PreprintPDF Available

Sampling Bias, Case Fatality Rates and COVID-19 Testing: A Further Analysis

Authors:
  • WardEnvironment
Preprints and early-stage research may not have been peer reviewed yet.

Abstract

Why do estimated Case Fatality Rates (CFRs) for COVID-19 vary so much across countries? Many explanations have so far been proposed including varying: health and/or age of populations; access to and/or quality of health care; methods of recording deaths, and; amounts of testing. This analysis shows that variations in the extent of sampling bias in COVID-19 testing can account for a lot of the variation in estimated CFRs, both across countries and across time. Countries that have tested more, relative to the number of deaths, have much lower CFR estimates, and this is highly statistically significant across 60 countries. Moreover, changes in CFR estimates over time are mainly consistent with changes in the extent of sampling bias. Conversely, neither GDP per capita, nor percentage of population over 65, nor median age of population can account for current variations in CFR estimates, likely because any effects from varying age and GDP are being masked by much larger variations in testing. These findings imply that the Infection Fatality Rate (IFR) will likely be at the lower end of the current CFR range (i.e. much closer to 0.14% than 15%), and it is estimated that the IFR will likely be in the range 0.28% to 0.68%. These findings also imply that many, or even most, countries have substantially underestimated numbers of cases of COVID-19, and that the basic reproduction number (R0) may also have been underestimated during the early stages of the pandemic. COVID-19 likely has a lower fatality rate (IFR) than some fear, but it also likely has a much higher infection rate (R0) than many realise. Many people, especially the elderly and those with pre-existing conditions, are therefore still at risk of dying. All countries thus need to do more testing urgently, especially of those with only mild or no symptoms, and this applies particularly to countries with low tests/deaths ratios. Other factors may also be affecting CFR estimates than were analysed, and so the IFR could still lie outside the estimated range (i.e. 0.28% to 0.68%). The analysis was limited by the availability, accuracy and consistency of official reports.
Sampling Bias, Case Fatality Rates and
COVID-19 Testing: A Further Analysis
by Dan Ward, WardEnvironment.
Email: info@wardenvironment.ch
Abstract
Why do estimated Case Fatality Rates (CFRs) for COVID-19 vary so much across countries? Many
explanations have so far been proposed including varying: health and/or age of populations; access
to and/or quality of health care; methods of recording deaths, and; amounts of testing. This analysis
shows that variations in the extent of sampling bias in COVID-19 testing can account for a lot of the
variation in estimated CFRs, both across countries and across time. Countries that have tested more,
relative to the number of deaths, have much lower CFR estimates, and this is highly statistically
significant across 60 countries. Moreover, changes in CFR estimates over time are mainly
consistent with changes in the extent of sampling bias. Conversely, neither GDP per capita, nor
percentage of population over 65, nor median age of population can account for current variations
in CFR estimates. These findings imply that the Infection Fatality Rate (IFR) will likely be at the
lower end of the current CFR range (i.e. much closer to 0.14% than 15%), and it is estimated that
the IFR will likely be in the range 0.28% to 0.68%. These findings also imply that many, or even
most, countries have substantially under-estimated numbers of cases of COVID-19, and that the
basic reproduction number (R0) may also have been under-estimated during the early stages of the
pandemic. COVID-19 likely has a lower fatality rate (IFR) than some fear, but it also likely has a
much higher infection rate (R0) than many realise. Many people, especially the elderly and those
with pre-existing conditions, are therefore still at risk of dying. All countries thus need to do more
testing urgently, especially of those with only mild or no symptoms, and this applies particularly to
countries with low tests/deaths ratios. Other factors may also be affecting CFR estimates than were
analysed, and so the IFR could still lie outside the estimated range (i.e. 0.28% to 0.68%). The
analysis was limited by the availability, accuracy and consistency of official reports.
Key Words: Sampling Bias, Case Fatality Rate, COVID-19 Testing
Key Points:
Variations in Sampling Bias in COVID-19 testing can account for a lot of the variation
in current estimates of Case Fatality Rates (CFRs), across countries and across time.
Countries that have tested more, relative to the number of deaths, have much lower
CFR estimates, and this is highly statistically significant across 60 countries.
This implies that the Infection Fatality Rate (IFR) for COVID-19 will be at the lower
end of the current CFR range, and it is estimated that the IFR will be 0.28% - 0.68%.
The findings also imply that many countries have substantially under-estimated cases,
and that the basic reproduction number (R0) may also have been under-estimated.
All countries need to do more testing urgently, including of those with only mild or no
symptoms, and this applies particularly to countries with lower tests/deaths ratios.
This publication is part of a series of publications on COVID-19 by Dan Ward. Any comments or
questions are very welcome. Please email any comments or questions to: info@wardenvironment.ch
23 April 2020 Dan Ward – www.wardenvironment.ch 1
Citation: Ward, Dan (2020) Sampling Bias, Case Fatality Rates and COVID-19 Testing: A Further Analysis.
Email: info@wardenvironment.ch Disclaimer: This preprint has not been peer reviewed.
Sampling Bias, Case Fatality Rates and COVID-19 Testing: A Further Analysis
Many different hypotheses have been proposed in recent weeks to explain the wide, and apparently
confusing, range in estimated COVID-19 Case Fatality Rates (CFRs)1 across countries. Such
estimates have been calculated simply as reported deaths divided by reported cases, and the current
range, as of 21 April, is 0.14% to 15% – see table 2, p9. Hypotheses proposed to explain this wide
variation include varying: access to, and quality of, health care2; age demographics of populations3;
methods for recording deaths4; health of populations5, and; amounts of testing6. Whilst all of these
hypotheses deserve important consideration, some publications about these hypotheses have risked
presenting a misleading interpretation, as they have either (1) drawn conclusions from a very small
sample size (e.g. just two countries7) and/or have (2) failed to adequately consider the likely large
impact of wide variations in sampling bias in testing on CFR estimates8. Some other publications
have proposed the hypothesis that sampling bias may be an important factor. However, these
publications have generally not been able to provide clear evidence to support that claim9. It is thus
the aim of this report to provide this key evidence, regarding the impact of sampling bias on CFRs.
Sampling Bias
Sampling bias occurs when test subjects are not randomly selected, and thus do not represent the
true distribution in the population being analysed10. This has likely occurred in many countries with
respect to COVID-19, because most countries have only tested a tiny minority of their populations
to datea, and many tests that have been done, have been prioritised on the seriously ill11, due to the
clinical priority of diagnosing such individuals (so they can be treated appropriately). Moreover,
some countries have had official policies not to test those with only mild or no symptoms12,13, whilst
it is already well-established that most people infected with COVID-19 do not get seriously ill.14,15
Previous Analysis
A statistical report published on 10 April 2020 used ratios between reported tests and reported
deaths (tests/deaths ratios) to estimate the amount of sampling bias in COVID-19 across countries.16
Countries with a high tests/deaths ratio have less sampling bias in their testing, as they are testing
more widely, and are thus better able to detect the majority of infected people, who only have mild
or no symptoms. Conversely, countries with low tests/deaths ratios are focusing their testing mostly
on the seriously ill and dying, and thus have much more sampling bias. Tests/deaths ratios were
then used to test an hypothesis that varying degrees of sampling bias are significantly impacting
CFR estimates. This hypothesis was supported in that, when 20 countries were ranked by
tests/deaths ratios, there was a significant difference in estimated CFR in the upper and lower
quartile of countries: countries with higher tests/ratios (less sampling bias) had significantly lower
CFR estimates (p=.016). Moreover, when this analysis was repeated two weeks later (9 April), for
the same 20 countries, the same result was found, further supporting the hypothesis tested.
New Analysis
It is now possible to test this hypothesis in much more detail, as the pandemic has spread much
more widely, and far more countries are now reporting up-to-date precise numbers of tests, deaths
and cases. This data was all taken from https://www.worldometers.info/coronavirus/ on 7 April
2020. The analysis was restricted to those countries with at least 5 deaths that report daily precise
numbers of tests, deaths and cases.b This yielded 60 countries. Moreover, a second analysis was
also performed, on the same 60 countries, using data taken from the same source on 21 April 2020.
a Maximum 12.8% (Iceland), and a mean of 1.24%, across the 60 countries analysed in this report, by 21 April 2020.
b These criteria meant that several key countries could not be included – e.g. China, France, Germany and Spain.
23 April 2020 Dan Ward – www.wardenvironment.ch 2
New Results
As with the previous report, it was again found that, when countries were ranked by tests/deaths
ratios, there was a significant difference (p=.000028)c in estimated CFR between the upper quartile
(mean estimated CFR 1.41%) and lower quartiles (mean estimated CFR 7.11%); see table 1, p8.
Moreover, the significance level of this result was much higher than in the previous report (p=.016),
probably because more countries could now be analysed. In addition, a similar result was found for
both datasets: 7 April (p=.000028) and 21 April (p=.000031)d; see table 2. R 2
calculations17 for
power equations for estimated CFRs v. tests/deaths were 0.75 and 0.67e; see figure 1 and 2f. This
suggests that sampling bias can account for a lot of the variation in CFR estimates across countries.
c Welch's t-test (29) = 7.87, p=.000028
d Welch's t-test (29) = 7.73, p=.000031; upper quartile (Mean CFR 1.96%) v. lower quartile (Mean CFR 7.66%).
e For the power equations shown on figure 1 and 2. R2 indicates how much variation these equations can explain.
fFor sources: see tables 1 and 2. The top and bottom four countries, for reported tests/deaths on 7 April, are labelled.
23 April 2020 Dan Ward – www.wardenvironment.ch 3
1 10 100 1,000 10,000 100,000
0%
1%
10%
100%
f(x) = 0.62 x^-0.49
= 0.67
Figure 2: Estimated CFR against Reported Tests / Deaths, by country (log scales) - 21 April
Reported Tests / Reported Deaths, by country (log scale)
Estimated CFR, by country (log scale)
1 10 100 1,000 10,000 100,000
0%
1%
10%
100%
f(x) = 0.69 x^-0.53
= 0.75
Figure 1: Estimated CFR against Reported Tests / Deaths, by country (log scales) - 7 April
Reported Tests / Reported Deaths, by country (log scale)
Es timated CFR, by country (log scale)
Qatar
Iceland
Belgium UK
Australia
Kazakhstan
Italy
Guyana
Qatar
Iceland
Kazakhstan
Australia
Guyana
Italy
UK
Belgium
f
f
It should also be noted that there have been some obvious changes during this two week period: 7
April to 21 April. Overall, the mean CFR for these 60 countries has increased (from 3.85% to
4.32%), and the mean tests/deaths ratio has decreased (from 975 to 939) – suggesting an inverse
relationship; calculated using numbers as shown on tables 1 and 2. In 44 (i.e. most) countries, CFR
estimates have increased. However, in 16 countries, CFR estimates have decreased (see table 3,
p10). Moreover, in 32 of the 44 countries with increased CFR, tests/deaths ratios have decreased,
and in all but two of the 16 countries with decreased CFR, tests/deaths ratios have increased.
Clearly testing is not the only factor influencing CFR estimates over time, and neither is it likely to
be the only factor influencing CFR estimates across countries. However, taken together, these
findings add a lot more support to the main hypothesis tested in that sampling bias (as estimated by
tests/deaths ratios) can account for a lot of the variation in CFR estimates, both across countries,
and across time. Countries with lower CFRs have less sampling bias (higher tests/deaths ratios), and
where CFRs have decreased, sampling bias has also mostly decreased. Conversely, countries with
higher CFRs have more sampling bias (lower tests/deaths ratios), and where CFR estimates have
increased, sampling bias has mainly increased too. In general, there is an inverse relationship
between CFR and tests/deaths, both across countries and time, which supports the main hypothesis.
Most CFR estimates have increased over this two-week period, but this change is mainly consistent
with (and thus can be mainly explained by) increases in sampling bias in most of these countries.
The other factors analysed in this report (age of population and GDP – see below) cannot account
for these changes over time, as neither would have changed in just two weeks. Other factors that
may also explain changes over time in CFRs might include: changes in the way that countries are
recording and/or reporting numbers of cases, deaths and/or tests, and/or; increases in the degree to
which health systems may be becoming over-burdened, which – if true – would be extremely
concerning. Both these factors could account for why CFR estimates in 12 countries have increased,
despite those countries now doing more testing, relative to deaths, but this remains to be shown.
Further Analysis
It was also possible to analyse these 60 countries further, using data from both 7 and 21 April, for
three other possible factors that may be affecting CFR estimates: GDP per capita18, median age19 of
population and % of population over 6520. GDP per capita may be important because richer
countries likely have better health systems, and are thus better able to detect and treat COVID-19
cases. Both median age and % of population over 65 may be important, because it is already well-
established that, once infected, elderly people are much more likely to become seriously ill.21 Many
other factors are also likely to be affecting CFR estimates. However, given the availability of data,
and the urgent need to provide more clarity, this analysis was restricted to these factors.
Further Results
Analysing the 60 countries on 7 and 21 April showed that neither median age of population nor %
of population over 65 have a significant impact on CFR estimates at present. When ranked by both
these variables, there was no significant difference found in estimated CFR in the upper and lower
quartiles of countriesg. These findings are consistent with findings of the previous report, and are
likely due to any effect from varying age being masked by much larger variations in testing. It was
also found that, when countries were ranked by GDP per capita, there was also no significant
difference in CFR estimatesh. This contradicts the finding of the previous report, and may be
g Median Age: 7 April - Welch's t-test (29) = 3.20, p = .11; 21 April - Welch's t-test (29) = 0.05, p = .98
% population over 65: 7 April - Welch's t-test (29) = 1.35, p = .50; 21 April - Welch's t-test (29) = 2.23, p = .26
h 7 April: Welch's t-test (29) = 3.02, p = .13; 21 April: t-test (29) = 0.62, p = .76
23 April 2020 Dan Ward – www.wardenvironment.ch 4
because variations in testing, and/or other factors, are now masking the effect of GDP, which was
then significant. Both GDP and age of population are still likely to have some impact on some CFR
estimates, but neither can account for the wide variation in current CFR estimates across countries.
Conclusions
The main finding of this update report is that it is even more apparent now that sampling bias can
account for a lot of the variation in CFR estimates. Countries that test more, relative to the
number of deaths, have much lower CFR estimates, probably because they are much more able to
detect the majority of infected people who only have mild or no symptoms. Conversely, countries
that have tested less, relative to the number of deaths, have much higher CFR estimates. In addition,
changes in CFR estimate over time are mainly consistent with changes in tests/deaths ratios (i.e.
changes in the extent of sampling bias). In most countries, CFR estimates have increased recently,
and this seems to be mainly because most of these countries now have more sampling bias in their
testing. This is because, as deaths have increased, testing has mainly increased at a slower rate, thus
increasing sampling bias. The very wide variation in current CFR estimates (0.14% to 15% – i.e.
105 fold) may at first seem very confusing. However, once the even larger variation in sampling
bias is taken into account (28 to 7,414 – i.e. 265 fold; see table 2), the current variation in CFRs can
be much more easily explained. Moreover, these findings have very significant policy implications:
1. Given that countries with less sampling bias have significantly lower CFR estimates, and
that changes in CFR estimates over time are mainly consistent with changes in sampling
bias, all countries need to do more testing urgently, and especially those countries with low
tests/deaths ratios (see tables 1 and 2). Moreover, this is important, not only to have more
reliable statistics, but also to save lives and the economy. Countries that have done the most
testing to date seem to be the ones that are managing the pandemic better, e.g. Iceland22,
Germany23 and South Korea24. In particular, more testing allows countries to do more robust
quarantining of infected individuals, and tracking and tracing of contacts25. In addition,
increasing testing, especially of those with only mild or no symptoms, may also be key to
ensuring that economically painful restrictions26 can be reduced sooner rather than later27.
2. Given that countries with the least sampling bias have the lowest CFR estimates, and that
changes in CFR estimates over time are mainly consistent with changes in sampling bias,
this implies that the underlying Infection Fatality Rate (IFR)i for COVID-19 will likely be
at the lower end of the current range of estimated CFRs across countries (i.e. much closer to
0.14%, than 15%) – see table 2. New estimates analysing data from China suggest an IFR of
0.66%28. Similarly, the results of antibody tests on blood donors in the Netherlands29
(announced on 16 April) suggest a very similar IFR estimate (0.64%)j. Some other IFR
estimates are lower, such as the findings from the first random testing for COVID-19 in
Germany, which suggest an IFR of 0.37%30. Moreover, a number of factors (not analysed in
this report) could be causing this variation in estimates, including differences in the degree
to which some countries may be under-reporting deaths, which may be by around a factor of
two in some places31,32,33. It therefore seems prudent to conclude a range for the final IFR,
which is here estimated to be 0.28% to 0.68% (which has a midpoint of 0.48%). The low
point in this range is twice the lowest current CFR (i.e. Qatar with 0.14% – see table 2),
whilst the high point takes into account the higher IFRs estimated from China and the
i The IFR can be thought of as the likelihood that an individual infected with COVID-19 will die.
j 3,315 deaths, divided by the estimation of 514,005 infections (3% of population infected) – resulting in 0.64%, as on
16 April 2020. COVID-19 deaths, and population, for Netherlands both taken from www.worldometers.info. This
assumes that the blood donors were representative of the wider population. However, seriously ill people may not
have donated blood, meaning actual infections may have been higher, and the IFR could thus be lower than 0.64%.
23 April 2020 Dan Ward – www.wardenvironment.ch 5
Netherlands. This range is also consistent with the current estimated CFR from Iceland
(0.56% – see table 2), which is significant because Iceland has tested far more of their
population than any other country (i.e. 12.84%)k. It should also be noted that CFRs in
particular countries will likely turn out to either slightly above or below the finally
concluded IFR, particularly given variations in GDP per capita and percentage of population
over 65 across countries, which are likely to have some impact on particular countries: GDP
per capita will likely reduce some CFR estimates, and percentage of population over 65 will
increase some CFR estimates. But neither of these factors will be as significant as testing,
and only testing has been here found to account for the wide variations in estimated CFRs.
3. Given that countries with more sampling bias have significantly higher CFR estimates, this
means many countries have been substantially under-estimating cases, because they have
done far too little testing. By 21 April, no country had tested more than 12.8% of their
population, and on average (across the 60 countries analysed) countries have only tested
1.24%l. Moreover, many countries are routinely only testing the seriously ill and dying, and
some countries have had official policies not to test those with only mild or no symptoms.34
The resulting sampling bias means that many – or even most – reported cases are substantial
underestimates35 compared with the likely number of actual infections. If it is assumed that
reported deaths are reasonably accurate, and that the IFR is 0.48% (i.e. the midpoint of
0.28% - 0.68%; see conclusion 2), this would mean that, by 21 April, 37 million people had
already been infected globally, compared with 2.5 million reported cases. Moreover, in
countries with higher tests/deaths ratios, this disparity is even more profound (see table 2).
Italy, under these assumptions, would have had 5 million infections by 21 April, as opposed
to 184,000 reported cases (estimated infections calculated as reported deaths divided by
assumed IFR)m. As a result, many healthy, young or middle-aged people are likely to be
infected without even knowing it, and could easily be passing on the virus to the elderly and
the vulnerable. Governments need to address and communicate the under-estimation of
cases urgently – especially on official websites and regular briefings – and individuals need
to understand that they are much more likely to be infected than they may realise.
4. Given that many countries have been under-estimating cases, due to sampling bias, and that
this sampling bias has largely been increasing (as deaths have increased faster than tests),
this also suggests that the basic reproduction number (R0)36 may have been under-
estimated, especially during the early stages of the pandemic. R0 can be thought of as the
number of people one infected person infects, on average, without any interventions in place
– e.g. no social distancing, travel restrictions, quarantining etc. The World Health
Organisation (WHO) initial R0 estimate on 23 January was 1.4 to 2.5.37 However, a re-
evaluation of the likely spread of the COVID-19 during the early stages of the pandemic,
published on 7 April (which sought to control for biases and inaccuracies in early reporting),
estimated an R0 of 5.7.38 This much higher estimate is also consistent with the first findings
of random testing announced on 10 April in Germany: testing of 1,000 randomly-selected
people in the region of Gangelt found that 15% of people were already infected by that
date.39 This finding is significant, because flu pandemics usually only infect up to 20% of
people over the entire pandemic, lasting many months, and with few interventions (e.g. no
social distancing or travel restrictions).40 COVID-19 is already approaching that level, at
least in parts of Germany, in a matter of weeks, and even with stringent measures in place.
k This compares with an average of just 1.24% of populations tested, by 21 April 2020, for the 60 countries analysed.
l Top country is Iceland. Populations from: www.worldometers.info/world-population/population-by-country/
mNote: these estimates of infections would be reduced by a higher IFR, and increased by higher death estimates.
Reported numbers of cases and deaths taken from www.worldometers.info/coronavirus on 21 April 2020.
23 April 2020 Dan Ward – www.wardenvironment.ch 6
This means that the R0 for COVID-19 will indeed likely be much higher than flu (0.9 -
2.1)41, and an R0 of around 5.7 may well be more accurate. This has big implications for
how the pandemic is being managed. In particular, much more attention needs to be paid to
the likely high majority of cases that have gone undetected. COVID-19 is killing many
thousands of people, even with a likely relatively low IFR (see conclusion 2), because the
R0 is likely very high – and that is the key problem that needs to be focused on. The fatality
rate (IFR) is likely to be much less than some fear, but the infection rate (R0) is likely to be
much higher than many realise. As a result, the pandemic is still very dangerous, and many
people, particularly the elderly and those with pre-existing conditions, are at risk of dying.
This means that control measures and increases in testing are even more important.
Extra Note: Why use Tests/Deaths, and not Test/Cases or Tests per capita?
Tests/deaths is a more useful estimate of sampling bias than tests per capita42, because countries
with more infections need to do more testing to obtain the same level of reliability in their testing. If
we imagine two countries with the same population, but one has a lot more infections, the country
with more infections will need to do a lot more testing, to reduce their sampling bias to the same
level, particularly because most countries are prioritising tests on the seriously ill43 (mainly due to
the clinical need to do so, so that such individuals can be diagnosed and treated appropriately).
Moreover, deaths are a more accurate predictor of the actual number of infections than cases,
because deaths are likely to have been much more accurately reported than cases in most countries.
In particular, whilst deaths may be being under-estimated in some countries by a factor of two44,
cases may be being under-estimated by up to a factor of 60n, 45, 46. Or, to put it simply: although both
deaths and cases may be being under-estimated, it is much more likely that countries are failing to
notice and count people with an unusual cough (or no symptoms at all) as opposed to those who die.
Tests/deaths is thus a better estimation of sampling bias than either tests/cases or tests per capita.
Limitations
Many other factors may also be affecting CFR estimates than were analysed, and thus the IFR for
COVID-19 could still turn out to be outside the range estimated by this report (i.e. 0.28% - 0.68%).
This analysis was also limited by the availability, accuracy and consistency of official reports.
Acknowledgements
Thanks to Sarah Tschopp, Neal Ward, Ian Conlon, Eoghan Flanagan, Joel Phillips, Tobias Rudolph,
Christopher Leffler and Sandro Mendonça for valuable comments on earlier versions of this report.
n Italy, 28 March: 5.9 million infections estimated by Imperial College London, compared with 92,472 reported cases.
23 April 2020 Dan Ward – www.wardenvironment.ch 7
Sources: Cases, deaths and tests from www.worldometers.info/coronavirus on 7 April at 23:59 GMT. GDP from:
www.worldometers.info/gdp/gdp-by-country/. Median age: https://worldpopulationreview.com/countries/median-age/
Percentage of population over 65: https://data.worldbank.org/indicator/SP.POP.65UP.TO.ZS - 2018 data.
23 April 2020 Dan Ward – www.wardenvironment.ch 8
Upper Quartile (15 countries)
Mean CFR estimate = 1.41%
Lower Quartile (15 countries)
Mean CFR estimate = 7.11%
Table 1: Countries ranked by decreasing Tests/Deaths ratios, as of 7 April 2020
Country Test/Deaths
Kazakhstan 697 6 0.86% 54,767 $9,009 30.6 7.39% 9,128
Qatar 2,057 6 0.29% 41,818 $61,264 33.2 1.37% 6,970
Aus tralia 5,988 49 0.82% 310,700 $53,831 38.7 15.66% 6,341
Iceland 1,586 6 0.38% 28,991 $73,233 36.5 14.80% 4,832
Thailand 2,258 27 1.20% 71,860 $6,579 37.7 11.90% 2,661
S. Korea 10,331 192 1.86% 477,304 $29,958 41.8 14.42% 2,486
Lithuania 880 15 1.70% 27,807 $16,709 43.7 19.71% 1,854
Israel 9,248 65 0.70% 117,339 $42,852 29.9 11.98% 1,805
Cyprus 494 9 1.82% 13,017 $18,695 36.8 13.72% 1,446
Chile 5,116 43 0.84% 57,122 $15,001 34.4 11.53% 1,328
Norway 6,086 89 1.46% 113,896 $75,428 39.2 17.05% 1,280
Estonia 1,149 21 1.83% 23,546 $20,170 42.7 19.63% 1,121
5,017 88 1.75% 91,247 $20,291 42.1 19.42% 1,037
Canada 17,897 381 2.13% 346,507 $44,841 42.2 17.23% 909
Kenya 172 6 3.49% 4,973 $1,578 19.7 2.70% 829
Slovenia 1,059 36 3.40% 29,455 $23,488 44.5 19.61% 818
Uruguay 424 7 1.65% 5,727 $16,341 35.0 14.81% 818
Poland 4,848 129 2.66% 92,215 $13,871 40.7 17.52% 715
Pakis tan 4,035 57 1.41% 39,183 $1,467 23.8 4.31% 687
Croatia 1,282 18 1.40% 12,322 $13,200 43.0 20.45% 685
Armenia 853 8 0.94% 5,140 $3,918 35.1 11.25% 643
Cuba 396 11 2.78% 7,054 $8,541 41.5 15.19% 641
Colombia 1,780 50 2.81% 30,445 $6,429 30.0 8.48% 609
Japan 4,257 93 2.18% 55,311 $38,214 47.3 27.58% 595
Luxembourg 2,970 44 1.48% 25,702 $105,280 39.3 14.18% 584
Lebanon 548 19 3.47% 10,221 $7,857 30.5 7.00% 538
Hungary 817 47 5.75% 23,746 $14,364 42.3 19.16% 505
Aus tria 12,639 243 1.92% 115,235 $47,261 44.0 19.00% 474
Paraguay 115 5 4.35% 2,039 $5,776 28.2 6.43% 408
Iraq 1,122 65 5.79% 26,331 $5,114 20.0 3.32% 405
Niger 278 11 3.96% 4,199 $376 15.4 2.60% 382
Tunisia 623 23 3.69% 8,274 $3,494 31.6 8.32% 360
Greece 1,832 81 4.42% 28,584 $19,214 44.5 21.66% 353
Portugal 12,442 345 2.77% 121,256 $21,316 42.2 21.95% 351
Turkey 34,109 725 2.13% 228,868 $10,498 30.9 8.48% 316
Denmark 5,071 203 4.00% 58,419 $57,545 42.2 19.81% 288
Bangladesh 164 17 10.37% 4,289 $1,564 26.7 5.16% 252
Moldova 1,056 22 2.08% 5,108 $2,002 36.7 11.47% 232
North Maced. 599 26 4.34% 5,879 $5,418 37.9 13.67% 226
Romania 4,417 197 4.46% 43,578 $10,781 41.1 18.34% 221
764 33 4.32% 6,911 $5,387 42.1 16.47% 209
Switzerland 22,253 821 3.69% 167,429 $80,296 42.4 18.62% 204
Ireland 5,709 210 3.68% 42,484 $69,727 36.8 13.87% 202
Peru 2,954 107 3.62% 21,555 $6,723 28.0 8.09% 201
Argentina 1,715 60 3.50% 11,778 $14,508 31.7 11.12% 196
Mexico 2,439 125 5.13% 20,475 $9,224 28.3 7.22% 164
USA 400,335 12,841 3.21% 2,075,739 $59,939 38.1 15.81% 162
Serbia 2,447 61 2.49% 9,626 $4,692 42.6 18.35% 158
Ukraine 1,462 45 3.08% 6,385 $2,521 40.6 16.43% 142
Albania 383 22 5.74% 2,753 $4,521 32.9 13.74% 125
107 8 7.48% 878 $15,952 36.0 10.73% 110
Andorra 545 22 4.04% 1,673 $39,128 44.3 16.18% 76
Ecuador 3,995 220 5.51% 14,406 $6,214 27.7 7.16% 65
Indones ia 2,738 221 8.07% 14,354 $3,837 30.2 5.86% 65
Morocco 1,184 90 7.60% 5,437 $3,083 29.3 7.01% 60
Iran 62,589 3,872 6.19% 211,136 $5,628 30.3 6.18% 55
Italy 135,586 17,127 12.63% 755,445 $32,038 45.5 22.75% 44
UK 55,242 6,159 11.15% 266,694 $39,532 40.5 18.40% 43
Belgium 22,194 2,035 9.17% 80,512 $43,325 41.4 18.79% 40
Guyana 33 5 15.15% 132 $4,671 26.2 6.45% 26
Reported
Cases
Reported
Deaths
Es timated
CFR
Reported
Tests
GDP per
capita
Median Age
(years )
% of pop.
over 65
Czechia
Bos. & Her.
Trin. & Tob.
Sources: Cases, deaths and tests from www.worldometers.info/coronavirus on 21April at 23:59 GMT. GDP from:
www.worldometers.info/gdp/gdp-by-country/. Median age: https://worldpopulationreview.com/countries/median-age/
Percentage of population over 65: https://data.worldbank.org/indicator/SP.POP.65UP.TO.ZS - 2018 data.
23 April 2020 Dan Ward – www.wardenvironment.ch 9
Upper Quartile (15 countries)
Mean CFR estimate = 1.96%
Lower Quartile (15 countries)
Mean CFR estimate = 7.66%
Table 2: Countries ranked decreasing by Tests/Deaths ratios, as of 21 April 2020
Country Tests/Deaths
Qatar 6,533 9 0.14% 66,725 $61,264 33.2 1.37% 7,414
Kazakhstan 1,995 19 0.95% 126,727 $9,009 30.6 7.39% 6,670
Australia 6,645 71 1.07% 439,000 $53,831 38.7 15.66% 6,183
Iceland 1,778 10 0.56% 43,831 $73,233 36.5 14.80% 4,383
Cyprus 784 12 1.53% 37,081 $18,695 36.8 13.72% 3,090
Thailand 2,811 48 1.71% 142,589 $6,579 37.7 11.90% 2,971
S. Korea 10,683 237 2.22% 571,014 $29,958 41.8 14.42% 2,409
Lithuania 1,350 38 2.81% 70,753 $16,709 43.7 19.71% 1,862
Israel 13,942 184 1.32% 240,303 $42,852 29.9 11.98% 1,306
Uruguay 543 12 2.21% 13,923 $16,341 35.0 14.81% 1,160
Kenya 296 14 4.73% 14,704 $1,578 19.7 2.70% 1,050
Lebanon 677 21 3.10% 21,764 $7,857 30.5 7.00% 1,036
Estonia 1,552 43 2.77% 42,213 $20,170 42.7 19.63% 982
7,033 201 2.86% 178,617 $20,291 42.1 19.42% 889
Chile 10,832 147 1.36% 122,357 $15,001 34.4 11.53% 832
Cuba 1,137 38 3.34% 30,416 $8,541 41.5 15.19% 800
Norway 7,241 182 2.51% 145,279 $75,428 39.2 17.05% 798
Paraguay 208 8 3.85% 5,878 $5,776 28.2 6.43% 735
Iraq 1,602 83 5.18% 60,837 $5,114 20.0 3.32% 733
Armenia 1,401 24 1.71% 13,929 $3,918 35.1 11.25% 580
Poland 9,856 401 4.07% 224,355 $13,871 40.7 17.52% 559
Slovenia 1,344 77 5.73% 42,976 $23,488 44.5 19.61% 558
Pakis tan 9,565 201 2.10% 111,806 $1,467 23.8 4.31% 556
Croatia 1,908 48 2.52% 26,610 $13,200 43.0 20.45% 554
Tunisia 901 38 4.22% 18,165 $3,494 31.6 8.32% 478
Greece 2,401 121 5.04% 55,666 $19,214 44.5 21.66% 460
Luxembourg 3,618 78 2.16% 34,962 $105,280 39.3 14.18% 448
Japan 11,512 281 2.44% 124,550 $38,214 47.3 27.58% 443
Ukraine 6,125 161 2.63% 61,997 $2,521 40.6 16.43% 385
Aus tria 14,873 491 3.30% 189,018 $47,261 44.0 19.00% 385
1,342 51 3.80% 19,195 $5,387 42.1 16.47% 376
Portugal 21,379 762 3.56% 271,962 $21,316 42.2 21.95% 357
Serbia 6,890 130 1.89% 45,355 $4,692 42.6 18.35% 349
Colombia 4,149 196 4.72% 65,169 $6,429 30.0 8.48% 332
Peru 17,837 484 2.71% 155,724 $6,723 28.0 8.09% 322
Turkey 95,591 2259 2.36% 713,409 $10,498 30.9 8.48% 316
Canada 38,422 1834 4.77% 569,878 $44,841 42.2 17.23% 311
Denmark 7,695 370 4.81% 100,543 $57,545 42.2 19.81% 272
Bangladesh 3,382 110 3.25% 29,578 $1,564 26.7 5.16% 269
Argentina 3,144 151 4.80% 36,611 $14,508 31.7 11.12% 242
Niger 657 20 3.04% 4,832 $376 15.4 2.60% 242
Hungary 2,098 213 10.15% 50,052 $14,364 42.3 19.16% 235
Albania 609 26 4.27% 6,017 $4,521 32.9 13.74% 231
North Mac. 1,231 55 4.47% 12,340 $5,418 37.9 13.67% 224
Romania 9,242 498 5.39% 101,552 $10,781 41.1 18.34% 204
115 8 6.96% 1,393 $15,952 36.0 10.73% 174
Moldova 2,614 72 2.75% 11,763 $2,002 36.7 11.47% 163
Switzerland 28,063 1478 5.27% 227,554 $80,296 42.4 18.62% 154
Ireland 16,040 730 4.55% 111,584 $69,727 36.8 13.87% 153
Morocco 3,209 145 4.52% 18,100 $3,083 29.3 7.01% 125
USA 818,744 45318 5.54% 4,187,392 $59,939 38.1 15.81% 92
Indones ia 7,135 616 8.63% 50,370 $3,837 30.2 5.86% 82
Mexico 8,772 712 8.12% 49,570 $9,224 28.3 7.22% 70
Iran 84,802 5297 6.25% 365,723 $5,628 30.3 6.18% 69
Ecuador 10,398 520 5.00% 33,389 $6,214 27.7 7.16% 64
Italy 183,957 24648 13.40% 1,450,150 $32,038 45.5 22.75% 59
Guyana 66 7 10.61% 328 $4,671 26.2 6.45% 47
Andorra 717 37 5.16% 1,673 $39,128 44.3 16.18% 45
UK 129,044 17337 13.43% 535,342 $39,532 40.5 18.40% 31
Belgium 40,956 5998 14.64% 167,110 $43,325 41.4 18.79% 28
Reported
Cases
Reported
Deaths
Estimated
CFR
Reported
Tests
GDP per
Capita
Median Age
(years )
% population
over 65
Czechia
Bos. & Her.
Trin. & Tob.
23 April 2020 Dan Ward – www.wardenvironment.ch 10
Notes: Countries with decreased CFR estimate coloured GREEN. Countries with increased
CFR estimate coloured PURPLE. Most countries show an inverse relationship between CFR
and tests/deaths over time. However, 14 countries do not, and these countries are highlighted.
Sources: Reported cases, deaths and tests from www.worldometers.info/coronavirus
Table 3: Countries ranked by increasing percentage change in CFR estimate
Country
Bangladesh 10.37% 3.25% 252 269 -68.62%
Qatar 0.29% 0.14% 6,970 7,414 -52.77%
Morocco 7.60% 4.52% 60 125 -40.56%
Guyana 15.15% 10.61% 26 47 -30.00%
Albania 5.74% 4.27% 125 231 -25.68%
Peru 3.62% 2.71% 201 322 -25.09%
Serbia 2.49% 1.89% 158 349 -24.31%
Niger 3.96% 3.04% 382 242 -23.07%
Cyprus 1.82% 1.53% 1,446 3,090 -15.99%
Ukraine 3.08% 2.63% 142 385 -14.60%
4.32% 3.80% 209 376 -12.02%
Paraguay 4.35% 3.85% 408 735 -11.54%
Iraq 5.79% 5.18% 405 733 -10.57%
Lebanon 3.47% 3.10% 538 1,036 -10.53%
Ecuador 5.51% 5.00% 65 64 -9.19%
7.48% 6.96% 110 174 -6.96%
Iran 6.19% 6.25% 55 69 0.97%
North Mac. 4.34% 4.47% 226 224 2.93%
Italy 12.63% 13.40% 44 59 6.07%
Indones ia 8.07% 8.63% 65 82 6.96%
Kazakh stan 0.86% 0.95% 9,128 6,670 10.63%
Turkey 2.13% 2.36% 316 316 11.18%
Japan 2.18% 2.44% 595 443 11.73%
Greece 4.42% 5.04% 353 460 13.98%
Tunisia 3.69% 4.22% 360 478 14.24%
S. Ko rea 1.86% 2.22% 2,486 2,409 19.37%
Denmark 4.00% 4.81% 288 272 20.11%
Cuba 2.78% 3.34% 641 800 20.32%
UK 11.15% 13.43% 43 31 20.50%
Romania 4.46% 5.39% 221 204 20.82%
Ireland 3.68% 4.55% 202 153 23.73%
Andorra 4.04% 5.16% 76 45 27.84%
Portugal 2.77% 3.56% 351 357 28.54%
Aus tralia 0.82% 1.07% 6,341 6,183 30.57%
Moldova 2.08% 2.75% 232 163 32.21%
Uruguay 1.65% 2.21% 818 1,160 33.86%
Kenya 3.49% 4.73% 829 1,050 35.59%
Argentina 3.50% 4.80% 196 242 37.28%
Switzerland 3.69% 5.27% 204 154 42.75%
Thailand 1.20% 1.71% 2,661 2,971 42.80%
Luxembourg 1.48% 2.16% 584 448 45.52%
Iceland 0.38% 0.56% 4,832 4,383 48.67%
Pakis tan 1.41% 2.10% 687 556 48.76%
Estonia 1.83% 2.77% 1,121 982 51.59%
Poland 2.66% 4.07% 715 559 52.90%
Mexico 5.13% 8.12% 164 70 58.37%
Belgium 9.17% 14.64% 40 28 59.72%
Chile 0.84% 1.36% 1,328 832 61.46%
1.75% 2.86% 1,037 889 62.94%
Lit huania 1.70% 2.81% 1,854 1,862 65.14%
Colombia 2.81% 4.72% 609 332 68.18%
Slovenia 3.40% 5.73% 818 558 68.53%
Aus tria 1.92% 3.30% 474 385 71.71%
Norway 1.46% 2.51% 1,280 798 71.88%
USA 3.21% 5.54% 162 92 72.56%
Hungary 5.75% 10.15% 505 235 76.48%
Croatia 1.40% 2.52% 685 554 79.18%
Armenia 0.94% 1.71% 643 580 82.66%
Israel 0.70% 1.32% 1,805 1,306 87.77%
Canada 2.13% 4.77% 909 311 124.22%
7 Apri l CFR
Es timate
21 April CFR
Es timate
7 Apri l
Tests/Death
21 April
Tests/Deaths
% Change in CFR
Es timate
Bos. & Her.
Trin. & Tob.
Czechia
1"case fatality rate | Definition, Example, Equation, & Facts". Encyclopedia Britannica. Retrieved 18 April 2020.
2"Coronavirus: Why Germany has such a low COVID-19 death rate". Sky News. Retrieved 18 April 2020
3 Onder, Graziano; Rezza, Giovanni; Brusaferro, Silvio (2020-03-23). "Case-Fatality Rate and Characteristics of
Patients Dying in Relation to COVID-19 in Italy". JAMA. doi:10.1001/jama.2020.4683. Retrieved 20 April 2020.
4 Mary Van Beusekom | News Writer | CIDRAP News |; 24 March 2020. "Italian doctors note high COVID-19 death
rate, urge action". CIDRAP. Retrieved 18 April 2020.
5 Brake, Samuel James; Barnsley, Kathryn; Lu, Wenying; McAlinden, Kielan Darcy; Eapen, Mathew Suji; Sohal,
Sukhwinder Singh (2020/3). "Smoking Upregulates Angiotensin-Converting Enzyme-2 Receptor: A Potential
Adhesion Site for Novel Coronavirus SARS-CoV-2 (Covid-19)". Journal of Clinical Medicine. 9 (3): 841.
doi:10.3390/jcm9030841. Retrieved 19 April 2020.
6 Rema Nagarajan / TNN / Updated. 11 April 2020. "Why death rates vary widely in states: Testing could be key |
India News - Times of India". The Times of India. Retrieved 23 April 2020.
7"Comparing South Korea and Italy's healthcare systems and initiatives to combat COVID-19". April 2020.
www.paho.org. Retrieved 19 April 2020.
8 Dowd, Jennifer Beam; Andriano, Liliana; Brazel, David M.; Rotondi, Valentina; Block, Per; Ding, Xuejie; Liu, Yan;
Mills, Melinda C. (2020-04-16). "Demographic science aids in understanding the spread and fatality rates of
COVID-19". Proceedings of the National Academy of Sciences. doi:10.1073/pnas.2004911117. ISSN: 0027-8424.
PMID 32300018. Retrieved 19 April 2020.
9 Courage, Katherine Harmon (2020-04-01). "The stark differences in countries' coronavirus death rates, explained".
Vox. Retrieved 18 April 2020.
10 Panzeri, Stefano; Magri, Cesare; Carraro, Ludovico (1 September 2008). "Sampling bias". Scholarpedia. 3 (9):
4258. doi : 10.4249/scholarpedia.4258 . ISSN 1941-6016. Retrieved 19 April 2020.
11 "Amid a surge in COVID-19 cases, officials prioritize testing for health care workers, critically ill".
www.advisory.com. Retrieved 18 April 2020.
12 "Bund sucht nicht mehr alle Corona-Infizierten". Der Bund (in German). 6 March 2020. ISSN 0774-6156. Retrieved
19 April 2020.
13 Mary Van Beusekom | News Writer | CIDRAP News | 24 March 2020. "Italian doctors note high COVID-19 death
rate, urge action". CIDRAP. Retrieved 18 April 2020.
14 " Non-severe vs severe symptomatic COVID-19: 104 cases from the outbreak on the cruise ship “Diamond Princess”
in Japan". Sakiko Tabata, Kazuo Imai, Shuichi Kawano, Mayu Ikeda, Tatsuya Kodama, Kazuyasu Miyoshi,
Hirofumi Obinata, Satoshi Mimura, Tsutomu Kodera, Manabu Kitagaki, Michiya Sato, Satoshi Suzuki, Toshimitsu
Ito, Yasuhide Uwabe, Kaku Tamura, medRxiv preprint. 18 March 2020. Retrieved 19 April 2020.
15 CDC (2020-02-11). "Coronavirus Disease 2019 (COVID-19)". Centers for Disease Control and Prevention.
Retrieved 20 April 2020.
16 Ward, Dan. (2020). Sampling Bias: Explaining Wide Variations in COVID-19 Case Fatality Rates. doi:
10.13140/RG.2.2.24953.62564/1 Retrieved 19 April 2020.
17 Hayes, Adam. "R-Squared". Investopedia. Retrieved 23 April 2020.
18 "GDP by Country - Worldometer". www.worldometers.info. Retrieved 21 April 2020.
19 "Countries by Median Age 2018”. worldpopulationreview.com. Retrieved 21 April 2020.
20 "Population ages 65 and above (% of total population) | Data". data.worldbank.org. Retrieved 21 April 2020.
21 "What explains Covid-19's lethality for the elderly?". STAT. 2020-03-30. Retrieved 19 April 2020.
22 "COVID-19: First results of the voluntary screening in Iceland". Nordic Life Science—the leading Nordic life
science news service. 23 March 2020. Retrieved 18 April 2020.
23 “Germany’s virus response shines unforgiving light on Britain”. Financial Times. 3 April 2020. Retrieved 19 April
2020.
24 Normile, Dennis. "Coronavirus cases have dropped sharply in South Korea. What's the secret to its success?".
Science | AAAS.17 March 2020. Retrieved 18 April 2020.
25 "Why are Germany and Austria's coronavirus death rates so low?". www.newstatesman.com. Retrieved 18 April
11
References
2020.
26 "Virus means millions will be left in poverty". BBC News. 31 March 2020. Retrieved 19 April 2020.
27 Parker, George (4 April 2020) "Top UK adviser says mass tests key to ease lockdown" Financial Times. Retrieved 9
April 2020.
28 Verity, Robert; Okell, Lucy C.; Dorigatti, Ilaria; Winskill, Peter; Whittaker, Charles; Imai, Natsuko; Cuomo-
Dannenburg, Gina; Thompson, Hayley; Walker, Patrick G. T.; Fu, Han; Dighe, Amy (2020-03-30). "Estimates of the
severity of coronavirus disease 2019: a model-based analysis". The Lancet Infectious Diseases. 0 (0).
doi:10.1016/S1473-3099(20)30243-7. ISSN 1473-3099. PMID 32240634. Retrieved 20 April, 2020.
29 "Dutch study suggests 3% of population may have coronavirus antibodies". Reuters. 2020-04-16. Retrieved 20 April
2020.
30 "Ergebnisse der Heinsberg-Studie machen Hoffnung". www.t-online.de (in German). Retrieved 19 April 2020.
31 Giles, Chris (2020-04-22). "UK coronavirus deaths more than double official figure, according to FT study | Free to
read". www.ft.com. Retrieved 23 April 2020.
32 "Covid-19's death toll appears higher than official figures suggest". The Economist. ISSN 0013-0613. Retrieved 18
April 2020.
33 "Excess mortality caused by the novel coronavirus (COVID-19) | RIVM". www.rivm.nl. Retrieved 23 April 2020.
34 "Bund sucht nicht mehr alle Corona-Infizierten". Der Bund (in German). 6 March 2020. ISSN 0774-6156. Retrieved
18 April 2020.
35 "Report 13 - Estimating the number of infections and the impact of non-pharmaceutical interventions on COVID-19
in 11 European countries". Imperial College London. 30 March 2020. Retrieved 19 April 2020.
36 “Basic reproduction number”, ScienceDirect, Retrieved 18 April 2020.
37 "Statement on the meeting of the International Health Regulations (2005) Emergency Committee regarding the
outbreak of novel coronavirus 2019 (n-CoV) on 23 January 2020". www.who.int. Retrieved 15 April 2020.
38 Sanche, Steven; Lin, Yen Ting; Xu, Chonggang; Romero-Severson, Ethan; Hengartner, Nick; Ke, Ruian. "Early
Release - High Contagiousness and Rapid Spread of Severe Acute Respiratory Syndrome Coronavirus 2 - Volume
26, Number 7—April 2020 - Emerging Infectious Diseases journal - CDC". doi:10.3201/eid2607.200282. Retrieved
18 April 2020.
39 "Ergebnisse der Heinsberg-Studie machen Hoffnung". www.t-online.de (in German). Retrieved 18 April 2020.
40 "Burden of influenza". www.euro.who.int. Retrieved 18 April 2020.
41 Thomas Stephens. "Coronavirus: Head-shaking doctors and non-hand-shaking workers". SWI swissinfo.ch. 26
March 2020. Retrieved 19 April 2020.
42 Gould, Aylin Woodward, Skye. "One chart shows how many coronavirus tests per capita have been completed in 8
countries. The US is woefully behind". Business Insider. 10 March 2020. Retrieved 19 April 2020.
43 "Doctors Starting to Limit Who Gets Tested for COVID-19". Healthline. 24 March 2020. Retrieved 19 April 2020.
44 Giles, Chris (2020-04-22). "UK coronavirus deaths more than double official figure, according to FT study | Free to
read". www.ft.com. Retrieved 23 April 2020.
45 "Report 13 - Estimating the number of infections and the impact of non-pharmaceutical interventions on COVID-19
in 11 European countries". Imperial College London. 30 March 2020. Retrieved 19 April 2020.
46 “AGGIORNAMENTO 28/03/2020 ORE 17.00”, PCM DPC dati fornitidal Ministero della Salute. Italy. Retrieved
19 April 2020.
12
... There also seems to be a general trend of over-stating the role of age in explaining key aspects of COVID-19, given that high profile statements concerning the supposed role of age in explaining infection rates and estimated Case Fatality Rates have also been shown to be false. 19,20 It should also be noted that the extent of government actions cannot explain all of the variation in reported deaths, even in high income countries (see figure 2). A number of other factors, not analysed in the report, will likely explain the remainder of this variation. ...
... However, as regards the impact of GDP, this does not mean that poorer countries will necessarily be spared high mortality, as poorer countries also likely have weaker health systems and also worse population health, both of which will likely increase COVID-19 deaths. In addition, the response of countries to their own statistics of reported cases is problematic, as reported cases are substantially impacted by the amount of testing undertaken: countries that test more, relative to the number of deaths, report far more cases, probably because they are far better able to detect the majority of infected people who only have mild or no symptoms 26,19,20 . It is therefore likely that countries that tested less, detected less cases and thus responded slower to the pandemic, regardless of how many actual infections those countries in fact had. ...
Technical Report
Full-text available
Why do reported COVID-19 deaths vary so much across countries, and is population age really a plausible explanation? This report shows that variations in the extent to which countries responded to the pandemic can explain a lot of the variation in reported deaths, whereas population age cannot. Countries that enacted more stringent policies, at a relatively early stage in the pandemic (when each had one death per 8 million), went on to have a lot less reported deaths per million over the subsequent 6 weeks. Conversely, there was no significant correlation found between population age and deaths per million, across the 26 high income countries analysed, despite the fact that these countries varied substantially in percentage of population over 65 (i.e. 1% to 28%). This is likely to be because any effect from varying population age is being masked by much larger variations in the speed with which governments responded to the pandemic. It is thus concluded that population age cannot account for why some particular countries (such as Italy and Spain) have been worse hit than others by the pandemic, and that having younger populations will not necessarily protect countries against high numbers of deaths. All countries, regardless of population age, thus need to respond appropriately and rapidly to the pandemic. It is also concluded that some experts and publications seem to have over-stated the role of population age in explaining various aspects of the pandemic. The analysis was limited by the availability, accuracy and consistency of official reports.
... 9 This has likely occurred in many countries with respect to COVID-19, and variations in sampling bias have already been shown to account for much of the wide variation in Case Fatality Rates across countries. 8,10 Most countries have only tested a small minority of their populations to date (4% on average, for the countries on figure 1), and many tests that have been done, have been prioritised on the seriously ill 11 , due to the clinical priority of diagnosing such individuals (so they can be treated appropriately). Moreover, some countries have had official policies not to test those with only mild or no symptoms 12,13 , whilst it is already wellestablished that most people infected with COVID-19 do not get seriously ill 14,15 . ...
... Moreover, some countries have had official policies not to test those with only mild or no symptoms 12,13 , whilst it is already wellestablished that most people infected with COVID-19 do not get seriously ill 14,15 . The resulting sampling bias has likely led to significant under-estimations of cases in many countries 10 . Moreover, given that it is already well-established that younger people are much less likely than elderly people to become seriously ill if infected 16 , it also seems very likely that sampling bias in testing has also significantly impacted the age distribution of reported cases 17 , which is the hypothesis of this report. ...
Technical Report
Full-text available
Why does the age distribution of COVID-19 cases vary so much, and are the elderly really more likely to become infected? This report shows that countries that test more, relative to the number of deaths, have a much higher proportion of young (20-29 year old) cases, and a much lower proportion of elderly (70 years and over) cases, relative to the wider population. Moreover, these countries will likely have much more accurate statistics, because they have much less sampling bias in their testing. These findings thus imply that many countries, and especially those with low tests/deaths ratios, have substantially underestimated the number of young adults infected, in particular. Conversely, the disproportionately high percentage of elderly cases in countries with low tests/deaths ratios is likely to be mainly just an artefact of too little testing in these countries, and is not indicative of a disproportionately high number of actual infections in this age group: i.e. elderly people are not more likely to be infected, although they are more likely to become seriously ill, if infected. It is thus concluded that all countries, and particularly those with low tests/deaths ratios – such as the UK, Algeria, France, Sweden, the Netherlands, Mexico and Brazil – need to do massively more testing urgently, especially of young adults. Any young adults working with the elderly, in particular, need to be tested at the highest priority, in all countries, as this may substantially reduce mortality. Children also need to be specifically tested in all countries because, as they are more likely to be asymptomatic, they are much less likely to have been tested so far. There is thus insufficient information to date to determine infection rates amongst children.
... (The use of COVID-19 CFR has, in any event, been questioned because it is highly sensitive to the number of tests carried out resulting in significant sampling error. [36] Other vaccinations, including Hib, diphtheria-tetanuspertussis (DPT), measles-containing vaccines (MCV), BCG (tuberculosis), and oral polio vaccine did not demonstrate any significant correlation, positive or negative, with COVID-19 case rates or rates of death. The lack of correlations with both Bacillus Calmette-Guerin (BCG) and oral polio vaccine are notable since several research groups are currently investigating the use of these vaccines as means to prevent COVID-19 infection. ...
Article
Full-text available
Two conundrums puzzle COVID‐19 investigators: 1) morbidity and mortality is rare among infants and young children and 2) rates of morbidity and mortality exhibit large variances across nations, locales, and even within cities. It is found that the higher the rate of pneumococcal vaccination in a nation (or city) the lower the COVID‐19 morbidity and mortality. Vaccination rates with Bacillus Calmette–Guerin, poliovirus, and other vaccines do not correlate with COVID‐19 risks, nor do COVID‐19 case or death rates correlate with number of people in the population with diabetes, obesity, or adults over 65. Infant protection may be due to maternal antibodies and antiviral proteins in milk such as lactoferrin that are known to protect against coronavirus infections. Subsequent protection might then be conferred (and correlate with) rates of Haemophilus influenzae type B (Hib) (universal in infants) and pneumococcal vaccination, the latter varying widely by geography among infants, at‐risk adults, and the elderly. A strong inverse correlation exists between decreased case and death rates from COVID‐19 and increased rates of pneumococcal vaccinations among children and adults. These correlations are independent of percentage of the population over the age of 65, diabetic, and/or obese. No significant correlations are found between COVID‐19 case or death rates and Bacillus Calmette–Guerin, polio, measles‐mumps‐rubella, or diphtheria‐tetanus‐pertussis vaccination rates.
... (The use of COVID-19 CFR has, in any event, been questioned because it is highly sensitive to the number of tests carried out resulting in significant sampling error. [36] Other vaccinations, including Hib, diphtheria-tetanuspertussis (DPT), measles-containing vaccines (MCV), BCG (tuberculosis), and oral polio vaccine did not demonstrate any significant correlation, positive or negative, with COVID-19 case rates or rates of death. The lack of correlations with both Bacillus Calmette-Guerin (BCG) and oral polio vaccine are notable since several research groups are currently investigating the use of these vaccines as means to prevent COVID-19 infection. ...
Preprint
Full-text available
Various studies indicate that vaccination, especially with pneumococcal vaccines, protects against symptomatic cases of SARS-CoV-2 infection and death. This paper explores the possibility that pneumococcal vaccines in particular, but perhaps other vaccines as well, contain antigens that might be cross-reactive with SARS-CoV-2 antigens. Comparison of the glycosylation structures of SARS-CoV-2 with the polysaccharide structures of pneumococcal vaccines yielded no obvious similarities. However, while pneumococcal vaccines are primarily composed of capsular polysaccharides, some are conjugated to CRM197, a modified diphtheria toxin, and all contain about three percent protein contaminants, including the pneumococcal surface proteins PsaA, PspA and probably PspC. All of these proteins have very high degrees of similarity, using very stringent criteria, with several SARS-CoV-2 proteins including the spike protein, membrane protein and replicase 1a. CRM197 is also present in Hib and meningitis vaccines. Equivalent similarities were found at statistically significantly lower rates, or were completely absent, among the proteins in diphtheria, tetanus, pertussis, measles, mumps, rubella, and poliovirus vaccines. Notably, PspA and PspC are highly antigenic and new pneumococcal vaccines based on them are currently in human clinical trials so that their effectiveness against SARS-CoV-2 disease is easily testable.
Article
Full-text available
Governments around the world must rapidly mobilize and make difficult policy decisions to mitigate the coronavirus disease 2019 (COVID-19) pandemic. Because deaths have been concentrated at older ages, we highlight the important role of demography, particularly, how the age structure of a population may help explain differences in fatality rates across countries and how transmission unfolds. We examine the role of age structure in deaths thus far in Italy and South Korea and illustrate how the pandemic could unfold in populations with similar population sizes but different age structures, showing a dramatically higher burden of mortality in countries with older versus younger populations. This powerful interaction of demography and current age-specific mortality for COVID-19 suggests that social distancing and other policies to slow transmission should consider the age composition of local and national contexts as well as intergenerational interactions. We also call for countries to provide case and fatality data disaggregated by age and sex to improve real-time targeted forecasting of hospitalization and critical care needs.
Technical Report
Full-text available
(This technical report is part of a series of reports on sampling bias in COVID-19 testing. Please also see the subsequent report: "Sampling Bias, Case Fatality Rates and COVID-19 Testing: A Further Analysis", also by Dan Ward). This statistical report seeks to explain the wide variation in currently estimated Case Fatality Rates (CFRs) for COVID-19 (coronavirus) across countries. Based on a statistical analysis of 20 countries, it was found that countries that have tested more, relative to the number of deaths, have significantly lower CFR estimates. This is likely to be because these countries are able to detect more people with only mild or no symptoms. Moreover, as these countries are testing more widely, their CFR estimates are also likely to be more reliable, as they are subject to less sampling bias. It was also found, based on an analysis of 60 countries, that countries with higher GDP per capita also have significantly lower CFR estimates. This may be because richer countries are likely to have better health systems, and are thus better able to detect and treat COVID-19 cases. Finally, median age of populations was found not to be currently having a significant impact on CFR estimates across countries, probably because any effect from varying age is being masked by much larger variations in the amount of testing. Key conclusions are that the underlying Infection Fatality Rate (IFR) for COVID-19 will likely be at the lower end of the analysed range (i.e. 0.25%, not 10.1%), and that far more people are likely to have been infected than officially reported numbers of cases suggest: 26 million globally by 9 April 2020, not 1.4 million. It is also concluded that the basic reproduction number (R0) will likely be higher than initially estimated, as many R0 estimates may have been based on under-estimates of numbers of infections. In addition, all countries need to do more testing, especially of those with only mild or no symptoms, as only then will it be possible to detect and quarantine most infected people, and trace and test contacts. Finally, special help needs to be given, in particular, to poorer countries. Many other factors may also be affecting estimated CFRs than were analysed, and the IFR could thus still turn out to be higher than 0.25%. In particular, some under-reporting of deaths may be occurring, possibly by a factor of two. It is thus likely that the IFR will be in the range 0.25% - 0.50%. There is also no room for complacency, as many people are still at risk of dying, even if the IFR is 0.25%, because COVID-19 does seem to be very infectious. The analysis was limited by the availability, accuracy and consistency of official reports.
Italian doctors note high COVID-19 death rate, urge action
  • Mary Van Beusekom
Mary Van Beusekom | News Writer | CIDRAP News |; 24 March 2020. "Italian doctors note high COVID-19 death rate, urge action". CIDRAP. Retrieved 18 April 2020.
Smoking Upregulates Angiotensin-Converting Enzyme-2 Receptor: A Potential Adhesion Site for Novel Coronavirus SARS-CoV-2 (Covid-19)
  • Sukhwinder Sohal
  • Singh
Sohal, Sukhwinder Singh (2020/3). "Smoking Upregulates Angiotensin-Converting Enzyme-2 Receptor: A Potential Adhesion Site for Novel Coronavirus SARS-CoV-2 (Covid-19)". Journal of Clinical Medicine. 9 (3): 841. doi:10.3390/jcm9030841. Retrieved 19 April 2020.
Demographic science aids in understanding the spread and fatality rates of COVID-19
  • Melinda C Mills
Mills, Melinda C. (2020-04-16). "Demographic science aids in understanding the spread and fatality rates of COVID-19". Proceedings of the National Academy of Sciences. doi:10.1073/pnas.2004911117. ISSN 0027-8424. PMID 32300018. Retrieved 19 April 2020.
Non-severe vs severe symptomatic COVID-19: 104 cases from the outbreak on the cruise ship "Diamond Princess" in Japan
  • Sakiko Tabata
  • Kazuo Imai
  • Shuichi Kawano
  • Mayu Ikeda
  • Tatsuya Kodama
  • Kazuyasu Miyoshi
  • Hirofumi Obinata
  • Satoshi Mimura
  • Tsutomu Kodera
  • Manabu Kitagaki
  • Michiya Sato
  • Satoshi Suzuki
  • Toshimitsu Ito
  • Yasuhide Uwabe
  • Kaku Tamura
" Non-severe vs severe symptomatic COVID-19: 104 cases from the outbreak on the cruise ship "Diamond Princess" in Japan". Sakiko Tabata, Kazuo Imai, Shuichi Kawano, Mayu Ikeda, Tatsuya Kodama, Kazuyasu Miyoshi, Hirofumi Obinata, Satoshi Mimura, Tsutomu Kodera, Manabu Kitagaki, Michiya Sato, Satoshi Suzuki, Toshimitsu Ito, Yasuhide Uwabe, Kaku Tamura, medRxiv preprint. 18 March 2020. Retrieved 19 April 2020.
Top UK adviser says mass tests key to ease lockdown
  • George Parker
Parker, George (4 April 2020) "Top UK adviser says mass tests key to ease lockdown" Financial Times. Retrieved 9 April 2020.
Coronavirus: Head-shaking doctors and non-hand-shaking workers
  • Thomas Stephens
Thomas Stephens. "Coronavirus: Head-shaking doctors and non-hand-shaking workers". SWI swissinfo.ch. 26 March 2020. Retrieved 19 April 2020.
One chart shows how many coronavirus tests per capita have been completed in 8 countries. The US is woefully behind
  • Aylin Gould
  • Skye Woodward
Gould, Aylin Woodward, Skye. "One chart shows how many coronavirus tests per capita have been completed in 8 countries. The US is woefully behind". Business Insider. 10 March 2020. Retrieved 19 April 2020.