Machine Learning–Based Risk Stratification for Gastrointestinal Bleeding in ICU Patients with Cirrhosis: Evidence from the MIMIC Database

Yuxin DUANWeifan SUIJianyun LIZefeng CAIYimao XIAJianhua FU^*

• Jiangsu University Affiliated People's Hospital, Zhenjiang, China

Background: Gastrointestinal bleeding (GIB) is a common complication in critically ill patients with cirrhosis, significantly impacting clinical outcomes. Early identification of high-risk patients is crucial for guiding appropriate interventions and improving outcomes. Objective: To develop and externally validate a machine learning model for predicting in-hospital GIB in ICU patients with cirrhosis, identify key predictors, and assess its clinical utility for risk stratification. Methods: This retrospective cohort study included 3,160 ICU patients with cirrhosis from the MIMIC-IV database, divided chronologically into training (n=2,528) and testing (n=632) cohorts. External validation was performed on 523 patients from the eICU database. Predictive variables were identified using the Boruta algorithm, correlation analysis, and variance inflation factor (VIF) assessment. Six machine learning algorithms—logistic regression, k-nearest neighbors, support vector machine, random forest (RF), multilayer perceptron, extreme gradient boosting, and gradient boosting machine—were trained and evaluated through 10-fold cross-validation. Model performance was assessed using AUC-ROC, accuracy, sensitivity, specificity, precision, F1-score, calibration curves, and decision curve analysis (DCA). Shapley additive explanations (SHAP) analysis identified key predictors. Multivariable logistic regression examined the relationship between anticoagulant therapy and in-ICU GIB incidence. Results: The RF model achieved AUC of 0.86 (95% CI: 0.84–0.88) in training and 0.72 (95% CI: 0.68–0.76) in testing, with sensitivity 0.68, specificity 0.71, and precision 0.47. Key predictors included red blood cell count, hemoglobin, platelet count, and anticoagulant therapy. DCA indicated meaningful clinical utility for risk stratification. Anticoagulant use was independently associated with lower GIB risk (OR: 0.29; 95% CI: 0.24–0.34), with consistent findings across gender, age, weight, and other subgroups. Conclusions: The RF model demonstrated stable discrimination for predicting GIB risk in ICU patients with cirrhosis across multiple cohorts. Built on readily available clinical data, it enables timely risk stratification and informs individualized preventive interventions in critical care settings.