Interpretable machine learning for prognostic prediction in critically ill patients with coronary artery disease: a multicenter study

Read full text Summarise with VIRAT
Authors: Shuai Yang, Shuo Zhang, Lianzheng Ma, Jinfang Zeng, Mengdi Wang, Shunbin Huang, Xun Zhang, Xiao Liang, Minmin Zhu
Cited by: 0
References: 33

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Background: Coronary artery disease (CAD) ranks among the most prevalent and clinically challenging cardiovascular disorders encountered in the intensive care unit (ICU). Patients with CAD admitted to the ICU typically exhibit elevated mortality rates, intricate pathophysiological alterations, and a high likelihood of adverse outcomes. This study aims to develop and validate a prognostic prediction model for ICU-admitted CAD patients using machine learning (ML) methodologies. Methods: The data were retrieved from two independent cohorts within the Medical Information Mart for Intensive Care (MIMIC) database: MIMIC-IV was utilized for model training, while MIMIC-III served as an external validation dataset. The primary endpoints of the prediction were the 28- and 365-day mortality risks in this patient population. Feature selection was performed using LASSO regression integrated with commonality analysis, and feature importance was quantified via the SHapley Additive exPlanations (SHAP) approach to identify critical risk factors. Subsequently, short-term and long-term mortality risk prediction models for patients with coronary artery disease were developed based on seven interpretable machine learning algorithms. Results: A total of 15,930 patients with coronary artery disease were enrolled in this study (mean age, 70.3 ± 12.1 years; 5,055 females, accounting for 31.7%). To evaluate the mortality risk of patients across different time horizons, we developed predictive models incorporating 40 and 41 feature variables, respectively. Comparative analyses with six other machine learning algorithms revealed that the RandomForest algorithm exhibited the optimal performance in predicting both short-term and long-term mortality risks among patients with coronary artery disease [28-day mortality risk: Internal validation: AUC = 0.858, 95% CI: 0.843-0.872; Accuracy = 88.2%; External validation: AUC = 0.914, 95% CI: 0.904-0.923; Accuracy = 91.4%] [365-day mortality risk: Internal validation: AUC = 0.851, 95% CI: 0.840-0.863; Accuracy = 79.6%; External validation: AUC = 90.1, 95% CI: 0.893-0.909; Accuracy = 85.3%]. Conclusion: The random forest model developed in this study exhibited robust predictive performance and generalization capability in evaluating short-term and long-term mortality risks among critically ill patients with CAD. As a promising predictive tool, it offers data-driven decision support for clinicians to conduct early identification of high-risk patients and perform risk stratification, while its ultimate clinical utility remains to be further validated by prospective studies.

Fields of Study & Subjects

Machine Learning in Healthcare Artificial Intelligence in Healthcare Sepsis Diagnosis and Treatment

Publication details

Journal / source
Frontiers in Medicine
Volume
13
Pages
1794827–1794827
Publication date
30 March 2026
Language
EN