What data and research is available on the mortality risk (infection fatality rate) of COVID-19?

Question

Given controversy and confusion over aspects of COVID-19, particularly over comparisons and contrasts with the mortality risk of seasonal flu, this question seeks data and findings on the mortality risk (Infection Fatality Rate) of Covid-19.

The mortality risk, or Infection fatality rate (IFR), sometimes known as ‘mortality rate’, ‘death rate’, or 'lethality' represents the risk of dying after contracting the virus, and is calculated as the number of deaths divided by the number infections.

The mortality risk (IFR) of COVID-19 is often overlooked in mainstream news media, which tends to focus on the case fatality rate (number of deaths divided by the number of CASES), or falsely equate/conflate the two.

The case fatality rate is in itself not a valid indicator of the mortality risk of a virus as it does not account for unreported/asymptomatic cases and therefore the true number of infections, or estimates of them. Hence the interest in data/studies on the mortality risk.

Ref:
https://medicalxpress.com/news/2020-05-team-covid-infection-fatality.html
http://jdmichel.blog.tdg.ch/archive/2020/03/24/covid-19-the-game-is-over-305275.html?fbclid=IwAR2xfWuLuVodAVLk8bZ9VzDTGsa_7CS0B0xmlZw6AqskRNRjjGLDAi4Mp2c

Source / information criteria:

For the purposes of this question, all answers must come with specific, referenced data/findings based on (or at least explicitly indicating) the number of infections (either directly known through testing of the entire study population, as in confinement/isolation situations, or based on modeled/estimated/extrapolated number of infections), and the number of deaths in each case.

This question does not seek answers containing personal estimates from respondents, and sources providing guesses based on anecdotal evidence or impressions/experience do not meet the criteria for this question.

Information based entirely on the CASE fatality ratio (i.e which ignores the number of unreported infections/asymptomatic infections) is not of interest for the purposes of this question for the reasons mentioned.

To avoid statistical errors due to small sample sizes, please limit sources to only those findings which are based on populations/sample sizes (total number of infections, either estimated or known) greater than 500.

Other coronavirus strains already circulating among population before COVID-19 (eg. 229E, HKU1, NL63, OC43) are not of interest for the purposes of this question.

In the case of news articles, obviously references should be included in support of data/findings, or at least traceable from the article, otherwise the information is not useful.

Goes without saying that the ratio/risk presented in any given source does not, in itself, determine the quality of the answer, and all answers meeting the given criteria are encouraged, regardless how consistent or inconsistent the findings they contain may be with other findings.

Feel free to indicate the following data points if they are known when indicating any sources of information, as they are of obvious relevance for the mortality risk (IFR). In any case, I will read all posts and summarize the data under various headings as I have done in my own answer to the question, which I will update as answers come in.

• Average age of the people considered in the study, and age stratified mortality risk (mortality risk for different age groups) if known.
• How much time has passed since all persons considered in the study became infected if known, as deaths can occur after data is collected.
• Whether Outcome delay adjustments (adding a percentage increase of deaths due to possible deaths after releasing data) are included in the information, if known.
• Possible underlying illnesses which can increase likelihood of death if known.
• Other factors.

Below I have provided an answer (now a wiki) where all the sources which I have found or have been provided here meeting the given criteria are summarized with the corresponding mortality risk.

Edit (in light of one answer below):

Please note, answers based on 'common sense', or which cite or are based on data pertaining to cases (rather than infections) do not meet the criteria for this question. For the sake of the clarity and relevance of the data found here, please take the necessary time to research and understand these terms and the criteria specified for the data provided here before posting any data / papers.

Answer 1

Well, there is one meta-analytical estimate of the IFR for Covid-19 out already albeint only as a draft paper:

there were 13 estimates of IFR included in the final meta-analysis, from a wide range of countries, published between February and April 2020. The meta-analysis demonstrated a point-estimate of IFR of 0.75% (0.49-1.01%) with significant heterogeneity (p<0.001). Conclusion: Based on a systematic review and meta-analysis of published evidence on COVID-19 until the end of April, 2020, the IFR of the disease across populations is 0.75% (0.49-1.01%). However, due to very high heterogeneity in the meta-analysis, it is difficult to know if this represents the "true" point estimate. It is likely that different places will experience different IFRs. More research looking at age-stratified IFR is urgently needed to inform policy-making on this front.

A couple of additional points from the paper:

Analysing by country of origin did not appear to have a substantial effect on the findings, with both those studies from within and outside of China showing similar aggregate estimates [...] There was very significantly lower heterogeneity in studies published using Chinese data (I² = 0%, p>0.5)

On the other hand, they found that IFR estimates (insofar) increased by month, in April in particular, although this refers to the date of publication of the study rather than the time interval spanned by the study's observations. (Personally, I find this a little intriguing, as during the H1N1/09 pandemic, I've read--albeit not in great detail-- that the CFR estimates mostly went down over time.)

There's a summary table with the exact findings of each of the 13 studies at the end of this meta-analysis. Alas it's in a somewhat too gaudy of a format to include here; that table spans 5 pages in the draft paper.

Answer 2

Below is a summary of all data and findings answering this question meeting the question criteria which I have found or which have been pointed out in answers to this question. The mortality risk given is that provided in the research paper cited, or, where no paper is cited (sources 6,7 and 9), the number of deaths divided by number of infections based on the raw data.

1. 'Estimates of the severity of coronavirus disease 2019: a model-based analysis'. Robert Verity, PhD et al.
   https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(20)30243-7/fulltext

Mortality risk (overall) = 0.657%

Mortality risk (age stratified):
0–9 = 0.00161%
10–19 = 0.00695%
20–29 = 0.0309%
30–39 = 0.0844%
40–49 = 0.161%
50–59 = 0.595%
60–69 = 1.93%
70–79 = 4.28%
≥80 = 7.80%

---

2. LA COUNTY PUBLIC HEALTH DEPARTMENT / UNIVERSITY OF SOUTHERN CALIFORNIA STUDY FINDINGS.
   https://qz.com/1841445/covid-19-may-be-undercounted-50-fold-in-la-antibody-surveys-show/
   https://reason.com/2020/04/20/l-a-county-antibody-tests-suggest-the-fatality-rate-for-covid-19-is-much-lower-than-people-feared/
   Infections (in US, extrapolated) = 221,00 to 442,000.
   Age factors: Age adjusted estimate for entire US population.

Mortality risk (IFR) = 0.1% - 0.3%.

---

3. GANGELT, GERMANY STUDY.
   https://www.land.nrw/sites/default/files/asset/document/zwischenergebnis_covid19_case_study_gangelt_0.pdf
   https://www.technologyreview.com/2020/04/09/999015/blood-tests-show-15-of-people-are-now-immune-to-covid-19-in-one-town-in-germany/?fbclid=IwAR1P-zqpfx6ATyVsGYa_9EHVgr3aY0ryKDh_uuC90xKtXMWXKR4fR4OMYKI
   Age factors: Age adjusted estimate for entire adult population of Gangelt.
   Infections (approx.) = 2% of adult population of Gangelt.

Estimated mortality risk (IFR) = 0.37%

---

4. DIAMOND PRINCESS OUTBREAK.
   https://www.statista.com/statistics/1099517/japan-coronavirus-patients-diamond-princess/
   Date of last possible infection: March 1st, 2020
   Deaths to date: 13 Diamond Princess study #1 (Mortality risk):
   'Estimating the infection and case fatality ratio for COVID-19 using age-adjusted data from the outbreak on the Diamond Princess cruise ship'. Timothy W Russell et al.
   https://www.medrxiv.org/content/10.1101/2020.03.05.20031773v2
   Average age: Age adjusted estimate for entire population of China.
   CI= 95%.

Estimated mortality risk (IFR) = 0.5%

Diamond Princess study #2 (infections estimate):
‘The contribution of asymptomatic SARS-CoV-2 infections to transmission -a model-based analysis of the Diamond Princess outbreak’. Jon C . Emery et al.
https://cmmid.github.io/topics/covid19/reportsEmery_Transmission%20from%20asymptomatic%20SARS-CoV-2.pdf
Average age = 65.
Total estimated infections: 1,304 (1,198-1,416).

Mortality risk (IFR) based on mean infections estimate= 0.99%

---

5. MORTALITY RISK ESTIMATE FROM JEAN-DOMINQUE MICHEL.
   https://phusis.ch/2020/03/25/covid-19-il-sagit-dune-epidemie-banale/
   Age factors: Statistics cited for entire population of china as of time of publishing (25th March, 2020).

Mortality risk (IFR) =< 0.3%

---

6. STOCKHOLM STUDY.
   Infections data: ‘Estimates of the peak-day and the number of infected individuals during the covid-19 outbreak in the Stockholm region, Sweden.
   February – April 2020’.
   https://www.folkhalsomyndigheten.se/publicerat-material/publikationsarkiv/e/estimates-of-the-peak-day-and-the-number-of-infected-individuals-during-the-covid-19-outbreak-in-the-stockholm-region-sweden-february--april-2020/
   Type: Modelling based estimate.
   CI: Unknown.
   Deaths data: https://c19.se/en/Sweden/Stockholm.
   Age factors: Age adjusted estimate for entire population of Stockholm.
   Number of infections as of April 8th, 2020 = 70,500
   Number of deaths as of May 1st, 2020 = 1,417.

Mortality risk (IFR)= 2%

---

7. USS THEODORE ROOSEVELT OUTBREAK.
   https://navylive.dodlive.mil/2020/03/15/u-s-navy-covid-19-updates/
   Outbreak arrival date: 24th March, 2020.
   Number of infections = 2,141
   Number of deaths = 1
   Average age = ?

Mortality risk (IFR) = 0.046%

---

8. DRAFT OF META-STUDY ESTIMATING IFR OF CORONAVIRUS.
   'A systematic review and meta-analysis of published research data on COVID-19 infection-fatality rates', Gideon Meyerowitz-Katz et al.
   https://www.medrxiv.org/content/10.1101/2020.05.03.20089854v1
   Age factors: Meta-study based on studies with varying age groups/average ages. Average age of meta-study not indicated.

Estimated mortality risk = 0.75% (0.49-1.01%)

---

9. CHARLES DE GAULLE AIRCRAFT CARRIER OUTBREAK.
   https://en.wikipedia.org/wiki/COVID-19_pandemic_on_Charles_de_Gaulle
   Arrival date: April 10th.
   Average age = ?
   Number of infections = 1,046
   Deaths: 0

Mortality risk (IFR) = 0%

---

10. Bulletin of the World Health Organization, 99 (‎1)‎: 19 - 33F, Ioannidis, John P A. (‎2021)‎. Infection fatality rate of COVID-19 inferred from seroprevalence data.

Generalized Mortality Risk (IFR): 0.23%

---

Answer 3

Around 1 to 2%.

While @Fizz and @Dale Newton have already provided a nice collection (with statistics even), I'd like to add one more, which is based on common sense as well as statistics.

First off, the source should be such that it could reasonably be expected to report truthful data; that pretty much means democratic governments which are taking this seriously and not trying to minimize it for various reasons. Second, the number of infections (as the most likely source of error) should be estimated from as large a dataset as possible; so only data from countries that have a very large ratio of tests to positive results, and a rigorous test program. Third, the test should have a sensitivity and specificity which is well known.

The one source that best meets these criteria is South Korea. South Korea has done 802k tests, found 11142 confirmed infections, and had 264 deaths (as of 2020-05-22), for a raw infection fatality rate of 2.37%.

Australia and New Zealand both provide approximate confirmation of the South Korea-based estimate. Australia has done 1192k tests, found 7106 confirmed infections, and had 102 deaths, for an IFR of 1.43%. New Zealand has done 259k tests, found 1504 confirmed infections, and had 21 deaths, for an IFR of 1.39%.

In South Korea, the ratio of tests to confirmed infections at 72:1 is very high; Australia and New Zealand are even higher, but (perhaps) with less good tracing. The definition of "confirmed" is a positive PCR test, with a positive re-test. South Korea also has a robust contact tracing program, and surveilance for people presenting with symptoms which would be suspicious even if not linked to a known cluster. Recently, just one new case led to running 45k tests on possible contacts. Of course that doesn't mean they've caught every single infection, but they would not be far from it; it is reasonable to think they have found most of the infections, symptomatic or not. "Most" is a bit hard to quantify, but over 50% is virtually certain, and over 80% is quite likely.

The 2.37% fatality rate is quite disturbingly high, and much higher than most other reports, but it does in my opinion come from the most trustworthy (and largest) dataset. There are a few things that may account for that: demographics (older population), most cases occured quite early in the timeline (less worldwide experience on effective treatment), or possibly a few large clusters that just happened to be in an older population; and yes, of course they missed some infections. Because of that I think it's reasonable to think a population-average infection fatality rate ought to be a bit lower; 1% to 2% is a good common-sense range. A somewhat earlier but more detailed South Korean report including demographics is here: https://www.medrxiv.org/content/10.1101/2020.03.15.20036368v1.full.pdf (if anyone can find a more recent one, please add a link to it)

I consider most of the sources collected in other answers as not credible, for various reasons. The Lancet article is based primarily on Chinese data. The LA County study was done by Eran Bendavid (who did the similarly flawed Santa Clara study which was thoroughly trashed by just about everyone link link and subsequently retracted), used an antibody test that may have a fairly high false positive rate (which itself is based on a tiny validation data set; 3 false positives out of 401 known negative samples) and found 35 positives out of 863 (also quite small sample). Common sense says that conclusions based on a whole country's worth of data and based on ultra-reliable PCR tests trumps conclusions based on a few hundred tests based on unreliable antibodies, any day. Similarly, all of the studies cited in Meyerowitz-Katz's meta-analysis linked above either (a) draw conclusions from a tiny number of cases/infections/tests, or (b) use completely unreliable methods to estimate the rate of infections, or both.

I would give the whole-country data from South Korea (and other countries with thorough testing) a huge weight in any meta-analysis; data from large seroconversion studies a medium weight (but there aren't any of those yet); and small seroconversion studies (or studies that just estimate prevalence using models) essentially no weight. I think trying to claim we know a more precise rate than the 1-2% range I described, or that the rate is significantly less than 1% for an average population, is simply not supported by the data available now.

NOTE This assumes a health care system which is not overwhelmed. For an example of what happens if the system is overwhelmed, see Italy: also from Statista, Lombardy had 228k tests, 86k reported infections, 15.8k deaths. The ratio of tests to positive results is less than 3:1; the infection fatality rate based on known infections is 18%. How many total infections were there that are not counted? Well, Italian data on seroconversion is pretty sparse, but at least one source says 4-11% (presumably: 400k to 1.1M infections in Lombardy, with very wide error bars). That gives an IFR of 2 to 5% (also with very wide error bars).