Substack cometh, and lo it is good. (Pricing)

COVID-19 at the beginning of the 2020 “Holiday Season”


Like most people I initially underestimated coronavirus. Unlike most people I have a blog where I can see what I actually thought. My first mention of coronavirus is on January 26th, 2020. This is what I said:

In Coronavirus, a ‘Battle’ That Could Humble China’s Strongman. One thing I will say is that public health professionals are focused on the tail risk. The risks are real. But please note that the worst-case scenario may not be the most likely scenario.

My worries about tail risk increased gradually until the middle of February (I was still relatively sanguine in early February, though that is when we began to stock up). On February 24th I sent out a very alarmed tweet, and several people privately have told me that that’s when they also became alarmed (I tweeted in reaction to a private query from “Default Friend”).

So where are we at? I’ve been wary about giving predictions for a while because though the worst, worst, case scenarios were avoided (people have taken precautious), there are some pretty grim numbers out there. All that being said, I’m going to be cautiously optimistic. I don’t think we’ll double the death toll over the rest of the pandemic for a variety of reasons. If I had to bet. Unfortunately, I might turn out to be wrong. Who knows?

Finally, there is an intense bittersweet aspect to the stories about China in January and February. I’m glad China didn’t collapse in the plague. But as 2021 starts China is in a good position to keep pushing ahead in the great power race. I have a piece to come out in City Journal soon that ruminates on this tentatively titled “Twilight Empire.”

12 thoughts on “COVID-19 at the beginning of the 2020 “Holiday Season”

  1. “Twilight Empire.”

    Well, that is depressing. I worry constantly about what kind of a society my children (and their future children) will inherit.

    Nonetheless, I look forward to your article.

  2. Slightly off topic, but GNXP is a fairly good go to for smart comments, I have had a look at the cumulative CFR rate in USA and EU on OWiD, they’ve both tended to converge towards the estimated Covid IFR – https://tinyurl.com/y3md9hg8 (albeit USA more)

    Be really useful if in that Explorer I could look easily at the rolling 7 day average CFR, comparing cases to deaths, which I think tend to be on about a 14 day lag(?). As the cumulative CFR is weighted down by very low detection rates in early pandemic (in NE USA and W Europe, mostly).

    In the absence of being able to do that on their website, downloaded their data and, here’s some graphics on how this property changes across the pandemic for the USA alone*: https://imgur.com/a/CbMOamT

    Since this is kind of converging towards 1, which is close to an estimated Covid-19 IFR, does this imply that the US is actually converging towards identifying all cases?

    And shouldn’t this sort of convergence to capturing somewhere in the range of 50% -> 100% of cases be improving the R0…? E.g. if you capture most of the cases, then people would at least avoid causing these high dispersal super-spreader events that drive most of spread, even if they don’t properly actually fully isolate, and then spread slows down a bunch…

    Or have I added 2+2 together and got 5?

    One obviously flaw here is if true deaths from Covid are waaay higher than recorded deaths, then this isn’t a well estimated ratio…

    OTOH, test positivity rates have risen in USA (and probably EU, but OWiD doesn’t allow this to be aggregated, probably because of how different EU states differently record positive tests) – https://tinyurl.com/y5fzxppw

    That’s not incompatible with an actual improvement in case detection – it could be that test positivity rises because, well, the US is just getting much better at identifying the right people to test.

    But if you take the standard interpretation that positivity = more undetected cases, that would indicate that improving CFR reflects improvements in real IFR, through age shifting and successful shielding (younger populations being infected), morbidity shifting (less morbid populations being infected), or better treatments…

    *The data file OWiD provides doesn’t aggregate EU deaths + cases, this must be done via some coding in their Explorer, so it’s harder for me to do the same exercise for EU…

  3. I’ve actually found it tough to decide whether we’ve been doing well or badly with Covid. A lot hangs on what you’re comparing things to. Eg some were predicting Sweden might see 96k deaths by summer unless they imposed strict lockdown measures like their southern neighbors. They are now at 6.5k. If you compare them to their neighbors, they look bad. If you compare them to the predictions, on the other hand, they’re doing very well. I think we first need to decide on metric of success and failure since both sides of debate seem to be deriving radically opposing conclusions from the same facts.

  4. I didn’t underestimate coronavirus at all. I basically understood what kind of thing we were dealing with before the first case was reported in New York. On February 29th, I said maybe 10-20% of the U.S. population will get it, New York will without doubt be the worst hit state, and the death toll will be “maybe” half a million. Prior to mid-to-late February, however, I greatly overestimated the intelligence, wisdom, morality, and competence of our leaders. Compare the sublime patriotism and empathy of a Xi Jinping or Prayut to the odiousness of a Donald Trump or Macron.

  5. Also, does anybody doubt that China has multiplied funding for its biological warfare program by a thousandfold in the past year? If it has any way to cripple the United States, it’s with a pandemic with ~15% IFR.

  6. @Harding: “Also, does anybody doubt that China has multiplied funding for its biological …”

    Previously I questioned Alina Chan’s reasoning ability but I did not post it.

    Alina Chan also seemed to be jumping to early conclusion in her chosen university major. She asserted that she chose biology because she did not want to directly compete with those so called Chinese math wizs. She did not understand that in China the top down streaming of student come very early in high schools as the reputation of high schools are dependent on how many of their students enter prestigious university STEM courses including biology. The top down streaming might already happening in primary schools since only elite students can get into prestigious high schools.

    The Chinese priority on research disciplines is reflected in the number of prestigious state financed of the so called “Double First” disciplines and biology ranked third with 16 university departments is ahead of mathematics and computer science each with only 14 chosen university departments, i.e. biology is more prestigious than mathematics and computer science.

    https://en.wikipedia.org/wiki/Double_First_Class_University_Plan/

    Ecology is ranked separately at fifth place. Information and communication engineering (which is responsible for the 5G network research) ranked far behind at number 8. Even those with biology minor also play dominant role in life science research. The Chinese Academy of Sciences division responsible for cloning primates, the principle investigator (ex Berkeley) is based in Shanghai while the primate lab is in Zhejiang and the director of the primate lab is a bio-engineer as well as the senior co-researcher which might be responsible for the day to day supervision and operation of the researchers. A second separate primate lab is also run by by a bio-engineer. The researcher responsible for carrying out the biology experiments at the dark side of the moon is a civil engineer.

  7. @Dux.ie, link? 🙂

    1) Following on from my upthread comment about case detection rates, here’s a quick eyeball to estimating death lag to cases for the EU set (with data could more systematically identify with +ndays is best correlation): https://imgur.com/a/UTdhpd0

    Cases start accelerating on Oct 6, then deaths around Oct 18 (but smoother curve makes it a little difficult to eyeball). Then a change to slower growth in cases around Oct 31, while deaths around Nov 10. So if trends continue, death curve should lag cases by about 10-12 days.

    Now cases started to slow down around Nov 10, but we’re about 20 days later now and death rate continues to rise, slowly…. Maybe this is a mirror image of the slower rise in deaths compared to cases.
    Simply superimposing the recent EU death curve over the recent case curve – https://imgur.com/a/ahMe6Xe.

    It looks to me like the curves, after accounting for lag, may have become somewhat uncorrelated (inversely correlated even) in the last week… Deaths continue to rise after, according to the prior-trend in cases, they should be falling.

    Needs further monitoring over the next week. Perhaps this is just a small change in death reporting lag, perhaps a deeper emerging problem with EU countries case detection rates.

    2) Published paper with a similar approach to what I’ve used to estimate detection rates – https://royalsocietypublishing.org/doi/10.1098/rsos.200909

    They use a lower IFR (0.68%) so estimate a lower detection than I thought, about 22.8% for the US by 31st August.

    Extrapolate from this by continued changes in 14day lagged CFR, and would expect US should be capturing about 1/3 of cases now, up from 1/5 in August.

    That’s comparable to the best countries at the start of the outbreak (Norway/South Korea), and still today as they haven’t really improved much too since then. But there may be a date difference where countries with similar rates of detection detect at different times and this has an impact on cutting spread…

    The only country with a steady decline in case detection rates according to their methodology is Australia, which had very high detection rates at the start of the epidemic, but has now reduced to levels which are still fairly high by European standards but relatively low. Hopefully this won’t pose much of a future problem for them.

    Case detection rates are still fairly low in absolute terms even for successful responders though – Norway and South Korea only rose to about 30% at the start of the outbreak, Germany only 15%.

  8. @Matt: “@Dux.ie, link?”

    Hmm. I have to find them again.

    Raw and processed data for [dsbbfinddx/FINDCov19TrackerShiny](https://github.com/dsbbfinddx/FINDCov19TrackerShiny)

    There are national as well as aggregated continent data,

    data_all.csv

    Reg|Name|Unit|Time|CumTestsOrig|NewTestsOrig|Pop100k|NewCasesOrig|NewDeathsOrig|CapCumCases|CapNewCases|CapCumDeaths|CapNewDeaths|CapCumTests|CapNewTests|AllCumCases|AllNewCases|AllCumDeaths|AllNewDeaths|AllCumTests|AllNewTests|Pos

    region|Africa
    region|Asia
    region|Europe
    region|North America
    region|Oceania
    region|South America

    I think he mislabled the Cap*’s as per capita but his Pop unit is 100k so it should be per 100k capita. I rescaled the test data to per 1000 capita so that they can be plotted in the same chart.

  9. @Matt: Deaths continue to rise after, according to the prior-trend in cases, they should be falling.

    I don’t think there’d be a 1-to-1 correlation at a fixed “lag” distance because it can take anywhere from 1 to 4 weeks to die. eg. If you get a 10,000 new cases today at a 10% fatality rate, you might get something like ~250 deaths/week for 3-4 weeks starting 1 week later, rather than 1,000 deaths all at once exactly 10-12 days later.

  10. @Tobus, I think that’s a frequent argument and logically sensible but empirically I don’t really know that this is the case in data; for ex, Random Critical Analysis has argued that new hospitalisation and new death data pretty much has the same shape with no lag.

    E.g. partly though deaths *could* show that sort of pattern, in practice many people die way faster than 1 week and almost instantly after hospitalisation, while conversely few people die 4 weeks out? Then once reporting lag included, this shows strong trend to a standard lag.

    Some of this may be artificiality in data (e.g. I know in UK there is controversy over whether deaths over 28 days out should be counted).

  11. Some had criticized the early Italian results. Now you have accepted manuscript from US CDC personnel (though not official position) that SARSCoV2 might be in US as early as December 13–16, 2019. The leak of US data have started.

    https://academic.oup.com/cid/advance-article/doi/10.1093/cid/ciaa1785/6012472

    Accepted manuscript
    Serologic testing of U.S. blood donations to identify SARS-CoV-2-reactive antibodies: December 2019-January 2020

    Discussion: These findings indicate that SARS-CoV-2 reactive antibodies were detected in 106 specimens, a small percentage of blood donations from California, Oregon, and Washington as early as December 13–16, 2019.

Comments are closed.