In a large, real‑world study across 18 hospitals, software that reads the first electrocardiogram (ECG) matched a leading bedside score at identifying heart attacks and improved risk triage when combined with it.
What’s new
Researchers in the Republic of Korea tested an artificial‑intelligence tool that analyzes a standard 12‑lead ECG to estimate a patient’s likelihood of acute myocardial infarction (heart attack). Among 8,493 adults who arrived at emergency departments within 24 hours of chest‑pain symptoms (March 2022–October 2023), 1,586 (18.6%) were ultimately diagnosed with heart attack. The AI‑ECG’s overall accuracy for diagnosing heart attack was essentially the same as the widely used HEART score (a composite that includes blood troponin testing), and better than several other clinician tools.
According to the study’s structured graphical abstract on page 2, the AI result is available in about 10 minutes from arrival—before lab results return—positioning it as an early triage aid.
Key findings (in plain language)
- Early “rule‑out” with very few misses: Using its pre‑set low‑risk threshold, AI‑ECG classified 8.2% of patients as low risk with a negative predictive value of 99.1% (≈ <1% missed heart attacks)—a commonly accepted safety benchmark in emergency medicine. No STEMI (major, artery‑blocking heart attack) occurred in the AI‑defined low‑risk group.
- Early “rule‑in”: AI‑ECG labeled 22% of patients high risk with a positive predictive value of 60.4%, helping clinicians prioritize who needs urgent attention. (See the flowchart on pages 9–10.)
- As good as the HEART score—faster: Overall diagnostic performance (AUROC 0.878) matched the HEART score (0.877), which typically requires an hour or more for troponin lab results. (See Table 2 on page 7 and ROC curves on page 8.)
- Better together: When the team combined AI‑ECG with the HEART score, triage improved further—reclassifying patients more appropriately (net reclassification improvement +19.6%) and boosting discrimination (C‑index 0.926 vs. 0.877 for HEART alone). (See Table 3 on page 10.)
- 30‑day safety signal: The AI‑ECG also predicted 30‑day major cardiac complications with accuracy similar to HEART (AUROC 0.866 vs. 0.858).
Why it matters
In overcrowded ERs, minutes count. Today’s triage often waits on blood tests; AI that reads the first ECG could help clinicians sort low‑ and high‑risk patients faster—speeding care for those who need it and easing bottlenecks for those who don’t. The diagram on page 2 highlights this potential “front‑door” role for AI‑ECG in the first ~10 minutes of care.
Important caveats
- Who was studied: All patients were treated in Korea; results need validation in other countries and health systems.
- Not perfect for everyone: Performance dipped somewhat in adults ≥65 and in people with certain conditions (e.g., chronic kidney disease, prior heart disease) or complex ECG patterns (e.g., bundle branch block, pacemaker rhythm). Clinicians should interpret AI results in context. (See subgroup notes around pages 8–11.)
- Conflicts and funding: Several authors are employees or stakeholders of the company that developed the AI model (AiTiAMI), which also helped fund the work.
- Short‑term outcomes: The study tracked diagnosis during the initial visit and complications through 30 days; longer‑term impact and real‑world workflow effects still need study.
Bottom line
An AI tool that reads the first ECG can help ER teams rule out heart attacks safely in a subset of patients and flag those at high risk within minutes. Used alongside existing protocols—not as a replacement—it could sharpen triage and accelerate care, pending broader validation.
Source: European Heart Journal, ROMIAE multicentre study on AI‑ECG for acute myocardial infarction (2025).
Editor’s note: This article is for information only and is not a substitute for professional medical advice.