Development and validation of machine learning models for predicting acute kidney injury in acute-on-chronic liver failure: a multimodel comparative study

Acute kidney injury (AKI) is one of the serious complications in acute-on-chronic liver failure (ACLF), and the mortality rate is very high. Early identification of high-risk patients is critical. Therefore, this study aimed to develop prediction models for AKI in ACLF patients based on machine learning (ML) algorithms. This retrospective study enrolled 1,076 adult patients diagnosed with ACLF, with AKI defined according to the International Club of Ascites criteria. Participants were randomly allocated into training (n = 753, 70%) and test (n = 323, 30%) sets. Six ML models were developed: logistic regression (LR), random forest (RF), k-nearest neighbors (KNN), support vector machine (SVM), decision tree (DT), and extreme gradient boosting (XGBoost). The performance of each model was assessed using the area under the receiver operating characteristic curve (AUC-ROC), the area under the precision-recall curve (AUC-PR), the calibration curve, and decision curve analysis. Among participants, 250 (23.2%) developed AKI during hospitalization. Multivariate LR analyses identified ten significant variables in the training set: age, hypertension, total bilirubin, blood urea nitrogen, serum creatinine, blood uric acid, international normalized ratio, hepatic encephalopathy, abdominal infection, and sepsis. The RF model performed best in the test set (AUC-ROC = 0.899; AUC-PR = 0.806). The ML models can be reliable tools for predicting AKI in patients with ACLF. The RF model performed the best and can help medical clinicians to better identify patients with high risk of AKI in ACLF.