Predicting the Risk of Decompensation in Patients With Cirrhosis: A Validated Machine Learning Approach.

Area of Science:

• Hepatology
• Machine Learning
• Predictive Analytics

Background:

• Compensated cirrhosis patients lack simple tools to predict hepatic decompensation.
• Early prediction is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a straightforward machine learning model for predicting liver decompensation in compensated cirrhosis.
• To create a tool utilizing routinely available clinical and laboratory data.

Main Methods:

• Applied a random forest classifier to clinical and laboratory data from 575 patients in training cohorts (UK, Italy).
• Validated the model across independent international cohorts totaling over 2,100 patients (Ireland, Egypt, UK, Japan).

Main Results:

• Developed RODIC (Risk of Decompensation in Cirrhosis), a model based on albumin, bilirubin, and Fib-4 score.
• Achieved high performance in training (AUC-ROC=0.86, F1=0.82) and validation cohorts (AUC-ROC 0.67-0.80).
• Model demonstrated effectiveness across diverse cirrhosis etiologies and in HCV patients with sustained virologic response.

Conclusions:

• A validated machine learning model (RODIC) accurately quantifies decompensation risk in compensated cirrhosis.
• The model uses readily available clinical and laboratory features, offering a practical tool for risk assessment.