基于机器学习建立老年慢性心力衰竭患者 1年全因死亡预测模型

Development of a machine learning-based model to predict one-year all-cause mortality in elderly patients with chronic heart failure

摘要

目的 基于机器学习建立老年慢性心力衰竭（CHF）患者1年全因死亡预测模型。方法 回顾性分析902例老年CHF患者（782例训练集和120例验证集）1年全因死亡的影响因素，随访1年，根据是否全因死亡分为死亡组和存活组，并建立5种机器学习预测模型。采用受试者操作特征（ROC）曲线、校准曲线、决策曲线对模型进行验证。结果 训练集782例老年CHF患者1年全因死亡率为12.28%；验证集120例老年CHF患者1年全因死亡率为10.83%。LASSO回归和多因素Logistic回归分析显示：体重指数增加、左室射血分数升高为老年CHF患者1年全因死亡的独立保护因素，纽约心脏协会心功能分级Ⅳ级、C反应蛋白升高、D-二聚体升高、氨基末端脑钠肽前体（NT-proBNP）升高为独立危险因素（P<0.05）。ROC曲线显示，训练集和验证集极端梯度提升（XGBoost）预测模型的曲线下面积（AUC）分别为0.897、0.864，均优于多因素Logistic回归（0.860、0.822）、决策树（0.767、0.761）、随机森林（0.875、0.818）、支持向量机（0.859、0.788）预测。验证集校准曲线显示，XGBoost预测模型的预测概率与实际曲线贴合，预测概率>0.10可为临床带来净收益。结论 体重指数、纽约心脏协会心功能分级、左室射血分数、C反应蛋白、NT-proBNP为老年CHF患者1年全因死亡的独立影响因素，基于此建立的XGBoost预测模型的预测效能最佳。

Abstract

Objective To develop a machine learning (ML)-based model for predicting 1-year all-cause mortality in elderly patients with chronic heart failure (CHF). Methods A ML-based model was built for predicting 1-year all-cause mortality in elderly patients with CHF. Results The 1-year all-cause mortality rate was 12.28% in the training set and 10.83% in the validation set. LASSO regression and multivariate logistic regression (LR) identified body mass index (BMI) and left ventricular ejection fraction (LVEF) as independent protective factors, while New York Heart Association (NYHA) class Ⅳ, elevated C-reactive protein (CRP), D-dimer (D-D), and N-terminal pro-brain natriuretic peptide (NT-proBNP) were independent risk factors (P < 0.05). Receiver operator characteristic (ROC) curve showed that the extreme gradient boosting (XGBoost) model had the highest area under the curve (AUC) in both training and validation sets (0.897 and 0.864, respectively), outperforming LR (AUC = 0.860, 0.822), decision tree (DT) (AUC = 0.767, 0.761), random forest (RF) (AUC = 0.875, 0.818), and support vector machine (SVM) (AUC = 0.859, 0.788). The calibration curve in the validation set indicated that the XGBoost model closely aligned predicted and observed probabilities. DCA showed clinical benefit when predicted probability exceeded 0.10. Conclusion BMI, NYHA class, LVEF, CRP, and NT-proBNP were independent predictors of one-year all-cause mortality in elderly CHF patients. The XGBoost model based on these variables demonstrated superior predictive performance.

关键词

老年 / 慢性心力衰竭 / 机器学习 / 全因死亡 / 预测模型 / 极端梯度提升算法 / 左室射血分数

Key words

elderly / chronic heart failure / machine learning / all-cause mortality / predictive model / extreme gradient boosting algorithm / left ventricular ejection fraction

参考文献

[1] 郇轩， 田梅香， 尹江燕， 等. 心力衰竭合并糖尿病住院患者临床特点的回顾性分析及意义[J]. 兰州大学学报（医学版）， 2023， 49（8）：68-75.

[2] PAVÍA—LÓPEZ A A, MAGAÑA—SERRANO J A, CIGARROA—LÓPEZ J A, et al. Clinical practice guidelines for diagnostic and treatment of the chronic heart failure[J]. Archivos de cardiologia de Mexico, 2024, 94(Supl 1): 1-74.

[3] LIU Z Y, LI Z P, LI X Q, et al. Global trends in heart failure from 1990 to 2019: an age—period—cohort analysis from the global burden of disease study[J]. ESC heart failure, 2024, 11(5): 3264-3278.

[4] WANG H, LI Y Y, CHAI K, et al. Mortality in patients admitted to hospital with heart failure in China: a nationwide cardiovascular association database—heart failure centre registry cohort study[J]. The lancet global health, 2024, 12(4): e611-e622.

[5] SUN Z Y, WANG Z H, YUN Z J, et al. Machine learning—based model for worsening heart failure risk in Chinese chronic heart failure patients[J]. ESC heart failure, 2025, 12(1): 211-228.

[6] 中华医学会老年医学分会心血管疾病学组， 《老年慢性心力衰竭诊治中国专家共识》编写组. 老年人慢性心力衰竭诊治中国专家共识（2021）[J]. 中华老年医学杂志， 2021， 40（5）：550-561.

[7] TUERSUN R, ABUDOUWAYITI A, LI Y X, et al. Serum CA125: a prognostic biomarker for mortality in chronic heart failure[J]. BMC cardiovascular disorders, 2025, 25(1): 227.

[8] 李晨， 江珊， 汪欣， 等. 三种衰弱评估工具对老年慢性心力衰竭患者不良结局预测能力的比较[J]. 中华老年心脑血管病杂志， 2024， 26（10）：1125-1129.

[9] MONTALTO M, D’IGNAZIO F, CAMILLI S, et al. Heart failure in older patients: an update[J]. Journal of clinical medicine, 2025, 14(6): 1982.

[10] LAU G H W, AU YEUNG T W, JOYCE E, et al. Consolidating concepts from recent international expert consensus on sarcopenia: implications for heart failure management[J/OL]. (2025—06—16)[ 2025—07—10]. Journal of cardiac failure, 2025: https://doi.org/10.1016/j.cardfail.2025.03.021.

[11] JENSEN G L, CEDERHOLM T, CORREIA M I T D, et al. GLIM consensus approach to diagnosis of malnutrition: a 5—year update[J]. Journal of parenteral and enteral nutrition, 2025, 49(4): 414-427.

[12] 赵灿， 王刚， 刘霄燕， 等. BMI对慢性心力衰竭的预后价值[J]. 临床心血管病杂志， 2024， 40（3）：194-198.

[13] XU Z, YUE Y B, XU M F, et al. The role of cut—off values for creatinine, blood urea nitrogen, and uric acid in prognostic assessment of chronic heart failure: a retrospective cohort study[J]. BMC cardiovascular disorders, 2025, 25(1): 209.

[14] 中国老年医学学会心电及心功能分会， 中国医师协会心血管内科分会， 中国心衰中心联盟专家委员会. 慢性心力衰竭加重患者的综合管理中国专家共识2022[J]. 中国循环杂志， 2022， 37（3）：215-225.

[15] 中华医学会超声医学分会超声心动图学组， 国家超声诊断专业医疗质量控制中心专家委员会. 经胸超声心动图检查规范化应用中国专家共识（2024版）[J]. 中华超声影像学杂志， 2024， 33（1）：1-13.

[16] ROMAN—PEPINE D, SERBAN A M, CAPRAS R D, et al. A comprehensive review: unraveling the role of inflammation in the etiology of heart failure[J]. Heart failure reviews, 2025, 30(5): 931-954.

[17] FANG Z, RAZA U, SONG J, et al. Systemic aging fuels heart failure: molecular mechanisms and therapeutic avenues[J]. ESC heart failure, 2025, 12(2): 1059-1080.

[18] SEGRETI A, MASTROBERARDINO S, FRAU L, et al. Severe heart failure and intracardiac thrombosis: going beyond the appearance for diagnosis and treatments[J]. Monaldi archives for chest disease, 2025, 95(2): 1-10.

[19] 中国医师协会检验医师分会心血管专家委员会. B型利钠肽及N末端B型利钠肽前体实验室检测与临床应用中国专家共识[J]. 中华医学杂志， 2022， 102（35）：2738-2754.

[20] SHANG S, WEI M, LV H S, et al. Construction and validation of a hospital mortality risk model for advanced elderly patients with heart failure based on machine learning[J]. International journal of general medicine, 2025, 18: 3277-3288.

[21] LI G J, ZHAO Z Q, YU Z L, et al. Machine learning for risk prediction of acute kidney injury in patients with diabetes mellitus combined with heart failure during hospitalization[J]. Scientific reports, 2025, 15(1): 10728.

基金资助

首都全科医学与社区卫生研究专项资助项目(2023-2Y-014)