Corona CFR Creeps Up To 2% in South Korea

The novel coronavirus is a long disease. That’s one of the things that makes it so problematic. Apart from having being at least 10x as lethal as the standard flu, and people having no herd immunity against it (so potentially up to 5x as many infectees as during a typical flu season), the average hospital stay lasts several times as long as for flu as well. Multiplied together, the cumulative strain on healthcare systems goes up by two orders of magnitude.

The mortality profile is distinct over time. It is concentrated in the first few days, as the really hopeless cases die off quick. Then recovery improves. But after 3-4 weeks, it begins to go down again. I suppose there’s only so much a body can withstand, such as fighting off a virus while hooked up to a ventilator. We are beginning to see a demonstration of that in South Korea.

As Robert Rohde points out, the CFR in South Korea – which started off really small, thanks to its extensive testing regimen (almost all cases were caught) and excellent healthcare system – has been creeping upwards as of the past week:

A couple of weeks ago, people were excited that South Korea was only showing a 0.6% case-fatality rate.

Today that rate has risen to 1.4%, and the progression analysis suggests that it could still rise over 2%.

South Korea has the second largest number of hospital beds of any OECD country after Japan, and ahead of Germany in third place. 9,478 cases largely concentrated in Daegu was not a large strain on its system and already looks decidedly modest relative to what’s unfolding in Europe and the US. We need to start assuming a CFR of ~2% – not 1% – as the base case scenario in which healthcare systems hold up.


  1. Please keep off topic posts to the current Open Thread.

    If you are new to my work, start here.

  2. for-the-record says

    Korea has tested (as of 2 days ago) 365,000 people out of a population of 51.5 million. As of today it has around 9,500 cases of whom a maximum of 203 may die (144 already dead +59 critical/serious), which does indeed give 2% CFR.

    Korea has a large number of asymptomatic cases (“significantly higher” than other countries, according to Jeong Eun-kyeong, director of Korea Centers for Disease Control and Prevention, presumably due to its more extensive testing). Yet it has currently tested less than 0.8% of the population. Surely if the testing were extended to the entire population, tens of thousands, if not hundreds of thousands, of asymptomatic cases would be identified, i.e. people who are infected but are not going to die from Covid-19.

    So let’s make some relatively conservative assumptions. According to this article from Deutsche Welle, “Up to 30% of coronavirus cases [are] asymptomatic”:

    So let’s assume that the asymptomatic rate in Korea is 25% (definitely conservative, since Korea is supposed to be “significantly higher” than other countries). So of the 9,500 postive cases so far, at least 2,300 are asymptomatic. If there had been 365,000 tests as of 2 days ago, there have presumably been around 400,000 by now. So 2,300 asymptomatics out of 400,000 tests is a bit less than 0.6%, again to be on the conservative side we will call it 0.5%.

    We should also note that this asymptomatic rate is likely a considerable underestimate, since many who were tested were presenting symptoms (this is why many sought out testing in the first place). Suppose for example that of the 400,000 tests, 100,000 were of people feeling unwell and the other 300,000 were “random”. Then the asymptomatic rate would be 2,300 out of 300,000, or nearly 0.8%.

    But sticking with a “conservative” 0.5% asymptomatic ratio, this implies that there must be more than 250,000 asymptomatics in the overall population!

    So we could then estimate the total number of infections to date as 9,500 (reported) + 250,000 (untested asymptomatics), giving us (a minimum of) around 260,000 people to date in Korea who have thus far been infected. Of these 260,000 infected to date, the maximum who will die are 203.

    203/260,000 is less than 0.1% , i.e. no more (and in fact a bit less) than ordinary flu.

  3. Unsettling but hardly surprising, there’s been data coming out of China for several weeks now suggesting that once someone needs a ventilator, their prognosis is not good. Even in mid January, the recovery rate suggested this.

  4. reiner Tor says

    Surely if the testing were extended to the entire population, tens of thousands, if not hundreds of thousands, of asymptomatic cases would be identified, i.e. people who are infected but are not going to die from Covid-19.

    A small percentage of those would then develop pneumonia, and would be hospitalized and tested. Then we would keep hearing about mysterious new outbreaks in other cities. So why is it not happening? Why are all of the known pneumonia cases connected to the known infections, and geographically concentrated in just a few places?

    You just didn’t address this point, already made in earlier threads.

    Suppose for example that of the 400,000 tests, 100,000 were of people feeling unwell and the other 300,000 were “random”.

    Also each contact (based on cellphone data) of anyone testing positive was tested (otherwise how to stop such an epidemic spread by asymptomatic virus carriers?), which would presumably catch each one of the asymptomatic cases. Or else we’d see unexplained pockets of pneumonia outbreaks elsewhere. Or do you assume that those asymptomatic carriers are essentially spreading a different strain of the virus, one which only gives immunity but no illness?

  5. for-the-record says

    I can see your point. However, the only way one could really “test” (and hence disprove) the hypothesis of 0.5% asymptomatic in the entire population is to do random testing of people not showing symptoms to see what percentage, if any, are asymptomatic. If you are right, then this percentage should be only a (very) small fraction of 0.5%. Perhaps such testing has been done in Korea, in which case someone can provide us with the results.

  6. reiner Tor says

    Well, the simple consequence of your theory would be several unexplained cases of pneumonia, several clusters of unexplained such cases. This in itself makes your theory untenable, unless you can find a good explanation.

  7. for-the-record says

    This in itself makes your theory untenable, unless you can find a good explanation

    Let’s try this:

    1. From my original comment, there has in fact already been a “random” survey carried out — namely among those not presenting symptoms who were tested, whose exact number we do not know but were surely the overwhelming majority of the 400,000 tested to date. So for the sake of argument let’s assume 300,000-350,000 “random” tests, of which we have 2,300 asymptomatic cases, or around 0.7%. This would translate to 350,000 on a national level. But as you point out, Covid-19 is not uniformly present throughout the country, so this could well give an overestimate. But looking at the map of the infections

    I would hazard a guess that at least 60% of the population (30 million) lives in infected zones, so that we could reduce our estimate of asymptomatics from 350,000 to 210,000.

    1. As you rightly point out, I was implicitly assuming that none of these would ever develop symptoms (and hence be at risk of death), which is clearly not the case. Some of them, albeit a very small percentage, will eventually develop symptoms. This is presumably what we are currently seeing, as new cases are being reported at the rate of c.150 per day.
    2. So how many of these 210,000 current asymptomatics might we expect to develop symptoms, and what percentage of these will eventually die?

    3. Answering the second question first, of the 7,200 (presumed) current symptomatic cases (including deaths), 203 have died or are at risk of doing so, hence we might expect (a maximum of) 3 percent of those developing symptoms to die.

    4. So returning to the first question, how many of the estimated asymptomatics are likely to come down with symptoms serious enough to be classified as cases of Covid-19? Based on the current numbers, my guess is that this is no more than 2% (by comparison, approximately 1% of flu cases in the US require hospitalization). This would imply that of the 210,000 current asymptomatics, 4,200 would come down with symptoms of whom (at most) 126 would die.

    5. So of the c. 220,000 current cases in Korea (those with symptoms, diagnosed asymptomatics, and estimated undiagnosed asymptomatics) we would expect total deaths to be no more than the current 203 + 126 = 339, or 0.15%.

    Obviously my conclusion would be altered if a significantly higher percentage (than 2%) of asymptomatics develop symptoms. So if you have evidence that a higher percentage of asymptomatics are likely to eventually develop symptoms, please let me know.

  8. You seem to be making formally correct argument but I suspect that sampling (testing) was not done randomly over 51.5 mil population but rather within the circles of social/geographic vicinity of symptomatic cases, so then your extrapolation would not be valid unless you would propose that the ratio of symptomatic/asymptomatic cases varies in different regions which, I do not think, you could argue for.

  9. for-the-record says

    I have at least partially addressed that point in my later comment, in which I reduced the effective (or affected) population to 30 million.

  10. sudden death says

    Obviously my conclusion would be altered if a significantly higher percentage (than 2%) of asymptomatics develop symptoms. So if you have evidence that a higher percentage of asymptomatics are likely to eventually develop symptoms, please let me know.

    77% of asymptomatic patients developed symptoms one week after:

    One week after testing, the 13 residents who had positive test results and were asymptomatic on the date of testing were reassessed; 10 had developed symptoms and were recategorized as presymptomatic at the time of testing (Table 2). The most common signs and symptoms that developed were fever (eight residents), malaise (six), and cough (five). The mean interval from testing to symptom onset in the presymptomatic residents was 3 days. Three residents with positive test results remained asymptomatic.

  11. reiner Tor says

    I guess a more refined map would further reduce the affected population.

    The new cases are always followed up by finding out the source of infection, and it’s always, invariably found. Moreover, they test each contact of the patient, to find out who he infected – or if the source of infection is not immediately determined, to find out where he was infected. How do you think this kind of highly refined epidemic control is done? It’s not just testing the most severe cases in the hospitals, which only happens after the healthcare system has been overwhelmed, like in Lombardy, or in Wuhan in January, and ideally the healthcare system should try to get out of this state, as it did in Wuhan. (Wuhan had the big advantage of receiving enormous resources from the rest of China, a country of 1.5 billion people, obviously not really replicable in Italy.)

  12. While the differences between CFR% in various regions/countries has been commented on in various places (though IMO not convincingly explained), virtually everywhere the CFR% has increased over time. This does not appear to be just because deaths at any given point are of cases reported (say) 10 days on average earlier. (So for example deaths will keep being reported for a couple of weeks or more after new cases drop to 0 – as in Hubei for example – simply because some of those in hospital and already counted will eventually die).

    Even German CFR jumped from 0.3% to 0.6%.

  13. showmethereal says

    Very true…. It would have to be random testing to get the most accurate picture… Though I doubt it would change that much. The increase in rate in South Korea is not surprising.

  14. Another way to get the CFR is with the following calculation:

    (Total Deaths) divided by (Total Deaths + Total Recovered), which is:
    (152) divided by (152 + 5033) = 0.0302 (in other words, 3.0%, using data as of 28-Mar-2020)

    The above calculation is using the resolved cases. If a person has died or recovered, then the case has been resolved. The problem with using total cases is that we don’t what the resolution will be for active cases. But here is the calculation with total cases:

    (Total Deaths) divided by (Total Cases), which is (as of 28-Mar-2020):
    (152) divided by (9583) = 0.01586 (in other words, 1.6%)

    Of course, these are just the cases we know about, because they got tested. In South Korea, they have an aggressive testing policy. But let’s say that roughly half of infected people are aymptomatic or with only minor symptoms, and so do not bother to get tested. Well, that would cut the CFR in half to something like 1.5% (down from 3.0%).

    In any case, even if it’s only 1%, this is pretty bad, especially since I’m sure South Koreans are healthier than we Americans. Also, because of the high cost of health care, Americans are less likely to want to go to the hospital. So, the CFR could be much higher in the U.S.

    Of course, any predictions could quickly be invalidated if the Hydroxychloroquine / Arithromycin works. But assuming that nothing helps much, my guess is 3% CFR rate in the U.S. If only 150 million Americans are infected this year, this would be 4.5 million deaths. In a typical year, about 2.7 million Americans die. So, instead of 2.7 million dying, we would have 7.2 million dying.

  15. Korea is the country where people drop dead from never leaving their chair for days during online gaming sessions.

    It has implemented a people isolate themselves at home policy, though I’m unsure they needed to because of the above!

    The newspapers seem to report that it is actually contact tracing that has kept things under control. This seems incredibly unlikely given the very many false negative tests and assymptomatic cases there (uncontroversially) are. The newspapers may be more motivated by embarrassing their domestic political opponents than getting to the truth.


    The post above was by me not “Jiminy”, a poster I hadn’t even noticed previously. I suppose it did so because their details autopopulated into my browser. How? I have no idea. I do now have their email address though but I won’t do anything weird with it, of course.

    Apologies to Jiminy for not noticing before posting! I’ve got all the fun symptoms of Chinavirus if that’s any excuse!

  17. This happens sometimes. I once posted a comment that got credited to Ron Unz(!) No idea how that happened.

  18. LondonBob says

    Here in Britain three doctor deaths, one a GP and two hospital, all ethnic minority immigrants. Ethnic minorities are disproportionately being hit, so is it an urban thing except they don’t ski, is it family structure, is it unhealthy upbringing, underlying health issues or something like dormant TB?

  19. LondonBob says

    Folks need to understand the difference between CFR and IFR, IFR includes asymptomatic and those unrecorded.

  20. A novel coronavirus is not all that novel in South Korea , 8 people died killed in an outbreak of MERS-coronavirus (AKA Camel Flu) during 2015 , and there was a case in 2018.
    Due to the clinical similarity between MERS-CoV and SARS-CoV, it was proposed that they may use the same cellular receptor; the exopeptidase, angiotensin converting enzyme 2 (ACE2).[28] However, it was later discovered that neutralization of ACE2 by recombinant antibodies does not prevent MERS-CoV infection.[29] Further research identified dipeptidyl peptidase 4 (DPP4; also known as CD26) as a functional cellular receptor for MERS-CoV

    Dr. Anthony S. Fauci of the National Institutes of Health in Bethesda, Maryland, stated that as of now MERS-CoV “does not spread in a sustained person to person way at all.” Dr. Fauci stated that there is potential danger in that it is possible for the virus to mutate into a strain that does transmit from person to person.[33]

    China said on 7 Jan that the virus could not be transmitted between people. The stupid bastards assumed it was like Camel Flu.

  21. reiner Tor says

    This seems incredibly unlikely given the very many false negative tests and assymptomatic cases there (uncontroversially) are.

    It doesn’t seem unlikely at all. Tests are not perfect (nothing is in this world), but it’s enough to reduce R0 to below 1.0 to arrest the spread of the epidemic.

  22. sudden death says

    Taiwan, the best performing and containing country in the world from the start, at the moment has 1,6 % CFR, so everybody else will be much higher.

  23. South Korea was forewarned with MERS in 2015 (18 dead) plus a case in 2018, Taiwan was forewarned with SARS is 2003 (37 deaths). With COVID-19, Taiwan has already racked up 3 deaths with under 300 confirmed cases of infection. Far from impressive.

  24. sudden death says

    And which countries are close to impressive then? If nearly 24 million population Taiwan, which managed to keep case numbers in low hundreds without full shutdowns&guarantines is far from that?

  25. Someone’s shifting feet; the comment I made was a rejoinder to your pronouncement about the case-fatality ratio in Taiwan. Thailand seems to have relatively little COVID-19 so far, but the King has fled to Germany.