Work

Prediction of acute kidney injury in ICU patients with cirrhosis and sepsis: a multicenter machine learning study

Year: 2026
Type: article
Study design: Observational Study
Abstract: Objective To develop and validate a machine learning model based on routinely available clinical variables for early risk stratification of acute kidney injury in intensive care unit (ICU) patients with cirrhosis and sepsis. Methods In this retrospective multicenter cohort study, data were derived from the Medical Information Mart for Intensive Care IV (MIMIC-IV v3.1) database and an independent ICU cohort from the Second Affiliated Hospital of Wenzhou Medical University. Adult patients with cirrhosis who met Sepsis-3 criteria at ICU admission were included. Clinical variables obtained at the early stage of ICU admission were used as candidate predictors. Feature selection was performed using least absolute shrinkage and selection operator (LASSO) regression combined with the Boruta algorithm. Seven machine learning algorithms, including random forest (RF), extreme gradient boosting (XGBoost), artificial neural network (ANN), light gradient boosting machine (LightGBM), support vector machine (SVM), decision tree, and logistic regression (LR), were developed and compared. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration, and decision curve analysis. Model interpretability was assessed using Shapley Additive Explanations. Results A total of 2,491 ICU patients with cirrhosis and sepsis were included, among whom 61% developed AKI. Fourteen variables were selected for model construction, including ascites, vasopressors, SOFA, SAPS II, weight, hemoglobin, platelet count, WBC, albumin, sodium, creatinine, prothrombin time, total bilirubin, respiratory rate. In the internal validation cohort, RF and XGBoost demonstrated comparable discrimination, both achieving an AUC of 0.839, while the RF model showed good calibration. In the external validation cohort, the RF model achieved the highest AUC (0.755; 95% CI, 0.686–0.820) among the evaluated models and provided a favorable clinical net benefit across a clinically relevant range of threshold probabilities. SHAP analysis indicated that creatinine, disease severity scores, coagulation parameters, and inflammatory markers were the most influential predictors, with effect directions consistent with known hepatorenal pathophysiology. Conclusion A machine learning–based model using routinely available clinical variables can provide accurate and interpretable early prediction of AKI in ICU patients with cirrhosis and sepsis. This model may facilitate timely risk stratification and support clinical decision-making. Further prospective studies are warranted to validate its clinical utility and real-world implementation.
Language: English
Subfield: Nephrology
Field: Medicine
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7225993557
FWCI:
Cites: 41
Cited by:
Related to: 10
Parts: 7