Objective: To study the utility of artificial intelligence (AI)-enabled electrocardiograms (ECGs) in patients with Graves disease (GD) in identifying patients at high risk of atrial fibrillation (AF) and heart failure with reduced ejection fraction (HFrEF), and to study whether AI-ECG can reflect hormonal changes and the resulting menstrual changes in GD.
Patients and methods: Patients diagnosed with GD between January 1, 2009, and December 31, 2019, were included. We considered AF diagnosed at 30 days or fewer before or any time after GD and de novo HFrEF not explained by ischemia, valve disorder, or other cardiomyopathy at/after GD diagnosis. Electrocardiograms at/after index condition were excluded. A subset analysis included females younger than 45 years of age to study the association between ECG-derived female probability and menstrual changes (shorter, lighter, or newly irregular cycles).
Results: Among 430 patients (mean age, 50±17 years; 337 (78.4%) female), independent risk factors for AF included ECG probability of AF (hazard ratio [HR], 1.5; 95% CI, 1.2 to 1.6 per 10%; P<.001), older age (HR, 1.05; 95% CI, 1.03 to 1.07 per year; P<.001), and overt hyperthyroidism (HR, 3.9; 95% CI, 1.2 to 12.7; P=.03). The C-statistic was 0.85 for the combined model. Among 495 patients (mean age, 52±17 years; 374 (75.6%) female), independent risk factors for HFrEF were ECG probability of low ejection fraction (HR, 1.4; 95% CI, 1.1 to 1.6 per 10%; P=.001) and presence of AF (HR, 8.3; 95% CI, 2.2 to 30.9; P=.002), and a C-statistic of 0.89 for the combined model. Lastly, of 72 females younger than 45 years, 30 had menstrual changes at time of GD and had a significantly lower AI ECG-derived female probability [median 77.3; (IQR 57.9 to 94.4)% vs. median 97.7 (IQR 92.4 to 99.5)%, P<.001].
Conclusion: AI-enabled ECG identifies patients at risk for GD-related AF and HFrEF and was associated with menstrual changes in women with GD.
Copyright © 2022. Published by Elsevier Inc.