Your observation about pleural fluid analysis for TB is interesting, because I didn’t read it thinking that the AI did your thinking for you. I saw that you looked something up and learned new information that you will use in the future, because we are also systems that can learn and adapt. This is no different than looking something up online or in a book - more efficient, sure, which is one of its strengths, but still a human act that expanded a human’s knowledge. In the future, you will know to order an ADA on that pleural fluid (and maybe get a pleural biopsy).
I am 51 years old - old enough to remember paper charts, but young enough to have come up with a variety of evolving digital tools. I am a bit of a Luddite when it comes to some tools in the ICU. (Arterial lines are not as useful as some people think; before you do your cleverly-acronymed POCUS study, have you considered taking a history and performing a physical exam?) But as I’ve used these systems a bit more, I see their utility and will work to find ways to improve care by making me better, not by replacing me.
The AI clinical notes are also interesting. I have found that the outpatient notes they generate are poorly-written, do not explain clinical reasoning well, and have a prose style not to my liking. However, they do take notes very well. I have taken to using DAX as my note taker in clinic, rather than my note writer. It doesn’t speed me up dramatically, but it does mean that I can focus my attention entirely on the patient in the office and not on my notepad or a computer screen. After clinic, I can use the notes it took to write a proper note.
Decreasing the time it takes me to find needed data serves my patients and serves my competence. Taking notes for me in the office means that I can focus my attention more directly on the patient. I will not delegate the core clinical decision-making and personal relationships that make critical care what it is. But if the machine can speed up my lit search, I can live with that.
Yes, this is the question … are LLMs just another tool, just an advanced search engine that we should accept and adopt just like other technologies we have become dependent on (and in some ways less skilled as a result) — video laryngoscopy, ultrasound etc.
For LLMs, the temptation to stop thinking or at least stop critically thinking is very high, speaking from experience.
OTOH, it is a fantastic ‘partner’, a curbside consult that is always available and probably correct almost as often (or maybe more often) as a real colleague.
True enough. I remember some analogous issues when you could first access UpToDate on pre-iPhone PDAs. Or when we wrote down the Iliad instead of memorizing it in the bardic tradition.
That had to be much bigger inflection point for civilization and the history of knowledge. People were newly able to go look things up in books instead of having to memorize everything? Must have been seen as potentially calamitous. Great reference.
I finished med school in 2010...my first 2 years of residency were still on paper...oh how I miss paper charting; but even then, I knew a good deal of the technology available and how to use it. I even remember the specific moment when my MA show me a function I overlooked on the computer...something I did with my attendings and others still to this day, so it hit me like a ton of bricks when that happened.
With the LLM's, I'm absolutely lost. I downloaded to my phone the ChatGPT app, but really didn't find a use for it enough to replace just using google search (though I admit, I do like - most of the time - the google's AI summary). After a month of not using it more than 2 or 3 times, I deleted the app.
For the last several years, I've been full time locums, and use multiple different EMRs. I've seen some with the disclaimer that AI was utilized in the development of the progress note (specifically in an office), but have yet to really see AI, knowingly, in any of the systems I've used that I can easily access within the systems (much less to be able to use and make things run smoother).
What I am curious about, what apps/websites/etc, are others using for AI or that integrate AI and how are they using it - like I do Google Search - or ?.
Like am precepting journal club on brinsupri and using for non cf bronchiectasis.
Asked how like much of immune suppression how can you harm host defense versus corralling inflammation?
Not much of such in the trial itself (aspen)
Why does inhibiting neutrophil elastase in brinsupri not cause more infections? Neutrophils are important!
Chat
Great question — this gets at the heart of why DPP-1 inhibitors like brensocatib (formerly brinsupri) are being pursued despite neutrophils being frontline defenders.
Why infections aren’t dramatically higher
1. Target is upstream, not downstream
Brensocatib doesn’t wipe out neutrophils. Instead, it blocks dipeptidyl peptidase-1 (DPP-1), an enzyme required to activate neutrophil serine proteases (neutrophil elastase, cathepsin G, proteinase 3) inside maturing neutrophils in the bone marrow. Neutrophils are still produced, migrate, phagocytose, and generate oxidative bursts normally — the antibacterial core functions remain intact.
2. Partial inhibition, not complete blockade
In trials, protease activity is suppressed but not eliminated. Enough residual elastase activity is preserved for host defense, while the excessive tissue-destructive protease release that drives bronchiectasis flares is reduced.
3. Redundancy in neutrophil defenses
Neutrophil elastase is one part of a large antimicrobial arsenal (ROS, MPO, antimicrobial peptides, NETs, phagocytosis). Blocking it doesn’t completely paralyze the cell’s ability to kill bacteria.
4. Clinical trial safety data
In the phase 3 ASPEN trial, rates of serious infection (e.g., pneumonia, bacteremia) were similar between brensocatib and placebo arms. Mild respiratory infections were common across both groups — but not increased with drug. That’s why the NEJM paper emphasized a “favorable safety/tolerability” profile despite theoretical concerns.
5. Chronic vs. acute setting
The bronchiectasis population studied already has high baseline infection/exacerbation risk. The drug’s net effect was to reduce exacerbations because less protease-mediated airway damage and inflammation meant fewer infection-triggered spirals.
⸻
✅ Key tension: You’re absolutely right that neutrophil elastase helps kill pathogens (especially Gram-negative bacteria). But the human body has multiple redundant antimicrobial mechanisms, and brensocatib’s inhibition is incomplete and selective. In effect, it blunts collateral tissue damage without completely disarming neutrophils.
Would you like me to also pull a direct table/figure from the NEJM trial on infection adverse events, so you can see how the infection rates stacked up numerically?
Um like wow. But feel my question and understand tension came from experience.
Excellent! As usual. Would add…it’s a tool like any other. A critical engaged thinker will use it better.
A caveat even the AI community mentions - age matters. If AI use is literally shaping our brains differently - those of us from the microfiche! or look up sources and quote and integrate argument era have a huge benefit. Was even joking with a Genz! resident listen we (am late 40s) didn’t have our dating pool in our pocket on a phone but we were forced to get critical thinking before AI.
In full transparency I would have likely relied on AI like a madman given how good it’s gotten if I was 26 and going through training. But I didn’t have that luxury.
I feel like LLMs are going to shape our brains differently as you say, and moreso the next generation. They are just so embedded and will become inextricably embedded in all information seeking. Our teachers have a profound influence on us. And for many people, LLMs will be their primary teacher—and for certain domains, there is nothing wrong with that—but it will be a big and unpredictable change I think.
Your observation about pleural fluid analysis for TB is interesting, because I didn’t read it thinking that the AI did your thinking for you. I saw that you looked something up and learned new information that you will use in the future, because we are also systems that can learn and adapt. This is no different than looking something up online or in a book - more efficient, sure, which is one of its strengths, but still a human act that expanded a human’s knowledge. In the future, you will know to order an ADA on that pleural fluid (and maybe get a pleural biopsy).
I am 51 years old - old enough to remember paper charts, but young enough to have come up with a variety of evolving digital tools. I am a bit of a Luddite when it comes to some tools in the ICU. (Arterial lines are not as useful as some people think; before you do your cleverly-acronymed POCUS study, have you considered taking a history and performing a physical exam?) But as I’ve used these systems a bit more, I see their utility and will work to find ways to improve care by making me better, not by replacing me.
The AI clinical notes are also interesting. I have found that the outpatient notes they generate are poorly-written, do not explain clinical reasoning well, and have a prose style not to my liking. However, they do take notes very well. I have taken to using DAX as my note taker in clinic, rather than my note writer. It doesn’t speed me up dramatically, but it does mean that I can focus my attention entirely on the patient in the office and not on my notepad or a computer screen. After clinic, I can use the notes it took to write a proper note.
Decreasing the time it takes me to find needed data serves my patients and serves my competence. Taking notes for me in the office means that I can focus my attention more directly on the patient. I will not delegate the core clinical decision-making and personal relationships that make critical care what it is. But if the machine can speed up my lit search, I can live with that.
Yes, this is the question … are LLMs just another tool, just an advanced search engine that we should accept and adopt just like other technologies we have become dependent on (and in some ways less skilled as a result) — video laryngoscopy, ultrasound etc.
For LLMs, the temptation to stop thinking or at least stop critically thinking is very high, speaking from experience.
OTOH, it is a fantastic ‘partner’, a curbside consult that is always available and probably correct almost as often (or maybe more often) as a real colleague.
True enough. I remember some analogous issues when you could first access UpToDate on pre-iPhone PDAs. Or when we wrote down the Iliad instead of memorizing it in the bardic tradition.
That had to be much bigger inflection point for civilization and the history of knowledge. People were newly able to go look things up in books instead of having to memorize everything? Must have been seen as potentially calamitous. Great reference.
I finished med school in 2010...my first 2 years of residency were still on paper...oh how I miss paper charting; but even then, I knew a good deal of the technology available and how to use it. I even remember the specific moment when my MA show me a function I overlooked on the computer...something I did with my attendings and others still to this day, so it hit me like a ton of bricks when that happened.
With the LLM's, I'm absolutely lost. I downloaded to my phone the ChatGPT app, but really didn't find a use for it enough to replace just using google search (though I admit, I do like - most of the time - the google's AI summary). After a month of not using it more than 2 or 3 times, I deleted the app.
For the last several years, I've been full time locums, and use multiple different EMRs. I've seen some with the disclaimer that AI was utilized in the development of the progress note (specifically in an office), but have yet to really see AI, knowingly, in any of the systems I've used that I can easily access within the systems (much less to be able to use and make things run smoother).
What I am curious about, what apps/websites/etc, are others using for AI or that integrate AI and how are they using it - like I do Google Search - or ?.
Thanks!
Like am precepting journal club on brinsupri and using for non cf bronchiectasis.
Asked how like much of immune suppression how can you harm host defense versus corralling inflammation?
Not much of such in the trial itself (aspen)
Why does inhibiting neutrophil elastase in brinsupri not cause more infections? Neutrophils are important!
Chat
Great question — this gets at the heart of why DPP-1 inhibitors like brensocatib (formerly brinsupri) are being pursued despite neutrophils being frontline defenders.
Why infections aren’t dramatically higher
1. Target is upstream, not downstream
Brensocatib doesn’t wipe out neutrophils. Instead, it blocks dipeptidyl peptidase-1 (DPP-1), an enzyme required to activate neutrophil serine proteases (neutrophil elastase, cathepsin G, proteinase 3) inside maturing neutrophils in the bone marrow. Neutrophils are still produced, migrate, phagocytose, and generate oxidative bursts normally — the antibacterial core functions remain intact.
2. Partial inhibition, not complete blockade
In trials, protease activity is suppressed but not eliminated. Enough residual elastase activity is preserved for host defense, while the excessive tissue-destructive protease release that drives bronchiectasis flares is reduced.
3. Redundancy in neutrophil defenses
Neutrophil elastase is one part of a large antimicrobial arsenal (ROS, MPO, antimicrobial peptides, NETs, phagocytosis). Blocking it doesn’t completely paralyze the cell’s ability to kill bacteria.
4. Clinical trial safety data
In the phase 3 ASPEN trial, rates of serious infection (e.g., pneumonia, bacteremia) were similar between brensocatib and placebo arms. Mild respiratory infections were common across both groups — but not increased with drug. That’s why the NEJM paper emphasized a “favorable safety/tolerability” profile despite theoretical concerns.
5. Chronic vs. acute setting
The bronchiectasis population studied already has high baseline infection/exacerbation risk. The drug’s net effect was to reduce exacerbations because less protease-mediated airway damage and inflammation meant fewer infection-triggered spirals.
⸻
✅ Key tension: You’re absolutely right that neutrophil elastase helps kill pathogens (especially Gram-negative bacteria). But the human body has multiple redundant antimicrobial mechanisms, and brensocatib’s inhibition is incomplete and selective. In effect, it blunts collateral tissue damage without completely disarming neutrophils.
Would you like me to also pull a direct table/figure from the NEJM trial on infection adverse events, so you can see how the infection rates stacked up numerically?
Um like wow. But feel my question and understand tension came from experience.
Excellent! As usual. Would add…it’s a tool like any other. A critical engaged thinker will use it better.
A caveat even the AI community mentions - age matters. If AI use is literally shaping our brains differently - those of us from the microfiche! or look up sources and quote and integrate argument era have a huge benefit. Was even joking with a Genz! resident listen we (am late 40s) didn’t have our dating pool in our pocket on a phone but we were forced to get critical thinking before AI.
In full transparency I would have likely relied on AI like a madman given how good it’s gotten if I was 26 and going through training. But I didn’t have that luxury.
I feel like LLMs are going to shape our brains differently as you say, and moreso the next generation. They are just so embedded and will become inextricably embedded in all information seeking. Our teachers have a profound influence on us. And for many people, LLMs will be their primary teacher—and for certain domains, there is nothing wrong with that—but it will be a big and unpredictable change I think.