In a prospective screening study of 3,779 adults in two Brazilian cities, echocardiography found structural heart disease (SHD) in 24.2%—about 1 in 4—while an AI-enabled 12-lead ECG identified higher-risk patients with 87.3% sensitivity and a 92.6% negative predictive value. The study suggests conventional ECG findings alone are no longer enough for echo triage in low-access settings, with AI-guided prioritization cutting the estimated time to diagnose half of SHD cases from 407 to 172 days.

Why It Matters To Your Practice

• SHD was common in this screened population, yet echo capacity is often the bottleneck in public and underserved systems.

• The AI-ECG model outperformed a conventional model built from clinical variables and major ECG abnormalities (AUC 0.80 vs. 0.75).

• A high negative predictive value suggests potential usefulness for ruling out many patients when deciding who needs earlier transthoracic echocardiography.

Clinical Implications

• Among 3,779 participants, 58.7% screened AI-ECG-positive and 24.2% had echo-detected SHD, yielding a number needed to screen of 4.7.

• At the prespecified threshold, performance was 59.4% accuracy, 87.3% sensitivity, and 50.4% specificity.

• In practice, this is better suited to prioritization than stand-alone diagnosis: more patients will be flagged for echo, but fewer cases should be missed.

• Patients with major ECG abnormalities, older age, male sex, Chagas disease, coronary disease, and heart failure had higher odds of SHD.

Insights

• Each 10% increase in AI-ECG-SHD probability was associated with higher odds of SHD (OR 1.55, 95% CI 1.51-1.61).

• SHD in this study included moderate/severe ventricular dysfunction or hypertrophy, wall-motion abnormalities, moderate/severe valvular disease, congenital heart disease, and pericardial disease.

• The workflow used digital tele-ECG analyzed through an API, illustrating how AI screening could plug into existing remote ECG infrastructure.

The Bottom Line

• For clinicians tracking how AI will affect practice, this is a practical triage use case: AI on routine ECG may help decide who should get scarce echo slots first.

• The tradeoff is clear: strong sensitivity and rule-out performance, but modest specificity, so local workflow design and downstream echo capacity still matter.

• Best near-term role: augment clinician judgment and conventional ECG review, especially in resource-limited screening programs.

Reference: Identifying Structural Heart Disease and Prioritizing Echocardiography Through Artificial Intelligence-Enabled Tele-ECG: An External Validation of the PROVAR+ Study.