Development and Validation of a Long-Term Prognostic Model for Cirrhotic Patients after Endoscopic Variceal Ligation

Dongxue YaoJianmei Pan^*

• Central Hospital Affiliated to Shandong First Medical University, Jinan, China

Objective: To develop and validate a machine learning model incorporating liver function, variceal characteristics, and hemodynamic parameters for individualized prediction of long-term prognosis in cirrhotic patients undergoing endoscopic variceal ligation (EVL), thereby providing a reference for clinical outcome assessment and treatment decision-making. Methods: Cirrhotic patients who underwent EVL between January 2022 and June 2024 were retrospectively enrolled and randomly divided into a training set and a validation set at a 7:3 ratio. In the training set, univariate analysis, Least Absolute Shrinkage and Selection Operator regression and multivariate logistic regression were used to identify independent predictors. Key predictive variables were used to construct machine learning models (random forest (RF), support vector machine (SVM), and k-nearest neighbors (KNN)). Model performance was evaluated by the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis, with the optimal model selected. SHapley Additive exPlanations (SHAP) method was employed to assess model interpretability, analyzing the direction and magnitude of variable contributions to predictions. Results: A total of 342 patients were retrospectively enrolled and randomly divided into a training set (n=239) and a validation set (n=103). Baseline clinical characteristics showed no significant differences between the training and validation sets (all P>0.05). In the training set, multivariate logistic regression showed that Model for End-Stage Liver Disease (MELD) score, total bilirubin, maximum variceal diameter, and hepatic venous pressure gradient (HVPG) were independent risk factors for poor prognosis (all P<0.05), whereas serum albumin, platelet count, and hemoglobin at admission were protective factors (all P<0.05). Among machine learning models, the SVM model demonstrated superior predictive performance, with an AUC of 0.859 in the training set and 0.797 in the validation set, outperforming RF and KNN models. SHAP interpretability analysis confirmed that MELD score, HVPG, and total bilirubin contributed most strongly to increased risk, while serum albumin, platelet count, and hemoglobin at admission exerted protective effects. Conclusion: This study successfully developed and validated a predictive model incorporating MELD score, variceal diameter, and HVPG. The model accurately predicts long-term survival in cirrhotic patients after EVL, serving as a practical tool for individualized prognostic assessment.