Improving Arrhythmia Diagnostics in Primary Care: The Potential Role of Artificial Intelligence in ECG Interpretation

Charlotte Eklund Gustafsson, Pyotr G Platonov, Josefin Kron, Hans Thulesius, Erik Ljungström, Olle Pahlm, Thomas Lindow

Keywords: Arrhythmia diagnostics, artificial intelligence, electrocardiogram (ECG), primary care, atrial fibrillation

Background:

Algorithm-based ECG interpretations are widely used but frequently misclassify arrhythmias such as atrial fibrillation/flutter (AF), often leading to inappropriate anticoagulant treatment. Recent advancements in artificial intelligence (AI), particularly deep learning and neural networks, show promise for improving ECG interpretation. Telehealth integration further enhances access to AI-assisted diagnostics, addressing geographic barriers in primary care. However, the application of AI in primary care for arrhythmia diagnostics remains underexplored.

Research questions:

Can AI-based tools improve the diagnostic accuracy of atrial fibrillation/flutter (AF) in primary care by reducing misclassifications from algorithm-based ECG systems and enhancing telehealth capabilities?

Method:

This study analyzed 980 ECGs performed in primary care and flagged with an algorithm-generated AF diagnosis (Glasgow ECG analysis program, version 28.5.1). The ECGs were reformatted, photographed, and reanalyzed by the AI tool PMcardio. Expert readers provided a reference standard, identifying 89 false positive AF diagnoses in the initial algorithm-generated dataset.

Results:

PMcardio correctly classified 94% of all ECGs, outperforming the algorithm-based system’s 91% accuracy.
It reclassified 84% of false positive AF diagnoses as non-AF. However, PMcardio misclassified 2% of true positive AF diagnoses as non-AF.

Conclusions:

The AI tool PMcardio shows potential for reducing false positive AF diagnoses and inappropriate anticoagulant treatments in primary care.
However, its 2% misclassification rate for true positive AF underscores the need for further studies to validate its diagnostic accuracy before clinical implementation.

Points for discussion:

1. How can AI tools be integrated into primary care workflows to enhance arrhythmia diagnostics?

2. What are the risks of relying on AI-based ECG interpretation, and how can these be mitigated?

3. What is the potential impact of AI-enhanced ECG interpretation on clinical outcomes and patient safety?

#198