Development and Validation of an Explainable Machine Learning Model for Warning of Hepatitis E Virus-Related Acute Liver Failure.

BACKGROUND AND AIMS: Early identification of patients with acute hepatitis E (AHE) who are at high risk of progressing to hepatitis E virus-related acute liver failure (HEV-ALF) is crucial for enabling timely monitoring and intervention. This multicentre retrospective cohort study aimed to develop and validate an interpretable machine learning (ML) model for predicting the risk of HEV-ALF in hospitalised patients with AHE in tertiary care settings.

METHODS

The study cohort included patients admitted to seven tertiary medical centers in Jiangsu, China, between 01 January 2018 and 31 December 2024. Multiple ML algorithms were applied for feature selection and model training. The predictive performance of the models was evaluated in terms of discrimination, calibration and clinical net benefit. The interpretability of the final model was enhanced using the SHapley Additive exPlanations.

RESULTS

A total of 1912 participants were included in the study. Ten ML models were developed based on seven consensus-selected baseline features, with the survival gradient boosting machine (GBM) demonstrating superior performance compared to the traditional Cox proportional hazards regression model and other relevant models or scores. The GBM model achieved a Harrell's concordance index of 0.853 (95% CI: 0.791-0.914) in the external validation set. To facilitate clinical application, the GBM model was interpreted globally and locally and deployed as a web-based tool using the Streamlit-Python framework.

CONCLUSIONS

The GBM model demonstrated excellent performance in predicting HEV-ALF risk in hospitalised patients with AHE, offering a promising tool for clinical decision-making.