A MAJOR international study has found that machine learning can significantly improve predictions of in-hospital death for patients with cirrhosis, a chronic liver condition affecting millions worldwide. The findings, published by researchers from the CLEARED study group, show that AI-powered tools outperform traditional prediction methods and could help doctors make more timely, life-saving decisions.
The team analysed data from 7,239 patients hospitalized with cirrhosis across 115 centres globally, including countries of all income levels. Using information available on the first day of hospital admission, researchers compared how well different models predicted a patient’s risk of dying during their hospital stay.
The machine learning model known as Random Forest stood out for its accuracy. It correctly predicted patient outcomes more often than standard statistical tools across all income settings—from low-income to high-income countries. The model achieved an AUC (area under the curve) of 0.815, a key measure of prediction accuracy, and remained reliable when tested separately in each economic group.
To further test its real-world performance, the model was applied to a separate group of 28,670 U.S. veterans hospitalized with cirrhosis. In this large external validation, the tool again performed strongly, with an AUC of 0.859.
“This model could help clinicians worldwide better identify high-risk patients early and guide care decisions more effectively,” said Dr. Jasmohan Bajaj, senior author of the study.
The study highlights how artificial intelligence may transform hospital care by helping predict patient outcomes more accurately, even in low-resource settings. Researchers believe this approach could pave the way for more personalized and efficient liver disease care around the world.
Reference
Silvey S et al. Enhancement of Inpatient Mortality Prognostication with Machine Learning in a Prospective Global Cohort of Patients with Cirrhosis with External Validation. Gastroenterol. 2025;DOI: 10.1053/j.gastro.2025.07.015.
