Artificial intelligence-powered electrocardiogram is superior to NT-proBNP in predicting left ventricular systolic dysfunction in atrial fibrillation with rapid ventricular response

Q: What are the key findings of this study?

Among patients with rapid atrial fibrillation, AI-ECG was superior to NT-proBNP in predicting a left ventricular ejection fraction of <35% (AUC 0.72 vs 0.59, accuracy 77% vs 47%).

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Atrial fibrillation with rapid ventricular response (n=11801). Intervention: Artificial intelligence-powered electrocardiogram (AI-ECG) vs. N-terminal pro-B-type natriuretic peptide (NT-proBNP). Primary outcome: Prediction of left ventricular ejection fraction <35% (AUC 0.72 vs 0.59).

Q: What question did this study set out to answer?

To assess whether an artificial intelligence-powered electrocardiogram model can accurately predict left ventricular systolic dysfunction in patients presenting with atrial fibrillation with rapid ventricular response compared to NT-proBNP.

Omar Baqal; Claire I. Yee; Jaxon Quillen; Xuan Ci Mee; Ghee Kheng Lim; Chadi Ayoub; Reza Arsanjani; Dan Sorajja; Peter Noseworthy; Hicham Z. El Masry

doi:10.1016/j.hroo.2026.01.028

April 1, 2026Heart Rhythm O2Open Access

Artificial intelligence-powered electrocardiogram is superior to NT-proBNP in predicting left ventricular systolic dysfunction in atrial fibrillation with rapid ventricular response

View Full Paper

Ask AI

Bookmark

Share

Key result

AI-ECG outperforms NT-proBNP in identifying LVEF <35% in rapid AF, reaching ~77% accuracy.

AUC 0.72 vs 0.59
n=11,801

Figures

Graphical abstract published by Heart Rhythm O2 — Graphical abstract · Published by Heart Rhythm O2 · CC BY-NC-ND

Why the study?

Does an artificial intelligence-powered electrocardiogram improve the prediction of reduced left ventricular ejection fraction compared to NT-proBNP in patients with atrial fibrillation with rapid ventricular response?

Key Points

To assess whether an artificial intelligence-powered electrocardiogram model can accurately predict left ventricular systolic dysfunction in patients presenting with atrial fibrillation with rapid ventricular response compared to NT-proBNP.
Included patients with a 12-lead ECG processed by an AI-ECG tool and an echocardiogram performed within 7 days, excluding patients with known heart failure with reduced ejection fraction (<50%).
Compared the predictive performance of AI-ECG against NT-proBNP blood concentrations (available for 5,279 patients) across multiple diagnostic cutoff thresholds.
AI-ECG was superior to NT-proBNP in predicting an LVEF <35% (area under the curve, 0.72 vs 0.59), achieving an accuracy of 77% vs 47% at their respective Youden cut points (17% AI-ECG probability vs 1767 pg/mL NT-proBNP).
An AI-ECG low ejection fraction probability cutoff of 75% achieved a specificity of 96% and a negative predictive value of 98% in ruling out an LVEF <35%.
An AI-ECG probability cutoff of 17% provided the best balance of sensitivity (66%) and specificity (78%) for detecting an LVEF <35%, with an overall accuracy of 78%.

Study Design

Type

Observational (n=11,801)

Multicenter

Yes

Structured PICO

Does an artificial intelligence-powered electrocardiogram improve the prediction of reduced left ventricular ejection fraction compared to NT-proBNP in patients with atrial fibrillation with rapid ventricular response?

P

Population

11,801 adult patients (mean age 73.2 years, 48.6% female) with atrial fibrillation and rapid ventricular response, without known heart failure with reduced ejection fraction, evaluated retrospectively.

E

Exposure

Artificial intelligence-powered electrocardiogram (AI-ECG) analysis to predict reduced left ventricular ejection fraction

C

Comparator

N-terminal pro-B-type natriuretic peptide (NT-proBNP) levels (available in 5,279 patients, 48% of cohort)

O

Outcome

Prediction of reduced left ventricular ejection fraction (<35%) confirmed by echocardiogram within 7 dayssurrogate

Main Result

Effect estimate: AUC 0.72 vs 0.59

Absolute Event Rate: 0.72% vs 0.59%

AI-ECG provides a rapid, noninvasive, and superior alternative to NT-proBNP for ruling out reduced LVEF in patients presenting with atrial fibrillation with rapid ventricular response.

Limitations

Retrospective nature of data
Possibility of misclassification owing to the use of diagnostic codes
Lower incidence of reduced LVEF in the cohort may affect positive predictive value
Exact timing of echocardiography could influence LVEF classification and AI-ECG performance analysis
Predominantly white cohort limiting racial generalizability
The AI-ECG model used was not developed exclusively on a cohort with AF
Lower incidence of reduced LVEF affecting positive predictive value
Exact timing of echocardiography could influence LVEF classification
AI-ECG model was not developed exclusively on a cohort with AF

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

AI-ECG may better identify reduced LVEF than NT-proBNP in rapid AF; leaves open prospective validation before clinical adoption.

Abstract

Despite its prevalence, risk stratification for tachycardia-induced cardiomyopathy remains poorly understood. A deep learning model developed at Mayo Clinic uses standard 12-lead electrocardiograms (ECGs) to predict reduced left ventricular ejection fraction (LVEF) (≤35%) in the general population with high accuracy (area under the curve, 0.93).1 The performance of artificial intelligence (AI)–powered ECG (AI-ECG) in predicting systolic dysfunction among patients with atrial fibrillation (AF) with rapid ventricular response (AF RVR) has not been previously explored.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Baqal et al. (2026) conducted an observational in Atrial fibrillation with rapid ventricular response (n=11,801). Artificial intelligence-powered electrocardiogram (AI-ECG) vs. N-terminal pro-B-type natriuretic peptide (NT-proBNP) was evaluated on Prediction of left ventricular ejection fraction <35% (AUC 0.72 vs 0.59). Among patients with rapid atrial fibrillation, AI-ECG was superior to NT-proBNP in predicting a left ventricular ejection fraction of <35% (AUC 0.72 vs 0.59, accuracy 77% vs 47%).

synapsesocial.com/papers/69df2a4be4eeef8a2a6af792 https://doi.org/10.1016/j.hroo.2026.01.028

View Full Paper

Ask AI

Bookmark

Share

Artificial intelligence-powered electrocardiogram is superior to NT-proBNP in predicting left ventricular systolic dysfunction in atrial fibrillation with rapid ventricular response

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Artificial intelligence-powered electrocardiogram is superior to NT-proBNP in predicting left ventricular systolic dysfunction in atrial fibrillation with rapid ventricular response

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading