Skip to main content

Wearable sensor data and self-reported symptoms for COVID-19 detection

Our take —

Researchers enrolled 333 persons in the United States with fitness tracking devices (e.g., FitBits) who had COVID-like symptoms and received a SARS-CoV-2 test. Variations in sleep and activity, alongside self-reported symptoms can suggest a possible positive COVID-19 case and may make individuals more aware of possible infective status prior to confirmed testing. However, it is also conceivable that participants changed their lifestyles as a result of learning that they had been exposed to a person with COVID-19 or that they were positive. The methods in this paper do not explicitly account for either of these possibilities and therefore, should be interpreted with caution.

Study design

Prospective Cohort, Retrospective Cohort

Study population and setting

This analysis was restricted to 333 people with fitness trackers (Fitbit, Apple HealthKit, Google Fit) in the United States who reported COVID-like symptoms, sought testing for SARS-CoV-2, and were enrolled in the study between March 25 and June 7, 2020. The authors evaluated whether the addition of sensor data (i.e., changes in resting heart rate, sleep, and physical activities) to symptoms could improve detection of people who reported a positive test result for SARS-CoV-2. Authors evaluated the ability of each model to correctly classify people with and without the disease using receiver operating curves (ROC) and the area under the curve (AUC). AUC is a common metric for simultaneously considering the sensitivity and specificity of tests that attempt to diagnosis a disease.

Summary of Main Findings

Fifty-four (16.2%) participants reported being positive for SARS-CoV-2. Symptomatic people who reported receiving a positive test result were significantly more likely to get more sleep and take fewer steps daily than those who reported a negative result. Combining both sensor and symptom data resulted in an AUC of 0.80, which was significantly better than self-reported symptom or sensor data alone. An AUC of 0.8 suggests an 80% chance that the test will correctly distinguish an infected from a non-infected patient

Study Strengths

This study drew participants from across the United States and gathered data from objective, real-time fitness trackers worn by participants.

Limitations

The most pressing limitation of this study is whether participants changed their sleeping and activity as a result of receiving their diagnosis. If participants altered their habits after their diagnosis, then the proposed diagnostic method (i.e., detecting changes in sensor data) would not be useful in predicting infection. The authors acknowledge that participation in this study was limited to persons with fitness trackers, who are likely not representative of the general population.

Value added

People diagnosed with COVID-19 may have COVID-specific symptoms and change their sleeping and physical activity habits, as measured by a fitness tracker. Fitness trackers and self-reporting may offer new ways to potentially identify COVID-19 positive individuals before RT-PCR testing is done.

This review was posted on: 14 January 2021