Machine Learning Models predicting Decompensation in Cirrhosis.

Area of Science:

• Hepatology
• Machine Learning in Medicine
• Biostatistics

Background:

• Cirrhosis decompensation significantly reduces patient survival.
• Preventing cirrhosis complications is crucial for improving outcomes.
• Machine learning offers novel approaches to predict decompensation risk.

Purpose of the Study:

• To identify key parameters predicting cirrhosis decompensation using machine learning.
• To evaluate the performance of different machine learning models in predicting decompensation.
• To explore the utility of laboratory, clinical, and genetic data in risk prediction.

Main Methods:

• Applied various machine learning techniques (Random Forests, Support Vector Machines) to a database of 983 patients.
• Utilized hierarchical clustering and Permutation Feature Importance for parameter evaluation.
• Analyzed retrospective and prospective data, including laboratory, clinical, and genetic information.

Main Results:

• Random Forests achieved 81.6% accuracy retrospectively; Support Vector Machines achieved 78.6% prospectively.
• Key predictors identified include baseline albumin, baseline bilirubin, and maximum bilirubin.
• NOD2 genotype and inflammatory markers were significant predictors beyond established score parameters.

Conclusions:

• Laboratory parameters, genetic variants, and infections are valuable in predicting cirrhosis decompensation risk.
• This study provides a foundation for developing advanced predictive models.
• Early identification of high-risk patients can facilitate timely interventions.