29 April 2020
There’s a persistent, recurrent idea out there that COVID-19 isn’t really that dangerous after all, and that we are taking economic pain unnecessarily.
I won’t wade into the important and difficult debate about what the right tradeoffs are between minimizing excess deaths and minimizing economic pain, a debate that is becoming more pronounced as many countries see new cases level off and decline and begin to consider next steps.
But I do want to share the best current evidence and logic for why the view that we always could have allowed, or even could now allow,m the virus to run through the population unchecked is wrong.
Yes, it’s much deadlier than the flu
Two weeks ago, I wrote a long post about the question of whether it could turn out that COVID-19 has already infected vastly more people than we think — even the majority of the population in some countries / regions / cities. I concluded that while it’s certain that many more people have been infected than the number of reported cases, it was very unlikely (but not impossible) that, say, 50% of the US or UK has already been infected.
While this point of view appears to be shared by the vast majority of epidemiologists and infectious disease specialists who publish papers or give interviews in the mainstream press, there continues to be a small minority of commentators (some more qualified than others) articulating the alternative view, a view I call the “ubiquity hypothesis.”
For this group, the point of arguing that (in their view) a very large number of people have already been infected almost always follows the following logical path:
- If the number of people who have been infected is (say) two orders of magnitude higher than the number of reported cases, the Infection Fatality Rate (IFR) — the measure of the percentage of people who are infected (whether detected or reported or not) who ultimately go on to die — will turn out to be much lower (two orders of magnitude lower) than the Case Fatality Ratio (CFR), which by definition is calculated by dividing the number of people confirmed to have died with COVID-19 by the number of confirmed and reported cases.
- If the IFR is vastly lower than the CFR, the disease is less dangerous than we think.
- Therefore, we should re-open the economy rapidly and let the virus spread through the population relatively rapidly.
The first point is correct (as a conditional statement by definition).
The second point is wrong, or at least misleading, in two senses.
First, if by “we”, we mean experts, it’s simply wrong. Experts know, and take into account, that crude CFRs are not helpful as a forecast of ultimate mortality. There are a huge number of studies out there that try to estimate the true Infection Fatality Rate, and this is the basis on which professionals are forming a view of how “dangerous” the virus is, and what the risks would be of letting it run unchecked through the population.
Secondly, the point of comparison is almost always to seasonal influenza. The problem is, there’s another logical fallacy that creeps in here. Many people have heard a figure like 0.1% as the fatality rate for flu. And the advocates of the minority view above point out that if many more people have been infected with COVID-19 than we think, then the IFR of COVID-19 could be something like 0.1%, or even lower.
Now, it’s possible that the IFR of COVID-19 could be as low as 0.1%, though recent mainstream estimates typically put it in the 0.3%-0.8% range, with 0.5% probably close to a crude consensus estimate. And of course if it turned out that 100x the number of people have been infected compared to what we thought, the IFR would be even lower.
But it’s important to recognise that isn’t the Infection Fatality Rate for the flu; it’s much closer to something like the the Case Fatality Rate. That’s because the vast majority of cases of seasonal flu are never reported.
The epidemiologist Adam Kucharski explains this very clearly here. If we wanted to estimate the IFR for flu, it would be something closer to 0.02-0.05% of cases are fatal. Order of magnitude, that would make COVID-19 10x more deadly than flu.
Finally, as a sanity check, think what it would mean if the IFR of COVID-19 turned out to be 0.1%. If 60% of the world population ultimately was infected, that’s (60%*7.8B) = 4.7 billion cases. 0.5% of that is 24 million deaths.
Yes, excess mortality is high
I’ve discussed in several earlier posts why there are factors that lead to both over- and -under-attribution of deaths to COVID-19.
At the end of the day, when we talk about “attributing” deaths to COVID-19, what we care about is how many more people die because of COVID-19 than would have otherwise.
On this topic, the evidence is mounting every day that excess mortality is very high indeed. Here are a few examples of the evidence, among many.
Since 2008, the European Mortality Monitoring project (EuroMoMo), supported by WHO and ECDC, has been tracking “excess number of deaths related to influenza and other possible public health threats across participating European Countries.” Their data for 24 EU countries shows significantly higher-than-expected deaths:
In this dataset, the UK looks particularly badly impacted.
I’ve previously shared The Economist’s estimates of excess mortality (article dates from 16 April); this analysis is based on the EuroMoMo dataset as well.
The Financial Times has also been estimating excess mortality for a number of countries.
Looking at the US, the New York Times yesterday reported excess mortality estimates for seven states based on CDC data, suggesting :
The Washington Post published a similar analysis based on a Yale study:
The New York Times also looked at excess mortality in other countries, estimating at least 40K higher deaths, above and beyond those attributed to COVID-19, across just 12 countries.
Of course, there could be other sources of excess mortality besides COVID-19. Some people may be dying of conditions like heart disease or cancer who otherwise would have sought, and been able to obtain, treatment had lockdowns not been in place, had the healthcare system not been overwhelmed in some cases, and so on.
On the other hand, we also have evidence that mortality rates from certain other causes, such as car accidents, are lower — and maybe much lower– because of lockdowns.
So while there’s no way of being certain of the excess mortality that is directly due to the coronavirus, there’s strong evidence that it is significant.
No, the data do not suggest that the majority of people have already been infected.
The limited serological data we have continues to suggest low prevalence in most cases
In my earlier post referenced above, I shared a number of analyses and studies that suggested that the prevalence of COVID-19 was likely still low (<10%, often <5%) in most countries, though there are individual, highly-impacted cities and regions where it is likely higher. The few studies that show higher rates appear to have significant methodological issues.
We’re getting more data by the day from serological studies, and the weight of evidence is adding up against the “ubiquity hypothesis.” Here is a table summarizing findings from a number of studies:
Note that all but one of the cases with prevalence rates above 10% are in highly impacted areas, or small, contained populations.
A study that seems particularly well designed is in Miami Dade County, because it is a random sample of the general population, reasonably large, adjusted statistically for representativeness, and ongoing. This has found 6% incidence.
Still, there are reasons to continue to be sceptical about serological studies of any kind. Many tests appear to be flawed; and even tests that work as advertised suffer from high false positive rates. A good summary of some of the issues is here.
What do other coronaviruses tell us about immunity?
On a different topic, this MIT Technology Review article (paywall, but three free articles per month with registration) worried me. Apparently, for a number of common coronaviruses that cause colds, immunity is only partial and quite short-lived.
Subjective Probability 