More than one, always more than one to address the real uncertainty.

The OHDSI study-a-thon group has a pre-print An international characterisation of patients hospitalised with COVID-19 and a comparison with those previously hospitalised with influenza.

What is encouraging with this one over yesterday’s study, is multiple data sources and almost too many co-authors to count (take that Nature’s editors).

So an opportunity to see the variation and some assurance that many eyes had an opportunity to see and question the protocol and the study work.

Results: 6,806 (US: 1,634, South Korea: 5,172) individuals hospitalised with COVID-19 were included. Patients in the US were majority male (VA OMOP: 94%, STARR-OMOP: 57%, CUIMC: 52%), but were majority female in HIRA (56%). Age profiles varied across data sources. Prevalence of asthma ranged from 7% to 14%, diabetes from 18% to 43%, and hypertensive disorder from 22% to 70% across data sources, while between 9% and 39% were taking drugs acting on the renin-angiotensin system in the 30 days prior to their hospitalisation. Compared to 52,422 individuals hospitalised with influenza, patients admitted with COVID-19 were more likely male, younger, and, in the US, had fewer comorbidities and lower medication use.

Now, it may be important to note that none of the authors had direct access to the very confidential patient data. They write analysis scripts which the data holders run (separately) and return data quality diagnostics and the summaries reported in the paper. Some ability to query the study here as well a access protocol and code.

Now this a live entity, the scripts can be run by any data holder at least after the data has been transformed into a standard format. Hopefully that can be done for the enterprise electronic health record (Sunrise Clinical Manager; Allscripts) reporting database from yesterday’s study.

If not, why not?

p.s. CS Peirce quote – “one man’s [group’s] experience is nothing if it stands alone. If he sees what others cannot see, we call it hallucination” (CP5.402n)

This post is by Keith O’Rourke and as with all posts and comments on this blog, is just a deliberation on dealing with uncertainties in scientific inquiry and should not to be attributed to any entity other than the author.

 

39 thoughts on “More than one, always more than one to address the real uncertainty.

  1. “If not, why not?”

    Exactly! This stuff should all be planned out in advance, and if it is not, well the effort fell short of producing a publishable paper, didn’t it?

  2. Thanks Kieth. I was especially interesting Asthma, COPD, and smoking (not present). The Columbia University data is the only one that looks similar to what was seen elsewhere. I wonder if using hospitalization as a proxy for severe illness may have issues because the criteria for that differ so much depending on the situation. Ie, in NYC you really were only hospitalized for severe illness but not elsewhere.

    Also, I can’t wait for the WHO’s response to the panic buying of nicotine products around the world that is just beginning: https://www.bbc.com/news/world-europe-52415793

    • Also, I was listening to the emcrit podcast and the critical care doctors were mentioning how strange it was they were not seeing anyone with asthma.

      • The CDC’s weekly FluView report shows data that may indicate excess hospital admissions due to frightened people, stating:

        “The COVID-19 pandemic is affecting healthcare seeking behavior. The number of persons and their reasons for seeking care in the outpatient and ED settings is changing. These changes impact data from ILINet in ways that are difficult to differentiate from changes in illness levels…” . . https://www.cdc.gov/flu/weekly/

        I wonder whether the Covid symptoms are similar to clinical symptoms of panic or mass hysteria, particularly the body’s allocation of oxygen under panic stress. Perhaps smokers appear to have some Covid protection (indicated by lower hospital admission rate) because they are calmer? The lack of asthma, as you have noted from the Emcrit podcast, may be a clue also.

    • > because the criteria for that differ so much depending on the situation.
      Definitely, for instance I have two relatives, one in Toronto and one in Ottawa, both very likely had Covid19. One was seen by there GP, the other by no one. One took six weeks to recover, the other still not fully recovered after four weeks. But nothing to identify them as cases.

    • Anoneuoid, In case you haven’t seen it, another data source for you with low smoking rates among cases. As usual, the sample of patients could have high selection bias though, and that could explain the entire smoking-covid relationship seen.

      https://covidibd.org/current-data/

      It seems this shot in the dark was worth further study, and the nicotine patch trial will help resolve the question.

      • Thanks, I hadn’t seen that. About 5% were smokers and only 1/36 of them got put on a ventilator but 4 died. I’d say this is entirely consistent with what we’ve been seeing. But I guess we need to compare to smoking rates in IBD patients. But now I am seeing some reports that IBD is helped by smoking as well, so maybe they smoke more often than the general population?

        I’m not too optimistic about nicotine patches/gum.

        I think there are *at least* two things going on, probably more given how huge this effect is. I don’t think I’ve ever seen anything in medicine that is so repeatable, it is almost unheard of. And this has to be in the face of intense bias for the data to show the opposite.

        1) There is an acute effect of smoking for high altitude sickness. It is said to be helped by smoking a single cigarette in both mountaineers and cyclists.

        2) But high altitude is also used as a treatment for asthma, which indicates something chronic (I even wonder if asthmatics are less likely to have an asthma attack when they have this illness). Something like airway modelling, or adaptation to low blood oxygenation that we would require chronic smoking to see. Another thing is former smokers also seem to be rare (but sometimes the data just says “smoking” and groups them together).

        Maybe the acute effect is mediated by nicotine, but applying it via a patch rather than topically to the airway seems like it wouldn’t work as well (also what are the relative doses?). Vaping would be a better model if you want to isolate nicotine.

        • I just realized the above assumes you are familiar with the similarity of covid-19 to high altitude sickness. I’m not sure how much that has trickled into the mainstream consciousness yet. It was first noted in the literature here:
          https://www.ncbi.nlm.nih.gov/pubmed/32226695

          Then this NYC critical care doctor brought it to the public attention: https://www.youtube.com/watch?v=k9GYTc53r2o&feature=emb_logo

          And now hyperbaric oxygen therapy is reported to be extremely and immediately effective, resolving a diverse set of symptoms at once that are otherwise being attributed to the virus infecting this or that tissue:

          https://www.msn.com/en-us/health/medical/louisiana-hospital-using-hyperbaric-oxygen-therapy-to-treat-extreme-cases-of-covid-19/ar-BB133OVV

          https://www.ihausa.org/covid19-hyperbaric-therapy/

        • And I see you won’t find any of this life saving information from the WHO. They are still recommending a protocol of early intubation and chasing spO2 that causes unnecessary ventilator induced injury:

          Tips for managing respiratory distress
          • Keep SpO 2 > 92–95%.
          • Do not delay intubation for worsening respiratory distress. Be prepared for difficult airway!

          https://www.who.int/publications-detail/clinical-care-of-severe-acute-respiratory-infections-tool-kit

        • It is helpful to know that COVID should be treated as acute hypoxia. But how do you scale that up? Normally, acute hypoxia is rare and an odd diver or climber here and there is not an issue.

          WHO probably know there are even fewer hyperbaric chambers than ventilators on this planet, so they don’t see it as anything viable.

          We don’t even have enough wooden sticks with a cotton ball on top to implement wide-reaching testing, let alone something more fancy.

        • Look at the protocol. Each patient only needs to be in there for an hour or so each day for five days, and some are big enough to hold 10 patients. So one of the larger hyperbaric chambers can treat hundreds of patients each day and requires less specialized skill than ventilators. They can also pressurize airplanes and use those.

          So I don’t know all the details but this seems much more scalable than putting everyone on ventilators (which was is a bad idea for these patients anyway).

        • Does it even require hyperbaric? It certainly seems reasonable to consider creating 100% oxygen tents. Although it might not be as effective, it might give relief to those who are not quite as severe. anyway, just thoughts.

          You could pressurize an airplane to around 1.4 atmospheres, on the basis of calculations here: https://www.engineeringtoolbox.com/air-altitude-pressure-d_462.html (cruising altitude is about .4 Bar, and they’re pressurized to internal pressure around .8 Bar when cruising, so pressure differential of 0.4, so at ground level you could probably get them up to 1.4 without bursting anything.

          so yes, you could clear out all the contents of airplanes (fabrics etc, anything flammable), and pressurize them with pure oxygen to 1.4 atm for a couple hours at a time, maybe, if flammability isn’t an issue, and if you can hook up pure oxygen to the pressure system. I don’t know how they’re pressurized, so I don’t know if it’s really feasible.

        • Does it even require hyperbaric? It certainly seems reasonable to consider creating 100% oxygen tents. Although it might not be as effective, it might give relief to those who are not quite as severe. anyway, just thoughts.
          […]
          so yes, you could clear out all the contents of airplanes (fabrics etc, anything flammable), and pressurize them with pure oxygen to 1.4 atm for a couple hours at a time, maybe, if flammability isn’t an issue, and if you can hook up pure oxygen to the pressure system. I don’t know how they’re pressurized, so I don’t know if it’s really feasible.

          I don’t think it works that way, but am far from an expert. Eg,

          In the larger multiplace chambers, patients inside the chamber breathe from either “oxygen hoods” – flexible, transparent soft plastic hoods with a seal around the neck similar to a space suit helmet – or tightly fitting oxygen masks, which supply pure oxygen and may be designed to directly exhaust the exhaled gas from the chamber. During treatment patients breathe 100% oxygen most of the time to maximise the effectiveness of their treatment, but have periodic “air breaks” during which they breathe chamber air (21% oxygen) to reduce the risk of oxygen toxicity. The exhaled treatment gas must be removed from the chamber to prevent the buildup of oxygen, which could present a fire risk. Attendants may also breathe oxygen some of the time to reduce their risk of decompression sickness when they leave the chamber. The pressure inside the chamber is increased by opening valves allowing high-pressure air to enter from storage cylinders, which are filled by an air compressor. Chamber air oxygen content is kept between 19% and 23% to control fire risk (US Navy maximum 25%).[76] If the chamber does not have a scrubber system to remove carbon dioxide from the chamber gas, the chamber must be isobarically ventilated to keep the CO2 within acceptable limits.[76]

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

        • Thanks. Makes sense that they are using high pressure, but the oxygen is still localized to the breathing space. I was imagining a truly scary fire risk. Yes you could definitely pressurize airplanes with air, and then have people breathing from a pure oxygen hood or mask while under the pressure. the point about Decompression sickness risk was useful as well. This all would require careful protocols. 1.4 bars is not a particularly “deep” dive though. I think it’s like 13 feet of seawater. About the level that the shallower decompression stops occur. I don’t know what pressure you could put an airplane to at max. I wouldn’t guess more than 2 bar.

        • If you want more information, try pulmcrit: https://emcrit.org/category/pulmcrit/

          It isn’t true altitude sickness. The description is that blood ends up moving from the right to left ventricle without sufficient oxygenization. At altitude, that’s because there is less available. Down here, it’s for a couple of reasons, and the speculation is that there are alveolar patches which are not obstructed, while other patches are.

          Now to get personal: my wife had this. Very low oxygen, but able to breathe. In other words, what is now a fairly typical. And I believe but I’m not sure this tends to go with an absence of fever because there is no white cell crash. If these two things happen together, my guess is the risk of death is much higher. The basic fear with low oxygen is the buildup of fluid. That kills at altitude. The basic treatment is relatively frequent but low energy movement. We followed that protocol and my wife recovered.

        • The basic treatment is relatively frequent but low energy movement. We followed that protocol and my wife recovered.

          So, playing darts?

        • Anoneuoid, Given the direction of your comments, I want to add points about how very bad smoking is for health. Smoking is responsible for about 1 in 6 of all deaths world wide, as found by Global Burden of Disease studies. It is a scourge on the world.

          I assume (?) you agree about how terrible smoking is, but I think it is worth repeating often because of the many deaths smoking causes, and because tobacco companies have used so much money to stealthily fund research about “silver linings” of smoking, and about how “more evidence is needed” before smoking can be judged responsible for bad outcome X. This is PR for their scourge on the world.

          The points you make about similarity to high altitude sickness are interesting, but I haven’t had time to check to see if the evidence looks reliable or unreliable to me.

          Thanks.

        • If global burden of disease studies is some aspect of WHO (which it looks like after a quick search) I would be inclined to believe the opposite of whatever they say.

          That said, smoking is obviously bad for most people. It was for me and I stopped, you don’t need the WHO to tell you that.

        • “In the present study we analyze the epidemiologic data of COVID-19 of Tibet and high-altitude regions of Bolivia and Ecuador, and compare to lowland data, to test the hypothesis that high-altitude inhabitants (+2500 m above sea-level) are less susceptible to develop severe adverse effect in acute SARS-CoV-2 virus infection. ”
          Geographical selection. These regions are remote, are expected to have been later than average, and thus have lower infection prevalence.

          “A representative cohort of 67 patients (only two imported cases) who were diagnosed with COVID-19 in Sichuan reveals that 54% were completely asymptomatic (no cough, fever, or headache), and less than 10% of the patients presented fever. Nevertheless, 10% of the SARS-CoV-2 positive cohort developed severe medical condition, however, all of these patients fully recovered after treatment, resulting in no mortality. Moreover, 29% of all the patients were potential high-risk due to predisposition with chronic respiratory and/or cardiovascular disease at the time of COVID-19 diagnosis (Gelek 2020; World-Health-Organization 2020). Thus, it appears that both the pathogenesis of the SARS-CoV-2 virus and the general prevalence of infection in Tibet does not correspond to global trends.”

          That corresponds exactly to global trends: 10% hospitalization rate, 29% predisposition would depend on age group but doesn’t seem out of line, with <1% mortality no death in 67 is not significant.

          The Bolivia/Ecuador discussion seems to focus on number of cases and does not consider population density at all.

          There's correlation, but I don't see conving evidence for causation.

        • FWIW I had a similar hypothesis regarding altitude in US a few weeks back that I examined privately (stay in your lane and all that, which I’m all for generally). I became curious due to the data coming out of mountain west, in particular (having lived there). I scraped nyt state data and ran some simple models in brms comparing reported cfr (deaths/cases) to mean state elevation (Wikipedia). Generally there was a meaningfully sizeable but not highly certain slope there. The model was:

          (deaths_now) | trials( cases lagged ) ~ 1 + standardized( mean_elev_state ), family = binomial(logit)

          Results e.g. here for 14 day lag for apr 18 deaths / apr 5 cases.

          Intercept est: -2.47 (0.07)
          u_elev_std: -0.52 (0.14)
          sd( Intercept ): 0.48 (0.06)

          I used various lags (and no lag) and different days for deaths for “cfr” and generally got similar estimates. Also used kfold for comparisons to the model without the effect and that generally supported the effect, though not super consistent.

          There’s a good bit of variability in cfr on the upper end of state mean elev gradient and small n of course. I speculated then that addressing variability in elev within states might be more informative (ie going to fips level data). I didn’t have the time to look that deep with a model unfortunately. I did check some of the fips data more closely within some of those mountain states and the pattern was suggestive that in fips in these mountain states where most of population looked situated below 6000-7000 ft the cfr was more similar to the rest of the country. Again, I did not model any of this within state stuff, so grain of salt and all that. In fact I mostly took all of this as grain of salt given uncertainty, potential confounders (eg pop density, lifestyle), etc.

          Also, when looking into potential altitude related mechanisms, I found this cool paper. Not covid related directly but an interesting study related to metabolic adaptation to hypoxia at altitude.

          https://pubs.acs.org/doi/abs/10.1021/acs.jproteome.6b00733

          Way out of my lane at this point, but its a fun study regardless of any potential value to this covid topic ;)

        • Sorry, the model was actually

          (deaths_now) | trials( cases lagged ) ~ 1 + standardized( mean_elev_state ) + ( 1 | state ), family = binomial(logit)

          Forgot to include the random effect in original post

        • Anoneuoid, The inuitive (and possibly wrong) thought is that smokers are less-likely to get altitude sickness because they have acclimated their bodies to low oxygen receipt, as a consequence of… smoking. If this is the explanation, being a long- or medium-term smoker could also result in less risk of serious covid, but a non-smoker taking nicotine would not have their covid risk affected. Have you seen evidence that distinguishes between this possible explanation vs. the possible explanation that nicotine itself is biochemically responsible? I’m guessing if its there you’ve found it, and am curious.

          I don’t want to get off topic into discussion of the WHO, but one can go to any number of other sources to find evidence of the many health harms of smoking. iirc, the original Doll-Hill doctor’s study showed about 50% increased risk of death in smokers. Loosely extrapolating to the world total population with 20% of people smokers, that would mean about 1 in 11 deaths due to smoking. Whether its 1 in 6 or 1 in 11 or 1 in 4 or whatever, it’s a scourge.

        • Oops, I noticed your science direct link is actually related to the question I asked. However, the evidence level there is more hypothesis generating than useful. Anything else?

        • Thanks. It’s a chain of ideas building into a theory, which I like, but there is little evidence behind some chain links, which I dislike. More evidence would be nice, as I think you’d agree, and hopefully will be forthcoming.

  3. Just watched a clip of Birx on Fox. She seemed to suggest data showing a difference disease profile in US youth (or at least younger people) when compared to other countries.

    Has anyone seen anything about this?

  4. Without even looking at the author name of the post, I’ve come to recognize articles by Keith based on the presence of a CS Peirce quote or mention. Currently reading about him (Peirce) again and the use of statistics/experimental design in psychology in my second read of ‘Statistics on the Table’ by Stigler

Leave a Reply

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