Community Screening for Undiagnosed Left Ventricular Systolic Dysfunction Using Single-Lead ECG-AI: The RURAL Cohort Study

Background&nbsp;Left ventricular systolic dysfunction is treatable when found early, but detection requires echocardiography, which is scarce across the rural US South. Artificial-intelligence electrocardiogram analysis (ECG-AI) can identify reduced ejection fraction from a single lead, but has been validated only in clinically indicated recordings from referral populations; its performance for community screening is unknown.<br><br>Methods&nbsp;We applied a validated single-lead (lead I) ECG-AI model, without retraining, to 1276 RURAL Cohort Study participants from rural Alabama and Mississippi, USA. All underwent protocolized echocardiography irrespective of clinical indication, with centralized core-laboratory interpretation. Reduced ejection fraction (rEF) was LVEF &lt;40%. We assessed discrimination, average precision, screening yield, unrecognized disease, and reproducibility across three serial ECGs. Reported per STARD.<br><br>Findings&nbsp;Of 1276 participants (mean age 48·5 years [SD 11·0]; 836 [65·5%] women; 814 [63·8%] Black), 15 (1·2%) had rEF. Most had no previously reported heart failure diagnosis: about one new case per 100 adults screened. ECG-AI detected rEF with an AUC of 0·91 (95% CI 0·80–1·00; average precision 40%, sensitivity 86·7%, and specificity 95·4%, without sex or race differences. Predictions were reproducible across three recordings (r&gt;0·93). Of 58 screen-positives without rEF, 12 (21%) had LVEF 40–49%.<br><br>Interpretation In an unselected rural cohort with limited imaging access, most systolic dysfunction was previously unrecognized, and a single-lead ECG identified it with high discrimination and reproducibility. Given the low prevalence, ECG-AI is best positioned as triage that concentrates scarce echocardiography where it will change management, not as a diagnostic test.

Duke Scholars

Author Gerald Bloomfield Medicine, Cardiology