Development and validation of a machine learning model to predict prognosis in liver failure patients treated with non-bioartificial liver support system

Shi Shi ^1,2, Yanfen Yang ³, Yuanli Liu ³

⁰Department of Health Statistics, School of Public Health, Shanxi Medical University, Taiyuan, Shanxi, China.

Frontiers in Medicine +

|

March 28, 2024

View abstract on PubMed

Summary

This study developed a Random Survival Forests (RSF) model to predict liver failure prognosis. The RSF model, incorporating key clinical factors, demonstrated superior predictive accuracy compared to existing methods, aiding in risk stratification for better patient outcomes.

Area Of Science

Hepatology
Medical Informatics
Prognostic Modeling

Background

Liver failure prognosis with non-bioartificial support is often poor.
Identifying risk factors and developing predictive models are crucial for improving outcomes.
Current prognostic tools require enhancement for better clinical application.

Purpose Of The Study

To identify independent prognostic factors for liver failure.
To develop and validate novel prognostic models, including Nomogram and Random Survival Forests (RSF).
To compare the predictive performance of these models against established scores like MELD.

Main Methods

Retrospective analysis of 215 liver failure patients treated with non-bioartificial liver support.
Construction of Nomogram and RSF models using identified prognostic factors.
Evaluation of model performance using AUC in training and validation sets; risk stratification into low- and high-risk groups.

Main Results

Etiology, hepatic encephalopathy, total bilirubin, alkaline phosphatase, platelets, and MELD score were independent predictors of short-term prognosis.
The RSF model achieved superior predictive accuracy (AUC 0.863/0.792) over Nomogram (AUC 0.816/0.756) and MELD (AUC 0.658/0.700).
Patients stratified into a low-risk group exhibited significantly better prognoses than those in the high-risk group.

Conclusions

A novel RSF prognostic model was successfully developed for liver failure.
This RSF model demonstrates enhanced predictive power compared to Nomogram and MELD scores.
The model aids clinicians in optimizing treatment decisions and improving patient management.

Keywords:

liver failure multivariate cox regression nomogram non-bioartificial liver support system random survival forests