Related Experiment Videos
Development and external validation of a machine learning model for cardiovascular risk prediction in individuals
Ankang Zhu1, Shuai Wei2, Haobo Wang2
1Department of Thoracic Surgery, Shengli Clinical Medical College of Fujian Medical University; Fuzhou University Affiliated Provincial Hospital, Fuzhou, Fujian, China.
Insights
A new machine learning model accurately predicts cardiovascular disease (CVD) risk in patients with chronic lung disease (CLD). This tool aids early detection and personalized interventions for this high-risk group.
Area of Science:
- Cardiology
- Pulmonology
- Artificial Intelligence in Healthcare
Background:
- Patients with chronic lung disease (CLD) face elevated cardiovascular disease (CVD) risk.
- Existing risk assessment tools are insufficient for this population.
- A need exists for tailored CVD risk prediction in CLD patients.
Purpose of the Study:
- To develop and validate a machine learning (ML) model for predicting concurrent CVD risk in CLD patients.
- To interpret the ML model's predictive mechanisms using SHapley Additive exPlanations (SHAP).
- To provide a practical tool for CVD risk stratification in primary care.
Main Methods:
- Utilized the China Health and Retirement Longitudinal Study (CHARLS) cohort (n=2,639) for model development.
- Employed logistic regression for feature selection and compared seven ML algorithms.
- Validated the optimal model externally using the English Longitudinal Study of Ageing (ELSA) cohort (n=1,303).
- Applied SHAP for model interpretability and developed a web application.
Main Results:
- Identified 8 core predictors: age, BMI, depression, hypertension, dyslipidemia, IADL, and medication history.
- The XGBoost model achieved AUCs of 0.838 (training), 0.797 (testing), and 0.695 (external validation).
- SHAP analysis highlighted hypertension, depression, and age as key predictors, revealing synergistic effects.
Conclusions:
- The XGBoost-based ML model accurately predicts CVD risk in CLD patients.
- SHAP analysis provides crucial insights into predictive factors.
- The developed tool offers decision support for risk stratification and personalized interventions in primary care.
Background:
Patients with chronic lung disease (CLD) are at a significantly increased risk of developing cardiovascular disease (CVD); however, specific risk assessment tools tailored for this high-risk population are currently lacking. This study aimed to develop, validate, and interpret a machine learning model specifically designed to predict the risk of concurrent CVD in patients with CLD.
Methods:
Based on the China Health and Retirement Longitudinal Study (CHARLS) cohort, 2,639 patients with CLD were included. Core features were selected using univariate and multivariate logistic regression. Seven machine learning algorithms were systematically compared. After identifying the optimal model, external validation was conducted using the English Longitudinal Study of Ageing (ELSA) cohort (n = 1,303). The SHapley Additive exPlanations (SHAP) framework was employed to interpret the model's predictive mechanisms, and an interactive web application was developed based on the optimal model.
Results:
The study ultimately identified 8 core predictors: age, body mass index (BMI), depression score, hypertension, dyslipidemia, impaired instrumental activities of daily living (IADL), and medication history for lung diseases and lipid-lowering drugs. The XGBoost model demonstrated the best performance, achieving Area Under the Curve (AUC) values of 0.838, 0.797, and 0.695 in the training, testing, and external validation sets, respectively, while exhibiting excellent calibration and clinical net benefit. SHAP analysis revealed that hypertension, depression score, and age were the primary contributing variables, and confirmed a significant synergistic amplification effect between lipid metabolism and psychophysical functional indicators.
Conclusion:
The model constructed based on the XGBoost algorithm can accurately and robustly predict CVD risk in patients with CLD. Coupled with SHAP interpretability analysis and the online prediction tool, this study provides reliable digital decision support for CVD risk stratification, early identification, and personalized intervention among patients with CLD in primary care settings.
