Tracking back the COVID cat as a research problem
May 25, 2020 12:35 PM   Subscribe

Since many of our fearless leaders worldwide don't want to bother ensuring widespread COVID testing and would rather understate the impact: how will researchers of the future estimate the rates of infections / death toll?

I've been thinking about this for a couple of days, especially after the New York Times' highlighting of 1,000 obits yesterday.

One way that researchers could go about it, I guess, would be to gather obit numbers and compare them to similar periods (March 2020 -> average of March 1990-2019, for example) in years gone by.

In terms of rates of infection: if data on emergency room / hospital / doctor visits is publicly available, perhaps do the same? This would be particularly telling in the United States where so many avoid seeking medical help because of the cost (or maybe it would be particularly skewed).

I'm not an epidemiologist or a scientist, far from it... but I wonder if anyone knowledgeable could chime in on potential methodologies for getting more reliable numbers.
posted by Sheydem-tants to Science & Nature (5 answers total) 1 user marked this as a favorite
You're right -- a lot of researchers look at excess deaths for the death toll. How many deaths are expected in a given timeframe, vs. how many actually happened.
posted by k8lin at 12:55 PM on May 25, 2020 [8 favorites]

The Economist has done a bit of work on this topic. Here is an explanation of their approach and calculations.

And the Financial Times also reports on excess mortality, with a separate article on their methods.
posted by lulu68 at 1:30 PM on May 25, 2020 [2 favorites]

It will take a lot of work and the end result will be a confidence interval, not a number.

Before testing became widespread people were being diagnosed with "atypical pneumonia"; what percent of those do you retroactively declare to be covid? Researchers will have to sort it out. There's a lot of talk in denialist circles about people who are diagnosed with covid but die from something else being included in official death counts -- this not a problem in the sense that denialists think it is, but it is something that has to be accounted for.

Then you have the effects of the pandemic + lockdown and on non-covid deaths. In hard-hit areas there will have been otherwise preventable deaths that weren't prevented because of the overloaded healthcare system. These can be attributed to the pandemic but not specifically to covid. On the other hand reduced travel and economic activity mean fewer traffic deaths than there would have been otherwise. And so on.

In the coming years lots of public health theses will be written trying to sort out different aspects of this by combing through electronic medical records and death certificates, and statisticians will try to model the actual number of covid deaths in a jurisdiction based on reported cases, excess mortality, demographics, lockdown level etc etc. Eventually systematic reviews will be done to pull it all together and come out with a sensible range. Excess mortality is a good heuristic for the scale of the problem but it's still not the whole story.
posted by no regrets, coyote at 2:00 PM on May 25, 2020 [3 favorites]

I think the traffic mortality might come out about even based on the news here. Less people on the road means a lot of people who drive faster and accidents are even more fatal due to the speed involved.

Ditto the excessive deaths, they'll look at the past and the time to come and WTF is that spike in 2020? Unless there's some other reason for a bunch of people to pick 2020 to die for no apparent reason, that would be the covid.
posted by zengargoyle at 3:00 PM on May 25, 2020

One way that researchers could go about it, I guess, would be to gather obit numbers and compare them to similar periods (March 2020 -> average of March 1990-2019, for example) in years gone by.

We don't have to gather obit numbers - that would be so imprecise as to be useless. The public health system builds what are called "surveillance systems," which includes hospital-submitted standardized reports on patients to departments of public health, and access to death records. So when people talk about "confirmed cases," they're not talking about an estimate - they're talking about the actual number of people who have a positive test in their hospitalization record (these data aren't publicly available, and they're only reported in the aggregate, because a medical record is private information).

But this isn't an adequate count of the number of covid cases that exist, because in places with strong outbreak, people were told to ONLY seek testing and care if they were VERY sick - the numbers of confirmed cases were undercounts of the number of actual cases. But now that the severity of the outbreak has lessened in a lot of places, and serology (testing the blood for antibodies) tests are getting rolled out. This can help us estimate how widespread infections are: what percent of the population HAS BEEN infected, whereas our previous numbers were just about how many serious cases there are RIGHT NOW.

These are all blunt tools. They involve some complicated statistics around uncertainty. They're not going to wind up with precise numbers. But they don't have to.

For deaths, yes - excess mortality will be helpful. In some places there is talk about counting "probable covid deaths" which tends to mean people who have died and who have "covid-19" listed as a cause of death, but don't actually have a positive lab test (yet) - so a judgement call made by the medical examiner. Excess mortality, though - as others have pointed out - will capture both people who died with untested covid, as well as people who have died of other things because, say, they weren't able to seek care during the outbreak.
posted by entropone at 4:34 PM on May 25, 2020 [5 favorites]

« Older Community Lego sorter?   |   Enhance! Newer »
This thread is closed to new comments.