Unlocking the link: predicting cardiovascular disease risk with a focus on airflow obstruction using machine learning

Area of Science:

• Cardiology and Respiratory Medicine
• Data Science and Machine Learning

Background:

• Cardiovascular diseases (CVD) and respiratory diseases frequently coexist.
• Airflow obstruction (AO) severity is linked to CVD incidence and mortality.
• Current CVD risk models do not adequately account for AO as an independent risk factor.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting CVD risk.
• To assess the independent contribution of AO to CVD risk.
• To identify key predictors of CVD in the general population and in individuals with AO.

Main Methods:

• Utilized NHANES III and NHANES 2007-2012 datasets for participants over 40 with complete AO and CVD data.
• Employed logistic regression to analyze the AO-CVD association and six ML models (XGBoost optimized via RandomizedSearchCV) for CVD risk prediction.
• Evaluated models using AUC, accuracy, precision, recall, F1 score, Brier score, and SHAP for explainability.

Main Results:

• A significant positive correlation was found between AO and CVD prevalence (P < 0.05).
• The XGBoost model demonstrated optimal CVD risk prediction in the general population (AUC=0.7508) and those with AO (AUC=0.6645).
• Key predictors for the general population were age, hypertension, and PIR; for AO patients, education, gender, and race were most impactful.

Conclusions:

• AO is a significant independent risk factor for CVD.
• ML models, particularly XGBoost, effectively predict CVD risk, incorporating AO as a predictor.
• Identifying key demographic and clinical factors can facilitate early CVD risk assessment and intervention in at-risk populations.