Predicting short-term mortality in severe cirrhosis: An interpretable machine learning model integrating routine
Shun Zhang1,2, Rui Liu2, Zhengjie Li1,2
1Department of Gastroenterology, Chengdu University of Traditional Chinese Medicine Affiliated Hospital of Integrated Traditional Chinese and Western Medicine, Chengdu, China.
Plos One
|March 3, 2026
Summary
A new machine learning model accurately predicts short-term mortality in severe liver cirrhosis patients. Key predictors include age, INR, creatinine, and AST/ALT, aiding clinical risk stratification and intervention.
Area of Science:
- Critical care medicine
- Machine learning in healthcare
- Liver disease research
Background:
- Severe liver cirrhosis presents high 30-day mortality rates, necessitating precise risk stratification.
- Effective clinical interventions depend on reliable tools for predicting patient outcomes.
Purpose of the Study:
- To develop a machine learning (ML) prognostic model for predicting short-term mortality in decompensated cirrhosis.
- To leverage critical care data for enhanced patient risk assessment.
Main Methods:
- Retrospective analysis of 1,044 patients from the MIMIC-IV database, split into training and validation sets.
- Development of an ML model using multidimensional clinical parameters.
- Internal validation and Cox proportional hazards regression for survival analysis.
Main Results:
- The ML model identified eight key predictors: age, INR, creatinine, platelets, white blood cell count, total bilirubin, peptic ulcer, and AST/ALT ratio.
- The model achieved an AUC of 0.846 in the training cohort, demonstrating strong predictive accuracy.
- Significant variations in mortality were observed across different international normalized ratio (INR) ranges.
Conclusions:
- The developed ML model effectively identifies high-risk cirrhotic patients for timely intervention.
- Key prognostic factors identified include laboratory markers and clinical complications, emphasizing close patient monitoring.
- External validation in multi-center studies is recommended to expand the model's clinical applicability.
