Machine Learning-Based Prediction of Decompensation in Hepatitis B Virus-Related Cirrhosis

Area of Science:

• Hepatology
• Medical Informatics
• Machine Learning

Background:

• Decompensation significantly increases mortality in cirrhotic patients.
• Limited research exists on machine learning for predicting cirrhosis decompensation.
• Hepatitis B virus (HBV)-related cirrhosis presents a critical need for predictive models.

Purpose of the Study:

• To develop and validate machine learning models for predicting complications in HBV-related cirrhosis.
• To assess the predictive capabilities of various algorithms for decompensation events.
• To evaluate the role of antiviral therapy in predicting cirrhosis decompensation.

Main Methods:

• Utilized electronic health records from 50,047 HBV patients.
• Trained and tested 32 machine learning models (SVM, LR, DT, RF) for predicting variceal bleeding, ascites, jaundice, and multiple complications.
• Assessed model performance using AUROC and accuracy, considering entecavir (ETV) and lamivudine (LAM) use.

Main Results:

• SVM and RF models demonstrated strong predictive performance (AUROCs 0.85-0.93, accuracy 0.77-0.88) for ascites, jaundice, and multiple complications.
• SVM and LR showed excellent differentiation of ascites in ETV users (AUROC 0.93, 0.92).
• Antiviral treatment details and clinical data are significant predictors of decompensation.

Conclusions:

• Machine learning models can aid clinicians in decision-making for cirrhosis decompensation.
• Electronic health records are a valuable resource for developing predictive ML models.
• Antiviral therapy is a crucial predictor for decompensation in HBV-related cirrhosis.