HYPERTHYROIDISM is known to be associated with both atrial fibrillation (AF) and heart failure, two conditions that can cause substantial morbidity and mortality. For this reason, academics at the Mayo Clinic campus in Rochester, Minnesota, USA, sought to determine whether people with Graves’ disease (characterised by the overproduction of thyroid hormones), who are at very high-risk for these cardiac disorders, could be identified.
Jwan A. Naser, first author on the study, highlighted the relevance of the research: “It is clinically pertinent to identify the subset of patients with Graves’ disease at increased risk of developing AF and heart failure who may benefit from closer surveillance and prompt restoration of euthyroidism.”
In a large population of individuals with Graves’ disease undergoing clinically indicated ECGs, artificial intelligence (AI)-enabled ECGs using a convolutional neural network were able to identify the signature of silent AF in ECGs obtained while patients were in sinus rhythm. In addition, the signature of low ejection fraction was also determined for each ECG performed, with areas under the curve for silent AF and low ejection fraction of 0.87 and 0.93, respectively. According to Naser, this enhanced ECG is “now available in the electronic medical record at Mayo Clinic and gives a probability number between 0 and 1 for having silent AF and low ejection fraction.”
Marius N. Stan, a Consultant at Mayo Clinic in Rochester, provided an overview of the methodology and principal findings: “We applied this AI ECG tool to more than 400 patients diagnosed with Graves’ disease over the past decade at Mayo Clinic who had ECGs around the time of diagnosis. We found that the AI ECG-derived probabilities of silent AF and low ejection fraction was an independent risk factor for the development of incident AF and heart failure with reduced ejection fraction (HFrEF), respectively.”
Sorin V. Pislaru, a Cardiologist and Echocardiographer at Mayo Clinic in Rochester, noted that the predictive ability of the AI-enabled ECG could be significantly improved by utilising multivariate models that not only included the AI ECG tool itself but also conventional risk factors for AF and HFrEF in hyperthyroidism such as older age, overt hyperthyroidism, and comorbidities. “We are currently planning to build a risk score tool to identify patients at highest risk of AF and HFrEF,” said Pislaru.
