JMIR Publications

ABSTRACT

Background:

Hepatic encephalopathy (HE) accounts for a significant mortality risk in patients with liver cirrhosis. Early identification of their prognosis and incidence of complications are crucial, but challenging, for physicians in decision making and treatment strategy, especially for treating non-cancer-related cirrhotic patients due to unpredictability of their disease trajectories.

Objective:

Therefore, this study aimed to develop a novel machine learning (ML) model to enhance predictability of HE in patients with non-cancer-related cirrhotic patients.

Methods:

A multi-centre, retrospective cohort study was conducted from January 2010 to December 2017 across all Chang Gung Memorial Hospital branches in northern, middle and southern Taiwan. We applied several ML models to evaluate HE predictability and compared their performance in the training dataset and testing datasets. Optimal sensitivity and specificity were determined using Youden’s index. The best ML model was interpreted by SHapley Additive exPlanations Plot.

Results:

A total of 4080 subjects were enrolled. The eXtreme gradient boosting algorithm showed the best performance in predicting HE incidence (area under the curve 0.85, 95% confidence interval: 0.834–0.879) compared with other ML models and model for severity of cirrhosis score. Key variables included ammonia, aspartate aminotransferase, prothrombin time, alanine aminotransferase, and serum potassium. The cut-off value for HE discrimination was 0.25 (sensitivity: 80%, specificity: 81%), providing a high negative predictive value (0.94) in the training dataset.

Conclusions:

We developed a novel ML model for predicting HE in non-cancer-related cirrhotic population, thereby providing a practical guide to help physicians and these patients in share-decision-making for treatment strategy to improve patient care and reducing suffering from morbid complications.