Using machine learning models to predict coronary artery calcium scores in firefighters.

Area of Science:

• Cardiovascular disease prediction
• Occupational health
• Machine learning in healthcare

Background:

• Firefighters face increased cardiovascular disease risk.
• Early detection of coronary artery calcium (CAC) is crucial for preventive strategies.
• Traditional risk prediction models may have limitations in this population.

Purpose of the Study:

• To develop and compare machine learning (ML) models for CAC prediction in firefighters.
• To evaluate the cross-validated performance of ML models against binary logistic regression (BLR).

Main Methods:

• Utilized health records of 416 firefighters.
• Assessed CAC using cardiac computed tomography and Agatston scores.
• Developed and compared XGBoost, Random Forest, SVM, Naïve Bayes, and KNN models against BLR using 17 clinical and lifestyle variables.

Main Results:

• Age, glucose, monocyte percentage, and systolic blood pressure were positively associated with CAC.
• Sodium levels, GFR, and maximum oxygen volume were inversely associated with CAC.
• XGBoost achieved the highest cross-validated AUC (0.770), outperforming other ML models and BLR (0.658).

Conclusions:

• ML algorithms, particularly XGBoost, are effective in predicting CAC in firefighters.
• These models enhance early detection and preventive strategies for cardiovascular health in this occupational group.
• Proactive health management using advanced predictive tools is vital for firefighters.