Interethnic Validation of Artificial Intelligence for prediction of Atrial Fibrillation Using Sinus Rhythm Electrocardiogram
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Previous research has demonstrated acceptable diagnostic accuracy of AI-enabled sinus rhythm (SR) electrocardiogram (ECG) interpretation for predicting paroxysmal or incident atrial fibrillation (AF). However, interethnic validations of these AI algorithms remain limited. We aimed to develop and comprehensively evaluate our AI model for predicting AF based on standard 12Dlead SR ECG images in a Korean population, and to validate its performance in Brazilian patient cohorts.
Methods
We developed a modified convolutional neural network model using a dataset comprising 811,542 ECGs from 121,600 patients at Seoul National University Bundang Hospital (2003–2020). Ninety percent of the patients were allocated to the training dataset, while the remaining 10% to the internal validation dataset. The model outputs a risk score (from 0 to 1) indicating the probability of concurrent paroxysmal or incident AF within 2 years, using standard-format 12Dlead SR ECG images. External validation was performed using the CODE 15% dataset, an open ECG dataset from the Telehealth Network of Minas Gerais, Brazil, by applying a 1:4 (AF:Non-AF) random sampling strategy.
Results
In the internal validation, our AI model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.907 (95% CI: 0.897-0.916), with a sensitivity of 80.6% and a specificity of 85.0% for AF prediction. Subgroup analyses showed an AUROC of 0.874 (95% CI: 0.856-0.891) for patients in routine health checkups or outpatient settings, and 0.852 (95% CI: 0.824-0.880) for patients with "Normal ECG" interpretations. In the external interethnic validation with the CODE 15% dataset, the AI model exhibited an AUROC of 0.884 (95% CI: 0.869-0.900), which increased to 0.906 (95% CI: 0.893-0.919) when adjusted for age and sex. In the subset of patients with "Normal ECG" interpretations, the AUROC was 0.826 (95% CI: 0.769-0.883), increasing to 0.861 (95% CI: 0.814-0.908) after applying the same adjustments.
Conclusions
Our AI-powered SR ECG interpretation model demonstrated excellent performance in predicting paroxysmal or incident AF, with valid performance in the Brazilian population as well. This suggests that the model has potential for broad application across different ethnic groups.
