A San Francisco smart thermometer company says its data suggest social distancing policies enacted to slow the spread of COVID-19 are working.
“Restrictions are Slowing Coronavirus Infections, New Data Suggest,” the headline on a widely circulated story in the New York Times that cited data from Kinsa, a manufacturer of smart thermometers.
Kinsa found that after spiking to atypical levels in the middle of March, the share of the U.S. population with a fever has dropped in the last week and a half. That might mean social distancing measures are reducing the share of Americans who are sick.
Policymakers in Minnesota say social distancing policies enacted in recent weeks will slow the spread of COVID-19. Data from Kinsa’s smart thermometers seems to back that up.
Smart thermometers
Kinsa’s thermometers take people’s temperatures and send them, along with location data, back to the company, where it is anonymized and combined at broader geographic levels.
“We are able to aggregate an illness signal, essentially,” said Nita Nehru, the company’s director of communications and partnerships. “It’s where and when are fevers starting and spreading across the U.S. to help give us a sense of where something unusual might be happening so public health first responders can target those areas before an outbreak becomes an epidemic.”
Kinsa has more than a million thermometers across the U.S. Since COVID-19 hit the U.S., the company has been getting as many as 162,000 temperature readings daily.
Public health officials have praised Kinsa’s data for being able to track spikes in fevers quickly. In past flu seasons, the company’s data has tracked closely to the Centers for Disease Control’s, but is collected in real-time versus being reported through health care providers.
To check for atypical illness levels, Kinsa measures the share of illness in a given geography against a baseline level it developed with the help of Benjamin Dalziel, a professor at Oregon State University who studies infectious disease.
“You can call it an illness signature of a geography. So in particular geographies, illness — like flu, for example — will behave in a particular way,” that often differs from place to place, Nehru said.
“So what we were able to do was look at the illness signature of geographies across the U.S. and come up with a forecast,” she said — a prediction of what illness would look like at any given time based on typical behavior.
If fevers are above the expected level for a given point in time, Kinsa calls it atypical illness.
[image_credit]Kinsa[/image_credit]
If fevers are below expected levels, that suggests something is causing less disease overall than is expected for a given time of the year. That may mean social distancing measures are working — not just to slow the spread of COVID-19, but also other diseases.
Fevers falling
The U.S. saw a spike in atypical illness — denoted in red in the chart above — in the second week of March. Illness levels dropped back to typical levels, then below them, in the third week of March.
[image_credit]Kinsa[/image_credit]
[image_credit]Kinsa[/image_credit]
In Minnesota’s most populous county, Hennepin, the level of illness has not veered into atypical territory, but it has been below what Kinsa would expect since early March. The share of illness in the county leveled off a bit mid-March and declined again around March 19 — two days after Gov. Tim Walz’s executive order closing bars, restaurants and other places of amusement took effect. (You can look up Kinsa’s data on any county in the Lower 48 here).
[image_credit]Kinsa[/image_credit]
“People are very inconvenienced and, for some people, it’s very hard, very lonely,” Nehru said. “The key takeaway is that it’s working, the sacrifices are working.”
What can Kinsa’s smart thermometers say about our future with COVID-19?
It’s not a predictive technology, but Nehru said we’d expect illness levels to stay low as people practice social distancing.
“I think the question that we’re all going to be grappling with is when social distancing starts to let up which has to happen at some point, are illness levels going to spike back up again or will we be able to find a balance?”
Couldn’t this simply be a matter of more people taking their temperatures who are not sick, but only cautious? Their normal temperatures would bring the average down.
Perhaps. One might also consider that only fairly well-off people, middle or upper class, would be able to afford these smart thermometers, which apparently cost $35-$70. Such folks may be more easily able to self-isolate during the pandemic, in comparison to the low-wage Wal-Mart, fast food, and grocery store employees who are still expected to go to work and interact with the public. We also don’t know the geographical distribution of how many of these devices are out there – which might explain why these maps show Miami doing much worse than NY, Louisiana, or Seattle, when this is almost certainly not the case.
Not to say this isn’t useful info – in a crisis all information is useful – but to be taken with a grain of salt.
It would be good to know what thermometers are reporting to them. It seems likely to only be those in some care facilities, since I doubt home thermometers have that capability.
We need to take into account that this is data only about people who own this particular type of thermometer. From my quick searches, Kinsa thermometers are not very widely available—they’re sold at Best Buy and CVS (currently sold out), but not Target or Walgreens or Walmart. And they’re not the cheapest available thermometer; at CVS, the Kinsa thermometer is $21.99, which isn’t crazy, but there are others that cost half as much.
This means that people with limited budgets are probably not buying Kinsa thermometers, and Kinsa’s data probably comes from people somewhat towards the wealthier end of the spectrum—who are also probably more likely to have the option of working from home.
Hello Elsa,
Thank you for your research on these thermometers. I wasn’t certain who were using them and I neglected to go online to find the sellers. However, I am also interested to know of what quality they are, at only $21. I am not upper income; however, I do work from home and have been wondering where to buy a thermometer as I do not live close to a commercial district and don’t often get out of my home.
Thanks for this, Greta. I was only vaguely aware of the thermometers, and didn’t know the data was being collected and aggregated at the manufacturer. For those making policy, this should be very useful information.
The last paragraph is important. The math shows that it will kick back up because we weren’t allowing immunity to build up. The Bayesian analysis of the Spanish Flu also pointed out that social distancing doesn’t work. It’s straight forward math as well. What ends up happening is a 2nd wave that’s worse than the 1st.
I was thinking the same thing as Thomas…I first got a Kinsa thermometer because of COVID, and since March 1 have been taking my temp twice daily….ranges from 98.2 to 99.1 consistently….
To Greta Kaul,
This is an excellent article, and I began to forget it was not written by a national or international agency. However, Thomas Quinn’s curiosity is something which I share.
Who owns these smart thermometers? Are they private citizens, clinics and hospitals and school, or both sets?
I hope you have time to respond. Your use of graphics and sharing timelines was excellent! Your comment with regard to the shut downs are not for naught was an excellent opinion and feathered into your article in a nuanced manner so as not to offend anyone and to highlight and identify that these measures are working. Well done!
This said, it is clear that we are not over the hump, yet. We mustn’t let a week or two of success interfere with our attention to the likelihood of second and third waves of the virus striking our population centers.
Governor DeSantis, of Florida, is wrong to not impose a statewide stay at home order. I have a medical student in Florida who is very concerned about his family. A healthcare researcher for twenty years, now aiming to go into the practice of medicine, he questions the wisdom, or the lack thereof, of the governor’s not placing all counties in a stay at home measure. Currently, only some of Florida’s counties are being shut down in a manner similar to what we in Minnesota are experiencing at this moment.
We should remember that travel is not terribly restricted in our nation, and that people traveling from highly infected regions to lesser or non-infected regions, can spread this deadly virus (again, ten times deadlier that other coronaviruses we know, such as various forms of the flu and common colds).
The overall writing, again, was excellent, but please let us know who owns the smart thermometers.
Ms. Kaul,
Please tell us who are the primary users of the smart thermometers. This would do much to qualify the quantity of data you mentioned in your article.
Thank you.
I’m sorry but this just plain silly. This is a press release pretending to be legitimate epi surveillance and monitoring. What public health officials are applauding this company? Unless these temps are being taken with scientifically designed sampling methods they are almost certainly not getting a reliable sample. This is a retail product that is distributed according to sales, and those sale represent a particular customer profile, not a reliable sample of the population. I’m surprised so many news agencies are reporting this. It’s more likely that the customer profile will tell you more about who these people are and why they’re recording these temps than exposure to COVID 19.
Mr. Udstrand,
Please tell us who are the primary users of these smart thermometers. Are they school, hospitals, clinics or private individuals and families. You seem to have a lot of knowledge about this, so I would appreciate your response. I hope you monitor your submissions to MinnPost, as I do.
In my opinion, the only weak point in the author’s article was not stating who are the users of this product.
Finally, I hope that you, like myself, are finding ways to cope with being shut into your home. Taking walks, watching movies on You Tube and Netflix are a couple of ways that I have been relaxing, as well as corresponding with family and friends, and enjoying conversations with others by telephone.
With best wishes.
Barry, the primary users are those who purchase this product. That’s why it’s not a representative sample, the product has not been distributed in any methodical way to monitor the population, this customer base is almost certainly not a reliable population sample.
Hello,
The company outlines statistical modeling techniques it uses to account for how often people are taking their temperature and the density of its thermometers in its methodology report (https://content.kinsahealth.com/covid-detection-technical-approach)
The short answer, from its FAQ:
Q: Are you seeing increased activity and does this affect your illness signal?
A: As of March 2020, we are seeing 2-3x the number of users taking temperatures than we’ve tracked in previous flu seasons. This does not impact our illness signal, as our modeling already accounts for rapid changes in our user base. We also benchmark our signal against the Centers for Disease Control (CDC) at the end of every flu season when the CDC has finalized their illness reporting.
As for users, the thermometer is sold in retail stores, and the company has a schools program (https://www.kinsahealth.co/enterprise/kinsa-for-schools/kinsa-for-nurses/)
Thanks for the additional information Ms. Kaul.
The problem however is that this is NOT legitimate population research, so the statistical methods the company used can’t make this claim valid. This company isn’t doing epidemiology, they’re just collecting customer data, which is a very different animal.
As I pointed out previously, this companies customer base and profile are almost certainly NOT representative of the population at large. There’s reliable way to extrapolate from this customer base to the general population, much less a population of COVID infected individuals. It doesn’t really matter how you analyze your sample if the the sample isn’t reliable.
In other words, you’d have to distribute these thermometers to a methodologically reliable population sample, marketing and sales are not scientific methodology.
One other problem I didn’t mention earlier is that you can’t just measure “fever’s” because not all fevers are associated with COVID 19. Even if you see a drop in temps across the population that’s not necessarily telling us anything about the pandemic. Those temperature drops might be associated with the passing of seasonal flu illnesses for instance. Flu incidence might well be dropping faster than COVID 19 is increasing at the moment, so even if you these things were distributed scientifically you might see a drop in fevers, but that drop wouldn’t have anything to do with COVID 19.
This ends up not being relevant or reliable “data” for two basic methodological reasons:
a) These thermometers are not monitoring a representative sample of any kind, it’s just company tracking their own customers.
b) The data being collecting isn’t necessarily associated with the pandemic, therefore it doesn’t tell us anything about the pandemic. Even if you see temperatures drop, unless you know those temps are dropping among COVID 19 cases, you can’t suggest you’re seeing a drop in temps associated with COVID 19. There are other reasons you might be seeing temps drop, and you have to rule those out before make COVID 19 claims.