Counting COVID-19 deaths

06 Jan 2022

There has been a great deal of controversy over the way COVID-19 deaths are calculated. Initially in the UK, the count that was most widely reported consisted of all who died within 60 days of a positive PCR test, or who died more than 60 days after a positive test but had COVID-19 listed as a cause of death on their death certificate.

The main objection to this is that, for example, someone could test positive for COVID-19, recover fully, then die in e.g. a car crash and still be counted as a COVID-19 death.

What are we trying to do when we count COVID-19 deaths anyway?

It’s important to be clear what we’re trying to achieve with the COVID-19 death count. I think most people agree that we should be aiming to count the number of deaths caused by COVID-19. But what does that mean exactly? In the Scottish data, the average number of comorbidities in people who died from COVID-19 is upwards of 3. How do we assign causation when there are multiple contributing factors?

Ideally, what we would like to do is a controlled experiment. We would ‘run’ a parallel version of the world that is completely identical, except SARS-CoV-2 is non-existent. And when I say identical, I really mean it. Everyone believes just as much in the reality of SARS-CoV-2 as they do now, people are furloughed from work en masse, governments impose social distancing and lockdown measures etc. Everything the same, it’s just that there ain’t no such thing as SARS-CoV-2.

The point is that we only changed one thing in our experiment. Therefore any difference in outcome can definitively be attributed to that one change.

Of course this wouldn’t really be feasible (not to mention ethical), but at least now we know what we’re aiming for.

Hasn't someone spent a lot of time thinking about this problem before?

Yes. There is a large literature on, for example, how to define deaths due to influenza. The proximate cause of death in most people who die as a result of influenza is pneumonia. But they also often have several comorbidities. One might typically see a flu death defined as death after a hospital admission with flu recorded as the reason for admission. However, the question is far from being settled.

What should we do with COVID-19 then?

On August 12 2020, Public Health England changed their main definition of a COVID-19 death to be a death within 28 days of a PCR positive test. This led to the primary COVID-19 death count dropping by about 13% overnight. Still, it remains possible for car crash victims to be counted as COVID-19 deaths, and there is the unfortunate fact that the rate of COVID-19 deaths increases as we do more testing, even if the rate of deaths actually caused by COVID-19 remains constant.

This is also quite different from how a flu death is usually defined. Perhaps there’s a defensible rationale for that; COVID-19 and flu are different diseases, and there is a world of difference in how we have reacted to them. For example, it might be reasonable not to require a positive PCR test in the definition of a flu death because PCR testing is quite rare in a typical flu season, which would likely lead to a significant under-count. On the other hand, in the UK we have ramped up PCR testing for COVID extraordinarily rapidly, from 10s of thousands per day in April 2020, to well over half a million per day at the time of writing. In my eyes, the more pertinent danger with our current definition of a COVID-19 death is over-counting. I think there is a reasonable case for further modifying the defintion of a COVID-19 death, to be death within 28 days of a postive PCR test and COVID-19 listed as the main cause of death on the death certificate, or something similar.

I really wouldn’t want to be the PR person who deals with the reaction to that, though.