Sequential AI-ECG Diagnostic Protocol for Opportunistic Atrial Fibrillation Screening: A Retrospective Single-Center...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Atrial fibrillation (AF) episodes are often asymptomatic, delaying crucial anticoagulation treatment.
• Current screening methods may miss intermittent AF, impacting patient outcomes.
• Resource limitations necessitate efficient and accurate screening tools.

Purpose of the Study:

• To evaluate a two-stage artificial intelligence (AI) electrocardiogram (ECG) screening protocol for detecting atrial fibrillation (AF).
• To enhance diagnostic accuracy and optimize resource allocation in AF screening.
• To assess the performance of a serial ECG AI model following initial screening.

Main Methods:

• Analysis of 248,612 12-lead ECGs from 164,793 adults for AI model development.
• Implementation of a two-stage protocol: initial single-ECG AI model followed by a serial-ECG AI model at three months if needed.
• Performance evaluation using metrics like AUROC, sensitivity, specificity, accuracy, and F1 score in 11,349 eligible patients.

Main Results:

• The protocol achieved a high area under the receiver operating characteristic curve (AUROC) of 0.908.
• Sensitivity was 88.1%, specificity 78.7%, and negative predictive value (NPV) 98.4%.
• The protocol correctly identified 84.9% of AF-positive patients with a history of stroke.

Conclusions:

• A sequential AI ECG strategy effectively maintains a high NPV and improves positive predictive value (PPV) through longitudinal confirmation.
• This AI-driven approach can prioritize ambulatory monitoring for patients most likely to benefit from AF detection.
• Further prospective, multi-center validation and cost-effectiveness studies are warranted.