Original TOC summary: We used SHAP, an explainable machine learning (ML) technique, to determine the risk predictive value and age interaction of coronary artery calcium (CAC) characteristics among 63,215 asymptomatic patients in the CAC consortium. The addition of CAC density and number of calcified vessels to an ML model with clinical characteristics + CAC did not improve prediction for all-cause mortality (p = 0.23), but did improve for cardiovascular mortality (p = 0.03). Lower CAC density increased mortality, particularly very low CAC density ≤0.75, which occurred predominantly in CAC1-100. Explainable ML should be applied in clinical research for transparent predictive modeling.