3 ms·

Assuming that someone who died from coronavirus is far more likely to have been tested for the virus than someone who's been infected by it, the true death rate

by httpz 7y ago

Assuming that someone who died from coronavirus is far more likely to have been tested for the virus than someone who's been infected by it, the true death rate of someone infected by the virus is probably far lower than the calculated death rate. (# of confirmed death / # of confirmed infections). S.Korea did extensive testing, which reduces the number of unreported infections. So the calculated death rate in S. Korea is 4x lower than a comparable country like Spain, which has similar # of confirmed cases, population, GDP, and GDP per capita. If we assume S. Korea's calculated death rate(<1%) is closer to the true death rate and apply it to other countries to derive the total # of infections (reported + unreported) based on the reported death count, we can see there may be far more infections than we know.

cycrutchfield 7y ago
This is not really valid due to fatality rate being heavily dependent on age, comorbidities, and healthcare quality (which is itself dependent on case load), and the distributions of these differ greatly across countries. But, sure, you could try to normalize these to infer the actual case load.
- httpz 7y ago
  There is a lot of assuming going on so this doesn't actually prove anything but I think it's something worth thinking about. Is Spain and S. Korea really that much different to have a 4x difference in death rate?
- danielheath 7y ago
  Right - eg when the ICU is full and refusing new patients, the death rate is going to be higher.
- TheOtherHobbes 7y ago
  It would be useful if this could be broken down into separate categories - estimated CFR without medical intervention, and likely CFR at various levels of intervention. The former suggests what your population mortality rate will be if your healthcare system is overwhelmed, and could possibly be derived from Italian data. The latter gives you a resources vs effectiveness sweet spot that maximises health system throughput, so you can save the maximum number of lives over time given the resources you have (or can build/find.) The age profile is relevant, but worryingly there seem to be a number of reports that it's not just the over-65s who are at risk. I haven't seen anything more recent about this than the Chinese estimates from a month or so ago.
  - wahern 7y ago
    CFR can't be used to determine population mortality rate. It in no way reflects r0, and only tangentially at best hints at percentage of asymptomatic infections (and that's recent as there was never drive-by testing for any previous epidemic). Those two dimensions add unfathomable complexity to the equation. In many ways it's better to leave those dimensions out to improve the utility of CFR. A doctor doing triage or health department coming up with orders wants a simple, basic number, not a panoply of options that require data they won't have at hand. We by all means should collect more data, just don't shoehorn it into such a very primitive yet very essential statistic. Give them new names and new, more appropriate semantics. I suspect that after this is over they're going to tighten the criteria around CFR, and in particular exclude by definition asymptomatic cases and possibly non-hospitalized cases. They never had to do that before as "cases" usually implied someone sick enough to be given a diagnosis, which in turn implied someone at least moderately sick--e.g. actual or suspected case of pneumonia.
- scythe 7y ago
  It seems like the biggest confounder is age. If we stratify across age, we might be forgiven for ignoring differences in healthcare quality. The death rates by age in Korea are given here: https://www.businessinsider.com/coronavirus-death-rates-by-age-south-korea https://www.businessinsider.com/coronavirus-death-rates-by-a... 20-29: 0% 30-39: 0.1% 40-49: 0.1% 50-59: 0.4% 60-69: 1.5% 70-79: 4.3% 80+: 7.2%
  - anewguy9000 7y ago
    honest question: what is the death rate for an average person in their 80s? ie: if you have cov2 you have a 7% chance of dying from it but dont you already have say a 50% chance of dying in your 80s anyway? to be strictly logical and not emotional about this moral question: are we unhinging society (effects we cannot begin to calculate) to keep the fatality rate for the population 80+ at 50% instead of 57%? in terms of costs to the healthcare system, might this equate to a net relief if the peak were accelerated instead of flattened?
    - abtinf 7y ago
      Here are the death probabilities for males in their 80s based on the 2016 social security life table: 80 5.8% 81 6.4% 82 7.1% 83 7.8% 84 8.7% 85 9.7% 86 10.7% 87 11.9% 88 13.3% 89 14.8% https://www.ssa.gov/oact/STATS/table4c6.html https://www.ssa.gov/oact/STATS/table4c6.html
    - Majromax 7y ago
      > honest question: what is the death rate for an average person in their 80s? That's a standard actuarial problem, answered with life tables (https://www.cdc.gov/nchs/data/nvsr/nvsr61/nvsr61_03.pdf https://www.cdc.gov/nchs/data/nvsr/nvsr61/nvsr61_03.pdf -- pdf, US 2008). In said life tables, there is a 7% death rate between ages 83 and 84. To an order-of-magnitude approximation, infection with the novel coronavirus seems to impose roughly one year's worth of mortality risk on the victim.
      - nostromo 7y ago
- 465023485544 7y ago
  Don't you first of all have to offset the infection rate by the time-to-death? I've been puzzling over all these extremely bad estimates we hear, and the only reason for it that I can come up with is that nobody wants to point at the much darker truth. One study I've seen pinned the median time-to-death after symptom onset at 18 days. Even in Korea you won't get less than a 2-3% CFR with that estimation. edit: I don't know anything about statistics by the way, please tell me that I'm wrong.
  - cycrutchfield 7y ago
    Yes, absolutely. But that's why we should use case fatality rate and not just divide #deaths/#infected
nopinsight 7y ago
South Korea's CFR of <1.0% is a very optimistic figure. It is a naive calculation based on current # deaths / # infections. 1) But # infections were growing exponentially, we need to use numbers from the same cohort, which implies much lower # infections & higher CFR. Naive CFR will go up once infections grow more slowly. (It is already going up over time.). 2) South Korean confirmed cases are much younger than their median age, mainly between 20-29 years old (perhaps because of where superspreading events happen—that church). This age group has a much lower fatality rate from Covid-19. 3) # hospital beds per capita in South Korea is second highest among OECD countries (1st is Japan) and ~4 times that of the US. They already have patients waiting for beds. Most countries will do much worse if they reach the same # infections per capita, since fatality rate surges to multiple times as high without proper care (as can be seen in several places around the world now, unfortunately). (SK has ~12 beds/1000 capita; China ~4.3; US ~2.8)
- nopinsight 7y ago
  Also, a very recent Lancet report calculates global CFR at 5.7%. CFR (Case Fatality Rate) is based on the # people who were tested. IFR (Infection Fatality Rate) which includes people with no or mild symptoms will be lower. "On this basis, using WHO data on the cumulative number of deaths to March 1, 2020, mortality rates would be 5·6% (95% CI 5·4–5·8) for China and 15·2% (12·5–17·9) outside of China. Global mortality rates over time using a 14-day delay estimate are shown in the figure, with a curve that levels off to a rate of 5·7% (5·5–5·9)" https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(20)30195-X/fulltext https://www.thelancet.com/journals/laninf/article/PIIS1473-3...

by httpz 7y ago

cycrutchfield 7y ago

This is not really valid due to fatality rate being heavily dependent on age, comorbidities, and healthcare quality (which is itself dependent on case load), and the distributions of these differ greatly across countries. But, sure, you could try to normalize these to infer the actual case load.

httpz 7y ago
There is a lot of assuming going on so this doesn't actually prove anything but I think it's something worth thinking about. Is Spain and S. Korea really that much different to have a 4x difference in death rate?
danielheath 7y ago
Right - eg when the ICU is full and refusing new patients, the death rate is going to be higher.
TheOtherHobbes 7y ago
It would be useful if this could be broken down into separate categories - estimated CFR without medical intervention, and likely CFR at various levels of intervention. The former suggests what your population mortality rate will be if your healthcare system is overwhelmed, and could possibly be derived from Italian data. The latter gives you a resources vs effectiveness sweet spot that maximises health system throughput, so you can save the maximum number of lives over time given the resources you have (or can build/find.) The age profile is relevant, but worryingly there seem to be a number of reports that it's not just the over-65s who are at risk. I haven't seen anything more recent about this than the Chinese estimates from a month or so ago.
- wahern 7y ago
  CFR can't be used to determine population mortality rate. It in no way reflects r0, and only tangentially at best hints at percentage of asymptomatic infections (and that's recent as there was never drive-by testing for any previous epidemic). Those two dimensions add unfathomable complexity to the equation. In many ways it's better to leave those dimensions out to improve the utility of CFR. A doctor doing triage or health department coming up with orders wants a simple, basic number, not a panoply of options that require data they won't have at hand. We by all means should collect more data, just don't shoehorn it into such a very primitive yet very essential statistic. Give them new names and new, more appropriate semantics. I suspect that after this is over they're going to tighten the criteria around CFR, and in particular exclude by definition asymptomatic cases and possibly non-hospitalized cases. They never had to do that before as "cases" usually implied someone sick enough to be given a diagnosis, which in turn implied someone at least moderately sick--e.g. actual or suspected case of pneumonia.
scythe 7y ago
It seems like the biggest confounder is age. If we stratify across age, we might be forgiven for ignoring differences in healthcare quality. The death rates by age in Korea are given here: https://www.businessinsider.com/coronavirus-death-rates-by-age-south-korea https://www.businessinsider.com/coronavirus-death-rates-by-a... 20-29: 0% 30-39: 0.1% 40-49: 0.1% 50-59: 0.4% 60-69: 1.5% 70-79: 4.3% 80+: 7.2%
- anewguy9000 7y ago
  honest question: what is the death rate for an average person in their 80s? ie: if you have cov2 you have a 7% chance of dying from it but dont you already have say a 50% chance of dying in your 80s anyway? to be strictly logical and not emotional about this moral question: are we unhinging society (effects we cannot begin to calculate) to keep the fatality rate for the population 80+ at 50% instead of 57%? in terms of costs to the healthcare system, might this equate to a net relief if the peak were accelerated instead of flattened?
  - abtinf 7y ago
    Here are the death probabilities for males in their 80s based on the 2016 social security life table: 80 5.8% 81 6.4% 82 7.1% 83 7.8% 84 8.7% 85 9.7% 86 10.7% 87 11.9% 88 13.3% 89 14.8% https://www.ssa.gov/oact/STATS/table4c6.html https://www.ssa.gov/oact/STATS/table4c6.html
  - Majromax 7y ago
    > honest question: what is the death rate for an average person in their 80s? That's a standard actuarial problem, answered with life tables (https://www.cdc.gov/nchs/data/nvsr/nvsr61/nvsr61_03.pdf https://www.cdc.gov/nchs/data/nvsr/nvsr61/nvsr61_03.pdf -- pdf, US 2008). In said life tables, there is a 7% death rate between ages 83 and 84. To an order-of-magnitude approximation, infection with the novel coronavirus seems to impose roughly one year's worth of mortality risk on the victim.
    - nostromo 7y ago
465023485544 7y ago
Don't you first of all have to offset the infection rate by the time-to-death? I've been puzzling over all these extremely bad estimates we hear, and the only reason for it that I can come up with is that nobody wants to point at the much darker truth. One study I've seen pinned the median time-to-death after symptom onset at 18 days. Even in Korea you won't get less than a 2-3% CFR with that estimation. edit: I don't know anything about statistics by the way, please tell me that I'm wrong.
- cycrutchfield 7y ago
  Yes, absolutely. But that's why we should use case fatality rate and not just divide #deaths/#infected

nopinsight 7y ago

South Korea's CFR of <1.0% is a very optimistic figure. It is a naive calculation based on current # deaths / # infections. 1) But # infections were growing exponentially, we need to use numbers from the same cohort, which implies much lower # infections & higher CFR. Naive CFR will go up once infections grow more slowly. (It is already going up over time.). 2) South Korean confirmed cases are much younger than their median age, mainly between 20-29 years old (perhaps because of where superspreading events happen—that church). This age group has a much lower fatality rate from Covid-19. 3) # hospital beds per capita in South Korea is second highest among OECD countries (1st is Japan) and ~4 times that of the US. They already have patients waiting for beds. Most countries will do much worse if they reach the same # infections per capita, since fatality rate surges to multiple times as high without proper care (as can be seen in several places around the world now, unfortunately). (SK has ~12 beds/1000 capita; China ~4.3; US ~2.8)

nopinsight 7y ago
Also, a very recent Lancet report calculates global CFR at 5.7%. CFR (Case Fatality Rate) is based on the # people who were tested. IFR (Infection Fatality Rate) which includes people with no or mild symptoms will be lower. "On this basis, using WHO data on the cumulative number of deaths to March 1, 2020, mortality rates would be 5·6% (95% CI 5·4–5·8) for China and 15·2% (12·5–17·9) outside of China. Global mortality rates over time using a 14-day delay estimate are shown in the figure, with a curve that levels off to a rate of 5·7% (5·5–5·9)" https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(20)30195-X/fulltext https://www.thelancet.com/journals/laninf/article/PIIS1473-3...