DescriptionAtherosclerotic cardiovascular disease (ASCVD) and subsequent adverse cardiovascular events remain highly prevalent in the U.S., making primary prevention an important goal. While the 2013 ACC/AHA Pooled Cohort Equations (PCE) remains the gold standard for cardiovascular event prediction, not represented in the model is cardiac electrophysiology, a major cause of sudden cardiac death. The electrocardiogram (ECG), a routinely available test that reflects one’s electrophysiologic health, may thus be useful for cardiovascular risk stratification in addition, and in comparison, to the PCE. Given the automated and highly correlated nature of its measurements, ECG data are well suited for analysis via machine learning. In this work, the value of aggregated ECG measurements for prediction of cardiovascular mortality is assessed in a nationwide cohort (NHANES III), via a comparative analysis of traditional survival analysis and machine learning methods. Overall, machine learning models could predict 10-year cardiovascular mortality with superior accuracy and event detection capacity compared to the PCE. Interestingly, only demographic and ECG data were necessary for such improved performance. Variable comparison between different prediction models provided insight into the relative importance of specific ECG components and the detection of silent myocardial infarctions as a possible underlying mechanism.