Subspecialties most impervious to AI and midlevel+AI creep?

[damien_chazelle_fan]

Hello everybody. Incoming M1 interested in ophtho. After today seeing OpenAI's demo of GPT-4o, I am very apprehensive about my future in medicine and how much AI is going to impact it. ChatGPT has gone from a very impressive text-based model to Jarvis from Iron Man in literally 2.5 years. AGI is right around the corner.
I don't know how much replacing AI will be doing of clinicians (I don't think that it will replace all of them, but it might indirectly replace some (ie radiologists become 2x or 3x more efficient, so hospital doesn't need as many)), but my main worry is that it is going to bolster midlevels. The intuitions of an OD + the knowledge base of GPT5 or GPT6 might be sufficient to rival the care of a well-seasoned comp ophthalmologist. Because of this, my specialty and subspecialty decision is going to be HEAVILY influenced by how essential it is to have a competent, human being performing the task (which I think involve delicate, procedural specialties).
Of the ophthalmic subspecialties, which do you all think is the least likely to be impacted by AI, AI + OD's, or AI robots (ie, no way that a non-human entity will be able to perform a certain procedure)? My initial thought is retina, both medical and surgical? Maybe oculoplastics as well? Thanks

TLDR: Title

---

Members don't see this ad.

 

[RetinaDude]

Every time a new piece of technology comes out (ie, OCT), we always thought “oh no, the ODs will refer less because they’ll have this new technology to help them understand what’s going on at the macula”. Instead, it increased our referrals because a lot of them purchased the new technology but then didn’t know what to do with it once they had it. I’m really not concerned about AI taking over for ophthalmology any time in the near future. Even if AI reads the OCTs, or FAs, humans still have to give the injections, remove the FBs, repair the RD, etc…

 

[eyeeye_captain]

ODs would take the biggest hit. A reliable autorefractor and AI doing the screening exam could take away a lot of their routine visits. Retina and plastics will be fine due to the high number of procedures.

 

[damien_chazelle_fan]

Every time a new piece of technology comes out (ie, OCT), we always thought “oh no, the ODs will refer less because they’ll have this new technology to help them understand what’s going on at the macula”. Instead, it increased our referrals because a lot of them purchased the new technology but then didn’t know what to do with it once they had it. I’m really not concerned about AI taking over for ophthalmology any time in the near future. Even if AI reads the OCTs, or FAs, humans still have to give the injections, remove the FBs, repair the RD, etc…

I selfishly want you to be right, however this technology is going to impact ODs in a much different way. As an anecdote, I know very little about statistics and even less about coding, yet with my intuitions and ChatGPT's knowledge base, I was able to successfully generate code in R to run multivariate regression analyses and to generate a professional looking graph (keep in mind, I know a whopping two R commands off the top of my head). In that same sense, it stands to make ODs much "smarter" and better at ophtho than they are. I agree in that I don't think it will ever take over, not procedures and not even diagnostics (as there has to be a human being there for legal liability), but I definitely think that it will decrease the sum total of referrals, at least for comp.

ODs would take the biggest hit. A reliable autorefractor and AI doing the screening exam could take away a lot of their routine visits.

Insofar that this comment is right, this is all the more reason that ODs would be looking to expand their scope of practice. You all know the OD lobby much better than I do, but I do not get the sense that they would passively accept defeat. When reliable AGI arrives, I strongly anticipate that they (along with all the midlevels for all other specialties) would rapidly embrace its knowledge base to stay alive.

 

[solitude]

If you think there are a lot of hurdles to implementing truly self-driving cars, which are not yet a reality, despite arguably capable technology, imagine the regulatory hurdles of replacing physicians and surgeons. AI is not a new concept, it has been segmenting our OCT thickness maps, etc for many years. It will augment our workflow. Hopefully it will make being a doctor easier and more fun, as we can spend more time with patients and trainees instead of doing EMR documentation.

The population is aging, more people are getting diabetes, etc. Ophthalmologists are retiring in spades. We will all have plenty of work to do.

 

[LightBox]

If you think there are a lot of hurdles to implementing truly self-driving cars, which are not yet a reality, despite arguably capable technology, imagine the regulatory hurdles of replacing physicians and surgeons. AI is not a new concept, it has been segmenting our OCT thickness maps, etc for many years. It will augment our workflow. Hopefully it will make being a doctor easier and more fun, as we can spend more time with patients and trainees instead of doing EMR documentation.

The population is aging, more people are getting diabetes, etc. Ophthalmologists are retiring in spades. We will all have plenty of work to do.

Agreed. There are so many patients out there now and less doctors. In my area alone, I think 3-4 eye surgeons left in the past 5 years. We have >6 month waiting lists now for many patients.

 

[Slide]

Personally I think the low hanging fruit (refractions, screenings) will involve AI in a significant way. I don't think it'll replace ODs in the immediate future, but I am willing to wager in 5-10 years it's going to take demand away from the number of ODs needed. The argument from years past of it actually generating more human need I think is not going to work this time - the differences now that will change things are 1) remote work/supervision is now fairly cheap and easy to set up, 2) the advancements in AI models and how they are trained are going to likely be parabolic, not linear, and 3) the desire to drive down the cost of healthcare and accept the drawbacks that come with it are going to allow this to flourish.

There are people that will need subjective refractions, but a majority of people don't need complex refractions that AI models can't address. As far as screenings go, the strides in widefield imaging and OCT can probably screen the vast majority of patients fine. The only things keeping it from being super widespread are cost and liability, and with how chipsets and hardware advance, the cost may not be as much an issue in the future.

 

[damien_chazelle_fan]

This is somewhat reassuring to hear. Thank you all for your perspectives. Are you all aware of any data that exist regarding projected demand for cataract surgeries? Another practical thing for me (and all my peers pursuing ophtho) to consider in terms of whether or not to become a comp ophthalmologist is what the surgical volume would look like 8 years from now versus 28 years from now for cataract surgeries given that most of the boomers will not be eligible for cataract surgery in the future (euphemism intended)

 

[Slide]

This is somewhat reassuring to hear. Thank you all for your perspectives. Are you all aware of any data that exist regarding projected demand for cataract surgeries? Another practical thing for me (and all my peers pursuing ophtho) to consider in terms of whether or not to become a comp ophthalmologist is what the surgical volume would look like 8 years from now versus 28 years from now for cataract surgeries given that most of the boomers will not be eligible for cataract surgery in the future (euphemism intended)

I wouldn't necessarily let the latter affect your decision. The demographic cliff is going to be an issue for most fields that's not pediatrics, OBGYN, or psychiatry.

 

[RetinaDude]

I saw an article the other day that said the supply of ophthalmologists will only be about 70% of what is needed by the year 2035 (I think). The age group, for cataract surgeons, is booming and will continue to grow for years to come. Combine these, and you have a scenario of more than enough cataracts. Only issue is it may create scenarios where the ophthalmologist has to spend all day in surgery while the ODs run the clinics, because the surgical demand is so high. I’m seeing that already in some smaller to medium sized cities as too many ophthalmologists retire and not enough fresh docs come in to help out/replace

 

[Dusn]

They’ve been able to send radiology images to India and other countries for decades now. 20 years ago they said that radiology jobs would be outsourced. Did it happen? No. Why not? Because you can’t sue a non-US radiologist.

At the end of the day someone needs to take the liability and AI companies aren’t doing it. They will need some way to put the responsibility on a physician.

The way AI could cause problems is by creating high volume automated billing machines for the companies that drain scarce healthcare dollars - leading for further cuts in reimbursement. Then, once the billing has occurred, the machines will defer to the doc anyway.

 

[Trefoil]

The other issue with AI is that it hallucinates. Medicine has a language, but it is reality based. You can answer history questions based on language. Images can also be learned by AI with enough practice. The thing about ophthalmology is that we are exam based. It takes a high level of coordination and motor skills to even do our exam. Sure, you can take pictures of the retina and optic nerve with minimal skill. You need motor skill to do a good peripheral retinal exam. You need motor skill to gonio a patient. You need the ability to interpret the eye in motion for a sensorimotor exam.

Sure, if I feed the info into a large language model, it might guess the diagnosis. Some reasons why I don’t think that is likely.

1) Liability: humans have malpractice insurance. We make mistakes, but we have insurance to cover us. I don’t imagine insurance companies can actually cover AI programs. I might be wrong about that, but a company could potentially be sued for the AI getting a wrong diagnosis.

2) Humans actually distrust machines more than humans and prefer human error over machine error.

3) You need a human to explain the diagnosis, treatment, and answer questions.

4) Most of ophthalmic surgery requires a very high level of visual interpretation during the procedure. I think we call this “subtle.” Everything in ophthalmology is precise. AI actually learns differently than humans. It’s doubtful that AI will have the capability to process the amount of visual information at the speed necessary to make movements that are less than a mm. It requires fast reaction time and flexibility. I don’t think AI will be as flexible as humans for a long while.

 

[i-doctor]

Another vote for being primarily a cataract surgeon. There are thousands of nuances and individualized set of circumstances in cataract surgery, which will be impossible to program A.I. for... Even surgeons who have 2 or 3 decades of experience encounter new challenges and unique circumstances.
The margin for error in intraocular surgery is the key here. Things can go south real quickly.

 

[kevlar22]

Another vote for being primarily a cataract surgeon. There are thousands of nuances and individualized set of circumstances in cataract surgery, which will be impossible to program A.I. for... Even surgeons who have 2 or 3 decades of experience encounter new challenges and unique circumstances.
The margin for error in intraocular surgery is the key here. Things can go south real quickly.

Purported to be better than an experienced surgeon with better reaction times with constant monitoring by OCT, so I guess we'll see.

 

[i-doctor]

We’ll see!

 

[Dusn]

The key to training AI is that they need the training data.

So step one to training AI to do surgery is to get surgeons performing surgery with robotics and using imaging technology so that both the sensory input and output (motor) data involved in performing the surgery can be recorded in large numbers

Once the input and output data is there, AI can be trained to replace the surgeon. If there’s no realistic way to get that data though then training AI with current ML technology will be very difficult.

 

[voxveritatisetlucis]

The key to training AI is that they need the training data.

So step one to training AI to do surgery is to get surgeons performing surgery with robotics and using imaging technology so that both the sensory input and output (motor) data involved in performing the surgery can be recorded in large numbers

Once the input and output data is there, AI can be trained to replace the surgeon. If there’s no realistic way to get that data though then training AI with current ML technology will be very difficult.

Doesn’t DaVinci already record everything? Intuitive is definitely aiming for AI driven robotic surgery or their PE ratio wouldn’t be >100

Probably not applicable to opthalmology right now but it’s probably in the works

 

[solitude]

The key to training AI is that they need the training data.

So step one to training AI to do surgery is to get surgeons performing surgery with robotics and using imaging technology so that both the sensory input and output (motor) data involved in performing the surgery can be recorded in large numbers

Once the input and output data is there, AI can be trained to replace the surgeon. If there’s no realistic way to get that data though then training AI with current ML technology will be very difficult.

We can't even train AI to drive a car, or solve the associated legal/liability concerns. IMO it will be decades before AI can competently do simple procedures, let alone complex intraocular surgery.