Artificial intelligence-enabled electrocardiography to triage echocardiography for structural heart disease diagnosis in a low-resource setting.

BACKGROUND

Timely diagnosis of structural heart disease (SHD) is often constrained by limited access to echocardiography in low-resource settings. Although electrocardiography (ECG) is widely available, traditional interpretation poorly reflects underlying SHD risk, leading to inefficient referral for echocardiography.

METHODS

We used data from PROVAR+, a large community-based cardiovascular screening program in operation since 2014 in Brazil. In this program, adults underwent 12-lead ECGs paired with either screening point-of-care ultrasound (POCUS) (screening cohort) or comprehensive transthoracic echocardiography (TTE) (imaging cohort). We evaluated the performance of a validated artificial intelligence model in identifying probable SHD from ECG images compared with Minnesota Code-defined major ECG abnormalities, with POCUS as a screening reference. We then assessed the health-system impact of AI-ECG-guided echocardiography referral using paired ECG-TTE data by simulating standard versus AI-prioritized referral strategies under an established health-system echocardiography capacity.

RESULTS

Among 3282 individuals in the screening cohort, AI-ECG demonstrated substantially higher discrimination for probable SHD than Minnesota Code-based major ECG abnormalities (area under the curve 0.73 vs 0.52), with higher positive predictive value and significant net reclassification improvement (35%), driven primarily by improved identification of individuals without disease. Decision-curve analysis showed consistently greater net benefit for AI-ECG across clinically relevant referral thresholds. In the imaging cohort (n = 1475), AI-ECG-guided prioritization accelerated SHD diagnosis under a previously known system capacity of 200 echocardiograms per month. The time to identify 50% of SHD cases was shorter with AI-guided care, with 19 days using AI-ECG-based triage vs 81 days under standard referral, which was consistent across demographic and clinical subgroups.

CONCLUSION

AI-enabled ECG interpretation improved the identification of SHD beyond conventional ECG assessment. Applying AI-ECG to echocardiography referral workflows can substantially accelerate diagnosis under limited imaging capacity, supporting more efficient and equitable use of cardiovascular diagnostic resources.