Machine Learning Adds to Clinical and CAC Assessments in Predicting 10-Year CHD and CVD Deaths

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Coronary artery calcium (CAC) scoring predicts cardiovascular disease (CVD) risk but current methods don’t integrate all available data.
• Comprehensive risk assessment requires integrating diverse clinical and imaging variables.

Purpose of the Study:

• To evaluate if machine learning (ML) using noncontrast computed tomographic (CT) and clinical variables enhances prediction of atherosclerotic cardiovascular disease (ASCVD) and coronary heart disease (CHD) deaths.
• To compare ML model performance against traditional coronary artery calcium (CAC) Agatston scoring and clinical data.

Main Methods:

• A cohort of 66,636 asymptomatic subjects without established ASCVD was analyzed.
• An ensemble boosting ML approach incorporated 77 variables, including CAC score, plaque characteristics, and extracoronary scores.
• Model performance was assessed using 10-fold cross-validation and area under the curve (AUC), comparing against ASCVD risk score, CAC score, ML clinical, and ML CT models.

Main Results:

• The comprehensive ML model (ML all) achieved superior AUC for predicting CVD death (0.845) versus ASCVD risk (0.821), CAC score (0.781), and ML CT (0.804).
• Similarly, for CHD death prediction, ML all (0.860) outperformed ASCVD risk (0.835), CAC score (0.816), and ML CT (0.827).
• All comparisons showed statistically significant improvements (p < 0.001).

Conclusions:

• A comprehensive ML model integrating CT and clinical data significantly improves the prediction of CVD and CHD death.
• This advanced ML approach demonstrates superior prognostic utility compared to traditional risk stratification methods.
• The findings support the use of integrated ML models for more accurate cardiovascular risk assessment.