A machine learning model for 90-day mortality prediction in hepatitis B virus-related acute-on-chronic liver failure: the pivotal role of CALLY index

Background: Hepatitis B virus-related acute-on-chronic liver failure (HBV-ACLF) is a life-threatening syndrome, the condition can deteriorate rapidly, and the 90-day mortality rate is high. Due to the rapid changes in the clinical course, early and accurate risk stratification is crucial for timely decision-making and resource allocation in the ICU. This study has developed and verified a machine learning framework that integrates the C-reactive protein-albumin-lymphocyte (CALLY) index to predict the 90-day mortality rate of HBV-ACLF patients. Methods: = 142) according to the ratio of 7:3. In addition, we performed temporal external validation using a later cohort of 46 patients with HBV-ACLF admitted between 2024 and 2025. Six machine-learning algorithms including logistic regression, support vector machine, K-nearest neighbors, Extra Trees, XGBoost, and LightGBM, were trained to develop predictive models. SHAP was applied to assess feature importance and model interpretability. Results: Among the evaluated algorithms, the LightGBM model showed the strongest overall discriminative performance, with AUCs of 0.940 (95% CI, 0.916-0.964) in the training set, 0.825 (95% CI, 0.757-0.894) in the internal validation set, and 0.804 (95% CI, 0.669-0.939) in the external validation set. SHAP analysis identified international normalized ratio (INR) as the strongest predictor of mortality, followed by the CALLY index, log-transformed total bilirubin (TBIL), age, and creatinine. The CALLY-based model significantly improved risk stratification compared with the Model for End-Stage Liver Disease (MELD) score, as demonstrated by a continuous net reclassification index of 0.6223 and an integrated discrimination improvement of 0.0937. To facilitate clinical implementation, a user-friendly online tool was created to predict mortality rates. Conclusions: Our machine-learning framework integrating the CALLY index provides a high-precision and transparent decision-support tool for predicting 90-day mortality with HBV-ACLF patients. By quantifying the immune-nutritional-inflammatory axis, this approach may facilitate earlier risk stratification and personalized interventions, potentially improving survival in high-risk patients.