How can we spot the "silent spreaders" of SARS-CoV-2 -- the people who are infected and contagious, but have no symptoms? [1] Our digital medicine group at Scripps Research Translational Institute has just launched a health study [2] to explore one possible approach: crowdsourcing data from wearable devices like the Apple Watch and Fitbit, and then attempting to use machine learning (or less grandly, linear regression) to identify COVID-19 infections.
In January, my colleagues at Scripps published a study in the Lancet that used heart rate and sleep data from about 47,000 Fitbit users in the U.S. to predict which areas would be hardest hit by flu. [3] The correlation with actual reports from the Centers for Disease Control was very high: 0.9.
We’re hoping that we can do something similar for COVID-19 infection. In particular, elevated resting heart rate appears to be a reliable indicator of viral infection. [4] We’re wondering whether it might also be a leading indicator, i.e, spiking to a detectable degree before any measurable fever. If it were, we might be able to identify COVID-19 infections in individuals -- and areas with outbreaks -- before people even have any symptoms.
To figure this out, we’re going to need a lot of participants. And that’s really the greatest challenge for us now. A similar project in Germany [5], which was inspired by our Lancet flu study, already has more than 160,000 volunteers. But they have the advantage of being sponsored by a national research institute. We don’t.
If you have suggestions about how we might reach people with wearables who want to help out -- or if you have questions about the DETECT study -- please add them to the comments.
I really wish Apple, Garmin, Google, Fitbit would start working on body temperature sensors. It doesn't need to be realtime like a heartbeat sensor to be useful, but seeing a 0.5-1.0 average daily body temperature increase could encourage A LOT of people to stay home or go get tested
My Garmin watch has a thermometer, a heart rate sensor and an oxymeter.
Temperature is only recorded during activities, and as it's measured inside the watch the value is somewhere between body and environment temperature. Unless environment and fit of the watch are controlled, I would not expect useful readings.
The oxymeter is also not very good, probably because the position at the wrist is suboptimal and the sensor isn't opposite of the LED. The light shines into the skin, not through it, as with the oxymeters that are clipped onto a finger. Measurements take about 20 seconds and are wildly off and inconsistent. I just measured 99-100% saturation with a clip-on oxymeter and 92% with my watch.
The value I trust most is the resting heart rate. It's not just the minimum value recorded, the watch seems to apply some heuristics to figure out whether a measurement was clean or not. The sensor data itself must be very noisy. When the "abnormal HR" feature is turned on, I get false alarms multiple times per day. I had a colleague who thought her HR was pretty constantly at 60, because that was the default value of the Fitbit that she wore like a loose bracelet.
To provide some anecdotal data, I had a bronchitis last year and the RHR went from 49 to 69 at its peak. At the same time, the "Body Battery", which is calculated from HR data, also went down to 0%. I was surprised that the feature actually works, but definitely didn't need it to figure out I am sick.
Yeah, there would haven't to be some sort of correction from "temperature measured at the wrist" to "estimated core temperature" using some fancy models or ML. I have a feeling there hasn't been a lot of research done by Garmin into doing this.
I suspect one problem is that their current health wearables don't sit on generally accepted core temperature measurement points. Wireless earbuds would be a great place to stick an infrared temperature sensor, though.
Not exactly wearable, but I have learned in the midst of all this that fingertip pulse oximeters cost only $20-40 retail. Has there been any work to see if falling oxygenation readings can serve as a precurser? In any event I have started charting my own oximeter readings along with my temperature, both for baselines. It looks like I should add pulse rate as well.
Finger pulse oximeters can definitely provide useful data, but I suspect that they're lagging indicators for Covid-19, because typically, severe respiratory symptoms that affect blood oxygen levels appear later in the course of the disease. (But my background is in behavioral science, not pulmonology, so I might be wrong!)
Agreed, but I saw an unverified comment somewhere suggesting someone had a less than <90% reading and was still mobile. I don't know if that is possible, but I mentally filed that next to incidents of people just collapsing on the street. To the extent hemoglobin disorder might??? be involved (per Chris Martenson), there could? be a mechanism for low blood oxygen without lung/respiratory disorder per se.
For those of you that are located in Germany and want to help with their data: the RKI has an app[1] like that and uses heart rate, resting heart rate, sleep, and steps to determine the spread throughout the country
As you might be aware, California Governor Newsom in his TV briefing yesterday referred to encountering an abundance of tech-inspired proposals -- as I understood this, in the areas of monitoring and tracking. I also heard him to say there will be some attempt made to "pick one" so-to-speak, for a focus in participation, tech and data.
Thanks, I hadn't heard that yet. This might be one of those situations where it's preferable to "let a thousand flowers bloom."
For example, in another comment, I mentioned a study that's using the Oura ring, which is a great product, but has a far smaller user base than the mass-market Fitbit and Apple Watch audience that our study is focused on. So it's really additive to have both of our research efforts going forward.
To be fair Newsom might have implied several winners, but same point taken.
ABC7 in SF just reported a collaboration as "Stanford and Fitbit", giving innovators.stanford.edu (not working) as a site to sign up. Relatedly I find the following:
I could imagine participants in any device-specific work opting-in to sharing anonymized data with one-or-more broader-based studies/analyses, if that is not part of opting-in to the device-specific work in the first place.
It would be great if there were a way to join multiple studies at the same time. At the moment, though, you have to sign up separately for the Stanford COVID-19 Wearables Study:
https://innovations.stanford.edu/#section-wearables
If you have a Fitbit that includes heart rate monitoring, please consider joining the study.
Unfortunately, as a non-profit academic institution, we don't have any budget for advertising. I wish we could run millions of dollars of ads online and on TV to encourage people to join. Instead, I guess we'll have to rely on word of mouth.
Why not cut through the red tape and collaborate with research groups in other countries? These groups can release apps for their respective regions, and as long as the data generated by the apps is compatible, it can be aggregated any way you want.
That's a great suggestion. To do that, though, I'm pretty sure that we'd still need approval from our IRB. When you do research within an academic institution, everything needs to be looked over by a relatively impartial third party. And in general, I think that's a good thing -- especially when the research involves human or animal subjects.
28 comments
[ 3.4 ms ] story [ 66.1 ms ] threadIn January, my colleagues at Scripps published a study in the Lancet that used heart rate and sleep data from about 47,000 Fitbit users in the U.S. to predict which areas would be hardest hit by flu. [3] The correlation with actual reports from the Centers for Disease Control was very high: 0.9.
We’re hoping that we can do something similar for COVID-19 infection. In particular, elevated resting heart rate appears to be a reliable indicator of viral infection. [4] We’re wondering whether it might also be a leading indicator, i.e, spiking to a detectable degree before any measurable fever. If it were, we might be able to identify COVID-19 infections in individuals -- and areas with outbreaks -- before people even have any symptoms.
To figure this out, we’re going to need a lot of participants. And that’s really the greatest challenge for us now. A similar project in Germany [5], which was inspired by our Lancet flu study, already has more than 160,000 volunteers. But they have the advantage of being sponsored by a national research institute. We don’t.
If you have suggestions about how we might reach people with wearables who want to help out -- or if you have questions about the DETECT study -- please add them to the comments.
Thanks, Danny
[1] Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus. https://science.sciencemag.org/content/early/2020/03/24/scie...
[2] Pivoting health research to take on COVID-19. https://www.weforum.org/agenda/2020/03/pivoting-health-resea...
[3] Harnessing wearable device data to improve state-level real-time surveillance of influenza-like illness in the USA. https://www.thelancet.com/journals/landig/article/PIIS2589-7...
[4] Fever and cardiac rhythm. https://jamanetwork.com/journals/jamainternalmedicine/articl...
[5] Germany launches smartwatch app to monitor coronavirus spread. https://www.nytimes.com/reuters/2020/04/07/technology/07reut...
Temperature is only recorded during activities, and as it's measured inside the watch the value is somewhere between body and environment temperature. Unless environment and fit of the watch are controlled, I would not expect useful readings.
The oxymeter is also not very good, probably because the position at the wrist is suboptimal and the sensor isn't opposite of the LED. The light shines into the skin, not through it, as with the oxymeters that are clipped onto a finger. Measurements take about 20 seconds and are wildly off and inconsistent. I just measured 99-100% saturation with a clip-on oxymeter and 92% with my watch.
The value I trust most is the resting heart rate. It's not just the minimum value recorded, the watch seems to apply some heuristics to figure out whether a measurement was clean or not. The sensor data itself must be very noisy. When the "abnormal HR" feature is turned on, I get false alarms multiple times per day. I had a colleague who thought her HR was pretty constantly at 60, because that was the default value of the Fitbit that she wore like a loose bracelet.
To provide some anecdotal data, I had a bronchitis last year and the RHR went from 49 to 69 at its peak. At the same time, the "Body Battery", which is calculated from HR data, also went down to 0%. I was surprised that the feature actually works, but definitely didn't need it to figure out I am sick.
https://support.garmin.com/en-US/?faq=VOFJAsiXut9K19k1qEn5W5
I wish that the formula used to calculate it were less opaque. If it were open source, researchers might make use of it.
You might get more volunteers and data if you remove the geographic restrictions.
[1]: https://corona-datenspende.de
For example, in another comment, I mentioned a study that's using the Oura ring, which is a great product, but has a far smaller user base than the mass-market Fitbit and Apple Watch audience that our study is focused on. So it's really additive to have both of our research efforts going forward.
ABC7 in SF just reported a collaboration as "Stanford and Fitbit", giving innovators.stanford.edu (not working) as a site to sign up. Relatedly I find the following:
https://hitconsultant.net/2020/04/14/fitbit-scripps-stanford...
I could imagine participants in any device-specific work opting-in to sharing anonymized data with one-or-more broader-based studies/analyses, if that is not part of opting-in to the device-specific work in the first place.
Unfortunately, as a non-profit academic institution, we don't have any budget for advertising. I wish we could run millions of dollars of ads online and on TV to encourage people to join. Instead, I guess we'll have to rely on word of mouth.