Development and validation of an interpretable machine learning model for predicting in-hospital mortality among critically ill patients with liver cirrhosis

This study aimed to develop and validate an interpretable machine learning model for early prediction of in-hospital mortality in critically ill patients with liver cirrhosis to optimize risk stratification and individualized clinical intervention. A total of 5358 cirrhotic ICU patients were retrospectively selected from the MIMIC-IV database and randomly divided into training and internal validation cohorts at a 7:3 ratio. Twelve machine learning algorithms were constructed and compared with the conventional SAPS II score. Model performance was evaluated via AUC-ROC, and SHAP analysis was used to improve model transparency and clinical interpretability. The CatBoost model showed the best predictive performance with an AUC of 0.79, significantly superior to SAPS II (AUC = 0.75, p < 0.001). SHAP analysis identified minimum lactate, body temperature, blood urea nitrogen and white blood cell count as the top four influential predictors. Using routine clinical indicators collected within 24 h of ICU admission, this interpretable model showed better discrimination than SAPS II for predicting in-hospital mortality in critically ill patients with cirrhosis. It may assist early risk stratification, but further external validation is needed before clinical application.