LAUSR

Abstract Funding Acknowledgements Type of funding sources: None. Background There were no artificial intelligence (AI) tools to predict the severity and prognosis of patients hospitalized with COVID-19 using initial electrocardiography (ECG). Purpose We assessed whether AI using 12-lead ECG could assist in the early determination of COVID-19 severity and predict prognosis of hospitalized patients. Methods Patients diagnosed with COVID-19 via polymerase chain reaction and admitted to our institution during February 2020–October 2021 were enrolled. We identified 642 patients (mean age, 56±19 years; 41.7% male) who underwent 12-lead ECGs within 2 days of admission for COVID-19. Patients were divided according to severity: mild-to-moderate illness (oxygen or low-flow oxygen therapy not required) and severe-to-critical illness (required high-flow oxygen, invasive mechanical ventilation, or extra corporeal membrane oxygenation). Results Severe-to-critical illness occurred in 243 (37.8%) patients, including 23 (3.6%) in-hospital mortalities. Experimental results using our developed AI enhanced ECG demonstrated better outcomes for mild-to-moderate illness based on an area under the curve (AUC), sensitivity, specificity, and F1 score of 0.71, 0.95, 0.70, and 0.81, respectively. The AUC for detecting severe-to-critical illness was 0.71 (95% confidence interval [CI]: 0.58– 0.82; sensitivity: 0.70; specificity: 0.35; F1 score: 0.35; and overall accuracy: 0.70). When analyzing individuals without COVID-19, the AI algorithm for identifying COVID-19 showed an AUC of 0.85 (95% CI: 0.82–0.88; sensitivity: 0.60; specificity: 0.84; F1 score: 0.70; overall accuracy: 0.85). Conclusion AI-enhanced ECG could reasonably predict severity in patients hospitalized for COVID-19, suggesting that AI-ECG assists in the early determination of COVID-19 severity, helping physicians improve patient management.