This research was conducted as a multicenter retrospective study, drawing data from seven tertiary hospitals located in Southwest China. The study analyzed a total of 17,567 patients with cirrhosis recorded between the years 2012 and 2023. Following initial data collection, a total of 13,490 eligible patients were included in the analysis. Of this cohort, 1,250 patients, which constitutes 9.27% of the total eligible population, developed ACLF. The cohort was subsequently divided into a training set, consisting of 9,758 patients, and a test set comprising 4,182 patients.

The methodological approach involved a rigorous machine learning pipeline. Feature selection was initially performed using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Multiple machine learning algorithms were then developed and compared for ACLF prediction, including the Light Gradient Boosting Machine (LightGBM) model. Model performance was assessed and validated using several standard metrics: the area under the receiver operating characteristic curve (AUC–ROC), the area under the precision–recall curve (AUC–PR), calibration curves, and decision curve analysis. Furthermore, the TreeSHAP framework was applied to the model to facilitate interpretation and understand the contribution of individual features.

Among the evaluated algorithms, the LightGBM model demonstrated the best predictive performance across both the training and test sets. In the training set, the LightGBM model achieved an AUROC of 0.960 and an AUPRC of 0.727. Other performance metrics for the training set included an accuracy of 0.882, a sensitivity of 0.954, a specificity of 0.875, and an F1 score of 0.593. When the model was applied to the independent test set, the corresponding performance values were 0.948 for AUROC, 0.655 for AUPRC, 0.866 for accuracy, 0.912 for sensitivity, 0.861 for specificity, and 0.549 for the F1 score.

To provide clinical interpretability, SHAP analysis identified the nine most important predictive features, ranked by their contribution to the model’s predictive accuracy. These critical variables included: international normalized ratio, total bilirubin, prealbumin, hepatitis B virus infection status, alanine aminotransferase, platelet count, total bile acid, hepatic encephalopathy, and alcoholic liver disease. The successful development of this LightGBM-based diagnostic model, which incorporates these nine key clinical variables, facilitates the timely identification and risk stratification of ACLF in patients with cirrhosis.

The findings of this multicenter retrospective study suggest that the LightGBM model offers a robust computational framework for clinicians. By enabling the timely identification and risk stratification of ACLF in patients with cirrhosis, this model provides a tool that may support clinical decision-making at the point of presentation and facilitate targeted clinical interventions.

Subject of Research: Development and validation of a machine learning model for the prediction of Acute-on-Chronic Liver Failure in cirrhosis patients.

Article Title: Development and validation of a LightGBM-based diagnostic model for acute-on-chronic liver failure in patients with liver cirrhosis: a multicenter retrospective study in an HBV-predominant cohort

Article References: Xiao, M., Zhang, X., Tan, J., Tan, Q., Li, S., Li, R., Luo, Q., & Zhong, X. (2026). Development and validation of a LightGBM-based diagnostic model for acute-on-chronic liver failure in patients with liver cirrhosis: a multicenter retrospective study in an HBV-predominant cohort. Frontiers in Medicine, 13, 1922075. https://doi.org/10.3389/fmed.2026.1922075

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1922075

Keywords: Acute-on-chronic liver failure, Liver cirrhosis, Machine learning, LightGBM, Diagnostic model, Hepatology, Clinical prediction, LASSO regression, Development, validation, LightGBM-based, diagnostic