AI-enhanced electrocardiography models achieved concordance correlation coefficients of 0.90 for both signal- and image-based composite structural heart disease detection across 731,466 ECG pairs from three major health systems. The image-based ensemble model demonstrated 87-89% categorical screen-status concordance when ECGs were repeated 1-30 days apart in the same patients. This represents a significant advance in cardiovascular AI reliability, addressing a critical barrier to clinical implementation. The consistency rivals traditional diagnostic tests, suggesting AI-ECG could serve as a dependable screening tool for structural heart disease in routine clinical practice. Particularly noteworthy is that patients showing negative-to-positive screen status changes faced 1.67-2.37 times higher risk of developing heart failure, indicating these AI fluctuations capture meaningful physiological changes rather than mere noise. However, younger patients under 65 showed higher discordance rates, and inpatient ECGs were less reliable than outpatient recordings. As a preprint awaiting peer review, these findings require validation before widespread clinical adoption. The work appears confirmatory rather than paradigm-shifting, strengthening the evidence base for AI-ECG as a reliable clinical decision support tool in cardiovascular care.
AI-ECG Models Show 90% Reliability for Heart Disease Screening
📄 Based on research published in medRxiv preprint
Read the original research →⚠️ This is a preprint — it has not yet been peer-reviewed. Results should be interpreted with caution and may change following peer review.
For informational, non-clinical use. Synthesized analysis of published research — may contain errors. Not medical advice. Consult original sources and your physician.