Development and validation of a prognostic model for acute-on-chronic liver failure

Area of Science:

• Hepatology
• Machine Learning in Medicine
• Prognostic Modeling

Background:

• Prognostic assessment for acute-on-chronic liver failure (ACLF), especially in Hepatitis B virus (HBV)-endemic areas, is challenging.
• Conventional models show limited accuracy in predicting short-term outcomes for HBV-ACLF patients.

Purpose of the Study:

• To develop and validate a novel machine learning model for improved individualized prediction of short-term outcomes in HBV-ACLF.
• To incorporate liver reserve function indicators into the prognostic model.

Main Methods:

• Retrospective collection of data from 496 HBV-ACLF patients for training and 52 for external validation.
• Systematic evaluation of 12 machine learning algorithms, with the optimal model selected using the LASSO-RF approach.
• Identification of key predictive variables using SHAP values and comparison with MELD and CTP scores.

Main Results:

• The final LASSO-RF model achieved high predictive accuracy with an AUC of 0.99 in the training cohort and 0.98 in the validation cohort for 90-day mortality.
• The model significantly outperformed established scoring systems like MELD and CTP.
• An interactive web calculator was developed for clinical application, providing real-time risk scores.

Conclusions:

• Liver reserve function indicators are crucial for prognosticating HBV-ACLF outcomes.
• The developed machine learning model and online tool offer accurate risk stratification for HBV-ACLF patients.
• The tool facilitates timely and individualized clinical management decisions.