Artificial intelligence (AI) may provide a new way to precisely determine if an individual is not infected with COVID-19. An international retrospective study found that infection with SARS-CoV-2, the virus responsible for causing COVID-19, makes slight electrical changes in the heart.
An AI-enhanced EKG has the ability to detect these changes and could be utilized as a quick, reliable COVID-19 screening test to rule out COVID-19 infection. It was also able to detect COVID-19 infection in the test with a positive predictive value ― people infected ― of 37% and a negative predictive value ― people not infected ― of 91%.
When more normal control subjects were added to reflect a 5% prevalence of COVID-19 — analogous to a real-world population — the negative predictive value increased to 99.2%. The study results were published in the Mayo Clinic Proceedings journal.
COVID-19 has an incubation period of 10 to 14 days, which is long compared to other kinds of common viruses. A majority of people do not show symptoms of infection, and they could put others at risk without being aware of it. Moreover, the clinical resources and turnaround time that were required for present-day testing techniques are significant, and access could be an issue.
If validated prospectively using smartphone electrodes, this will make it even simpler to diagnose COVID infection, highlighting what might be done with international collaborations.
Paul Friedman, MD, Study Senior Author and Chair, Department of Cardiovascular Medicine, Mayo Clinic
The recognition of a global health crisis brought together stakeholders from across the world to design a tool that could tackle the need to quickly, cost-effectively and non-invasively rule out the existence of acute COVID-19 infection. The study, which incorporated data from racially diverse populations, was carried out via a global volunteer consortium covering 14 countries and four continents.
The lessons from this global working group showed what is feasible, and the need pushed members in industry and academia to partner in solving the complex questions of how to gather and transfer data from multiple centers with their own EKG systems, electronic health records and variable access to their own data. The relationships and data processing frameworks refined through this collaboration can support the development and validation of new algorithms in the future.
Suraj Kapa, M.D, Cardiac Electrophysiologist, Mayo Clinic
The team chose patients with EKG data from around the time their COVID-19 diagnosis was verified by a genetic test for the SARS-Co-V-2 virus. This data was control-matched with analogous EKG data from patients who were not infected with COVID-19.
Over 26,000 of the EKGs were used by the researchers to train the AI and almost 4,000 others to confirm its readings. Eventually, the AI was tested on 7,870 EKGs that were not used before. In each of these sets, the presence of COVID-19 was approximately 33%.
To precisely reflect a real-world population, over 50,000 normal EKGs were further added to achieve a 5% prevalence rate of COVID-19. This increased the negative predictive value of the AI from 91% to 99.2%.
Zachi Attia, PhD, a Mayo Clinic engineer in the Department of Cardiovascular Medicine, described that prevalence acts as a variable in the calculation of negative and positive predictive values. Particularly, as the prevalence reduces, the negative predictive value increases. Dr. Attia is the co-first author of the study with Dr. Kapa.
Accuracy is one of the biggest hurdles in determining the value of any test for COVID-19. Not only do we need to know the sensitivity and specificity of the test, but also the prevalence of the disease. Adding the extra control EKG data was critical to demonstrating how a variable prevalence of the disease ― as we have encountered with regions having widely different.
Zachi Attia, PhD, Engineer in Department of Cardiovascular Medicine, Mayo Clinic
“This study demonstrates the presence of a biological signal in the EKG consistent with COVID-19 infection, but it included many ill patients. While it is a hopeful signal, we must prospectively test this in asymptomatic people using smartphone-based electrodes to confirm that it can be practically used in the fight against the pandemic. Studies are underway now to address that question.” concluded Dr. Friedman
Attia, Z., et al. (2021) Rapid exclusion of COVID infection with the artificial intelligence ECG. Mayo Clinic Proceedings. doi.org/10.1016/j.mayocp.2021.05.027.