<p>Esophagogastric varices (EGV) in liver cirrhosis patients within the intensive care unit (ICU) is a significant medical concern. This study aims to develop and validate a machine learning (ML) model to predict the early mortality of those patients. Medical information was extracted from Intensive Care (MIMIC)-IV database, and 793 cirrhotic patients accompanied with EGV were enrolled, randomly assigned to the training group and the test group in a 7:3 ratio. For external validation, 100 cirrhotic patients with EGV hospitalized in ICU in our institution were retrospectively analyzed, The least absolute shrinkage and selection operator (LASSO) method and Logistic Regression(LR) analysis were applied for variable selection and predictive signature building, and four predictive models - LR, Support Vector Machine (SVM), Naive Bayes (NB), and Random Forest (RF) were conducted, and their performance in predicting 28-day all-cause mortality in the patients was evaluated using area under the receiver operating characteristic (AUROC), and decision curve analysis (DCA). five predictors associated with 28-day all-cause mortality in cirrhotic patients with EGV were identified based on LASSO and regression analysis, including MELD score, SOFA score, admission age, esophagogastric Variceal bleeding (EVB) and white cell-platelet ratio Z-score (WPR Z-score). Forest Plot and survival analysis showed WPR Z-score is strongly associated with 28-day all-cause mortality in those patients. The model based on LR showed the best predictive performance in the training set and test set with AUROC (0.833, 95% CI: 0.793–0.873) vs. (0.854, 95% CI༚0.795–0.913). For external validation, AUROC was (0.882, 95% CI༚0.795–0.924). LASSO-based predictive model, especially the LR model, showed promise in predicting early mortality in critically ill patients with cirrhosis and EGV. WPR Z-score showed strong association with early mortality in those patients.</p>