Thread by @WhitneyEpi

OK, bc of Stanford Med #COVID19 vaccine algorithm fiasco & N@te Silver’s recent tweet re: vaccine priority guidelines, I have to talk about one of my favorite topics! I even drew an illustrative Venn diagram! (I’m already imagining @rdpeng rolling his eyes. I’m sorry, Roger!) 1/

Key background 1: Because of limited supplies of #COVID19 vaccine, a federal advisory board making recommendations about the order in which people should get vaccinated (don’t get distracted: these recs are non-binding; real action is at state level) 2/
nytimes.com/2020/12/20/hea…

Key background 2: Infection fatality rates (IFRs) for #COVID19 increase markedly with age. Note the y-axis is on a log-10 scale. The IFR for COVID-19 for an 80 yo is literally ~1,000 times greater than for a 5 yo & ~100X greater than for a 40 yo
3/

https://twitter.com/trvrb/status/1336841338923278336?s=20

So that should settle it, right!? Age is the most important consideration for COVID-19 death burden so the older people should have first dibs on vaccine, right? Right…? Well, now let’s talk about Stanford Health Care!
4/

Key background 3: Most states have started vaccination with hospital-based health care workers.
5/

Stanford’s hospital had to decide which HCWs would get vaccine first. Stanford decided to prioritize older people over younger people in first round of vaccination. Cool, right?

WRONG!
Everyone is outraged!

Huh. What went wrong?!

https://twitter.com/NPR/status/1340126008406667269?s=20

6/

To explain, here’s the promised Venn diagram! I drew three circles: 1 circle for high risk of exposure to virus (left), 1 for high risk of infection *if* exposed (right), and 1 for high risk of death/grave sickness *if* exposed *and* infected (bottom). 7/

The Venn diagram distinguishes btw exposure and infection because #SARSCoV2 is not highly infectious if proper mitigation is in place. Exposure is not enough to cause infection. Infectiousness thrives in particular conditions:

https://twitter.com/booksnips/status/1311788238256955397?s=20

8/

Further, high IFR for older age groups is a “conditional” statistic. IFR = risk of death *IF* one is exposed to #SARSCoV2 virus & *IF* one is infected with that virus. This Venn diagram makes this explicit: Bottom circle is IFR. But 2 other circles also matter for death risk
9/

Now back to the Stanford! Many hospital MDs have high exposure risk (left circle) but low workplace infection (right circle). By now, most US hospitals have decent mitigation (ventilation & adequate-although limited-PPE), so infection risk relatively low during clinical care 10/

If argument to prioritize vaccine by age based on the bottom circle (theoretically high IFR), the decision to prioritize HCWs is based on the left circle (theoretically high exposure). Others have written interesting threads about this moral tension.11/

https://twitter.com/_stah/status/1338487545416527872?s=20

In general US popn, when you prioritize by age (bottom circle), you prioritize lots of folks at virtually no risk of exposure & infection (not ideal). *But* you also sweep in high-exposure, high-infection-risk older folks actually at highest risk of #COVID19 death (dark green)12/

But what happens when you apply the age priority scheme (select by IFR, conditional risk of death) to a population selected on the basis of their theoretical level of *exposure*? You get this!

https://twitter.com/propublica/status/1340046664875319297?s=20

13/

You get ludicrous results like a wealthy 68 yo Stanford Health Care radiologists and executives who work from home in spacious houses getting prioritized ahead of 29 yo residents with two roommates who work in the ED every day. 14/

(#Epitwitter nerds, I argue that this is a classic case of structurally defined selection bias. Fight me! Others, feel free to ignore this tweet)
cdn1.sph.harvard.edu/wp-content/upl… 15/

Once you restrict to hospital-based MDs (purple in Venn diagram), age prioritization loses advantage of getting some of the highest risk folks (dark green) incidentally. With Stanford MDs, you’re not getting any of highest risk people--but you get older WFH Stanford docs! 16/

I wrote a related @Biostatistics commentary. Applying algorithms (like one that prioritizes vaccination) to a new context can give really biased results if you don’t understand the mechanisms by which the algorithm worked in your original population 17/

https://twitter.com/KatieMollan/status/1221973713308078080?s=20

This thread was originally going to include a whole other general discussion of tension btw prioritizing based on age vs exposure & infection risk, but I’ll save that for another day. Basically: while prioritizing by age seems like a no-brainer, it’s often more complicated 18/