Skip to content
 

A tale of two epidemiologists: It was the worst of times.

A couple of commenters pointed me to the story of John Ioannidis and Gideon Meyerowitz-Katz. David Gorski tells the tale. Ioannidis still seems to be dealing with the after-effects of his extrapolation last year that there might be 10,000 coronavirus deaths in the United States. This was just a back-of-the-envelope calculation, and as Gorski says, there’s no shame in being off by “by a factor of ten or even 50.” The problem is that Ioannidis didn’t seem to be able to let go of his initial downplaying of the risk. I agree with Ioannidis that so many of the deaths are elderly people so the loss of qualys is much less than from a flu that kills the same number of younger people, but still it is a lot of mortality and morbidity, not to mention all the disruption of people’s lives (much of which cannot be attributed to “lockdowns,” given that at this time last year people were pulling their kids out of schools, cancelling trips, etc.). It seems in retrospect that we could’ve saved a lot of lives by just all wearing masks, but it’s not like Ioannidis said, “only 10,000 deaths if we all wear masks”; no, it was the opposite, he said that if we’d just been living our lives normally we’d barely have noticed coronavirus at all.

This 10,000 extrapolation was based on an underestimate of the fatality rate of coronavirus and also a guess that only 1% of the U.S. population would get infected. The odd thing is, the very same March 2020 article with that calculation also said, “we don’t know if we are failing to capture infections by a factor of three or 300,” and a month later Ioannidis was a coauthor of a papers estimating that some counties in California already had infection rates over 2% in early April. I had skepticism about the 2% estimate—not that I thought it was too high, necessarily, just that I thought the estimate was less certain that the authors of that study had claimed—but, in any case, there was a lot of incoherence in the position that (a) infection rates were much higher than reported, but (b) only 1% of the U.S. population would get infected. I think we can all agree it’s been more than 1% by now.

Again, getting a forecast wrong is fine. It’s the nature of probabilistic forecasts that there will be uncertainty. We forecasted that Joe Biden would win between 259 and 415 electoral votes. He actually got 306, so that was in our range, but it was lower than our point prediction. I can’t fault Ioannidis for considering 10,000 deaths as a possibility; as Gorski says, the problem came later with a failure to fully reassess his models in light of the forecast’s (inevitable) errors. This is not a problem unique to any particular epidemiologist; similar issues arose in the other direction with Imperial College epidemiologist Neil Ferguson. Again, the point is not that Ferguson was unreasonable, just that making a prediction in a time of uncertainty is step 1 of a two-step process. Step 2 is going back afterward and assessing how your forecast did.

I wonder if some of these problems can be attributed to Jensen’s inequality. When forecasting the death toll from a pandemic, there are literally orders of magnitude of uncertainty. Some of this relates to the feedback in the stochasticity of the spread—an epidemic can die out or it can explode, and if it explodes it can become uncontainable—and some of it is just uncertainty about how fast it will spread and how many people it will kill. Suppose the forecast of deaths for a particular pandemic is in the range {100, 1000, 10,000, 100,000, 1,000,000}, with the lower and higher ends being less likely. To fix ideas, suppose the probabilities of these 5 outcomes are 5%, 20%, 50%, 20%, and 5%, respectively. Due to the practical difficulties of diagnosis and reporting, it might be that the three outcomes on the low end will be hard to tell apart, but let’s not worry about this here. In any case, the point is that the best estimate or most likely outcome here is 10,000, but the expected number of deaths is 75,000. And major economic and social damage could occur at the very high end. So it can make sense to prepare for a million deaths and to forecast an expected 75,000 deaths, even if 10,000 is the most likely. And all of that is with the probabilities known. With unknown probabilities, everything is that much harder. And then, after the fact, if the death toll is only 10,000, lots of people can run around accusing the “75,000” people and the “1,000,000” people of being unnecessary alarmists. My point here is not that it was wrong for Ioannidis and others at the beginning of the pandemic to consider low-end possibilities. Rather, I just want to emphasize the inherent difficulties of handling this sort of variation over orders of magnitude, and the importance of going back and understanding where past guesses were wrong.

P.S. There’s also some discussion about Ioannidis criticizing Meyerowitz-Katz for wearing big glasses and having a cat. That wasn’t cool for Ioannidis to do that, but I think that’s pretty much separate from the issues of statistics and epidemiology discussed above. We all do things that aren’t cool sometimes; Ioannidis just had the misfortune not to have an editor on that article who could’ve suggested he remove those passages.

235 Comments

  1. Michael Weisman says:

    It’s somewhat worse than you depict. Immediately when that first article came out, I had an email exchange with an epidemiologist pointing out that JI had combined a low-ball IFR that required assuming a high cryptic infection rate with an ultra-low limiting infection rate, the opposite limit. Either one could conceivably have been right, but the combination was essentially impossible.(You make the same point above.) Combined with other tendentious parts of the same paper, it’s clear this was not just an ordinary erroneous estimate. JI had some unstated agenda.

    • Joshua says:

      There’s more that goes beyond merely falling victim to orders of magnitude in uncertainty and lacking an alert editor.

      The worst, imo, is a fundamental epidemiological and scientific and logical error: Ioannidis went on national TV to promolgate an generalized IFR estimate that was derived by extrapolating from non-randomized and unrepresentative convenience sampling without even adjusting post-stratification for common predictors of health outcomes like race/ethnicity. I have yet to have someone explain to me how that is even remotely acceptable epidemiological practice.

      There’s more, that I’m not entirely sure is less aggregious. For example, when talking about the uncertainty in quantifying COVID deaths, Ioannidis focused on uncertainty in one direction only – the potential mixing of “died with” vs. “died from” COVID – and ignores the uncertainty that might have led to undercount. Another example is when talking about the Danish mask study, he ignored the significant limitations in the finding (see Andrew’s post on the topic) to focus only on a conspiratorial framing of the publication process.

      With Ioannidis’ science on COVID, there has been a uniform alignment in relation to the uncertainties – towards limiting the seriousness of the pandemic. I wouldn’t assume an “agenda,” as that would require mind-reading – but at some point when viewed in the context of more than one projection that were off by orders of magnitude, there is a systemic pattern in his direction of error that implies more than just a random distribution within the range of uncertainty.

      Then you mix that with his obvious and explicit affiliation with a particular group of policy advocates, and a consistent leveraging of political media to push his policy preferences.

      And then you add in his bizarre use of the scientific literature to launch a cheap personal attack. In the very least, this all undermines his self-cleansing claims of “pure” science focus and no political interest, and gives his editorial framing more than a tinge of sanctimony.

      • Hi Joshua,

        RE: “There’s more, that I’m not entirely sure is less aggregious. For example, when talking about the uncertainty in quantifying COVID deaths, Ioannidis focused on uncertainty in one direction only – the potential mixing of “died with” vs. “died from” COVID – and ignores the uncertainty that might have led to undercount.
        —–
        John does go into more detail in his July 2020 article, that I cited earlier, as to why quantifying COVID deaths has been so challenging. He has noted, for example, that there can be substantial error rate. Heck even Sander Greenland has a post on this blog about this problem.

        https://pubmed.ncbi.nlm.nih.gov/1871957/

        A subset of medical experts have pointed to overdiagnoses and other such issues in medicine, the concerns of which have been aired in so many science and medical journals. John Ioannidis has presented his perspectives in many scientific fora. So ok he has a cautionary approach to projections made. That is one approach. The other side of it is that there have been many doomsday projections as well. Our propensity to align with one approach is determined by many heuristics and biases that we acquire through our lives. A good number of physicians have written about their disillusionment with medical treatments. It’s worth reading such critiques. The Evidence Based Movement has continued to echo degrees of disillusionment and frustrations.

        We may not ever come close to calculating an accurate IFR, unless we devise a testing surveillance public health program. Now that may be a possibility as a conwquence of Michael Mina’s efforts.

        The state of knowledge is itself in flux in so many endeavors. Also we know far less than we want to.

        • Sorry, I meant I was referring to the accuracy of DEATH CERTIFICATES Eeekkkk!

        • Here is a response to the BuzzFeed article about John Ioannidis

          The John Ioannidis Affair: A Tale of Major Scientific Reaction

          https://www.scientificamerican.com/article/the-ioannidis-affair-a-tale-of-major-scientific-overreaction/

          • Andrew says:

            Sameera:

            I think that article is absolutely horrible. It was so bad that Scientific American followed up to disown many problematic things about it.

            • Hi Andrew,

              I was a member of Right Care Alliance, a project of the Lown Institute. Shannon Brownlee was its founder. I did not have an occasion discuss the article nor anything else about John Ioannidis at the time b/c I was too busy with some other projects. I was surprised that they knew each other as well as they did. I did know that John Ioannidis had given a keynote at the Lown Institute which was well received and that Shannon had co-authored an article which I haven’t read yet.

              I have not been actively involved with the Right Care Alliance for a year and a half.

              John has loyal friends and colleagues, as you and others who post here do. So I didn’t view the defense of John as a political maneuver but as one of enduring loyalty.

              It’s not my way to besmirch people with whom I disagree with or even dislike. I saw too much of that and it strikes me as toxic.

              • Rahul says:

                Sameera:

                I think loyalty isn’t what’s at question here but full disclosure by interested parties. If I write a defense of a long time collaborator readers ought to know this fact.

              • Andrew says:

                Sameera:

                Tha authors of that article misrepresented what I wrote and they misrepresented the reporting of Stephanie Lee, and then when I contacted them to ask them to correct these misrepresentations, they refused. Then a few months ago they did it again. Perhaps they could find a way to be loyal to their friends and colleagues while still writing the truth about others. Or perhaps they do not have the ability to thread this needle, and to them loyalty is more important than truth. In that case I think they’d do better to just not write about the topic at all, if they would otherwise find it necessary to misrepresent the actions of others in order to make their point. In any case, I’m glad that Scientific American called them on it. That showed integrity on the part of Scientific American’s leadership.

              • Hi Andrew,

                Sorry to hear that you were not able to resolve the misunderstandings. I know that can be very. frustrating. The progression of the situation seems so complex. Without John directly addressing the critiques, some of us can only second guess the behind the scenes dynamics.

                My brief experience of Shannon Brownlee [ don’t know Jeanne Lenzer] is that she is a straightforward person. I strongly doubt that she intentionally hid that the fact that she and Jeanne co-authored articles with John. In reviewing John’s keynote at Lown Institute, I would guess that Shannon and John share similar viewpoints about the challenge of conflicts of interests and constraints posed by the lack of reliable data in so many cases.

                .

                Scientific American added that disclosure right off the bat. So good for the magazine.

                I will reread that UnDark article if it is still available. I never heard of UnDark until the controversy erupted.

                And it’s unfortunate when people misquote or misinterpret our statements.

    • Rahul says:

      What would this unstated agenda be? Anyone know or have a guess in hindsight?

      • Joshua says:

        To the extent that Ioannidis has an “agenda,” my assumption is that it is to do good science and save lives.

        It isn’t his agenda that I question. I question his science. And I question the degree to which he has allowed “motivated reasoning” to affect his science.

        Another example – one of many: his comparison of COVID to the seasonal flu: It’s almost as if he doesn’t understand the basic dynamics of compounding growth – which isn’t plausible.

        Let’s say that he’s right about the IFR of COVID. So then he says that it’s not that far away from the flu (in absolute terms). But the problem is that a small difference in IFR (coupled with a somewhat small difference in R0 in absolute terms), and you wind up, after months of unchecked growth, a huge difference in the number of deaths.

        But then we need to go further. From what I’ve seen Ioannidis almost always concentrates on deaths. He ignores the hospitalization rate, which is obviously important – and he ignores the much higher infection hospitalization rate from COVID when he compares COVID to the flu.

        As I mentioned above, he only talks about the uncertainties that might lead to an overcount and ignores the uncertainties that could lead to an undercount.

        He makes facile statements attributing negative outcomes to COVID without speaking to how such an assessment is based on counterfactual assumptions (that things wouldn’t have been worse absent interventions) for which he has very little evidence.

        There’s a long list of examples for how poorly he’s conducted science related to support his COVID activism. I don’t begrudge him wanting to be an activist. I wouldn’t expect someone to sit on their hands in such an important situation.

        But I’m rather shocked at how bad his science has been.

        • Chris Wilson says:

          +1 and to a lot of the commentary below.

          Uncertainty in multiplicative risks, ipso facto supports stronger precautionary approaches than uncertainty in additive domains. This is the core insight of Taleb – although far from unique, and as others point out, he gets a lot of unnecessary mileage for making the same point in various ways. But it is still absolutely correct! And Ioannidis, for all his heavyweight background, still either doesn’t get it, or doesn’t want to.

          Moreover, like Joshua I fault Ioannidis for going out pre-emptively to press with his mistaken analyses (e.g. that Santa Clara study that took many of us 5-10 minutes to show its analytical flaws), and even sought out contrarian Covid Truther podcasts to share his point of view.

          I don’t think he has malign intent. I think he had a gut intuition that he couldn’t let go of, and runs in some Stanford circles who lean politically independent to conservative. Those guys (not necessarily him) got politicized and it may have leached off on him, but this is going into realms of speculation that are not necessarily helpful. So I’ll stop there :)

  2. Kam says:

    >>> “My point here is … to emphasize the inherent difficulties of variation over orders of magnitude, and the importance of going back and understanding where past guesses were wrong.”

    OK, so looking back 13+ months at official US government guesses on coronavirus mortality — were they wrong ?
    If so, why ?

    And what is the margin of error (if any) in officially designating Covid-19 as the cause of death for hundreds of thousands of deceased people ?

    • Andrew says:

      Kam:

      Oh, yeah, the U.S. government response was terrible. See the above-linked article by Lawrence Wright for lots of details. To get more statistical, here’s an example of a government forecast that was clearly ridiculous at the time it was produced. What was particularly bad about this case was that the government official in question attacked his critics for not having “peer-reviewed scientific work and academic appointments.” As if having a government job, peer reviewed papers, and a fancy academic chair is more important than getting things right. I can’t stand that swaggering attitude. If these dudes had spent less time preening and more time telling us all to wear masks, maybe we wouldn’t be in the bad place we are today.

    • Anoneuoid says:

      And what is the margin of error (if any) in officially designating Covid-19 as the cause of death for hundreds of thousands of deceased people ?

      We know George Floyd met the criteria for a covid death.

      Lots of people died early on due to inappropriate medical interventions, maybe they still are. I am not up to date on what they are doing to covid patients these days.

      Also stress and poverty are no longer considered to increase mortality rates apparently.

      Finally, I don’t know where Ioanndis got his 1% infected assumption, but I am sure he did not account for sending the covid patients into nursing homes.

      • Phil says:

        Anon,

        Let me take these in order:

        1. As far as I can determine, George Floyd did not “meet the criteria for a COVID death”. https://www.medpagetoday.com/blogs/working-stiff/88141 for example. And https://www.cdc.gov/nchs/data/nvss/coronavirus/cause-of-death-data-quality.pdf I think you’re just wrong about this one. By extension, I think you’re wrong about what I think you’re implying, which is that there’s a significant overcount of COVID deaths. If you think COVID deaths are significantly overcounted — by enough to change the way we should think about Ioannidis’s projections and his certainty in them, which would require a huge, huge overcount — then I suggest you say “I think US COVID deaths are over-counted by more than 400,000 [or whatever you think the number is], and here’s why I think this:” and then list your reasons.

        2. “Lots of people died early on due to inappropriate medical interventions, maybe they still are. I am not up to date on what they are doing to covid patients these days.”

        When you say “lots”, I’m not sure whether you’re talking about a percentage or an absolute number. If 50K people died due to inappropriate medical interventions, that’s a lot in absolute terms — the number of U.S. combat deaths in Vietnam — but it’s less than 10% of official US COVID deaths at this point. Do you really think the number of COVID deaths due to inappropriate medical interventions is so large that it accounts for a substantial part of the difference between Ioannidis’s projections and the actual course of the pandemic? That seems to me to be a ridiculous suggestion, that something like 90-95% of COVID deaths are due to that. I don’t even think 20% would be a reasonable estimate.

        I realize there’s some interaction with hour first point: maybe you think COVID deaths are over-counted by a factor of 3, and of the remaining third, you think 80% of them are “due to inappropriate medical interventions”, so when you put these together, the count of COVID deaths that aren’t due to inappropriate medical interventions is only 40K or whatever. This still seems ridiculous to me. Is this really what you believe, and if so, could I trouble you to explain why?

        3. Stress and poverty… I have no idea what you’re saying there. It’s widely recognized that stress is harmful, and that poverty is associated with poor health and shorter lifespans for a variety of reasons. When you say these things are “no longer considered”, by whom do you think they are no longer considered? Or are you just saying “poverty” is not listed on death certificates, or something like that? It never was, as far as I know.

        4. “Finally, I don’t know where Ioanndis got his 1% infected assumption, but I am sure he did not account for sending the covid patients into nursing homes.”

        I agree that Ioannidis probably didn’t think in detail about what might or might not happen with elderly sick patients in hospitals and where they might be sent as the hospitals filled up with patients. For one thing, it seems that he didn’t think hospitals anywhere would be overwhelmed with seriously ill people. Indeed, his failure to imagine such a scenario is of a piece with his dismissal of the suggestion that things might get quite bad. But even if Ioannidis had considered the possibility that there might at least be local saturation of critical care capabilities, he was just trying to get an order-of-magnitude estimate and I doubt he would have thought things through in so much detail. So, as a criticism of Ioannidis this comment seems beside the point: the problem with his estimates, and his presentation of his thoughts, is not that he failed to consider how some locales might respond if they had more patients than they could handle.

        But I don’t think your comment is intended to be a criticism of Ioannidis. I’m not sure what it’s intended to be. To try to be extremely generous to you, I’ll assume you are trying to contribute to this discussion and you’re pointing out the difficulty in predicting how governments, hospital administrators, nursing home managers, etc., might respond to a pandemic. If that’s your point then: yes, sure, when Ioannidis was making his predictions a year ago it was very unclear how all aspects of society might respond to the coronavirus outbreak. But, logically, that should have made Ioannidis _less_ sure about his projections.

        Andrew generally discourages personal comments on this blog, but I’m going to put one here anyway:

        Anon, I think I am not the only one who finds your comments frustrating. You sometimes make genuine contributions to this blog and its discussions — I learned some things from you about Vitamin D, for example — and I hope you continue to do that. But often…far, far too often…your posts just seem pointlessly snarky. Here I’ve pretended that your snarky comments are intended to contribute to the discussion of how hard it is to forecast something like the course of a pandemic, and how to handle the resulting uncertainties, but I don’t think that’s actually the case. I don’t think you were trying to contribute in any way. Sometimes you just seem mad at the world and your response is mockery and derision. Hey, it’s the Internet, I’m well aware that there are always going to be trolls who want to get a rise out of people. It usually doesn’t bother me, but with you it does bother me because I know you’re capable of doing so much better.

        • Anoneuoid says:

          Start with #1:

          DR. DEBORAH BIRX: So, I think in this country we’ve taken a very liberal approach to mortality. And I think the reporting here has been pretty straightforward over the last five to six weeks. Prior to that when there wasn’t testing in January and February that’s a very different situation and unknown.

          There are other countries that if you had a preexisting condition and let’s say the virus caused you to go to the ICU and then have a heart or kidney problem some countries are recording as a heart issue or a kidney issue and not a COVID-19 death. Right now we are still recording it and we will I mean the great thing about having forms that come in and a form that has the ability to market as COVID-19 infection the intent is right now that those if someone dies with COVID-19 we are counting that as a COVID-19 death.

          https://m.youtube.com/watch?v=blZpgra3XbU

          So, you are wrong. And if this policy changed at some point then all these charts looking at death rates over time are misinformation.

          • Joshua says:

            Anoneuoid –

            Intersting thst you’re taking the word of a politically appointed public health official as gospel. I wouldn’t expect that from you.

            Regardless, given that uncertainties about “COVID deaths” run in both directions – is your argument that in the US, overall there is a significant overcount?

            Part of the rhetorical playbook of many people who think that interventions have cost more lives than they’ve saved has been to address uncertainties related to counting “COVID deaths” in a very one-sided manner (mixed in with promoting conspiracy theories about how medical professionals have been falsifying deaths to hurt Trump and profit financially). Ioannidis has treated the uncertainties very selectively.

            No doubt, I’d guess that some number of people have been counted as dying from COVID where that designation could be fairly questioned. But I’d say the salient question here is how big that number is in comparison to the number running the other way – such as those who died at home or on the streets or in a nursing home but never had an attending physician, never took a COVID test, etc.

            I thought that Phil was asking you to be more specific related to that comparison.

            (Andrew – I’m hoping that’s not overstepping you’re request that I not belabor points of disagreement with Anoneuoid)

          • Phil says:

            Anon,
            The only statements I can find about Floyd and COVID say exactly the opposite of what you’re saying: they say that the coroner determined that although Floyd tested positive for COVID it was NOT a contributing factor in his death. You may be right that Floyd’s death certificate lists COVID as a cause but you’ve presented no evidence that that’s the case, and it contradicts what I’ve seen.

            As for how COVID tests are counted: If you are genuinely interested in this issue, which now seems to me unlikely, you might start here: https://www.aamc.org/news-insights/how-are-covid-19-deaths-counted-it-s-complicated and follow the links.

            Here’s one section:

            The CDC uses different sources to post slightly different fatality figures.

            Death certificates provide the data for the agency’s daily updates of COVID-19 deaths. In mid-February, the total stood at 462,000.
            At the same time, the agency’s COVID Data Tracker reported 486,000. That data comes from the National Notifiable Diseases Surveillance System (NNDSS), which gathers information from state and local health departments when a disease is diagnosed in someone in a health care setting. (Estimates vary among media outlets and other organizations because they use various sources, including the CDC.)

            Why use two counting methods? The NNDSS provides “real time” awareness of fatalities linked to a disease, Robert Anderson, PhD, chief of the Mortality Statistics Branch at the CDC’s National Center for Health Statistics, says via email. He explains that it takes about two weeks longer for death certificate information to work its way up to the daily updates.

            “Ultimately, the official numbers will be based on the death certificates,” Anderson says.

            If you care to read about this, you’ll find that some states initially counted everybody who died with COVID as a COVID-related death, but then changed that policy…and went back and changed the numbers so that the time series would be correct. Here’s Colorado, for example: https://www.coloradoan.com/story/news/2020/05/16/colorado-changes-how-coronavirus-deaths-state-counted/5198485002/

            • Anoneuoid says:

              You may be right that Floyd’s death certificate lists COVID as a cause but you’ve presented no evidence that that’s the case, and it contradicts what I’ve seen.

              I never said this. I said he met the criteria of dying with covid.

              • Phil says:

                Anon, you said “We know George Floyd met the criteria for a covid death.”

              • Anoneuoid says:

                Anon, you said “We know George Floyd met the criteria for a covid death.”

                Yes, which was announced to world as dying with a positive covid test. Personally, I think it was unlikely he got included in the stats, but I have no idea how to tell.

                What states report and what the cdc reports is different, as you say they are changing the criteria and past numbers *somehow*, etc.

                Return to the original post:

                And what is the margin of error (if any) in officially designating Covid-19 as the cause of death for hundreds of thousands of deceased people ?

                We know George Floyd met the criteria for a covid death.

            • Anoneuoid says:

              If you care to read about this, you’ll find that some states initially counted everybody who died with COVID as a COVID-related death, but then changed that policy…and went back and changed the numbers so that the time series would be correct.

              I hope readers will focus on this. You don’t even realize you are making my point for me.

              • dhogaza says:

                You are asking us to focus on the point that states like Colorado went back and made their numbers more accurate?

              • Andrew says:

                Anon:

                Stop, please! A bit of discussion is fine, but too much on one point just clutters the comment thread.

              • Phil says:

                Anon,
                I find it pretty much impossible to tell what your point is with any of this stuff. This is why I suggested that, instead of making a cryptic and apparently erroneous statement about George Floyd, you tell us why you think there is a significant over-count of COVID deaths, if that is in fact what you believe. Ditto for all of your other statements.

                Or, you can just keep being snarky and argumentative without contributing anything. You’re allowed to do that. I just wish you wouldn’t.

              • Anoneuoid says:

                you tell us why you think there is a significant over-count of COVID deaths, if that is in fact what you believe.

                I never said this!

                This strawman stuff is getting so extreme on this blog.

                Can you share sources that account for these other factors?

                The answer is no. They all just set them to zero like the one andrew linked to in the OP. I am sorry that science is becoming uncomfortable for people but your choice is science or ignoring inconvenient info. Choose the latter and more people will die.

              • Anoneuoid says:

                Choose the latter and more people will die.

                Not you of course, or me. We are largely irrelevant. But the collective choice.

                If you want to save lives, encourage sharing of accurate information so people can make informed cost-benefit based decisions.

              • Anoneuoid says:

                I mean, here is what I was concerned about March 2nd 2020:

                https://old.reddit.com/r/COVID19/comments/fc8f7z/what_the_role_of_mechanical_ventilation_on_the/

                Then look what happened. Were you concerned about this at that time?

                My goal is to be informed as share that information to save lives.

              • Andrew says:

                Anon:

                No more. I appreciate discussions, but this is just making the blog less useful when you fill up the comment thread like this.

              • Anoneuoid says:

                No more. I appreciate discussions, but this is just making the blog less useful when you fill up the comment thread like this.

                Sorry about that, I do understand and didn’t see your other comment before all those posts.

        • Anoneuoid says:

          Btw, this longstanding concern has finally been confirmed:

          Altogether, our results suggest that vaccinees probably do not elicit an early humoral response detectable at mucosal surfaces. They strengthen the hypothesis that some vaccines may not protect against viral acquisition and infection of the oral–nasal region, but may prevent severe disease associated with viral dissemination in the lower respiratory tract. Our results are in line with those obtained in nonhuman primates, in which vaccinated and then challenged animals display detectable viral loads in nasal swabs but not in lower airways37.

          https://www.nature.com/articles/s41591-021-01318-5

          So the vaccines do not induce mucosal immunity.

          • Navigator says:

            The paper you linked states:

            Our results indicate that B1.351, but not B.1.1.7, may increase the risk of infection in immunized individuals.

            I believe it’s been known for a while that there is no sterilization of the virus, just decreased load and, most importantly, protection from hospitalization and death.

            • Anoneuoid says:

              Here we demonstrate that two recently emerging mutants in the receptor binding domain of the SARS-CoV-2 spike protein, L452R (in B.1.427/429) and Y453F (in B.1.298), can escape from the HLA-24-restricted cellular immunity. These mutations reinforce the affinity to viral receptor ACE2, and notably, the L452R mutation increases protein stability, viral infectivity, and potentially promotes viral replication. Our data suggest that the HLA-restricted cellular immunity potentially affects the evolution of viral phenotypes, and the escape from cellular immunity can be a further threat of the SARS-CoV-2 pandemic.

              https://www.biorxiv.org/content/10.1101/2021.04.02.438288v1

              Here they come…

              Still, I am pretty sure it can’t be as bad as last year unless some new poorly thought out intervention is added.

        • Anoneuoid says:

          Let us go to point #2:

          When you say “lots”, I’m not sure whether you’re talking about a percentage or an absolute number. If 50K people died due to inappropriate medical interventions, that’s a lot in absolute terms — the number of U.S. combat deaths in Vietnam — but it’s less than 10% of official US COVID deaths at this point. Do you really think the number of COVID deaths due to inappropriate medical interventions is so large that it accounts for a substantial part of the difference between Ioannidis’s projections and the actual course of the pandemic? That seems to me to be a ridiculous suggestion, that something like 90-95% of COVID deaths are due to that. I don’t even think 20% would be a reasonable estimate.

          Fine, what is the percent due to that? Currently it is set to zero in most discussions, eg: https://sciencebasedmedicine.org/what-the-heck-happened-to-john-ioannidis/

          • Anoneuoid says:

            Btw, I don’t mean forum speculations like 10-20% based on unclear methods. I mean is there a single published paper that tries to account for this?

            Can you share a single one?

  3. Michael Weissman says:

    Joshua- We’re in agreement. But IIRC in that first paper JI did adjust for some characteristics of the cruise ship population, in particular age. So that lowered the estimated population IFR. He didn’t adjust for other variables, like not being too frail to go on a cruise, that would have raised the estimated IFR. Completely tendentious, as we both agree. Again, it was obvious at the time.

    • Joshua says:

      Michael –

      > But IIRC in that first paper JI did adjust for some characteristics of the cruise ship population, in particular age.

      Sure – but the very idea of extrapolating from a cruise ship population – and living conditions on a cruise related to spread of an infectious disease – to a general population in normal living conditions seemed like a scientific parody to me.

      > He didn’t adjust for other variables, like not being too frail to go on a cruise, that would have raised the estimated IFR. Completely tendentious, as we both agree. Again, it was obvious at the time.

      Yup. Yes, they did adjust for age and I’m not saying don’t look at the DP and calculate the IFR, but trying to say such a situation is generalizable, even with post-stratification adjustments for age, looks to my non-PhD-in-epidemiology-eyes as bizarre.

      But then he later extrapolated from the Santa Clara – obviously non-random – sample w/o even adjusting for such an obvious correlate with health outcomes as race/ethnicity? I mean what? And then went on a national TV campaign to promote that IFR and ridicule other scientists (for promoting “science fiction and projecting from inadequate data) as he said that COVID was basically as much of a threat to public health as the seasonal flu?

      • Rahul says:

        Hypothetically, had one of us had to issue our best forecast on the day JI did, how would you have gone about it?

        I would love to hear the alternative analysis with the constraint that you should put yourself ( best as you can) at that point of time sincerely: you are not allowed to even implicitly color your forecast with anything you know since then.

        I think this will be an interesting thought experiment.

        • Andrew says:

          Rahul:

          Speaking just for myself, I would’ve found it very difficult to issue a forecast, even a probabilistic forecast, given my state of knowledge at that time. And, as I wrote in my above post, I don’t hold it against Ioannidis that his published forecast (or estimate, or projection, or hypothetical scenario, or whatever you want to call it) was way off. It’s the nature of highly uncertain outcomes that point forecasts can be way off. I do think that Ioannidis could’ve done better at assessing where and how his forecast went wrong. For example, his projection of only 1% of the U.S. population ever being exposed to the virus was pretty much directly contradicted by the high-profile claims of the Stanford study he was involved in the very next month. It would’ve been appropriate for him to acknowledge that contradiction, work through its implications, and learn from the discrepancy.

          • Rahul says:

            Let’s assume no forecast is not an option on the table.

            Just for Fun, imagine your state of mind a year ago: what number would you have picked for say total deaths had you been forced to.

            • confused says:

              At the time, I would have looked at the estimates of IFR, assumed they were probably somewhat high due to under-reporting of mild/asymptomatic cases, compared that to past pandemics, and said (as I did say in April) that this would probably be comparable to 1957/1968, maybe a bit worse, but clearly less severe than 1918.

              Adjusting 1957 death rate to current population numbers would suggest about 200,000 deaths in the US and maybe 2.5 to 10 million worldwide.

              This is a very unscientific way of doing it, but doesn’t seem to have been much farther off than the serious professional estimates! (Too low for the US – though maybe not for the world depending on how quickly vaccination spreads, and whether the low per-capita deaths in the Old World tropics are real vs. an effect of underreporting.)

          • Joshua says:

            Andrew –

            > For example, his projection of only 1% of the U.S. population ever being exposed to the virus was pretty much directly contradicted by the high-profile claims of the Stanford study he was involved in the very next month.

            His mention of 1% was probabilistic. It’s hard to tell from that article how likely he felt that % might be. I think it’s important to be clear about that. Since Ioannidis always stresses one end of the spectrum of the probabilities (e.g., in April 2020 he said that we had probably peaked in COVID deaths in Europe and the US) the temptation is to peg him as not actually considering a range. One problem with Ioannidis, IMO, is that he treats the probabilistic aspects in such a fashion. He did that when he said that the IC modelers “predicted” 2 million deaths (and that such a prediction was “science fiction”). It bugs me when people like Ioannidis ignore the probabilistic aspect of the Imperial College projections.

            But in the climate change world, there’s a pattern where scientists make probabilistic statements about uncertainty, consumers of that information then ignore the probabilistic aspect, cherry-pick the low or high end of the range as suits their “motivations,” and then shout that the scientists were wrong, and complain that the scientists are over-certain and don’t properly treat uncertainties.

            • confused says:

              Yes, he did mention that the high end (which he didn’t believe) was something like 40 million deaths worldwide.

              The problem I have is that the 1% infection is incompatible with the low IFR assuming high rates of undetected infection, so multiplying those two doesn’t work.

              It would have made sense to give *separate* estimates for those *different* scenarios (say 1% infected at 1%-2% IFR; 20%-33% infected [comparable to flu pandemics, IIRC] at 0.3% IFR).

  4. Ron Kenett says:

    To clarify the dissuasion with respect to COVID19 and consider the role of Ioannidis, I would distinguish between:
    1. The various stages of statistical intervention
    2. The methods and tools for statistical intervention
    3. The role of the statistician and biostatistician

    My take, in considering the role of Ioannidis in all this, is mixed. I cannot avoid the thought that attracting media attention seems to have played a major role in terms of motivation.

    Here is a brief of what I experienced in Israel, with some generalization.

    The Covid-19 pandemic is still considered as an emergency. However, one cannot really talk about an emergency, when a health situation lasts for more than one year and includes an evolution of policy strategies, population behavior and scientific knowledge. The role of a medical researcher, in general, has been evolving from handling a first panic phase (March 2020), when the search for diagnostic tools, medication approaches and hospitalization procedures was almost emotional and based on trial and error strategies, until today, when the focus is on understanding the vaccine effect and on planning the best vaccine strategies. In parallel, the role of statisticians and biostatisticians has been evolving to face different challenges and needs of biomedical research, in the hope of mitigating the dramatic impact of COVID-19 on society, the health and the economic systems.

    Phase I: Data collection and storage
    During the first phase (February 2020-June 2020) statisticians working in hospitals have been involved in designing and building a biobank for storing COVID 19 data. Initially, this was complicated by the lack of accuracy of diagnostic tools. No unique criteria were provided for defining variables (even the definition of “positive” was not unique and comparable). At that stage, statisticians were coordinating data collection and storage trying to ensure some quality control. In Israel, Prof Gamzu was appointed as COVID19 project coordinator in July 2020. One of his first steps was to reorganize the ministry of health data base. Because of politics, Gamzu was not able to implement differential restrictions (the red light system) and restrictions were applied universally, if needed, or not. Overcontrol was also prevalent due to lack of awareness and methods for process control methods such as change point detection.

    Phase II: Data analysis
    During a second phase (June 2020-November 2020), statisticians have worked on analysing clinical data and data from basic research. Basically, this was a very important phase where many statisticians tried to find a modelling approach, and extrapolate from information on epidemiology parameters such as mortality, fatality rate, incidence rate and predictors of disease evolutions. The search for appropriate statistical approaches, in dealing with highly correlated covariates and variables, has been a primary goal to help biomedical research identify COVID19 risks factor. In Israel, physicists and computer scientists lead modeling efforts. A call by the Israeli National Statistician, and general manager of the central bureau of statistics, to run a national serological survey to map infection rates was largely ignored. In fact, Seigal Sadeszki, who was the director of public health in the Health Ministry, stated on TV that testing is not useful as it does not cure people.

    Phase III: Design and surveillance
    In a third phase, statisticians have been involved in study design (planning new prospective studies) and surveillance studies. This includes i) handling issues on sampling (that is designing a representative statistical sample to follow longitudinally evolution of immunity related factors) ii) identification of individual predisposing factors to define not only surveillance procedures for frail groups of infected people but to develop vaccination strategies and iii) to identify priority groups for vaccination. In Israel, with a world record deployment of vaccinations (Pfizer), a major study by the Clalit healthcare system assessed vaccination efficiency.

    Again, considering such phases can help assess the writings of Ioannidis.

    This blog, although not specifically addressing epi and medical topics, seems to have done much more….

  5. Emilio Lobato says:

    I think the attacks on the grad student are related, given that the overall message of Gorski’s article is that Ioannidis seems to be suffering from a bad case hubris.

  6. Roy says:

    Not commenting on Ioannidis but your comment in the paragraph starting with “Jensen’s inequality” lines up a lot with what Taleb has argued about the pandemic. In an exponential or multiplicative situation, even small errors in a parameter estimate can lead to large differences in projected outcomes.

    • Rahul says:

      Which may be true but ( like most points Taleb makes) so what? Is it actionable in any way?

      How do we use that fact to make better forecasts or even to remotely help our fight against covid in any way. All that says is forecasts will be difficult.

      To me this is a recurrent theme with Taleb: apparently profound but eminently useless.

      • Barney says:

        It was very much actionable at that time! Ioannidis was suggesting that absent better quality data, the lockdown measures were an overreaction (we couldn’t be sure it was bad enough to warrant a lockdown given the trade-offs). Taleb agreed with the point that there was a lot of uncertainty, but argued that uncertain situations are precisely those in which it is crucial to be cautious. Caution consists in taking the action that minimizes the possibility of the worst outcome, ie lockdown.

        • jim says:

          But doesn’t it have another more significant meaning in the context of the time and for the future?

          If your data are highly uncertain your model doesn’t tell you anything. You can’t use it to justify any action. Its outcome is meaningless.

          • Barney says:

            I don’t see why you can’t use uncertainty to justify action. If you are uncertain about the depth of a fast flowing river, don’t try to cross it. When there was massive uncertainty about the coronavirus, the prudent thing to do was take preventative action by locking down. From a decision theoretic point of view, the expected value is so massively dominated by worlds in which the worst things are true that one needs to act so as to avoid catastrophe in those worlds.

            • Joshua says:

              +1

              One thing this pandemic underscores is what a difficult time some people have evaluating “far tail” risks in the face of high uncertainty.

              • jim says:

                “what a difficult time some people have evaluating “far tail” risks”

                Yes, and some people have a difficult time making any useful analysis whatsoever because they’re so busy relying on cliches about analysis.

            • jim says:

              “if you are uncertain about the depth of a fast flowing river, don’t try to cross it. “

              I made no such statement nor anything similar.

              Through this entire pandemic there is not one single data analysis that I ever saw that could reasonably contribute to decision making. The data analysis were entirely useless.

  7. Alain says:

    >>> The problem is that Ioannidis didn’t seem to be able to let go of his initial downplaying of the risk.

    “For whoever exalts himself will be humbled, and whoever humbles himself will be exalted.” Matthew 23:12

  8. It is a little strange that Ioannidis is writing papers like this. After all, it isn’t neccessary to downplay the effects of the virus in order to oppose lockdowns.

    That said, I take issue with the “From my reading, it was a solid meta-analysis and review” of the Meyerowitz-Katz paper. I have a blog post from a few months ago that looked at this paper at https://categoricalobservations.com/2020/10/26/replication-when-it-matters/, and I don’t think the meta analysis is particularly good. Then again, back then I set out to do it better myself with Bayesian techniques and quickly gave up because I couldn’t find the relevant data.

    • Joshua says:

      > Gideon Meyerowitz-Katz co-authored a paper which came up with a mainstream estimate, as I understand it…

      There are others. The “mainstream estimate” I’d say comprises a number of meta-surveys in addition to thst one.

      In all fairness, Ioannidis’ meta-survey “global” estimate includes very large populations of much younger median age (like all of Africa and India) and as such, that estimate shouldn’t be considered applicable to the US.

      That said, he himself has been very inconsistent in how he has applied his IFR estimates – sometimes applying a single estimate to widely varying populations. Of note, he did that early on to promote an of IFR around 0.23% (as I recall) and imply during his TV publicity campaign that number applicable for the US generally, which seems clearly implausible for the whole population given it would mean that some 80% of Americans have been infected, and obviously more are going to die. And he did that back when, given the data he was using, the IFR should have been even higher than it would be now, since treatment has improved.

      • Navigator says:

        Joshua,

        Improved treatment led to fatality decrease among hospitalized patients, but that doesn’t mean IFR increased, decreased or stayed the same as we don’t know the ‘dark data’ at any given time (true number of infected people, which varies region to region).

  9. dhogaza says:

    I don’t think the current kerfuffle has much, if anything, to do with Ioannidis’s original article which included 10,000 as a possible lower bounds on the deaths the US would experience.

    It centers around his continued publishing of papers which some find poorly done which purport to show that the IFR of covid-19 is, in actuality, no worse than a bad seasonal flu.

    The latest paper which includes the personal attacks mentioned above concludes that global IFR is 0.15%. This point estimate applied to the confirmed covid deaths in the US implies that virtually everyone in the US has had the disease at this point. It is nowhere near the mainstream estimates of about 0.5%-1%, a range that is consistent with mainstream estimates that 20%-30% of the US at this point has had either an asymptomatic or symptomatic case of covid-19.

    Gideon Meyerowitz-Katz co-authored a paper which came up with a mainstream estimate, as I understand it, and in the past has publicly criticized Ioanndis’s lowball IFR numbers on his popular twitter account.

    Anyway here’s Gideon Meyerowitz-Katz’s twitter thread on the subject.

    https://twitter.com/GidMK/status/1376304539897237508

    • Carlos Ungil says:

      > The latest paper which includes the personal attacks mentioned above concludes that global IFR is 0.15%. This point estimate applied to the confirmed covid deaths in the US implies that virtually everyone in the US has had the disease at this point.

      To be fair, that 0.15% is “global” and the US population is quite different from the global population (16% and 8% over 65 respectively).

      • Joshua says:

        Oops. This comment should have been nested here. And apologies to Carlos as I didn’t see your comment before I wrote mine.

        https://statmodeling.stat.columbia.edu/2021/03/30/a-tale-of-two-epidemiologists/#comment-1775044

        • dhogaza says:

          “In all fairness, Ioannidis’ meta-survey “global” estimate includes very large populations of much younger median age (like all of Africa and India) and as such, that estimate shouldn’t be considered applicable to the US.”

          It is primarily US policy that he’s been focused on influencing.

          • Joshua says:

            dhogaza –

            > It is primarily US policy that he’s been focused on influencing.

            No doubt. But that doesn’t mean he’s arguing that the 0.15% IFR would apply for the US.

            • dhogaza says:

              I’ve never seen him argue that the IFR in the US is in the range of 0.5-1% either …

              Didn’t the original draft of the Santa Clara study come up with a point estimate of something like 0.17% or the like?

            • confused says:

              Not in defense of Ioannidis’ low estimates for the US (eg Santa Clara): but the much lower per-capita fatality rate in the Old World tropics, vs Europe, US, or the New World tropics, *could* mean that the global IFR is indeed much lower than US/Europe IFR – as a large part of the world’s population is in South Asia, Southeast Asia, and Africa.

              (It could also mean that the infection rate there is surprisingly low.)

      • Steve says:

        To be extra fair, how on earth is Ioannidis estimating the global IFR of Covid when we know that lots of places on earth don’t even have good data? Ioannidis is a contrarian who came up with a bad estimate early in the outbreak and has dug in based on even weaker arguments.

        • Steve,

          John Ioannidis has continued to highlight that the data is not reliable. I think that it is pointed out in that article as well. The quality of the data has been a problem all along. Why? The experts did not see the need to test asymptomatics, as John Ioannidis maintained from the start of the pandemic.

          • dhogaza says:

            “John Ioannidis has continued to highlight that the data is not reliable.”

            But in ways that are favorable to his assertion that the IFR is close to that of the seasonal flu and favorable to his positions on policy.

      • dhogaza says:

        What Steve said.

  10. Steve says:

    Andrew writes: “My point here is not that it was wrong for Ioannidis and others at the beginning of the pandemic to consider low-end possibilities. Rather, I just want to emphasize the inherent difficulties of handling this sort of variation over orders of magnitude, and the importance of going back and understanding where past guesses were wrong.”

    I think that you are too charitable to Ioannidis. He and others were clearly engaged in advocacy not armchair science. If a fire alarm goes off, and I tell everyone that I’ve been a fire warden and 90% of the time those alarms are false, just sit put, I’m sure it will shut down soon. If people die, I am morally and perhaps legally responsible. In the face of assymetric risks, it is immorral to advise people to take the least safe path. On the one side, the pandemic can explode (in January, you could already see in China’s data that deaths were doubling every 3 days). On the other, you have an economic slow down that can be immediately reversed if suddenly its discovered that Americans are magically immune to a virus that kills other humans. This was not a guy who said, “We have estimated the risk and it is much lower than other estimates, but our data is too uncertain to be used for policy decisions and the risks if we are wrong are exponential.” He was saying quite the opposite.

    • dhogaza says:

      “If a fire alarm goes off, and I tell everyone that I’ve been a fire warden and 90% of the time those alarms are false…”

      Off topic, but oddly enough I had that experience on an upper floor of a business hotel in the DC area. Breakfast room full of Colonels and Lt. Colonels in uniform every morning, on their way to meet with Congressional staff etc.

      Anyway first morning I was awoken at sunrise by a fire alarm. Looked outside and could see fire trucks on the way, put a wet towel under the door jam, waited. They stood around, looked, did nothing else, alarm went off, they drove away.

      I was there for an entire business week and the same thing happened every morning …

  11. michael schwartz says:

    Last summer, when it was clear he was very wrong, I had an email exchange with Ioannidis about his estimates, and his whole “aggresively optimistic” view and approach (lots of interviews, op eds, etc…).

    In our personal exchange, he seemed very forthcoming, and openly admitted he got it way wrong.

    I guess it’s a lot easier to do that one-to-one than go on a big mea culpa tour on t.v.

    Of course, he was perfectly willing to go on a big t.v. tour when he was spouting media-friendly, attention-grabbing, contrarian opinions.

    Not sure what my point is here – just that humans are complex. He certainly seems like a very decent and smart dude from all of his work and my limited interactions. I don’t know how well I would react to a massive public humiliation like this either.

    • John has admitted he has made mistakes in several public interviews. I think that John’s being featured on Fox news raised angst But the authors of the Santa Clara Study did incorporate its critiques subsequently.

      • Joshua says:

        Sameera –

        They never walked back extrapolating an IFR from non-random convenience sampling and compounding that error by extrapolating from their methdologlically flawed estimate based on the Santa Clara data to a more broadly applied IFR without adjusting those data to make them representative.

        That’s the kind of science that led Ioannidis to compare COVID to the flu.

        Do you find that justifiable?

        Errors are one thing. Bad science is a other.

        • dhogaza says:

          Timeline error – he compared covid-19 to a bad seasonal flu season first, Santa Clara followed and was meant to support the assertion.

        • Hi Joshua,

          As I have suggested, that comparison to the flue was context specific to the Diamond Princess Case, Each consecutive month since March, John Ioannidis has continued to say that COVID19 is more lethal especially as it affects vulnerable and elderly population. In short there is a steep age gradient.

          Moreover, John has continued to highlight that we are drawing hypotheses from unreliable and little data. I don’t think there is real disagreement on that point.

          Finally, even good science can be misapplied to a specific situation. It’s a complex epistemic environment.

          Addendum: I don’t agree with John Ioannidis always. I am much less of a proponent of Statistical Significance, as one example.

  12. xyu says:

    just some wild thoughts. Is it possible that Dr. JI has been so carried away by his years of being different that his mindset became fixed to be different from others? True, his many articles criticized wrong practices in medicine and epidemiology, which I agreed with him most of them.

    I was very optimistic when I learned COVID in late Dec 2021(yes, the underground Chinese media were already circulating bad news before the official announcement). I didn’t believe there would be 100,000 deaths in the US, and I thought the Imperial College estimation is a bluff to paint the worst case scenario. I believed it would go away by May or June, just like SARS in 2003. If checking the history, virus jumps onto human beings all the times, and small outbreaks of new virus (or new variants) occur every few years. Most epidemic will die away without the public knowing it. The public health department will handle it quietly and the reports remain internal, and even if made online (or to the public), nobody cares.

    But reality kicked in and I was puzzled by the erratic behaviors and beliefs of the public, the government, and worst of all, the CDC and other acclaimed experts. Not wearing masks, deriding lockdown, and so on are unthinkable. These traditional NPIs worked since 1918, and nothing is new. However, watching the number of deaths and hospitalizations going up continuously is very saddening and painful. It is a sign of failure in epidemiology and public health system and also an indication of selfishness of many Americans.

    • Andrew says:

      Xyu:

      One thing that bugs me is that the anti-lockdown people aren’t more vocally recommending masks. I can see that anti-lockdown can be a reasonable position, but if you’re not going to have restrictions on how people interact face to face, then masks are even more important, right?

      • Michael Weissman says:

        There seems to be something even sicker than that. E.g. a while ago Martin Kulldorff was arguing that herd immunity was the way to go. His argument was indifferent to human life but at the time not as obviously wrong technically as some of the other arguments. Now he’s urging everyone who is not at high personal risk not to get a vaccine. What happened to promoting herd immunity? It seems like the performative callousness is a more consistent thread than any particular technical claim.

        • The irony is that since March until December, some rate of herd immunity has been progress. Paul Offit, Michael Osterholm, and others have guessed that upwards of 100 million in US has had COVID; most perhaps asymptomatic. I think even CDC Director Walensky put out a figure of 20 % of population has had COVID19. There may be an undercount of COVID infected pop. Some may suggest that John Ioannidis was instrumental in forging COVID policy under Trump administration. But he did insist that asymptomatics get tested. That was not heeded. Why we have the case of the elusive COVID19 denominator.

          In regard to Who should or should not get vaccinated. There is intense disagreement. I do think that we could saved many more lives had we tested Americans back in April or May. Not with the PCR; with the Rapid Antigen tests for infectivity.

          • dhogaza says:

            “In regard to Who should or should not get vaccinated. There is intense disagreement. I do think that we could saved many more lives had we tested Americans back in April or May.”

            For that to be effective you have to be willing to take action on the information that is made available. That wasn’t going to happen in the US at the federal level, certainly, and in many states.

          • Michael Weissman says:

            Absolutely agreed re rapid antigen tests. It’s still not too late for them to be somewhat useful.

            Don’t just about everybody agree that ~100M have had Covid in the US?

            I found my old email from 3/17/20″ “a neat trick: get a low CFR by assuming many undiagnosed cases and a low infection plateau by assuming very few undiagnosed cases.”

            You’re bending over backward to give JI credit for saying in passing “we need more data”, as if that weren’t on the stationary letterhead for any scientist. His overwhelming message was: meanwhile, no biggy, don’t worry, do nothing.

            • Hi Michael,

              I have watched nearly all of the interviews with John Ioannidis and read a good number of his articles. I will add that John is most understandable when he is lecturing or being interviewed. The articles can get a tad too technical. I have to rely on commentaries to make sense of the technicalities.

              Yes, I do give John ioannidis credit for highlighting the complex epistemic environment environment in medicine. I have followed the evidence based medicine movement since the 90s. I’m disappointed to find that it has been co-opted in ways that translate into marketing products and treatments that of marginal value. It takes considerable courage to convey this to the public and to the medical establishment.

              BTW, here is the article that seemed to have caused such controversy. When John mentioned the 10,000 figure, in what context did it come up? I think that readers have taken the 10,000 out of context. My interpretation was that John was suggesting that even the Diamond Princess projections were unreliable due to scant data. I took the trouble to ask some of my friends to read the article. Even they did not think that John Ioannidis was making a categorical claim that the IFR was as low as the flu. He couldn’t have makde that claim given the totality of the argument he made.

              https://www.statnews.com/2020/03/17/a-fiasco-in-the-making-as-the-coronavirus-pandemic-takes-hold-we-are-making-decisions-without-reliable-data/

              • Michael Weissman says:

                Hi Sameera- I read that article at the time, with great alarm. At every stage where there was any uncertainty he always chose the lowball number, even when that required inconsistent assumptions. He never gave a hint that for exponentials early action is cheap and effective while late action is expensive and ineffective. It was a complete hack job, decorated with a few scientific bromides, like how more data would be nice.

                This is not Monday morning quaterbacking. I wrote people on 3/17/20.

                One important thing to do under uncertainty is to ask whether the second derivative of the utility function is positive or negative wrt estimation error. JI implicitly got the sign wrong.

              • Joshua says:

                Sameera –

                > Even they did not think that John Ioannidis was making a categorical claim that the IFR was as low as the flu. He couldn’t have makde that claim given the totality of the argument he made

                2:36 in this video… “the same ballpark as the seasonal influenza.”

                https://youtu.be/jGUgrEfSgaU

              • Joshua says:

                And anyone can be wrong.

                The problem is the methodology behind why he was so wrong. He got it so wrong because he extrapolated from unrepresentative and non-random and ill-suited and insufficiently comprehensive data.

                That isn’t just missing on pegging the uncertainties. That’s bad science. What’s odd is that (1) he warned about that kind of bad science vis a vis COVID and, (2) he made his stellar reputation by criticizing bad science, exactly like the bar scoence he’s done with COVID.

              • Joshua says:

                Sorry, that’s 2:30 in the video.

              • Good Morning Joshua,

                The video that you posted doesn’t capture the progression of John Ioannidis’ hypotheses. The video below is clearer, I think.

                https://www.youtube.com/watch?v=QUvWaxuurzQ&t=1s

                What is the thesis of the original article which has resulted in such expert angst? It’s in answering that question that lends itself to appropriately contextualizing the 10,000 estimate associated with the Diamond Princess Cruise situation.

              • Carlos Ungil says:

                How is the 10,000 estimate associated with the Diamond Princess Cruise situation?

                The main problem with the 10,000 estimate is the assumption that 1% of the U.S. population gets infected which is introduced without any justification.

          • xyu says:

            as of beginning of March, my own data on seroprevalence of SARS-CoV-2 infection is 20-27% in the community, depending on the samples. My estimation of overall infection in the US as of March, 2021 (when US reached 500,000 deaths mark) is 23%.

          • Rahul says:

            Sameera:

            What do you mean by some rate of herd immunity? We definately don’t seem to be at it? If you mean we are 30% there, sure.

            Second, 100 million versus 20% of US population is a sizable difference. In any case neither of those numbers seem anywhere close to herd immunity.

            So basically I struggle to understand your claim.

            • Hello Rahul,

              To clarify, earlier I meant to qualify the characterization ‘herd immunity’ to natural infection. By that I mean that herd immunity can also be reached when a sufficient number of people in the population have recovered from a disease and developed antibodies against future infection.

              Note what I posted earlier:
              “he irony is that since March until December, some rate of herd immunity has BEEN IN PROGRESS. Paul Offit, Michael Osterholm, and others have guessed that upwards of 100 million in US has had COVID; most perhaps asymptomatic.”

              So I was inferring from the opinions based on those COVID experts that roughly a 1/3 of US population had been infected. Rate is a ratio.

              In this case calculating an accurate denominator has been a challenge. Now CDC director Walensky estimates 20 %. Back in the fall, The previous CDC director estimated that there were between 90 to 105 million.

              • Navigator says:

                Sameera,

                There is a 100-fold variability in antibodies among those previously exposed to covid. True, for some people on the higher end of the spectrum, the vaccine won’t add much, but without complicated and expensive tests we can’t know where any of us stand.

                The chance for a second infection and long covid in those previously infected is a real danger. There were cases of second infection fatalities in Italy.

                Whether vaccinate or not, 1918 flu is a pretty good indicator. I believe you mentioned your reaction to vaccine some time ago, and in those cases probably best to wait for a milder one, which may come soon, but don’t let your n=1 sway your opinion in general.

                AFAIK, there are no preservatives, alluminum, eggs, mercury and other stuff in these new ones, so you may as well be ok with it (depending on your age and if you need it in the first place, of course).

              • Anoneuoid says:

                AFAIK, there are no preservatives, alluminum, eggs, mercury and other stuff in these new ones, so you may as well be ok with it (depending on your age and if you need it in the first place, of course).

                The main risk along those lines is anti-PEG antibodies:

                Despite this success, an emerging body of literature highlights the presence of antibodies produced by the immune system that specifically recognize and bind to PEG (anti-PEG Abs), including both pre-existing and treatment-induced Abs. More importantly, the existence of anti-PEG Abs has been correlated with loss of therapeutic efficacy and increase in adverse effects in several clinical reports examining different PEGylated therapeutics.

                https://pubmed.ncbi.nlm.nih.gov/27369864/

              • Rahul says:

                Sameera:

                So I hear that fact about herd immunity often but don’t understand why that matters.

                Sure, you can reach herd immunity by natural infection. If you wanted to go that route we could all shed masks, stop social distancing and go about life as usual. Maybe even congregate around super spreaders and our herd would reach immunity faster!

                Problem is, natural infection carries with it a mortality / morbidity risk which is way higher than building up your immunity by using a vaccine.

                That’s where vaccines have a huge advantage over the natural infection route. A secondary point is that the antibody titres post vaccination seem higher and sustained than post natural infection ( on average).

                Finally, even taking your estimate that 30% of the US population has been infected, it’s still a far cry from herd immunity and vaccines are the only credible way to get there, IMO.