Fatality Rates Aren’t Fixed: They Depend on What We Do
Fundamentally, there is no such thing as “the” fatality rate for a virus. How likely people are to die from COVID-19 (just like many diseases) depends on whether they can get access to treatment.
There is no such thing as “the” death rate for a virus. It changes based on what we do.
If the medical system is able to provide adequate care — as it mostly has in South Korea, China outside of Hubei, and for cruise ship passengers in Japan — fatality rates will be lower.
But if hospitals and ICUs are suddenly swamped with cases — like in Wuhan and Italy — deaths will rise, even though the virus itself isn’t behaving any differently. Patients who need a ventilator to breathe and don’t get it have a fatality rate of 100%. [source]
The fatality rate of coronavirus is under the microsope. It appears to be anywhere between 0.5% and 3.5%. More on this in a second. This is made worse when considering the state of different healthcare systems throughout the world, giving rates anywhere between 0.6% and 5% or higher. It also depends on the “when” factor – when in the crisis you measure. Regular flu has a fatality rate of just less than 0.1%, which means that Covid-19 seems vastly more deadly.
This, and its infection rate, is what has kicked the world into action and governments into issuing lockdowns.
The problem, as many have pointed out, is that, without accurate testing, the infection rate is skewed, and so the fatality rate is skewed. The whole thing depends on knowing how many people have contracted the disease, including asymptomatic and mild sufferers. Because people are only being tested when they are pretty ill, it has been hard to get a proper sense of how many people have actually got this disease. Therefore, working out fatality rates is difficult, and the real rates are potentially wildly wrong, since we don’t know how many asymptomatic or mild sufferers there are.
Fear of Covid-19 is based on its high estimated case fatality rate—2% to 4% of people with confirmed Covid-19 have died, according to the World Health Organization and others. So if 100 million Americans ultimately get the disease, two million to four million could die. We believe that estimate is deeply flawed. The true fatality rate is the portion of those infected who die, not the deaths from identified positive cases. The latter rate is misleading because of selection bias in testing. The degree of bias is uncertain because available data are limited. But it could make the difference between an epidemic that kills 20,000 and one that kills two million. If the number of actual infections is much larger than the number of cases—orders of magnitude larger—then the true fatality rate is much lower as well. That’s not only plausible but likely based on what we know so far.
Population samples from China, Italy, Iceland and the U.S. provide relevant evidence. On or around Jan. 31, countries sent planes to evacuate citizens from Wuhan, China. When those planes landed, the passengers were tested for Covid-19 and quarantined. After 14 days, the percentage who tested positive was 0.9%. If this was the prevalence in the greater Wuhan area on Jan. 31, then, with a population of about 20 million, greater Wuhan had 178,000 infections, about 30-fold more than the number of reported cases. The fatality rate, then, would be at least 10fold lower than estimates based on reported cases.
Next, the northeastern Italian town of Vò, near the provincial capital of Padua. On March 6, all 3,300 people of Vò were tested, and 90 were positive, a prevalence of 2.7%. Applying that prevalence to the whole province (population 955,000), which had 198 reported cases, suggests there were actually 26,000 infections at that time. That’s more than 130-fold the number of actual reported cases. Since Italy’s case fatality rate of 8% is estimated using the confirmed cases, the real fatality rate could in fact be closer to 0.06%.
In Iceland, deCode Genetics is working with the government to perform widespread testing. In a sample of nearly 2,000 entirely asymptomatic people, researchers estimated disease prevalence of just over 1%. Iceland’s first case was reported on Feb. 28, weeks behind the U.S. It’s plausible that the proportion of the U.S. population that has been infected is double, triple or even 10 times as high as the estimates from Iceland. That also implies a dramatically lower fatality rate.
The best (albeit very weak) evidence in the U.S. comes from the National Basketball Association. Between March 11 and 19, a substantial number of NBA players and teams received testing. By March 19, 10 out of 450 rostered players were positive. Since not everyone was tested, that represents a lower bound on the prevalence of 2.2%. The NBA isn’t a representative population, and contact among players might have facilitated transmission. But if we extend that lower-bound assumption to cities with NBA teams (population 45 million), we get at least 990,000 infections in the U.S. The number of cases reported on March 19 in the U.S. was 13,677, more than 72-fold lower. These numbers imply a fatality rate from Covid-19 orders of magnitude smaller than it appears.
As the article continues, this does not mean it is a nonissue because healthcare systems are tangibly being swamped and people are dying, but is the modelling for these deaths accurate?
This video is very interesting, getting onto the overestimation towards the end:
True Case Fatality Rates
This is the difference between (confirmed) Case Fatality Rate (CFR) and the equivocation that is Fatality Rate (tCFR being a true case fatality rate). As the video explains, there is a mitigation of the much lower rate when considering lag times between catching and dying, so you need to match up when people die to when they caught it for accurate statistical purposes, but this will still not be nearly enough to compensate for the difference between the selection biased case fatality rate and an overall fatality rate. The difference, it suggests, could be as much as 19 times.
That said, a draft paper by Alexander Lachmann of the Icahn School of Medicine tries to model the true number of coronavirus infections in China and finds the true number 9 times higher, finding the true fatality rate in China to be 0.46% (this depends on where and at what point).
But a different model published in Nature Medicine concludes that the symptomatic case fatality rate in Wuhan was actually 1.4%.
The true stats will only be available after conducting far-reaching antibody studies.
This is somewhat taken into account in this informative video, discussing the lag issue as well:
It might be worth taking a look at what contributors over at Watts Up With That (a global warming denial site) have concluded. Looking at a closed (albeit small) group – the passengers on the Diamond Princess – and adjusting for all sorts (and these could be contested), they conclude the tCFR is 0.34%. I would add a caveat that it is still too soon to get an accurate appraisal on deaths; put simply, people are still possibly going to, or even likely, die in that cohort. That would put an absolute worst-case scenario, if this analysis is remotely accurate, of Covid-19 being over three times as fatal than the flu.
This article looks both at lag times causing underestimation and true case numbers causing overestimation of fatality rates. Concerning work on the cruise liner, it states:
A draft paper by a team of researchers in the UK (which is not yet peer-reviewed) attempts to use the cruise ship’s death and infection data to estimate China’s true fatality rate, accounting for undiagnosed cases. Using their model, they estimate China’s real coronavirus fatality rate was about 0.5%, with a range from 0.2–1.2%.
But the cruise ship data is also severely limited. The quarantine was ended after two weeks because people on board kept falling ill, and follow-up data is not available for many patients.
Moreover, the ship saw just six total deaths during the quarantine, which is about 1% of known cases. But it’s hard to draw conclusions from a small sample size like that. In fact, the data in the UK draft study was taken on March 3; four days later, a seventh former passenger died from COVID-19. (Johns Hopkins’ database now also lists an eighth death.)
This throws all of the models based on the ship’s earlier fatality rates into question.
This would call into question the validity of the Watts Up With That piece.
Here’s the rub
These sorts of problems exist for the flu, too. Estimations for infection numbers in the US are that somewhere between 3% and 11% of the population get it. But this is only for symptomatic infection. In other words, it seems to me that the claim that influenza has a ~0.1% fatality rate wouldn’t take into account asymptomatic carriers, the point that the WSJ is making above for coronavirus. Indeed, one flu metastudy concludes:
Influenza infection manifests in a wide spectrum of severity, including symptomless pathogen carriers. We conducted a systematic review and meta-analysis of 55 studies to elucidate the proportional representation of these asymptomatic infected persons. We observed extensive heterogeneity among these studies. The prevalence of asymptomatic carriage (total absence of symptoms) ranged from 5.2% to 35.5% and subclinical cases (illness that did not meet the criteria for acute respiratory or influenza-like illness) from 25.4% to 61.8%.
These ranges are pretty substantial. Therefore, in a case of being good enough for the goose and thus the gander, you would need to apply the same criteria for claiming the coronavirus stats are overestimated to the flu stats as well. There are vastly more people who get the flu than we record and this will skew the fatality rates, meaning they are, too, overestimated.
I’m not sure what the answer is, and I’m certain my astute readers will help out here, but I can guarantee that the overall fatality rates of both Covid-19 and the flu are overestimated. The question, for me, is whether the overestimation for Covid-19 is in line with the overestimation for flu, or more, or less. In other words, is the true fatality rate actually that much worse than flu or not?
Much of the answer will be “it depends on the context”: where, when, what healthcare system and how is the system coping?
There are, as the WSJ admits, very tangible problems facing healthcare services around the world, and people are dying. This is important, because as Tomas Pueyo’s piece, on which the last video is based, states:
This is what you can conclude:
- Excluding these, countries that are prepared will see a fatality rate of ~0.5% (South Korea) to 0.9% (rest of China).
- Countries that are overwhelmed will have a fatality rate between ~3%-5%
Put in another way: Countries that act fast can reduce the number of deaths by a factor of ten. And that’s just counting the fatality rate. Acting fast also drastically reduces the cases, making this even more of a no-brainer.
There is a huge variable involved in the fatality rate – the state of the healthcare system in question. In other words, it is tangibly the case that huge amounts of people will die if we don’t flatten the curve to allow for greater hospital capacity. Pueyo accentuates the importance of acting hard and acting quickly, something that the US is not doing:
In this theoretical model that resembles loosely Hubei, waiting one more day creates 40% more cases! So, maybe, if the Hubei authorities had declared the lockdown on 1/22 instead of 1/23, they might have reduced the number of cases by a staggering 20k.
And remember, these are just cases. Mortality would be much higher, because not only would there be directly 40% more deaths. There would also be a much higher collapse of the healthcare system, leading to a mortality rate up to 10x higher as we saw before. So a one-day difference in social distancing measures can end exploding the number of deaths in your community by multiplying more cases and higher fatality rate.
This is an exponential threat. Every day counts. When you’re delaying by a single day a decision, you’re not contributing to a few cases maybe. There are probably hundreds or thousands of cases in your community already. Every day that there isn’t social distancing, these cases grow exponentially.
My conclusion is that Covid-19 is certainly more fatal than flu, but to what degree is uncertain. It is more infectious and more fatal and our bodies are not used to it and we have no vaccine. Does the reaction warrant the number of deaths it will save? Time will tell, but what is certain is that if healthcare systems get overwhelmed, lots of people die, and not just from coronavirus, but from all manner of other things not effectively treated by swamped hospitals. Do we look at the fatality stats and say, “well, they’re overstated” and stand by and do nothing, or do we go to the other extreme and shut down economies and societies to flatten the curve, erring on the side of caution?
I guess you could ask this to a doctor in Cremona hospital. The simple fact is that, so far, some 50 doctors in Italy have died of coronavirus. I can’t imagine that this is comparable to flu.
My next post will concern the cost of lives: how much is a life worth? How much is a government willing to pay to save them?
Stay in touch! Like A Tippling Philosopher on Facebook: