Early Identification Model of Diuretic Resistance in Heart Failure Patients based on Machine Learning Methods
10.11783/j.issn.1002-3674.2025.04.008
- VernacularTitle:基于机器学习方法构建心衰患者利尿剂抵抗早期识别模型
- Author:
Haozhe HUANG
1
;
Jing GUAN
;
Chao FENG
Author Information
1. 天津大学数学学院 300350
- Publication Type:Journal Article
- Keywords:
Heart failure;
Diuretic resistance;
LASSO regression;
Feature selection;
Machine learning
- From:
Chinese Journal of Health Statistics
2025;42(4):519-526
- CountryChina
- Language:Chinese
-
Abstract:
Objective This study aims to explore the diagnostic value of medical indicators in identifying diuretic resistance(DR)among patients with heart failure(HF)using machine learning methods such as Least Absolute Shrinkage and Selection Operator(LASSO)regression.An early recognition model for diuretic resistance in heart failure patients is constructed to provide clinicians with timely intervention opportunities and reduce hospitalization duration.Methods 267 heart failure patients from Tianjin Chest Hospital were included in the study.Standard urine output and DR status were employed as dependent factors,whereas 64 clinical indicators were gathered as independent variables.Depending on the kind of dependent variable,two modeling strategies were used.The first method used LASSO regression to fit the model,and 10-fold cross-validation was used to find the ideal regularization parameter.Standard urine output was used to predict diuretic resistance in the test set.Root mean square error(RMSE),coefficient of determination(R2),and classification accuracy were employed to assess the model's performance.In the second method,the most pertinent subset of features in relation to DR status was selected utilizing the max-relevance and min-redundancy(mRMR)algorithm.After that,three machine learning classifiers were used for classification:support vector machines(SVM),random forests(RF),and logistic regression(LR).Accuracy,precision,recall,area under the receiver operating characteristic curve(AUC),and F1-score were used to evaluate the model's performance.Results The optimal LASSO parameter was 0.0161.On the test set,the RMSE was 0.3489,the R2 was 0.7200,and the classification accuracy was 91.36%.The AUCs for mRMR+LR,mRMR+RF,and mRMR+SVM were 0.94,0.89,and 0.95,respectively.Their accuracies were 0.86,0.77,and 0.90;precisions were 0.87,0.80,and 0.90;recalls were 0.86,0.75,and 0.90;and F1-scores were 0.86,0.75,and 0.90,respectively.In the validation set,the LASSO regression model outperformed the other machine learning methods in predictive performance.Conclusion Diuretic resistance in heart failure patients was successfully detected by both the LASSO regression model and mRMR-based machine learning classifiers.The LASSO regression model performed better than the mRMR-based classifiers,providing more accurate scientific evidence for early clinical intervention.
