COVID-19 mortality, that is, the likelihood of dying, is of vital interest to anyone who contracts this disease. This blog quantifies what we know, and explains why things are, yes, getting better. This is the latest in a series of blogs discussing COVID-19.
As usual, you may jump ahead to major sections with these links:
– A Consensus Picture of COVID-19 Mortality
– A Deeper Dive Into COVID-19 Mortality
Although COVID-19 cases are ballooning since the recent holidays, deaths have not increased to the same degree. In other words, the mortality rate of the virus seems to be declining, at least in the US.
It’s difficult to find data that quantifies COVID-19 mortality, that is, what percentage of people who get infected then die from the infection. One thing that is certain: social media posts quoting survival rates of 99.9% are fake and FALSE (for example, see Politifact and FullFact).
Experts give many reasons why COVID-19 mortality is difficult to calculate. They particularly blame inconsistencies in “when-where-who” gets tested, gets hospitalized, or gets reported as a COVID-related death.
Art Attempts Analysis
Finding mortality data so scarce, I made a modest (ahem!) effort to analyze the CDC total US data to compute COVID-19 mortality. I felt that if I took such a large sample, inconsistencies in the data might average out and allow us to identify big trends with a bit of confidence.
The CDC data tracker allows downloading daily infections and daily deaths for the entire pandemic. They also provide 7-day averages to smooth out the fluctuations.
Infections precede deaths of course, so I correlated the sequence of cases and the sequence of deaths with varying delays from cases to deaths. I found a maximum correlation of 79% with a 9 day offset.
Then came a surprise: I calculated COVID-19 mortality as week-average daily deaths divided by week-average daily new cases from 9 days earlier. I found that mortality rates started out high and dropped rapidly until about mid-July, after which they leveled out at a low level. Something changed substantially, mid-year.
– COVID-19 Mortality Before and After Mid-Year
I repeated the correlations but this time treated dates before and after July 15 as separate sets. I found, aha!, that the correlations peaked with a 6 day delay before mid-July, and a 26-day delay after that time. Moreover, the correlations were a very high: 95% for each of these time intervals!
What’s going on, anyway? This is what I suggest: Early in the pandemic we had very little virus testing. The only way we learned that someone had coronavirus was when they developed symptoms, generally severe symptoms. So their disease was pretty advanced by the time it was recognized, and those serious cases died pretty quickly.
However, later in the year, testing was available, not only for people with symptoms but also those who had been exposed to a probable COVID case. This allowed doctors to identify cases earlier and treat them before they became serious. Those who advanced to a serious disease died relatively later, 26 days according to the correlations.
A Consensus Picture of COVID-19 Mortality
Armed with this knowledge, I re-computed the mortalities with both 6 and 26 day lags. Because of the significant differences before and after mid-year, I defined a Consensus mortality using the 6-day delay prior to July, the 26-day delay after July, and the average of the two for July. Here are the results:
I’ll be the first to admit that this calculation is not “good science.” This is the work of a scientist trying to understand contradictory and uncertain data to glean some understanding of trends. However, since the UK science advisor is quoting a death rate of 1.0 to 1.4% for a 60-year old infected person, these COVID-19 mortality numbers are certainly in the right ballpark.
– Insights from COVID-19 Mortality Data
This analysis, rough though it is, offers some useful and hopeful information:
- In the first few months of the pandemic, if you were diagnosed with coronavirus, you had a 4 to 8% of dying from it!
- Most recently, your odds of dying have declined are now around 1.5%. In round numbers, this is a 4 to 1 improvement in mortality. We’ll see later that this may be due to testing that identifies many mild infections that not diagnosed early in 2020.
- Since October we have seen a huge increase in daily cases, likely due to holiday travel and gatherings. However, our caregivers are more efficient at treating patients, so the mortality is still only 1.5%.
– Why Things Are Getting Better
Why is the survival percentage so much better now than in the early days of the pandemic? Here are some reasons:
- Many therapies have been tried for all stages of the coronavirus disease. Enough of them have been successful that comprehensive treatment protocols are available to physicians such as those provided by the Front Line COVID-19 Critical Care Alliance and by the National Institutes of Health. Besides recommending “what,” these guidelines also suggest “when”: for example, permitting some patient hypoxia without rushing them into intubation.
- Coronavirus testing is more widely available, so that doctors can treat and isolate patients early in their disease.
- Treating patients early reduces the number of victims who progress to serious cases requiring hospitalization. And drugs such as dexamethasone significantly reduce mortality for patients with more advanced disease.
- Hospital procedures such as “prone positioning” have helped patients breathe better without assistance, reducing the demand for mechanical ventilators and intensive care.
- The combination of the factors above has helped reduce demand for medical services, allowing patients to receive better care.
A Deeper Dive Into COVID-19 Mortality
If this is already Too Much Information for you, I apologize. But I have one more chart to show and this one is fascinating to study. I happened upon the nation’s most complete data set at CovidTracking.com, a volunteer data compilation hosted by The Atlantic magazine.
The most complete data set lists 40 data sets from 56 US states and territories from March 16, 2020 through the present. (Data for fewer states extends all the way back to January 13, 2020.) Data definitions help clarify the many categories.
The quality of data and the definitions used vary from state to state, so the data must be used with care. I found one obvious inconsistency, in that the number of patients currently on a ventilator (6,925) exceeds the total that have ever been on a ventilator (3,959). I am hoping that hiccups of this type don’t invalidate the broad totals that I present here but hey, it’s not surprising if sometimes data is worth only what you pay for it.
– Explanation of the Data
I downloaded the entire data set, calculated the total-US sum for each date and then averaged each entry over a 7-day period. The averaging is necessary because some officials don’t report certain data, such as deaths, on weekends, causing a false bumpiness in the numbers. I extracted the following data sets (the field names used by CovidTracking are in parentheses):
- New cases of COVID-19 (positive increase)
- New deaths (death increase)
- Hospital admissions (hospitalized increase)
- Intensive Care Unit (ICU) admissions (daily changes in InIcuCumulative)
- Ventilator admissions (daily changes in onVentilatorCumulative)
I plotted the results on a logarithmic scale, which has its advantages and disadvantages. The obvious disadvantage is that a log scale minimizes the tragedy of the infection and disruption of millions of our fellow Americans and their families. However, the advantage is that we can easily compare the trends in disparate and widely different numbers.
Here is the plot. If you want to see the underlying data set, click on the plot for a CSV file.
Click to see Covid Tracking source data.csv.
– New Cases and New Deaths
The data shows three peaks in new cases, and we may right now be at the third peak in new deaths. It’s informative to consider the ratio of these numbers, which is “sort of” a measure of mortality, the likelihood that an infection leads to death.
- First peak: New cases peak at 31,111 on 4/12/20. New deaths peak at 2,116 on 4/21/20. This is a ratio of 6.80% and a delay of 9 days from peak to peak.
- Second peak: New cases peak at 66,564 on 7/23/20. New deaths peak at 1,145 on 8/1/20. The ratio is 1.72% and the delay is 8 days.
- Third peak: New cases peak at 244,707 on 1/11/21. New deaths may be peaking at the third-from-the-last data point, 3,304 deaths on 1/26/21. The ratio is 1.35% and the delay is 15 days.
On the surface, this trend shows a five-to-one drop in case mortality, as we saw before. However, early in 2020 we had very little testing, so the only way we knew someone was sick was when they acquired symptoms that were serious and unmistakable. Today, we test many more people and detect many cases that are mild: they don’t lead to death, or even to hospitalization.
Let’s defer for the moment what this data means and move to some other measures.
– Hospitalizations and New Cases
The hospitalization data is somewhat irregular, showing spurious-appearing peaks around June 1 and November 1. Otherwise, hospitalizations follow a three-peak curve that resembles the new cases. Some hospitals were over-filled and their data may not reflect all the people who needed to be hospitalized. Nevertheless, the nationwide total may still be approximately correct.
- First peak: Hospitalizations peak at 4,119 on 4/6/20. The ratio of hospitalizations to new cases at the peak is 13.24%.
- Second peak: Hospitalizations peak at 3,132 on 8/7/20. The ratio of peaks is 4.71%.
- Third peak: Hospitalizations peak at 4,843 on 1/10/21. The ratio of peaks is 1.98%.
We would like to believe that a serious coronavirus infection will consistently lead to the patient being hospitalized. If this is the case, then we conclude that even as we identify more new cases, a smaller fraction of them are strong infections: a decline of more than 6 to 1 since early 2020.
– Hospitalizations and Deaths
Therefore, let us assume that the number of hospitalizations is a reasonable measure of the number of serious, that is, strong infections. Using hospitalizations eliminates a large number of mild infections that count because those people test positive for the virus. We’ll compare hospitalizations with deaths:
- First peak: The ratio of deaths to hospitalizations is 2,116 to 4,119: 51.4%. A huge mortality rate for hospitalized patients.
- Second peak: The ratio of deaths to hospitalizations is 1,145 to 3,132: 36.6%. Somewhat smaller but still distressingly large.
- Third peak: The ratio of deaths to hospitalizations is 3,304 to 4,843: 68.2%, larger than ever.
What does that 68.2% number signify?
It’s not likely that the quality of coronavirus care is falling. Therefore, I propose that this number implies that there are so many coronavirus cases now that many victims die before they enter a hospital, either because of crowded hospitals, or because patient care less often requires hospitalization.
– Hospitalizations, Intensive Care and Ventilators
Over the time frame of the plot, the rate of new coronavirus cases has increased by almost 700%. The number of new deaths has increased by 56%. Hospitalizations have varied, but are creeping up. However, the utilization of intensive care and ventilators holds steady, although hospitals have added more of both these facilities. That suggests that more hospitalized patients are responding to less heroic treatments; therefore fewer hospitalized patients require intensive treatments. This is a bit of good news in a generally troubling prospect.
The data is very complex, as you see. Here is my summary:
- The effective mortality, that is, your chance of death if diagnosed with COVID-19, has dropped a factor of 5 in the past year.
- Medical treatments have evolved, improving patient outcomes and reducing the demand on hospitals, especially for intensive care.
- Serious infections are becoming rarer, and infections requiring hospitalization are becoming less common. However, mild infections continue to boom.
Overall, COVID-19 mortality is improving. The disease is still rampant, either out of control or barely stable. Nevertheless, we are making progress: if you test positive for coronavirus today, you are more likely to have a mild case of the disease; and if you have a serious case, you are more likely to survive it than one year ago.
Mortality Rate Trend chart and COVID-19 Data graph by Art Chester from referenced sources.
This remaining images come from a wordless novel by artist Lynd Ward (1905-1985): Wild Pilgrimage (1932). Although most famous for his wood engravings, Ward also worked in other media and as a book illustrator. Many of Ward’s engravings are dark in content, telling the stories of troubled lives (“He whom the gods favor, dies young“). However, he also communicates very effectively when he depicts hope and possibility, as in the works shown here. I appended my own names to these untitled images.
I was not able to learn who currently owns the rights to these images. Therefore, I will quote Wikipedia’s principle of fair use under United States copyright law, asserting that showing low-resolution images (in this case, maximum 275 pixels) constitutes fair use when accompanied by a discussion of the artist and his works.
Thanks for trying to quantify this difficult (and politically charged) subject. I believe that the poorly coordinated US response has been an important part of the problem. Masked, social distancing, etc. have not been adequately imposed. Also, this opens the door to new variants of the virus — which may turn out to be a big problem — or it may just mean that we have to get annual vaccinations, as in the case of the flu. Many thanks for leading us through this maze. Dick
Thanks for your comment and feedback, Dick! It’s true that our lack of a national response may be part of our problem, but not all the countries with national standards have fared well either. Think of the problems plaguing UK and Europe. The greatest successes seem to have been in a few countries who imposed strict rules that would not have been well-accepted in the US: total border closure, mandatory tracing, strong stay-at-home orders (I’m thinking of New Zealand, Japan, South Korea as examples).
You’re quite right, this whole situation is a maze and we cannot yet see our way through it. I hope that common sense, persistence and an optimistic outlook will help us win the day! – Art