Curious what people's thoughts are on this?

They are currently testing out systems now where patients come in and explain their problems to mid level providers (PA's and NP's). Then the info gets entered into an AI program, a work up and treatment plan is generated and implemented.

(This is just one that I'm aware of in Texas. I'm sure there are many others elsewhere involving MD's. I do not know all the ins and outs of this protocol, I just heard from a colleague in a different field that it's being done with a few mid level providers for primary care in this particular healthcare system)

Obviously this protocol still has a human health care professional involved, so it's not as if the patient is directly interacting with AI, and it's not as if a human can't intervene with the plan if they feel the need to go in another direction

All that being said, who would feel comfortable having first line medical follow care being driven by AI?

I think we're a long way from non-humans being there to do the physical exam and implement physical treatments, but the day is coming that many of these decisions could be made without human involvement. As a cost saving strategy, the pressure will be there.

I personally feel that the technology will be there before people's comfort level will be. I imagine that patient comfort will be largely correlated with age (older people prob slower to accept that)

Aside from direct patient contact, I can easily see AI reading images (xray, CT, MRI) and pathology slides.

Several years ago I saw a presentation where AI was shown a tremendous amount of images of skin cancer and non cancerous moles. Then the moles were tested (confirmed by path reports after biopsy). The AI was able to identify cancer in moles that most human docs thought were not likely to be cancerous.

The Grinder (99) said:

---

Curious what people's thoughts are on this?

They are currently testing out systems now where patients come in and explain their problems to mid level providers (PA's and NP's). Then the info gets entered into an AI program, a work up and treatment plan is generated and implemented.

(This is just one that I'm aware of in Texas. I'm sure there are many others elsewhere involving MD's. I do not know all the ins and outs of this protocol, I just heard from a colleague in a different field that it's being done with a few mid level providers for primary care in this particular healthcare system)

Obviously this protocol still has a human health care professional involved, so it's not as if the patient is directly interacting with AI, and it's not as if a human can't intervene with the plan if they feel the need to go in another direction

All that being said, who would feel comfortable having first line medical follow care being driven by AI?

I think we're a long way from non-humans being there to do the physical exam and implement physical treatments, but the day is coming that many of these decisions could be made without human involvement. As a cost saving strategy, the pressure will be there.

I personally feel that the technology will be there before people's comfort level will be. I imagine that patient comfort will be largely correlated with age (older people prob slower to accept that)

---

As far as using AI to augment care, I think it's a reasonable possibility. However I think the reality is that AI is essentially just a really good search engine, and is highly dependent on inputs (from a good history and physical, as well as what resources it has access to).

There is a lot of nuance to medicine. The longer you're practicing, the more gestalt you develop, and the more accurate it becomes. Can also become inaccurate if you aren't staying up to date on research and advances.

There have been plenty of times where I know docs that did everything by the book, but something didn't feel right, so they did an extra test and were able to make a diagnosis that would have otherwise been missed.

I think AI will be adequate for 90% of medicine one day. But it'll take a long time for it to approach that 100%.

And I agree, cost cutting will be a big push for this. As it is, people are trying to cut costs by allowing mid-level/non-physician providers to practice medicine independently. You frequently hear the anecdotes of one of them finding something a physician missed. But you rarely hear about all the missed diagnoses, and medical errors, and not to mention excessive amounts of unnecessary tests. There are plenty of good PAs/NPs, not trying to hate on them. But they aren't doctors. I think slowly you're going to see a degradation of care and outcomes, but it'll be too slow to realize, and too late to quickly fix.

Quote:

---

Aside from direct patient contact, I can easily see AI reading images (xray, CT, MRI) and pathology slides.

Several years ago I saw a presentation where AI was shown a tremendous amount of images of skin cancer and non cancerous moles. Then the moles were tested (confirmed by path reports after biopsy). The AI was able to identify cancer in moles that most human docs thought were not likely to be cancerous.

---

I'm sure AI might eventually reach the point where it's fully capable of this, but at the current moment I'd be curious to see how accurate this AI was. Sometimes it may be overly sensitive at the expense of specificity, resulting in over testing. Also, who takes liability for an AI missing a lesion that needs biopsy?

https://radiologybusiness.com/topics/artificial-intelligence/after-seeing-17-different-doctors-boy-rare-condition-receives-diagnosis-chatgpt

harge57 said:

---

https://radiologybusiness.com/topics/artificial-intelligence/after-seeing-17-different-doctors-boy-rare-condition-receives-diagnosis-chatgpt

---

Yeah, I've seen this case before. She (and the doctors) likely had the diagnosis staring at them in the MRI. Not sure if it was miscommunication, or someone couldn't understand the MRI results, or what. Since no one can have access to this kid's medical record to review, no one can identify who missed it. Would be interesting to see what the MRI actually said, since the neuro surgeon was obviously able to see it on the scan.

The thing about medicine these days is that it is focused on throughput. Everyone wants to make money, but no one wants to pay for services, and everyone is trying to get their hands on a piece of the money. The focus in medicine these days is "when you hear hooves, think horses, nor zebras." So there's a tendency for people to run the numbers and base treatments and decisions off statistical likelihood. If you have chest pain that gets worse when you walk, and relieved with rest, and you have high cholesterol and blood pressure, and are over 55, it's probably your heart. But every now and again you'll see an 18 year old with a heart attack. Do you test every 18 year old the same way you'd test every 55 year old?

I think AI definitely can provide good differentials and statistical likelihood, especially for more obscure disease/events. But I think it has a long way to go before it can replace physicians. But I may be biased.

AggieOO said:

---

i'm in the IT world and AI is all the rage. There are some amazing applications for it, and it is doing some incredible things. It also have some terrible, terrible usages, which need to be talked about so people are aware, but its scary. Its not all of my job, but I sell a lot of things that are the underlying infrastructure for AI. I say all this to preface my stance...

Zero chance I want AI to be my doctor. People can make mistakes or miss things, but there are things people can literally see and hear that AI "never" will be able to. AI can/will see/hear, but its not the same. I think AI can be a massive tool that doctors can use, but it should only be one tool in their arsenal. AI is only as good as its algorithms and inputs.

Also, AI seems to be used these days to cover anything where a computer is "thinking." In reality, there are multiple things that fall under that umbrella like: Machine Learning (ML), Artificial Intelligence (AI), and Generative Artificial Intelligence (Gen AI). All three of those have similarities but are also very different. And the applications within the medical field obviously vary. I was at a hospital today discussing AI.

---

Great post. I think AI will have tremendous influence on determining best pathways, but the problem with using AI for diagnosis is that the "inputs" will be terribly flawed. A lot of medical school is learning a new language, in a way. But it takes a long time to go through medical school and complete a residency for a reason. You have to figure out what "normal" is and what's abnormal. I have people coming to see me all of the time thinking that they have X, Y, or Z because they felt A, B, or C. It isn't their fault, they just aren't used to looking at, feeling, etc., and comparing that to normal. Nor can a person use the equipment I have in my office, such as fiberoptic scopes, on themselves in order to see areas not otherwise visible. I spend a pretty significant amount if time talking people off of ledges they have no business being on.

I think AI will be able to determine best treatment algorithms, best medications, etc. for a given diagnosis or condition once all of the appropriate information is input.

Quote:

---

Aside from direct patient contact, I can easily see AI reading images (xray, CT, MRI) and pathology slides.

---

i missed this on my first read of the OP. this is absolutely an area where AI could be quite useful and the patient would potentially never even know AI was involved. Technically, this would be a subset of AI, which is Machine Learning. ML is where data and instructions are given to the "computer" and it is trained specifically to do certain tasks and only those tasks. It doesn't "think" for itself. Reading medical images is a similar use case to another customer I'm working with who is in O&G. They collect seismic data to identify where companies should drill. But before they can take their machinery out into the field, aerial drone footage has to be manually reviewed for things like ponds, fence posts, sinkholes, creek beds, existing pipeline, telephone poles, etc. We are looking at training a model to identify these things in the imagery. Its not only a cost saving exercise (over time, big up front cost), but it saves A LOT of time, as the model can review imagery in a fraction of the time. A very similar algorithm could be applied to medical imagery.