Infection Fatality Rate (IFR)

Covid-19 discussion, bring your own statistics
PeteB
Stargoon
Posts: 141
Joined: Wed Nov 13, 2019 1:02 pm

Infection Fatality Rate (IFR)

Post by PeteB » Fri Nov 20, 2020 11:32 am

After seeing jimbob's response to Toby Young's hilarious misestimate of IFR I was wondering if we have got a better handle on this yet. I have seen lots of claims that we are getting better at treating Covid-19 in hospitals, but I wondered if that was just that the people in ICU were a younger age, so lower fatality rate (as per heat maps showing Covid-19 infection rates gradually moving up to older age brackets)

I did a back of the envelope calculation based on antibody % and current death rate - I've tried this a few times on different data sets and always get a slightly higher IFR based on antibody testing (possibly because antibodies fall over time)

ANTIBODY

https://www.gov.uk/government/news/larg ... -over-time

6% antibodies in June for England (population 55 million)

Guess at deaths for England in June = 45,000 with Covid on death cert + at least 5000 undiagnosed Covid

IFR = 0.9%


CURRENT DEATH RATE

If I look at the current death rate for England over the last couple of weeks

https://coronavirus.data.gov.uk/details/deaths

By England - death by reported date Average Last 7 Days = 338, previous 7 days 334

If I look at ONS infection survey (which excludes hospitals and care homes) for the period where the deaths would come from

https://www.ons.gov.uk/peoplepopulation ... usReleases

10 to 16 October 2020 - 35,200 new cases per day
17 to 23 October 2020 - 51,900 new cases per day
25 to 31 October 2020 - 45,700 new cases per day

IFR roughly = 336/44,000 = 0.76

Herainestold
Snowbonk
Posts: 436
Joined: Mon Nov 25, 2019 1:23 pm

Re: Infection Fatality Rate (IFR)

Post by Herainestold » Fri Nov 20, 2020 12:35 pm

Most of the literature I have seen indicates something around 0.7%

So your numbers seem to be about right.

User avatar
headshot
Fuzzable
Posts: 354
Joined: Tue Nov 12, 2019 9:40 am

Re: Infection Fatality Rate (IFR)

Post by headshot » Fri Nov 20, 2020 12:35 pm

That’s pretty close to the early reports that came out of China isn’t it?

User avatar
bob sterman
Fuzzable
Posts: 307
Joined: Mon Nov 11, 2019 10:25 pm

Re: Infection Fatality Rate (IFR)

Post by bob sterman » Fri Nov 20, 2020 6:01 pm

Well - we can put a lower bound on the IFR for the sort of population structure you see in Western urban locations.

COVID-19 has already killed about 0.3% of the entire population of New York City - about 25,000 deaths for 8.4 million residents.

So in a city with a population structure like NYC - the IFR cannot be < 0.3%

User avatar
shpalman
After Pie
Posts: 2239
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: Infection Fatality Rate (IFR)

Post by shpalman » Fri Nov 20, 2020 6:43 pm

Official deaths in Lombardy are about 20,000 out of a population of 10 million. The first wave officially killed 16,000-17,000 (depending on when you consider that it ended) but then this says the excess deaths were actually about 24,000 (but not all from covid).

The last time I saw mass serological testing I think it found ~10% of the population had antibodies: link from July.

So it would be in the range 0.3-3%. (The CFR is around 6% but falling).
molto tricky

User avatar
sTeamTraen
Dorkwood
Posts: 1189
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: Infection Fatality Rate (IFR)

Post by sTeamTraen » Fri Nov 20, 2020 7:25 pm

I'm really not certain how useful the IFR is, in general, since as well as the problem of determining what the infection rate is, you also have to deal with the fact that mortality is so variable across age groups. Do we calculate it based on an equal chance of every person in the population becoming infected, as if there was some kind of random assignment process, or do we take into account that different segments of the population spend more time indoors and in close contact with others? Statistics that are (in effect) based on arithmetic means don't work well when the numerator is so variable.

So I think that as a minimum, for something like this to be useful other than for intellectual curiosity, you would need half a dozen or more IFR bands, one for each age group.
Something something hammer something something nail

Herainestold
Snowbonk
Posts: 436
Joined: Mon Nov 25, 2019 1:23 pm

Re: Infection Fatality Rate (IFR)

Post by Herainestold » Fri Nov 20, 2020 10:36 pm

sTeamTraen wrote:
Fri Nov 20, 2020 7:25 pm
I'm really not certain how useful the IFR is, in general, since as well as the problem of determining what the infection rate is, you also have to deal with the fact that mortality is so variable across age groups. Do we calculate it based on an equal chance of every person in the population becoming infected, as if there was some kind of random assignment process, or do we take into account that different segments of the population spend more time indoors and in close contact with others? Statistics that are (in effect) based on arithmetic means don't work well when the numerator is so variable.

So I think that as a minimum, for something like this to be useful other than for intellectual curiosity, you would need half a dozen or more IFR bands, one for each age group.
Point.

I have seen quoted everything from <0.1% (lockdown skeptics) to 2-3% . It seems like a society with a higher median age should have a higher IFR than a young healthy society,..but it is so hard to get a realistic number for those infected. I dont know whose numbers to believe. Or does it matter?

User avatar
jdc
Hilda Ogden
Posts: 608
Joined: Wed Sep 25, 2019 4:31 pm
Location: Your Mum
Contact:

Re: Infection Fatality Rate (IFR)

Post by jdc » Sat Nov 21, 2020 1:27 am

sTeamTraen wrote:
Fri Nov 20, 2020 7:25 pm
I'm really not certain how useful the IFR is, in general, since as well as the problem of determining what the infection rate is, you also have to deal with the fact that mortality is so variable across age groups. Do we calculate it based on an equal chance of every person in the population becoming infected, as if there was some kind of random assignment process, or do we take into account that different segments of the population spend more time indoors and in close contact with others? Statistics that are (in effect) based on arithmetic means don't work well when the numerator is so variable.

So I think that as a minimum, for something like this to be useful other than for intellectual curiosity, you would need half a dozen or more IFR bands, one for each age group.
This doesn't give the IFR at all (due to the problem of determining what the infection rate is) but it does have that banding by age group using CFR data: https://ourworldindata.org/mortality-ri ... -19-by-age

User avatar
shpalman
After Pie
Posts: 2239
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: Infection Fatality Rate (IFR)

Post by shpalman » Sat Nov 21, 2020 9:04 am

sTeamTraen wrote:
Fri Nov 20, 2020 7:25 pm
I'm really not certain how useful the IFR is, in general, since as well as the problem of determining what the infection rate is, you also have to deal with the fact that mortality is so variable across age groups. Do we calculate it based on an equal chance of every person in the population becoming infected, as if there was some kind of random assignment process, or do we take into account that different segments of the population spend more time indoors and in close contact with others? Statistics that are (in effect) based on arithmetic means don't work well when the numerator is so variable.
It depends what question you are trying to answer. "Do we calculate it based on an equal chance of every person in the population becoming infected" helps answer the question of what are the chances that a given person of a certain age will die, if they contract the virus while "do we take into account that different segments of the population spend more time indoors and in close contact with others" goes towards answering the question what are the chances that a given person of a certain age will die, given that this virus exists.

The whole "some countries have higher death rates because they have older populations" bothers me a bit. Age surely needs to be evaluated in context of the local life expectancy? MD wrote that a person's risk of dying from covid parallels their risk of dying anyway, just compressed into a much shorter time. So in a country in which there are lots of 80+ people with a high risk of mortality, you'll see a high death rate from covid in the 80+ range but not so much in the 60-69 range (because they are generally healthy enough to keep going until 80+). In another country in which there are very few people at 80+ because the mortality rate in the 60-69 range is already very high, you'll see lots of covid deaths in the 60-69 range.

tl;dr to compare different populations, covid mortality shouldn't be banded by how long since a person was born but how long it would have been before they died.
molto tricky

User avatar
bob sterman
Fuzzable
Posts: 307
Joined: Mon Nov 11, 2019 10:25 pm

Re: Infection Fatality Rate (IFR)

Post by bob sterman » Sat Nov 21, 2020 10:23 am

shpalman wrote:
Sat Nov 21, 2020 9:04 am
The whole "some countries have higher death rates because they have older populations" bothers me a bit. Age surely needs to be evaluated in context of the local life expectancy? MD wrote that a person's risk of dying from covid parallels their risk of dying anyway, just compressed into a much shorter time. So in a country in which there are lots of 80+ people with a high risk of mortality, you'll see a high death rate from covid in the 80+ range but not so much in the 60-69 range (because they are generally healthy enough to keep going until 80+). In another country in which there are very few people at 80+ because the mortality rate in the 60-69 range is already very high, you'll see lots of covid deaths in the 60-69 range.
An important thing to bear in mind is that local life expectancy depends on age-specific mortality rates across the entire lifespan - not just the final decades of life.

In populations with low life expectancy - important contributors to this are high rates of extrinsic mortality early in the lifespan (e.g. infectious disease, environmental hazards etc and high infant mortality). Adults in such populations don't necessarily senesce much more rapidly. So a 60 year old in a low life expectancy population won't necessarily be more vulnerable to COVID-19 than a 60 year old in a high life expectancy population.

In fact, there are plausible reasons why they could be less susceptible - e.g. in high life expectancy populations we often see higher levels of obesity, metabolic syndrome, cardiovascular disease etc.

User avatar
shpalman
After Pie
Posts: 2239
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: Infection Fatality Rate (IFR)

Post by shpalman » Sat Nov 21, 2020 11:18 am

Life expectancy and healthy life expectancy at age 65 (pdf from 2013)

http://dx.doi.org/10.1787/health-data-en
Life expectancy at age 65 has increased significantly for both men and women during the past 50 years across both OECD countries and emerging economies. Some of the factors explaining these gains in life expectancy at age 65 include advances in medical care combined with greater access to health care, healthier lifestyles and improved living conditions before and after people reach age 65.
life-expectancy-at-65.png
life-expectancy-at-65.png (109.77 KiB) Viewed 242 times
http://dx.doi.org/10.1787/888932919194
Increased life expectancy at age 65 does not necessarily mean that the extra years lived are in good health. In Europe, an indicator of disability-free life expectancy known as “healthy life years” is calculated regularly, based on a general question about disability in the European Survey of Income and Living Conditions (EU-SILC). Given that this indicator has only recently been developed, long-time series are not yet available and efforts continue to improve its comparability.
healthy-life-expectancy-at-65-Europe.png
healthy-life-expectancy-at-65-Europe.png (69.34 KiB) Viewed 242 times
http://dx.doi.org/10.1787/888932919213

The hypothesis would be something like "Italy had lots of deaths because of its aging population". The data here show that Italy has a relatively long life expectancy at age 65 but not so many "healthy life years". But then Germany's "health life years" is right next to Italy's and Germany had a really low CFR in the first wave.

I think you can only end up concluding that countries with a high CFR in the first wave were the ones with the capacity to only test the most severe cases.

Slovakia is right at the low end and they've just tested their entire population so that could be useful data.
molto tricky

User avatar
shpalman
After Pie
Posts: 2239
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: Infection Fatality Rate (IFR)

Post by shpalman » Sat Nov 21, 2020 11:29 am

Slovakia has 91578 cases and 579 deaths out of a population of 5.45 million.

579/91578 = 0.63%.

57,462 cases were found by recent mass rapid antigen testing following 77,123 cases previously found by PCR testing. The antigen test is of course less accurate but the order of magnitude seems to match other results.
molto tricky

User avatar
sTeamTraen
Dorkwood
Posts: 1189
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: Infection Fatality Rate (IFR)

Post by sTeamTraen » Sat Nov 21, 2020 11:55 am

shpalman wrote:
Sat Nov 21, 2020 9:04 am
It depends what question you are trying to answer.
Indeed. I suppose it means that I don't think that the kinds of questions one can answer with a single IFR number are very relevant.
Something something hammer something something nail

User avatar
shpalman
After Pie
Posts: 2239
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: Infection Fatality Rate (IFR)

Post by shpalman » Sat Nov 21, 2020 11:59 am

sTeamTraen wrote:
Sat Nov 21, 2020 11:55 am
shpalman wrote:
Sat Nov 21, 2020 9:04 am
It depends what question you are trying to answer.
Indeed. I suppose it means that I don't think that the kinds of questions one can answer with a single IFR number are very relevant.
A single IFR for a country will help answer the question of what sort of order of magnitude of total deaths are we looking at if we take no measures whatsoever.
molto tricky

User avatar
sTeamTraen
Dorkwood
Posts: 1189
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: Infection Fatality Rate (IFR)

Post by sTeamTraen » Sat Nov 21, 2020 12:12 pm

shpalman wrote:
Sat Nov 21, 2020 11:59 am
A single IFR for a country will help answer the question of what sort of order of magnitude of total deaths are we looking at if we take no measures whatsoever.
Yes, but that "we" then has to include the idea that people don't take measures as individuals either. We know from Sweden, and personal experience elsewhere, that even if the government does nothing or very little, a lot of people will take quite substantial measures to protect themselves. So I don't see how an IFR can ever be much more useful than a model of what would happen if there was some kind of experiment on non-reasoning beings. (Insert joke about mailing packages to the members of lockdownskeptics.org here.)
Something something hammer something something nail

User avatar
shpalman
After Pie
Posts: 2239
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: Infection Fatality Rate (IFR)

Post by shpalman » Sat Nov 21, 2020 12:25 pm

sTeamTraen wrote:
Sat Nov 21, 2020 12:12 pm
shpalman wrote:
Sat Nov 21, 2020 11:59 am
A single IFR for a country will help answer the question of what sort of order of magnitude of total deaths are we looking at if we take no measures whatsoever.
Yes, but that "we" then has to include the idea that people don't take measures as individuals either. We know from Sweden, and personal experience elsewhere, that even if the government does nothing or very little, a lot of people will take quite substantial measures to protect themselves. So I don't see how an IFR can ever be much more useful than a model of what would happen if there was some kind of experiment on non-reasoning beings. (Insert joke about mailing packages to the members of lockdownskeptics.org here.)
They take those measures to protect themselves because they have a feel for the order of magnitude of the fatality rates even if they don't know the exact number and the rates feel skewed because it's on the the news all the time in a way which other causes of death aren't.
molto tricky

User avatar
shpalman
After Pie
Posts: 2239
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

Re: Infection Fatality Rate (IFR)

Post by shpalman » Sat Nov 21, 2020 1:14 pm

Personally I was more interested in the IFR during the first wave because the CFR looked so high, especially in Italy and the UK, that I wanted to get an idea of how many cases there were going untested in the population.

But the high CFR at the beginning of the first wave was in large part due to the infection getting into care homes and hospitals.

The IFR of covid seems to be in that zone in which it's high enough to make a serious dent in the demographics but not so high that people feel they can't live with that risk.
molto tricky

User avatar
bob sterman
Fuzzable
Posts: 307
Joined: Mon Nov 11, 2019 10:25 pm

Re: Infection Fatality Rate (IFR)

Post by bob sterman » Sat Nov 21, 2020 8:03 pm

I just did a quick calculation...

Not IFR but simply COVID-19 deaths per 100,000 population per day since date of first COVID-19 death in each of the badly affected European countries...

France 0.262
Italy 0.293
Spain 0.347
UK 0.308

I was surprised at how similar the figures are - given the much more substantial variation in diagnosed cases per capita.

User avatar
sTeamTraen
Dorkwood
Posts: 1189
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: Infection Fatality Rate (IFR)

Post by sTeamTraen » Sat Nov 21, 2020 9:49 pm

I wrote some code to try and estimate the CFR from the ECDC data. I assume that deaths lag cases by about 14 days. That's a parameter that I can tweak, but it doesn't make much difference if I use 7 or 10.

I'm starting in July because prior to that, especially in March/April, the rate of testing probably varied a lot more across countries than it does now; plus, the number of cases being tested was far lower, which biases the apparent CFR upwards.

So these numbers are, very roughly, the chance of someone dying given that they get a positive test, which seems to be of practical use for those of us who observe policy, because we are (well, I am anyway) as interested in what governments will make of the numbers, as I am in the direct health implications of the numbers.

The quick version is: "A bit more than 1%". If we are missing half of all cases (due to false negatives, people not bothering to get tested because "I know I have COVID" or "I know I don't have COVID [even though they do have it]", etc), that would give an IFR of about 0.7%. But as with the IFR discussion, this takes no account of age grouping; perhaps Norway's lower death rate is due to better protection of the elderly rather than better ICU treatment.

Code: Select all

European second wave CFR estimate
Cases from 2020-07-01 to 2020-11-06
Deaths from 2020-07-15 to 2020-11-20
Country            Cases  Deaths   Rate%
----------------------------------------
Italy             584443   12903   2.21%
Czechia           380140    6521   1.72%
Portugal          119438    2039   1.71%
United Kingdom    839656   12920   1.54%
Spain            1079561   14210   1.32%
Belgium           432258    5664   1.31%
France           1437107   17098   1.19%
Germany           424830    4562   1.07%
Switzerland       170232    1771   1.04%
Sweden             76659     758   0.99%
Slovakia           67069     551   0.82%
Netherlands       339800    2624   0.77%
Ireland            38584     264   0.68%
Finland             9910      46   0.46%
Denmark            39002     163   0.42%
Norway             13720      52   0.38%
----------------------------------------
Overall          6052409   82146   1.36%
Something something hammer something something nail

User avatar
sTeamTraen
Dorkwood
Posts: 1189
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: Infection Fatality Rate (IFR)

Post by sTeamTraen » Sun Nov 22, 2020 11:45 pm

It looks like new UK cases have finally plateaued - average just over 20,000 per day for the past 7 days, compared to 25,000 the week before. Deaths are up from 412 to 441, but the growth is slowing. With luck they will peak at around 450/day this week. The image here shows the 14-day moving averages.

The next problem can be seen in the rather gentle downward slope from the first wave: The UK numbers came down much slower than in comparable countries. If the same thing happens this time, I would be surprised if new cases are below 15,000 per day at the scheduled end of Lockdown 2 (2 December, I think) and --- even if the lockdown is maintained --- also surprised if they are much below 10,000 per day by Christmas.

(That said, one thing that might help would be if the slower downward slope in the first wave reflected a strong increase in testing in May, because the true number of cases in March-April was clearly a lot higher than those that were counted. That would imply that the actual drop in May was faster than it appears.)

UK.png
UK.png (20.59 KiB) Viewed 115 times
Something something hammer something something nail

Millennie Al
Fuzzable
Posts: 306
Joined: Mon Mar 16, 2020 4:02 am

Re: Infection Fatality Rate (IFR)

Post by Millennie Al » Mon Nov 23, 2020 4:47 am

sTeamTraen wrote:
Sun Nov 22, 2020 11:45 pm
It looks like new UK cases have finally plateaued
That'll make the nutters claim that we no longer need lockdown.
(That said, one thing that might help would be if the slower downward slope in the first wave reflected a strong increase in testing in May, because the true number of cases in March-April was clearly a lot higher than those that were counted. That would imply that the actual drop in May was faster than it appears.)
You could go by deaths instead. Testing for the disease may vary, but there's no doubt if someone is dead.
Covid-19 - Don't catch it: don't spread it.

User avatar
Martin_B
Snowbonk
Posts: 592
Joined: Mon Nov 11, 2019 1:20 pm
Location: Perth, WA

Re: Infection Fatality Rate (IFR)

Post by Martin_B » Mon Nov 23, 2020 5:38 am

Millennie Al wrote:
Mon Nov 23, 2020 4:47 am
You could go by deaths instead. Testing for the disease may vary, but there's no doubt if someone is dead.
Yes, but people argue over why someone died. If you had a heart condition (even one which you are keeping under control) and developed Covid and died, people have argued that they didn't really die of Covid, they died of the heart condition.

People who want to believe/want to convince other people will make any argument try and stick. Maybe they should claim that they can't be counted as having Covid if there was no Covid-denialist allowed in the hospital to count them and verify the diagnosis.
"My interest is in the future, because I'm going to spend the rest of my life there"

User avatar
sTeamTraen
Dorkwood
Posts: 1189
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: Infection Fatality Rate (IFR)

Post by sTeamTraen » Mon Nov 23, 2020 11:14 am

Millennie Al wrote:
Mon Nov 23, 2020 4:47 am
sTeamTraen wrote:
Sun Nov 22, 2020 11:45 pm
It looks like new UK cases have finally plateaued
That'll make the nutters claim that we no longer need lockdown.
As I predicted here: https://twitter.com/sTeamTraen/status/1 ... 27267?s=20
Millennie Al wrote:
Mon Nov 23, 2020 4:47 am
(That said, one thing that might help would be if the slower downward slope in the first wave reflected a strong increase in testing in May, because the true number of cases in March-April was clearly a lot higher than those that were counted. That would imply that the actual drop in May was faster than it appears.)
You could go by deaths instead. Testing for the disease may vary, but there's no doubt if someone is dead.
But since the frootloops are claiming that all the cases are false positive tests, they must have died of something else. Just try to get your head round this, from an actual pathologist: https://twitter.com/ClareCraigPath/stat ... 59045?s=20
Something something hammer something something nail

User avatar
Woodchopper
After Pie
Posts: 2080
Joined: Sat Oct 12, 2019 9:05 am

Re: Infection Fatality Rate (IFR)

Post by Woodchopper » Mon Nov 23, 2020 11:23 am

sTeamTraen wrote:
Mon Nov 23, 2020 11:14 am
But since the frootloops are claiming that all the cases are false positive tests, they must have died of something else. Just try to get your head round this, from an actual pathologist: https://twitter.com/ClareCraigPath/stat ... 59045?s=20
Yes, I was arguing with an anti-lockdown person over the summer. When asked what was behind the spike in excess deaths across Western Europe his reply was that all the people were being killed by the lockdowns themselves.

User avatar
sTeamTraen
Dorkwood
Posts: 1189
Joined: Mon Nov 11, 2019 4:24 pm
Location: Palma de Mallorca, Spain

Re: Infection Fatality Rate (IFR)

Post by sTeamTraen » Mon Nov 23, 2020 11:31 am

Woodchopper wrote:
Mon Nov 23, 2020 11:23 am
sTeamTraen wrote:
Mon Nov 23, 2020 11:14 am
But since the frootloops are claiming that all the cases are false positive tests, they must have died of something else. Just try to get your head round this, from an actual pathologist: https://twitter.com/ClareCraigPath/stat ... 59045?s=20
Yes, I was arguing with an anti-lockdown person over the summer. When asked what was behind the spike in excess deaths across Western Europe his reply was that all the people were being killed by the lockdowns themselves.
It's well known that spending four weeks watching Netflix leads to symptoms such as bilateral interstitial pneumonia and kidney failure.
Something something hammer something something nail

Post Reply