A smartphone app that combines passively collected physiologic data from wearable devices, such as fitness trackers, and self-reported symptoms can discriminate between COVID-19–positive and –negative individuals among those who report symptoms, new data suggest.
After analyzing data from more than 30,000 participants, researchers from the Digital Engagement and Tracking for Early Control and Treatment (DETECT) study concluded that adding individual changes in sensor data improves models based on symptoms alone for differentiating symptomatic persons who are COVID-19 positive and symptomatic persons who are COVID-19 negative.
The combination can potentially identify infection clusters before wider community spread occurs, Giorgio Quer, PhD, and colleagues report in an articleOct. 29 in Nature Medicine. DETECT investigators note that marrying participant-reported symptoms with personal sensor data, such as deviation from normal sleep duration and resting heart rate, resulted in an area under the curve (AUC) of 0.80 (interquartile range [IQR], 0.73-0.86) for differentiating between symptomatic individuals who were positive and those who were negative for COVID-19.
“By better characterizing each individual’s unique baseline, you can then identify changes that may indicate that someone has a viral illness,” said Dr. Quer, director of artificial intelligence at Scripps Research Translational Institute in La Jolla, Calif. “In previous research, we found that the proportion of individuals with elevated resting heart rate and sleep duration compared with their normal could significantly improve real-time detection of influenza-like illness rates at the state level,” he said in an interview.
Thus, continuous passively captured data may be a useful adjunct to bricks-and-mortar site testing, which is generally a one-off or infrequent sampling assay and is not always easily accessible, he added. Furthermore, traditional screening with temperature and symptom reporting is inadequate. An elevation in temperature is not as common as frequently believed for people who test positive for COVID-19, Dr. Quer continued. “Early identification via sensor variables of those who are presymptomatic or even asymptomatic would be especially valuable, as people may potentially be infectious during this period, and early detection is the ultimate goal,” Dr. Quer said.
According to his group, adding these physiologic changes from baseline values significantly outperformed detection (P < .01) using a British model described inby by Cristina Menni, PhD, and associates. That method, in which symptoms were considered alone, yielded an AUC of 0.71 (IQR, 0.63-0.79).
According to Dr. Quer, one in five Americans currently wear an electronic device. “If we could enroll even a small percentage of these individuals, we’d be able to potentially identify clusters before they have the opportunity to spread,” he said.
DETECT study details
During the period March 15 to June 7, 2020, the study enrolled 30,529 participants from all 50 states. They ranged in age from younger than 35 years (23.1%) to older than 65 years (12.8%); the majority (63.5%) were aged 35-65 years, and 62% were women. Sensor devices in use by the cohort included Fitbit activity trackers (78.4%) and Apple HealthKit (31.2%).
Participants downloaded an app called MyDataHelps, which collects smartwatch and activity tracker information, including self-reported symptoms and diagnostic testing results. The app also monitors changes from baseline in resting heart rate, sleep duration, and physical activity, as measured by steps.
Overall, 3,811 participants reported having at least one symptom of some kind (e.g., fatigue, cough, dyspnea, loss of taste or smell). Of these, 54 reported testing positive for COVID-19, and 279 reported testing negative.
Sleep and activity were significantly different for the positive and negative groups, with an AUC of 0.68 (IQR, 0.57-0.79) for the sleep metric and 0.69 (IQR, 0.61-0.77) for the activity metric, suggesting that these parameters were more affected in COVID-19–positive participants.
When the investigators combined resting heart rate, sleep, and activity into a single metric, predictive performance improved to an AUC of 0.72 (IQR, 0.64-0.80).
The next step, Dr. Quer said, is to include an alert to notify users of possible infection.
Alerting users to possible COVID-19 infection
In a similar study, an alert feature was already incorporated. The study, led by Michael P. Snyder, PhD, director of the Center for Genomics and Personalized Medicine at Stanford (Calif.) University, will soon be published online in Nature Biomedical Engineering. In that study, presymptomatic detection of COVID-19 was achieved in more than 80% of participants using resting heart rate.
“The median is 4 days prior to symptom formation,” Dr. Snyder said in an interview. “We have an alarm system to notify people when their heart rate is elevated. So a positive signal from a smartwatch can be used to follow up by polymerase chain reaction [testing].”
Dr. Snyder said these approaches offer a roadmap to containing widespread infections. “Public health authorities need to be open to these technologies and begin incorporating them into their tracking,” he said. “Right now, people do temperature checks, which are of limited value. Resting heart rate is much better information.”
Although the DETECT researchers have not yet received feedback on their results, they believe public health authorities could recommend the use of such apps. “These are devices that people routinely wear for tracking their fitness and sleep, so it would be relatively easy to use the data for viral illness tracking,” said co–lead author Jennifer Radin, PhD, an epidemiologist at Scripps. “Testing resources are still limited and don’t allow for routine serial testing of individuals who may be asymptomatic or presymptomatic. Wearables can offer a different way to routinely monitor and screen people for changes in their data that may indicate COVID-19.”
The marshaling of data through consumer digital platforms to fight the coronavirus is gaining ground. New York State and New Jersey aresmartphone apps to alert individuals to possible exposure to the virus.
More than 710,000 New Yorkers have downloaded the COVID NY Alert app, launched in October to help protect individuals and communities from COVID-19 by sending alerts without compromising privacy or personal information. “Upon receiving a notification about a potential exposure, users are then able to self-quarantine, get tested, and reduce the potential exposure risk to family, friends, coworkers, and others,” Jonah Bruno, a spokesperson for the New York State Department of Health, said in an interview.
And recently the Mayo Clinic and Safe Health Systemsto store COVID-19 testing and vaccination data.
Both the Scripps and Stanford platforms are part of a global technologic response to the COVID-19 pandemic. Prospective studies, led by device manufacturers and academic institutions, allow individuals to voluntarily share sensor and clinical data to address the crisis. Similar approaches have been used to track COVID-19 in large populations in Germany via theapp.
The study by Dr. Quer and colleagues was funded by a grant from the National Center for Advancing Translational Sciences at the National Institutes of Health. One coauthor reported grants from Janssen and personal fees from Otsuka and Livongo outside of the submitted work. The other authors have disclosed no relevant financial relationships. Dr. Snyder has ties to Personalis, Qbio, January, SensOmics, Protos, Mirvie, and Oralome.
A version of this article originally appeared on.