Iffy preprint on COVID-19 risk?

Get your science fix here: research, quackery, activism and all the rest
Post Reply
User avatar
jimbob
Dorkwood
Posts: 1385
Joined: Mon Nov 11, 2019 4:04 pm
Location: High Peak/Manchester

Iffy preprint on COVID-19 risk?

Post by jimbob » Tue Sep 08, 2020 8:20 pm

https://www.medrxiv.org/content/10.1101 ... l.pdf+html
Abstract Objective: Our objective is to demonstrate a method to estimate the probability of a laboratory confirmed COVID19 infection, hospitalization, and death arising from a contact with an individual of unknown infection status. Methods: We calculate the probability of a confirmed infection, hospitalization, and death resulting from a county-level person-contact using available data on current case incidence, secondary attack rates, infectious periods, asymptomatic infections, and ratios of confirmed infections to hospitalizations and fatalities. Results: Among US counties with populations greater than 500,000 people, during the week ending June 13,2020, the median estimate of the county level probability of a confirmed infection is 1 infection in 40,500 person contacts (Range: 10,100 to 586,000). For a 50 to 64 year-old individual, the median estimate of the county level probability of a hospitalization is 1 in 709,000 person contacts (Range: 177,000 to 10,200,000) and the median estimate of the county level probability of a fatality is 1 in 6,670,000 person contacts (Range 1,680,000 to 97,600.000). Conclusions and Relevance: Estimates of the individual probabilities of COVID19 infection, hospitalization and death vary widely but may not align with public risk perceptions. Systematically collected and publicly reported data on infection incidence by, for example, the setting of exposure, type of residence and occupation would allow more precise estimates of probabilities than possible with currently available public data. Calculation of secondary attack rates by setting and better measures of the prevalence of seropositivity would further improve those estimates.
I think this is attempting to answer the wrong question.
Annotation 2020-09-08 211325.png
Annotation 2020-09-08 211325.png (137.94 KiB) Viewed 207 times
Annotation 2020-09-08 211125.png
Annotation 2020-09-08 211125.png (171.1 KiB) Viewed 207 times
Starting with the minimum of the infections, then with the 10% infection rate from an infectious contact, and assuming there's no clustering - so indeed, if the overall rate is 6/100k uniformly spread and people interact uniformly, and you have a 0.3% chance of death if infectious, then it is going to be very low.

I think the main problem is that it's missing the impact of cluster events and it's actually better to think what the chance is of being in a group with an infectious person.

Any views... It's being shared by the founder of the "Keep Britain Free" hashtag - who lives in Monaco
Have you considered stupidity as an explanation

User avatar
sTeamTraen
Catbabel
Posts: 718
Joined: Mon Nov 11, 2019 4:24 pm

Re: Iffy preprint on COVID-19 risk?

Post by sTeamTraen » Tue Sep 08, 2020 8:52 pm

Any article that uses county-level data without discussing why the ecological fallacy does not apply should go straight on the junk pile.

This also illustrates the limitations of preprints. I don't think the creators of medRxiv imagined having anti-science trolls moving into the comments.
Careering along in my creosote car

Millennie Al
Fuzzable
Posts: 249
Joined: Mon Mar 16, 2020 4:02 am

Re: Iffy preprint on COVID-19 risk?

Post by Millennie Al » Wed Sep 09, 2020 1:36 am

It takes a big bunch of made up figures and pretends that doing some maths on them tells you something. It doesn't. For example, there's no such figure as "Probability of infection per exposure". That depends on the nature of the exposure (which is even explicitly mentioned in the paper), which depends on how people are behaving. If that was the only figure with substantial uncertainty, it might have been possible to produce a meaningful analysis, but if you read the paper there's an awful lot of estimation and assuming in there. And the different estimates are not independent, so having so many just makes the whole thing meaningless.
Covid-19 - Don't catch it: don't spread it.

User avatar
jimbob
Dorkwood
Posts: 1385
Joined: Mon Nov 11, 2019 4:04 pm
Location: High Peak/Manchester

Re: Iffy preprint on COVID-19 risk?

Post by jimbob » Wed Sep 09, 2020 12:31 pm

Millennie Al wrote:
Wed Sep 09, 2020 1:36 am
It takes a big bunch of made up figures and pretends that doing some maths on them tells you something. It doesn't. For example, there's no such figure as "Probability of infection per exposure". That depends on the nature of the exposure (which is even explicitly mentioned in the paper), which depends on how people are behaving. If that was the only figure with substantial uncertainty, it might have been possible to produce a meaningful analysis, but if you read the paper there's an awful lot of estimation and assuming in there. And the different estimates are not independent, so having so many just makes the whole thing meaningless.
Yup - I just need something that's easy for Twitter.

Maybe this bit " We conceptualize this probability under steady state conditions (e.g. no epidemic growth or decline) as a..."
Have you considered stupidity as an explanation

User avatar
sTeamTraen
Catbabel
Posts: 718
Joined: Mon Nov 11, 2019 4:24 pm

Re: Iffy preprint on COVID-19 risk?

Post by sTeamTraen » Thu Sep 10, 2020 8:59 pm

I've done a bit more reading.

First, they estimate the chance of death from COVID-19 after a single social encounter with a person of *unknown COVID-19 status". That's bonkers.

Second, they estimate a population prevalence of 7 per 100,000 (the work was done in June). That scales up to a total of 23,000 cases for the whole of the Yoooonited States. They are currently having 40,000 cases a day, and people remain "a case" for at least two weeks. So that's an underestimate by a factor of 25 or more.

This will hopefully never make it past peer review, but it's already doing damage. I'm having interesting discussions with other scientists about which is the more f.cked, preprints or the peer-reviewed literature. I think it's neck and neck at the moment. 😐
Careering along in my creosote car

Post Reply