Welcome to this Medmastery

coronavirus update. I’m Franz Wiesbauer. I’m an internist, trained in epidemiology

and public health at Johns Hopkins and the founder of Medmastery where we teach

important clinical skills to doctors and other healthcare providers

around the world. Today we’re going to talk about case

fatality rates and how to estimate and calculate them. Let’s go to the Johns

Hopkins coronavirus map. There we have the total number of

confirmed cases, total deaths, and total cases recovered. Please note that these are the

data as of March 3rd, 2020. So here you can find the total confirmed

cases and the total number of deaths. Now you could calculate a case

fatality rate with these numbers. You could divide the number of deaths

by the number of total confirmed cases. Multiply it by 100 to receive a case

fatality rate in this present situation of 3.4%, but this way of doing

it is slightly erroneous. I have constructed a fictitious epidemic

showing new daily cases and new daily deaths of a certain disease in

green and red, respectively. Now, as you can see from the graph and

probably from your own reasoning, is that someone needs to be sick or classified as

a case for some time before they can die. In this fictitious population, the time

from being identified as a case to the time of death is two days on average. You can see that the peak of cases was

reached on day seven to nine whereas the peak of deaths occurred

between days nine and 11. Here you can see the same two curves of

daily deaths and daily new cases in light red and light green on the very bottom. These are the daily new deaths. These are the daily new cases. In addition, you can now also see the

cumulative cases and deaths in dark red and dark green. As you can appreciate when we are in

the ascending limb of the epidemic, the current number of cumulative deaths is

also delayed by two days when compared to the number of cumulative cases. So deaths that occurred over here

correspond to cases that occurred over here. So one would have to compare the

cumulative deaths from day ten to the number of cumulative cases from day

eight in order to arrive at an accurate comparison or a case fatality rate. So what’s that time delay in COVID-19? So over here, you have the time of

infection, the time of symptom onset, then the time when the case is

reported, and the time of death. Now the time between infection and symptom

onset is called the incubation period. The incubation period of COVID-19 has

been estimated to be around 5.2 days on average. However, there are cases with

the lower incubation period. But there are also patients with a much

longer incubation period of up to 21 days, possibly even 24 days have been described. So this interval is five days. And what’s the time from symptom

onset to the reporting of a case? The authors of this study found

a mean duration of 7.1 days. Obviously, this will be quite a bit

different from country to country. So let’s say that this period is seven

days; and what’s the time from symptom onset to eventual death? In an analyses done at the Imperial

College London, authors found that the time from disease onset to eventual death

was 22.3 days, very similar to the time of recovery. So this interval is 22 days

long according to this paper. So 22 minus seven days equals a time delay

between reporting and death of 17 days, and that’s the interval one should use to

determine where to take the current case number from. So instead of comparing the current total

deaths to the current total number of confirmed cases, one should compare the

current total deaths to the total number of confirmed cases from 17 days prior. This number can be retrieved from

enlarging this inset down here. So we said, these numbers were taken from

March three, going back 17 days means we need to retrieve the number of

total cases from February 15th. So on that day, the number of total cases

in mainland China was 68,300, and the total number of confirmed cases

in other locations was 685. So 68,300 plus 685 equals 68,985

total cases on February 15th. So what’s the new case fatality

rate using this lag time of 17 days? Remember this was the calculation from

the beginning of the video, which did not account for the lag time. And now we have 3,117 divided by

68,985 times 100, which equals 4.5%. Now, you should not only use a lag time of

17 days – but also see what a shorter lag time of say eight days will bring up. This is called doing a sensitivity

analysis or testing your numbers. Now, the problem in all of this, as

described by the researchers from Imperial College London, is these calculations are

likely skewed because we don’t know how many asymptomatic cases or mild cases

are out there whom we don’t know off. They argue that mostly the very severe

cases have been detected in mainland China. Whereas it appears that internationally,

all symptomatic cases, meaning deaths, severe cases, and symptomatic

cases are picked up. If there truly is a substantial number of

cases in the dark blue bottom part of the pyramid, then the true case fatality rate

will be much lower than what we have just calculated. The authors of this paper found that we

might be underestimating the true number of cases by a factor of 19. Which would mean that we would have to

divide our calculated case fatality rate by 19 in order to arrive at the true rate. Which would obviously make the

mortality rates seem much less scary. One argument in favor of

that view can be seen here. This graph from nucleus wealth shows that

there might be significant under-reporting of cases in most Chinese provinces,

but also in places like Iran. The bars depict the reported COVID-19

cases per million inhabitants. South Korea, which has done the most

extensive testing up until now, reports around 80 cases per million. The Chinese provinces are way lower when

they should be really higher, especially those provinces next to Hubei,

where the outbreak started. So this should be an indicator that

the true case fatality rate is lower. In reality. We will only be able to tell once

we measure antibodies in the general population in order to tell what

the true number of cases was. That’s it for now. If you want to improve your understanding

of epidemiology, make sure to register for a free Medmastery trial account and

attend our epidemiology essentials course. We’ve just opened it up to trial

users due to the huge demand. So stay safe and talk soon.

## 4 Comments

amazing video you deserve more subscribers

Many people are focusing on assessing how dangerous this virus is in terms of deaths toll. Many say it's like a "silly influenza" comparing the two viruses in terms of fatality rates. In my opinion this is all misleading. What should be worried now is the lack of a therapeutic drug that targets the virus and blocks its replication in the body. Without this weapon all pneumonias caused by the virus that develops ARDS need ICU interventions (high flow oxigenation, mechanical ventilation, positive pressure assisted respiration and ECMO). Since the number of cases that suddenly appear in the population, the demand for this kind of interventions can overwhelm the numbers of beds, anesthesiologists, ECMO specialists available. Imagine a glass of water already full that needs to hold additional water. Here in Italy the number of cases seeking ICU attention in some part of my country are already overwhelming the hospitals capacity. Planned surgical interventions have already been postponed to add more anesthesiologists to the ICU wards. It should be also considered that some of the infected are the same health care workers who assists the patients. So it's a virus that suddenly overwhelms a health care system and weakens its capacity to react to the virus. That's a new view that should be taken into account when assessing how dangerous this virus is. In my humble opinion.

How about we apply a similar analysis to south Korea specifically? Right now there 7500 cases, with a growth rate of 10 percent per day we get 1384 cases 17 days ago. This means a mortality rate of 3.9 percent using this method, but let me be clear that this will not be accurate because it doesn't really work when you go that far back and the number of cases is so low. Since the average number of cases back then was so low many of the deaths we have seen now probably didn't test positive 17 days ago (which would be the average) but rather from the larger group of people that have tested positive a bit later and with a bit out of the average time for the lag between positive test and death.

22-7=15 not 17!

Would you please explain?