Mount Sinai Researchers Develop Machine Learning Model that Can Detect and Predict COVID-19 from Collected Data on Wearable Devices

Paper Title: Evaluation of a machine learning approach utilizing wearable data for prediction of SARS-CoV-2 infection in healthcare workers

Journal: Journal of the American Informatics Association (JAMIA) Open, July 2022 issue

Authors: Robert P. Hirten, MD, Associate Professor of Medicine (Gastroenterology) at the Icahn School of Medicine at Mount Sinai; Zahi Fayad, PhD, Director of the BioMedical Engineering and Imaging Institute and Professor of Diagnostic, Molecular & Interventional Radiology and Medicine (Cardiology) at the Icahn School of Medicine at Mount Sinai; and other coauthors.

Bottom Line: There continues to be high rates of COVID-19 in the population, significantly disrupting many industries and resulting in widespread illness. Mount Sinai researchers have developed a machine learning algorithm that can determine if an individual has SARS-CoV-2 infections, the virus that causes COVID-19—with a high sensitivity and specificity—from the data collected by wearable devices. This machine learning model works in both symptomatic and asymptomatic people.

How: The researchers enrolled more than 400 participants, with 49 (about 12%) having a positive nasal COVID-19 PCR test during the follow up. Participants downloaded a custom smart phone application, wore an Apple Watch and completed a daily questionnaire about how they felt and whether they had been diagnosed with COVID-19. The research team then examined several machine-learning approaches to determine which performed best to predict positive COVID nasal PCR results.

Results: They found that the machine learning algorithm called gradient-boosting machines (GBM) had the most favorable validation performance out of all models. The model was calibrated to value sensitivity over specificity, achieving an average sensitivity of 82% and specificity of 77%. They found markers of heart rate variability—or the calculation of the small-time differences between each heartbeat, to be most important in identifying infections.

Why the Research Is Interesting: The researchers say this is one of the only studies to be able to determine if someone has COVID-19 from metrics collected by commonly used wearable devices such as smartwatches. The algorithm can be used by groups, industries, or healthcare organizations to monitor people for infections, using an easy to implement algorithm that can be run on data collected from wearable devices. People who may be infected can then be referred for PCR testing to confirm the diagnosis.

This is one of the only algorithms to predict the presence of a disease or condition from wearable device outputs. Similar approaches can be applied to other diseases to monitor and predict their development or worsening. This approach may have large implications for the management of many chronic diseases.

Said Mount Sinai’s Dr. Robert Hirten of the research:
We have developed an easily implemented algorithm that can be applied to the information collected by wearable devices and can reliably predict if someone has COVID-19. This provides an opportunity for improved monitoring and control of infections, as well as the possibility to apply similar techniques to the management of other diseases.

Said Mount Sinai’s Dr. Zahi Fayad of the research: 
The algorithm helps to identify and direct resources towards high-risk individuals, which can in turn prevent further spread of infection and keep communities safe. This research has important implications for the potential of wearable devices to support public health in the face of an infectious disease. 


To request the full paper or schedule an interview with the researchers, contact the Mount Sinai Press Office at [email protected] or 347-346-3390.