Back in August, we asked: can we monitor COVID-19 patients with smartwatches? Three months later, a research team from Stanford asked something even extreme: Can smartwatches detect COVID-19 even before symptoms arise?
Under one of Stanford’s long-standing digital health protocols, researchers had access to medical records of 5,262 participants. This group of people were a sample of the general population, with the difference that they regularly contacted Stanford to inform of their health status. Of these, 3,325 wore Fitbit devices. And out of all those Fitbit wearers, 27 got COVID-19 over the course of the last year. This gave researchers access to real-world Fitbit data from people that got COVID, before and during the disease!
What kind of data? Well, everything they could get their hands on. Heart rate, steps, sleep quality (however unreliable that may be) were derived from Fitbit. Thanks to Stanford’s protocol, the researchers also had access to disease symptoms and severity, tracked daily.
A common pattern was detected – at some point during the incubation period, patients started to have increased resting heart rate. This became even more obvious when adjusted for steps. If an increase in average heart rate happens because of a more active day, it is only logical that the number of steps will also be high. This marker was named Heart Rate Over Steps (HROS) by researchers. Quite expectedly, this variable could easily tell separate illness-related increase in heart rate over other causes.
The CuSum COVID-19 detection tool
Using data from these 32 COVID-19 patients and 73 matched Fitbit control wearers, the researchers managed to create the CuSum algorithm. The goal of this tool is to sound the alarm when abnormal readings arise. In a simulation that ran with the data from these patients, a COVID alert fired for 66% of COVID cases during the incubation period. Along with a further 17% that received an alert in the first days of symptomatology, Fitbit-derived data detected 83% of COVID cases!
Smartwatches detect illness, not COVID
Did CuSum detect COVID-19? No. As per the researchers, the algorithm detected the patterns of illness, not COVID specifically. This high sensitivity but low specificity tool could be really handy in the current COVID landscape. Regular testing of people whose smartwatches detect increased heart rate and increased HROS could lead to more efficient testing. In the coming months, pinpoint accurate testing will be necessary to root out pockets of disease in the community. It will definitely be interesting to see whether Fitbit capitalizes on these findings…