COVID-19 (artist formerly known as Wuhan strain novel Corona virus)

No I am not we cannot assign one outcome to one case. I'm saying what we *have* done in the press is aggregate it over time.

Typically when a case fatality rate is reported in the press, they are taking total cases to date and total deaths to date (aggregation over time).

This has trended down. This is because case capture rate has gone up (primary) and infection fatality rate has likely gone down (secondary) due to treatment and demographics of the infected.

However, if we were instead to report case fatality rate as one case and one outcome with no time impact, I still wouldn't agree with your point.

My point is pretty simple. I would not have disagreed with you had you said infection fatality rate in your initial post. Instead you said case fatality rate. And I am saying when higher capture rates due to more testing include more asymptomatic and mildly symptomatic INFECTIONS in your CASE dataset, your CASE fatality rate will go down. Even if measured as one case, one outcome - no time effects.

 

nothing gets by you.

 

And I'm saying, again, that's a stupid measure. As reported. By the press. Stupid measure.

Case fatality rate can go up. Even with including asymptomatic infections.

Not when using totals, which is stupid, as stated. But when looking at it at a case by case by date level, which is what it should be.

Example already given, but here is another:

Monday: 3 cases
Tuesday: 0 cases
Wednesday: death Outcome from 1 Monday case.
Thursday: 1 case
Friday: non death outcomes from final 2 Monday cases.

On Friday, the case to death rate would be 1/3.

Not 1/4. Thursday case has unknown outcome.

But that is not what you are telling me this measure does.

What you are telling me it does is:

Monday: 0/3
T: 0/0 or 0/3, depending
W: 1/0 or 1/3, depending
R: 0/1 or 1/4, depending
F: 0/0 or 1/4, depending

 

Do it your way. It has still gone down since March because we are capturing so many more cases. We shouldn’t even discuss case fatality rates anymore. We have infectio. Fatality rate estimates now.

 

I don't want to count from March. I want to know what the outcome of a case is, from only cases 4 weeks ago that have had an outcome vs the outcome of a case 3 months ago that have had a known outcome, to determine if our outcomes are actually getting better, or if we're just burying the outcome in a mountain of new cases. A measure that COULD be going up.

And I believe that such a measure would validly be called death to case rate, when looking only at the death outcomes.

 

And I am saying in your example, if case capture rate is higher and more asymptomatic and mildly symptomatic are being captured as a percentage of total new cases that the case fatality rate will fall without being any indication of improving outcomes. The only way it will indicate improved outcomes is if the severity of cases in the case pool is not largely changed.

From three months ago to today, my guess is that's an OK comparison.

It's a horrible comparison of April data to today. Even when calculated the way you suggest.

That is why infection fatality rates for a specific demographic over time are really the only solid way to investigate improved outcomes. And, of course, this is challenged by the difficulty in teasing out total infections vs. measured cases.

 

Not necessarily the only way, and FULL contradiction of the following BINARY "will fall." You are contradicting yourself with your first two sentences.

It isn't a horrible comparison of April to today. People can't die twice.

 

Correct on binary will fall. The point that I intended to convey is that infection fatality rate can be identical but case fatality will drop given the conditional I provided.

I will provide numbers to illustrate why I think it is a horrible comparison. They are made up but illustrate the conditional I applied in my first sentence. And there are data to support the notion that the condition is applicable.

April:
Total infections: 1,000,000 (among a demographic representative of the population)
Measured cases: 100,000
Percent of total cases that become severe or heavily symptomatic: 3 percent
Percent of severe cases captured in testing: 80 percent
Percent cases that are severe: 24 percent
Percent asymptomatic or mildly symptomatic: 76 percent
Infection fatality rate: 0.5 percent
Total deaths: 5,000
Deaths among cases: 4,000 - that is, we missed twenty percent of the COVID deaths
Case fatality rate: 4 percent

October:
Total infections: 1,000,000
Measured Cases: 350,000
Percent of total cases that become severe or heavily symptomatic: 3 percent
Percent of severe cases captured in testing: 95 percent
Percent cases that are severe: 8 percent
Percent asymptomatic or mildly symptomatic: 92 percent
Infection fatality rate: 0.5 percent
Total deaths: 5,000
Deaths among cases: 4,750 - that is, we missed five percent of the COVID deaths
Case fatality rate: 1.4 percent

In this example, despite the fact the we measured the death outcome more effectively in October than April and recorded more deaths, the case fatality rate dropped due to the case capture rate. However, the infection fatality rate did not change.

Because this case capture effect has been strong from April to today, it makes using case fatality rate to track outcomes rather useless. You have to rely on modeling of infection fatality rate to try to do that, and it is difficult because error bars are large.

 

Case fatality rate can rise.

What does missing deaths have to do with anything? You are measuring the outcome of a CASE. This is standard morbidity and mortality.

Let's start over. What are the differences between an infection and a case, for you?

 

By definition a case is a measured infection.

 

I only added the death capture as a matter of intellectual honesty. If you have enough testing to capture more cases you are at a point that you are capturing more of the deaths as well, most likely.

 

So infections is unmeasured?

 

Yes. That is our reality. It has to be modeled.

 

Ignore reality, and view test design through the lens of the ideal, that everything is possible. As such, the case status on any given day of any given person is known. Following? In April, the case to death ratio would be some value. In November, the case to death ratio would be some other value. And it could be higher or lower.

 

I'm talking solely MEASURED things. Modeling is prediction.

I don't measure a sapling with a tape and say its 145 feet tall, in 45 years. Those are predictions. I am talking about measurement. It's 3 feet.

 

Well yes. But that by definition is the infection fatality rate. In its limit the case fatality rate will equal the IFR when case capture is 100%.

Case capture rate has moved so much that it overwhelms changes in IFR this making the very rea comparison that you are taking about misleading

 

Lordy.

My example laid out why the measurement you are asking for will likely not reveal their signal you are looking for due to noise introduced by changes in case capture rate.

Since you are apparently fine talking in hypotheticals where we have perfect data then just say infection fatality rate. That is the true measure of the virus’ lethality for a given demographic.

Case fatality rates are things we talk about in the absence of IFR studies.

 

No, that cannot be the infection fatality rate if infection is not measured.

 

You just told me that infection means unmeasured.

 

It is.

Where do you think the flu symptomatic infection fatality rate of 0.1% comes from? That’s modeled.

The actual case fatality rate (measured cases/death outcomes) is much higher. Like over 3% I think.