12,399 views

, 27 tweets, 21 min read

@TheEconomist

@TheEconomist

THREAD re the "everybody's-got-it" school of covid truthery, which my story this wk in @TheEconomist economist.com/graphic-detail…, based on study by @inschool4life & @Alex_Washburne, supports. Most impt pt is that these results do NOT mean we should end/loosen lockdowns now/soon. 1/n

By now, every1 following this knows that only a minority of people w/ covid-19 are getting tested, & that a substantial share have mild or totally asymptomatic cases. The official case-fatality rate (CFR), which is deaths divided by CONFIRMED cases, of 6% is obv way too high. 2/n

To estimate infection-fatality rate (IFR), which is deaths divided by total infections, have 2 estimate number of undiagnosed cases. Evidence is dribbling in that it's v hi. A review: 1. Entire pop of Italian town of Vò was tested by Mar 6. 3% infected independent.co.uk/news/world/eur… 3/n

2. Serology on 1/3 of the pop of Italian town of Robbio, found 13-14% w/ antibodies to covid-19, 30x num of confirmed cases unz.com/isteve/even-a-…. 3. Serology on 1K people around Copenhagen found 2.7% w/ antibodies, despite hi false negative rate sst.dk/-/media/Udgive… 4/n

4. Serology on n=500 in Gangelt, Germany found 14% w/antibodies + 2% w/active virus. Used test w/ >99% specificity, making base-rate fallacy (en.wikipedia.org/wiki/Base_rate…), where false positives dwarf real ones when true cases are rare, less of a concern. google.com/amp/s/www.tech… 5/n

5. Analysis of deCODE Genetics sample in Iceland finds ~10x undercounting of covid-19 infections, even in country w/ highest testing rate in world (6% of pop already tested). Implies 4-5% of country infected. igmchicago.org/wp-content/upl… 6/n

6. Aforementioned Silverman/Washburne paper estimates increase from Mar 8-28 in share of Americans contacting doctors about flu-like symptoms w/o having flu, relative to historical baseline, at 7% of population. (medrxiv.org/content/10.110…) 7/n

7. Imperial College researchers, using pure epidemiology based on the number of deaths, estimated 5% infection rate by late March across European countries. imperial.ac.uk/media/imperial… 8/n

8. San Miguel County Colorado (centered on Telluride) is testing the entire pop of 8K. There have been delays w/processing. But 2% either positive or borderline/need followup so far. coloradosun.com/2020/04/09/san… 9/n

There's obv yuge uncertainty around these estimates. But the 2 studies on small towns that got slammed w/ covid both came out ~15%; others are all low-mid single digits. By contrast, just 0.1% of Americans & Britons, 0.2% of Italians & 0.3% of Spaniards have tested positive 10/n

So undercounting is prob 10x or more. This has both encouraging & disturbing implications. Good: if we've counted deaths right, IFR is 1/10 of CFR or less. 1/10 of current CFR would be 0.6%. That happens to be near one 0.66% estimate of China's IFR. medrxiv.org/content/10.110… 11/n

2nd encouraging consequence is that it puts us a bit closer to eventual herd immunity. If only 0.2% of pop has already had covid-19, then we're 9-10 doubling periods away. If 5% has already had it, we're "only" 7 doubling periods away. 12/n

HOWEVER, remember that IFR has both a numerator and a denominator. We're also under-counting the numerator—covid-19 deaths—albeit prob by a smaller magnitude. Excess mortality in parts of Spain, Italy and France is 2-3x official covid-19 death toll economist.com/graphic-detail… 13/n

Moreover, many people now infected haven't died yet, but will. You need to compare deaths today to estimated infections 3-4 weeks ago (depending on reporting lag) to guess at the true share. Recorded US deaths the past few days have been ~50% higher than in previous week 14/n

But let's just explore consequences of a low-ish IFR scenario—say 0.3%, so "just" ~3x flu. What would this mean? First, it means covid-19 spreads v quickly. One new paper on spread in China gets r0 of 5.7, DOUBLE earlier estimates of reproductive num wwwnc.cdc.gov/eid/article/26… 15/n

Very fast growth means extreme, economically destructive lockdowns are needed to stop explosive spread, and that the % of pop infected needed for herd immunity is higher. For those misguidedly arguing to lift lockdowns now, the "everyone" in "everyone's got it" is relative. 16/n

Just b/c recent data implies we're clos-ER to herd immunity than official case counts show doesn't mean we're actually anywhere CLOSE to herd immunity. If covid-19 has really spread undetected like wildfire, then maybe 10% of the pop has had it—which means that 90% hasn't! 17/n

But if covid-19 were not so bad—maybe somewhat deadlier than the flu, but not a full order of magnitude or worse—then why wouldn't we just treat it like chicken pox, and hope everyone gets it so we get to herd immunity ASAP? 18/n

2 reasons. 1st, low-IFR scenarios presume everyone w/ severe cases gets good supportive care. If hospitals get swamped, doctors have 2 triage & death rate soars, like in Italy. And w/ spread as fast as studies imply, hospitals would def overflow. NY is already close to this. 19/n

2nd, getting 2 herd immunity w/o a vaccine, in a low-IFR, fast-spreading scenario means maybe 80% of pop has 2 b infected. & a 0.3% IFR on 80% of pop means that ~1 in 400 people die—like 800K deaths in the US. There's no way we would or should tolerate that kind of mortality 20/n

So yeah. The vast majority of covid-19 cases are prob undiagnosed, & the disease is prob much less deadly than official CFR. Under extreme assumptions, it could even have similar mortality to flu. The proper response 2 this is what govts are already doing: keeping lockdowns 21/n

in place until the rate of new cases drops enough 4 hospitals to give every1 best possible care, & mass testing becomes available so we can rely on contact tracing to open up in appropriately small increments until we get a vaccine. Even if the IFR were just 0.1%, a world 22/n