AI in EM

[EctopicFetus]

---

Members do not see this ad.

It’s not what people want to hear but AI is coming and the proper program is pretty good. There are some AI scribe companies that are more geared to EM. Also, I recently started using open evidence a little bit.. it’s pretty darn good. The AI scribe program I use has built in EM specific calculators. NIHSS, PERC, Heart etc..

Open evidence answers clinical questions if you have them.. One other thing I like about ChatGPT is if there is a procedure I did but dont have a macro/procedure note it will generate a decent one.. Requires editing but is decent enough.

I dont think you can use a generic LLM or a generic AI scribe and get what we want out of it. I also think one of the issues of most LLMs is they are really designed for clinic notes.. not for what we want in EM.

 

[TooMuchResearch]

It’s not what people want to hear but AI is coming and the proper program is pretty good. There are some AI scribe companies that are more geared to EM. Also, I recently started using open evidence a little bit.. it’s pretty darn good. The AI scribe program I use has built in EM specific calculators. NIHSS, PERC, Heart etc..

Open evidence answers clinical questions if you have them.. One other thing I like about ChatGPT is if there is a procedure I did but dont have a macro/procedure note it will generate a decent one.. Requires editing but is decent enough.

I dont think you can use a generic LLM or a generic AI scribe and get what we want out of it. I also think one of the issues of most LLMs is they are really designed for clinic notes.. not for what we want in EM.

What proper program?
I would try to use it if I could figure it out.

 

[xaelia]

This is a good read re: continued advancements and the underlying "agent" architecture powering the next generation.

From diagnosis to treatment: Advancing AMIE for longitudinal disease management

research.google

The path forward will be a bit bumpy – but we're definitely on the transition curve where humans function increasingly to gather data points to feed AI and rely on the outputs to guide management.

 

[Major]

Anyone use the PalmER AI? It’s pretty steep at $199 month

 

[Restart13]

I’m a big AI scribe believer. I think AI will revolutionize medicine in the medium term once the tech is finessed and EMR’s use AI as physician extenders as one other put it. If I all i have to do is interview patients and do procedures while the AI does the rest of the stuff (writing a note, putting in orders, sending admit texts/messages, organizes transfers) the job could be a lot better.

 

[Tiger26]

We just got access to Abridge and I was impressed using it for the first time yesterday. It did a better job than expected of not including the meaningless babble of an encounter while also crafting a thoughtful and organized narrative.

It was more refreshing to have my note done by the time I got back to the computer and it was freeing to focus less on trying to remember key points, so I felt like I could spend more time connecting with the patient.

 

[gutonc]

We just got access to Abridge and I was impressed using it for the first time yesterday. It did a better job than expected of not including the meaningless babble of an encounter while also crafting a thoughtful and organized narrative.

It was more refreshing to have my note done by the time I got back to the computer and it was freeing to focus less on trying to remember key points, so I felt like I could spend more time connecting with the patient.

Right now, with current tech, I think this is where AI in clinical medicine is most helpful. I only use it for HPI. It's CDM and A/P was mediocre at best (at least Nuance Dax implemented in Epic).

I look forward to the day when it orders up my labs, imaging and follow up.

 

[EctopicFetus]

Right now, with current tech, I think this is where AI in clinical medicine is most helpful. I only use it for HPI. It's CDM and A/P was mediocre at best (at least Nuance Dax implemented in Epic).

I look forward to the day when it orders up my labs, imaging and follow up.

I wonder if it will be able to order tests or if it will just pend them like med students do.

I could see a world where it pends them and then on my phone i ok them. It would be helpful. Thinking of a code and I want to order an x ray. Or post reduction ordering of films etc. I tend to agree. I think it is best for the HPI and to some degree the physical exam.

I think low hanging fruit would be to maximize billing, and EKG interpretation. There are a few Apps out there to help interpret EKGs. I think they could act as a second check and pop out the billable info to include in the note so i don’t have to dictate it.

Obviously would want this integrated and not to use another app to do this work.

The possibilities are vast.. execution is potentially gonna be hard. Im not an IT person so i dont know.

 

[xaelia]

I wonder if it will be able to order tests or if it will just pend them like med students do.

I could see a world where it pends them and then on my phone i ok them. It would be helpful. Thinking of a code and I want to order an x ray. Or post reduction ordering of films etc. I tend to agree. I think it is best for the HPI and to some degree the physical exam.

I think low hanging fruit would be to maximize billing, and EKG interpretation. There are a few Apps out there to help interpret EKGs. I think they could act as a second check and pop out the billable info to include in the note so i don’t have to dictate it.

Obviously would want this integrated and not to use another app to do this work.

The possibilities are vast.. execution is potentially gonna be hard. Im not an IT person so i dont know.

It's coming sooner than later (and some is here already) –

Some of what you describe is a problem of integration, and Epic has decided to just forge ahead with their own ambient technology:

https://www.fiercehealthcare.com/ai-and-machine-learning/epic-adds-ambience-healthcares-ai-its-toolbox-program

That'll ease development of all those idealised augmentations and decision-support shortcuts you've envisioned.

I believe ambient AI is also integrated into the Oracle EHR, with the same goal.

 

[WhatJobDoIPick]

I’m a big AI scribe believer. I think AI will revolutionize medicine in the medium term once the tech is finessed and EMR’s use AI as physician extenders as one other put it. If I all i have to do is interview patients and do procedures while the AI does the rest of the stuff (writing a note, putting in orders, sending admit texts/messages, organizes transfers) the job could be a lot better.

I can handle the orders and documentation. Can it please deal with the patient and family? Thanks.

 

[RustedFox]

I can handle the orders and documentation. Can it please deal with the patient and family? Thanks.

You beat me to it.
I want a hologram version of me (just handsomer) to answer all their ridiculous questions and educate them on to how to act like a reasonable person while I actually get work done.

 

[EctopicFetus]

It's coming sooner than later (and some is here already) –

Some of what you describe is a problem of integration, and Epic has decided to just forge ahead with their own ambient technology:

https://www.fiercehealthcare.com/ai-and-machine-learning/epic-adds-ambience-healthcares-ai-its-toolbox-program

That'll ease development of all those idealised augmentations and decision-support shortcuts you've envisioned.

I believe ambient AI is also integrated into the Oracle EHR, with the same goal.

Yeah Oracle / Cerner is coming. My main worry is the data the hospital will capture that they just cant right now. I also envision they will have more data than they know what to do with but I also worry they will weaponize the data to get the outcomes they want.

 

[Restart13]

It's coming sooner than later (and some is here already) –

Some of what you describe is a problem of integration, and Epic has decided to just forge ahead with their own ambient technology:

https://www.fiercehealthcare.com/ai-and-machine-learning/epic-adds-ambience-healthcares-ai-its-toolbox-program

That'll ease development of all those idealised augmentations and decision-support shortcuts you've envisioned.

I believe ambient AI is also integrated into the Oracle EHR, with the same goal.

Would love to see this. Realistically probably still some years away and then each health system will have it's own bureaucratic process that will take forever to adopt it, but one day🤞

 

[thegenius]

I wonder if it will be able to order tests or if it will just pend them like med students do.

I could see a world where it pends them and then on my phone i ok them. It would be helpful. Thinking of a code and I want to order an x ray. Or post reduction ordering of films etc. I tend to agree. I think it is best for the HPI and to some degree the physical exam.

I think low hanging fruit would be to maximize billing, and EKG interpretation. There are a few Apps out there to help interpret EKGs. I think they could act as a second check and pop out the billable info to include in the note so i don’t have to dictate it.

Obviously would want this integrated and not to use another app to do this work.

The possibilities are vast.. execution is potentially gonna be hard. Im not an IT person so i dont know.

The world will come to a time when you are in a room....and you say "Computer, order a left shoulder xray, limited, indication: reduction:" And like Star Trek the computer will hear your voice, you won't have to put in a silly password, and it will just get done.

That world should come tomorrow. That's what we need.

"Computer: order the TheGenius Chest Pain orderset"

Why can't we have this now. This should be my new company idea.

 

[thegenius]

I’m a big AI scribe believer. I think AI will revolutionize medicine in the medium term once the tech is finessed and EMR’s use AI as physician extenders as one other put it. If I all i have to do is interview patients and do procedures while the AI does the rest of the stuff (writing a note, putting in orders, sending admit texts/messages, organizes transfers) the job could be a lot better.

The big thing to revolutionize medicine is stop spending 1/3 of all our health care money on those in the last 3 months of their life.

AI will make health care be more like Star Trek, which is what it should be now.

 

[Apollyon]

The big thing to revolutionize medicine is stop spending 1/3 of all our health care money on those in the last 3 months of their life.

AI will make health care be more like Star Trek, which is what it should be now.

I thought it was 1/2 of Medicare money in the last 6 months of life. I've been wrong before.

 

[sloh]

I thought it was 1/2 of Medicare money in the last 6 months of life. I've been wrong before.

25% of all Medicare spending goes to patients in their last year of life. More specifically, ~10–12% of total Medicare spending happens in the last 30 days of life, and roughly 25% in the last 90 days. If you extend it to the last 6 months of life, estimates are around 25–30% of Medicare spending.

Still a lot of wasted money either way.

 

[RustedFox]

The world will come to a time when you are in a room....and you say "Computer, order a left shoulder xray, limited, indication: reduction:" And like Star Trek the computer will hear your voice, you won't have to put in a silly password, and it will just get done.

That world should come tomorrow. That's what we need.

"Computer: order the TheGenius Chest Pain orderset"

Why can't we have this now. This should be my new company idea.

I'm on board only if instead of "computer", we can name it what we want.

Think of the possibilities.

 

[VA Hopeful Dr]

25% of all Medicare spending goes to patients in their last year of life. More specifically, ~10–12% of total Medicare spending happens in the last 30 days of life, and roughly 25% in the last 90 days. If you extend it to the last 6 months of life, estimates are around 25–30% of Medicare spending.

Still a lot of more wasted money either way.

Is it though? I think we overlook that it's not all 2 month ICU stays where family flogs meemaw to death with repeated codes, g-tubes, and dialysis.

When my grandmother died 20 years ago, she spent 3 days in the hospital. Comfort care only. The previous 15 years were twice a year doctor visits and 2 prescription meds per day (both generic). So easily 90% of her Medicare costs were in her last 90 days. Wouldn't call it wasteful.

Daily in home hospice care is expensive as well, way cheaper than the hospital certainly but still very expensive. Wouldn't call that a waste either.

 

[EctopicFetus]

The world will come to a time when you are in a room....and you say "Computer, order a left shoulder xray, limited, indication: reduction:" And like Star Trek the computer will hear your voice, you won't have to put in a silly password, and it will just get done.

That world should come tomorrow. That's what we need.

"Computer: order the TheGenius Chest Pain orderset"

Why can't we have this now. This should be my new company idea.

It exists.. hospitals wont allow it to happen (for now).

 

[EctopicFetus]

25% of all Medicare spending goes to patients in their last year of life. More specifically, ~10–12% of total Medicare spending happens in the last 30 days of life, and roughly 25% in the last 90 days. If you extend it to the last 6 months of life, estimates are around 25–30% of Medicare spending.

Still a lot of wasted money either way.

I think the other part is the first yea of life. @Apollyon Something like half of all spending is in the first and last 6 months of life.

 

[Apollyon]

Is it though? I think we overlook that it's not all 2 month ICU stays where family flogs meemaw to death with repeated codes, g-tubes, and dialysis.

When my grandmother died 20 years ago, she spent 3 days in the hospital. Comfort care only. The previous 15 years were twice a year doctor visits and 2 prescription meds per day (both generic). So easily 90% of her Medicare costs were in her last 90 days. Wouldn't call it wasteful.

Daily in home hospice care is expensive as well, way cheaper than the hospital certainly but still very expensive. Wouldn't call that a waste either.

"One got wasted/And the other's a waste"

- Offspring

 

[thegenius]

I thought it was 1/2 of Medicare money in the last 6 months of life. I've been wrong before.

It’s a lot of money in the last little bit of life. We are both right with that statement.

 

[thegenius]

I'm on board only if instead of "computer", we can name it what we want.

Think of the possibilities.

Thinking

My wake would would probably pay homeage to the MILTF from star trek, Dr. Crusher.

"Crusher: Order the cardiac heparin order set. Default settings."

Or maybe I would go with "Picard". But picard didn't know anything about medicine. So not as funny.

But just saying Picard and having the computer sound like Picard would be wonderful.

 

[xaelia]

Thinking

My wake would would probably pay homeage to the MILTF from star trek, Dr. Crusher.

"Crusher: Order the cardiac heparin order set. Default settings."

Or maybe I would go with "Picard". But picard didn't know anything about medicine. So not as funny.

But just saying Picard and having the computer sound like Picard would be wonderful.

I jailbroke our AI scribe pilot to give me a full differential diagnosis and treatment plan ... in the Star Trek universe:

 

[greatnt249]

I'm on board only if instead of "computer", we can name it what we want.

Think of the possibilities.

 

[Hamhock]

The "cerebral" aspect EM docs enjoy is over, within five years.

Find a 'need', like supervising midlevels.

There are a few procedures that require more than a trained monkey -- and a bit of human judgement is needed, occasionally.

Across USA, it will take 10-15 years...but it's inevitable

HH

 

[Mount Asclepius]

The "cerebral" aspect EM docs enjoy is over, within five years.

Find a 'need', like supervising midlevels.

There are a few procedures that require more than a trained monkey -- and a bit of human judgement is needed, occasionally.

Across USA, it will take 10-15 years...but it's inevitable

HH

I don’t fully agree. Either way, if that’s your opinion, the ICU won’t be spared.

 

[HemeOncHopeful19]

25% of all Medicare spending goes to patients in their last year of life. More specifically, ~10–12% of total Medicare spending happens in the last 30 days of life, and roughly 25% in the last 90 days. If you extend it to the last 6 months of life, estimates are around 25–30% of Medicare spending.

Still a lot of wasted money either way.

I’m not convinced this number would be very different in other countries though? Forgive me if I misunderstood but I am assuming this is a criticism of US Healthcare

Almost by definition the last year of someone’s life would be by far the most medically complex/expensive.

 

[xaelia]

I’m not convinced this number would be very different in other countries though?

Costs and dependency goes up for sure – but, yes, other countries are required to ration limited resources, and the rationing becomes more aggressive in tandem with increasing dependence and lower quality of life.

It is routine in my hospital to unilaterally impose DNR orders on elderly patients being admitted, as well as limiting access to ICU resources. There are also specific programs to keep patients in "hospital-level care" in the community with non-transport conditions so they can be palliated or receive maximal non-invasive medical therapy without consuming additional resources.

 

[Hamhock]

I don’t fully agree. Either way, if that’s your opinion, the ICU won’t be spared.

The ICU won't be spared, in my opinion. Indeed, I anticipate earlier consolidation in critical care medicine than emergency medicine.

In my hospital, there are currently three CCM attendings daily. This is unnecessary and wasteful. There needs to be one attending (perhaps remote) supervising residents or midlevels. I am expendable immediately.

I spend most of my clinical time in a MICU and less time in a SICU (both with trainees). Especially in the MICU, current LLMs (eg Gemini) are equivalent or better the majority of the time than my meat brain. I have not been asked a question, in recent memory, by a resident that couldn't be more accurately and more deeply answered by LLMs.

My current usefulness is mostly in just three areas; two of which will be assumed by AI soon (5-10 years):

1. Patter recognition requiring vision and sometimes hearing or palpation. Most clinical diagnosis is pattern recognition, obviously...but most of that is based upon quantifiable objective data already analyzed superiorly by LLMs and similar algorithm-based models. The remaining subset of pattern recognition to which I am referring here, requires 'senses' -- which is waiting for optimization of 'real-world AI' (think Tesla autonomous driving with only cameras and neural nets). Currently, I am better than AI, residents, and -- I claim with cock-sure attitude reflected here -- most critical care attendings at determining need for intubation, readiness for extubation, and indentification of rare but obvious diagnoses with "typical presentation" (eg aortic dissection extending to the carotids often mistaken for embolic ischemic stroke). Other examples are available with moments of thought. All of my "superiority" will be replaced by real-world AI within 10 years effectively; though Luddites may briefly delay implementation at the expense of life and prosperity. I am an analytical betting man, but I never bet against human stupidity.

2. Procedures. I am trained in emergency medicine and critical care medicine. Not general surgery or related subspecialties. Within my fields of expertise, it is easy to train midlevels and residents/fellows to be better than me with my 10+ years of experience. I speculate, based upon already publicly demonstrated humanoid robotics and brain-computer interface demonstrations, that all EM and CCM procedures will be performed better by machines in 10 years. This is actually the cutting edge, in my opinion. The blend of humanoid robotics and real-world AI is more complex than "AGI". The reason my assertion with which I started this post (that my usefulness, and that of all other EM and CCM attendings, will be eliminated within 5-10 years) has such a uncertain range (ie 5-10 years) is not my doubt in AI; but rather the assimilation of real-world AI and humanoid robotics.

3. Palliative care. Human touch. Love and ****.

Respectfully to @Mount Asclepius @Apollyon @xaelia @Zweihander @Steve_Zissou @Nutmeg @vector2 @nimbus @sylvanthus @CCM-MD,
HH

 

[Steve_Zissou]

The ICU won't be spared, in my opinion. Indeed, I anticipate earlier consolidation in critical care medicine than emergency medicine.

In my hospital, there are currently three CCM attendings daily. This is unnecessary and wasteful. There needs to be one attending (perhaps remote) supervising residents or midlevels. I am expendable immediately.

I spend most of my clinical time in a MICU and less time in a SICU (both with trainees). Especially in the MICU, current LLMs (eg Gemini) are equivalent or better the majority of the time than my meat brain. I have not been asked a question, in recent memory, by a resident that couldn't be more accurately and more deeply answered by LLMs.

My current usefulness is mostly in just three areas; two of which will be assumed by AI soon (5-10 years):

1. Patter recognition requiring vision and sometimes hearing or palpation. Most clinical diagnosis is pattern recognition, obviously...but most of that is based upon quantifiable objective data already analyzed superiorly by LLMs and similar algorithm-based models. The remaining subset of pattern recognition to which I am referring here, requires 'senses' -- which is waiting for optimization of 'real-world AI' (think Tesla autonomous driving with only cameras and neural nets). Currently, I am better than AI, residents, and -- I claim with cock-sure attitude reflected here -- most critical care attendings at determining need for intubation, readiness for extubation, and indentification of rare but obvious diagnoses with "typical presentation" (eg aortic dissection extending to the carotids often mistaken for embolic ischemic stroke). Other examples are available with moments of thought. All of my "superiority" will be replaced by real-world AI within 10 years effectively; though Luddites may briefly delay implementation at the expense of life and prosperity. I am an analytical betting man, but I never bet against human stupidity.

2. Procedures. I am trained in emergency medicine and critical care medicine. Not general surgery or related subspecialties. Within my fields of expertise, it is easy to train midlevels and residents/fellows to be better than me with my 10+ years of experience. I speculate, based upon already publicly demonstrated humanoid robotics and brain-computer interface demonstrations, that all EM and CCM procedures will be performed better by machines in 10 years. This is actually the cutting edge, in my opinion. The blend of humanoid robotics and real-world AI is more complex than "AGI". The reason my assertion with which I started this post (that my usefulness, and that of all other EM and CCM attendings, will be eliminated within 5-10 years) has such a uncertain range (ie 5-10 years) is not my doubt in AI; but rather the assimilation of real-world AI and humanoid robotics.

3. Palliative care. Human touch. Love and ****.

Respectfully to @Mount Asclepius @Apollyon @xaelia @Zweihander @Steve_Zissou @Nutmeg @vector2 @nimbus @sylvanthus @CCM-MD,
HH

Eh, Geoffrey Hinton said the same years ago, but turned out to be wrong. In radiology specifically, there are only a very small number of tools that are useful to any degree and the majority are either useless or actively detrimental (koios thyroid ultrasound AI was actively detrimental, and the neuro-triage tool they added crashed our system and was removed within an hour). Outside of utility, they’re finding it difficult to actually monetize these tools to justify expense, they’re primarily 1-dimensional so IT has to now deal with a bunch of individual programs all trying to work together on platforms that third parties have built to try and help them work together, they’re prone to going down (at least our main AI used for fractures is), they don’t work well when there is artifact or imperfect imaging, and there is drift that occurs (our AI for dictation summarization starts to say crazy **** after a while). Additionally, radiology is more complex than those outside of the field realize. It’s not simple pattern recognition, there is a significant amount of reasoning that needs to happen often utilizing inference, comparison to prior exams, and data found elsewhere in the notes or labs to create a good, coherent impression.

All that to say, I’ll be surprised if it has supplanted 10-20% of my work in the next decade. I think a decade from now people will be saying that radiology will be replace in another 5-10 years.

 

[joooul]

There are cheaper alternatives for the pattern recognition besides LLMs and may also help execution in very specific areas Pigeons (Columba livia) as Trainable Observers of Pathology and Radiology Breast Cancer Images - PMC

 

[EctopicFetus]

The ICU won't be spared, in my opinion. Indeed, I anticipate earlier consolidation in critical care medicine than emergency medicine.

In my hospital, there are currently three CCM attendings daily. This is unnecessary and wasteful. There needs to be one attending (perhaps remote) supervising residents or midlevels. I am expendable immediately.

I spend most of my clinical time in a MICU and less time in a SICU (both with trainees). Especially in the MICU, current LLMs (eg Gemini) are equivalent or better the majority of the time than my meat brain. I have not been asked a question, in recent memory, by a resident that couldn't be more accurately and more deeply answered by LLMs.

My current usefulness is mostly in just three areas; two of which will be assumed by AI soon (5-10 years):

1. Patter recognition requiring vision and sometimes hearing or palpation. Most clinical diagnosis is pattern recognition, obviously...but most of that is based upon quantifiable objective data already analyzed superiorly by LLMs and similar algorithm-based models. The remaining subset of pattern recognition to which I am referring here, requires 'senses' -- which is waiting for optimization of 'real-world AI' (think Tesla autonomous driving with only cameras and neural nets). Currently, I am better than AI, residents, and -- I claim with cock-sure attitude reflected here -- most critical care attendings at determining need for intubation, readiness for extubation, and indentification of rare but obvious diagnoses with "typical presentation" (eg aortic dissection extending to the carotids often mistaken for embolic ischemic stroke). Other examples are available with moments of thought. All of my "superiority" will be replaced by real-world AI within 10 years effectively; though Luddites may briefly delay implementation at the expense of life and prosperity. I am an analytical betting man, but I never bet against human stupidity.

2. Procedures. I am trained in emergency medicine and critical care medicine. Not general surgery or related subspecialties. Within my fields of expertise, it is easy to train midlevels and residents/fellows to be better than me with my 10+ years of experience. I speculate, based upon already publicly demonstrated humanoid robotics and brain-computer interface demonstrations, that all EM and CCM procedures will be performed better by machines in 10 years. This is actually the cutting edge, in my opinion. The blend of humanoid robotics and real-world AI is more complex than "AGI". The reason my assertion with which I started this post (that my usefulness, and that of all other EM and CCM attendings, will be eliminated within 5-10 years) has such a uncertain range (ie 5-10 years) is not my doubt in AI; but rather the assimilation of real-world AI and humanoid robotics.

3. Palliative care. Human touch. Love and ****.

Respectfully to @Mount Asclepius @Apollyon @xaelia @Zweihander @Steve_Zissou @Nutmeg @vector2 @nimbus @sylvanthus @CCM-MD,
HH

4. Med mal.

 

[Hamhock]

All that to say, I’ll be surprised if it has supplanted 10-20% of my work in the next decade. I think a decade from now people will be saying that radiology will be replace in another 5-10 years.

Although I think your estimate (10-20%) is woefully low, even that estimate supports my assertion: AI will cause a shocking consolidation in all of medicine -- radiology included.

If a radiologist has 20% of his work "supplanted", he can't claim that AI isn't coming for his job because AI isn't doing the 80% leftover. That's the wrong way to look at the "20% supplanted".

Instead, view that radiologist as 20% more efficient; which means that 20% less radiologists need to be hired. AI may not take your job but it will take one of your partner's jobs. Twenty percent less radiologists at least if AI can do 20% of a radiologist's work.

HH

 

[Steve_Zissou]

Although I think your estimate (10-20%) is woefully low, even that estimate supports my assertion: AI will cause a shocking consolidation in all of medicine -- radiology included.

If a radiologist has 20% of his work "supplanted", he can't claim that AI isn't coming for his job because AI isn't doing the 80% leftover. That's the wrong way to look at the "20% supplanted".

Instead, view that radiologist as 20% more efficient; which means that 20% less radiologists need to be hired. AI may not take your job but it will take one of your partner's jobs. Twenty percent less radiologists at least if AI can do 20% of a radiologist's work.

HH

Ahhh but therein lies a whole mess of unknowns. AI is also speeding up MRIs, new technology is decreasing the amount of radiation we need to get decent quality CTs, and there are a whole host of new avenues for radiology to expand with AI and new tech. I anticipate that, until the population starts declining (and maybe not even then) we will continue to get more availability and more usage of imaging that requires a real person to lay eyes on. Add in midlevels who can’t think clinically worth a damn so they order every imaging study under the sun, and you likely end up with even more demand.

I used to be somewhat pessimistic about AI and rads, then I started doing it and started learning about trends regarding imaging usage in a litigious environment, and now I don’t worry so much.

It’s like a recent article asked, what if AI is just a “normal” technology? What if it does have true, extremely difficult to solve limitations in both production and implementation? I think we’re seeing that now and I don’t anticipate it will so rapidly change the landscape of radiology.

Also I said “I’ll be surprised if it has supplanted 10-20% of my work” as in I expect it to be even less than that. I do think it’ll be helpful for improving workflows and possibly optimizing lists though, which will save more headaches for me and make me more efficient, but not so much that there won’t be enough work left to go around. See above regarding unknowns.

We could make a bet, maybe a 6 pack of beer if you happen to drink alcohol, that in 5 or 10 years we can revisit this topic and see what it has functionally done to radiology. I’ll even put it in my google calendar to remind me, if you agree. I’m open to terms and fairly confident.

 

[Hamhock]

Ahhh but therein lies a whole mess of unknowns. AI is also speeding up MRIs, new technology is decreasing the amount of radiation we need to get decent quality CTs, and there are a whole host of new avenues for radiology to expand with AI and new tech. I anticipate that, until the population starts declining (and maybe not even then) we will continue to get more availability and more usage of imaging that requires a real person to lay eyes on. Add in midlevels who can’t think clinically worth a damn so they order every imaging study under the sun, and you likely end up with even more demand.

I used to be somewhat pessimistic about AI and rads, then I started doing it and started learning about trends regarding imaging usage in a litigious environment, and now I don’t worry so much.

It’s like a recent article asked, what if AI is just a “normal” technology? What if it does have true, extremely difficult to solve limitations in both production and implementation? I think we’re seeing that now and I don’t anticipate it will so rapidly change the landscape of radiology.

Also I said “I’ll be surprised if it has supplanted 10-20% of my work” as in I expect it to be even less than that. I do think it’ll be helpful for improving workflows and possibly optimizing lists though, which will save more headaches for me and make me more efficient, but not so much that there won’t be enough work left to go around. See above regarding unknowns.

We could make a bet, maybe a 6 pack of beer if you happen to drink alcohol, that in 5 or 10 years we can revisit this topic and see what it has functionally done to radiology. I’ll even put it in my google calendar to remind me, if you agree. I’m open to terms and fairly confident.

I appreciate your thoughtful reply. I am swayed a bit by the potential for new or expanding imaging studies; but not enough to change my prediction.

Although I would be more confident betting on critical care (radiology is not my area of expertise), I am open to a bet regarding radiology. Let's make it simple: The number of attending radiologists in the US will decrease by 25% or more within 10 years.

Before you make this bet, I think it needs to be acknowledged that the decreased number of radiologists may not be attributable to AI -- but I would still win the bet! 😉 For example, there could be "a shortage of radiologists" (not enough residency spots; pay inadequate to attract new grads; etc)...but those factors have been around for decades and the number of radiologists and the demand has been increasing over the past 20 years, as best as I can tell (quick LLM search).

HH

 

[Steve_Zissou]

I appreciate your thoughtful reply. I am swayed a bit by the potential for new or expanding imaging studies; but not enough to change my prediction.

Although I would be more confident betting on critical care (radiology is not my area of expertise), I am open to a bet regarding radiology. Let's make it simple: The number of attending radiologists in the US will decrease by 25% or more within 10 years.

Before you make this bet, I think it needs to be acknowledged that the decreased number of radiologists may not be attributable to AI -- but I would still win the bet! 😉 For example, there could be "a shortage of radiologists" (not enough residency spots; pay inadequate to attract new grads; etc)...but those factors have been around for decades and the number of radiologists and the demand has been increasing over the past 20 years, as best as I can tell (quick LLM search).

HH

Hmmm are we talking absolute numbers? As in 25% fewer radiologists than there are in the US today? If so then I’ll take that bet.

I think all specialties that we’re not in tend to seem deceptively simple over time and we tend to have a blind spot for the things that we have done so often that they are rote, ignoring the complexities of things like decision making with a multitude of factors influencing that decision. I know I look at Derm and think “man they just look at skin and sometimes cut it out, what’s the big deal?” But I imagine it’s far more complex than that.

Edit: I enjoy engaging the AI debate when people do it in good faith, which I think you are. There are many people who seem to take perverse glee in the idea that I’m pursuing something that they believe will be obsolete in a few years. Often, I think they’re upset that I might actually work a chill job, not having to deal with patients, getting to go home every night to put my kids to bed, and still make more than them, so they try and level the field in their head. When I see that, I generally just don’t engage. In this case I am interested hearing your perspective on your own subspecialty. I have a feeling it won’t happen to you or your field either, but that’s just gut feeling more than anything real.

 

[Hamhock]

Hmmm are we talking absolute numbers? As in 25% fewer radiologists than there are in the US today? If so then I’ll take that bet.

I think all specialties that we’re not in tend to seem deceptively simple over time and we tend to have a blind spot for the things that we have done so often that they are rote, ignoring the complexities of things like decision making with a multitude of factors influencing that decision. I know I look at Derm and think “man they just look at skin and sometimes cut it out, what’s the big deal?” But I imagine it’s far more complex than that.

Edit: I enjoy engaging the AI debate when people do it in good faith, which I think you are. There are many people who seem to take perverse glee in the idea that I’m pursuing something that they believe will be obsolete in a few years. Often, I think they’re upset that I might actually work a chill job, not having to deal with patients, getting to go home every night to put my kids to bed, and still make more than them, so they try and level the field in their head. When I see that, I generally just don’t engage. In this case I am interested hearing your perspective on your own subspecialty. I have a feeling it won’t happen to you or your field either, but that’s just gut feeling more than anything real.

Again, your engaged tone is appreciated. I so wish this was more common on SDN (I would return here more often).

That said, I remain in disagreement and a bit antagonistic. For our bet, let's stick with my orginal proposition; regardless of my potential for blindness to the complexities of radiology or derm. Please make a proposal. I think there's something like a 6-pack or BTC on the line, right?

Will "the number of attending radiologists in the US will decrease by 25% or more within 10 years"?

Regarding my specialty...phew, I feel more comfortable here. There is no doubt that critical care docs are not necessary in numbers nationwide currently; before the implementation of AI. Once simple LLMs and data-spinners are fully active, our numbers should -- and I hope -- plummet 25-50%.

I firmly believe I am unnecessary as a present physician more than 90% of the minutes I am in-house.

HH

 

[Steve_Zissou]

Again, your engaged tone is appreciated. I so wish this was more common on SDN (I would return here more often).

That said, I remain in disagreement and a bit antagonistic. For our bet, let's stick with my orginal proposition; regardless of my potential for blindness to the complexities of radiology or derm. Please make a proposal. I think there's something like a 6-pack or BTC on the line, right?

Will "the number of attending radiologists in the US will decrease by 25% or more within 10 years"?

Regarding my specialty...phew, I feel more comfortable here. There is no doubt that critical care docs are not necessary in numbers nationwide currently; before the implementation of AI. Once simple LLMs and data-spinners are fully active, our numbers should -- and I hope -- plummet 25-50%.

I firmly believe I am unnecessary as a present physician more than 90% of the minutes I am in-house.

HH

I’m down for that bet. A 6 pack on “the number of radiologists in the US will decrease by 25% or more within 10 years.” I’ll mark my calendar. I wish there was a way to set reminders here like on Reddit so it would just ping me back to this post in 10 years.