Prediction of Death Risk in Heart Failure Patients Based on Explainable Machine Learning Models

DOI: https://doi.org/10.62517/jmhs.202605101

Author(s)

Yue Yan, Hao Zhu*, Luyao Zhou, Qi Yang, Canyan Liao, Wei He, Baolin Zhou

Affiliation(s)

College of Physics and Information Engineering, Zhaotong University, Zhaotong, China *Corresponding Author

Abstract

This study was designed to construct an interpretable machine learning model for predicting mortality risk in intensive care unit (ICU) patients with heart failure. Data were obtained from the MIMIC-IV and eICU databases, comprising 30,411 heart failure patients, and were subsequently split into training and testing sets. The study employed four machine learning algorithms—eXtreme Gradient Boosting (XGBoost), random forest (RF), Classification and Regression Trees (CART), and logistic regression (LR)—to build predictive models for adverse outcomes. The SHapley Additive exPlanations (SHAP) method was applied to interpret the model and identify key prognostic factors. The optimal model was selected according to its predictive accuracy and the area under the receiver operating characteristic curve (AUC). Among the four models, the XGBoost model demonstrated the highest predictive performance, achieving an AUC value of 0.88. Finally, the optimal model is explained using the Shapley Additive Explanation value. The SHAP value shows that the average value of APSIII is the most important predictive variable. An interpretable machine learning model not only performs well in predicting mortality rates in heart failure patients, but is also crucial for clinicians to develop personalized prevention and treatment plans.

Keywords

Heart Failure; Machine Learning; Extreme Gradient Boosting; Shapley

References

[1]Alexander R L. Heart failure resulting from cancer treatment: still serious but an opportunity for prevention. systematic reviews, 2018, 7(1):313-323. [2]Mahoto K. The Concept of Heart Failure: Chronic Diseases Accompanied by an Attack of Acute Exacerbation. Tokyo: Therapeutic Strategies for Heart Failure. 2018:1-15. [3]Peng Peichi, Liu Wenxian, Zhao Han, et al Analysis of prognostic factors for heart failure caused by hypertension Chinese Journal of Modern Medicine, 2018 (4): 9-13. [4]Chen L M, Levine D A, Hayward R, et al. Relationship between Hospital 30-Day Mortality Rates for Heart Failure and Patterns of Early Inpatient Comfort Care. Journal of Hospital Medicine, 2018, 13(3):170-176. [5]Rohan K, Kumar D, Harlan M. et al. Rising Mortality in Patients With Heart Failure in the United States. Jacc Heart Failure, 2018, 6(7):610-612. [6]Ma Liyuan, Wu Yazhe, Wang Wen, et al Interpretation of Key Points in the 2017 China Cardiovascular Disease Report Chinese Journal of Cardiology, 2018, 23 (1): 3-6. [7]Aaronson K D, Schwartz J S, Chen T M, et al. Development and prospective validation of a clinical index to predict survival in ambulatory patients referred for cardiac transplant evaluation. Circulation, 1997, 95(12):2660-2667. [8]Gao, Z. Y. Research on Deep Learning Methods for Heart Failure Risk Assessment Based on Electronic Health Records. University of Science and Technology Beijing, 2025. DOI: 10.26945/d.cnki.gbjku.2025.000095. [9]ZENG Jing, HE Xiaolong, HU Huajuan, et al., Construction of risk prediction model for predicting death or readmission in acute heart failure patients during vulnerable phase based on machine learning, Journal of Army Medical University, 2024, 46(7): 738-745, http://dx.doi.org/10.16016/j.2097-0927.202312147. [10]Xu Qian, Xu Cuirong, Cai Xue, et al. Research Progress of Machine Learning-Based Prediction Models for Heart Failure Risk. Journal of Nursing Science, 2024, 52(5): 807-815. [11]Xie Qiuhua, Lu Zuohua, Deng Shengqiong, et al. Evaluation of Heart Failure Discrimination Model for NT-proBNP Grey Value Patients Based on Machine Learning Algorithm. Journal of Tongji University: Medicine, 2021, 42 (3) : 6. DOI: 10.12289 / j.i SSN. 1008-0392.20433.