Construction of a 30-day readmission risk prediction model for COPD patients based on multiple machine learning algorithms
10.3760/cma.j.cn115682-20250107-00110
- VernacularTitle:基于多种机器学习的COPD患者30 d再入院风险预测模型研究
- Author:
Yujing SHI
1
;
Yuanguo WANG
;
Yu SHI
;
Shufang WANG
;
Li WEI
Author Information
1. 天津医科大学总医院心胸外科，天津 300052
- Publication Type:Journal Article
- Keywords:
Chronic obstructive pulmonary disease;
Readmission;
Machine learning;
Risk prediction
- From:
Chinese Journal of Modern Nursing
2025;31(31):4239-4247
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop and validate a 30-day readmission risk prediction model for patients with chronic obstructive pulmonary disease (COPD) using multiple machine learning algorithms.Methods:Convenience sampling was used to select 1 450 COPD patients hospitalized at Tianjin Medical University General Hospital from January 2017 to December 2023 as study subjects. Twenty-nine variables associated with readmission were included. LASSO was used to screen for primary characteristic variables associated with 30-day readmission. The 1 450 patients were divided into a training set ( n=870) and a test set ( n=580) in a 6∶4 ratio. Ten machine learning methods, including random forest, AdaBoost, extreme radient Boosting (XGB), decision tree and so on, were used for model training and testing to identify the optimal prediction model. The optimal model and SHAP were employed to analyze key features and rank characteristic importance. Results:Among 1 450 COPD patients, the 30-day readmission rate was 24.48% (355/1 450). There were no significant differences in the general information between patients in test set and training set ( P>0.05). LASSO regression analysis identified seven variables with the highest predictive value, namely regular weekly exercise, hospital stay, mean arterial pressure, forced expiratory volume in one second/forced vital capacity (FEV1/FVC), C-reactive protein, body mass index, and medication adherence. Machine learning showed that in the training set, XGB had the highest area under the receiver operating characteristic curve ( AUC), sensitivity, and F1 score of 0.943, 0.926, and 0.930, respectively. In the test set, the AUC and accuracy of XGB were 0.882 and 0.858, respectively, and XGB's various scores showed that it had good generalization and predictive performance. XGB analysis showed that medication adherence, FEV1/FVC, and regular weekly exercise were negatively correlated with the 30-day readmission risk, while body mass index, C-reactive protein, mean arterial pressure, and hospital stay were positively correlated. The characteristics ranked in order of importance were medication adherence, body mass index, C-reactive protein, mean arterial pressure, FEV1/FVC, hospital stay and regular weekly exercise. Conclusions:The XGB model has strong predictive performance and good generalization ability, which can effectively predict the 30-day readmission risk of COPD patients, assist in clinical identification of high-risk patients, implement nursing interventions, and reduce readmission rates.
