RApid Throughput Screening for Asymptomatic COVID-19 Infection With an Electrocardiogram: A Prospective Observational Study

Objective: To evaluate the ability of a neural network to identify severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection using point-of-care electrocardiography obtained with a portable device. Patient and Methods: We enrolled 2827 patients in a prospective observational study, from December 10, 2020, through June 4, 2021, to determine the accuracy of a point-of-care, handheld, smartphone-compatible, artificial intelligence–enabled electrocardiography (ECG) (POC AI-ECG) in detecting asymptomatic SARS-CoV-2 infection using a modified version of an existing deep learning model framework trained on 12-lead ECG data. Results: Study participants were 48% (n=1067) female, 79% (n=1749) White, and 7% (n=153) endorsed previous COVID-19 infection. We found the POC AI-ECG algorithm was ineffective for detecting asymptomatic SARS-CoV-2 infection (area under curve, 0.56; 95% CI, 0.46-0.66), failing to adequately discriminate between ECGs performed among participants who tested positive compared to those who tested negative. Conclusion: Contrary to the prior 12-lead ECG study, a POC AI-ECG failed to reliably identify asymptomatic SARS-CoV-2 infection among adults. This study underscores the importance of prospective testing, assuring similar populations, and using similar signals or data when developing AI-ECG tools. Trial registration: clinicaltrials.gov Identifier: NCT04725097