【摘要】 目的 利用4种机器学习算法，构建针对心力衰竭患者认知衰弱的风险预测模型，并通过模型评价指标，选出具有最佳预测性能的模型算法，为患者的精准医疗管理与干预提供有力支持。方法 本研究采用便利抽样法，选取2023年3月至12月佛山市2所三级甲等医院的心力衰竭患者为研究对象，通过一般资料调查表、蒙特利尔认知评估量表、Fried衰弱表型量表、主观认知下降量表、微型评估精法、患者健康问卷对患者进行问卷调查，基于Logistic回归、极端梯度提升算法、随机森林、决策树4种机器学习算法，按照7∶3比例，将数据集划分为训练集和测试集，并利用这些数据构建HF患者认知衰弱的预测模型。采用准确率、召回率、精确率、F1分数和受试者工作特征曲线下面积评价预测模型的性能。结果 共回收247份有效问卷，HF患者认知衰弱患病率为44.94%。预测因子有NYHA心功能分级、日常生活活动能力、营养情况、抑郁状况、每周运动次数、年龄及NT-proBNP。随机森林模型准确率、召回率、精确率、F1分数和受试者工作特征曲线下面积均最高。结论 基于机器学习算法构建的心力衰竭患者认知衰弱的预测模型中，随机森林模型更适合用于心力衰竭患者认知衰弱风险的预测。更多还原

【Abstract】 Objective Four machine learning algorithms were used to construct a risk prediction model for cognitive frailty in heart failure patients. The model algorithm with the best prediction performance was selected through model evaluation indexes to provide strong support for precise medical management and intervention for patients. Methods In this study, 247 heart failure patients from two tertiary A hospitals in Foshan City from March 2023 to December 2023 were selected by using the convenience sampling method. In this study, patients were surveyed by means of a general information questionnaire, Montreal Cognitive Assessment, Frailty Phenotype, Subjective Cognitive Decline Questionnaire, Mini-Nutritional Assessment Short-Form, and Patient Health Questionaire-9. To build the predictive model for cognitive frailty in heart failure patients, Four Machine Learning algorithms, including Logistic Regression, Extreme Gradient Boosting Random Forest, Decision Tree, are employed. The dataset is divided into a training set and a test set in a 7 ∶3 ratio, and each algorithm was trained and evaluated accordingly. The efficiency of the prediction model performance was evaluated using Accuracy, Recall, Specificity, Precision, F1-Score, and Area Under Curve. Results A total of 247 valid questionnaires were returned, and the prevalence of cognitive frailty in heart failure patients was 44.94%. Predictors were NYHA cardiac function class, grip strength status, nutritional status, depression status, number of exercise sessions per week, age and NT-proBNP. Random forest model had the highest accuracy, recall, precision, F1 score and area under the curve of the subjects’ work characteristics. Conclusion Based on the prediction models constructed based on Machine Learning algorithms for cognitive frailty in heart failure patients, the Random Forest model is more suitable for the prediction of the risk of cognitive frailty in heart failure patients.更多还原