Post #579: No inflection point for Virginia today, but maybe the Kinsa data show us where we’re headed.

Posted on March 28, 2020

Source:  Kinsa health weather map.  Blue = declining trend in abnormal amounts of fever, compared to what would be expected at this time.

Source:  Kinsa health weather map.

Kinsa makes an internet-connected thermometer, and they use the resulting data for, among other things, monitoring trends in flu.  Lately, it seems to have value in monitoring trends in coronavirus.

Above, the dashed blue line is the amount of fever they expect to see at this time of year.  Band around that is (probably) the 95% confidence interval (what you could reasonably expect to occur, purely by chance).  Orange/red line is the actual observed level of fever.  Red is “atypical”, meaning above that 95% confidence interval band.  The little peak of the red portion is literally the day after Fairfax County shut the schools.  So, FWIW, their count of atypical fevers has been declining pretty much ever since Fairfax County closed the schools.

Caveat:  You can never be sure whether that’s a true result related to coronavirus, or just an artifact of odd changes in people’s behavior in these odd times.  That solid-blue map just looks too good to me, and I wonder whether (e.g.) the lack of kids catching the flu, with schools closed, is swamping any signal they might be getting from a change in coronavirus cases.  At this point, I don’t see anyone pointing to this and jumping up and down, other than a single business publication (see below).  And Kinsa itself is not saying that this heralds a decline in future coronavirus cases.

It is what it is:  it’s a general, non-disease-specific measure of the level of fever in the US population.  Plausibly, social distancing may have perturbed the calibration of their system.  (In effect, suppressing the normal levels of normal flu).  But at the minimum, it shows that any continued growth in coronavirus is too small to be seen against that background.

In terms of real-time data, it looks like that’s the only source.  It’s the only source I’ve found so far.  And for what it’s worth, it’s showing declines in unusual amounts of fever all across the US.  On the way up, this thing did seem to be flagging hotspots of coronavirus activity.  You have to cross your fingers and hope that it’s flagging the reduction of those hotspots, on the way down.  But you have to be aware of the possibility that the vast change in population behavior with our “social distancing” orders might produce a false signal.

The maps above were updated as of about midnight yesterday (midnight of 3/27/2020), per the Kinsa website.

 No inflection point in cases yet, but maybe there’s a long lag.

Even with just a part-day update, it’s clear that Virginia is not going to see an inflection point in case counts today.  The increase (though noon today) was almost as large as yesterday’s entire increase.  The Fairfax County number had not yet been updated from yesterday’s count of 124 cases.

That said, since this string of posts is all about finding the first signs of the inflection point (the point at which daily increase in cases slows), I thought I’d at least seek out some people who can offer a hopeful viewpoint.  Any sign at all that some inflection points are on the horizon.

This is, after all, a disease.  It spreads by real, knowable vectors.  It does not spread by magic.

First, I now suspect that my simple arithmetic, based on the Chinese experience, was simply too optimistic.  I now suspect that, in the US, there are longer lags between the time of infection, and the time that testing gets done and centrally reported.  If true, it’ll take longer to reach that point — not because the severity of the epidemic is any worse, just because it takes so long to get the cases reported.

Second, not that Business Insider is my go-to source for science news, but they do have a couple of nice cuts at potentially interesting data.  You might put eyes on this article if you want at least a hint of some light at the end of the tunnel.  The gist of it is, maybe the pool of sick-but-not-yet-diagnosed cases that is feeding overall COVID-19 growth may be shrinking, based on real-time data from Kinsa, a “smart thermometer” maker.

For what it’s worth, Kinsa’s “health weather map” shows that atypical rates of fever are trending downward in Fairfax County.   (Note:  That site only works with Chrome for me, not with Firefox.)  I note, however, that the Kinsa sample size appears quite small, for doing small-area estimates.

Based on this New York Times article, their system appears to have a proven record for predicting flu outbreaks, and appears to have identified future coronavirus hotspots before those places were identified via testing.

And, there have been a couple of peer-reviewed published studies of their system.  You can find the links to that on their website (link given under the map).  Offhand, I’d say they have done the homework to back up their claims.

Currently, they show that reported atypical levels of fever are declining all over the US.  But to me, the picture looks a little too good.  It’s hard to believe that this is starting to be beaten back everywhere.

But this appears to be the only real-time indicator out there.  If I can find any other real-time indicators, I’ll add them to this post later today.