Objective:

To develop and verify a machine learning framework that integrates the CALLY index for predicting the 90-day mortality rate in patients with HBV-ACLF.

Approach:

• Study Design: A retrospective single-center study involving 471 patients with HBV-ACLF, divided into training and validation sets.
• Machine Learning Algorithms: Six algorithms were evaluated: logistic regression, support vector machine, K-nearest neighbors, Extra Trees, XGBoost, and LightGBM.
• Model Evaluation: Performance was assessed using AUC, and SHAP was used for feature importance analysis.

Key Findings:

• The LightGBM model demonstrated the best 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.
• INR was identified as the strongest predictor of mortality, followed by the CALLY index, log-transformed total bilirubin, age, and creatinine.
• The CALLY-based model significantly outperformed the MELD score in risk stratification.

Interpretation:

The machine-learning framework integrating the CALLY index provides a high-precision decision-support tool for predicting 90-day mortality in HBV-ACLF patients.

Limitations:

• The study was conducted at a single center, which may limit generalizability.
• The retrospective design may introduce biases.

Conclusion:

The integration of the CALLY index into a machine learning framework enhances early risk stratification.