UPDATED! April 3, 2020. The coronavirus mortality rate varies significantly by country. In this post, I look at the mortality rates for ten countries and assess factors that affect these numbers. After discussing the trends, I provide a rough estimate for where the actual fatality rate might lie.
To start, take a look at the current mortality rates for ten countries and the worldwide rate. These countries are the same ones I’ve been tracking in my article about coronavirus curves and different outcomes.
That’s a large range of values! However, we can learn even more when we look at these numbers over time.
This chart is a bit messier than my graph of confirmed cases. Don’t worry. I’ll cover what to look for below!
I start each country on the graph at the point when they have at least 20 cases—just as I do for my chart of confirmed cases. I deleted 0% values that occur very early to declutter the graph. The X-axis indicates the number of days since 20 cases. The Y-axis is the mortality rate as a percentage. The formula for calculating mortality rates is the following:
Johns Hopkins provided the data and they are current up to March 27, 2020.
How Do We Interpret these Mortality Rates?
What do these numbers tell us? Here are some important caveats!
These data reflect a sampling of individuals who sought testing, tested positive, and then subsequently died. That’s a very specific group of people. Consequently, mortality rates calculated from these data represent the percentage of individuals who died after testing positive.
That’s not quite what we want to know. We’d really like to know the overall mortality rate in the population—which brings us to the statistical concept of generalizability. Can we take the results of this sample and generalize beyond it?
When scientists conduct studies, their goal isn’t to understand just a particular group of participants. For example, in clinical trials for a new medication, scientists want to understand its effectiveness in the general population rather than just the relatively small sample of study subjects. In other words, they want to generalize the results from the sample to the population.
In statistics, generalizability depends on certain conditions. The most commonly known of these conditions is random sampling. However, there are various other conditions, which I discuss in detail in my Introduction to Statistics ebook.
The conditions under which health care workers collected the coronavirus data satisfy generalizability requirements to varying degrees in different countries. Consequently, countries with similar circumstances can appear to have different mortality rates. While these numbers are not perfect, we can learn a lot from them and even derive a rough estimate of the actual mortality rate.
Sampling Problems and Selection Bias in the Mortality Data
I’m sure you’ve heard there has been inadequate testing in various countries. However, this problem doesn’t only cause underreporting. It also biases the results. Individuals seeking testing are nowhere near to being a random sample that reflects the larger population.
When a disease is new and tests are scarce, there is no routine testing of individuals. Instead, severely ill people are likely to show up at a hospital seeking a coronavirus test. In more severe conditions, hospitals won’t even test all incoming patients, but only a subset of those with the most critical symptoms. Meanwhile, those with milder conditions are not tested.
Under these conditions, the mortality rate among those who are tested is higher than those who are not tested. For the fatality rate, the number of confirmed cases in the denominator does not include milder cases. Consequently, many of these rates calculated from the available data are higher than the real value for a general population that includes milder cases. This positive bias means we can’t always generalize from the sample data (tested individuals) to the general population.
As the scope of testing increases, countries can include milder cases, which lowers the mortality rate. The graph shows falling mortality rates for both South Korea and the United States, and increased testing is a likely reason.
On the other hand, Germany has consistently tested individuals both with severe symptoms and those with milder cases, which introduces less bias. On the graph, Germany has a relatively low mortality rate, which has increased from 0.17% to 0.69%. Germany might provide a good indication of the true mortality rate for coronavirus.
Similarly, South Korea has had large-scale testing very early on. Their mortality rate ranges from 0.46% to 1.38%. Interestingly, it started at 1% on February 6th, decreased to a minimum of 0.46% on March 1st, and has since increased to 1.52%.
Overloaded Health Care Systems
In other countries, overwhelmed health care systems can increase the coronavirus mortality rate. In these countries, coronavirus patients are more likely to die because the inundated systems don’t have the resources to provide lifesaving care. For example, there can be shortages of ICU beds and ventilators, among other critical resources.
Shortages in medical care likely explain the sharply increasing mortality rates in both Italy and Spain.
The early rates in China were similarly caused by the inability of hospitals to handle the high rate of patients. I’ve read reports that the initial Chinese rate was 4%. However, I don’t have data for those early days. My data begin on January 22, 2020. Interestingly, like South Korea, the Chinese mortality rate shows a similar pattern of an initial decrease followed by a rise. China had a 3.1% mortality rate on January 22nd. That rate decreased to a minimum of 2.05% on February 5th, after which it increased up to 4.03%.
Undoubtedly, both overwhelmed hospitals and selection bias play a role in the high rates for Italy, Spain, and China.
Related post: Coronavirus: Exponential Growth and Hospital Beds
Coronavirus Mortality Rates and Population Characteristics
The characteristics of the infected population can affect mortality rates. For example, the first cases in Washington were among the elderly in nursing homes, which raised the fatality rate. As the coronavirus spread to younger populations, that rate has fallen.
As the virus moves from between various subpopulations with a country, it can affect the overall mortality rate. Perhaps this explains the mortality rate curves on the graph that change directions?
Mysteries and the True Coronavirus Mortality Rate
As you can see, a variety of factors affect the mortality rate within countries. Additionally, we can’t always generalize the mortality rates from the sample data to the general population. These data aren’t perfect and it’ll take much more study to determine the real coronavirus mortality rate. However, we are able to glean significant trends and information from these data.
A mystery for me are countries like Japan, South Korea, and China that had a declining mortality rate, reached a minimum, and then started a systematic increase. Perhaps that reflects the virus spreading to more vulnerable groups?
How do we estimate the coronavirus mortality rate that we’d see when the data aren’t biased and health care systems are not overwhelmed? What’s the rate we’d calculate if we could include all coronavirus cases in our dataset—from the very mild to extremely severe? We can get some hints about that from the graph!
The United States had inadequate testing and is beginning to experience health care shortages, and yet its rate had gone down to 1.2%. Let’s call that a good upper-limit for now. Additionally, countries that have had widespread testing and no overburdening of health care systems have sustained rates that fall between 0.5% and 1%. Let’s consider that a most likely range. Finally, and tantalizingly, Singapore has a tiny mortality rate that is currently at 0.27%. For now, let’s say that this is the lower-bound.
Based on these data, a rough estimate is that the coronavirus mortality rate falls between 0.3% and 1.2%, with it more likely falling between 0.5% and 1%. If it falls within the narrower range, that’s 5 to 10 times more fatal than the flu! However, these ranges are educated guesses. Scientists will require much better data to nail it down!