| Title: | Electrocardiogram-based deep learning score for coronary artery calcification reclassifies cardiovascular risk and identifies screening candidates |
| Journal: | BMC Medical Informatics and Decision Making |
| Published: | 21 Feb 2026 |
| Pubmed: | https://pubmed.ncbi.nlm.nih.gov/41723484/ |
| DOI: | https://doi.org/10.1186/s12911-026-03400-9 |
| URL: | http://dx.doi.org/10.1186/s12911-026-03400-9 |
Publication 17688
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Abstract
BackgroundCoronary artery calcium (CAC) scoring is included in major guidelines to guide statin decisions when conventional cardiovascular risk assessments are inconclusive. While computed tomography for CAC is impractical for routine use, electrocardiograms (ECGs) are widely available, offering broader opportunities for early detection. Extending our prior work demonstrating an ECG-based deep learning model for CAC prediction (ECG-CAC model), we aimed to develop a more robust model, while also comprehensively assessing its clinical utility.MethodsUsing nearly 200,000 standard 12-lead ECGs from Severance Hospital (SH), we developed an ECG-CAC model producing a risk score reflecting the likelihood of CAC. We utilized data from three health checkup centers and the United Kingdom Biobank (UKB) to evaluate its performance in predicting CAC, potential for guiding targeted CAC screening and statin therapy decisions, and association with major adverse cardiovascular events (MACE). MACE was defined as a composite of myocardial infarction, ischemic stroke, and cardiovascular death.ResultsThe ECG-CAC model achieved areas under the receiver operating characteristic curve of 0.741 (95% confidence interval: 0.732-0.749)/0.837 (95% confidence interval: 0.821-0.851) for predicting CAC scores > 0/≥400, with robust external validation. Among Pooled Cohort Equations (PCE) low-risk individuals, for whom statins are not recommended unless CAC-positive, CAC prevalence rose from 24.9% to 49.8% in those reclassified as ECG-CAC model high-risk (P < 0.001), and among PCE moderate-risk individuals, for whom statins are recommended unless CAC-negative, CAC absence rose from 35.4% to 53.9% in those reclassified as ECG-CAC model low-risk (P < 0.001), supporting targeted CAC screening in these groups. MACE incidence rate per 1000 person-years was higher in PCE low-risk individuals reclassified as ECG-CAC model high-risk than in PCE moderate-risk individuals reclassified as ECG-CAC model low-risk (SH: 6.4 vs. 3.4, P = 0.026; UKB: 8.1 vs. 5.1, P = 0.183), suggesting that statin therapy could be considered in the former and potentially deferred in the latter, pending further clinical evaluation. ECG-CAC model score was an independent MACE risk factor and improved prediction when combined with conventional tools.ConclusionsThe ECG-CAC model serves as an indicator for CAC and shows potential as a biomarker of coronary-cerebrovascular atherosclerotic burden, providing incremental value to conventional cardiovascular risk stratification tools.</p>
11 Authors
- Changho Han
- Seng Chan You
- Hyung-Chul Lee
- Jin Young Park
- Hong-Seok Lim
- ChulHyoung Park
- Hui-Nam Pak
- Oh-Seok Kwon
- Songsoo Kim
- Jung-Sun Kim
- Dukyong Yoon
1 Application
| Application ID | Title |
|---|---|
| 85037 | Prediction of drug response using pharmacogenomics-driven machine learning approach |
Enabling scientific discoveries that improve human health