Abstract
BACKGROUND: Variceal rebleeding is a significant and potentially life-threatening complication of cirrhosis. Unfortunately, currently, there is no reliable method for stratifying high-risk patients. Liver stiffness measurements (LSM) have been shown to have a predictive value in identifying complications associated with portal hypertension, including first-time bleeding. However, there is a lack of evidence to confirm that LSM is reliable in predicting variceal rebleeding. The objective of our study was to evaluate the ability of generating a extreme gradient boosting (XGBoost) algorithm model to improve the prediction of variceal rebleeding. METHODS: This retrospective analysis examined a cohort of 284 patients with hepatitis B-related cirrhosis. XGBoost models were developed using laboratory data, LSM, and imaging data to predict the risk of rebleeding in the patients. In addition, we compared the XGBoost models with traditional logistic regression (LR) models. We evaluated and compared the two models using the area under the receiver operating characteristic curve (AUROC) and other model performance parameters. Lastly, we validated the models using nomograms and decision curve analysis (DCA). RESULTS: During a median follow-up of 66.6 weeks, 72 patients experienced rebleeding, including 21 (7.39%) and 61 (21.48%) patients who rebleed within 6 weeks and 1 year, respectively. In brief, the AUC of the LR models in predicting rebleeding at 6 weeks and 1 year was 0.828 (0.759-0.897) and 0.799 (0.738-0.860), respectively. In contrast, the accuracy of the XGBoost model in predicting rebleeding at 6 weeks and 1 year was 0.985 (0.907-0.731) and 0.931 (0.806-0.935), respectively. LSM and high-density lipoprotein (HDL) levels differed significantly between the rebleeding and nonrebleeding groups, with LSM being a reliable predictor in those models. The XGBoost models outperformed the LR models in predicting rebleeding within 6 weeks and 1 year, as demonstrated by the ROC and DCA curves. CONCLUSION: The XGBoost algorithm model can achieve higher accuracy than the LR model in predicting rebleeding, making it a clinically beneficial tool. This implies that the XGBoost model is better suited for predicting the risk of esophageal variceal rebleeding in patients.