Show notes
Scot Morris, OD, and Rehan Ahmed, MD, on new AI developments and what they mean for the future of the medical field.
- AI-powered patient recruitment: The University of Colorado introduced an AI tool that scans electronic health records to match complex retinal scans with clinical trials, which could sharply cut recruitment costs.
- Early glaucoma detection: New models show AI screening outperforming expert clinicians at spotting early-stage glaucoma by analyzing OCT imaging at the pixel level.
- The “referral fallacy”: Highly sensitive algorithms can trigger a flood of false-positive referrals that overload specialists. The fix is smarter risk-stratification workflows.
- Doctor shortage vs. AI efficiency: Will AI, robotics and long-acting treatments create a physician surplus, or will demographic shifts demand even more specialized care?
- Reimagining the system: Industry disruptions that could put user experience and outcomes ahead of legacy frameworks.
Transcript
Welcome to Real Talk on AI in Eye Care with your host Dr. Scot Morris. And I'm co-host Dr. Rehan Ahmed. In each episode we discuss current news, add in a little AI education, and debate innovative AI tools and topics that will change the industry. Well, hey everybody, this is Scot Morris and my co-editor Dr. Rehan Ahmed. Thank you for joining us for Real Talk number 64. Can you believe that Rehan? I've been doing 64 of these. Crazy, pretty crazy. Keep on chugging as AI changes, we continue to discuss it and talk to so many of our listeners about it. So excited to be here. Yeah, so I have some interesting news of the day, and this is actually local news for me, Rehan.
It was actually, I was flipping through channels this morning and on the local news, which I heard they ever watched the news, but it just happened. It was timing, I guess. And they were just posting that there's some researchers at the University of Colorado at Onshoots Medical Campus that just debuted an imaging AI tool that can scan electronic health record databases to match things like complex retinal scans with all the clinical trials and match that to what people are finding and documenting EHR. I almost think it is going to be, you know, I think they didn't really say this, but if I read between the lines, if you think about how that's going to change patient recruitment for studies and doing all that kind of stuff, when they can go, we can look for, we can go dig through an EHR database and find out the exact people that might match who might be good candidates for this study.
How much would that change the cost of study recruitment? I think it could be a really big thing. I mean, we've talked about it before, Rayon, is that I think the future of clinical research is going to be very dramatic, not just in like, hey, we can find new drugs or we can find new things, but we might find dramatically decreased the amount of money it takes to actually conduct a study because we're going to be able to recruit very specific subgroups of people through the EHR databases. I think that's fascinating. And then the other thing, and this is one of the things I read yesterday, was that this was something that came out of the European Society of Cataract and Refractive Surgeons, the ERCRS, that showed that now AI screening models are now outperforming many expert human clinicians in early, and that's a key word, early, glaucoma detection.
Now, not so much in the modern advance these study shows, but in early detection. Because you and I have talked about this a little bit with oculomics and some of those things, is that some of these OCTs are able to see things at pixels, not microns, but pixels, and they may be able to pick up RNFL lost or GCC lost at the macula much, much earlier than we're able to pick it up. So I don't know. I mean, we're starting to see cases where these screening models are going to assist in finding cases much earlier. And will that result in us having less advanced glaucoma because we're catching it 5, 10, 15 years earlier? Maybe. I think that's one of the things we'll find out.
So that's my news of the day, Rayon. Fill me in. That last one was a great segue to this week's knowledge by talking really about referrals. And it kind of leads into McCollum this month. At the American Academy of Ophthalmology meeting, there was a ton of talk about AI and oculomics. And the takeaway for me basically was the science is real. The science is there. We all agree on that. Now it's about implementation. And that's where you just brought up. These screening models are very, very good. But then we have to talk about how is this going to work in practice. And we talked about this with Rumasa Chana, an ophthalmologist in Wisconsin. Please take a look at that video podcast.
That was fantastic. But just go a little bit a deeper dive in the implementation. So you mentioned the screening algorithm. So take a good algorithm. Suppose it's 90 percent sensitive, 90 percent specific, which is sort of like the baseline with the FDA has been improving. But say you've run around 10,000 people. And say that disease has a 2 percent prevalence rate, which is like per block home or something. So yeah, great. But when you do the math, that's 200 people who actually have it. And the algorithm catches 180. So that's fantastic. That's the 90 percent sensitivity. But it also, it also flags 10 percent of the 9,800 who don't. That's 980 healthy people called back out of that 10,000.
So that's more than five false alarms for every real one. So that's basically base theorem. And no better model sort of repeals that. So you do that across like a large retail network or imagine these screening programs in malls. Everywhere. We're talking about not just 10,000 patients, maybe talking about hundreds of millions of patients. You haven't really built a referral screening platform. You built a referral flood. And if you think the cardiologists and primary care doctors are having too many patients right now, I mean, that would definitely not work in their system. Right. But there's an other side of the coin that we also talk about.
We also talk about the patients who should have been sent but then never actually go to that appointment. And we've talked about this before as well as well as that the the clinic in China that we've talked about where they had a number of patients who were actually referred. But they never actually followed up or a clinic in Southern California that run it ran a telemedicine diabetic ran up these screening clinic for about five years. And it found that only five percent made it to within the recommended window of time. And what's interesting is that oftentimes it soaks to their primary care doctor, but they just never followed through with the ophthalmologist or optometrist referral.
So basically, we have problems on two ends. There's too many in and too few through. And so I sort of sort of view them as the same problem. This is a referral stream full of false alarms. Well, it could be full of false alarms. And it really risks this from really materializing, right? People stop listening. It never really takes off. So what are the fixes and how should we think about this in advance? And it's kind of like the three R's that I'll be talking about. We have to risk stratify and restrict, right? And so basically, you have to somehow increase the prevalence of that disease. I mentioned something like two percent beforehand. You have to increase that to say 10 percent.
So maybe you do an automated workflow where you ask about family history or you ask about medication use or other risk factors that are associated. Maybe with something like glaucoma or diabetes. And then you also rationalize. That's the last star. You refer the irreversible high harm findings. And so maybe something like if you find an ocular melanoma, right? You're okay with referring if it's a false positive, right? But you do not want to miss any of those, right? So you don't want to miss that. And so that's sort of where I'm sort of thinking about this. I think we have to be very, very understanding of this whole referral fallacy. Just because we have a screening system that, as you said, is significantly better than humans.
It's how do we scale that? And so for the knowledge about this week, I want to just focus. I guess this is not really telling you what to think necessarily, but just that this is a call out for our audience to start thinking about this in more details. When you get amazing sensitivity numbers and specificity numbers, you have to think, well, how is that going to be implemented? What's the prevalence of the disease? What are we talking about here? How is this going to impact our primary care colleagues and cardiologists and whoever we're referring to? And what happens when those patients go and what happens when the patients don't go? So just to call out there, I think it's really interesting and something we'll be talking about a whole bunch.
Well, I think that's a good transition, Rehan, into something I want to talk about for our fireside. And for the audience, we're probably going to ruffle some feathers on this one. So when we start having that kind of referral pattern, do we have too many doctors or do we not have enough? And I often, you know, I sit in the hallways of meetings or I sit in consulting meetings or month, you know, group chats or whatever. And I hear this and, you know, one group of people go, oh, we have too many doctors and then too many ophthalmologists, too many ophthalmologists. And then I have other people going, we don't have enough ophthalmologists. We don't have enough ophthalmologists.
And, Rehan, I want to kind of attack that today. And so I'm going to start with this and I'm going to pick on you a little bit. And I'll let you pick on me. Now, man, you're really hitting the third ground. But that brings us, I think about retina, right? If you now all of a sudden are finding all of these potential, you know, early macular degenerative patients or early GA, are we now going to completely flood retina specialists? I don't know if we have enough retina specialists now. Are we going to so totally flood them that we're just never going to have enough retina specialists? But on the other hand, I wonder, but are we going to have such a decrease in cataract surgeons?
Because with robotic cataract surgery and all that kind of stuff that may be taken where one surgeon's doing what used to take three or four or five surgeons to do can do with the same amount of time. So do we need more ophthalmologists? Yeah. Or do we have too many ophthalmologists? I think a classical, you know, if you talk to folks like Richard Lowe, the economist who's based out on the East Coast, or a lot of these sort of demographic trends. Classically, we always say that the number of ophthalmologists basically staying flat or going down actually a little bit and our population is increasing. And so we are in the midst of a demographic crunch with the baby boomers still getting cataract surgeries and needing retina care.
And that's just going to increase. And so classically, we say, well, we need to provide more tools to make doctors and the whole clinic experience much more operationally efficient. I think that's generally true. I think your example for retina is probably a good one. But I always like to think about sort of black swan type of situations or events or technologies. And you mentioned one. How would robotic cataract surgery change things? Right. And in terms of retina, how could gene therapy change things? Imagine you're getting injections every month. What if there's an injection that you just give once a year or maybe once and you're done and it just produces anti-veg F.
And they're working on this stuff is in clinical trials. Or, you know, I mean, when we're able to diagnose this stuff decades earlier, do we prevent a whole bunch of people from ever getting the severe stage? So it's very easy. And so I was like that we I love the example of, you know, it's hard to predict short term. It's much easier. You know, the long term, we know sort of directionally we're headed. And this, I think it's the inverse for here. I think the short term is pretty clear. I mean, we're going to need more ophthalmologists just because of the demographic trends. And these these systems are going to come online the next five to 10 years, about five years.
But I think about this, whether you're ophthalmologist, optometrist or any part of medicine, you know, I wonder is the lifespan of any of us is maybe 20, 25 years. If we look, if I look back 25 years ago, you know, where I was at, what I was doing in, would that be 2001? Yeah. None of the stuff we're talking about ever existed. Right. And the demographic of the iPhone wasn't even out 25 years ago. No, not even a little bit. Right. We were still using pagers. And that was the advanced high tech group were using pagers. You know, so you wonder and look at, okay, so over the lifespan of a retina specialist or a glaucoma specialist or, you know, even a general, a general surgeon, general ophthalmologist.
Are we training? You're right. I agree with you on the baby boomer thing, right? I think for retina and for cataract surgery and that kind of stuff. But do you look 10, 15 years down the road and there's a big, you know, as Richard Edlow said, there's a big, huge population drop of people who are going to be needing cataract surgery. Yeah. Yeah. And so now what happens? Do we have all these general ophthalmologists that don't have anything to do? And now are they not doing cataract surgery, but doing some other type of surgery or, you know, I mean, there's a whole bunch of questions there, but then we could say the same thing for optometry. Right. Now, I'll let you pick on me for a little bit. Do we have too many?
And the answer is, it depends who you ask. You know, you listen to the schools and they're like, we don't have enough. You know, we're going to be of such a shortage. And I'm like, on the other hand, I go, yes, but you got to remember that there are going to be practices who are using artificial intelligence and augmented intelligence to see two to three times more people than we're seeing now. So instead of needing 2,000 docs, maybe we only need 1,000 docs. And instead the schools are going, we need 3,000 or 4,000. Right. So I don't know. I look at this and go, I think, as you said, and we don't have a crystal ball. I personally think we're turning out way too many students, way too many young doctors, and it's diluting the entire market in the optometric space.
You know, there's all kinds of people going, I'm happy to put this job and this job and this job and this job together. They're working three or four jobs to earn a living. That means we have too many doctors. But I'm sure I was sure that you know, in terms of these sort of black swans, one thing I could push back a little bit is that there's always legislation and we're talking about UBI and things like that. I could see a future where we mandate that certain healthcare professions or certain exams, even though they can be done by an AI or an agent and be done more efficiently. That's still a doctor has to sign off on the exam or do some parts of it or you're not just not allowed to use it or there's some tax on using an AI.
So the doctor still has a competitive role to play compared to that. So, you know, there's a black swan going each way. So it's kind of hard to predict. Instead of predicting, I think my advice to our audience is to build it. Let's build the future, right? We are in a whole position to do it. Getting involved with your local professional society and your group and these types of things and talking to patients are really important. That's what I try to do because you can't, it's hard to predict it, especially now. It's really difficult. Well, and, you know, there's a great article that should be coming out of this. I think it'll be next week after maybe even the same week. This podcast comes out, Rehan, written by Andrew Gay, who is a doctor out of it was Australia.
Yeah, Australia. And boy, he posed an interesting question about what happens if there's a totally different model. You know, I've kind of joked about that a little bit on the air and off air, like we're all thinking about it working in the model of today. But what happens if there's a disruptive new model where vision insurance doesn't exist and private pay is different and we're not doing histories and diagnostic testing in the office. And it's about the patient experience and not about the provider experience. It's a great, it's a great article. I look forward to hearing everyone's comments, basically re-imagining the whole eye care system from the outcome that we want, i.e., we want people to see.
We want people to see well. Straight, just flat and flat. And more backwards from there and not trying to retrofit things onto our current system, which we often end up talking about and doing. So that's probably not the right way. So super interested. I'm really excited to hear everyone's feedback on that. It's cool. Hope you guys enjoyed today's show. This is You Can Reach Me, S-C-O-T at a-i-n-i-care.ai or rayon, R-E-H-A-N at a-i-n-i-care.ai. You can check out us, any of our podcasts, our innovators series. We have a couple of really cool innovators series coming out here in the near future. Listen to our real talk, all 64 of them, if you really have a few minutes to sit down, as well as lots and lots of articles about different concepts within AI.
And of course, we want to address that. You can go to our website, Rayon. I spent a lot of time building a program called Minutes Matter. It is the all-encompassing comprehensive guide of how you implement AI into your practice. Downloadable now on aiineyecare.com. Sounds great to talk to you next week.