Friday, March 20, 2020

Stats And Truth Re Covid-19

No, this won't be a stat heavy blog, but it will reproduce Greg Cochran's blog which strongly questions John Ioannidis' pooh-poohing the danger of Covid-19. I believe he makes a strong case for why public authorities need to take this very seriously.

John Ioannidis is sort of a rock star among medical statisticians--which means he's not an epidemiologist. Here is a link to his original article, which I haven't read: A fiasco in the making? As the coronavirus pandemic takes hold, we are making decisions without reliable data. Ioannidis' thesis is largely based on the cruise ship experience. Those readers who regard this as a hoax pandemic, or maybe just an overreaction, will take comfort from it.

For my part, I finished Michael Osterholm's book last night, which concludes with chapters on his number one priority--preparing for a new flu pandemic, which he regards as a certainty (including a lucid explanation for why). It's very sobering and informative, as is his chapter on the prospects for a truly effective flu vaccine--vaccines in general are difficult, and in the case of rapidly mutating viruses, particularly so.

Greg Cochran takes out after Ioannidis in his blog today, which I reproduce. Some of what Cochran uses from history--an invaluable tool of epidemiological analysis--is also explained in Osterholm's book.

John Ioannidis
Posted on March 20, 2020 by gcochran9

He emphasizes the cruise ship (why ignore other whole countries?) and he distorts that example. Most never caught it: of those that did, all had excellent medical care. That’s no longer possible when many millions get infected – resources are limited.

He suggests a scenario in which 1% of the population gets infected and 0.3% of that 1% die. We are not seeing that low level of fatality in China, or Korea, or Italy.  As for 1% getting infected, where the f*ck does that come from? When a new pathogen shows up, no-one is immune, and the only limiting factor (barring intelligent human action) is having  enough contract it, recover, and develop immunity that the virus has trouble find vulnerable hosts. The limit, in a simple model, is when R0, multiplied by the fraction of never-infected people, is less than 1.0 .  Since R0 [of Covid-19] is something like 3 (assuming no major efforts at social distancing etc), that would require infection of 2/3ds of the population of the United States – with a death rate well above the 1% we’re seeing in places like South Korea.  More like 4%, or even higher.

[You can argue with Cochran's assumptions here--the R0=3, etc., the level of infection required for "herd" immunity, etc. The basic criticism is valid, IMO.]

Again, where in the name of God does this 1% come from?  Is Wuflu supposed to quit because he’s caught his limit?

The Faroe Islands had an epidemic of measles in 1846, investigated by Peter Panum, a young Danish doctor. There had been no measles in the Faeroes for 65 years – so only some elderly people were immune. Out of some 8000 inhabitants, how many ended up contracting measles? Was it 1%, due to some kind of  viral gentility?  2%? 10%?

No: 6000 out of 8000 faroese got the measles, the kind of result you expect from a simple model.  75%.  Measles has a high R0 ( ~10), which would have predicted 90% – but people over 65 had already had it, and even the Faroese don’t practice perfect panmixia.

The Spanish flu had an R0 around 2.0 . There had been an antigenically related flu around 1890, so older adults were  less vulnerable. The percentage infected  ranged up to 50% – which, against, is approximately the limit predicted by the value of R0.

[The Spanish flu, which many believe originated in Kansas, did target the young and healthy, and pregnant women. It also came in several waves.]

When a pathogen is NOT novel, sure, you can have lower fractions of people vulnerable, because many people are already immune,  and the fraction infected can be low.  If it depends upon some regionally varying vector, like mosquitoes, sure, it doesn’t have to sweep across the whole country.  If its R0 is only greater than 1.0 in some subpopulations, as was the case with HIV, it may spread only in those subpopulations.

But Wuflu IS novel, and does NOT depend upon a vector.  The vulnerable subpopulation is those that breathe.

So,  the surprise-free prediction is that it hits > 50% of the population –

Unless we stop it.


  1. I think Cochrane is overdoing it just as he suggests Ioannidis is underplaying it. For as much as the Princess Diamond is dismissed--"What about whole countries?"--Cochrane misses two factors. 1) The Princess Diamond was a complete census of an entire, isolated population, so the data is not nothing. 2) There is no data of complete countries, there is only sampling, limited by test availability.

    Cochrane's citing the Faroe Island measles epidemic of 1846 would tend to confirm the use of the Diamond Princess--an isolated entire population, where everyone was tested--not dismiss it.

    This doesn't make the DP data a gold standard, it makes it a starting point--confirmed and modified as more data accumulates. The Faroe Island is NOT a good example, if only because we better understand contagion.

    In fact, the Faroe Island outbreak in 1846 predates the germ theory of disease as identified by John Snow (London cholera--1854), Louis Pasteur (1860-1864), and Joseph Lister (1870s). Needless to say, the people conduct themselves differently in light of Snow, Pasteur, and Lister's findings 150 years on.

    Random sampling (testing) works to define the extent of the challenge, but we don't have random testing. Testing, because it's limited, has started with those ill and symptomatic, and at-risk. Those not a risk, like myself with symptoms of the common cold, need not draw on health care resources or a test.

    I understood Ioannidis to recommend better collection, collation, and assessment of the data, as we're making decisions in the absence of good quality data. That seems entirely within his field of expertise.

    1. Of course it would have helped if this had occurred just about anywhere other than a 'Socialist Paradise.'

      Tom S.

    2. It's true that we don't have complete data for any country--it's still early days. Nor is the cruise ship data nothing--but its isolated population was probably wildly unrepresentative of any country as well. The results are also counterintuitive in a number of respects. For example on the cruise ship the 20-40 age group were heavily affected. While all deaths were of older people, older people were a disproportionate group and even then were much healthier than an average group of oldies. The Faeroes, otoh, would have had a representative population. Cochran only used that example for a very limited purpose which, in retrospect, almost exactly confirms modern epidemiological expectations. Measles is so contagious--one of the most readily transmissible of all diseases--that behavior modification for avoidance is very difficult.

      The real question to be investigated re the DP is why such a small portion of the total population was infected, given the R0 of Covid-19 as we know it from countries around the world.

    3. @Tom

      Before going to bed last night I listened to some Youtubes with Osterholm. Among other matters of interest he was fairly adamant that he expects new outbreaks in China once China sends everyone back to work.

    4. The Diamond Princess was an odd case. Only 17% of people got infected. If Covid is so dangerous, how come more people didn't get sick?

      Rob S

    5. "Only 17% of people got infected. If Covid is so dangerous, how come more people didn't get sick?"

      Consider ...

      Covid-19 is classified as a form of SARS = officially, SARS Cov 2.

      The original SARS burst upon a world catching it totally unaware--for some time nobody knew what it was. 16 years later the official stats state that the overall mortality rate was 6%, but approached 50% among those over 60.

      Early thought is that Covid-19 is less lethal than SARS but quite a bit more infectious. Meaning: it's more likely to cause enough hospitalizations that it will readily overwhelm all medical services in any given country. We're seeing that in various countries around the world. Sick passengers on the DP all received excellent medical care, but that's not true around the world where medical services have been overwhelmed.

      The US population is ~ 330,000,000. If "only" 17% percent of that population got infected that would be ~ 56,000,000 people. Even a low rate of hospitalization among that population would totally swamp our hospitals. If the mortality rate for that infected population were only .01, that would work out to 560,000 deaths

      As a public health adviser to Trump, would you recommend that he go on TV and point out how much worse the 1918 flu pandemic was?

      But what if the projections of many others are closer to the truth than Ioannides? After all, the earlier example of SARS suggests that a complacent attitude could lead to catastrophic consequences.

      My point is that the DP example looks like an outlier compared to what's happening around the world and that Trump really has no alternative to the actions he has taken. He can always back down if it turns out to be overreaction, but underreaction risks unthinkable consequences.

  2. One would think that the Fake News would put aside their Trump hatred and try to save lives. No, not even the possibility of needless death justifies laying down their arms to help their countrymen.


    1. Joe, Joe, Joe--Orange Man Bad is all they know. You're far too hopeful that media will learn more than one trick...

  3. Touche'.

    No, I guess that I'm not surprised.

  4. Again- the case fatality rate is overstated everywhere- in China, in South Korea, in Italy, in Spain, in the US, etc.

    At this point, we are firmly identifying the dead since all respiratory related deaths are mandatory tests for virus, but we are identifying only a fraction of the people who have been infected in the past, now, and will continue to undercount in the future.

    Seriously, if you followed this in the US from the first cases confirmed by RT-PCR, the ratios of positives to negatives bounced from 10% to 30% for the first couple of weeks, but the overall trend has been pretty solid since early in the first week of March- about 10-15% of the tests everyday are positive. What you are seeing is an explosion in testing, not an explosion in infections. As of yesterday afternoon, the US had run 103,000 tests with about 12,000 positives. Today we saw about 5,000 new cases, and I bet the total tests run right now is close over 170,000 based on the day to day change this week in tests run.

    What we have been doing all along is slowly uncovering a very large pool of infected people, probably numbering at least a million or two in the US alone. I don't know if the infection is spreading rapidly or not- no one can know this when the testing regularly shows the same relatively low positive rate, but that rate isn't changing much, so what we do know is that percentage of coronavirus sufferers isn't changing with respect to the number of people with flu-like symptoms- the requirement, still, to actually get tested for most of us.

    What is actually needed now is an immunoassay that can randomly sample the population for COVID-19 specific antibodies- that is the only way we will really know how many have been infected.

    You want to know what Italy, China, Spain, and increasingly Great Britain and France are doing wrong? They aren't protecting the feeble elderly in hospitals and nursing homes. There is a reason you can't work in a nursing home or a hospital without agressive up to date vaccinations against all types of flus and pneumonia causing viruses and bacteria- those illness kill such residents.

    You want to limit the deaths, here is what you do- you isolate all nursing home residents from visitors until we have a vaccine- you test for coronavirus everyday the people who work in such facilities. You don't house COVID-19 victims in all the hospitals- you segregate them into their own facilities with their own medical personnel.

    With the changes in social distancing, better hygiene, and shutting down crowded places has probably already reduced the R0 to well less than one pretty much everywhere in the developed world, but it will take at least two weeks to see the large pool of infected to start to fall enough to be picked up in the limited testing that can be done.

    1. Most of the above I definitely agree with. One thing I would question is your assertion that we can't know whether the virus is spreading rapidly. It appears that the infections in the worst hot spots--NY and Seattle--mostly stem from single individuals, so I think we can safely say that it's spreading rapidly. Also, it appears--perhaps counterintuitively to me--that those infections didn't begin until about the end of January. With the lengthy incubation period that suggests that we may still be at an relatively early stage--great news, if true.

      Re hospitals, I read somewhere today that for quite a while the sick in Italy were being taken to hospitals and not segregated--either from workers, the general patient population, or other Covid-19 patients--and that's one reason for the very high number of deaths.

    2. What I mean is, it is probably spreading rapidly but from a quite small starting point. Time will tell of course.

    3. You can't know that for the whole, Mark, and probably not really for the hot zones either. In the case of Seattle/Washington state, they don't know where most of the infections came from- some can be linked back to China, a biology conference, but most contacts can't be traced.

      As for the recent hotspots, people are fooling themselves if they think this is being traced to single individuals- with a big enough pool of infected- not even as high as 1 million in the US- there will be overlaps between any two carriers once you get more than 3 contacts away.

      The testing is actually leading to some truly idiotic policy decisions because it is leading them to assume they can trace the contacts and isolate all those people who might test positive in a day or two- this is just wishful thinking. The policy you should spend almost all your effort in convincing people to follow is this one- if you have fever or a persistent cough, or just a damned head cold- isolate yourself until you have recovered, and then add 3 days just to be safe. Most people will do this if you emphasize it, but if you insist on trying determine whether or not someone has COVID-19 and must isolate vs having something else, you are undermining the best policy because you won't ever test more than 10% of the people with flu like symptoms at any give time. There are real physical and socialogical contraints on how many people can ever be tested here. I think the upper limit of tests/day that the US can run will be met in about a week, and be about 100,000-150,000 tests/day, plus people will eventually stop going for the tests since it takes hours in lots of places to get sample taken. At that point, it will look like the new cases have plateaued, but that will just be an artifact of the tests no longer increasing, and as people stop being tested altogether, it will look like the disease is abating. I strongly suspect that is what happened in both China and South Korea- testing limits were reached, and then people stopped getting tested at all, and just isolated themselves when they got sick.

    4. The stories I read indicated that in Seattle the outbreak was traced to one, or possibly two, Chinese individuals. In NY it was traced to an Iranian woman returning from Iran. I'm sure there were others as well, but just as Osterholm says that there were 'superspreaders' of SARS-classic, so too with SARS-2 we see the same phenomenon. If there were a large number of initial spreaders I believe we'd be seeing far more widespread evidence of infection. In my own locality we were doing testing fairly early, but the number of cases is much lower despite a large population and busy international airport.

    5. More statistics from perusing various State websites FWIW:


      So I'm going to guess overall 5% positive rate of tests.
      95% of people tested will have the flu, not COVID.

      And from
      "Public health experts have estimated up to 1% of the population could be infected after identifying multiple cases of community spread of coronavirus."

      So if 1% of Americans get it, and 1% of those die, from 327 million Americans, total 33 thousand deaths, which is the average annual deaths from influenza.

    6. Tx, dfp21. Stats like that are part of why I suspect that a relatively few individuals are behind this spread. Which provides some hope that our public health people may be able to control it. Hope.

      Nobody wants to see 33,000 deaths on top of the flu deaths. More importantly, nobody wants to see a killer like Covid-19 become endemic to the US, reemerging in the fall.

    7. Mark, no one is going to trace 5,000 new cases a day. No one. At best, they will test the household members, and that will be it.

      As for the claims about tracing local outbreaks to individuals- they offer no proof- I have read the details- they are just guessing in 90% of the cases that there is an actual link. And if you read people claiming to have followed the trail by tracking mutations- that is just a flat out fabrication for the most part.

      Yes, superspreaders exist, but there isn't anything particularly unique to them, they just don't give a crap about infecting other people- that is all it is- if you can find, good, but you won't find more than 10% of them.

      The disease will just end up running its course in the US- we don't have the draconian will or ability to shut people in their houses for the several months it would take to completely halt this- you would have to shoot people who didn't obey just to get a month's compliance.

      This doesn't make me a pessimist- all of the data suggest this virus isn't the calamity the media are selling 24/7, but I am getting increasingly pessimistic about the economic calamity we are bringing on with truly stupid and evil decisions to lock down entire states. This goes on for another month, you are going to have 50% of the people unemployed, and I don't care how much money the government prints- it won't be able to buy the goods that are no longer produced.

      All of these governors executing statewide shutdowns are going to be looking at this in a couple weeks and will have to declare victory with no better data than they had this week because the supplies of food and medicine will start to run out for the simple reason that no one is actually producing them here in the US.

    8. All I know is that I've read that genome sequencing can be used to help track the spread.

      This search provides links to articles:

      "using genome sequencing to track spreaders of coronavirus"

      As for the economic calamity, I agree to some extent. I suspect that in a few weeks the authorities will have a better handle on how we stand throughout the states and, hopefully, more preventive gear will be available--yes, face masks.

      For all the reasons provided in my posts from yesterday I disagree that "all the data suggest" that this isn't a big deal.

  5. COVID-19 raged through that Kirkwood (near Seattle) nursing home to a great extent because the staff believed it was the seasonal flu, a common occurrence in such facilities. On top of that three staff members went to work when they were ill.

    In Northern Italy the first outbreak was traced to a Chinese chef who was cooking Chinese food and delivering it to homes throughout a fairly wide area.

    They may not identify all of the Patients Zero, but they have identified more than a few in the areas where the illness is concentrated.

    1. Yes. In a way those events provide some hope that the outbreak here will be at least more limited and traceable.