Comparison of predictive performance of three machine learning algorithms for frailty risk in elderly heart failure patients
10.3969/j.issn.1009-0126.2025.10.010
- VernacularTitle:三种机器学习算法对老年心力衰竭患者衰弱风险预测性能的比较
- Author:
Xin ZHANG
1
;
Xuemei ZHOU
;
Meng LI
;
Jiamin TANG
;
Danni MA
;
Hong HE
Author Information
1. 223800 宿迁,江苏省人民医院宿迁医院心血管内科
- Publication Type:Journal Article
- Keywords:
heart failure;
frailty;
forecasting
- From:
Chinese Journal of Geriatric Heart Brain and Vessel Diseases
2025;27(10):1330-1334
- CountryChina
- Language:Chinese
-
Abstract:
Objectives To construct frailty risk prediction models based on logistic regression anal-ysis,decision tree and random forest algorithm in elderly patients with heart failure(HF),and to compare the predictive performance of three models.Methods A total of 426 elderly HF patients hospitalized in the Affiliated Hospital of Nantong University from September 2022 to October 2023 were selected using convenience sampling.Based on the results of frailty assessment,194 of them were classified into the frail group and the other 232 into the non-frail group.The 426 patients were divided into training(299 casses)and testing sets(127 cases)in a 7∶3 ratio.Three prediction models were then constructed in the training set,while the test set was used to validate the results.Area under curve and confusion matrix were used to measure performance of the mod-els.The optimal model was then selected by evaluating the performance on the testing set.Results The area under curve value of the logistic regression model,decision tree model and random forest model in the testing set was 0.898,0.825 and 0.903,with a classification accuracy of 84.25％,77.95％and 83.46％,a sensitivity of 82.76％,68.97％and 82.76％,a specificity of 85.51％,85.51％and 84.06％,a positive predictive value of 82.76％,80.00％and 81.36％,and a negative predictive value of 85.51％,76.62％and 85.29％,respectively.The factors that ultimately affecting frailty in elderly HF patients were age,left atrial diameter,depression,albumin,physical activity level and social support.Conclusion Among the three prediction models,the logistic regression model demonstrates best predictive performance for frailty risk in elderly HF patients than the decision tree and random forest models.
