Beer Virus V

[T.Lex]

The better question would be is it ever 'clinically' significant.

Again, the only response is yes, it does matter, sometimes.

Is ratio. Depends on the size of the numbers and what is being measured.

But, as I stated in the post that spawned this faux controversy, the change (from decreasing to increasing) can also indicate a trend.

 

[qwerty]

The study is an interesting read. I have always felt and experienced personally the inflation of cases, just makes some sense. Although the term "busted" in the article lends itself to hyperbole.

BUSTED: CDC Inflated COVID Numbers, Accused of Violating Federal Law

The Centers for Disease Control and Prevention (CDC) stands accused of violating federal law by inflating Coronavirus fatality numbers, according to stunning information obtained by NATIONAL FILE. CDC illegally inflated the COVID fatality number by at least 1,600 percent as the 2020 presidential...

nationalfile.com

 

[T.Lex]

So it appears to start with a procedural objection that, at the beginning of the pandemic caused by a novel virus, the CDC failed to go through the federal comment process to change reporting rules.

Ok.

I'm on board with the criticism that the CDC was woefully unprepared for the reporting obligations. They weren't even actually tracking the flu, but using modeling.

But I'll continue to wait for further studies to compare conclusions about how far off the numbers are.

 

[chipbennett]

The literal answer to your question, of course, is "yes."

Show me.

A change of 0.0003, using a standard significance level (say, 0.05), is very rarely, if ever, going to be statistically significant. And even where it might be, those circumstance likely wouldn't apply here.

That's a difference of 3 more or fewer people out of ten thousand testing positive. That's well - well - within the testing error range, much less, random chance.

 

[T.Lex]

Show me.

A change of 0.0003, using a standard significance level (say, 0.05), is very rarely, if ever, going to be statistically significant. And even where it might be, those circumstance likely wouldn't apply here.

That's a difference of 3 more or fewer people out of ten thousand testing positive. That's well - well - within the testing error range, much less, random chance.

Well, you used the word "ever" which extends the topic.

So, the Six Sigma people may bristle at only 4 significant figures. If .03% of all remdesivir produced caused death, I think some people would consider that significant. I dunno - maybe it does and no one is saying so.

And the financial service industry even has a name for that number: 3 basis points. For people who get paid by AUM or invested capital, that's an important number.

But whatever. This is a contrived controversy with no apparent utility other than to emphasize the importance of reading comprehension. In my original post, I conceded that the change in percentage was nominal. At no point did I say that the amount of the increase, in and of itself, was significant.

 

[chipbennett]

There is no controversy, contrived or otherwise.

If anything, I'm commenting on the utility (or futility) of discussing a change in positivity rate from 1.66% to 1.69% (or whatever the metric was - looks like it was fatality rate?).

And, this is the statement you made about it. I'll highlight the parts that are assertions/suggestions not supported by such a statistically insignificant change:

Nationally, our absolute CFR has crept up in the last 10 days from 1.66% to today potentially hitting 1.69%.

Difficult to tell if it is a change in test numbers (I don't track that) or what. I really don't think the virus itself is any more (or less) fatal.

And sure, it is a nominal increase, but it suggests a larger trend since the beginning of the year.

None of those is supported by a reported change from 1.66% to 1.69%.

 

[T.Lex]

We're really going to do this? Selective redlining is selective, but this is too easy for a Monday.

crept up/increase - any positive number going from 166 to 169 (regardless where the decimal is) is increasing; because that's the word we use for that

nominal - not looking it up, but those tools are available; it means "really small" (for purposes of this conversation)

larger trend - for a long time leading up to the beginning of January, the percentage of people dying compared to the number of people testing positive was decreasing; since then, it has remained static and increased (see above), thus "suggest[ing] a larger trend"

admitting I don't know if the test numbers change or if the virus is deadlier - I'm not even sure why that's redlined; I really don't know and you can't prove that I do know

You're just wrong about my post and it isn't even worth discussing anymore.

 

[nonobaddog]

There is no controversy, contrived or otherwise.

If anything, I'm commenting on the utility (or futility) of discussing a change in positivity rate from 1.66% to 1.69% (or whatever the metric was - looks like it was fatality rate?).

And, this is the statement you made about it. I'll highlight the parts that are assertions/suggestions not supported by such a statistically insignificant change:


None of those is supported by a reported change from 1.66% to 1.69%.

I suspect you are right. That 0.03% change in CFR does represent a 1.8% change from point A to point B. (0.03/1.66) That could be real, however I bet an error analysis would show that last digit is not a significant digit.
Any calculator or computer can poop out lots of digits but that does not make them meaningful.

 

[HoughMade]

Anyhoo...

The news in Indiana remains encouraging. Positives, deaths and hospitalizations are all falling. I don't usually put too much stock in Monday numbers as weekend reporting is less complete, but things still look good.

In particular, over 34% of ICU beds are available and only about 17% of ICU beds are being used due to Covid.

At this point, we are probably starting to see some effect from the more widespread vaccinations, even just the first dose.

 

[T.Lex]

I suspect you are right. That 0.03% change in CFR does represent a 1.8% change from point A to point B. (0.03/1.66) That could be real, however I bet an error analysis would show that last digit is not a significant digit.
Any calculator or computer can poop out lots of digits but that does not make them meaningful.

Ok.

Even that approach supports the observation that there is (at best) a stabilization of the CFR.

From 5/7/2020 (based on reported numbers, of course), when the CFR was at 6% (5.95%, but since we're rounding to tenths, that means 6%), the CFR was decreasing fairly steadily through December 27, 2020, when it hit 1.7% (actually, 1.74%). It has been at 1.7% (rounded) since then.

 

[chipbennett]

We're really going to do this? Selective redlining is selective, but this is too easy for a Monday.

crept up/increase - any positive number going from 166 to 169 (regardless where the decimal is) is increasing; because that's the word we use for that

nominal - not looking it up, but those tools are available; it means "really small" (for purposes of this conversation)

larger trend - for a long time leading up to the beginning of January, the percentage of people dying compared to the number of people testing positive was decreasing; since then, it has remained static and increased (see above), thus "suggest[ing] a larger trend"

admitting I don't know if the test numbers change or if the virus is deadlier - I'm not even sure why that's redlined; I really don't know and you can't prove that I do know

You're just wrong about my post and it isn't even worth discussing anymore.

You're still missing the point. A reported change of CFR from 1.66% to 1.69% is not evidence of any change whatsoever. That's the point. You can continue to split hairs all you want, but this is science, not a courtroom. Data within the margin of error do not represent a change. They are statistical noise. Thus, based on these data, nothing has crept up or increased. Nothing has changed, even nominally. Based on these data, we can make no evaluation of a change in the deadliness of the virus.

As for trend: perhaps, but you'd need much more. We have a rule of thumb in my world (a world that requires both science and risk basis): once is an accident, twice is a coincidence, three times is a trend. And because the variables can't be isolated/controlled (testing criteria change, testing rates change, the damn analytical methods for the very same tests are changing, positivity criteria are changing, etc.), there's no way to say that there is a trend of anything.

 

[HoughMade]

Anyhoo...

The news in Indiana remains encouraging. Positives, deaths and hospitalizations are all falling. I don't usually put too much stock in Monday numbers as weekend reporting is less complete, but things still look good.

In particular, over 34% of ICU beds are available and only about 17% of ICU beds are being used due to Covid.

At this point, we are probably starting to see some effect from the more widespread vaccinations, even just the first dose.

I forgot to say that we will likely see the vaccination's effects in hospitalizations and deaths first rather than positivity because while a fairly good percentage of the over 70 crowd is getting vaccinated, that is a small percentage of the overall population. However, they are also the ones most likely to get really sick, require hospitalization and die.

 

[nonobaddog]

I forgot to say that we will likely see the vaccination's effects in hospitalizations and deaths first rather than positivity because while a fairly good percentage of the over 70 crowd is getting vaccinated, that is a small percentage of the overall population. However, they are also the ones most likely to get really sick, require hospitalization and die.

Around here it seems like a very low percentage of the over 70 crowd is getting vaccinated.
We do have a large healthcare population and they are getting it first so there is very little spillover to the old people at this point.

 

[SheepDog4Life]

Again, the only response is yes, it does matter, sometimes.

Is ratio. Depends on the size of the numbers and what is being measured.

But, as I stated in the post that spawned this faux controversy, the change (from decreasing to increasing) can also indicate a trend.

T, it's all about when the cases and deaths are counted.

A spike in cases will have negative (decreasing) influence on that CFR calculation because the deaths associated with those cases occur in the future.

Once the case spike retreats, there are still daily fatalities from cases reported weeks, in some cases months prior, so that has an increasing affect on CFR as calculated.

This is due to using current total cases and fatalities for CFR calculation when those two time series have a delta time between them... they don't occur at the same time.

This would be true even if the case-by-case fatality rate was constant due to the time difference.

 

[SheepDog4Life]

Perhaps an "extreme" thought experiment would help.

What would happen if there was a super-duper super-spreader event? I dunno, maybe a national door handle and hand rail licking day or some such and 300 million people popped positive on the same day? Everyone in the US who hadn't had it already.

What would do to your CFR calculation? Drop it from 1.6% to 0.1% or so. Did the virus just become less deadly because everyone got it today? No, of course not.

And then months later, say exactly 1.6% of those 300 million cases proved fatal, what would happen to the CFR calculation? It would increase as the fatalities occurred... no more new cases because everyone has had it... and eventually end up at 1.6% again.

The "true" CFR never changed... just when cases and deaths occur in time. It's a flaw in the way it's calculated.

 

[T.Lex]

T, it's all about when the cases and deaths are counted.

A spike in cases will have negative (decreasing) influence on that CFR calculation because the deaths associated with those cases occur in the future.

Once the case spike retreats, there are still daily fatalities from cases reported weeks, in some cases months prior, so that has an increasing affect on CFR as calculated.

This is due to using current total cases and fatalities for CFR calculation when those two time series have a delta time between them... they don't occur at the same time.

This would be true even if the case-by-case fatality rate was constant due to the time difference.

I don't disagree. The deaths are always a lagging indicator of how bad things are.

[As an aside, there are some that seem to emphasize the CFR as a metric to compare to other nations; I am not in that group.]

But, even as a lagging indicator, one can observe broad trends over time in terms of how we area dealing with the virus.

Any given variant probably has an "absolute" CFR (my descriptor). That's not what we're shorthanding "CFR" to be. A better descriptor would probably be "perceived" CFR or "presented" CFR. There's a great deal we don't know, but one thing we know are the numbers being reported.

As always, that's what I'm using.

[As another aside, the daily average CFR is actually different, too.]

The timing flaw is what it is and has been this whole time. Same with the various levels of reporting flaws.

My point is that, I've tracked some of these things for nearly a year now. When I say the reported absolute CFR trended down from May to December (or January, depending on how many decimals you want to use), that's simply the truth. When I say that it has been stable (or is trending up, depending on the decimals) since the beginning of January, that's just the truth, too.

I specifically did not offer any explanation for it.

 

[BugI02]

Again, the only response is yes, it does matter, sometimes.

Is ratio. Depends on the size of the numbers and what is being measured.

But, as I stated in the post that spawned this faux controversy, the change (from decreasing to increasing) can also indicate a trend.

I just find it ... entertaining? ... that the long, continuous decline in that number from last October was largely ignored in favor of derived/contrived statistics about doubling times for deaths, the conclusion seeming that it was apropos of nothing. A slight increase in that statistic, however, has it heralded as a significant indicator of trend

Hmmm

 

[BugI02]

Around here it seems like a very low percentage of the over 70 crowd is getting vaccinated.
We do have a large healthcare population and they are getting it first so there is very little spillover to the old people at this point.

We are currently, in theory, vaccinating people 75 and up - with 65 and over opening up Feb 8. The reality is appointments for vaccination of the currently authorized 75 and ups are now extending into late March, so the reality doesn't match the hype very well

When Bezos said Amazon would assist with vaccine distribution I thought I'd be able to get vaccinated at Whole Foods after making my appointment on Prime

 

[BugI02]

T, it's all about when the cases and deaths are counted.

A spike in cases will have negative (decreasing) influence on that CFR calculation because the deaths associated with those cases occur in the future.

Once the case spike retreats, there are still daily fatalities from cases reported weeks, in some cases months prior, so that has an increasing affect on CFR as calculated.

This is due to using current total cases and fatalities for CFR calculation when those two time series have a delta time between them... they don't occur at the same time.

This would be true even if the case-by-case fatality rate was constant due to the time difference.

Since the worldometers-derived CFR was calculable over a continuous period of months, it seems that comparing one data point to another calculated the same way would still be relevant unless somehow the length of time between diagnosis and death was lengthening quite markedly. Recall that the steady, downward trend of the calculation was uninterrupted by the 'second spike' which at the time we were told was due to the 'death latency'. Are you suggesting that those increased deaths postulated from the increased cases in August and September are just now showing up in the numbers? That's quite a large latency

 

[T.Lex]

Perhaps an "extreme" thought experiment would help.

What would happen if there was a super-duper super-spreader event? I dunno, maybe a national door handle and hand rail licking day or some such and 300 million people popped positive on the same day? Everyone in the US who hadn't had it already.

What would do to your CFR calculation? Drop it from 1.6% to 0.1% or so. Did the virus just become less deadly because everyone got it today? No, of course not.

And then months later, say exactly 1.6% of those 300 million cases proved fatal, what would happen to the CFR calculation? It would increase as the fatalities occurred... no more new cases because everyone has had it... and eventually end up at 1.6% again.

The "true" CFR never changed... just when cases and deaths occur in time. It's a flaw in the way it's calculated.

Quick followup to this post.

I posted at various times about the serious/critical metric, which remains a black box. Assuming the criteria for that number was stable (I know, plenty to be skeptical about, but work with me), it continued to increase after the CFR max at ~6% in May. That is, it also lagged the cases, as its early max was about a month later in early June. In fact, it doesn't really appear to be linked to either CFR or daily deaths, frankly.

It peaked in early June, and ebbed and flowed through the summer until it was back under 15k in late September. Perhaps as a separate indicator of the "second" wave, it then increased to the current max of ~29k in mid-January. It has been falling since then.

To me, that indicates that we did do a better job of sustaining (maybe even curing) people until relatively recently.

Perhaps also noteworthy: using that serious/critical metric, using the reported daily deaths as a percentage of that daily number indicates that the overall average is that ~7.82% of the daily serious/critical number is the daily death number. That was falling roughly in line with the CFR. However, also recently (and I think I've posted this), that number has also increased to currently ~12.82% (7 day rolling avg).

Again, there's not much more to read into those stats than the reported numbers suggest a higher percentage of sick people are dying now than in the Fall. That's an observation; it is not a conclusion. I can't explain why, although there are lots of potential explanations.