An artificial intelligence-generated model predicts 90-day survival in alcohol-associated hepatitis: A global cohort...

Area of Science:

• Hepatology
• Artificial Intelligence in Medicine
• Prognostic Modeling

Background:

• Alcohol-associated hepatitis (AH) has high short-term mortality.
• Current prognostic models lack precision for predicting 90-day mortality.
• Need for enhanced models in a global context.

Purpose of the Study:

• To derive and validate an enhanced prognostic model for AH mortality using artificial intelligence.
• To improve precision in predicting 30-day and 90-day mortality.
• To compare the new model against existing prognostic tools.

Main Methods:

• Retrospective study of a global cohort (Global AlcHep initiative) across 23 centers.
• Utilized 3 AI algorithms (Random Forest, GBM, XGBoost) to create an ensemble model.
• Refined the model using Bayesian updating and integrated center-specific mortality rates.

Main Results:

• The ALCoholic Hepatitis Artificial INtelligence Ensemble (ALCHAIN) score integrates clinical and lab data with center-specific mortality.
• Achieved AUCs of 0.811 (30-day) and 0.799 (90-day) in the validation cohort, outperforming legacy models (p < 0.001).
• Demonstrated superior calibration and identified steroid therapy benefits for specific patient subgroups.

Conclusions:

• AI-driven ALCHAIN score offers superior prediction of AH mortality compared to traditional models.
• The score is valuable for clinical trials, guiding steroid therapy, and informing transplant decisions.
• An accessible online tool (https://aihepatology.shinyapps.io/ALCHAIN/) is available.