Metabolic factor-based machine learning model for mortality prediction in acute hepatitis E: Development and...

Area of Science:

• Hepatology
• Machine Learning in Medicine
• Predictive Modeling

Background:

• Hepatitis E virus (HEV) infection is a significant cause of liver failure with high mortality.
• Existing predictive models for HEV lack comprehensive systemic metabolic factors.

Purpose of the Study:

• To develop a machine learning model for predicting mortality in HEV patients.
• To integrate systemic metabolic parameters into a predictive model for HEV.

Main Methods:

• Retrospective analysis of 510 HEV patients across training, internal, and external validation cohorts.
• Development of a Metabolism Score using Support Vector Machines (SVM) with metabolic parameters.
• Selection of clinical variables and the Metabolism Score using LASSO regression.
• Construction and evaluation of five machine learning models for 28-day and 90-day mortality prediction.

Main Results:

• Logistic Regression (LR) model demonstrated superior performance in predicting 28-day and 90-day mortality across all cohorts (AUROCs ranging from 0.84 to 0.98).
• The LR model outperformed the MELD score in predictive accuracy.
• The developed LR model showed good calibration, significant net clinical benefits, and was visualized using a nomogram.

Conclusions:

• The LR model incorporating systemic metabolic factors provides accurate prediction of short-term mortality in HEV patients.
• This model can potentially enhance early risk stratification and guide personalized treatment strategies for HEV infection.