Faculty, Staff and Student Publications

Language

English

Publication Date

11-18-2025

Journal

Scientific Reports

DOI

10.1038/s41598-025-24369-x

PMID

41254126

PMCID

PMC12627085

PubMedCentral® Posted Date

11-18-2025

PubMedCentral® Full Text Version

Post-print

Abstract

Liver cancer has a high incidence and mortality rate globally, particularly in patients requiring intensive care unit (ICU) admission. Early prediction of in-hospital mortality for these patients is crucial, yet lacking reliable tools. This study aims to develop and evaluate machine learning (ML) models for predicting in-hospital mortality in critically ill liver cancer patients admitted to the ICU. This retrospective study used data from the MIMIC-III and MIMIC-IV databases, including 862 patients from MIMIC-III (training cohort) and 692 patients from MIMIC-IV (validation cohort). The study focused on patients diagnosed with liver cancer, identified by specific ICD codes. Four ML algorithms, namely logistic regression, random forest, XGBoost, and LightGBM, were used to predict in-hospital mortality based on clinical characteristics, laboratory results, and severity scores. Performance was evaluated using accuracy, AUROC, AUPRC, F1 score, and Kaplan-Meier curves. A total of 1,554 patients were included. The random forest model demonstrated the best performance, with an AUROC of 0.911 (95% CI: 0.855-0.956) and an AUPRC of 0.823 (95% CI: 0.718-0.905) in the internal test set, and an AUROC of 0.857 (95% CI: 0.826-0.889) and an accuracy of 0.828 (95% CI: 0.802-0.857) in the external validation set. Kaplan-Meier curves showed that all four models effectively stratified high-risk and low-risk groups. Key features influencing the prediction included APSIII, SAPSII, LODS, OASIS, and vital signs such as heart rate, temperature, and oxygen saturation. Feature importance analysis revealed that clinical severity scores played a major role in predicting mortality. This study demonstrates the potential of machine learning algorithms, particularly random forest, in predicting in-hospital mortality for critically ill liver cancer patients in the ICU. The identified clinical features provide valuable insights for clinicians in assessing patient risk and making timely interventions.

Keywords

Humans, Machine Learning, Female, Male, Middle Aged, Intensive Care Units, Hospital Mortality, Liver Neoplasms, Aged, Retrospective Studies, Prognosis, Algorithms, Liver cancer, In-hospital mortality, Machine learning, Intensive care unit

Published Open-Access

yes

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.