Pairing wearables data with self-reported symptoms could improve COVID-19 prediction

Symptom-based COVID-19 screening can be improved by incorporating data collected from wearable sensors into prediction algorithms, an approach that could complement ongoing testing efforts by spotting symptomatic and pre-symptomatic individuals early, according to a research letter recently published in nature medicine.

The findings are part of the DETECT (Digital Engagement and Tracking for Early Control and Treatment) program headed by Scripps Research Translational Institute that was announced in late March and received a major push from Fitbit in May.

The effort enrolled thousands of consumer participants, the researchers wrote, and combined self-reported symptoms and diagnoses with data pulled directly from either their Fitbit accounts (78.4%), Apple HealthKit (31.2%) or Google Fit (8.1%). However, relatively few of the study’s participants provided COVID-19 testing data to the analysis.

“Our results show that individual changes in physiological measures captured by most smartwatches and activity trackers are able to significantly improve the distinction between symptomatic individuals with and without a diagnosis of COVID-19 beyond symptoms alone,” the researchers wrote in nature medicine. “Although encouraging, these results are based on a relatively small sample of participants.”

TOPLINE DATA

Among the 30,529 device owners who enrolled in the study, 62% were female and 12.8% were aged 65 years or older. Twelve-and-a-half percent of the participants reported at least one COVID-19 symptom, with 54 symptomatic participants reporting positive COVID-19 test results and 279 reporting negative results.

Looking at the wearable sensor data, the researchers found significant differences in sleep and activity metrics between positive and negative cases. However, changes in resting heart rate greater than two standard deviations were detected in only 30.3% of positive cases, which the researchers wrote was not statistically strong enough to discriminate between cases on its own.

Combining these three measures into a single metric yielded an area under curve (AUC) of 0.72, greater than that of any individual wearable metric but still on par with a model that only considered the self-reported symptoms, age and sex (AUC = 0.71), the researchers wrote. Marrying these two approaches into a single analysis significantly improved (P < .01) on both models, achieving an AUC of 0.80.

HOW IT WAS DONE

Between March 25 and June 7 of this year, the researchers openly enrolled adults living in the U.S. through the study’s website and through collaborations with partners such as Fitbit, Walgreens and CVS/Aetna. Participants enrolled by downloading the study app and consenting to share some combination of their device data, symptoms, COVID-19 diagnostic test results and other information in the EHR.

WHAT’S THE HISTORY

The researchers wrote that their work builds on the COVID Symptom Study app, a tool deployed for UK and U.S. residents earlier this year. Data from the project published in May included more than 2.5 million downloads and more than 18,000 molecular test-confirmed cases, highlighted a link between certain self-reported symptoms and positive cases, and yielded a symptom-based prediction model.

But the Scripps-led DETECT study wasn’t the only health wearable study launched during the pandemic. UC San Francisco’s TemPredict Study, for instance, is using Oura Rings and participant surveys to similarly “identify patterns that could predict onset, progression, and recovery in future cases of COVID-19,” while some COVID-19 vaccine trials are using health wearables to monitor the health and fitness of their study populations.

The data from the DETECT study is early support for other programs Fitbit is kicking off to help detect COVID-19 before symptom onset. Among these are a project with the U.S. Army Medical Research and Development Command, and a study with Northwell Health’s Feinstein Institutes for Medical Research.

IN CONCLUSION

“These results suggest that sensor data can incrementally improve symptom-only-based models to differentiate between COVID-19-positive and COVID-19-negative symptomatic individuals, with the potential to enhance our ability to identify a cluster before more spread occurs,” the researchers wrote. “Such a passive monitoring strategy may be complementary to virus testing, which is generally a one-off, or infrequent, sampling assay.”