Coronavirus Quickies

This post is by Phil Price, not Andrew.

There a couple of things that some people who comment here already know, but some do not, leading to lots of discussion in the comments that keeps rehashing these issues. I’m hoping that by just putting these here I can save some effort.

1. The ‘infection fatality rate’ (IFR) is not an endogenous number that describes the virus: it depends on the people and their circumstances.

By ‘it depends on the people’ I mean it is a much higher number for a typical group of old people than for a typical group of young people. It is a much higher number for a typical group of diabetics than typical group of non-diabetics.

By ‘it depends on their circumstances’ I mean it is a higher number for people who get no medical help than for people who do.

2. The IFR does not characterize how dangerous the disease is at a societal level, even if we knew the accurate number for a given population and their circumstances. That’s because the IFR quantifies the probability of dying once a person is infected; it says nothing about how likely it is that they will get infected.

Even if COVID-19 is like a slightly-worse-than-average seasonal flu in terms of IFR, it would be much much worse from a societal standpoint: it seems that nearly nobody is immune (except perhaps those who have already had it), whereas in any given year many people are immune to the flu, either because they were vaccinated and the vaccine was effective for them or because they had a similar strain of the flu in the past and are still immune. (I’m aware that the distinction between ‘immune’ and ‘not immune’ is not so clear-cut, but that doesn’t invalidate the point).

A virus with an IFR of 40% in a given population, but that only infects 0.1% of the people exposed to it, would not become an epidemic because it would not infect enough people. But a virus with an IFR of 0.1% that infects 40% of the people exposed to it would be a public health disaster and would kill millions of people. If someone says ‘coronavirus is like the seasonal flu, just look at the IFR’, they do not understand. To be like the seasonal flu in the way they mean, it would need to be like the seasonal flu both in terms of IFR and in terms of the number of people it will infect.

Those are my main points. But since I’m here I’ll go on with one more thing…hey, the rule of three, gotta do it:

3. According to the “Worldometers” data aggregation site, coronavirus deaths per million in the Republic of San Marino is over 1200. Even under the assumption that everybody there has been infected, that implies that in that population, with whatever medical care they received, the IFR was 0.12%. It’s one of the wealthiest countries in the world, and does not have a population that is highly skewed towards old people, which suggests to me that for the U.S. population as a whole — if everyone were infected, or if a simple random sample of people were infected — the IFR would be over 0.1%, even if every infected person got good medical care. This is also suggested by data from Spain and Belgium, where deaths per million are above 500 even though (I think most people agree) fewer than half the people in those countries have been infected.

This post is by Phil.

194 thoughts on “Coronavirus Quickies

  1. Can I ask (non-snarkily, I hope) why is it important to make the point that IFR depends on people and their circumstances? In particular:

    (1) Isn’t it widely understood that the coronavirus is much worse for older people than for younger people?
    And (2) how is “IFR depends on people and their circumstances” different from saying that IFR is an average?

    I mean, say the average family has 1.9 children. We know that no family has exactly 1.9 children, and some have many more and some have none, and the number of children depends on circumstances in all sorts of predictable ways.

    • It’s important because Phil is right, you see people asking whether a rate seen in one country or city “could be right” given a rate reported in another place.

      • What’s wrong with asking such a question? If NYC has an IFR of 0.8%, could it be right that Santa Clara County has an IFR of 0.1%? No. The populations aren’t sufficiently different for that much disparity.

        Knowing that IFR varies by location is not the same as saying IFR info from one location has no predictive value in others.

        • “If NYC has an IFR of 0.8%, could it be right that Santa Clara County has an IFR of 0.1%? No. The populations aren’t sufficiently different for that much disparity.”

          Why do you believe your last sentence? What evidence can you give that would support it? Also, there is a lot more than just “the populations” that could be relevant — things like density, climate, historical quirks that affect disease transmission or severity, …

        • What does disease transmission have to do with it? We’re talking IFR.

          The population in NYC is older and sicker than Santa Clara County’s, but not by enough for a factor of 8 difference in deaths.

          It’s possible the attack rate is more heavily asymmetric than I’ve allowed for, but in any case my question to Kyle C was what’s wrong with asking such a question? You take demographics, adjust for them, and compare. I’m not saying I’m equipped to do it nor that I have access to the data necessary, but to a great extent the comparison is possible and the data exists.

        • Santa Clara is quite likely wrong. But we’ve seen better studies suggesting an IFR notably below 0.8%.

          The death rate of this is so strongly skewed by age that things like number of nursing homes/long-term care facilities and how well they were protected could make a huge difference. In MA, over 50% of deaths are in nursing homes/LTC facilities, and I think this is true of a number of other states as well. And NYC was a lot closer to an overwhelmed system… they weren’t turning people away, but I’m sure quality of care must have dropped. And (while it’s notably anecdotal at this point) apparently NYC over-ventilated patients.

          Iceland has a 0.56% fatality rate and, as of today, zero patients in ICU. They have the best testing in the world, but I don’t think even they caught every single case… so their real IFR must be less… sample size is small though (10 deaths).

      • There is though… there exists a global mean IFR and I want to know what it is, and how that compares to other diseases.

        Yes, the number of people infected is also hugely relevant, which is why I also want to know the global mean R0.

        • +1

          Maybe it would help if we got in the habit of being more specific — eg, saying things like, “The current global R0”, or “The global R0 last month”, or “the current US R0”

        • Have we seen any evidence of new treatments affecting the IFR? I mean theoretically it could happen, but do we think it has to any extent at this point?

          I punched in the 40 days of data that Cuomo shows in his daily briefing, and it looks like almost exactly 25% of the people hospitalized each day died after an average of about 5 days. Actually, deaths include nursing home deaths so it’s somewhat less than 25%, but the relationship is still there.

          Here’s the visualization, it tied much more closely than I expected. https://i.redd.it/26afhymyxov41.png

          Nothing here suggest any change to the IFR from treatment over the bulk of the time it’s been active in NY.

          There’s an aspect of IFR that’s not visible here, which is what percentage of infected get hospitalized. But even if that has been changing (with corresponding changes in IFR), I don’t think it could be attributable to treatment.

        • “Nothing here suggest any change to the IFR from treatment over the bulk of the time it’s been active in NY.”

          Have there been any changes in treatment? You can’t have changes to IFR from treatment if there haven’t been any changes in treatment.

        • There have been changes in treatment!

          many doctors and hospitals reduced the use of ventilators and employed different treatments, like turning patients onto their stomach instead of leaving them on their backs. I’m sure when the dust settles other changes will emerge; in a situation like this people have more leeway to act on hunches and therefore find more improvements.

        • Actually, the aggregate R0 is not so meaningful. The spatial distribution of Covid cases is *extremely* fat-tailed. The R0 parameter necessarily has only a mean-field interpretation – and yet, the spatial distribution violates the meaningfulness of the mean-field.
          I wrote about this here about a month ago, arguing that compartment models should be fit to maximally disaggregated data. The piece is still sketchy and rough and needs updating, but my central thesis stands as far as I can tell:

          https://rpubs.com/chwilson101/593913

          Comments, and error detection welcome. Note that I consider only the standard Pareto distribution in the essay. In reality, it is upper-truncated, and thus the moments will exist and be finite, but the severe bias that is induced remains.

    • The thing I would focus on is the use of the IFR in these discussions.

      Usually the context is on establishing the severity of covid-19, with people like Trump and Ioannidis claiming that if IFR is low enough to be “like flu” this would mean lockdowns are bad and unnecessary and should be ended. The importance of recognising the contextual nature of IFR is that it cautions us against accepting this argument by extrapolation – even if IFR is low in Santa Clara with a few infected, if the infection increases and everywhere becomes New York, the numbers will likely go up. Thus their claims can be extremely dangerous.

      • Zhou,

        My understanding is that the IFR is really a snapshot and not a measure from which to extrapolate an IFR nationally. I didn’t hear nor read anything in that study to suggest that. I think the more specific takeaway I got was that there were way more infected than originally held. So one can dispute that here on the blog.

        If audiences interpret it to equate to a national IFR, then experts aren’t being clear enough.

        Not sure I understand what you mean by ‘numbers will likely go up’. In which specific context do you mean? My apologies.

        Lastly, I gather that the antibody testing method was used in the Santa Clara Study, unlike most all other studies and critiqued by some for its likelihood of resulting in False positives.

        • Sameera –

          > I didn’t hear nor read anything in that study to suggest that

          In the interview, Ioannidis said that the study supports a conclusion that COVID is about as fatal as the seasonal flu. And he went on to engage in politically overlapping discussion about the harms caused by government mandated social distancing.

          Let me ask you if you think thsr

        • Oops…

          I understand and agree with the importance of extending “cognitive empathy” to Ioannidis and everyone engaged in these discussions.

          But let me ask you. Do you think that Ioannidis should *not* be explicitly criticized for what I described?

        • Hello Joshua,

          I was responding originally to Zhoe. I did hear John Ioannidis comparing Sars 2 to the ‘seasonal flu’. But he also conveyed a whole bunch of caveats about comparisons, stating that the seasonal flu produces severe or fatal symptoms in vulnerable populations too, namely the elderly and hospital workers exposed to higher viral loads. So what I mean is that is he was elaborating a cautionary tale at the time when the Imperial College figures were being adjusted down from its original results.

          In so far as John’s lack of or support of social distancing, John again noted that its social utility will have ben studied further, but expressed some support of it. Likely, he is more aligned with the European perspectives on some measures.

          To reiterate, I don’t see any comparison made between the IFR for the Santa Clara County and a national one. So in that sense, I think you are skirting my initial assertion which, by the way, I am happy to retract if I am wrong. I think that John is trying to broaden the dialogue, and may realize that some circles are aligned with other experts. Again, I think it is always a harder line to convey optimistic future. There are apocalyptic voices and as you know they can suck up the airwaves.

          However, I try to read stuff a couple of times since I am a consumer of statistics and medicine. I have little interest in making a splash in the statistics community.
          Finally, I have no objection to anyone critiquing the Santa Clara Study, LA study, or Imperial College results.

          I have though a great deal of regard for fair play b/c I have seen how we can misread and misinterpet what is being said or written.

          Ya wanna Poke and Play? Smile. lol just kidding.

        • Hi Sameera –

          Thanks for the reply.

          > I did hear John Ioannidis comparing Sars 2 to the ‘seasonal flu’. But he also conveyed a whole bunch of caveats about comparisons, stating that the seasonal flu produces severe or fatal symptoms in vulnerable populations too, namely the elderly and hospital workers exposed to higher viral loads. So what I mean is that is he was elaborating a cautionary tale at the time when the Imperial College figures were being adjusted down from its original results.

          So here’s my take. I listened to much of the interview with Katz, and my recollection was that Ioannidis said rather confidently that the COVID fatality rate was not something to be scared about (I think his exact words), and that part of the reason was because the Santa Clara study showed that the COVID fatality rate was about the same as the seasonal flu.

          Now I’m not going to back and listen again, and my recollection may well be a “mis-memory” because of my own biases – but I thought I heard him say that in a context.

          1) he was making a comparison to the flu within a politicized context, of which he MUST be aware. He was making a deliberate choice to enter himself into that context.

          2) he was engaging in a rhetorical exercise with the interviewer. The interviewer was seeking rhetorical ammunition for supporting a policy agenda. John knew this. And he didn’t correct for that rhetorical agenda, but instead went along. One example: he was asked about the Imperical College study “predictions.” (or many it was estimates?). He must know those were not predictions – they were a range of projections, which is quite a different thing. And he was asked about the high end of the worst case projections under the conditions that absolutely no interventions would take place, and he didn’t take pains to qualify what those projections were. The low end of that range of projections were a great deal lower, and the projections were given along with the caveat that the possibility of absolutely no interventions was extremely low.

          Another example: he talked about the uncertainty injected by the difficulty of distinguishing between “died of” and “died with” categorizations, but didn’t mention the uncertainty running in the other direction – a result of people dying at home or at LTCFs without tests being performed.

          IMO, a selective treatment of uncertainties of that sort, from within a frame that was clearly rhetorically stylized and targeted, is just not acceptable from a scientist. I come at this from the climate blogosphere, where climate scientists are told not to be advocates for policies. Now I strongly disagree with that chastisement, as I think that advocacy is important and scientists have every right to be advocates – even with respect to policy development that doesn’t lie directly in the center of their range of scientific expertise. But, my caveat there, is that they have a scientific obligation to take an even-handed approach towards uncertainties during the course of their advocacy.

          3) he was jumping from an *infection* rate based on a non-random, non-representative sample collected in a particular area to extrapolating a broadly applicable *fatality* rate. I consider that to be scientifically unacceptable – even if one adds in some caveats (and I didn’t really think that aspect of what he said was particularly caveated. He was clearly saying that we can reasonably extrapolate that broader fatality rate from the SC study of infection rate.

          So here is the bigger context for me. I think that scientists need to enter these engagements with care – and when they clearly put themselves in an advocacy role, they should expect that they will get a lot of pushback. This is the environment in which we live. That doesn’t justify all manner of pushback. I won’t do that. But it does mean that when you enter the discussion from within a particular rhetorical and policy-advocacy frame, you should do so with your eyes wide open and acknowledge your advocacy rather than hide behind some badge of scientific purity.

          I want to extend this discussion to that video you linked – where those two docs from Cali promoted some god-awful analysis to advocate for an anti-lockdown policy orientation. They received a lot of pushback, the video was taken down by Youtube, and what followed were cries of “censorship” from rightwing media personalities. It was major news on Fox on multiple shows.

          This is very much a related issue to me – and although of course Ioannidis has a great deal more scientific credibility than those docs, I think that the issues are necessarily related. IMO, John is no idiot and I think that just like the Cali docs, there absolutely should be an expectation that people (1) directly acknowledge their advocacy. Sometimes that means being directly accountable for the rhetorical frame through which they engage. It really, really annoys me when people either falsely or unknowingly hide behind a badge of scientific purity to obscure that frame. People have a responsibility to investigate the rhetorical frame they’re engaged in. (2) expect pushback without whining. I think that Ioannidis did that to some extent – that’s great. But in that case, the one getting the pushback has a responsibility to tell his/her circle of supporters to likewise be accountable. In this case, that might be something like telling his good friend and collaborator to not pen an editorial in his defense – by going back to the scientific purity frame as if John didn’t stake out a rhetorical, policy-advocacy territory in his own work.

          > To reiterate, I don’t see any comparison made between the IFR for the Santa Clara County and a national one.

          That’s interesting, because I did. So I will acknowledge that I may be in error. Or maybe it’s just interesting that you and I may see the same thing and come away with different impressions. John can’t necessarily be accountable for that – but then again, I think that often people certainly do need to be accountable for the predictable differences in how they might be heard.

          > So in that sense, I think you are skirting my initial assertion which, by the way, I am happy to retract if I am wrong.

          No, I don’t think I am skirting it. And I don’t think that I have been doing so. And I actually been thinking that you’ve been skirting it. And again, I am not referring to the study in itself, but to his interview with Katz. Like you, I am happy to be corrected.

          > I think that John is trying to broaden the dialogue, and may realize that some circles are aligned with other experts. Again, I think it is always a harder line to convey optimistic future. There are apocalyptic voices and as you know they can suck up the airwaves.

          I think he is too. And I applaud broadening the dialog. But I think that should be done with care, and with careful attention to easily predictable, but in balance negative, outcomes. One can’t prevent that entirely. But people can be proactive.

          > I have though a great deal of regard for fair play b/c I have seen how we can misread and misinterpet what is being said or written.

          No doubt.

          > Ya wanna Poke and Play? Smile. lol just kidding.

          I’m from Philly. I advise you against that.

        • I don’t agree that actors like Ioannidis do broaden the dialogue. Studies do, but when Ioannidis goes out confidently on right wing platforms to push a certain point of view, and in the end the discussion turns into “well the Imperial College says X, esteemed professor John Ioannidis says Y”, what is actually going on is that the dialogue ends, and the discussion turns not into a debate about what the truth is, but which side you take. Ioannidis’ actions (and I say Ioannidis as opposed to the Stanford study very deliberately) is to be a pseudo-academic fig leaf for denial.

          The fact that is missed in these defenses of Ioannidis is that Ioannidis’s criticism is confined to academic circles – outside of these circles, Ioannidis is getting a thousand times more positive exposure than his critics ever will. It is his critics that cannot be heard.

        • Zhou,

          John hasn’t gone on simply ‘right-wing’ platforms. Fox News has been recruiting Democrats and liberals to draw larger audiences. Besides the context should be heeded. John is a peacenik largely.

          John has been on NBC, John Kirby podcast and I think on CNN too. Mostly he is asked to speak at universities.

          Just to make clear, I haven’t spoken to John in many years. So no collusion here.

          I was on an email chain with Andrew Gelman, Sander Greenland, and John a few years ago. The exchanges were covered here on Andrew’s Blog.

        • Joshua,

          Would you agree that the Imperial College study was a forecast or a prediction? Do you think that it addressed uncertainties? And if feasible, let’s not, you and I, evaluate the study as a function of hindsight bias. There was a paucity of data at the time. Katz and Ioannidis characterize the Imperial Study as a product of inaccurate or unreliable data. Therefore arguing that more data are needed. Now that’s not some novel insight I admit. The Imperial Study authors had to downgrade their numbers. I guess one of the authors was Neil Ferguson who favors a lockdown strategy. Katz very ably explains what a draconian lockdown strategy would accomplish. In short kicking the can down the road. Short of a vaccine, which is anywhere from 18 to 24 months away at minimum, the option of a phased heard immunity strategy is inevitable. In this regard, both doctors place an emphasis on separating and protecting the most vulnerable. I assume you must have been presented with the rationales for and against a draconian strategy.

          Their explanations seem quite sensible to me, especially as it is claimed that a vaccine for the coronavirus is hard to develop. It has been tried before, with marginal success. There is a tendency to omit the potential side effects when announcing plans for it.

          The 2017-2018 flu vaccine was 30% & effective. Likewise, such concern pertains to a vaccine for the COV-I9. We can’t predict how successful or unsuccessful it will be. Now I think ‘uncertainty’ was entailed in the Imperial College Study. The results were dealt with a mark of certainty.

          Nearly every expert will be wrong and right in some dimension.

          Frankly by laundering uncertainty, nearly every expert will be able to contribute tautologies.

        • Sameera –

          > Would you agree that the Imperial College study was a forecast or a prediction?

          I think it was a projection. Different words can be used interchangeably, but from a scientific perspective we know that it wasn’t a “prediction.” Part of what I bring to this comes from the debate about climate change, where “skeptic” frequently turn projections into predictions, and focus only on the high end of those projections and not the full high confidence range, to say “See, the ‘predictions’ were wrong and that proves we can’t trust climate scientists.”

          Here, read this article:

          https://reason.com/2020/03/27/no-british-epidemiologist-neil-ferguson-has-not-drastically-downgraded-his-worst-case-projection-of-covid-19-deaths/

          I think it summarizes the problem fairly well.

          Now Katz asked John about the “predictions” of 2 million deaths with in a rhetorical fashion and John laughed and called them “science fiction.” Well, first, they were projections, with a very wide range and best case scenarios waaaaaaay below 2 million, and they themselves said that the projections based on no interventions or behavioral changes were simply not plausible. But it was spun in the interview, IMO, in a way that will lead to a counterproductive misunderstanding of how the projections were meant to be used.

          No spin happens in the other way. The projections were mischaracterized as scary predictions to justify mandated social distancing. I won’t rationalize that either – particularly not from a scientist, and particularly not from a scientist who plays a very important role in critiquing the role of science in our society.

          > Do you think that it addressed uncertainties? And if feasible, let’s not, you and I, evaluate the study as a function of hindsight bias.

          Of course it did. But it was also just fundamentally a projection based on some 750 highly uncertain parameters. Now maybe you want to argue that such projections don’t return a net benefit. OK, I can accept that as a viewpoint – even if I may not agree.
          But take that argument on in a scientific fashion. Don’t spin the projections in a way that enhances misunderstanding, in order to advance a policy-agenda. Not cool, in my book.

          > There was a paucity of data at the time. Katz and Ioannidis characterize the Imperial Study as a product of inaccurate or unreliable data.

          Well, that’s a spin. Yes, there is a lot of uncertainty, and it was quantified. That’s what legitimate science does. And you can disagree with the characterization of the uncertainty. That’s what legitimate science does also. But don’t just talk about it being “science fiction” in an off-handed way. And don’t just throw out your own product, also based on “inaccurate and unreliable data” without similarly characterizing your own uncertainty – even if you do handwave towards uncertainty in some blanket fashion.

          > Therefore arguing that more data are needed.

          No. They’re doing more than that. And they’re also arguing that they can draw conclusions based on the same quality of data.

          > Now that’s not some novel insight I admit. The Imperial Study authors had to downgrade their numbers.

          See above.

          > I guess one of the authors was Neil Ferguson who favors a lockdown strategy. Katz very ably explains what a draconian lockdown strategy would accomplish.

          This is getting too long. I don’t want to be rude by cutting off the rest of what you wrote. I have read it and I appreciate the engagement, but I feel like I’ll just be repeating myself by responding in depth to the rest.

          I’ll just stop here with my opinion that your use of “draconian” right there is in kind with exactly the same kind of rhetorical frame I’m talking about. “Draconian” depends on a value judgement, based on making assumptions about a lot of absolutely huge uncertainties, in my opinion. It’s certainly your right to use that term. But please don’t use it and then tell me that it’s merely a product of “pure” science. And please don’t complain to me if you get pushback on that.

          I won’t justify rude or personalized pushback. But I really, really object when people either deliberately engage from within a politicized frame, or just wander into such a frame without really being proactive about how they’re engaging, and trying to shield what they’re doing with claiming victimhood from censorship, or from behind a “pure science” stance, even as they often have disproportionate access to power. It’s like that intellectual dark web article in the NYTimes where people who get tons of mass media exposure complained about how they were being shut out of freedom of speech.

        • Sameera –

          > To do justice would require both us to point out precisely where in the Ioannidis interview, John made the assertion you claim.

          I don’t really think so – if only because I raised a number of points that are not directly contingent on his precise phrasing.

          As for the precise working, Katz had a rhetorical spin throughout the interview, and John accepted that framing without speaking to it. I guess that you accept the rhetorical framing because you agree with the point of view.

          IMO, people are certainly entitled to rhetorical spin, but scientists have an obligation to identify that spin and speak to it.

        • Just to get clarification, are you saying that John is parroting David Katz? I think David Katz has done one of the better jobs of explaining ‘herd immunity’.

          We are already forging herd immunity differently in different locales. State by state and country by country.

        • Sameera –

          > Just to get clarification, are you saying that John is parroting David Katz? I think David Katz has done one of the better jobs of explaining ‘herd immunity’

          No. Not parroting. But he accepted the rhetorical framing, the one-sided angle on uncertainty. I described a few examples.

          >We are already forging herd immunity differently in different locales. State by state and country by country.

          I don’t agree at all. But that’s a whole other argument, and a complex one.

          And I don’t begrudge anyone a rhetorical framing. But I think that a scientist has an obligation to identify them, and be accountable for them, and to address unertainty on all aides.

        • As for the precise working, Katz had a rhetorical spin throughout the interview, and John accepted that framing without speaking to it. I guess that you accept the rhetorical framing because you agree with the point of view.

          IMO, people are certainly entitled to rhetorical spin, but scientists have an obligation to identify that spin and speak to it.
          —–
          Well, who is a scientist in your view? What is the rhetorical spin to which you refer With what do you disagree?

        • Sameera –

          > Well, who is a scientist in your view?

          I’m confused by that question. I’m talking about Ioannidis in this case. But it happens a lot. It’s not unique to him or this context.

          > What is the rhetorical spin to which you refer With what do you disagree?

          The way they talked about the Imperial College projections is one example. The way thsr he talked about the uncertsinty related to identifying deaths (a one-way treatment of uncertainty) is another. I described them both in more detail above.

          More generally, Katz clearly has a rhetorical spign aligned to a particular view on policy. Jihn didn’t push back on thst spin at all.

          Ifnhe goes on Fox News, that’s fine with me. But if he does so he has an obligation to speak on the different directions of uncertainty. He knows of he goes on Fox (or NBC) what he’s going to say is going to be politicized. That doesn’t mean he shouldn’t go on to Fox at all, nor that he has absolute powwr to keep his participation from being politicized. But if he doesn’t proactively take care to influence the polirization – in this case by talking about the bilateral uncertainty – he is effectively directly taking part in then politicization. That’s if his eight. But if he does that then don’t hide his advocacy behind a shield of scientific purity.

          And even then, I think that as a scientist. In particular one who champions de-biasing science, he has an absolute obligation to not deal with uncertainty in a selective manner so as to push towards a particular rhetorical framing of a policy issue!

        • Joshua, Apologies for not responding before now. I’ve been busy with work and family.

          If you are suggesting is that scientists, statisticians, physicians, etc shouldn’t preload their evaluations with hyped adjectives, I agree. To be fair, the term ‘lockdown’ conjures up different images for different people. My friends and I have been discussing the uses of contrast/comparisons/analogies/images, etc. Some of us are screenwriting buffs or write poetry. So we pay attention to context quite closely.

          In my observations of and reading biographies about scientists, scientists can be quite dramatic. We’ve seen examples of this tendency on Twitter and Facebook. There seems no way that comparisons/contrasts can be avoided. Some draw them well; others may not. I think that nearly every scientist does engage in some rhetorical framing. Nor do I think it is a bad thing to do either if logically and accurately presented.
          I think this rhetorical framing is part and parcel of the ‘argument culture’ that has suffused nearly every field. And much has written about the use of it in science I understand. Binary modes of thinking prevail in conversations and debates.

          Now you did not use ‘rhetorical framing’ as such. Merely recharacterizing some parts of Katz and Ioannidis’ assertions as ‘spin’, b/c you disagree with their claims, is also a way to encourage one interpretation and discourage another interpretation, which translates into the definition of ‘rhetorical framing’.

          That said, I appreciate your points and don’t disagree vehemently. I don’t listen or read just Katz or Ioannidis. I follow many statisticians and other types of experts. I didn’t even see this exchange between Katz and Ioannidis to which u referred earlier.

          I am not sure who has given the most cogent pushback to the various forecasts either. The fact-finding process has been evolving. So I roll with it.

          Lastly, I never heard the word ‘projection’ used in such a context b/c the definition I’ve been most familiar with is: The state or fact of jutting out or protruding.

          Forecast: to predict a future condition or occurrence; calculate in advance; foreshadow. Similar definition for a prediction I believe.

          So thanks for sharing.

  2. [Copying this from the other thread because 0.1% gets bandied about, and it should be 0.025%]

    0.1% is a rough estimate for the incidence of *symptomatic* cases of influenza in the USA. https://www.cdc.gov/flu/about/burden/2018-2019.html

    “The fraction of persons with influenza virus infection who do not report any signs or symptoms throughout the course of infection is referred to as the asymptomatic fraction.”
    “estimates from studies that adjusted for background illnesses were more consistent with point estimates in the range 65%–85%”
    doi: 10.1097/EDE.0000000000000340 (Hongkong 2015)

    “The age-adjusted attributable rate of illness if infected was 23 illnesses per 100 person-seasons (13–34), suggesting most influenza infections are asymptomatic.”
    doi:10.1016/S2213-2600(14)70034-7 (London 2014 / Lancet RM)

    So if ~75% of flu infected are asymptomic (and don’t die), and 25% are symptomatic, we multiply the symptomatic CFR with 0.25 to get the IFR.

    So, 0.025% IFR for the flu, as a ballpark number, for the whole country.

    • Erratum: 0.1% is a rough estimate for the *case fatality rate* for *symptomatic* cases of influenza in the USA.

      (I shouldn’t have edited that line…)

    • I agree with your point in spirit, if not in specific. Not my field, but when I was skimming the literature to get a handle on what percent of flu case are asymptomatic, I found widely varying estimates. The CDC says that 3 to 11% of Americans will get symptomatic cases in any given year, but that a total of 5 to 20% will be infected with the flu (including asymptomatic) which gives an asymptomatic rate of closer to 40%. That would put the IFR for the flu close to 0.06.

      Not saying you’re wrong. I guess it’s more so to point out that we don’t even know what the IFR (as an average across all Americans, Phil) is precisely. We’ve had a century plus to study the flu and a bit less than 5 months to study COVID-19.

    • https://www.cdc.gov/flu/about/burden/index.html

      I didn’t calculate this precisely, but it appears that in the neighborhood of .13% per symptomatic case died due to seasonal flu during most of the seasons in the decade 2010-2019. Using Dalton’s estimate of an asymptomatic rate of 40% brings the IFR down to the neighborhood of .1, so perhaps that number being bandied about is a reasonable estimate of the IFR.

      Whatever. The worst flu season this past decade killed 61,000 and it looks like the US will pass that figure today. After three more days we’ll be saying that more than 61,000 died in just four weeks. All this despite the fact that many states are in substantial lockdown mode, and that even states like Georgia are mandating some level of distancing while reopening businesses.

      Meanwhile, the denialist argument is starting to focus on claims that covid-19 deaths are being intentionally overcounted in order to make it looks worse than it is …

  3. I have been trying to make a version of this argument non-stop to the “just the flu” crowd, in some cases my priors are that some make the comparison in IFRs disingenuously. Thank you for this post.

  4. IMO the Republic of San Marino and the differences among states in the US and countries in Europe suggest that there are major additional factors besides age, health care access circumstance and local containment measures controlling the spread and severity of the disease.

    Seattle and Vancouver BC, both cities with substantial connections to China, are relatively under control. WA state had the first case in the country and a major early nursing home outbreak too; yet TF:POP, WA, 700:7,000,000 = 0.0001; NY 22,000:20,000,000 = 0.0011, almost 10x higher for NY. Does NY have 10x more old people? I’m doubtin’ it.

    There have been some reports that virus particles cling to particulate matter in the air so the pollution in NY may be a factor; but that suggests California would be risky as well, and that doesn’t seem to be the case TF:POP, CA, 1700:40,000,000 = 0.00004, even less than WA! CA was pretty aggressive about closing down, it’s true, and WA lagged behind in that regard so that might explain CA’s better performance than WA.

    One previously overlooked factor is that WA and CA are both inhabited by substantial geek/modeler populations, whose screen tans indicate that they’ve been social distancing since before the Blackberry, so that may be a factor as well.

    Mere speculation I admit, but just the same there’s much still to be learned, so even if you’re breaking groups down for IFR, new risk factor discoveries could change the hole dang thing.

    • Differences in prevalence seem much more likely to explain the orders-of-magnitude differences you’re showing there. Just based on density, I would expect R0 to be higher in NYC than anywhere else in the US.

    • Jim, the main other factor is for how long COVID-19 was spreading in a community before it was detected.
      1 month delay in detecting that there is a problem gives you 1000x the number of infections.
      Most governments in Europe reacted at the same time, and most differences can be explained by the different number of cases at that time.

      Your arguing about China connections misses the fact that a large part of the initial COVID-19 infections in the US were imported from Europe. On the US East Coast cases imported directly from Asia are rare, the first cases imported from Italy to New York are estimated to have occurred in early/mid February before anyone even suspected that there were local transmissions in Italy.

      New York is both an extremely busy hub for airline passengers from Europe, and a popular destination for tourists from Europe.

      • To express it more mathematically, I would expect the same ratio between the number of deaths today and the estimated number of infections on March 15th for most European countries.

        UK and especially Sweden should be outliers due to herd immunity attempts.
        The overload of the healthcare systems in Italy/Spain/France might have resulted in more deaths.

      • “Jim, the main other factor is for how long COVID-19 was spreading in a community before it was detected.”

        True, but Seattle did have the initial cases.

        “a large part of the initial COVID-19 infections in the US were imported from Europe.”

        But it came to Europe from China, so China connections matter. True, population density is higher in Europe and NYC than Seattle or Vancouver, but OTOH both west coast cities have direct connections and Vancouver has very strong Chinese connections (it’s no accident that Huawei exec Meng Wanzhou was travelling through Vancouver when arrested) but remains almost untouched, with only a few thousand cases and 100+ fatalities.

        So while your arguments make some sense, they certainly don’t close the door on other contributing factors.

  5. I think that for your intended audience point (3) may work better if you talk about New York State (population fatality rate similar to San Marino) or New York City (almost the double).

    Given that we use the IFR to reason about what might happen if (when) everybody has been infected, I personally tend to think of it at the whole population level. Age structure is relevant (emerging countries are much younger => lower IFR) but the extent to which each group has been infected so far not (to estimate properly the population-level IFR we have to estimate it for each age group and take the weighted average, the ratio of total estimated deaths to total estimated infections will be biased).

  6. > But a virus with an IFR of 0.1% that infects 40% of the people exposed to it would be a public health disaster

    Sounds reasonable.

    > would kill millions of people

    This line caught my attention. If I’m thinking of the USA, then 40% is over 100 million, and 0.1% of 100 million is 100k.

    Would this kill millions of people come from assuming a global thing? Or is there an assumption about reinfections and non-immunity going on? Or am I doing the IFR thing wrong? Or should I not even try to convert IFR to total deaths without assumptions about immunity?

    • About Africa, here’s a recent article: https://www.ft.com/content/e9cf5ed0-a590-4bd6-8c00-b41d0c4ae6e0

      Yes, the lack of deaths in much of the world is interesting and under-reported. A few weeks ago one could reasonably claim that this was due to under-testing, but if that were the case the death toll would currently be glaringly high. Some attribute this to demographics, some to temperature, some to prior viral exposure, but in general it doesn’t get *nearly* the weight that the high-incidence places do in informing our response, even though the data are just as important.

    • I think was thinking millions of people worldwide.

      But really my point is not about the specific number, it’s that the IFR alone, even if known perfectly for a given population, doesn’t tell you how big the problem is. Low IFR with high infectivity can be worse than the other way around.

      • I am just not getting seeing something here.

        Let’s take the U.S. as an example. There is a population of about 329 million. If the infectivity is 40%, we would expect 131.6 million infections, assuming everyone is exposed. If the IFR is 0.1% of those infected, we expect 131.6 thousand deaths. Rough, but not much worse than the worst flu years.

        Now, let’s assume infectivity of 0.1%. We get 329 thousand infected. With an IFR of 40%, we expect 131.6 thousand deaths, exactly the same number.

        So why do you see them as different?

        • Stephen –

          > Rough, but not much worse than the worst flu years.

          The impact is not only measured in deaths. It’s also measured in development of, and worsening in, comormbiditites – which happens in much greater numbers. It happens in lost work hours lost economic production. It happens in medical workers getting dick and dying, etc. You say not much, but all at m at maybe twice a serious flu season.

          That doesn’t change your numerical argument of course. As for thst, Mendel’s point seems most relevant. And I would add that the likely differences in speed of spread associated with vastly different infection rates is relevant because with a slower spread there’s more time for developing better treatments.

    • Ben,

      The numbers you cite as examples seem pretty close to what’s in the process of happening in USA. We have about 330,000,000 people of which 40% would be 132,000,000 and 0.1% of those would be 132,000 deaths. So far the death count (which we are repeatedly told is supposdly a massive undercount) is half that number, with a couple thousand more dying per day.

      I think “public health disaster” is a fair enough assessment. But it does not involved “millions” of deaths or even a single “million” in USA. Probably take years to get to “millions”, plural, world-wide.

  7. What’s going on in the rest of the world?

    Worldometer shows about 1,000 deaths for India, and India’s first case was January 30. Should India have hundreds of thousands of deaths by now? Are their quarantine efforts really that effective? What about other parts of the world outside of Europe and North America? Worldometer numbers for Brazil show about 5,000 deaths. Will this soon balloon to hundreds of thousands, or is CV being effectively contained, even in the favelas?

    I have seen very little discussion of the pandemic beyond Europe and North America. Is there something we can learn from what is going on in the rest of the world?

    • Sorry — replied to the wrong comment. Now I’ll be redundant, which is hopefully not too annoying!

      About Africa, here’s a recent article: https://www.ft.com/content/e9cf5ed0-a590-4bd6-8c00-b41d0c4ae6e0

      Yes, the lack of deaths in much of the world is interesting and under-reported. A few weeks ago one could reasonably claim that this was due to under-testing, but if that were the case the death toll would currently be glaringly high. Some attribute this to demographics, some to temperature, some to prior viral exposure, but in general it doesn’t get *nearly* the weight that the high-incidence places do in informing our response, even though the data are just as important.

      • I would think that the rest of the world is the counter-factual to the lock-down-the-country-to-bend-the-curve strategy, because I would assume that, in many countries (or at least in certain parts of many countries), there is little in the way of effective intervention. Since we have convinced ourselves that anything less than drastic and immediate action will produce death counts in the millions in each country, we should use this natural experiment to test our models. But we are not. Lack of good data is not an explanation for why we are not thinking about this. No matter how terrible the record-keeping, millions of corpses should be observable by now.

        • This is a very interesting point.

          I wouldn’t draw any conclusions quite yet, for a couple of reasons:

          – since we don’t necessarily know when the disease arrived in different nations (the US just recently retrospectively found a COVID death in California over 2 weeks earlier than the previous earliest-known death; I’m sure other nations made mistakes as well)

          – because the “noise” might be higher in poorer nations (less access to healthcare means more deaths at home might not be recorded – and death reporting is lagged even in the US; baseline death rates may be higher so a “spike” of unexplained deaths might be missed for longer).

          – vastly lower median ages in some nations might mean it’s less deadly there

          But in 2-3 weeks we should know a lot more. There is a point where uncontrolled spread/deaths would become obvious despite the “noise”. And I can’t really believe that every single nation on Earth, including failed states, will effectively implement mitigation measures.

        • Well, since the total is already over 200,000 and some countries probably have very poor reporting, unfortunately I think there’s not much chance of the world total being less than 1 million :(

          Though it could be, actually, IF there is no second wave or it is milder than the first, and IF tropical countries are less hard hit.

          But what I’m getting at is, there seem to be huge differences in outcomes in US states that aren’t explainable just by looking at the dates of stay-at-home orders etc.

          Florida vs. NY for example: about the same population, yet a 20-fold difference in COVID deaths. Is it a matter of March and April weather being much better for outside activities in Florida than in NY? Is it a matter of mass transit/population density boosting spread in NY?

          IE – are some places/societies more at risk than others? And if so, why? That’s what I’m thinking that data from Latin America, Africa, and Central/South Asia might illuminate.

        • 1) The devloping world has long, sad experience dealing with large epidemic outbreaks. They might just be good at it.
          2) As a corollary, they’re less likely to reject WHO advice.
          3) The developing world is less connected to the countries that have driven the epidemic globally. A time delay causes a disproportionally lower spread, since the growth is exponential at first.
          4) Countries with fewer old people have fewer expected deaths.

          Brazil has 5000 deaths. This is 3-4 doublings behind the US death toll, in other words, less than two weeks for unconstrained spread.

  8. You mean San Marino has a IFR of 41 per 33,785. People rightly point out how small some European countries are when discussing numbers like these but they usually at least have a few millions for there to be deaths per.

  9. My own hypothesis is that there is substantial cross reactive immunity to other coronaviruses, and that populations that are exposed to a lot of respiratory viruses are relatively protected. There also is an intriguing possibility that vit D insufficiency is associated with the severe form of the disease.

    Both of these issues go in favor of poorer communities with more outdoor exposure, such as the large homeless population in Los Angeles which has NOT seen a tremendous attack rate of severe form. The hypothesis also predicts a higher severe disease prevalence in dark skinned people (lower vit D) and that is true.

    • Daniel,

      Good point. I’ve wondered how much cross immunity occurs with the more common coronavirus. According to this textbook (https://books.google.com/books?id=xnClBCuo71IC&pg=PA656&hl=en#v=onepage&q&f=false), “coronaviruses cause about 15% of common colds”. That seems like a small percent, but it actually could be important.

      Assuming the CDC’s statement “Every year, adults have an average of 2–3 colds, and children have even more” is accurate. Then assume the lower bound of 2 colds per person per year. With the population of the US being ~328M, that means each year there are 656M common colds. Given the aforementioned percent, that means 98M of them were caused by common coronavirus. Now assume the statement from the referenced textbook, “However, immunity that is developed in response to a coronavirus is relatively short-lived (~one year), and an individual can later be reinfected by the same antigenic type,” is accurate (seems reasonable to me). This means that each year, there could be 98.2M people with some immunity to coronaviruses.

      Finally, if assume there is some cross-immunity across antigenetic types. (This is still a debate; some research suggest yes, some suggests no; it is a hard problem to answer.) That means there could be 100M people who have some or full immunity to COVID-19.

      This is a big if, though. And not one the make policy on without substantial evidence. It does raise an interesting point, though.

    • What are your priors for that hypothesis? I disagree and my priors are: There are 4 endemic coronaviruses, only one of which (HCoV-NL63) also binds to ACE2 but it is an alphaCoV whereas SARS-CoV-2 is a betaCoV. Isolated neutralizing antibodies from recovered SARS-2 patients do appear to target the spike protein that binds ACE2.

      Actually, here is a review: https://www.medrxiv.org/content/10.1101/2020.04.14.20065771v1. From their Key Findings section: “… repeat human challenge experiments with single HCoV suggest individuals can be infected with the same HCoV one year after first challenge, but with possible lower severity. There is cross-reactivity within but minimal reactivity between Alpha-and Beta-CoVs. While endemic HCoVs rarely induce cross-reactive antibodies against emerging HCoVs, SARS-CoV-1 and MERS-CoV stimulate antibodies induced by prior HCoV infections.”

      So my updated prior is that previous HCoV infection is very unlikely to provide immunity against SARS-CoV-2 or protection from more severe forms of the disease.

      • The more challenges you’ve had, the more non-specific antibodies you’re likely to have, it would seem. The statement “SARS-CoV-1 and MERS-CoV stimulate antibodies induced by prior HCoV infections.” from your review suggests that if you have HCoV antibodies they’ll likely be stimulated by SARS-2 as well right? This means, suppose you’ve been living on the streets in LA, and gotten 17 coronaviruses in the last 4 years. Aren’t you more likely to ramp up some kind of non-specific immune response earlier than less challenged people? It certainly predicts the results we’ve seen where homeless people aren’t winding up in hospitals in droves, even though I certainly thought they’d be high risk at first.

        • Homeless aren’t usually
          — old
          — diabetic
          — surviving a disease (cardiovascular, immunedeficiency) that requires constant medication
          They’re not high risk?

          Where do you see numbers of how many homeless wind up in hospitals?

    • Daniel,
      just a day or two ago a team published some mechanistic research on exactly your point (“Presence of SARS-CoV-2 reactive T cells in COVID-19 patients and healthy donors”, https://www.medrxiv.org/content/10.1101/2020.04.17.20061440v1). Dr. Drosten has pretty good podcasts and ist quite popular here in Germany. I do not know anything about T-Cells, but the key message sounds pretty clear. Beware, this helps only in explaining infection paths, but does not make them different than observed so far.

      @All: Thanks for interesting discussion here & regards from Germany

  10. If everyone on San Marino also caught a cold, is the IFR of the cold virus .1%?

    I am being cute here, but it is an honest question: Does one need to make an attribution of cause of death in computing IFR? What if there was a toxic gas leak (or another virus, or food poisoning) which coincided with a COVID outbreak in a San Marino nursing home? Is IFR the same, or different?

      • Here you go daniel: https://www.sciencedirect.com/science/article/pii/S2213007120301350

        Also, Fauci is speaking on p-values. A 4 day shorter hospital stay in Remdesivir was highly significant, with p-value 0.001. This level of effect was already reported for IV vitamin C a month ago. It seems to me China is slow-walking the release of the results for whatever political reason. They didn’t release the antibody testing results until other countries started publishing them, so if no one else reports a vitamin C trial they will probably not let it get published.

        • I don’t understand your comment re: Fauci.

          The paper you cite leads to the same conclusion.
          The paper sees a shortening of time to recovery for less severe cases, and no effect or problems with the more severe cases.
          We can consider the course of Covid-19 in 3 phases, lasting ~1 week each:
          1 naso-pharyngal virus replication in the nose
          2 pulmonary virus replication in the lungs, immune response
          3 immune response
          Remdesivir affects virus replication, so it would help the most in phase 1, help some in phase 2, and be useless or harmful in phase 3.

          So it stands to reason that if you include mostly severe cases in your study, lie the Chinese, your results might not be significant, but “not significant” is not a contradiction to a study that saw significant effects, especially if the effect that wasn’t statistically significant was the same effect that was, and if part of why it was less significant was the lower N.

          The challenge for the medical profession is to understand what caused the elevated incidence of “adverse events” that caused the Chinese researchers to terminate remdesivir treatment early, and to narrow down the type of patient that benefits the most from this treatment (i.e. figure out how to predict who’s going to have a less severe course anyway), and figure out which dosage is the most effective.

        • NHST generates apparently conflicting results by design. This is what always happens, which should have proven to people it is a worthless method of learning about the world decades ago. Instead it became the “gold standard”.

        • Cost benefit analysis. You estimate the expected benefit along with the risks/costs for the individual patient then make an informed decision based on that. Whether p is less than some arbitrary cutoff plays no role at all.

        • No, NHST doesn’t offer any useful information at all. I have completely ignored every p-value for years without missing anything. Of course people still use them behind the scenes to filter what gets published so I am still less informed than I would be if that practice ended.

        • Anoneuid –

          Interesting. I’ll chew on that a bit. Too radical.to take in all at once.

          My initial reaction is that NHST shouldn’t be assumed as dispositive (especially a single study), but that when they’re done well they add value/inform probabilities (especially through well-conducted meta analyses that stratify by things like statistical power).

          Thus, it seems to me, they can inform cost/benefit analyses. But maybe I’m just resistant to change.

        • NHST amounts to comparing a p-value calculated for an arbitrary statistical model to an arbitrary cutoff. What information could it possibly provide?

          Instead come up with some model of the uncertainties in benefits, risks, and costs based on assumptions you put some thought into and make a decision. Eg smaller expected benefit could still be worthwhile for a cheap and safe treatment. But we would require large expected benefit for an expensive treatment with lots of dangerous side effects.

        • If you’re using an “arbitrary” statistical model, sure, then the p-value doesn’t mean much (e.g. if you did a study where your mathematical model of the test didn’t reflect the test in use).

          But doesn’t cost-benefit without looking at significance mean that if I flip a coin once and it lands heads, I should bet heavily in it landing heads again because the expected cost/benefit ratio is huge?

        • But doesn’t cost-benefit without looking at significance mean that if I flip a coin once and it lands heads, I should bet heavily in it landing heads again because the expected cost/benefit ratio is huge?

          No… Your uncertainty will be large.

        • The real question is: “Should I, as researcher or regulatory body, endorse this therapy?”
          The worst-case is that the therapy turns out to not work and an undiscovered side effect kills some users. The risk to me is my reputation and career, trust in me as a regulatory body erodes, people think I’m corrupt and making pharma companies rich, etc., and that has secondary public health consequnces. Treatments endorsed by the regulatory body must be paid by health ensurance, so this endorsement has an evonomic cost associated with it which is the reason insurance doesn’t usually pay for homeopathy. There are all sorts of secondary effects involved if you widen the frame of analysis (and the timescale). [How many commenters consider the effect of a “we don’t care about the elderly” policy on a society where everyone hopes to grow old and enjoy retirement? That could have strong economic implications.]

          A journalist or commenter endorsing a treatment does not run this risk, and a pharma company reaps a monetary benefit.

          But yes, sometimes when you know the risks, you can endorse something without a proven benefit, but even that is not neutral. Allowing compassionate use of hydroxichloroquine deprives patients of it that need it for other chronic conditions. That’s a good reason to cancel that permission again.

          Thankfully, no endorsement is needed for vitamins, since they’re not actually medication. Although you can overdose on them, and there have been NHST studies showing elevated risk from taking supplemental vitamins. I’ve also heardthat if you are taking high doses regularly and then stop, you may have withdrawal effects because your body is no longer used to the lower dose.

          That is why the medical profession unanimously endorses a BALANCED DIET containing a variety of fresh food; this takes care of your vitamin levels and keeps you healthy. Why pay money for vitamins to big pharma when you could give it to our farmers instead, and be healthier in the bargain?

        • To me, this right there is p-value reasoning.

          I am not comparing a p-value to an arbitrary cutoff, or checking to see if the uncertainty interval contains zero to make my decision. There is nothing NHST about it.

          Does your “p-value reasoning” include taking into account the costs/risks of a medical treatment or the clinical benefit/significance. No, the calculation entirely ignores all that crucial info.

        • That is why the medical profession unanimously endorses a BALANCED DIET containing a variety of fresh food; this takes care of your vitamin levels and keeps you healthy. Why pay money for vitamins to big pharma when you could give it to our farmers instead, and be healthier in the bargain?

          The medical profession used NHST to decide the standard should be to put all these patients on ventilators unnecessarily. This was the time for their “gold standard” international RCTs to shine and it killed a bunch of people.

          I have been avoiding the medical profession for years because I saw hwo they were arriving at their conclusions. And their “balanced diet” was mostly low-fat grain-based diet. Now we have an obesity epidemic.

        • Anoneuoid –

          > And their “balanced diet” was mostly low-fat grain-based diet. Now we have an obesity epidemic.

          FWIW, a comment like that makes it harder for me to put your other arguments in a context. A low fat (whole) grain-based diet does not cause obesity – especially when the recommendations for balance also included a lot of vegetables, eating fish, etc. So that looks like rhetorical spin to me. Rhetorical spin can sometimes reflect an agenda.

          Now obviously, there’s no particular reason why you should care about how I”m evaluating the context of your other arguments…but like I said, FWIW.

        • A low fat (whole) grain-based diet does not cause obesity

          Eat a high-fat meat-based diet for a week and you will see that what you thought was hunger was actually carb addiction. It is so obvious, anyone can check it for themselves but all the medical literature uses a ridiculous definition of “low-carb” (< 50% calories from carbs instead of < 5%).

          Once you experience that you will know there is no way this diet doesn’t tend to lead to obesity (of course it is not sufficient alone to cause obesity).

        • Anoneuoid –

          > Once you experience that you will know there is no way this diet doesn’t tend to lead to obesity (of course it is not sufficient alone to cause obesity).

          But that’s kind of the same spin.

          There’s no reason to think there’d be an obesity epidemic in this country with people eating a largely low meat, (whole) grain-based diet, mixed with a lot of vegetables and fish, if they reduced the amount of highly processed foods they eat, and reduced their sugar intake, and got more exercise, and ate smaller portions, etc.

          There seem to have been some biasing influences on the research process that led to recommendations for reducing animal fat intake – but that doesn’t mean that there’s zero value in reducing zero animal fat intake from the level typical for many Americans.

          > Eat a high-fat meat-based diet for a week and you will see that what you thought was hunger was actually carb addiction. It is so obvious, anyone can check it for themselves but all the medical literature uses a ridiculous definition of “low-carb” (< 50% calories from carbs instead of come up with some model of the uncertainties in benefits, risks, and costs based on assumptions you put some thought into and make a decision. Eg smaller expected benefit could still be worthwhile for a cheap and safe treatment. But we would require large expected benefit for an expensive treatment with lots of dangerous side effects.

          What I’m trying to interrogate is your argument advocating for the “instead” part of that earlier statement – to suggest replacing the one with the other. I’m wondering about the whole baby with the bathwater aspect.

          And so then, the pattern of your argument about diet raises some questions for me

        • There’s no reason to think there’d be an obesity epidemic in this country with people eating a largely low meat, (whole) grain-based diet, mixed with a lot of vegetables and fish, if they reduced the amount of highly processed foods they eat, and reduced their sugar intake, and got more exercise, and ate smaller portions, etc.

          Well, refined sugar was considered a medicine until recently. Like you were allowed to consume it during your lent fasts, etc. So taking that is more like an addictive drug. I really haven’t eaten a sweet since I was like 10 years old, soda grosses me out, etc so that did not affect my experience.

          What I am talking about is complex carbohydrates from grains/potatoes/etc. The point is that when you limit that to < 5% of your calories, then you consume less total calories… simply because you don’t feel like consuming as many calories. I’d even say at 20% you notice a difference in appetite. Like I said, that feeling you think is “hunger”, is not normal hunger.

          Try it for one week to see. The only problem is you basically can’t eat out so most people get lazy eventually and relapse into their addiction.

        • Anoneuoid –

          > Try it for one week to see. The only problem is you basically can’t eat out so most people get lazy eventually and relapse into their addiction.

          This goes to the part I messed up with that cut-and-paste.

          It’s almost as if you’re suggesting uncontrolled anecdote, or observation, as a replacement for NHST. And so I’m trying to interrogate the either/or pattern of the argument. My question is about the baby/bathwater aspect.

          > Well, refined sugar was considered a medicine until recently. Like you were allowed to consume it during your lent fasts, etc. So taking that is more like an addictive drug. I really haven’t eaten a sweet since I was like 10 years old, soda grosses me out, etc so that did not affect my experience.

          But that isn’t really germane to the point I was trying to make – which was that it seems to me that you’re arguing that the research results that suggested benefit to reducing animal fat intake, or dietary fat in general, is what has led to the obesity epidemic. And so the line of causation is essentially that [weaknesses of the NHST approach to research] ====>> [obesity epidemic in the US].

          I think it’s a factor, but that analysis is just too straight-line and simplistic.

          > What I am talking about is complex carbohydrates from grains/potatoes/etc.

          Sure, but complex carbs from grains and potatoes have been consumed in high value in a lot of cultures for a long period of time without the rates of obesity we have in this country.

          > The point is that when you limit that to simply because you don’t feel like consuming as many calories. I’d even say at 20% you notice a difference in appetite. Like I said, that feeling you think is “hunger”, is not normal hunger.

          Try it for one week to see. The only problem is you basically can’t eat out so most people get lazy eventually and relapse into their addiction.

          I am fairly familiar with many of the arguments your alluding to w/r/t high fat/low carb diet. I’m really not trying to debate their merits. I’m looking at your argument from more of a structural standpoint.

        • It’s almost as if you’re suggesting uncontrolled anecdote, or observation, as a replacement for NHST. And so I’m trying to interrogate the either/or pattern of the argument. My question is about the baby/bathwater aspect.

          I do consider anecdotes from random people on the internet to be a superior form of knowledge to NHST. I am not kidding. A more careful case study is obviously better than a random anecdote, a series of cases studies better than that, etc. But whether or not there is a statistically significant difference between groups is of absolutely zero interest.

          Sure, but complex carbs from grains and potatoes have been consumed in high value in a lot of cultures for a long period of time without the rates of obesity we have in this country.

          Obviously the problem is consuming too many calories for your activity level. That is the problem fixed by stopping the carb addiction, you will stop eating more than you need.

        • Joshua, Mendel,

          Anoneuoid needs to be taken in small doses, preferably with a pinch of salt ;-)

          He can speak up for himself if I’m misunderstanding his argument so I will try to summarize what I think the argument is:

          1) Good science is about good theory together with good measurement, and good theory evaluation.
          2) measurement is independent of evaluation
          3) NHST is a crappy evaluation technique
          4) Even poor measurement with good theory and good evaluation leads to better results than good measurement with bad theory and bad evaluation.
          5) Most modern medical research has good measurement, zero theory, and NHST evaluation
          6) Therefore if you have decent theories, and a good way to evaluate them, and you have bad data, you’re still likely to get a better result than the kind of stuff you see in medicine these days.

        • Daniel –

          > Anoneuoid needs to be taken in small doses, preferably with a pinch of salt ;-)

          Yeah, that’s exactly what I was trying to navigate.

        • Daniel –

          > 1) Good science is about good theory together with good measurement, and good theory evaluation.
          2) measurement is independent of evaluation
          3) NHST is a crappy evaluation technique
          4) Even poor measurement with good theory and good evaluation leads to better results than good measurement with bad theory and bad evaluation.
          5) Most modern medical research has good measurement, zero theory, and NHST evaluation
          6) Therefore if you have decent theories, and a good way to evaluate them, and you have bad data, you’re still likely to get a better result than the kind of stuff you see in medicine these days.

          Let me chew on that.

          FWIW, I have a what I consider to be a similar frame. My feeling is that people in general are naturally geared towards finding causation. We’re essentially pattern finding machines. That’s how we learn things, by finding patterns, building models, etc. But that often leads us to think we see patterns that actually don’t exist.

          And so we go around finding correlations and from them building causations – and that is something that confounds the scientific research process because without plausible mechanisms speculating about causation can be no different from navel gazing. Statistical methodology, designed to address the problem has severe limitations – but neither is it worthless. It brings some advantages.

          So for me, a key is to work on constructing and interrogating plausible mechanisms when you find associations. Maybe that’s backwards. Maybe you should stick to studying for mechanisms and then exploring how they play out with different associated components. I guess there should be some of both.

          As a teacher, I always focus on finding the nexus point between theory and practice. The practitioners always say that the theoreticians didn’t know what they were talking about and the theoreticians always say that the practitioners are, essentially, ignorant. For me, the key is to go back and forth between theory and practice to see how they inform each other.

        • NHST is a lousy way to interrogate theories, it’s “bad evaluation”. What is a better evaluation strategy:

          1) try to come up with as many plausible theories as you can, and have each of them make predictions about what should be seen in the data

          2) Find whatever data you can, or better yet collect good quality data, and see how well the theories predict the data

          3) Downweight those theories that fail to predict the data well, while upweighting those theories that do predict the data well.

          This is exactly what comes out of Bayesian reasoning, whether you consider a “theory” to be something like “the statement about what quantity an average will be” or something like “reaction diffusion transport equations”

        • > 1) try to come up with as many plausible theories as you can, and have each of them make predictions about what should be seen in the data

          One of these theories is the NH

          > 2) Find whatever data you can, or better yet collect good quality data, and see how well the theories predict the data

          compute p-values

          > 3) Downweight those theories that fail to predict the data well, while upweighting those theories that do predict the data well.

          don’t reject anything

          you don’t even need p-hacking here, if you’re the slightest bit creative and have listed enough theories in step 1, one of these will randomly win over the “placebo” theory even if there is no effect to be observed, in fact there is a 50% chance that ANY single theory will have better evidence going for it than the null hypothesis, and since you’re running on bayesian statistics, your end result only depends on your priors. Prejudice in, prejudice out.

          Honestly, the only difference here is how much weight you’ll accept and how you determine it, and you’re wishy-washy about it. In my single coinflip experiment, upweigh the theory “aleays geads”, downweigh “always tails”, but “it’s random” in the middle. What do, and why?

        • Anoneuoid –

          > I do consider anecdotes from random people on the internet to be a superior form of knowledge to NHST. I am not kidding. A more careful case study is obviously better than a random anecdote, a series of cases studies better than that, etc. But whether or not there is a statistically significant difference between groups is of absolutely zero interest.

          Yeah, part of what got lost earlier is that I don’t mean to diminish the importance of anecdote or simple observation – and I agree that there can be a false sense of security of thinking that controlled studies protect against the kind of bias one is subjected to with anecdotal reasoning.

          But were I get stuck on this is the either/or aspect of your argument. I don’t think there is particularly an advantage to deciding their’s no value to NHST. If only to look at it as being more careful about in what kind of context you take what kind of approach.

          But I’m going to be thinking about it.

          > Obviously the problem is consuming too many calories for your activity level. That is the problem fixed by stopping the carb addiction, you will stop eating more than you need.

          I’m pretty sure you’d agree that it’s more complex than that. The “energy balance” concept is overly simplistic as well, IMO. It isn’t simply a matter of calories in vs. energy expended through activity – although that of course is a factor.

        • What value does NHST have? I am sure once you share that we will see at least one logical flaw in your reasoning.

          Can it tell you whether a treatment is worthwhile?
          – No, we need to do some cost-benefit analysis that requires info on practical significance, and cost/risk. NHST tells us none of this.

          Can it tell you if your theory is promising?
          – No, to do that you need to test the predictions you deduced from your theory. Testing a strawman model doesn’t answer anything about our own theories.

        • Anoneuiod –

          > What value does NHST have? I am sure once you share that we will see at least one logical flaw in your reasoning.

          Ok, I’ll try. But be gentle. I’m a delicate soul and I am basically statistically illiterate.

          My sense is that when you conduct an individiual NHST you should look at it as extremely conditional. And you should employ methods with extreme diligence to alter the conditionality, at least around the edge. Control as best you can for randomness. Control as best you can for representativeness. Don’t work with cross-sectional data unless you can avoid it, always try to work with longitudinal data. Use the available statistical tools to increase power. Do a lot of statifications and adjustments in your anlaysis.

          And you never extrapolate from unrepresentative data. And, importantly, you don’t try to infer causality unless you have longitudinal data.

          I’m sure that if I knew statistics, I could describe other statistical tools people use to increase the power of their analysis.

          And then, with that every conditional result, you conduct a lot of other related studies. You put your work up for review from other people who have spend a lot of time studying similar material. Effectively, you do the best you can to replicate your findings.

          And then you perform meta-analyses to pool the results of many similar studies, and you perform analysis on the meta-analyses to see how things like sample size, or power, correlate with the findings over all those studies.

          And then you take all you find with a grain of salt. And you realize that at best, you’re hedging against uncertainty, with the somewhat dubious assumption that all that work will reduce uncertainties even if it isn’t dispositive and will likely prove wrong, at least to some extent, over time.

          And you try to remain humble. And you try not to get scared by uncertainty. And you have a sense of humor. And you get out and meditate sometimes. And you recognize the power of motivated reasoning. And you remember that you are the easiest person for you to fool.

          Now I have a sense that for all the problems people encounter with that recipe, we still have a “signal” of advance over the “noise” of the flawed system of NHST. And the “replicability crisis” for all of its troubling aspects, doesn’t mean that there is no such signal. It’s smaller than a lot of people think and it’s smaller than we’d like, but it exists. So, I think you don’t throw the baby signal out with the bathwater noise.

          > Can it tell you whether a treatment is worthwhile?

          I reject the binary question there. Can we hedge on the probabilities that the treatment is worthwhile given the results of our NHST process as I described above? That’s the question I would ask.

          – No, we need to do some cost-benefit analysis that requires info on practical significance, and cost/risk. NHST tells us none of this.

          Right. NHST don’t stand alone as a way to answer an either/or question. That’s what I’ve been trying to say.

          Can it tell you if your theory is promising?
          – No, to do that you need to test the predictions you deduced from your theory. Testing a strawman model doesn’t answer anything about our own theories.

          Well, I guess that’s less binary. I think you can gain information to help judge the relative potential.

        • Can we hedge on the probabilities that the treatment is worthwhile given the results of our NHST process as I described above?

          We don’t have to wonder. You can see the current NHST-driven medical system in the US is full of very expensive and dangerous chemicals/procedures of dubious benefit.

          Most of your post has nothing to do with NHST (checking a p-value from a strawman null model against an aribtrary threshold) but is just general “be cautious”.

        • Thanks! I’m not surprised really. Glutathione or NAC are similar to vit C in that not going to hurt anyone, they’re known safe, and whenever you have massive immune response the ability to protect against oxidative damage, and downregulate immune activity is going to be important.

          Have you seen any of the stuff on vit D insufficiency? Vit D is known to have a role in downregulating various cytokines, for example:

          https://www.ncbi.nlm.nih.gov/pubmed/25331710
          https://www.ncbi.nlm.nih.gov/pubmed/16600924

          Some are claiming that vit D deficiency is a strong risk factor for severe disease:

          https://www.bmj.com/content/369/bmj.m1548/rr-6
          https://www.ncbi.nlm.nih.gov/pubmed/32252338

          There are several trials that haven’t recruited patients yet.

        • I’m intrigued by the vitamin D angle, and curious to see results with IV ascorbate. My own guess here is that for preventing and ameliorating the cytokine storm mediated progression, a class of molecules called ‘pro resolving mediators’ are going to turn out to be very useful. These are downstream metabolites of long chain omega 3 fatty acids and turn out to be critical for successful resolution of innate immune driven inflammatory processes. Some Pubmed and Google Scholar searching will turn up a variety of interesting recent papers. Notably, I found some animal model work linking them to successful outcomes in other respiratory diseases like RSV (which I happened to have a bad bout with as an infant). To my knowledge, no one on the front lines is trying these, but I convinced a couple of my healthcare friends to start taking them prophylactically. I use them as supplements to improve exercise recovery and as alternative to NSAIDS and find the benefits quite impressive.

        • Chris, thanks for this info I’m very interested to learn more about this class of molecules can you email me some info? I suspect you have my email but if not I’ll try to look you up and send you a request first

        • I use them too. I also get an hour of vitamin D via sunlight.

          I haven’t had a respiratory infection and don’t get the flu shot either. Maybe one short duration of bronchitis about 15 years ago.

        • Haven’t really looked into vitamin D, but the little I did didn’t tell me why it would be low or why exactly it helps (and generally I don’t trust these non-quantitative chains of “x up/down-regulates y which does z” generated from stringing together NHST results).

          But, like all vitamins, if it is lower than it should be then correcting that is going to help.

        • Vit D is an immune system modulator, and deficiency is known to be a big risk factor for autoimmune disorders like MS, Rheumatoid Arthritis, etc as well as related to higher levels of respiratory infections. There are a number of tissues in the respiratory tract that directly utilize Vit D, including in the mucosa of the sinuses. An important role for Vit D is to downregulate cytokines:

          https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5188461/

          I know… NHST is everywhere, but I think the picture is more or less clear that vit D plays an immune modulation role, and this disease, the worst form of it, seems to be about a massive tissue-damaging immune reaction.

          This disease hit the US in late Jan early Feb, and peaked in April in the hard hit areas. Places like NYC or NJ it was covered in snow in Feb, Mar, no-one was outdoors. in nursing homes in general no one goes outdoors ever, and elderly people have difficulty synthesizing vit D anyway.

          It seems plausible to me that it’s a part of the picture, allowing the immune system to go out of control due to insufficient regulation.

          It’s cheap and essentially cost free to give 100,000 IU or more intramuscular vit D as soon as anyone tests positive for COVID-19 and it has essentially zero downside. vit D toxicity is unheard of, and studies have given 600k IU or more in one dose without issues.

        • There are food sources of vitamin D, for example eggs, fish, milk products. Since there’s a connection between vitamin D and osteoporosis, a well-run care home probably ensures their residents have a diet rich in vitamin D (and all other vitamins).

          I don’t see how this is a factor unless there is a specific malnutrition, and there shouldn’t be one in an organized setting unless there is malpractice.

        • vit D toxicity is unheard of, and studies have given 600k IU or more in one dose without issues.

          That is surprising given it is fat soluble. Apparently the conversion is 40 IU/ug, so that would be 15 grams. From table 5 here 15 g/day for 96 weeks yielded blood levels of 221 nM. For MM = 385 g/mol that is .085 mg/L. This seems really low since if you assume 5 L of blood volume 0.085*5 = 0.425 mg. So only ~0.003% was detectable in the blood.

          Looking at some other data in that paper it looks like over 0.5-15 g/day yields anywhere between 200-2000 nM blood concentration under different circumstances. So no matter what it is very low, it seems like much lower than I’d expect even for something fat soluble. It makes me wonder if the blood levels really represent anything.

          Also, 12 grams is like one entire bottle of these: https://www.amazon.com/NOW-Vitamin-Structural-Support-Softgels/dp/B001UZPY1O/

          So I think the relative lack of toxicity is more due to people taking very low doses.

          https://www.ncbi.nlm.nih.gov/pubmed/10232622

        • Mendel said, “a well-run care home probably ensures their residents have a diet rich in vitamin D (and all other vitamins)”

          Which raises the question of what percentage of care homes are well-run. When my mother needed to move into a nursing home (about 30 years ago), well-run ones were hard to find, and I would guess that the situation has not changed dramatically since then.

        • Joshua said,
          “> well-run [nursing homes] were hard to find,

          Obviously, depends almost entirely on your finances.”

          Not necessarily so. In my mother’s case, some expensive nursing homes were poor quality, and one of the best-run was run by the county, and less expensive than most privately run ones.

        • Martha –

          > Not necessarily so.

          Perhaps I was too categorical – but I’d suggest that maybe your experience from 30 years ago might reflect a different landscape.

          I have looked for “nursing homes” and their parallel of “life care communities” a fair amount over the last decade for my parents and extended families. And my life partner was a hospice nurse for decades. And her daughter was a visiting homecare nurse. Other community-based nurses in the family as well.

          From those experiences, I would say that if you have money, you can find institutions that provide a very high level of health services and other sorts of services. Maybe it could be different in other areas of the country (my experience was on both coasts)…

          On the other hand, my experiences also have shown me that the level of health services and other sorts of services is pretty poor in many institutions for people with less money. Shockingly so, sometimes. I never heard of anything similar to what is fairly typical in the lower cost institutions, among the higher cost institutions. Not to say that all was perfect. Far from it. There was incompetence and worse, not all that infrequently.

          But those issues notwithstanding, I would still say that if you have money, good ones are not hard to find. Not sure how to check if my impressions are consistent with the larger reality….

        • Food intake is sufficient to prevent rickets, but just about that’s it. It’s very hard to get “healthy” levels of vit D from food. In the US something like 40% of people are considered insufficient (less than 50ng/ml)

          anoneuoid: 600,000 IU / (40 IU/ug) = 15000 ug = 15mg so you’re off by a factor of 1000 ;-)

          Of course you could get toxicity if you worked at it, but you’d have to work at it. People don’t get toxicity in practice because standard pill doses are like 2000IU so you have to take 300 pills at once to get into the high range of things that are tested to be safe. If you come up with a test at something like 20nm/ml the standard response is to give something like 40,000 IU a week for 6 weeks and recheck.

          The thing about vit D is that there are multiple forms of it, the amount of active form in the blood is regulated by the kidneys. Giving large doses basically provides a reservoir which your kidneys can then use to regulate healthy levels of the active form.

        • anoneuoid: 600,000 IU / (40 IU/ug) = 15000 ug = 15mg so you’re off by a factor of 1000 ;-)

          Ah, something seemed off. Too many unit conversions… But so even a huge dose is only 15 mg.

    • Daniel,

      What if the “excess_deaths” count occurs among a specific sub-population? Do you think it is still valid to extrapolate the resulting IFR to the entire population?

      For example, NJ has 48% of its COVID-19 deaths (positive+suspected; https://www.nj.gov/health/cd/topics/covid2019_dashboard.shtml) being residents of longterm-care facilities. This seems to be the case with other states as well; among 23 states reporting, the mean is 34% for COVID-19 deaths being residents of longterm-care facilities, with 6 states having a 50%+ value (https://www.kff.org/medicaid/issue-brief/state-reporting-of-cases-and-deaths-due-to-covid-19-in-long-term-care-facilities/)

      Since residents of longterm-care facilities don’t interact with outside society much, then perhaps the IFR for this population should be separate to IFR for the rest of the population?

      • death is so strongly associated with age that I don’t think IFR should be discussed except when thought of as a function of age. Unfortunately there’s not always good data on deaths by age. The CDC is putting out something though.

        https://www.cdc.gov/nchs/nvss/covid-19.htm

        I wish they had a similar *cases* dataset… I can’t seem to find it, it’s hard for me to understand how crappy the CDC response has been. Why is covidtracking.com lightyears ahead of them?

        • > death is so strongly associated with age that I don’t think IFR should be discussed except when thought of as a function of age.

          +1

          I agree and I’ve been saying that for a while. Someone here said that we should consider it as one disease for younger people and a different disease for older people.

        • I completely disagree. To the extent that we want to use the flu as a comparison and to the extent we only care about IFR and not infectiousness, there is data to support the idea that COVID-19 is 5 – 10 times more deadly and more serious than the flu for all age groups but the very young (the flu IFR curve against age is somewhat U-shaped with higher infant mortality). In other words, consider relative risk not absolute risk. Death is strongly associated with age in the flu as well: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3873104/

          “Someone here said that we should consider it as one disease for younger people and a different disease for older people.” You can say the exact same thing about the flu. I’m 37, according the crude, but probably ballpark right look here: https://github.com/jbloom/CoV_vs_flu_CFR/blob/master/cfr_stats.ipynb, the IFR for my demographic (18-49) for the flu is 0.0103%

          The IFR for COVID-19 for my demographic is something more like 0.1%. That’s 10 times as deadly.

          For 75+ demographic, the flu IFR is 1.4%. It’s more like 10% for COVID-19. Again 10 times as deadly.

          I don’t want the flu even though I have less than a 1 in 10000 chance of dying from it. The flu sucks. COVID-19 sucks 10 times as much.

          I already posted this link, but I will again: https://github.com/clauswilke/COVID19-IFR

        • Dalton –

          > To the extent that we want to use the flu as a comparison and to the extent we only care about IFR and not infectiousness, there is data to support the idea that COVID-19 is 5 – 10 times more deadly and more serious than the flu for all age groups but the very young (the flu IFR curve against age is somewhat U-shaped with higher infant mortality).

          Sure. If I were appointed god of the universe I’d say we should both not focus on the overall IFR and not compare an aggregated IFR for COVID to an aggregated IFR for the seasonal flue.

          But back in the real world, sure, when someone says that the IFR for COVID is about the same as for the seasonal flu, by all means bring out the relevant data. Especially when they’re making the comparison to weigh in on policy for how to address the pandemic.

        • Joshua,

          I’m not saying that we should be comparing the IFR for COVID-19 to the flu, because as we’ve all been pointing out there are lots of differences between the flu and COVID-19. I’m definitely not saying that IFR for COVID is about the same as for the seasonal flu. I’m saying it’s very likely 10 times worse. Maybe I’m misinterpreting the comments I’ve been reading, but when I interpreted you comment as saying that COVID-19 has an IFR similar to the flu when you look at younger people (i.e. 0.1% or thereabouts). If I read that wrong, I apologize, but I wanted to comment on it because that seemed to me a misleading argument.

          What I’m taking issue with is this idea that this is somehow a disease that only the elderly need to worry about. Or that this is “two different pandemics” one affecting the elderly and at risk and one affecting the rest of us. An IFR of 0.1% for the “young and healthy” is very, very serious disease (especially when you consider it’s infectiousness). The flu comparison drives that point home. When you consider that the seasonal flu is the 7th leading cause of death in the United States and that is with a prevention program that focuses on vaccinating at-risk populations. If we simply multiply the annual death rate for the flu by 10, COVID-19 is in the running for the leading annual cause of death with a toll of more than half-million people.

        • Dalton –

          > I’m not saying that we should be comparing the IFR for COVID-19 to the flu, because as we’ve all been pointing out there are lots of differences between the flu and COVID-19. I’m definitely not saying that IFR for COVID is about the same as for the seasonal flu. I’m saying it’s very likely 10 times worse.

          I will admit to being a bit confused. I saw you as saying that we should make the comparison to the extend that doing so will debunk the talking point that “it’s just like the seasonal flu.” And I agree with that, even if I think that we shouldn’t be comparing COVID to the flu unless we do that in a sophisticated manner.

          > Maybe I’m misinterpreting the comments I’ve been reading, but when I interpreted you comment as saying that COVID-19 has an IFR similar to the flu when you look at younger people (i.e. 0.1% or thereabouts).

          No – I don’t have any reason to think it’s more or less similar with young people than it is with older people. If it is, then that’s maybe useful information, but its usefulness is very limited since the fatality (and morbidity) rate on young people explain so little of the overall impact. It’s a drop in the bucket and relatively unimportant.

          > If I read that wrong, I apologize, but I wanted to comment on it because that seemed to me a misleading argument.

          Now why would you apologize if what I write is unclear?

          > What I’m taking issue with is this idea that this is somehow a disease that only the elderly need to worry about.

          Yeah – I don’t think that it is. One thing that people do is use that target to focus on mortality – as if mortality is all we need to think about. We need to think about morbidity, and pressure on our healthcare workers, and time lost to work, and absenteeism at the workplace, and younger people getting infected and then spreading it to older people – all of which might be somewhat limited in their own right (except the last one) but added together seem to me to be quite important.

          > Or that this is “two different pandemics” one affecting the elderly and at risk and one affecting the rest of us. An IFR of 0.1% for the “young and healthy” is very, very serious disease (especially when you consider it’s infectiousness).

          Sure – but at some level, aggregating the mortality rates across age groups just strikes me as being somehow missing the point. Maybe I need to think about it more – but as bad as the impact is on young people, it is so much more devastating for older people. I guess I think that because the skew is so heavy, that aggregating the data leads towards people undervaluing the real level of impact.

          > The flu comparison drives that point home.

          Except I don’t think the do for the most part. They may do so for you, but there is a reason why those who think this is all just an alarmist panic keep trying to say it’s just like the flu. You can only counter that kind of informational frame to a limited degree by pointing to facts. If aren’t familiar with it, Google the problems with the “deficit model” in addressing these kinds of issues.

          > When you consider that the seasonal flu is the 7th leading cause of death in the United States and that is with a prevention program that focuses on vaccinating at-risk populations. If we simply multiply the annual death rate for the flu by 10, COVID-19 is in the running for the leading annual cause of death with a toll of more than half-million people. In that sense, focusing on the fatality for older people is also likely to be unconvincing form many. So I’m not suggesting that one approach is dramatically more effective than the others….I guess as I’m thinking about it, maybe I’m just saying that because of the skew, I feel like I get a clearer sense of the impact when I look at the impact on the vast majority of people who are the most directly affected?

          Right – but you have a lot of people extremely convinced that multiplying the flu X 10 is an alarmist attitude that leads to draconian measures limiting freedom.

        • Joshua,

          Agreed. In states like NJ and MA, there are two pandemics, it seems: one in the general population, where the IFR seems low and a second among LTCFs that is very high.

          In NJ, for example, there are 3247 deaths recorded who resided in a LTCF. Among all deaths in NJ (as of report 4 days ago; https://www.nj.gov/health/cd/documents/topics/NCOV/COVID_Confirmed_Case_Summary.pdf), ~80% were age 65+. Assuming this percent holds relatively constant, then residents of LTCFs account for 60% of COVID-19 deaths.

          Therefore, the LTCFs — whose population is isolated and distinct form the general population (this is a prior of mine) — are skewing the data upward. If you exclude them from calculating IFR, and also assume a ballpark percent of total infections at 15% (since NJ is densely is most areas; major exception being the Pine Barrens), then the age-stratified IFR for NJ would be 0.1% for those age 0-64, which includes those with risk-factors; calculation: (1334 deaths age 0-64) / (8,882,000 population * 0.87 age 0-64).

          Not claiming this is gospel whatsoever; rather it is an example how one factor like location can really skew results

        • Adding to my above comment for the sake of completing the though exercise.

          Assume that among the 1334 deaths age 0-64, 98% had 1-or-more underlying condition. (NJ is not providing this data; so I’m assuming data from NYC, Italy, MA, etc., track similarly). That is 27 deaths age 0-64 of otherwise healthy people. That means the IFR if you are a health person age 0-64 in NJ is 0.003%.

          Again, this is to illustrate the trappings of IFR, like Phil describes — it is relative to whatever bounds you define the infected population by.

          It is still an important value to measure, of course: a highly infectious pathogen having an 0.1% IFR for ALL people is very scary; a highly infectious pathogen having 0.01%-0.005% IFR or less for healthy people and 0.5% IFR for at-risk people is still a problem and not something to disregard, but one that is much less frightening than the former.

          P.S. Something that has really started troubling me: All this Social Distancing may have failed at its most important goal — protecting the most at-risk people, namely those living in LTCFs. As my above comment describes, the average number of deaths for 23 reporting states is 34%, with multiple states at 50%+. Something went wrong. Not sure what caused this failing and not placing any blame, but its upsetting that all this effort to Social Distance has not seemed effective at protecting those most at-risk. (A prior of mine is my grandparents are part of the LTCFs category, so the aforementioned statistics really scare me.)

        • Twain said,
          “Something that has really started troubling me: All this Social Distancing may have failed at its most important goal — protecting the most at-risk people, namely those living in LTCFs. As my above comment describes, the average number of deaths for 23 reporting states is 34%, with multiple states at 50%+. Something went wrong. Not sure what caused this failing and not placing any blame, but its upsetting that all this effort to Social Distance has not seemed effective at protecting those most at-risk.”

          Very good point!

          “(A prior of mine is my grandparents are part of the LTCFs category, so the aforementioned statistics really scare me.)”

          Understandably! I think we need much more discussion regarding the virus and people in LTCF’s. It might be going on in some other forum, but I don’t know where that might be. But things that need consideration (off the top of my head) include: What are the possible contributing factors to the higher coronavirus fatality rate of older people in LTCF’s? (e.g., how much is just age; how much is factors that cause people to be in LTCF’s rather than living independently; how might the care/policies in LTCF’s be changed to lessen the chances of infection and/or improve the prognosis if residents acquire the infection)

          For somewhat selfish reasons (I am over 75, but live independently), I would also like to see more data and comparisons (by age group) between people who live in LTCF’s and people who do not, stratified by other factors (e.g., disabilities; spousal disabilities; education; and/or other factors). Also comparisons between different (or different types of) LTCF’s. Lots of possible contributing factors!

        • Martha,

          Agreed! I’m frustrated by the lacking public discussion/information about why the virus causes so many deaths in LTFCs. Hopefully it will happen or is happening but we aren’t privy to it yet.

          I hypothesize the issue may be secondary-infections. LTCFs are notorious for them — they are an ideal environment for bacteria, viruses, and fungi to thrive. Loads of vulnerable hosts. Add COVID-19 to the situation and its a recipe-for-disaster — even if COVID-19 does not kill a resident outright, its enough to weaken them and give other pathogens the perfect opportunity to strike. But at present, we don’t know how much this actually is happening.

          Overall, it would be nice to have better data stratified by age, demographics, commodities, etc. Personally, I’d like to see more resources go toward helping physicians (who are overwhelmed and can’t be expected to thoroughly stratify charts) log/analyze their data; because said data really helps evaluate how “risky” the pathogen really is and can inform treatment. Something to consider and learn from for the future!

        • “death is so strongly associated with age that I don’t think IFR should be discussed except when thought of as a function of age”

          I feel like you could make that statement about almost any disease. Age and death have strikingly high correlation.

        • Dalton,

          I think what you are saying is Daniel’s point — it matters to what degree age (and any comorbitity, for the matter) correlates with death. To ignore age and try to represent risk of dying from infection for everyone does not tell the full story.

          For example: Does likelihood of dying from infection increase linearly with age? Exponentially? Something else? Determining that relationship seems crucial to evaluating the risk a disease poses to the general population.

        • Twain, Dalton –

          > I think what you are saying is Daniel’s point — it matters to what degree age (and any comorbitity, for the matter) correlates with death. To ignore age and try to represent risk of dying from infection for everyone does not tell the full story.

          Interesting points. How much does age explain health risk with cancer, or diabetes, or heart disease, obesity, or….

          So I read a lot of people saying that we should be spending more energy on evaluating the economic impact of avoiding death from COVID on the economy, and we should be considering if we should basically just accept the deaths and morbidity from COVID as inevitable, since it’s mostly just a bunch of old and sick people dying.

          So to what degree should we be thinking the same way about the amount we spend on those other diseases?

          I’m thinking that would could maybe save around $11k per American. Maybe we could just all upgrade to the premium classes on our vehicles and skip any spending on healthcare whatsoever?

        • >So I read a lot of people saying that we should be spending more energy on evaluating the economic impact of avoiding death from COVID on the economy

          Maybe you’re reading smarter people than I am (or maybe you’re being sarcastic?), but that seems to me a false (and hugely cynical) choice. People are somewhat rationale. We can’t just open the economy and expect everybody to participate in it when there is deadly serious pandemic going on, and when we don’t have effective testing to even know what the risk is. My wife and I are in our 30s. We have jobs and live in a location where we can minimize our social interactions. We are not willing to risk getting seriously sick and dying just to get a haircut or just to go to a restaurant. My in-laws and parents certainly aren’t. It’s not like we can draft people to participate in the economy when they don’t want to because of the incredibly rational fear of dying or getting seriously ill.

          >I’m thinking that would could maybe save around $11k per American. Maybe we could just all upgrade to the premium classes on our vehicles and skip any spending on healthcare whatsoever?

          So let me get this straight… you think “we” could save a bunch of money by not offering healthcare to any American with some level of inherent mortality risk and all those “savings” would be distributed evenly among the survivors? That idea is somehow and hilariously both pie-eyed socialist and terrifyingly anti-social. I’m going out on a limb and say you’re being sarcastic. But my sarcasm meter is broken. I could’ve sworn I head a significant authority figure suggest we could all inject disinfectant as a potential treatment without a hint of sarcasm, but it turns out I was wrong.

        • Dalton –

          Let me work in reverse order, as in the end I think it might be more efficient.

          > So let me get this straight… you think “we” could save a bunch of money by not offering healthcare to any American with some level of inherent mortality risk and all those “savings” would be distributed evenly among the survivors? That idea is somehow and hilariously both pie-eyed socialist and terrifyingly anti-social. I’m going out on a limb and say you’re being sarcastic. But my sarcasm meter is broken. I could’ve sworn I head a significant authority figure suggest we could all inject disinfectant as a potential treatment without a hint of sarcasm, but it turns out I was wrong.

          I would go with sardonic rather than sarcastic. They say that sarcasm gets lost in the Interwebs, but maybe sardonicism does as well?

          What I’d like you to do is put it in your head that I was being sardonic there, and then go back and read the rest of what I wrote and see if you read it any differently. If you think there’s any chance that it wasn’t essentially a Poe, then you’re working from a place where you may not understand anything I write. It was my intention that it was totally clear that I wasn’t suggesting that calculation to be a reasonable one in any fashion. It was basically a mocking ad absurdism. (I’m not suggesting that you’re at fault because my humor missed the mark).

          With that understanding, maybe then we’d have a better chance at connecting our thoughts through a less circuitous route.

        • Joshua,

          > So I read a lot of people saying that we should be spending more energy on evaluating the economic impact of avoiding death from COVID on the economy, and we should be considering if we should basically just accept the deaths and morbidity from COVID as inevitable, since it’s mostly just a bunch of old and sick people dying. […] So to what degree should we be thinking the same way about the amount we spend on those other diseases?

          I’ve heard of this rhetoric (I don’t watch much mainstream news) and its rather disconcerting. Yes, some morbidity and mortality is inevitable for disease. But that does not predicate leaving people to suffer and die so others can flourish. We should always try to protect and heal those who are vulnerable. (As I’m sure you agree!)

          We do have a tricky question to answer: When should we “shutdown” to combat a pathogen and when should we not? What is the cutoff? How to we determine this cutoff?

          COVID-19 has brought this question to the forefront at a scope and rate not seen in modern times (at least since I’ve been alive). Interesting to see how things develop.

        • > But that does not predicate leaving people to suffer and die so others can flourish. We should always try to protect and heal those who are vulnerable. (As I’m sure you agree!)

          I do agree. What I see is that some people consider that as sentimentalism. Mostly, I think, because they think that those people are going to suffer anyway, so there’s no point in spreading the suffering around by hurting everyone economically. I guess I can understand the “calculus” of that equation, but I see a few main problems: First, I think that we have an obligation to elevate the welfare of the people who are putting themselves on the front line of fighting this disease. In my book, they get some extra “post stratification” weighting. I also think we have a particular obligation to those who are the most vulnerable due to societal inequalities. I’m not a big believer in the view that people who are more vulnerable are more vulnerable because somehow this is all a meritocracy. Finally, I think it’s hard to tease out how much of that thinking is some kind of objective calculus, and how much of it is a tribalistic attitude that leads people to think that libz are just a bunch of bleeding-heart hypocrites who only care about those poor suffering people as a way to demean conz. I can kind of get that there is a baseline validity to that – afterall, we’re all kind of sanctimonious hypocrites in some ways, right? But from what I see, there is a kind of identity-aggression that is taking place here that drives much of the antipathy.

          > We do have a tricky question to answer: When should we “shutdown” to combat a pathogen and when should we not? What is the cutoff? How to we determine this cutoff?

          Yes, despite what I said above, I agree that these are real, and extremely difficult questions.

          > COVID-19 has brought this question to the forefront at a scope and rate not seen in modern times (at least since I’ve been alive). Interesting to see how things develop.

          Yes, I’m trying to look at this as an opportunity. I like to think that one outcome from this is that more people are going to have a greater sense of how we’re all connected.

        • > But from what I see, there is a kind of identity-aggression that is taking place here that drives much of the antipathy.

          Agreed. It is happening on all sides, Right, Left, Independent, etc. Dogma helps in small amounts, but too much leads to dissent. As we are clearly seeing presently.

          > I like to think that one outcome from this is that more people are going to have a greater sense of how we’re all connected.

          Also agreed. As I posted above, one thing I hope we discuss and learn from — with all the Social Distancing, how did COVID-19 still manage to ravage LTCFs and similar? What can we adjust for the future? I can’t pose an immediate answer due to lacking information, but this is not something we want to happen in future pandemics similar to COVID-19.

        • Twain –

          Another thought on this…just watching something on TV about how much hospitals are losing right now….Hmmm. Let me think about this.

          One of the major complaints I’m seeing from the “openists” is that hospitals are losing so much money and under so much financial pressure because elective surgeries aren’t getting done.

          They think that this huge loss of hospitals is a problem, and indeed I agree that it is.

          But what is really the root of that problem? Is the root that we’re keeping people at home because we’re trying to limit the spread of COVID?

          Or is the root of the problem the simple fact that we have a fee for service healthcare care system, where funding for medical care come from the revenue stream of fee for services?

        • > But what is really the root of that problem? Is the root that we’re keeping people at home because we’re trying to limit the spread of COVID? […] Or is the root of the problem the simple fact that we have a fee for service healthcare care system, where funding for medical care come from the revenue stream of fee for services?

          Both are likely factors, right? Some are probably scared of receiving elective procedure and contracting SARS-CoV-2; so less demand. The fee-for-service is certainly also a problem…but the issues with our medical system and their impact on COVID-19 (and the inverse) is A LOT to unpack — something for another day :-).

        • Twain,

          I guess what I’m saying is I don’t quite understand why people in the comments keep pointing to the patterns of mortality being concentrated among the elderly as if it’s some big revelation. The elderly and those in long-term health care already have a hugely elevated risk of dying compared to the young and fit. That’s why I think focusing on relative risk is more important than focusing on absolute risk. I don’t have data to support this idea, but I don’t think it’s a stretch to hypothesize that COVID-19 is increasing everyone’s (except maybe the very young) risk of mortality on a fairly even relative basis. That is consider an age-stratified curve for probability of surviving to the next year (see below for a link to such a curve on the log scale). Now take that static situation and introduce COVID-19. I hypothesize that the new curve would like a parallel line above the old curve (on a log-scale). If that hypothesis is evenly approximately correct, then it’s fair to say that this is not a disease that primarily affects the elderly. It affects everyone. We are all at a greater risk of dying. Yes, the absolute risk to the elderly is higher, but it was higher (alot higher) to start with.

          >For example: Does likelihood of dying from infection increase linearly with age? Exponentially? Something else?

          There’s a whole class of statistical models on this. Hell there’s an entire profession based on the pinning down these rates (and actuaries make good money!). I imagine if you graphed the curve of “annual mortality risk” where the y-axis is the 1 – probability being alive in one year against an x-axis of age, that curve would look roughly exponential. (Actually here’s one for France: https://www.ined.fr/en/everything_about_population/graphs-maps/interpreted-graphs/age-risk-mortality/ looks pretty darn exponential to me)

        • Dalton,

          > I don’t have data to support this idea, but I don’t think it’s a stretch to hypothesize that COVID-19 is increasing everyone’s (except maybe the very young) risk of mortality on a fairly even relative basis.

          As a counterpoint, consider NYC: 53 deaths of those age 0-64 with no underlying conditions; so healthy people. Assume the 20% infection value is accurate. Assume 7,224,000‬ people age 0-64 per 2017 Census. That means 1,444,800 infections age 0-64. Assume 40% have underlying conditions (using obesity as a proxy here; would like a better value for accuracy). That means 53/(1,444,800*0.6)=0.006% IFR for health people. In contrast, those with underlying conditions age 0-64, following the same analysis, have 12,217/(1,444,800*0.4)=2%.

          Following the above analysis for hospitalizations but replacing numerator of death with hospitalizations, the likelihood of hospitalization for all people age 0-64 is 1.4%. Now assume 80% of hospitalizations have underlying condition (cannot find data on this stratification for NYC; 80% is for the sake of analysis). Then the likelihood for healthy people drops to 0.5%.

          From this analysis, it seems healthy people have much, much less risk of dying and a lower (but not much lower) risk of hospitalization. If you see a flaw, don’t hesitate to note it.

          > Now take that static situation and introduce COVID-19. I hypothesize that the new curve would like a parallel line above the old curve (on a log-scale).

          If your hypothesis is true, then I agree. But what if the curves start and the origin and diverge at some rate with age (or other factor)?

          > There’s a whole class of statistical models on this. Hell there’s an entire profession based on the pinning down these rates (and actuaries make good money!). I imagine if you graphed the curve of “annual mortality risk” where the y-axis is the 1 – probability being alive in one year against an x-axis of age, that curve would look roughly exponential. (Actually here’s one for France: https://www.ined.fr/en/everything_about_population/graphs-maps/interpreted-graphs/age-risk-mortality/ looks pretty darn exponential to me)

          I’d love to hear an actuary’s opinion on how to approach all this. And I to assume such a curve would be exponential — saying linear was more for the sake of argument.

        • I have a long comment below, but Dalton, I just wanted to say briefly here that I wholeheartedly agree with you that you “don’t think it’s a stretch to hypothesize that COVID-19 is increasing everyone’s (except maybe the very young) risk of mortality on a fairly even relative basis.”

          So far I haven’t seen any subpopulation that is being affected disproportionately (when baseline all-cause mortality is the starting point).

          David Spiegelhalter from the UK is also in agreement when it comes to age as a mortality risk factor. He has a great article here:

          https://medium.com/wintoncentre/how-much-normal-risk-does-covid-represent-4539118e1196

          My long comment below concerns sex. Men are dying from COVID-19 moreso than women (higher IFR for males). But (!) if you consider the “prior” baseline risk, it is nothing different than what you’d expect to see. So giving men estrogen is unwarranted at this point (until someone gets a chance to look at the sex-disaggregated data).

        • Hi Twain, Daniel, Joshua, Dalton, et al.

          I’m new to this forum/blog. I love the discussion here. I’m a statistician working in epidemiology for Veterans Affairs.

          I thought I’d post this comment here because you all are really getting into IFR and whether it should be reported as a whole-population mean or if it is irresponsible or unwise to think of it that way. I would say reporting a whole-population mean is valid and legit, terms of a way to compare with the seasonal flu, as Dalton stated below (“the extent that we want to use the flu as a comparison and to the extent we only care about IFR and not infectiousness, there is data to support the idea that COVID-19 is 5 – 10”).

          Bear with me please because I do have a point that could have practical implications.

          It started with a question from my sister yesterday, who thinks she has potentially been exposed to coronavirus. She woke up yesterday with a loss of smell and taste. Even her fruit smoothie and freshly ground/brewed coffee did not register.

          She was worried and asked me, a numbers guy who works in this field, to give her the straight truth. Assuming she had contracted COVID-19, then what would be her chances of dying from it. Based on the case fatality rates by sex and by age group (she is 34) and from everything I know is biased and wrong about these rates, I came up with 1 in 1,000 as a rough estimate. But what does this 1 in 1,000 mean to her without context? She asked me to put it into context. So I said that her baseline “all-cause“ annual mortality rate as a 34-yo white female in an urban area was probably also, conveniently, around 1 in 1,000. So then it clicked for me and I said, “so it is like the same amount of risk of dying when you live normal, everyday life for a year, but all condensed down into the next 2-3 weeks. If you’re alive in 3 weeks, it will be like you’ve lived through a whole year. You’ll be a year older in terms of having cheated death.” She thought this was a good explanation and had not heard it (mortality risk) explained that way before.

          To confirm what I had told her, I looked up some life tables from the CDC and saw that I was pretty damn close – the probability of a 34-yo female dying (of any cause, based on 2017 life table) in the next year is 1 in 967. The salient observation for my comment post here is that the all-cause annual mortality for a 34-yo male is almost exactly 2 times higher (1 in 483). If you look over all ages, the male:female mortality ratio is always > 1 over all ages and usually hovers around 1.5 – 2.0.

          None of this is probably news to you. But what I’m getting at is something that I think is being widely misconstrued and misunderstood. The facts on the ground are that many more men are dying from COVID-19 than are women. No denying that; it is certainly true in absolute terms. The thing that is not being stated in the news is the background context – that men are continuously, every day of every year, dying more frequently than are women (the direct corollary of men having a shorter life expectancy on average than do women). If you think about it in that context, then it makes absolute perfect sense that we see more men dying from COVID-19! The article below, for example, doesn’t even mention this point of view or context at all.

          https://www.nytimes.com/2020/04/27/health/coronavirus-estrogen-men.html

          Thus, I think it is misleading when the article states the following (without giving the baseline mortality risk context as I showed above with the life table):

          “The gender gap in coronavirus survival became apparent early in the pandemic. Reports from China indicated men were dying at higher rates, but the disparity was attributed to higher smoking rates. But the outcomes were consistent in other countries, with men in Italy dying at higher rates than women, and men in New York City dying at nearly double the rate of women.”

          My response would be, “Of course it’s double – it’s always been double!” What would be actually surprising to me is if it were something like 5 times higher or, on the other hand, if it was equal for men and women.

          This next article makes my same point, not about sex differences, but about mortality risk by age. Open it up in Chrome to see the graphics:

          https://medium.com/wintoncentre/how-much-normal-risk-does-covid-represent-4539118e1196

          The point of that article is that we don’t see anything that we wouldn’t have expected to see when it comes to age being a risk factor for death from COVID-19. [This is unlike the 1918 Spanish Flu, when, unexpectedly, young people in their 20s and 30s were far more likely to die than would have been guessed based just on life tables.]

          I contend that the same goes for sex and probably for race and many other factors as well. The fact that we see disproportionately more African Americans dying from COVID-19 is horrifying – but their mortality (i.e. lower life expectancy) has always been horrifying and has always been disproportionate.

          I think a good analogy is a magnifying glass — that COVID-19 is uniformly magnifying all of the true *natural* differences (the older you are the more likely you are to die; underlying health conditions make you more likely to die, etc.), the so-called “natural” differences (men are more likely to die (all-cause mortality) than women), and the social-inequality-related differences (race, ethnicity, poor/wealthy, rural/urban, health care access, etc.) in all-cause baseline mortality. My hypothesis/contention is that the magnifying glass magnifies/multiplies everything pretty uniformly. It multiplies each individual’s baseline mortality risk by a constant. And that constant is the same for everyone — let’s say it is the number 26.3, for example. The crux of this analogy is that the magnifying glass is uniform and not a funhouse mirror. The object being magnified has disfigurations, but it is not the magnifying glass creating the disfigurations.

          I think a more statistically literate approach (than the approach I read in so many news articles) is to start with baseline mortality risk as the underlying context. I can’t speak to the biological efficacy of estrogen for boosting male immunity, but I can say that, to my mind, the epidemiological case for giving it to men is completely non-existent (as far as my research into it goes) and seems to be based on specious reasoning because that reasoning didn’t start with baseline mortality (i.e. use life tables). I’m all for trying new medicines and ways of doing things to save lives, but it really boggles my mind that the logic of the estrogen trial is so inane and flimsy.

          The practical implication – should it be VA policy to give estrogen to men with COVID-19? I would say, at this point, from an epidemiological point of view, no. In fact, maybe it is women that are dying disproportionately (given their baseline mortality) more frequently than men – we don’t know. We need the data and we need someone to use life table analysis on it. It’s crazy to me that we don’t have access to the number of US cases by sex and the number of US deaths by sex.

          CDC has cases broken out by race/ethnic but not by sex:

          https://www.cdc.gov/coronavirus/2019-ncov/cases-updates/cases-in-us.html

          CDC has deaths broken out by sex but not by race/ethnic:

          https://www.cdc.gov/nchs/nvss/vsrr/COVID19/

          Are they deliberately making this hard for researchers?

          This NPR interview is good but it also doesn’t talk about male-female differences in baseline risk directly (only indirectly in that men have more comorbidities than do women):

          https://www.npr.org/transcripts/842195564

          Germane quote from that: “… the profession of health researchers is actually remarkably bad at taking sex and gender into consideration in their health care provision. And we’re seeing that reflected now by the sheer paucity of sex-disaggregated data coming out from many major countries, for example.”

          This site is all about getting governments to release sex-disaggregated data. Some countries have provided it.

          https://globalhealth5050.org/covid19/#1586248980572-3839d9fe-3b88

          Finally (if you’ve ready this far – thank you!), I would propose a new measure, which is the multiplier I spoke of before. Not the multiplier for how much more deadly COVID-19 is than the seasonal flu, which is likely in the 5 – 10 range, as Dalton says below in his comment. The multiplier is for how much more deadly is this than baseline normal-life mortality risk. If my risk of dying in 2 weeks is the same as a year of normal-life risk, then that multiplier would be 52/2 = 26.0. Is this multiplier already a thing? Is it similar to a proportional hazard ratio? It seems to be, but it’s still different.

          Thanks for your thoughts on this!

        • Hi statflash, welcome to the forum! Thanks for this post. I agree with your analysis, but it *is* a method of discussing the *function* of age… Yes it’s a constant multiplier to the baseline function, but that baseline function is very relevant (ie. exponential increase!)

        • statflash,

          Welcome to the blog! I’m new here as well and pleased by the discussion — its hard to do this in most other socials.

          Great post! Your analog using all-cause mortality is well-done and something I’ll adopt for explaining to friends, colleagues, etc. Thank you!

          I also agree with your statement of using gross IFR for comparing to influenza; we really don’t have any other way to do so. Perhaps if this is the only metric, we really shouldn’t be comparing at all?

          > The crux of this analogy is that the magnifying glass is uniform and not a funhouse mirror. The object being magnified has disfigurations, but it is not the magnifying glass creating the disfigurations.

          I’m not sure if the data supports a uniform magnifying glass; a funhouse mirror (nice analogy, BTW!) may be more appropriate.

          Consider NYC. Assume proportion having underlying conditions is 0.6. Assume reported seroprevelance proportion is 0.2 is accurate. Knowns are (1) total population being 8,400,000 people and (2) proportion being age 0-64 is 0.88. Therefore, the IFR for healthy people is 68 deaths / (8,400,000 people * 0.88 * 0.6 * 0.2) * 100% = 0.007%. Using the same calculation, except using the 12571 deaths with underlying conditions (I’m being conservative and including all deaths with unknown conditions), the IFR is 2%. That is a massive difference. (Source of data: https://www1.nyc.gov/assets/doh/downloads/pdf/imm/covid-19-daily-data-summary-deaths-04302020-1.pdf)

          What are your thoughts on this? I’m a biomedical researcher and engineer, not a epidemiologist with years of experiece. So I very well could be thinking about all this wrong.

          P.S. Thank you for your service to the VA!

        • Well… a lot of the ‘underlying conditions unknown’ deaths might not have had any. So the risk profile might not be that dramatic.

          But it does seem that at least at the youngest ages there is a divergence. Under 18 seems to have an incredibly low risk – wouldn’t one expect a bit higher infant mortality if it was a pure “magnifying glass”?

          Seems like this could be significant.

        • Hi Daniel and Twain, thanks for the warm welcome!

          replying to Twain:
          > I also agree with your statement of using gross IFR for comparing to influenza; we really don’t have any other way to do so. Perhaps if this is the only metric, we really shouldn’t be comparing at all?

          Yeah, I personally agree. I echo what someone else said here that *we* shouldn’t, but everyone in the airwaves and MSM (and the Bakersfield doctors thing) keep wanting to put a number on how many times more deadly than the seasonal flu this is. So if you can’t beat ’em…

          > Consider NYC. Assume proportion having underlying conditions is 0.6. Assume reported seroprevelance proportion is 0.2 is accurate. Knowns are (1) total population being 8,400,000 people and (2) proportion being age 0-64 is 0.88. Therefore, the IFR for healthy people is 68 deaths / (8,400,000 people * 0.88 * 0.6 * 0.2) * 100% = 0.007%. Using the same calculation, except using the 12571 deaths with underlying conditions (I’m being conservative and including all deaths with unknown conditions), the IFR is 2%. That is a massive difference.

          You make a good point. Thanks for the calculation and the link.

          First of all, I guess I would modify my uniform magnifying glass analogy to say that for people under 20 years old, maybe this magnifying glass has a blind spot. They, one could argue, seem disproportionately unaffected by COVID-19.

          Secondly, for your calculation, I think you meant to put 0.4 instead of 0.6 which would yield = .0115%. The 0.6 you said was the proportion with underlying conditions, so 12571 / (8,400,000 people * 0.88 * 0.6 * 0.2) * 100% = 1.42%. Granted, still massively different. But what if you went with the 9,165 with confirmed underlying conditions, yielding = 1.03%. This is still 90 times higher than for the no-underlying-condition group. So I would say 10, even 20, times higher wouldn’t be unreasonable relative risk in baseline mortality. But 90X is too high, so I concede my analogy doesn’t hold up in uniformly magnifying the “underlying condition” vs “no underlying condition” baseline mortality risks.

          I still conjecture, until I see data otherwise (please direct me to any race or sex data), that it (nearly) uniformly magnifies natural differences due to aging (the older you are the more likely you are to die), the “natural” sex differences (men are more likely to die than women), and the social-inequality-related differences (race, ethnicity, poor/wealthy, rural/urban, health care access, etc.) in all-cause baseline mortality.

          Appreciate the kudos, I love my job!

        • statflash,

          Forgot to mention a prior: I have been monitoring the NYC data since they started publishing and have seen almost all “underling conditions unknown” go toward the count of “underlying conditions” over time. Granted, this is not ideal.

          > The 0.6 you said was the proportion with underlying conditions…

          Agh! Late-night typo. It should be 0.4 with underlying conditions and 0.6 without underlying conditions — i.e., following the ~40% of people age 0-64 are obese statistic.

          > I still conjecture, until I see data otherwise (please direct me to any race or sex data), that it (nearly) uniformly magnifies natural differences due to aging (the older you are the more likely you are to die), the “natural” sex differences (men are more likely to die than women), and the social-inequality-related differences (race, ethnicity, poor/wealthy, rural/urban, health care access, etc.) in all-cause baseline mortality.

          A fair conjecture that is likely true. The key is by how much for a given age; but assuming uniform magnifying is reasonable until we know otherwise (it mitigates risk). What Anoneuoid posts below is a nice start. The caveat being it may skew the increase higher than reality for young people, since death counts in the age 0-30 category are very low and therefore likely outliers not fully representing the general population.

          Another thing to consider that I have seen discussed much: Not all underlying conditions are “created equal”. Someone with well-managed hypertension, diabetes, asthma, etc., often have minimal morbidity and have overall healthy vitals; in contrast, those with ill-managed underlying conditions have high morbidity. So even the “underlying conditions” category can over-simplify things too much.

        • Anoneuoid,

          Thanks for sharing. I’d like to see this curve now.

          And yes — from the beginning the hype on ventilators was scary. I hoped it was just groupthink among politicians and pundits and assumed physicians would act differently. Some reports suggest otherwise. Which is sad, if true — the basic physiology of ARDS makes it clear that invasive ventilating cannot do much help, if any, while increasing risk of pressure-induced damage to lungs, kidneys, brain, etc., and increasing likelihood of severe pneumonia.

          Of course, I don’t blame physicians for this whatsoever. They were overwhelmed and overworked. I do blame DOHs and other Public Health officials — whose job is to act as “second brains” for healthcare, especially during times of crisis, to ensure we are making the right choices for treatment.

        • And yes — from the beginning the hype on ventilators was scary. I hoped it was just groupthink among politicians and pundits and assumed physicians would act differently.

          Unfortunately I don’t think we are done learning that the treatment is hurting the patients. How wise is it to put someone who adapted to low blood oxygen saturation on 100% oxygen?

          The only mention I see is from this weird twitter account: https://twitter.com/HMonster19/status/1256016098958311425/photo/1

          They really need to be more careful in distinguishing between side effects of the treatment and illness caused by the virus.

        • > Unfortunately I don’t think we are done learning that the treatment is hurting the patients. How wise is it to put someone who adapted to low blood oxygen saturation on 100% oxygen?

          Exactly. I understand its a may be a “last ditch effort” but the basic physiology of it just doesn’t make sense. So many things could go wrong: hyperoxia, alveolar fibrosis, vascular damage, ventilator-induced pneumonia, etc., are all likely outcomes. Clinical texts ubiquitously say something like, “only use invasive ventilating as a last-ditch effort and for a short during, like 1-3 days; beyond that duration, the risks almost always well outweigh the benefit to patients”.

          Just doesn’t make sense from the outside.

        • @ Twain. The ventilators are clearly dangerous. But I mean just standard cpap or high flow may also hurt these patients. By the time they get to the hospital tissues that have already adapted to the low oxygen saturation are suddenly hit by too much oxygen so you get a kind of reperfusion injury.

          And once again, vitamin c in massive amounts will help.

          https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5861638/

        • This is one reason I’m not totally willing to assume the low estimates of IFR from certain serology studies are wrong due to incompatibility with the NYC / Northern Italy experience — differences in treatment may significantly impact mortality.*

          Obviously it’s totally irresponsible to say ‘COVID death rates are comparable to the flu’ based on a serology study because in hard hit places that is obviously wrong; pretty sure New York State hasn’t seen 0.1% of its population die of seasonal flu. Similarly for northern Italy, etc.

          But it does seem like some places have not seen NYC- or Italy-level ratios of deaths to serology-determined population prevalence.

          *Also, what will IFRs turn out to be in, say, Latin America? I’d expect less ICU availability to raise it, but age demographics to lower it…

        • Thank you all for commenting. I really like the points you’ve made:

          -infant mortality from COVID-19 should be higher based on baseline mortality from life tables — like many of these questions, I think this will need to be analyzed from CDC NCHS mortality data broken out by single age down the road, once that is available, usually a year lag at least.

          -difference between well-managed underlying conditions and not-well-managed — this will be really hard to disentangle.

          I really appreciate the link to that reddit thread, Anoneuoid.

          https://i.reddit.com/r/COVID19/comments/fgydxo/ncov19_mortality_rate_by_age_vs_expected_from/

          Did you see the Medium article I linked to? It does the same thing for the UK.

          I finally found the age and sex breakout for COVID-19 deaths (no luck with cases) at the CDC:

          https://www.cdc.gov/nchs/nvss/vsrr/covid_weekly/index.htm#AgeAndSex

          A simple thing to do with the age-breakout data from CDC is just to take the count of deaths in one age group over the count of deaths in the preceding younger age group. When you get to (age group 55-64 / age group 45-54) and the oldest 3 age groups, the multiple_COVID and multiple_TOtal are very close to the same. Assuming the proportion of cases (cases/subpopulation) is the same within each of the 4 oldest groups (of course it’s not, but assume this for the sake of argument), then age as a risk factor in COVID-19 mortality follows nearly the exact same curve as does age as a risk factor in all-cause mortality.

          Age group COVID and COVID+pneumonia Total Deaths (excluding col to the left) multiple-COVID multiple-Total

          Under 1 year 5 3,720
          1-4 years 4 719 0.80 0.19
          5-14 years 3 1,069 0.75 1.49
          15-24 years 58 6,327 19.33 5.92
          25-34 years 395 13,137 6.81 2.08
          35-44 years 985 18,554 2.49 1.41
          45-54 years 2,721 35,178 2.76 1.90
          55-64 years 6,753 84,393 2.48 2.40
          65-74 years 11,545 130,011 1.71 1.54
          75-84 years 14,844 163,073 1.29 1.25
          85 years+ 16,559 209,385 1.12 1.28
          All Ages 53,872 665,566

        • I don’t know if anyone is still reading the comments to this, but there have been a few more articles that have come out on this theme.

          https://www.medrxiv.org/content/10.1101/2020.04.15.20067074v2

          >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

          Seljak says that getting COVID-19 doubles your chance of dying this year.

          “If you want to know what are the chances of dying from COVID-19 if you get infected, we observed that a very simple answer seems to fit a lot of data: It is the same as the chance of you dying over the next 12 months from normal causes,” said Seljak.

          Current uncertainties can push this number down to 10 months or up to 20 months, he added. His team discovered that this simple relation holds not only for the overall fatality rate, but also for the age-stratified fatality rate, and it agrees with the data both in Italy and in the U.S.

          “Our observation suggests COVID-19 kills the weakest segments of the population,” Seljak said.

          The paper was posted online last week on MedRxiv in advance of peer review and submission to a journal.

      • Agreed.

        To find age-specific and facility-specific data on cases to stratify, one has to sift through the releases from a specific state — whose quality and data vary wildly. MA’s is excellent, NYC’s is decent, NJ’s is awful, for example.

        As to the CDC lagging in reporting deaths (for example, why are KFF and COVID Tracking Project able to keep up-to-date), I’m not sure. It’s certainly frustrating, though. Perhaps the CDC must complete more red-tape to list a death?

  11. To say “many people are immune to the flu, either because they were vaccinated and the vaccine was effective for them or because they had a similar strain of the flu in the past and are still immune,” is to imply that we understand fully how immunity, infection and disease interact. It ignores, for example, innate immunity and more generally treats the hurly burly of life as easily simplifiable.

    • …which is one of the reasons I said “(I’m aware that the distinction between ‘immune’ and ‘not immune’ is not so clear-cut, but that doesn’t invalidate the point).”

      • Even if COVID-19 is like a slightly-worse-than-average seasonal flu in terms of IFR, it would be much much worse from a societal standpoint: it seems that nearly nobody is immune (except perhaps those who have already had it), whereas in any given year many people are immune to the flu, either because they were vaccinated and the vaccine was effective for them or because they had a similar strain of the flu in the past and are still immune. (I’m aware that the distinction between ‘immune’ and ‘not immune’ is not so clear-cut, but that doesn’t invalidate the point).

        To say that “immune” is ambiguous is good. You seem to use it to mean, roughly, “protected from”, “doesn’t get infected by” or perhaps “doesn’t get sick from.” Thus “whereas in any given year many people are (protected from) the flu.”

        Do you think there’s any chance some people are “protected from” infections independently of whether “they were vaccinated and the vaccine was effective for them or because they had a similar strain of the flu (or whatever) in the past and are still immune.” Or might there be more to learn about how life protects itself?

        • Oh sure, some people have stronger immune systems or immune systems that happen to be more effective against that particular virus in spite of lack of previous exposure, yada yada. An exhaustive list would make the point no more clear, indeed would simply distract from it. But feel free to make a more complete list if you like!

        • Yeah who cares about the medicine, let’s pick a bunch of numbers from really poor data to reinforce the status quo. #science

  12. There a couple of things that some people who comment here already know, but some do not, leading to lots of discussion in the comments that keeps rehashing these issues. I’m hoping that by just putting these here I can save some effort.

    And now, we are up to almost 100 comments rehashing these issues again.

    This is actually kind of funny.

    • I know what you mean, but actually more than half of the comments are about different things, like where to get data on the age distribution, and what decision rule to use for deciding whether to shut down, etc etc. Actually way more than half. I learn a lot from the comments, I’m glad to see ’em.

      Also want to call out Martha (Smith) for once again taking home the award for greatest contribution per character typed. Actually it may be time to retire the trophy. Nice to be able to skim through and say “ah, better read this one, Martha says it’s good.”

  13. Just to note that San Marino is in northern Italy. It’s not particularly young or old. It isn’t wealthy; it’s similar to Italy. I believe they have a hospital. It’s not much of an area; it’s near Rimini and relies on people coming through who are going to and from the beaches. Was the thought Lichtenstein? Or Luxembourg?

    My guess about nursing homes is the people both tend to be an illness away from severe consequences, and that there may be an infection effect caused by what appears to be repeated exposure from somewhat differing vectors. That would be bad news, but it fits with the general idea that exposure may only confer partial immunity.

    In case, I’m not clear: we know that a virus, any virus, constantly generates mutations. This is true within yourself. We also know from the so-called common cold that ‘common’ means it occurs frequently, and that it occurs frequently because it varies so much in its expression that it gets passed around and around and around. The really bad news would be if Covid-19 has ‘common’ characteristics, meaning it goes into people and comes out of people with sufficient differences that you can pick it up again. Many if not most people have experienced this with some minor virus like the cold where you seem to get sick over and over: you are being exposed to expressions of the virus that differ enough that you get sick in some form again. In some cases, you can get much sicker, partly perhaps because your system is worn down. In other cases, it can linger and linger because your immune system can’t shut it down, perhaps again because it’s been taxed but also perhaps because it hangs on so well.

    My guess is that Covid-19 expresses through people with such variation that it has some characteristics of being ‘common’. That would help explain nursing home deaths, deaths among providers, deaths among workers crammed together: repeated slightly different vectors of infection. It suggests, for example, that a person exposed at family home not only gets a smaller viral load on infection – which HIV research showed is very important – but that the virus expresses itself in ways that the ‘family’ immune system is generally better able to handle because it is genetically similar and probably experientially similar.

    I think it expresses with such variation because we see such a wide range of symptoms and consequences. And because I see cases and clusters outside of nursing homes which have a commonality involves interacting with a number of people. Exposure to a bunch of asymptomatically infected people may be very different from being exposed to the same asymptomatic person. I had hoped the issue was more viral load on infection.

    I’m sorry if this is a downer, but I don’t see much in the way of light. If I’m right, then building an effective vaccine will be more difficult. Worst case would be if exposure, meaning some level of infection, even entirely asymptomatic, could make you more susceptible to other exposures. I’m not talking about grand scale mutation into different strains: just regular mutation that occurs within each person who gets the virus. If it has ‘common’ characteristic, then I mean relatively subtle variations through which the virus ‘learns’ how to remain actively infectious.

    This would be perfect virus design: a virus that mathematically can remain alive because it not only kills at a relatively low rate and is easily spread but which only confers partial immunity that is directly rooted in how it is acquired. We’ve all seen the difference between someone with a whopping great cold and a minor one: if this is true with Covid-19, then it will be very hard to cage. In fact, it might become endemic until and unless we are able to substantially advance our ability to attack viruses.

    Not sure how to say this: at the social level, this virus is the product of our Information Age. It doesn’t kill that many people compared to other great epidemics but the news of it, and the ability to ‘do something about it’ has changed dramatically. In 1918, there was not only general ignorance if you weren’t directly infected (or your area wasn’t experiencing it), but what could be done? There was little to nothing in the way of social relief programs. Now we have great big public health measures taken everywhere. That may look terrific, and I’m not arguing that they aren’t, but they’re also not economic in the sense that the past went through things and continued. That is, in the relatively recent past – even just before I was born with polio – you suffered with these things and hoped you and your family weren’t affected. Now, we spread the pain and engage in what appears every day to be magical thinking about how an economy will work when it patently won’t.

    It isn’t a virus that brings down Western civilization but that we are able better to respond to it so we’ve lost the cruel resiliency of a more primitive world. It’s very strange to type that. But I no longer think worst case scenarios are all that worst case. As in, most businesses can’t survive based on 50% or 70% of their prior revenue. That means massive unemployment, tremendous pressure on commercial retail and office rents, and thus tremendous pressure on commercial lending, and thus development. All office buildings are constructed based on projected rents that in turn assume a certain density of employees (because that directly connects to revenue and space needs). People casually toss off that this means the end of open plan offices as though that has no economic consequences. The latest developments have all tried to incorporate greater sociability: food halls, beer gardens, ‘lifestyle’ shopping centers all orient around shared experiences. The net worth of most people is in their house. That could vanish. The states are losing billions in revenue: they need to make that up by raising taxes and/or cutting spending. If Covid-19 becomes endemic, will people value cities the same?

    Maybe the ‘levellers’ are right and this spells the end of an unfair system. Problem is that history suggests bad times means the people with power tend to grab what they can for themselves and the rest get poorer. It’s like the virus is pushing us toward a better world but without a reason to get there beyond the traditional platitudes about love.

    • Although I have been to Rimini, I did not make a visit to nearby San Marino and know nothing about it other than what it says on Wikipedia, which includes: “It is one of the wealthiest countries in the world in terms of GDP per capita, with a figure comparable to the most developed European regions. San Marino is considered to have a highly stable economy, with one of the lowest unemployment rates in Europe, no national debt and a budget surplus.” I agree there is some tension between “one of the wealthiest countries in the world” and “comparable to the most developed European regions.” But then, the most developed European regions are, objectively, among the wealthiest areas in the world, so I guess the issue here may be a lack of specificity. In GDP per capita, San Marino is 13th out of 192 countries (Wikipedia again), but that puts it below the U.S. and, well, 11 other countries.

    • Jonathan said,
      “If I’m right, then building an effective vaccine [for Covid-19]will be more difficult,”
      and
      “The really bad news would be if Covid-19 has ‘common’ characteristics, meaning it goes into people and comes out of people with sufficient differences that you can pick it up again. ”

      Comparing with influenza instead of the cold: The flu varies so much from year to year that a new flu vaccine is developed for each season. From what I have heard about the development of new flu vaccines, it involves a lot of guesswork, and something like a “committee consensus”. I would expect that a Covid-19 vaccines would also need to be constructed yearly (to adapt to new versions of the disease), and probably with just as much uncertainty as now happens with flu vaccines.

      • Martha, each year the Flu undergoes a complex zoonotic lifecycle. It enters birds, the birds migrate, their droppings wind up in pig farms, the pigs are infected, the flu mutates within the pigs, pig farmers are infected, and then there’s a human outbreak.

        On the other hand, this coronavirus is endemic to bats, and we don’t have massive bat farms worldwide. So there isn’t necessarily a reason to think that there will be a massive continuous annual cycle of coronavirus outbreaks.

        maybe, but not necessarily.

  14. I’ve been mulling a new metric to compare how we’re doing:

    Let’s chart the deviation/volatility between IFR and Case Fatality Rate (CFR)?

    In countries that are ahead in terms of public health responses (SK, Taiwan, Germany(?), New Zealand, Australia(?),…) CFR and IFR should be close to converging.

    I do question why we’ve all been fixated on the IFR.

    IFR as THE metric in the early stages of a novel virus pandemic in countries with wildly different testing systems tells us something, but not as much as also using the CFR.

    The story the global Case Fatality Rate figures tells us very interesting too.

    Using European Union CDC data, I did the calculations and found global daily average 1st April to 10th May is 21%: My data: https://medium.com/geopolitical-impact-of-early-warning-covid-19/two-tribes-the-case-fatality-rateites-the-ifr-ites-7f0a1b4b8bc2

    UK’s CFR last Wednesday was 29%!

    I notice a quite emotional response from the “It’s Just Like Flu” brigade time the suggestion that for all its faults (many of which it shares with IFR) CFR figures tell an important story.

    At the risk of comparing Apples with Oranges, IFR are (in the absence of a SKorea-like testing system) highly model based.

    CFR, by contrast, measures actual confirmed cases and actual deaths, as confirmed by nation state reporting systems.

    Both measures contain complexity and uncertainty to the same degree.

    Yet, CFR tell a fascinating and important story, especially in countries which aren’t dealing very well with things.

    In countries like HK, NZ, Aus, Taiwan, China etc that are testing massively, the CFR and the IFR converge.

    This is why I propose a new metric that measures deviation between CFR and IFR, for use in the early stages of a pandemic.

    Hypothesis: high and consistent deviations between the two measures suggests governance systems struggling to deal with a novel virus pandemic. Therefore, this is an early warning system indicator that countries/regions need help and/or ought to change tack.

Leave a Reply to Joshua Cancel reply

Your email address will not be published. Required fields are marked *