Construction of a COPD risk prediction model based on machine learning and the COPD-SQ questionnaire
10.19405/j.cnki.issn1000–1492.2026.07.014
- VernacularTitle:基于机器学习和COPD-SQ问卷的COPD风险预测模型构建
- Author:
Lin CHEN
1
;
Luna ZHAO
1
;
Yue ZHOU
2
;
Panpan WANG
3
;
Jingkun LI
2
;
Wenwen ZHANG
1
;
Xinxin ZHANG
1
;
Chao WU
1
;
Dong LIU
1
Author Information
1. Department of Respiratory and Critical Care Medicine， The First Affiliated Hospital of Shihezi University， Shihezi 832000
2. School of Clinical Medicine， Shihezi University， Shihezi 832000
3. School of Mathematics， Southwest Jiaotong University， Chengdu 611756
- Publication Type:Journal Article
- Keywords:
chronic obstructive pulmonary disease;
risk prediction model;
machine learning;
Logistic regression;
class imbalance;
screening
- From:
Acta Universitatis Medicinalis Anhui
2026;61(7):1261-1268
- CountryChina
- Language:Chinese
-
Abstract:
ObjectiveTo construct and evaluate various machine learning models for predicting the risk of chronic obstructive pulmonary disease （COPD） in individuals， thereby providing data support for early screening and intervention. MethodsA total of 823 subjects were selected for this study， comprising 142 individuals in the high-risk group for COPD and 681 individuals in the low-risk group. Data collected included demographic characteristics， smoking history， symptoms （such as cough and shortness of breath）， and scores from the Chronic obstructive pulmonary disease screening questionnaire. Four machine learning algorithms—Logistic regression， random forest， support vector machine， and XGBoost—were utilized to construct risk prediction models. The performance of these models was assessed using 5-fold cross-validation， with evaluation metrics including accuracy， precision， recall， F1-score， area under the receiver operating characteristic curve （AUC）， and average precision （AP）. Furthermore， a feature importance analysis was performed. ResultsThe Logistic Regression model exhibited superior performance， achieving an AUC of 0.982 and an AP of 0.939. This was closely followed by the Random Forest model， which recorded an AUC of 0.975 and an AP of 0.890. Feature importance analysis revealed that smoking history， symptoms of shortness of breath， and body weight were significant predictors. All models demonstrated robust performance in identifying low-risk populations； however， variations were observed in their efficacy in identifying high-risk populations. ConclusionMachine learning models have proven effective in identifying individuals at high risk for COPD. Among these， the Logistic regression model exhibits the best overall performance， efficiently identifying high-risk populations and serving as a valuable clinical auxiliary screening tool. Various models， each with distinct performance characteristics， are suited to different clinical screening scenarios， thereby offering targeted decision-making support for the establishment of a hierarchical and intelligent COPD screening pathway.
