Using machine learning methods to predict hepatic encephalopathy in cirrhotic patients with unbalanced data

Area of Science:

• Medical informatics
• Machine learning in healthcare
• Cirrhosis research

Background:

• Hepatic encephalopathy (HE) is a frequent complication of cirrhosis.
• Medical data for cirrhosis with HE is often imbalanced, challenging traditional prediction methods.
• Accurate prediction of HE is crucial for managing cirrhosis patients.

Purpose of the Study:

• To develop a machine learning-based risk prediction model for HE in liver cirrhosis patients.
• To improve the efficiency and accuracy of HE prediction using advanced algorithms.
• To address the challenge of unbalanced data in predicting HE.

Main Methods:

• Collected medical data from 1,256 cirrhosis patients, extracting 81 features.
• Compared logistic regression, weighted random forest (WRF), SVM, and weighted SVM (WSVM) for HE prediction.
• Validated the model using an additional 722 cirrhosis patients.

Main Results:

• WRF, WSVM, and logistic regression showed sensitivity > 0.70 for HE recognition.
• Specificity for identifying uncomplicated HE was approximately 85% across models.
• WRF demonstrated superior performance with G-means of 0.82, F-measure of 0.46, and AUC of 0.82.

Conclusions:

• The WRF model is well-suited for classifying unbalanced medical data in HE prediction.
• WRF can be utilized to build a robust risk prediction and evaluation system for cirrhosis with HE.
• Probabilistic WRF models assist clinicians in identifying high-risk HE patients.