基于机器学习构建和验证慢性心力衰竭恶化风险预测模型OA
Construction and validation of risk prediction models for chronic heart failure deterioration based on machine learning
目的 基于机器学习算法构建6种慢性心力衰竭患者心力衰竭恶化的风险预测模型,并对其预测性能进行比较分析.方法 回顾性收集2019年1月—2023年12月合肥市第二人民医院收治的608例慢性心力衰竭患者作为研究对象,将其分为建模组(n=486)与内部验证组(n=122).同时,以同期蚌埠医科大学第一附属医院收治的400例慢性心力衰竭患者作为外部验证组.采用LASSO回归分析筛选慢性心力衰竭患者心力衰竭恶化的关键变量进行多因素分析,基于慢性心力衰竭患者心力衰竭恶化的独立危险因素,采用机器学习算法构建6种风险预测模型,并验证其性能.结果 本研究纳入1 008例慢性心力衰竭患者,其中294例心力衰竭恶化,发生率为29.17％.LASSO回归分析共筛选出13个关键变量,基于13个关键变量进行多因素Logistic回归分析,结果显示,心房颤动、左室射血分数、N末端脑利钠肽前体、尿酸、肌酐、情绪领域为慢性心力衰竭患者心力衰竭恶化的独立影响因素(P＜0.05).基于6个独立危险因素构建Logistic模型、决策树模型、神经网络模型、支持向量机模型、随机森林模型、XGBoost模型,经验证6种预测模型的曲线下面积(AUC)均＞0.8,其中以Logistic模型和XGBoost模型的预测效能最佳.结论 基于机器学习算法构建的6个适用于慢性心力衰竭患者心力衰竭恶化的风险预测模型,均具有较好的预测性能,但考虑临床适用性和便捷性,Logistic模型可能具有更高的临床应用价值,可为心力衰竭恶化的早期识别及防治方案制订提供参考.
Objective To construct six risk prediction models for heart failure deterioration in patients with chronic heart failure based on machine learning algorithms and conduct a comparative a-nalysis of their predictive performance.Methods A retrospective collection of 608 CHF patients in Hefei Second People's Hospital from January 2019 to December 2023 was conducted as the study sub-jects,and they were randomly divided into modeling group(n=486)and internal validation group(n=122)in a 4∶1 ratio.Additionally,400 chronic heart failure patients in the First Affiliated Hos-pital of Bengbu Medical University in the same period were collected as external validation group.LASSO regression analysis was used to screen key variables for heart failure deterioration in chronic heartfailure patients for multivariate analysis.Based on the independent risk factors for heart failure deterioration in chronic heart failure patients,six risk prediction models were constructed using ma-chine learning algorithms,and their performance was validated.Results A total of 1 008 chronic heart failure patients were included in this study,among whom 294 experienced heart failure deterioration,with an incidence rate of 29.17％.LASSO regression analysis identified 13 key variables.Multivariate analysis based on these 13 key variables revealed that atrial fibrillation,left ventricular ejection fraction,N-terminalpro-brain natriuretic peptide,uricacid,creatinine,and the emotional domain were inde-pendent risk factors for heart failure deterioration in chronic heart failure patients(P＜0.05).Logistic,decision tree,neural network,support vector machine,random forest,and XGBoost mod-els were constructed based on these six independent risk factors.Validation showed that the area un-der the curve(AUC)of all six prediction models was larger than 0.8,with the Logistic and XGBoost models demonstrating the best predictive performance.Conclusion This study construc-ted six risk prediction models for heart failure deterioration in chronic heart failure patients based on machine learning algorithms,all of which exhibite good predictive performance.However,consider-ing clinical applicability and convenience,the Logistic model may have higher clinical application value and can provide a reference for the early identification of heart failure deterioration and the for-mulation of prevention and treatment plans.
王静;方宁;黄容
合肥市第二人民医院心内科,安徽 合肥,230000合肥市第二人民医院心内科,安徽 合肥,230000合肥市第二人民医院心内科,安徽 合肥,230000
医药卫生
心力衰竭疾病恶化机器学习预测模型风险评估列线图情绪障碍模型验证
heart failuredisease deteriorationmachine learningprediction modelrisk as-sessmentnomogramemotional disordermodel validation
《实用临床医药杂志》 2026 (6)
103-110,8
2024年度安徽省中医药学会中医药科研项目(2024ZYYXH044)2022年度蚌埠医学院科研课题(2022byzd206)
评论