COVID19: Let’s Talk About Death (Rates, That Is)

by Raywat Deonandan, PhD
Epidemiologist & Associate Professor
University of Ottawa
(I add my credentials to these COVID-19 blog posts in case they get shared. I want readers to know that my opinion is supposedly an educated and informed one)

One of the more frequent words being thrown about by journalists, politicians, and the public is “mortality”, especially as it pertains to the current COVID19 assault on our society. Sometimes they use “deadliness” or “lethality”, and often it’s in the context of comparing COVID19 to diseases that seem to the untrained eye to be similar, like seasonal influenza.

But there are actually many different types of mortality rates, epidemiologically speaking. And that will be the topic of today’s post, as I also weigh in on the COVID19 modelling that was recently released by the Canadian federal government.

But First…

Before we get there, this is still a personal blog and not an op-ed factory, so let’s get some housekeeping done.

Since last we spoke, I’ve been in the media a few more times talking about, you guessed it, COVID19:

By the way, I keep all this stuff here not to celebrate my ego, but so I have an archive for when I use it all to apply for promotion to full professor one day 🙂

Okay, Back to Dead People


The simplest measure of mortality is the “crude death rate” or CDR.  Technically, it’s “crude death RATIO”, since a rate must have “time” as its denominator. But do we really care?  Well, some do. (I don’t.)

The CDR is essentially the number of people who died in a given region over a set duration divided by the total number of people who lived in that region in the same time period. It doesn’t measure what they died from (cancer, injury, old age, etc), just that they died.

The CIA World Factbook estimates the CDR for the world to be 7.7 per 1000 people, or almost two deaths every second. (Which reminds me of this joke, which is surely apocryphal.)

It may surprise many of you to know that the global CDR has been dropping fast for decades, as per this line graph:

Changes in global CDR (source: World Bank)


It’s one of many signs that the world is actually on a positive health and development trajectory, which is a controversial and triggering statement for many. But let’s not get into that here today.

Cause-Specific Date Rate

Most of you reading this won’t care about the CDR, since it measures all causes of death. But you might have an interest in the cause-specific death rate (or ratio), which is exactly what it sounds like: the number of people who died of a specific cause divided by the total number of people living in the population.

Now, to be really specific –because I know there are some retentive nerds out there who will send me angry emails– the base population changes size. People migrate in and out, and some die. So over the period that that’we looking at, the denominator is actually the value at the middle of the interval. So if we’re looking at the CDR or cause-specific death rate of the USA in the year 2016, we would probably use the USA population at the end of June of that year, since that’s midway through the year.

Proportionate Mortality Rate

Now what if we looked only at the people who died and asked ourselves how many of those people died from the causes we care about? Then we’re asking what proportion of deaths are caused by that thing…. or a proportionate mortality rate (or ratio), or PMR.

So imagine that a community has 1000 people.  In one year, 100 of those people die. Eighty of those deaths were from heart attacks, and twenty were from COVID19.

The CDR would be 100/1000 people, or 10%. (The nerds like to give it as a rate per 1000 people, but I think percentages are just fine.)

The cause-specific mortality for COVID19 would be 20 deaths /1000 people, or 2%.

And the PMR for COVID19 would be 20 deaths /100 dead people, or 20%.

Case Fatality Rate 

The case-fatality rate, or CFR, or sometimes “death-to-case ratio”, is the thing that everyone really cares about right now.  It answers the question, “If I get this disease, what is the probability I’m going to die?” Technically, it’s a measurement of risk. But that’s a touch of sophistry that few would appreciate.

Just so the nerds don’t get mad at me, I’m going to get a little Byzantine for a moment. Technically, a CFR is neither a ratio nor a rate. It’s a proportion of incidence. So it’s always given as a percentage. So I don’t want to get any annoying comments insisting that I’ve used the wrong terminology…. I know. I’m trying to make this as simple as possible. And clearly, I’ve already failed.

The CFR is computed by looking at the number of deaths attributed to that disease in a given period, and dividing that number by all the people newly diagnosed with the disease over that same period –the incidence rate. We don’t care if those people recovered or died, only that they were newly diagnosed. (Actually, we do care, because we’re not heartless monsters. But math is a psychopath.)

So in Ontario this week, if 100 people were diagnosed with COVID19, and 3 people died of it during the same week, then the CFR is 3/100 or 3%.

Infection Fatality Rate

The infection fatality rate, or IFR is specific to infectious disease outbreaks. Clearly, COVID19 fits that bill. But diabetes or heart attacks do not.

The IFR is actually the real CFR. It attempts to measure the lethality of the infection in all cases, not the just the ones we know about. Ultimately, the IFR is what we want to end up knowing for population health purposes, and we will not really know it until all actual cases are counted.

The IFR for the 1957 USA flu epidemic was about 0.27%, and for the legendary 1918 flu it was 2.6%. We won’t know what it is for COVID19 until all is said and done. Until then, we rely on the CFR to estimate the IFR. As you’ll see below, the current CFR estimates for COVID19 are well above the IFRs for the diseases mentioned above…. so far.


Let’s Talk More About the CFR

Okay, so the IFR is ultimately what we really care about. But frankly, no one ever talks about that. So let’s focus on the CFR and pretend that is will ultimately allow us to estimate the IFR. That means that what we care about is getting an accurate and true picture of the CFR. Or, in other words: If I get this thing, am I going to die?

CFR changes over time a lot. It changes due to at least four factors, possibly more: (1) changes or mutations in the disease itself; (2) changes in our ability to treat the disease; (3) characteristics of the population; and (4) improved accuracy in measuring the denominator.

There is also something to be said for data quality. So let’s get to each of those in turn, as they pertain to our understanding of COVID19.

Mutations of the virus

Viruses mutate. That’s a truth of the world. The rate of mutation depends upon the number of generations of new viruses produced (more generations, more opportunities for mutation and natural selection), and presumably upon environmental factors, like radiation.

SARS-Cov-2 is mutating, no doubt. But so far it’s not mutating so fast that it’s noticeably changing how it affects its human victims, nor how our immune systems react to it.  This is largely a good thing, since a homogeneous immune response likely means we can make a single vaccine to  attack all the mutated strains.

And as I mentioned in this post, it’s possible that this virus will mutate toward lesser lethality, causing the CFR to drop. But don’t count on it in the near term.

So I don’t think mutation is a big cause of differences in CFR observed around the world.

Our Ability to Treat the Disease

Ebola used to have one of the highest CFRs. Around 90% of people who caught it would die. But after the 2014 outbreak in West Africa, it dropped to about 50%. That is partly due to some data collection and coding issues, but also due to scientists and doctors learning how to treat it better.

Similarly, every day our heroic clinicians and lab scientists are making great strides in improving the prognosis of people afflicted with COVID19. New ventilator protocols, new treatments, and new early treatment options are all showing moderate improvements in patient outcome. We’re even learning how to have patients lie down properly in order to improve their breathing, lessening their need for ventilators.

It’s not surprising that the CFR has been lowering around the world.

Here are some quick back-of-the-envelope CFR calculations based on COVID19 data from OurWorldInData, accessed April 20, 2020:

China: 3340  deaths / 82925 cases = 4.0%
Italy: 18281 deaths /143626 cases = 12.7%
Germany: 2767 deaths /121842 cases = 2.2%
Spain: 15238 deaths / 152446 cases = 10.0%
USA: 16690 deaths / 466033 cases = 3.6%
Canada: 557 deaths / 22059 cases = 2.5%
World: 102026  deaths / cases 1681000 = 6.1%

MotherJones found similar data, which they reproduced in this graph:

CFRs from around the world


Some countries clearly have better CFRs than others. But they’re all experiencing the same disease. Surely it can’t be more deadly in one country than another. People are people, right?

One reason for the improved CFR in the USA and Canada is that the epidemic came here later, so we had chance to learn from the harrowed experiences of Chinese and Italian doctors.

The low numbers in Germany might have to do with their health care system responsiveness, but probably has more to do with their better grasp of their denominator (see below).

Population Factors

Sometimes CFR can change from population to population because those populations are innately different.  We know that COVID19 is more likely to kill older people, for example. So which populations have more old people?

Compare the age pyramids of Italy and Canada:

Italy age pyramid 2018 (source: IndexMundi)


Canada age pyramid 2018 (source: IndexMundi)

Hopefully you can see that Italy has a higher and fatter bulge, meaning that its population very slightly skews older than Canada’s. This is a small thing, but it might have an effect on the numbers of older people getting sick, and therefore dying.

Other population factors include underlying comorbidites and behaviours: smoking, obesity, and diabetes, for example. Countries that have more of that will have higher CFRs, in theory.

But the biggest source of bias in the CFR is….

Measuring the Denominator

Remember that CFR is the number of deaths divided by the number of new cases. In rich countries, we pretty much know almost all of the deaths. Sure, some institutions might be opting for palliation rather than ICU care, and maybe those deaths slip through the data fingers. But for the most part, almost all COVID19 deaths are known and counted.

(This is more true in Canada than in the USA. Lack of socialized health care might compel many to avoid seeking care and might result in some uncounted deaths. More on that later.)

The same confidence in data cannot be expressed for the detection of COVID19 cases. And this is where testing is the issue.

Currently, in almost all countries, we test on a clinical basis: those who present with symptoms are tested so that we know what to do with them. We are not testing on a surveillance basis. This means we really don’t have a good grasp of the number of people who are actually infected; plenty of mild or symptomatic cases are being missed. I tend to look at the case estimates and multiply that by a factor to get a better grasp of the real case burden in the community.

The more we test, the more we will discover the many thousands of mild cases who recovered at home and who were never counted. That’s why the more we test, the more our case counts will go up and our CFR will go down.

That’s an artifact of data collection and does not represent a change on the ground. So as our testing ramps up in coming weeks, please do not be alarmed by the sudden changes in the numbers. We call this a “detection bias.”

But changes in CFR have implications for our projections of the terror the disease will wreak across our society.

This is where some new data are very exciting. As I’ve noted many times, Iceland is engaging in random screening, which allows them to get a good impression of their population prevalence.

Now Germany has entered the fray with a vengeance. They tested 80% of a certain region of Germany to determine true seroprevalence. When plugged into their CFR formula, they were able to compute a more accurate predictor of the true IFR for COVID19.

Their estimate for this true IFR is 0.37%. Stunning, huh?

Meanwhile, a Japanese study used some fancy statistical methods to model a true IFR.  Are you ready for this number?  It’s 0.04%.

Keep in mind that this is pre-peer reviewed, so take that for what it’s worth. I’m sure that number will change. What is certain, though, is that the true IFR in Canada and the USA is most certainly substantially lower than the currently reported CFRs of 2-4%.

So that is good news.  As testing results come in, the CFR will drop, and fast.

So All Those People Saying It’s Not As Bad as the Flu Were Right?

Hells to the no.

I personally don’t trust the Japanese estimate. My feeling has always been that when this is said and done, the IFR for COVID19 will be found to be 0.3-0.5%.  This compares to about 0.1% for the seasonal flu, so it’s at least three times more lethal than the flu.

The big issue with COVID19 has never been its lethality, per se, but its infectiousness. We like to use the term “R naught” which is R with a subscript 0 (I don’t know how to do that in WordPress) to describe how infectious a disease is.

R naught is essentially how many people a single infected person will infect for the duration of his infectiousness. The Measles R naught is anywhere from 3 to 200. The seasonal flu is around 1.3. And for this disease, it’s between 2 and 3.

Several factors affect the R naught. The one we can control is contact rate, which is why “social distancing” is such a powerful tool right now. It’s one of the few things we can cheaply, easily, and quickly do to shrink the R naught and slow the spread.

So to recap: the seasonal flu is at least a third as deadly as COVID19 (using the most forgiving estimate of the latter’s likely IFR) and it’s half as infectious. Those two factors alone mean that COVID19 will blaze through the population and kill a lot more people than the flu.

Numerator Issues

I mentioned that we mostly know of all the deaths. That’s not entirely true. Places that embrace palliation might have COVID19 deaths that we don’t pick up in the official count.

In addition, some forensic statistics are suggesting that some places, like New York, might be under-counting the actual COVID19 deaths, as per this graph from the New York Times:

All cause mortality in New York

More controversially is identifying which deaths are truly COVID19 deaths. This is a standard issue in data coding in population epidemiology. In the early 1980s, the number of HIV/AIDS deaths were vastly underestimated because people don’t die from AIDS, they die from “opportunistic infections”: fungal infections, TB, pneumonia, etc.

When coroners’ reports in the USA were modified to include complications by HIV/AIDS, only then were we able to see the true deadliness of that disease.

With COVID19, accusations have come from some political corners that doctors are “lying” about the COVID19 deaths because they are reportedly coding all hospital deaths as COVID19 deaths if the patient was indeed infected.

It’s not “lying”. It’s a basic artifact of data coding in the throes of the chaos of a pandemic. I guarantee you that we are not counting as COVID19 deaths any cases who died in car accidents or who were murdered. The quadruple-bypass patient on a ventilator for his COVID19 infection who succumbed to a heart attack is fair game for coding as a COVID death, as it is highly probable that the disease accelerated his demise.

These coding practices will vary from jurisdiction to jurisdiction, further complicating our ability to aggregate data from multiple sources.


What Does This Mean For Canada’s Model

First off, the model presented by the government earlier this week was for the general public, not scientists. It didn’t present a lot of nuance, or even numbers on the vertical axis in one of the charts.

Rather, its intent was to explain to the citizens why we need to be socially restricted for a weeks or months.

But I draw your attention to this table, which has been a terrifying bit of data for a number of journalists who’ve been asking me to comment:

COVID19 projections for Canada

Where did these numbers come from? Well, if the expected prevalence of the disease is 2.5% (at the low end for this chart), that means that 2.5% of our total population of 37.4 million gives us a case load of about 934000 infected people. And applying a CFR of 1.17% (which is what was being reported for Canada for the days prior to this report’s publication) we get about 11000 deaths.

It ain’t rocket science or complicated dynamic modelling that got us there.

While 2.5% prevalence is the medium estimate for hard public health controls, the 11000 deaths are still scary. But only if that CFR is accurate. And I don’t think that it is. A more realistic CFR of 0.5% brings the death toll below 5000, which is not quite as scary. And that’s over the duration of the whole pandemic, which will likely last two years.

So for this reason and others, I have a more optimistic view than the official model presents. But do keep in mind that many experts would argue that we shouldn’t be looking at the numbers anyway, just the trend lines. The takeaway message still holds: what we’re doing right now is saving a lot of lives that would have been lost had we done nothing.

And that’s really all that matters. That’s why we measure mortality: to figure out ways of reducing it!