ARVO '26: TKIs for Wet AMD and How We Got to the AI Era of Research

Greg Nothstein:

TKIs are all of the rage, but what do the data tell us about their ability to effectively treat wet AMD?

Scott Krzywonos:

I'm Scott Krzywonos here with Greg Nothstein, and you are listening to New Retina Radio from Retina Today, and Bryn Mawr Communications. Dr. Akshay Thomas sits down to outline findings from DAVIO 2, the Phase 2 study assessing the safety, and efficacy of EYP-1901 for the treatment of wet AMD.

Greg Nothstein:

And doctors Aaron Lee, and Cecelia Lee recap their Cogan Award lecture at this year's ARVO Annual Meeting. Hear how they envision the future of ophthalmic research driven by AI, and machine learning. We could be on the edge of a durability shift in wet AMD therapy, thanks to the advent of tyrosine kinase inhibitors. But what do the data tell us?

Scott Krzywonos:

Dr. Akshay Thomas shared findings from the Phase 2 DAVIO 2 study evaluating EYP-1901 in patients with wet AMD, and he's here to tell us what the researchers found. Dr. Thomas practices at Tennessee Retina in Nashville. Dr. Thomas, welcome to New Retina Radio.

Akshay Thomas, MD:

Hey, guys. Thanks for having me.

Scott Krzywonos:

Let's talk about why durability is even a factor here. Don't we already have effective therapies for wet AMD?

Akshay Thomas, MD:

Yes. We're always on the lookout for better therapeutics, of course. I mean, despite currently available therapeutics, there certainly remains a significant unmet need for effective durable disease control in not just AMD, but essentially all retinal-like state of diseases. We all know that the need for clinic visits as often as every four weeks for optimal outcomes places a high burden on patients, and caregivers, and of course, providers. This treatment burden is a significant driver of undertreatment.

If we look at our real world outcomes, they always pale in compares into study outcomes, and that's just because around 40% of patients with wet AMD, and DME in real world studies actually discontinue therapy. So, additionally, while our current therapeutics are effective, Anti-VEGF agents really only address one aspect of disease pathogenesis. We know that wet AMD, and DME are multifactorial. And so while Anti-VEGFs treat the excess VEGF that leads to angiogenesis, and leakage, it don't really address the inflammatory process. That also contributes in some way, shape, or form to these retinal diseases. So, essentially, this high burden, the high discontinuation rate means that roller outcomes just don't reflect clinical trial results. So, we are constantly on the lookout for more durable multimodal mechanism of action therapeutics that we can use for our patients.

Greg Nothstein:

Tell us about EYP-1901.

Akshay Thomas, MD:

Yeah. So, EYP-1901 is essentially a bio-erodable intravitreal insert, which combines vorolanib, which is the tyrosine kinase inhibitor with [inaudible 00:03:11] drug delivery technology, which is currently under study from EyePoint Pharmaceuticals. The insert is basically 94% drug, and 6% bi-rotable matrix, which doesn't include PEG, and PLGA, which are just some things which have, perhaps, been implicated previously in inflammatory events with therapeutics in the eye. So, the unique cell proposition, of course, is that vorolanib offers potentially a multiple mechanism of action, blocking VEGF, signaling PDGFR, signaling as well as IL-6 mediated inflammation.

And at least from the preliminary data we have, we know that EYP-1901 provides immediate release of vorolanib reaching target tissues within hours, and giving essentially a consistent daily dosing of vorolanib for at least six months based on some of the [inaudible 00:04:01] samplings data that we have. One important thing we have to think about with some of these more durable inserts is that with ELP-1901, the vorolanib elution precedes the matrix by erosion. So, essentially you don't have this generation of free floating drug particles in the eye as the medication kind of dissolves.

Scott Krzywonos:

And now I understand in the real world, if this drug were approved, it would be used alongside say an Anti-VEGF agent. It's not one, or the other, they would be used together. This one has an intracellular mechanism versus the extracellular mechanism you'd find with a more conventional therapy.

Akshay Thomas, MD:

Yeah. I mean, I think some of it is to be determined. I think if we look at some of the Phase 2 data, we'll have a better sense once we have the Phase 3 data results, but if you look at some of the Phase 2 data results, you'll see that there are some patients that just do phenomenally well. They get a single treatment, and then they don't require supplemental therapy at least for the six months of fall duration for which we have data. And there's other patients that still require therapy, but they perhaps need therapy less frequently than they needed therapy prior to a treatment with EYP-1901.

So, I think in the real world, how we probably will end up using it, hopefully assuming it comes to market is, is that we use our Anti-VEGF agents as we did previously, but perhaps it allows for us to treat, and extend them more further along than we might have been able to do with Anti-VEGF monotherapy. And I think we will have some of those super responders where we might be able to treat them with monotherapy with EYP-1901 as well, but I think we have to wait, and see what the Phase 3 data shows.

Greg Nothstein:

All right. Let's get to that Phase 2 study within DAVI-02. Tell us who was enrolled.

Akshay Thomas, MD:

Yeah. So, DAVIO-2 was a Phase 2 trial of EYP-1901 versus on label aflibercept two milligrams used in patients with previously treated wet AMD. So, essentially we're including patients that have had at least two Anti-VEGF injections for the wet AMD, and have actually demonstrated a response to Anti-VEGF therapy. So, they ended up including 161 eyes. Because these were previously treated eyes, their vision was pretty good. So, they were 20, 32 at steady entry. They had relatively thin retinas in the sense that they were mostly fluid-free because they were patients that responded to VEGF, and are treated. So, their average CST at study entry was about 265 microns. And these were patients that had a heavy burden of treatment. The mean number of injections that study S had received in the year prior to study enrollment was 10 Anti-VEGF injections. So, these were frequent flyers as they required a lot of therapy at the time of study entry.

Scott Krzywonos:

And now what about the design itself? Walk us through how the study was structured, and what the primary endpoint was.

Akshay Thomas, MD:

Yeah. So, patients were essentially randomly assigned to one of three treatment arms. They could have been assigned to low dose EYP-1901, which was a two milligram dose, a high dose EYP-1901, which was a three milligram dose, or standard of care aflibercept two milligrams. So, all patients received loading doses of aflibercept two milligrams essentially at day one, week four, and week eight, the patients in the EYP-191 arms additionally received a single dose of EYP-1901 at week eight. At that point onwards, the patients in the aflibercept arm received standard of care aflibercept two milligram injections every eight weeks, whereas the patients in the EYP-1901 arms didn't receive the only received sham injections. Starting at week 12, patients in the EYP-1901 arms could receive rescue, or supplemental treatment based on some protocol specified criteria, or even just investigator discretion. And the primary endpoint was your best corrective visual acuity changed from baseline to average the average acuity measured between weeks 28, and 32. So, essentially six, and seven months post dosing of EYP-1901.

Greg Nothstein:

And was that primary endpoint met?

Akshay Thomas, MD:

Yeah. So, the DAVIO-2 met its primary endpoint with statistically non-inferior BCVA outcomes with a single dose of EYP-1901 versus aflibercept Q8 weeks. And again, this was the month seven, month eight averaged visual acuity, or roughly six months post-dosing of the EYP-1901.

Scott Krzywonos:

Now, I assume this might have been because as you mentioned, some EYP-1901 patients received supplemental aflibercept along the way. Is that how the primary endpoint was able to be met?

Akshay Thomas, MD:

That's a great question. So, we actually did a subgroup analysis where we looked specifically at eyes that were supplement-free across all three arms. And that basically demonstrated that receipt of supplemental aflibercept injections did not drive visual outcomes. So, if we look at patients who did not receive supplemental injections, the BCAVA results were similar across arms. In fact, in both the EYP-1901 arms, between 85% plus of patients had stable, or improved vision by the end of the study.

Scott Krzywonos:

You mentioned that retinas were fairly thin at baseline. The baseline measurement was approximately 265 microns for central subfield thickness. Were there any changes in anatomy over the course of the study?

Akshay Thomas, MD:

Yeah, great question. I mean, if we look just at the CST at month eight, so again, roughly six months post-dosing EYP-1901 was non-inferior to aflibercept regarding CSD change at month eight, but I think what you're getting at is also what's the fluid doing over the course of the study. So, if you look at the aflibercept arm, we kind of see that saw tooth pattern of central cell field thickness changing, which we didn't appreciate quite as much in the EYP-1901 arms. And that saw tooth pattern of fluid, which means fluid's kind of coming, and then going, and coming, and going, that retinal thickness variability has been identified as a prognosticator of worse visual outcomes, including fibrosis, and atrophy over time. So, we actually did a post-hoc analysis of these study patients using AI-assisted OCT analysis, and we basically saw good stability of intraretinal fluid, subretinal fluid, and PD fluidics over time in the EYP-1901 arms without that saw tooth pattern, which we observed in the aflibercept Q8 week arm. So, both durability, and kind of stability of retinal anatomy over time is very convincing based on the data we have to date.

Greg Nothstein:

Dr. Thomas, congratulations on a wonderful presentation at this year's ARVO meeting, and thank you for joining us here on New Retina Radio.

Akshay Thomas, MD:

Thanks for having me, guys.

Scott Krzywonos:

Like it, or not, AI is a fact of life in 2026. To best learn where we can go, we need to take stop of where we are today, and where we were yesterday.

Greg Nothstein:

Doctors Aaron Lee, and Cecilia Lee delivered this year's Cogan Award lecture at the ARVO annual meeting, which is also available to view on YouTube. They covered how AI, and machine learning has, and will affect retina research. Dr. Aaron Lee, and Dr. Cecilia Lee, welcome to New Retina Radio.

Aaron Lee, MD, MSCI:

Thank you so much for having us on the radio.

Cecelia Lee, MD, MS:

Thank you so much for having us as well. We are very excited to join.

Greg Nothstein:

Well, we're glad to have you. Let's start here. AI is a whale sized topic, and we have to break it down, and you described it in five distinct eras of research. Can you tell us about them please?

Aaron Lee, MD, MSCI:

Sure. So, we were really wanting to walk people through the very recent history of how we went from scribbling undecipherable notes in a paper chart all the way to this new [inaudible 00:12:01] decision medicine enabled by AI seems to be on... We are right at the cusp of seeing that future come true. And so we had broken it out into these five distinct eras across the evolution of clinical research, and Cecilia, and I were fortunate enough to have been part of that all the way through in our research career. So, we were there when it first started with paper charts all the way through the advent of the electronic health record to this era of big data, and then with this multimodal deep learning AI era that we are still sort of in the midst in, and finally this new era of precision medicine that we are approaching.

So, that was our ability to recount our history, and our journey through the evolution of clinical research, and also to be able to thank all of the people who have mentored us along the way.

Cecelia Lee, MD, MS:

Yeah. I think what made the presentation really special for us is really to walk, and walk through our journey. So, we basically, I mean Aaron mentioned who he met, and he was mentored by from undergrad, and then what he did as a med student. And I also mentioned my mentors at Emory doing my research doing really paper-based research, then really loving the random images, and really being in love with everything that you can show that is beyond what is just described by a word, or two words. And then afterwards, as Aaron mentioned, that we are really fortunate to see in real action from paper charts to EHR era. And it was a painful transition really because everybody had to take a course, or two to see what is clicking like, and where you have to click, and all that. But really our foundation knowledge, and paper chart era really taught us what is really important to look for when our datasets became more enlarged at scale.

Greg Nothstein:

All right, let's start with the manual research era. Can you tell us about that?

Aaron Lee, MD, MSCI:

Sure. I think that was a time that was where clinicians had to be very clever, and really think deeply about the clinical hypothesis that they wanted to test, or gather data around because there was so much effort involved in extracting that from the charts. So, you would either have to find an army of poor medical students of which I was one of them to go, and pull the charts manually from the library, put them on a cart, take them down the hallway, open up each one of them, and then transcribe out on another piece of paper what the age of that patient was, and the visual acuity at a particular date. And then you would have to gather all of that, and then put that through one of these statistical approaches to arrive at an answer. And that whole process took a lot of time, and effort, and you had to be very parsimonious about what you decided to go after because of the cost of human labor, and capital that was involved in being able to test a single given idea.

Greg Nothstein:

And then the age of digital records began, what was that?

Cecelia Lee, MD, MS:

Well, I think we are very fortunate because Aaron, and I happened to be fellows at Moorfield's Eye Hospital in London, UK. So, the big data field was much more advanced in UK than in the States because they had many different hospital systems were actually using the same hospital records. So, all of a sudden we had access to millions of peoples of data that were clinically recorded, and then it was immediately accessible. And then so that really gave us the... We fell in love with, oh my gosh, we can actually replicate pseudoclinical trials in a matter of a week, or so. And really we were no longer limited by what type of, or how many charts we could review in a humane way possible, but it was really limiting to our ability to visualize the entire dataset, and really making meanings out of some values that were everything was going to be P-value based significant, but then we were really looking for the meanings that were clinically meaningful, and then physiologically plausible, and biologically meaningful.

Aaron Lee, MD, MSCI:

Yeah. And I would just add that particular era, there was different parts of the world were advancing at different rates in some ways. So, in the United States, what people called electronic health record sometimes was just scanned in paper. Literally, they would put it on a scanner, and put it into a PDF, and they would call that electronic health record. In the United Kingdom where we went, the field was much further ahead at that particular time where they had visual acuities in a particular part of the electronic health record, and then the macular exam of the other eye in a different part of the medical record, and you could start to aggregate, and gather all that data in discrete fields that would allow you to do that type of analysis that Cecilia was talking about at a much faster rate. And so it was a very interesting time in history where different parts of the world were advancing at different rates as they were transitioning from paper to an electronic media.

Cecelia Lee, MD, MS:

And I think I want to mention that another benefit that we were kind of unexpectedly finding is that the fact that they were EHR, we are no longer limited to ophthalmology charts. So, if our dataset actually included all the systemic data like Alzheimer's, or diabetes, heart attack, now we could use all this systemic data as either biomarkers of what could predict were somebody worsening in eye diseases, vice versa. Are there some eye characteristics, or eye disease features that could be predictive of somebody developing other diseases in the brain, or in the heart, or in the kidney, and et cetera? So, I think it really opened our eyes, and then also kind of the door to what was previously not possible to answer, but it really enabled many, many different questions.

Scott Krzywonos:

I see. Okay. So, we start in the manual research era, which was from time in memorium up until computers. And then like you described, we get into this digital records era that also quickly becomes the big data error, but different places are at different points in that. Aaron Lee, like you pointed out, just simply making a paper file, a PDF does not an electronic health record make, but maybe at the time that was good enough. So, I get how those two are a little fuzzy, and spongy, the digital records era, and the big data era, but it feels like there's a really big shift when AI is introduced, when there's this, what you guys call the imaging, and AI era. So, when does that shift occur? Is there some sort of seminal moment, and then how do you define it?

Aaron Lee, MD, MSCI:

Well, let me just real quickly help with the previous fuzziness a little bit. So, I think initially in the early phases of the electronic health record era, it was a single center still learning from their data. And in the big data era, we started to aggregate these datasets across many health systems, and like Cecilia was saying about many different medical practices across the human body, and we started to learn things at scale at population level type data that was not really available before. And then if you had asked me what was the marker, or the beginning of the machine learning, deep learning era, it was probably the advent of the computer vision revolution that occurred. So, it used to be that if you had a picture of a car, and you wanted a computer program to tell you that this was a picture of a car, it would take 10 years of somebody's poor graduate school PhD work to make that work maybe 30% of the time.

And we saw a fundamental shift in the way that computer vision occurred, and worked because there was this idea that you could get models, or AI to learn directly from the data itself. It would learn to do everything from start to finish, and all you needed was thousands of examples of cars, and thousands of examples of other things so that the model could just learn from this is a bucket of images of houses, this is a bucket of images of cats, this is a bucket of images of dogs, and this is a bucket of images of cars. And that's all you needed. It learned everything in between. And that was a completely trans [inaudible 00:21:36] idea in the computer vision world. And the first examples in ophthalmology that we were a part of was that same thing of taking millions, or hundreds of thousands of OCT scans, and saying, "This is our OCT scans of people with macular degeneration, and these are OCT scans of people who are normal. Can the model learn from beginning to end how to do this binary task of saying AMD yes, or no?"

And what blew my mind was the ability of these models to learn from that early on in the deep learning era. And so that I think is sort of the hallmark time when we first saw the power of these models, and this technology, and we all realized that it was going to change the world.

Cecelia Lee, MD, MS:

But Aaron, maybe we should tell them that honestly, we didn't know. So, our career really has been through serendipitous path. I mean, it was really because we happened to be in the right time, and the right people, and then encountering right datasets. Aaron, maybe you can mention that in 2015 that's when we had the initial discussion in deep learning, and then you didn't believe that, right? So, do you want to tell the story?

Aaron Lee, MD, MSCI:

It's funny because you should never listen to what I say about the future because I'm always wrong. I thought that when deep learning first happened, and I was reading the tech news about this, I thought it was another fad. I thought it was like we see this all the time in the internet 2.0 era that we were in at back then where there's a flash in the pan, and everyone thinks it's going to change the world, and it actually doesn't. And so when deep learning happened, I thought it was another technological fad, and it wasn't until we had the, like Cecilia was saying, being in the right time in the right place, knowing the right people who helped us along our way, that we ran our first experiment, and I was stunned at the power of these models, but it wasn't until I saw those results firsthand myself that I believed that this was going to be a real thing. And so, again, if you ask me what the future is going to be, maybe you shouldn't listen to what I have to say.

Greg Nothstein:

So, I guess the next question then to ask is what's next?

Aaron Lee, MD, MSCI:

So, yeah, I kind of think that we are seeing this confluence of technology, and data right now. A lot of people are saying that we're living in what is being called the fourth Industrial Revolution, that the convergence of large datasets, data sets that are digital in nature like imaging of the retina, the visual field data, all of that data is electronic, but it's captured out of velocity, and at a volume that is very hard for human beings to process. So, that convergence of data just coming in at a tremendous rate coupled with this technology that can actually learn new insights from it is going to put us in a place where we will be able to chart new ways of taking care of patients. This idea of personalized medicine of being able to tailor new interventions to an individual is, I think, we're really at just the beginning cusp of that occurring.

Cecelia Lee, MD, MS:

There's definitely a big discrepancy between where research is heading, and then what is actually available clinically in our routine clinical care, or what's available to consumers. So, we would love to see more of a merging, or implementation of what's happening in the research. I think as Aaron mentioned, I think the next era is the precision era where we could really leverage all the available data that we could gather from population, but then now at the individual level, and then predict somebody being at risk of developing X, Y, disease. And then how can we use that to provide some type of prevention therapies, or prevention methods? So, I do think that we are living in this era of excitement, and there are many, many possibilities, but there are a lot of work to be done.

Aaron Lee, MD, MSCI:

Yeah. And this idea is of predicting a given person's future is really broad, and really powerful, not just in everyday clinical practice, but even changing the way that we do clinical trials through this concept of synthetic control arms, or digital twins, there's a lot of enthusiasm around the application of this core idea of predictive analytics, or forecasting based on artificial intelligence. So, I do think we are living through this time when a lot of people would've called it science fiction. We're fortunate enough to be living through that revolution today.

Scott Krzywonos:

I know what you mean. Sometimes it sounds like something from Star Trek. It doesn't sound like something from our world, but here we are. I want to know what patients think of all this because this is certainly, this is not research for research's sake, this is research so that patient care can improve. And I'm wondering, this is really two questions, one, how you summarize this succinctly for the curious patient, and two, how you articulate the benefits of this technology, which I imagine for a lot of patients is intimidating, or unknowable, and they need to understand why there is a pot of goal at the end of this rainbow.

Aaron Lee, MD, MSCI:

I've been stunned at how eager the American public is to crawl into a self-driving car. I've been sort of stunned by that because maybe people just hate driving, and they hate traffic, and all this stuff, but there is a certain level of trust that you need to have with AI to put your life literally on the line, not just your life, but the 50 other people that you could endanger when your car does the wrong thing on the highway. Yet the NTSB, the average American public is so eager to crawl into a level five automated, fully autonomous vehicle. People can't wait for Waymo to get to their cities.

It is a time when there is substantial hunger for technology, and AI in certain parts of our everyday lives. And I think when patients start to think about AI, and in being able to help them with their medical care, they may already have crossed that bridge before the physicians get there. They're already using ChatGPT to understand what the doctor is saying. And rather than going on Google to look something up, they're using these frontier AI models to understand their medical history, and their medical journey. And so I think if doctors started to explain to patients that we're starting to use these tools to take better care of them, they might be totally on board with this idea.

Cecelia Lee, MD, MS:

Yeah. And my clinic, I've been surprised that patients have been really eager to donate their time, and they really participate in this study. We do have our AI-ready study that we've been so grateful for all the participants that have participated in our study, but some participants are very suspicious about, and then also are concerned about security, and privacy of the data, which is definitely has reasons to be worried about. So, I do think that more open dialogue in terms of open communication, and transparency, and then community engagement, and with the different stakeholders will be really helpful because without the trust of the community, and public, I think we won't be able to advance as fast, and then also we would love to see where the science takes us because ultimately we do want to improve the health of our patients. I think it will be really important to continue the discussions, and dialogue.

Scott Krzywonos:

Dr. Aaron Lee, and Cecilia Lee who received the Cogan Award lecture at this year's ARVO annual meeting. Thank you for coming on New Retina Radio, and giving us a summary of her talk.

Aaron Lee, MD, MSCI:

Thank you for having us again.

Cecelia Lee, MD, MS:

Thank you so much.

Greg Nothstein:

Thank you so much for joining us. That concludes our coverage of the 2026 meeting from ARVO that occurred a few weeks ago in Denver.

Scott Krzywonos:

If you missed our first episode, it was an interview with Dr. Baruch Cooperman, and Dr. Mu Autumn. You can find it earlier in your podcast feed.