Abstract 4361582: Artificial Intelligence-Enabled Electrocardiogram for Screening Asymptomatic Left Ventricular Systolic Dysfunction in Health Check-Up Populations

Abstract

Background: Asymptomatic left ventricular systolic dysfunction (LVSD) often precedes overt heart failure but frequently goes undetected in the general population. With the rise of artificial intelligence-enabled electrocardiogram (AI-ECG) now offer a scalable approach for early detection. Hypothesis: We hypothesize that AI-ECG is capable of identifying LVSD in apparently healthy individuals undergoing routine check-ups. Methods: This retrospective single-center study evaluated the performance of an AI-ECG model (AiTiALVSD) among 40,713 asymptomatic adults who underwent 60,711 paired ECG and echocardiography exams as part of self-referred health screenings from 2011 to 2023. LVSD was defined as left ventricular ejection fraction ≤40%. ( Figure 1 ) Model performance was compared against two conventional heart failure risk scores. Results: The AiTiALVSD model demonstrated excellent diagnostic accuracy with an area under the receiver operating characteristic curve of 0.973 and a precision-recall area of 0.328. ( Figure 2 ) At the prespecified threshold, the model achieved 90.6% sensitivity, 99.4% specificity, 7.7% positive predictive value, and 100% negative predictive value. In comparison, conventional models performed worse (area under the curve of 0.696 for MESA and 0.672 for PCP-HF). ( Figure 3 ) Simulation suggested that 1,841 ECGs and 13 echocardiograms would be required to identify one LVSD case. Notably, most false positives exhibited other echocardiographic abnormalities. Conclusions: The AI-ECG model showed high diagnostic performance in a real-world low-prevalence setting, supporting its potential as a cost-effective, noninvasive screening tool for early detection of LVSD. These findings warrant prospective validation and suggest a promising role for AI in community-based heart failure prevention strategies.