AI-Enabled ECG Screening Finds the Structural Heart Disease Echo Would Have Missed
A composite AI-ECG signal predicted new structural disease years before it appeared — and the technology behind it may soon trade on a public exchange.
Structural heart disease (SHD) remains one of cardiology's most common blind spots.
Valvular disease, left ventricular hypertrophy, reduced ejection fraction, and pulmonary hypertension frequently go undetected until a patient presents with overt heart failure.
The reference standard, echocardiography, is accurate but constrained by cost, sonographer availability, and specialist interpretation.
New multinational data presented at New York Valves 2026 suggest that a simple 12-lead ECG, read by an artificial intelligence (AI) algorithm, can flag patients who need that echo before disease progresses.
Composite-positive patients were up to 3.75 times more likely to develop new SHD over six years than composite-negative patients.
Case Vignette
A 61-year-old with well-controlled hypertension presents for an annual physical with no cardiac complaints.
A routine 12-lead ECG obtained for the visit is run through an AI-ECG structural heart disease algorithm and returns a composite-positive result.
The patient has no murmur, normal oxygen saturation, and an unremarkable physical exam.
Guided by the AI-ECG flag rather than symptoms, the primary care physician refers for transthoracic echocardiography, which reveals moderate aortic stenosis with preserved ejection fraction.
Without the AI-ECG signal, this asymptomatic valvular disease likely would not have been captured for years.
Why Structural Heart Disease Slips Through
SHD is a leading contributor to heart failure and cardiovascular death, yet it is frequently silent in its early stages.
Echocardiography remains definitive, but its cost and the need for trained sonographers and interpreting physicians limit how broadly it can be deployed as a screening tool.
ECGs, by contrast, are inexpensive, ubiquitous, and already collected during routine care in primary care offices, emergency departments, and even on consumer wearables.
The clinical opportunity is to use the ECG's existing signal, amplified by AI, to decide who actually needs the more expensive test.
A related approach from Columbia University Irving Medical Center's EchoNext model found that nearly half of high-risk patients would not have received an echocardiogram under routine care alone.
The Composite AI-ECG Signal
The model discussed at New York Valves 2026 combines two previously validated algorithms.
One component, trained to detect left ventricular systolic dysfunction (ejection fraction below 40%), correlates most strongly with reduced pump function.
The second component, trained to detect left ventricular diastolic dysfunction (septal E/e' above 15), correlates more broadly with valvular disease, hypertrophy, and pulmonary hypertension.
Combining the two into a single composite-positive/negative output was designed to capture the fuller spectrum of SHD rather than any single subtype.
Both underlying algorithms already carry regulatory clearance and are in clinical use in South Korea.
What the Multinational Data Showed
Investigators pooled retrospective data from three cohorts: a South Korean hospital network, an academic center in New York, and the UK Biobank.
SHD prevalence differed sharply by cohort, ranging from roughly one in eight patients to more than four in ten, reflecting differences in referral patterns and population risk.
Despite that variation, the composite AI-ECG score performed consistently across all three groups.
| Cohort | Sample Size | SHD Prevalence | Composite Sensitivity |
|---|---|---|---|
| Incheon Sejong Hospital (South Korea) | 46,082 | 12.7% | 71.8% |
| Columbia University Irving Medical Center (New York) | 36,286 | 43.6% | 76.1% |
| UK Biobank | 41,226 | Not separately reported | Consistent with above |
Beyond detecting prevalent disease, the score also predicted incident SHD over six years of follow-up.
Composite-positive patients had a 3.75-fold greater risk of developing new SHD in the South Korean cohort and a 2.75-fold greater risk in the UK Biobank, compared with composite-negative patients.
| Outcome Comparison | Composite-Positive vs. Negative |
|---|---|
| Incident SHD risk, South Korean cohort (6-yr) | 3.75-fold higher |
| Incident SHD risk, UK Biobank (6-yr) | 2.75-fold higher |
| Prior heart-failure risk data (separate US cohorts) | ~20-fold higher; outperformed the AHA PREVENT-HF equation |
The Implementation Debate
A discussant at the meeting noted that the vast majority of ECGs recorded in daily practice are never systematically acted upon.
AI, in his framing, is not a diagnostic leap so much as an efficient way to extract signal that was already sitting in a waveform physicians routinely order anyway.
He cautioned that the models still leave room for missed cases, meaning a negative AI-ECG result should not override clinical suspicion.
He also raised an unresolved medicolegal question that will matter to any cardiologist adopting this technology: liability when a model's judgment differs from a physician's, in either direction.
Both the presenting investigator and the discussant agreed that prospective trials, not retrospective analyses, are needed before broad workflow integration, and randomized trials are reportedly being planned in the US and South Korea.
The Investor Angle: Who Makes These Algorithms
The composite algorithm in this study, marketed as AiTiALVSD and AiTiALVDD, is developed by Medical AI Co.PRIVATE / PRE-IPO
Medical AI is a South Korean company spun out of Sejong Hospital that has publicly stated it is preparing for a technology-specific listing on the KOSDAQ exchange after receiving top ratings from Korea Exchange evaluators.
As of this writing, the company has no public ticker, so investors cannot yet buy shares directly; readers should treat any pre-IPO valuation chatter with appropriate skepticism.
For physician-investors tracking the broader Korean medical-AI sector, two already-listed comparators illustrate how these business models have performed once public, though neither is an ECG-specific competitor.
| Company | Ticker | Focus Area | Notes |
|---|---|---|---|
| Medical AI Co. | Private | AI-ECG for LV systolic/diastolic dysfunction | Preparing for KOSDAQ listing; maker of the algorithm in this study |
| Lunit Inc. | KOSDAQ: 328130 | AI for oncology imaging and pathology | Publicly traded Korean medical-AI peer; not an ECG company |
| Vuno Inc. | KOSDAQ: 338220 | AI for imaging (chest X-ray, bone age, fundus) | Publicly traded Korean medical-AI peer; not an ECG company |
What This Could Mean for Echo Volumes and Reimbursement
Every AI-ECG-driven referral for confirmatory imaging is, from a health-system standpoint, an additional transthoracic echocardiogram (CPT 93306).
Medicare's national average reimbursement for a complete echocardiogram with Doppler runs roughly $210 to $260 globally, split between technical and professional components, while commercially negotiated and self-pay prices vary far more widely across hospitals.
If AI-ECG screening is adopted broadly, echo labs and cardiology groups performing in-house imaging could see meaningful volume growth, an operational detail worth factoring into staffing and capital equipment planning.
Conversely, payers may eventually push back on expanded echo utilization driven by an unvalidated screening trigger, which is one reason prospective outcomes data will matter for coverage policy as much as for clinical guidelines.
Bottom Line
A composite AI-ECG score detected prevalent structural heart disease with roughly 72% to 76% sensitivity across three international cohorts.
The same score predicted new-onset disease years in advance, with composite-positive patients carrying up to nearly four-fold higher risk.
The tool is already regulatory-cleared and deployed in South Korea, but remains retrospectively validated, and prospective randomized trials are still pending before this becomes standard US workflow.
The company behind the algorithm is privately held and preparing for a stock exchange listing, so there is currently no way to invest in it directly; broader Korean medical-AI names offer indirect sector exposure but are not therapeutic equivalents.
Financial disclaimer: This article is for general informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security; all company, valuation, and reimbursement figures are subject to change, and readers should consult a licensed financial advisor before making investment decisions.
References
- AI-ECG Models Detect Structural Heart Disease, Identify At-Risk Patients. TCTMD, presented at New York Valves 2026.
- AI for Heart Failure Care Is Evolving Rapidly, THT 2026 Makes Clear. TCTMD.
- Predicting Risk of Cardiovascular Disease Events (PREVENT) Calculator. American Heart Association, Professional Heart Daily.
- Variation in Cost of Echocardiography Within and Across US Hospitals. PMC.
- Medical AI Co. — Company Overview. Medical AI corporate site.
- Lunit Inc. (KOSDAQ: 328130) Stock Overview. StockAnalysis.com.
- Vuno Inc. (KOSDAQ: 338220) Stock Overview. StockAnalysis.com.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.