Identification of exacerbation risk in patients with liver dysfunction using machine learning algorithms.

Area of Science:

• Medical Informatics
• Machine Learning in Medicine
• Hepatology

Background:

• Liver failure (LF) poses significant health risks, with timely diagnosis crucial for managing complications and disease progression.
• Improving treatment efficacy and reducing healthcare costs for LF patients are key clinical challenges.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting patient deterioration after hospital admission for liver failure.
• To compare the predictive performance of the ML model against the established Model for End-Stage Liver Disease (MELD) score.

Main Methods:

• Retrospective analysis of 348 liver failure patients from May 2011 to March 2018.
• Utilized 15 key clinical indicators as input features for machine learning algorithms.
• Compared the predictive accuracy of ML models (GLMs, CART, SVM, NNET) and the MELD score using 10-fold cross-validation.

Main Results:

• The developed machine learning models, particularly NNET (AUC 0.912, accuracy 0.912) and SVM (AUC 0.853, accuracy 0.853), demonstrated superior predictive performance compared to the MELD score (AUC 0.670, accuracy 0.669).
• All evaluated ML models, except GLMs, exceeded the predictive capabilities of the classic MELD model.
• The NNET model achieved the highest predictive accuracy and AUC among all tested methods.

Conclusions:

• Machine learning models show significant potential in accurately forecasting liver failure patient deterioration.
• These ML models can serve as valuable tools for physicians, enabling timely initiation of treatment for liver disease patients.
• The developed ML approach offers a promising advancement over traditional methods like the MELD score for managing liver failure.