PlagueWatch 2020-23 - Coronavirus

[YmoBeThere]

Long story slightly shorter - the virus may remain viable for a week or slightly more when on the surface of cold food. I'm not sure this is new news, but there appears to be some rigor around testing the viability of the virus, not just its presence. The study has not been peer reviewed.

"While the CDC and the FDA have been consistently firm in their stance that foodborne transmission of coronavirus is virtually nonexistent, there may be some new evidence to the contrary.

After finding traces of the virus on several imported foods and their packaging, like salmon from Norway and chicken wings from Brazil, Chinese authorities have been investigating the viability of the virus on food items since June.

Now, a new study out of South China Agricultural University and Guangdong Academy of Agricultural Sciences in Guangzhou says that traces of the virus lingering on salmon could not only be detected, but also may remain infectious for more than a week."



https://www.yahoo.com/lifestyle/popu...195427387.html

[rsvman]

I was donating blood recently and one of the perks is they screen all donors for COVID antibodies, partly so they can have a pool for convalescent plasma donors. I was particularly interested in this as it looks at potential exposure of healthy folks. So off to google I went looking for data on how many blood donors have had positive SARS-COV-2 antibody screens and could not find any overall reports on this. Has anyone seen these numbers reported? Or know where I should look for them?

I don't know the data, but I want to re-emphasize to be very careful interpreting antibody test results at this point, when the prevalence of the disease in the community is still relatively low. Even if a test is 95% sensitive and 95% specific, a lot of the positives are false positives when the prevalence of the disease is low.

Let's take an example in which the prevalence is, for example, 5% (just to make calculations easier). And then let's do antibody testing in 1000 people, and take a look at what happens.

You will have 950 people who have NOT had coronavirus, and 50 that have. 95% of those who have NOT had coronavirus will test negative, but 5% will test positive (47.5 people). Of the 50 who HAVE had the coronavirus, 95% will test positive (47.5 people). Therefore, when all is said and done, 95 people will test positive, but only half of them will actually have been infected with the virus. HALF of the positives are false positives.


In other words, if you go and get tested and you test positive, you can't let your guard down; there is a 50% chance that the test result is wrong.

[camion]

I don't know the data, but I want to re-emphasize to be very careful interpreting antibody test results at this point, when the prevalence of the disease in the community is still relatively low. Even if a test is 95% sensitive and 95% specific, a lot of the positives are false positives when the prevalence of the disease is low.

Let's take an example in which the prevalence is, for example, 5% (just to make calculations easier). And then let's do antibody testing in 1000 people, and take a look at what happens.

You will have 950 people who have NOT had coronavirus, and 50 that have. 95% of those who have NOT had coronavirus will test negative, but 5% will test positive (47.5 people). Of the 50 who HAVE had the coronavirus, 95% will test positive (47.5 people). Therefore, when all is said and done, 95 people will test positive, but only half of them will actually have been infected with the virus. HALF of the positives are false positives.


In other words, if you go and get tested and you test positive, you can't let your guard down; there is a 50% chance that the test result is wrong.

One of the under-reported complications of the COVID-19 pandemic is that we have to learn (or relearn) math and statistics.

Charlton Heston's last line in Planet of the Apes is appropriate here.

[PackMan97]

I don't know the data, but I want to re-emphasize to be very careful interpreting antibody test results at this point, when the prevalence of the disease in the community is still relatively low. Even if a test is 95% sensitive and 95% specific, a lot of the positives are false positives when the prevalence of the disease is low.

Let's take an example in which the prevalence is, for example, 5% (just to make calculations easier). And then let's do antibody testing in 1000 people, and take a look at what happens.

You will have 950 people who have NOT had coronavirus, and 50 that have. 95% of those who have NOT had coronavirus will test negative, but 5% will test positive (47.5 people). Of the 50 who HAVE had the coronavirus, 95% will test positive (47.5 people). Therefore, when all is said and done, 95 people will test positive, but only half of them will actually have been infected with the virus. HALF of the positives are false positives.


In other words, if you go and get tested and you test positive, you can't let your guard down; there is a 50% chance that the test result is wrong.

Math is crazy. There is a 5% chance your test is wrong, but a 50% chance your test is wrong if you test positive but only a 2.75% chance your test is wrong if it's negative.

[rsvman]

Math is crazy. There is a 5% chance your test is wrong, but a 50% chance your test is wrong if you test positive but only a 2.75% chance your test is wrong if it's negative.

Yep. That's because positive and negative predictive values depend on prevalence (or, in an individual, a priori likelihood that the person has the disease in question).

Therefore, a low prevalence improves the negative predictive value of a negative test the same way it decreases the positive predictive value of a positive test.

Obviously, the positive predictive value would be better in somebody who had a clinical disease compatible with the diagnosis. The calculations refer to the overall predictice values in an unselected population.

[CameronBornAndBred]

Now, a new study out of South China Agricultural University and Guangdong Academy of Agricultural Sciences in Guangzhou says that traces of the virus lingering on salmon could not only be detected, but also may remain infectious for more than a week."



https://www.yahoo.com/lifestyle/popu...195427387.html

hqdefault.jpg

[House P]

I don't know the data, but I want to re-emphasize to be very careful interpreting antibody test results at this point, when the prevalence of the disease in the community is still relatively low. Even if a test is 95% sensitive and 95% specific, a lot of the positives are false positives when the prevalence of the disease is low.

Let's take an example in which the prevalence is, for example, 5% (just to make calculations easier). And then let's do antibody testing in 1000 people, and take a look at what happens.

You will have 950 people who have NOT had coronavirus, and 50 that have. 95% of those who have NOT had coronavirus will test negative, but 5% will test positive (47.5 people). Of the 50 who HAVE had the coronavirus, 95% will test positive (47.5 people). Therefore, when all is said and done, 95 people will test positive, but only half of them will actually have been infected with the virus. HALF of the positives are false positives.


In other words, if you go and get tested and you test positive, you can't let your guard down; there is a 50% chance that the test result is wrong.

Excellent point about the accuracy (positive predictive value) of a test with moderate specificity (95%) in a low prevalence population. That being said, what do you know about the clinical specificity of COVID antibody testing?

Based on this website from the FDA, it appears that the analytical specificity of COVID antibody testing is quite high. I am not sure which antibody tests are most widely used, but the large diagnostic companies I am familiar with (Abbott, Roche, Siemens, Beckman Coulter, Ortho Clinical, BioMerieux) all report a specificity of 99% or greater - and 13 of the 14 antibody tests from these companies report a specificity greater 99.5% or greater. Quest's website mentions that the specificity of their antibody testing is "approximately 99% to 100%".

Now the clinical specificity may be lower than the analytical specificity, but if the clinical specificity is 99.5%, RSVman's example would result in a positive predictive value of 91%. This means that "only" 9% of positive tests are "wrong" (false positives). In a population with a 10% prevalence, the PPV would be 95.5%. Still not 100%, but better than 50% and much better than the pre-test probability of 5%.

I guess the real question is what would (or should) change if an individual knew they had a 91% or 95.5% probability of having antibodies?

Math is crazy. There is a 5% chance your test is wrong, but a 50% chance your test is wrong if you test positive but only a 2.75% chance your test is wrong if it's negative.

Anyone interested in the math can download this excel spreadsheet from the FDA which allows you to explore the relationship between prevalence, sensitivity, specificity, PPV, and NPV.

[MChambers]

GOP leader of the California Senate was under quarantine, but spoke at a large prayer event without a mask.

https://www.washingtonpost.com/natio...ne-california/

In front of thousands of worshipers packed shoulder-to-shoulder outside the Capitol, California Senate Minority Leader Shannon Grove (R) grabbed the microphone on Sunday and promised the huge church event would have a real impact.

“I declare that after all of this is over tonight, the remnant, the residue of this worship will saturate this ground and seep into that building,” Grove told the crowd, the Sacramento Bee reported.

But state leaders are warning the event’s impact could actually be a mass coronavirus outbreak. Although Grove’s permit allowed 1,000 people and required social distancing, the California Highway Patrol said three times as many showed up; videos showed virtually no social distancing or masks in the crowd.

[freshmanjs]

I don't know the data, but I want to re-emphasize to be very careful interpreting antibody test results at this point, when the prevalence of the disease in the community is still relatively low. Even if a test is 95% sensitive and 95% specific, a lot of the positives are false positives when the prevalence of the disease is low.

Let's take an example in which the prevalence is, for example, 5% (just to make calculations easier). And then let's do antibody testing in 1000 people, and take a look at what happens.

You will have 950 people who have NOT had coronavirus, and 50 that have. 95% of those who have NOT had coronavirus will test negative, but 5% will test positive (47.5 people). Of the 50 who HAVE had the coronavirus, 95% will test positive (47.5 people). Therefore, when all is said and done, 95 people will test positive, but only half of them will actually have been infected with the virus. HALF of the positives are false positives.


In other words, if you go and get tested and you test positive, you can't let your guard down; there is a 50% chance that the test result is wrong.

Everything here is correct, of course (and interesting!).

Also important to note that:

- It depends what the test is being used for. For an individual to know if he/she had the virus, this analysis is spot on. However, for aggregate assessment of how many people have been infected, the 95% sensitivity and specificity would be perfectly acceptable and give you a very accurate picture

- Also, the above assumes a random person is being tested (5% chance of having been infected). If a person is tested because they were symptomatic and/or because they are known to have been exposed, then their chance of having been infected is much higher than the 5% average and, therefore, the test is more useful

[rsvman]

...
- Also, the above assumes a random person is being tested (5% chance of having been infected). If a person is tested because they were symptomatic and/or because they are known to have been exposed, then their chance of having been infected is much higher than the 5% average and, therefore, the test is more useful

I mentioned that in post #10105, above. I agree.

But in the example we were talking about, the Red Cross is testing people who go to donate blood. This is pretty close to a random sample, because presumably the people who go to donate blood don't have any higher risk of having had Covid than other members of the community. One could even argue that perhaps their risk might be slightly lower, if anything, because health-conscious people probably have a slightly lower risk. Which, of course, would make the positive predictive value even worse.

[Bostondevil]

To point out the most obvious mistake, the say "assuming none of these cases were fatal," but we, in fact, know that one person who attended the event did succumb to the virus. Jus' sayin.'


Otherwise, it's a lot of conjecture, but it points out something that has been shown to be true over and over again, and that is that generally people can't really conceptualize exponential growth.

Yeah, but, doesn't that number seem just a bit too large to you? I will cop to having the initial reaction of OMG, what were they thinking? 250,000 cases! What the 2020!?! And then I went, what percentage of overall cases is that in the month since the rally? I think it's something like 20% of all cases nationally. I imagine any uptick in South Dakota numbers can be blamed on the rally, but I don't think we can reliably attribute that high a percentage of cases in somewhere like Arizona to the rally. Some, sure, but 20%? I'm not buying it.

[rsvman]

Dunno. I think in the article they said 16%. Like I said, it seems like a lot of speculation, but on the other hand a guy I went to Scout camp with when I was a kid still owes me something like 80 million dollars, and he racked up that debt fishing. The bet started with one penny. He bet one penny that he would catch a fish the next time he cast, and then said 'double or nothing' each time he cast after that. Over the course of the next little while, he ran up the stated debt. That's the nature of exponential growth, and, like I said before, most people can't grasp it. Again, not saying that I believe the numbers to be accurate, just saying that they are not out of the realm of possibility.

[sagegrouse]

Yeah, but, doesn't that number seem just a bit too large to you? I will cop to having the initial reaction of OMG, what were they thinking? 250,000 cases! What the 2020!?! And then I went, what percentage of overall cases is that in the month since the rally? I think it's something like 20% of all cases nationally. I imagine any uptick in South Dakota numbers can be blamed on the rally, but I don't think we can reliably attribute that high a percentage of cases in somewhere like Arizona to the rally. Some, sure, but 20%? I'm not buying it.

When I briefly looked into the story, it looked like some people did an arithmetic extrapolation (well might have involved logarithms or exponential functions) and got a number -- a really big number! But is has not been correlated or buttressed with actually reported case data.

[MChambers]

When I briefly looked into the story, it looked like some people did an arithmetic extrapolation (well might have involved logarithms or exponential functions) and got a number -- a really big number! But is has not been correlated or buttressed with actually reported case data.

I think the 260,000 estimate is described in the study as being a worst-case scenario. But they did look at the increase in counties that contributed the most people to the Sturgis motorcycle rally. From the summary on the cover of the study: "Finally, difference-in-differences (dose response) estimates show that following the Sturgis event, counties that contributed the highest inflows of rally attendees experienced a 7.0 to 12.5 percent increase in COVID-19 cases relative to counties that did not contribute inflows."

https://cheps.sdsu.edu/docs/Contagio...Dave_et_al.pdf

So the number may be a little high, but there is data that supports it.

[Kdogg]

I think the 260,000 estimate is described in the study as being a worst-case scenario. But they did look at the increase in counties that contributed the most people to the Sturgis motorcycle rally. From the summary on the cover of the study: "Finally, difference-in-differences (dose response) estimates show that following the Sturgis event, counties that contributed the highest inflows of rally attendees experienced a 7.0 to 12.5 percent increase in COVID-19 cases relative to counties that did not contribute inflows."

https://cheps.sdsu.edu/docs/Contagio...Dave_et_al.pdf

So the number may be a little high, but there is data that supports it.

South Dakota's GOP governor dismisses the massive outbreak tied to the Sturgis Motorcycle Rally as 'back-of-the-napkin math.

She should ask Arthur Laffer about the power of “napkin math.”

[Bostondevil]

Dunno. I think in the article they said 16%. Like I said, it seems like a lot of speculation, but on the other hand a guy I went to Scout camp with when I was a kid still owes me something like 80 million dollars, and he racked up that debt fishing. The bet started with one penny. He bet one penny that he would catch a fish the next time he cast, and then said 'double or nothing' each time he cast after that. Over the course of the next little while, he ran up the stated debt. That's the nature of exponential growth, and, like I said before, most people can't grasp it. Again, not saying that I believe the numbers to be accurate, just saying that they are not out of the realm of possibility.

Article said 19%, so, my guess-timate was close. I read it more closely and I kinda can't quite grasp the reasoning they used - is dose response really appropriate here? (I don't have an answer for that.) I think, at the very least, the local prevalence of the "home counties" needs to be accounted for. I mean, were the counties that sent people to the rally having larger increases pre-rally than neighboring counties that sent fewer or no people to the rally? Are there better social distancing measures in place in the counties with lower rates? I need to know these things before I believe these numbers!

[budwom]

South Dakota's GOP governor dismisses the massive outbreak tied to the Sturgis Motorcycle Rally as 'back-of-the-napkin math.

She should ask Arthur Laffer about the power of “napkin math.”

Question The Snow Queen at your peril, sir!

[MChambers]

Article said 19%, so, my guess-timate was close. I read it more closely and I kinda can't quite grasp the reasoning they used - is dose response really appropriate here? (I don't have an answer for that.) I think, at the very least, the local prevalence of the "home counties" needs to be accounted for. I mean, were the counties that sent people to the rally having larger increases pre-rally than neighboring counties that sent fewer or no people to the rally? Are there better social distancing measures in place in the counties with lower rates? I need to know these things before I believe these numbers!

Here's a thoughtful article on the Sturgis study with comments by a couple of folks who seem to know something about public health and modeling. Basically, the study is reasonable, but that doesn't mean the numbers are right or even close. https://gizmodo.com/how-to-read-that...per-1844987945

[camion]

Article said 19%, so, my guess-timate was close. I read it more closely and I kinda can't quite grasp the reasoning they used - is dose response really appropriate here? (I don't have an answer for that.) I think, at the very least, the local prevalence of the "home counties" needs to be accounted for. I mean, were the counties that sent people to the rally having larger increases pre-rally than neighboring counties that sent fewer or no people to the rally? Are there better social distancing measures in place in the counties with lower rates? I need to know these things before I believe these numbers!

From the Gizmodo article which discussed the Sturgis study:
“A key limitation is that these sets of counties appear to have been on different trajectories before the rally, which makes it much harder to compare the trends after the rally and attribute the difference to the rally,” he said. “So, the 250,000 estimate is a nice headline-grabber but pretty likely to be an overestimate.”

One thing to consider is that the counties that sent people to the rally might have been on different trajectories from other counties before the rally precisely because they contained the type of people who would go to the rally, people who would have a greater tendency to be spreaders in any event. In a study where variables aren't controlled it's tough to ensure that the variables are independent.

*I am not a statistician, though I did work for one for two decades. She tried to teach me the basics, but said her dogs were more trainable.

[JasonEvans]

The perils of school reopenings...
"South Carolina 3rd grade teacher, 28, dies from COVID-19" - https://apnews.com/4ee025ba6c4c19c024ce384503f23ef3

It is worth noting that she did not get it from the kids because they were doing virtual classes, though she might have caught it at some teacher training sessions done before school restarted.

Bannister was a third grade teacher starting her fifth year of teaching at Windsor Elementary School in Columbia, officials said.

She was at the school on Aug. 28, during a week of teacher workdays before the school year began. She started teaching the following week from home, officials said. The district said it is tracing anyone who had close contact with Bannister and instructed custodians to deeply clean the school.