AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
Artificial Intelligence-Enhanced ECG as an Echocardiography-Triage Tool for Early Detection of Left Ventricular Systolic Dysfunction: A Systematic Review and Meta-Analysis
By MEDA SM, Mishra SKP, Naaz S et al.
Background: Left ventricular systolic dysfunction (LVSD) often remains asymptomatic during its initial phases, delaying clinical detection and is frequently underdiagnosed until it progresses to symptomatic heart failure, despite current disease-modifying therapies. Although echocardiography effectively identifies early LVSD, it is impractical for mass screening. Artificial intelligence-enhanced electrocardiography (AI-ECG) may provide a scalable strategy for identifying patients who require confirmatory imaging. Hypothesis: AI-ECG provides clinically useful diagnostic accuracy for LVSD detection and may function as an echocardiography triage tool for the early detection of reduced left ventricular ejection fraction. Methods: We conducted a systematic review and diagnostic meta-analysis of studies that evaluated AI-ECG algorithms against echocardiography for the detection of LVSD. Studies of adults undergoing paired ECG and echocardiography with extractable diagnostic accuracy data were included. The primary outcome was the diagnostic accuracy for LVEF <40%. Pooled sensitivity, specificity, area under the receiver operating characteristic curve (AUC), likelihood ratios, diagnostic odds ratio, and predictive values were assessed using reported pooled estimates and observed disease prevalence. Results: Nine studies comprising 145,992 patients with paired AI-ECG and echocardiography were included. Echocardiographic LVEF <40% was present in 10,549 (7.2 %) patients. The AI-ECG achieved pooled sensitivity and specificity of 0.84 and 0.89, respectively, with an AUC of 0.917 . The positive likelihood ratio was 7.6, the negative likelihood ratio was 0.18, and the diagnostic odds ratio was 42.5. At the observed LVSD prevalence of 7.2%, the estimated positive and negative predictive values were 37.3% and 98.6% , respectively, supporting the use of AI-ECG as a rule-out tool for echocardiography triage. The findings were consistent with earlier mixed-threshold evidence indicating high discrimination for LVSD, although heterogeneity was observed across model architectures, validation settings, population risk profiles, and LVEF thresholds. Conclusions: AI-ECG has demonstrated high diagnostic discrimination and promising rule-out performance for LVEF <40% , supporting its role as a scalable screening and echocardiography triage tool for the early detection of LVSD, particularly in resource-limited settings, rather than a replacement for confirmatory imaging. Prospective implementation studies are required to assess the clinical, economic, and health equity implications of integrating AI-ECG into LVSD screening pathways.