10 ms·

One of the most interesting parts of this press release, following it on other boards, is the immediate "We already do that. Nothing new here." That kind of re

One of the most interesting parts of this press release, following it on other boards, is the immediate "We already do that. Nothing new here." That kind of reply would immediately lead me to think it was just a useless release, but then I see people posting comments along the lines of "No, we don't do that. In fact, we have a policy against doing that because of the danger of aerosolization" For my hacker/systems wonks, this a great example of group learning happening world wide. I imagine there are many ICUs that do this, and many that forbid it. In such an environment, releases like this aren't submarines or spam; they're recurring prompts to administrators elsewhere that they might be missing something important. I hate to sound cold, but damn this is an interesting example of how organizations learn. One commenter asked "I wonder how many of those hospitals who forbid intubation are doing it to patients with good insurance" For the record, as far as I know this is a horrible thing to suggest. But the overall point, that large organizations have incentives that are many times removed from the actual work being done, is a good one. This kind of conversation facilitation across borders is what the internet was supposed to be doing. I think this is the first time I've seen it working the way we had hoped. What's especially interesting to me is that many of the signals we look for in social forums, like "this is just a spam press release", "nothing new to see", or "there's some ulterior motivation here" voting up or down, etc., are actually counter-indicators and inhibitors of overall progress.

joshgel 6y ago
Even better would be randomized experimental data that this was better than an alternative... 'It works' can mean lots of different things...
- DanielBMarkham 6y ago
  It's an excellent point. Learning is not the same as science, and we need randomized trials to actually be sure of anything. Unfortunately, in the middle of a pandemic, that's not going to happen with enough speed to prevent a lot of people dying. The perfect can be the enemy of the good enough. We don't have to guess or learn about the mass of an electron; that's been well-established through great lab work. But not all decisions are like that. Many times the hardest decisions are those you make with incomplete and contradictory information ... and not making a decision is, in effect, making a decision. This is one of the things that's so interesting here. If it were simple yes/no, what's science and what's speculation, there wouldn't be much interesting learning or decision-making happening.
  - lonelappde 6y ago
    RCT is not all of science. RCT is just one (very powerful) tool for making information quantifiable and for making decisions under uncertainty.
    - raverbashing 6y ago
      This, exactly. The fixation on RCTs has kinda become "a meme" at this point. However it's lives on the line in some situations. Not everything can be double-blind. Not every experiment can be ethically done with a placebo (or done at all). Also the placebo effect is real but it is more pronounced in things like back pain or headaches, not so much in broken bones or aggressive viruses.
      - StavrosK 6y ago
        You don't need to have a placebo to do a controlled trial, you can just have a control arm of people you administer nothing to. How are you going to know if you have an improvement if you don't have something to compare your treatment to, though?
        fhars 6y ago
        You need the placebo for the double blind RCT, which is still kind of the gold standard. But pronating patients is hard to do in a double blind fashion, so it is less of an issue here.
        gus_massa 6y ago
        The double blind part is very difficult, but at least they can have a randomized control group [1]. It doesn't sound very difficult. Also, in most cases the control group get the "standard" treatment, because "do nothing" (taking literally the GP) is too bad. And you want to be sure that the new treatment is better than the "standard" treatment. [1] A real randomized control group, not an unrelated bunch of guys in another city of France.
        6y ago
- tomrod 6y ago
  Sure, but check out causal I ferwnce methods coming from Pearl, Gelman, Rubin, Imbens, etc. RCT is wonderful when we can get them and not always show stopper when not.
- BurningFrog 6y ago
  Randomized controlled trials are optimal for gathering science. They're not optimal for saving lives.
  - ineedasername 6y ago
    Well, in the long term they're probably better at saving lives than case studies such as this one. But until such time as they're available, collaboration, discussion, and cases studies are what we have to work with. Becoming paralyzed by uncertainty pending systematic studies will kill more people.
- newacct583 6y ago
  If that study takes a few months to finish, be reviewed and published, it's very clearly not "better". Science is important, but there are times like this where brute intuitive engineering is the only tool available.
  - ineedasername 6y ago
    >Science is important, but..." This article isn't not science. It encompasses a key pillar of medical knowledge and advancements: case studies.
  - randcraw 6y ago
    All engineering needs feedback to guide implementation, especially in complex systems. And it's essential in black box systems like biological organisms. This is why the words engineering and medicine intersect mostly in relatively uncomplicated areas like prosthetics or eyeglasses. In immune systems, engineering is used only in the earliest stages, like to reduce search spaces in development of synthetic antigens and antibodies for vaccines, never to guide clinical trials. There feedback drives all decisionmaking under the purview of statisticians only. Engineering, and especially intuition, plays no role al all.
- abakker 6y ago
  If the effect size is large, the benefits of randomization go down. A very effective treatment would be evidently very effective without the burden of statistical proof. The randomization is critical when you are looking to differentiate small effects from random noise. That said, large effect sizes in one hospital don’t control for confounding variables.
  - kgwgk 6y ago
    Randomization is not required to neutralize random noise, which is... random. It’s required to prevent systematic biases like giving the drug that you think might help to the patients with better prognosis (instead of wasting it on those that are very likely to die anyway).
- hutzlibu 6y ago
  "'It works' can mean lots of different things..." "The proning and the high-flow nasal cannulas combined have brought patient oxygen levels from around 40% to 80% and 90%, so it’s been fascinating and wonderful to see" People die, if their oxygen levels are getting too low. This method gets them up again. Sounds like "working" to me.
  - joshgel 6y ago
    We aren’t treating oxygen levels. Higher oxygen levels now that result in lower oxygen levels later might actually be a bad thing. I have no doubt that proning increases the readings we are getting on pulse ox monitors, sometimes pretty immediately. I’ve seen it, literally first hand, in a number of my patients. But I have no idea if it saves lives. Many of those same patients still got intubated and some died. Understanding impact on surrogate markers isn’t enough.
specialist 6y ago
"...interesting example of how organizations learn." Highest recommendation for Everett Rogers' Diffusion of Innovation, first published in 1963, 5th edition in 2003. https://en.wikipedia.org/wiki/Diffusion_of_innovations https://en.wikipedia.org/wiki/Diffusion_of_innovations It's seminal. Having read a lot of process and innovation books in 90s and aughts, I really wish I had known to just start with the original text. Knowledge sharing takes time. We know how to model it. This book (and others) helped me to calm down and try to hasten the process instead of wasting my energy railing and failing.
christkv 6y ago
This also probably explains the high death rates on ventilators now as only the near death patients are being put on the ventilator now.
- mistermann 6y ago
  Is this well known in media reports for a while? I stopped following closely quite a while ago so mostly just pick things up from forum, but this seems newish to me? Last I heard was that some doubts were arising here and there about the usage/safety for covid patients.
  - christkv 6y ago
    The on the ground reports from NYC in This week in virology has been talking about moving away from respirators for at least 3 weeks. Now the ability of reporters looking for sensational news to actually listen to the professionals or read studies is a completely different issue. I wish they could just ask some of them to write opinion pieces to lay out the facts as they exist.
    - _startingover 6y ago
      Do you have data? I’m wondering if you may be overestimating how quickly ideas are adopted en masse.
- cma 6y ago
  The headline that claimed 88% mortality was pretty misleading apparently: https://twitter.com/DrJohnScott/status/1253284899752009728 https://twitter.com/DrJohnScott/status/1253284899752009728

joshgel 6y ago

Even better would be randomized experimental data that this was better than an alternative... 'It works' can mean lots of different things...

DanielBMarkham 6y ago
It's an excellent point. Learning is not the same as science, and we need randomized trials to actually be sure of anything. Unfortunately, in the middle of a pandemic, that's not going to happen with enough speed to prevent a lot of people dying. The perfect can be the enemy of the good enough. We don't have to guess or learn about the mass of an electron; that's been well-established through great lab work. But not all decisions are like that. Many times the hardest decisions are those you make with incomplete and contradictory information ... and not making a decision is, in effect, making a decision. This is one of the things that's so interesting here. If it were simple yes/no, what's science and what's speculation, there wouldn't be much interesting learning or decision-making happening.
- lonelappde 6y ago
  RCT is not all of science. RCT is just one (very powerful) tool for making information quantifiable and for making decisions under uncertainty.
  - raverbashing 6y ago
    This, exactly. The fixation on RCTs has kinda become "a meme" at this point. However it's lives on the line in some situations. Not everything can be double-blind. Not every experiment can be ethically done with a placebo (or done at all). Also the placebo effect is real but it is more pronounced in things like back pain or headaches, not so much in broken bones or aggressive viruses.
    - StavrosK 6y ago
      You don't need to have a placebo to do a controlled trial, you can just have a control arm of people you administer nothing to. How are you going to know if you have an improvement if you don't have something to compare your treatment to, though?
      - fhars 6y ago
        You need the placebo for the double blind RCT, which is still kind of the gold standard. But pronating patients is hard to do in a double blind fashion, so it is less of an issue here.
        gus_massa 6y ago
        The double blind part is very difficult, but at least they can have a randomized control group [1]. It doesn't sound very difficult. Also, in most cases the control group get the "standard" treatment, because "do nothing" (taking literally the GP) is too bad. And you want to be sure that the new treatment is better than the "standard" treatment. [1] A real randomized control group, not an unrelated bunch of guys in another city of France.
        6y ago
tomrod 6y ago
Sure, but check out causal I ferwnce methods coming from Pearl, Gelman, Rubin, Imbens, etc. RCT is wonderful when we can get them and not always show stopper when not.
BurningFrog 6y ago
Randomized controlled trials are optimal for gathering science. They're not optimal for saving lives.
- ineedasername 6y ago
  Well, in the long term they're probably better at saving lives than case studies such as this one. But until such time as they're available, collaboration, discussion, and cases studies are what we have to work with. Becoming paralyzed by uncertainty pending systematic studies will kill more people.
newacct583 6y ago
If that study takes a few months to finish, be reviewed and published, it's very clearly not "better". Science is important, but there are times like this where brute intuitive engineering is the only tool available.
- ineedasername 6y ago
  >Science is important, but..." This article isn't not science. It encompasses a key pillar of medical knowledge and advancements: case studies.
- randcraw 6y ago
  All engineering needs feedback to guide implementation, especially in complex systems. And it's essential in black box systems like biological organisms. This is why the words engineering and medicine intersect mostly in relatively uncomplicated areas like prosthetics or eyeglasses. In immune systems, engineering is used only in the earliest stages, like to reduce search spaces in development of synthetic antigens and antibodies for vaccines, never to guide clinical trials. There feedback drives all decisionmaking under the purview of statisticians only. Engineering, and especially intuition, plays no role al all.
abakker 6y ago
If the effect size is large, the benefits of randomization go down. A very effective treatment would be evidently very effective without the burden of statistical proof. The randomization is critical when you are looking to differentiate small effects from random noise. That said, large effect sizes in one hospital don’t control for confounding variables.
- kgwgk 6y ago
  Randomization is not required to neutralize random noise, which is... random. It’s required to prevent systematic biases like giving the drug that you think might help to the patients with better prognosis (instead of wasting it on those that are very likely to die anyway).
hutzlibu 6y ago
"'It works' can mean lots of different things..." "The proning and the high-flow nasal cannulas combined have brought patient oxygen levels from around 40% to 80% and 90%, so it’s been fascinating and wonderful to see" People die, if their oxygen levels are getting too low. This method gets them up again. Sounds like "working" to me.
- joshgel 6y ago
  We aren’t treating oxygen levels. Higher oxygen levels now that result in lower oxygen levels later might actually be a bad thing. I have no doubt that proning increases the readings we are getting on pulse ox monitors, sometimes pretty immediately. I’ve seen it, literally first hand, in a number of my patients. But I have no idea if it saves lives. Many of those same patients still got intubated and some died. Understanding impact on surrogate markers isn’t enough.

specialist 6y ago

"...interesting example of how organizations learn." Highest recommendation for Everett Rogers' Diffusion of Innovation, first published in 1963, 5th edition in 2003. https://en.wikipedia.org/wiki/Diffusion_of_innovations https://en.wikipedia.org/wiki/Diffusion_of_innovations It's seminal. Having read a lot of process and innovation books in 90s and aughts, I really wish I had known to just start with the original text. Knowledge sharing takes time. We know how to model it. This book (and others) helped me to calm down and try to hasten the process instead of wasting my energy railing and failing.

christkv 6y ago

This also probably explains the high death rates on ventilators now as only the near death patients are being put on the ventilator now.

mistermann 6y ago
Is this well known in media reports for a while? I stopped following closely quite a while ago so mostly just pick things up from forum, but this seems newish to me? Last I heard was that some doubts were arising here and there about the usage/safety for covid patients.
- christkv 6y ago
  The on the ground reports from NYC in This week in virology has been talking about moving away from respirators for at least 3 weeks. Now the ability of reporters looking for sensational news to actually listen to the professionals or read studies is a completely different issue. I wish they could just ask some of them to write opinion pieces to lay out the facts as they exist.
  - _startingover 6y ago
    Do you have data? I’m wondering if you may be overestimating how quickly ideas are adopted en masse.
cma 6y ago
The headline that claimed 88% mortality was pretty misleading apparently: https://twitter.com/DrJohnScott/status/1253284899752009728 https://twitter.com/DrJohnScott/status/1253284899752009728