Excess Death Counts and COVID-19

How many people really die because of COVID-19 in the US? Reported numbers exceeded 100,000 a couple of days ago, but the reported numbers are subject to errors, biases, and even outright fraud. If someone dies of COVID-19 without ever getting tested, the death will not be included in the COVID-19 statistics in many states. That would cause the reported number of COVID-19 deaths to be lower than the actual number of COVID-19 deaths.

But on the other hand, some states will report "probably" COVID-19 deaths, even if there is no confirmation from a test. Hospitals may be able to get higher payments if an uninsured patient is reported as a COVID-19 patient, which could create an incentive to falsely report COVID-19 cases (although the opposite is true of uninsured patients have enough money to pay the hospital bill, which will be several times higher than what the government pays). Or a patient may be COVID-19 positive, but die from an unrelated disease - and still be counted as a COVID-19 death. All these examples could lead to over-reporting of COVID-19 deaths.

So - is the real number of COVID-19 deaths higher or lower? Or do over- and under-counting simple cancel each other out, and the reported number is correct? To answer this question, we can look at the total number of reported deaths. If COVID-19 really causes additional deaths, that should be reflected. That's how the estimates of annual deaths from influenza are calculated - the actual number of confirmed influenza cases is just a small fraction of actual cases, and "flu deaths" are usually not directly caused by influenza, but rather by weakening a person so that they then die from pneumonia or another "opportunistic" disease.

COVID-19 caused > 20,000 excess deaths

We can start by looking at the CDC web site, which gives us graphs and data links for death statistics. Here's a graph that shows total weekly deaths in the US since last October:

The orange line shows a "maximum expected" number of weekly deaths, based on averages from the previous five years. Towards the end of March, the reported number of deaths was significantly higher than the maximum expected number. The blue bars indicate deaths where COVID-19 was listed as one of the causes; it is easy to see that reported COVID-19 deaths were responsible for most of the excess deaths, but not for all of them. If you look at the original graph on the CDC web site and hover over a bar, you can see the expected and actual death counts:

For the week that ended April 11, the actual number of deaths was 77,759 - that's more than 21,513 more deaths that expected for this week, and increase of more than 25%. For this week, the Johns Hopkins COVID-19 data show fewer than 14,000 reported deaths. The analysis of death certificates (downloaded from here) gives a slightly higher number of 15,333 COVID-19 deaths. Adjusting for incomplete reporting (see the next section) increases that number to 16,203; that leaves an "excess" of 5,310 that were not reported as COVID-19 deaths.

The effect of delayed death certificate processing 

The CDC analysis is based on death certificate data submitted by the states and territories, which arrive with varying delays that can go up to a full year. Therefore, the numbers for the past ~10 weeks are known to be incomplete. The CDC tries to compensate for the not-yet-reported death certificates using elaborate algorithms described on this CDC web page, but it does so in a very conservative way that is much more likely to under-estimate the effect of missing reports. This is quite obvious when comparing the numbers before and after a weekly update; for the week of 4/11, the last weekly update increased the number of deaths by 1,017. Additional increases are very likely to occur in the next few weeks as more data come in, with the largest increases in the more recent weeks. To minimize these distortions, the next section will focus on the 4 weeks from 3/21/2020 to 4/11/2020.

More than 12,000 excess deaths in 4 weeks not reported as COVID-19

The following table shows the expected and actual weekly deaths for the 4-week period that ended on 4/11/2020:

The number of COVID-19 deaths in the table is based on the CDC mortality data downloaded from this CDC web page, and corrected for estimated under-reporting, which increased numbers by 1-5%. Of the 42,348 excess deaths in the 4-week period, only 29,453 deaths certificates list COVID-19 as a cause of death. This leaves 12,895 excess deaths not explicitly linked to COVID-19 for this period. The most likely explanation for these excess deaths is that they were also caused by directly or indirectly by COVID-19. The CDC page can also show graphs about select causes of deaths for 2020:

The graphs shows just a small uptick in deaths from respiratory diseases and cancer, but a larger uptick for dementia and circulatory diseases. The small uptick for respiratory diseases indicates two things: (1) that most COVID-19 deaths that were directly caused by acute respiratory failure were classified as COVID-19 related, and (2) that COVID-19 was not "always and automatically" assumed when people died of respiratory problems. The larger increases in deaths classified as dementia and circulatory diseases are in agreement with what we know about the disease: it primarily kills old people, and many of its most severe effects, like strokes and blood clots, are linked to the circulatory system. The onset of very severe COVID-19 symptoms is often very sudden, and also often seen after a patient seems to recover from the initial symptoms, which can easily be confused with flu or a common cold. Therefore, a diagnosis of dementia or circulatory disease as the cause of death is very plausible, especially when access to COVID-19 tests is limited.

Are there 43% more actual COVID-19 deaths than reported?

It is reasonable to assume that all excess deaths during the height of the COVID-19 epidemic in the US are caused by COVID-19, either directly or indirectly. If that is indeed the case, then the analysis for the 4 week period until 4/11/2020 shown above indicates that the actual number of COVID-19 linked deaths is 43% higher than the reported numbers. If that would apply to the widely noticed landmark of 100,000 COVID-19 deaths, then the actual number of deaths would exceed 143,000.

However, before jumping to a quantitative conclusion, there are several important facts that we need to consider. The increase in COVID-19 testing in recent weeks has reduced the chance that a symptomatic patient cannot get a COVID-19 test, which removes one of the reasons for an incorrect cause on a death certificate (or, more accurately, reduces its importance). Furthermore, the CDC has changed the counting of COVID-19 deaths on 4/14/2020 to include "probable" death; however, it is not clear is this also resulted in a reporting change on death certificates, and not all states in the US follow this practice.

While more tests and better reporting should decrease under-reporting of COVID-19 deaths, other factors may have the opposite effect - but that will be the topic of a different post. Ignoring these factors, we can speculate that the reporting of COVID-19 related deaths has improved in recent weeks. But whatever the current exact number may be, it is very likely that there is still very significant under-reporting of deaths caused by COVID-19. For the 4-week period that ended April 11, the death certificate analysis shows that without any doubt, the number of excess deaths was under-reported my more than one third.