bleeding after ESD trend from Japan (BEST-J) score. RESULTS: Among 178 patients who received ESD treatment, 29 cases (16.29%) had bleeding, and 149 cases (83.71%) had no bleeding. The average BEST-J score and the proportion of high-risk and extremely high-risk patients were higher in the bleeding group than in the non-bleeding group (P < 0.05). Multivariate logistic regression analysis showed that tumor size ≥ 3 cm, surgical bleeding, and C-reactive protein (CRP) were independent risk factors for bleeding after ESD in patients with early esophageal cancer (P < 0.05). The ROC curve showed that the area under the curve of the GBM prediction model based on the influencing factors was greater than that of the BEST-J score (0.818 vs 0.653, P < 0.05). CONCLUSION: The GBM prediction model based on tumor size ≥ 3 cm, surgical bleeding, and high CRP levels is more effective than the BEST-J score at predicting bleeding after ESD.