Ram Yalamanchili (00:02.83)

Hey, Dr. Barkat, how are you?

Mark Barakat (00:06.157)

Good, good, are you?

Ram Yalamanchili (00:07.638)

I'm doing good. Thanks for making time. So today I want to talk about the, I think an aspect which you and I spent a bunch of time talking about. know you're really excited about some of the future technologies coming out from AI and related technologies. So let's get started, right? So first off, tell me a bit about yourself and, you know, and your introduction to how you got into research, how you got into ophthalmology.

I'd love to hear a little bit more of your background.

Mark Barakat (00:38.863)

Yeah, sure. Thanks for having me. So yeah, I'm a founding director of Retina Macklin Institute of Arizona here in Enscosta, Arizona. you know, Retina specifically have been doing research for the better part of, probably 10, 15 years at this point. Clinical trials are very interesting to me. They're kind of stimulating, fascinating. It's a nice way of not making the same widget over and over and over and over.

gives us access and my patients access to newer cutting edge therapies and which kind of dovetails nicely into this conversation because it's all about getting access to cutting edge technology, cutting edge therapies and here we are. So, yeah.

Ram Yalamanchili (01:25.87)

Great. And just briefly, I know you mentioned that you have some background in computer science, which is kind of unique when we first spoke, but tell us more about that. Just from the background.

Mark Barakat (01:40.319)

I mean, so yes, absolutely. I come from a lot of math people. My dad is an engineer. His father was a PhD in mathematics. So, you know, I'm the kind of person that unfortunately, or fortunately, however, look at it, thinks of numbers. And so yes, computer science major in college, I found it fascinating. But at the same time, found that I enjoy interacting with patients as well. So I kind of found my way into medicine.

But you never lose that outlook, that way of thinking about problem solving, things like that.

Ram Yalamanchili (02:14.636)

Yeah, yeah, no, it's fascinating. That makes us two people coming from computer sciences into healthcare, but obviously you took the tough route. So, no, great. I'm really excited to hear about your thoughts on what's happening today, right? And where we are with AI. So maybe to start off, right? Tell us more about, know, where do you see AI's role in, you know,

Mark Barakat (02:20.782)

Haha.

Ram Yalamanchili (02:40.918)

generally maybe like applicable to the ophthalmology and the clinical trial space and then we'd love to hear more about your particular practice and yourself,

Mark Barakat (02:49.017)

No, no, sure. So, you know, I mentioned the reason why I like and what I actually love doing research. What I don't mention is all the other stuff that comes with it. Right. So there's there's there's a high burden that comes with, you know, obviously regulatory burden oversight in terms of data entry, you name it. I there are many, many different steps along the way to assure of the quality of the data that goes in and it's, it's, that's critical.

but that takes a lot of man hours and frankly repetitive tasks. So that's the underpinnings of getting to do all the fun stuff, all the cool stuff of bringing new therapies to patients. And so I think that is where there's a unique potential and possibility for artificial intelligence, right? Because artificial intelligence, as you know better than I do, helps to...

offset some of that burden, but also helps to basically collaborates with whoever's using it. makes it easier to do, makes it more complete, makes it more reliable, more accurate, has the potential to sort of revolutionize the underpinnings of research. So there's a lot of pain points here that can be addressed by AI.

Ram Yalamanchili (04:09.314)

And what do you say when people say, hey, research has been one of the slower parts of the industry, which in terms of technology adoption, right? I think there's even pre-AI, there's been other technologies which have kind of come in. Do you see anything different in terms of what's happening right now or do you have a different view on what's maybe about to happen?

Mark Barakat (04:31.001)

Well, you know, I haven't thought about that, but since you bring it up, I would say it makes, it makes some, it makes sense that in research, there's going to be a slower adoption curve. only because it's not just the head of clinics saying, I want to try a different software or something like that. There's, there's many, many different stakeholders here. Right? So there's the clinical trial site, there's the investigator, there's the coordinator team, there's the, central research organization.

There's the sponsor themselves. And of course, everything has to be audit ready if and when the FDA or the sponsor wants to look at all this stuff. So there's many, many different barriers, I guess you'd call it. But at the same time, for that very reason, there's a lot of opportunity. Because if you have to please that many stakeholders, that's kind of tough. That's kind of tough. And so I think that.

Ram Yalamanchili (05:26.691)

Yeah.

Mark Barakat (05:30.081)

is an opportunity.

Ram Yalamanchili (05:32.379)

And do you think that these sort of barriers have, you know, some ways curtailed the amount of research we could do? Do you see that at all in practice?

Mark Barakat (05:41.527)

No, I think so. think so. mean, you can do anything in a haphazard manner, but if you want to do something in a reliable and consistent manner, it takes a lot of time, a lot of man hours, a lot of effort. And of course, you're limited by that, right? So if I have the staff to run

five trials, by definition, I don't have time to run the six. And it may not be the amount of time it takes to actually see the subjects or the patients within those trials. It may also be all the back end stuff that needs to happen. So for every single visit you have, you have hours upon hours of data entry and query resolution and paperwork. And that's the stuff that

is a limiting factor in terms of how many subjects within a trial or for that matter how many trials you can accept at the site.

Ram Yalamanchili (06:44.35)

And have there been historically challenges around managing this status quo? One area I'm thinking about is what you just said, is how do you enable staff to do all this work in some kind of a scalable way? And do you have any insights around how that worked or how that worked out for you or places in the past?

Mark Barakat (07:06.735)

It's, mean, honestly, if you talk to other clinical trialists, that is one of the biggest problems that we have is finding and training good and great staff and then keeping them because there's a bottleneck there. There's only so, as you know, in any field, we talked about computer science, we talked about medicine, in any field, you're limited by the number of good people that you have. And I think that will always be the case, right? And so,

Really what you have here is kind like a force multiplier with AI. So you have good people, you want to keep training people. I'm not suggesting in the least that you want to replace good people, but you want to enable them to do what they're doing well at a faster, more reliable manner.

Ram Yalamanchili (07:52.738)

Yeah, and that kind of comes back to, I think something which gets a lot of sound bite lately, which is, you know, what is AI's place in society? And I don't know where you sit, but I'm more excited about the abundance which is coming. think there is an, like you said, it's a force multiplier for all the things we would want to do and couldn't do for all this while, right? And, you know, there's so many opportunities to do better research, more research, more drugs to market.

There's definitely an area where, I mean, I could definitely see the world turning towards a place where there's just a lot more opportunity than there is today. And we're able to manage that kind of opportunity. And today we're bottlenecked because of many, many other resources or whatever it might be, right? So it's interesting that you mentioned there's an aspect of training and then there's an aspect of retention. Why did you say that? I'm curious, you sort of brought up the retention and keeping them around.

And is there a reason why that's sort of a top of mind, guess?

Mark Barakat (08:55.151)

Quite simply, knowing how to work within the conference of clinical trial is a very specific skill set that's in high demand, right? So there are only so many people that are well-trained and are able to do that. And these people are being courted as they should be, are being courted by clinical sites. They're also being courted by CROs, possibly even by sponsors. And so they're

they're in high demand, right? And so you constantly have to try to replenish that pipeline as best as you can, but I don't have to tell you turnover is a killer. So much better to maintain the stuff that you have and enable them to be productive in the least amount of, you know, stressed environment as possible. And you know, sometimes anything can get hectic, right? And so you're trying to figure out, you know,

Staff Burnout is another thing, for example. We're seeing more and more trials coming out, which is, hey, that's fantastic. That's why we're doing it, right? We're getting these cool new mechanisms of actions. by the same time, the burden has begun to increase.

Ram Yalamanchili (10:09.282)

Right. You're talking about burden on the staff, on your staff.

Mark Barakat (10:12.655)

Well, the system as a whole, a system as a whole, right? A burden on the staff, but also sponsors are looking for more sites. There's more concurrent trials going on right now. So it's the entire system is trying to accommodate a higher demand than it appears was in the past, which is great. That means there's a lot of innovation in this field, which is fascinating. How do we accommodate that innovation? And how do we make sure that

We don't limit. Innovation should be limited only by your imagination, not by the logistics on the ground.

Ram Yalamanchili (10:50.542)

Makes sense. So that comes to one of the pointers. How did you end up saying, know, yes, we're in research. Yes, I have to be mindful of the technology I'm bringing in, but let's let's sort of try out something in the AI space. You you obviously are sort of an innovator yourself, but I'm curious, like how you think about adopting new technology, even potentially like, you know, why work with companies like ourselves in the space, right?

Mark Barakat (11:21.453)

I mean, ultimately, we'll come back to where we are right now, but I mean, ultimately, it's either you adapt and you adopt or you get left behind. And I think the writing is on the wall. AI already in other fields, even in day-to-day areas, AI has already made huge inroads in facilitating what we do on a daily basis. So I cannot imagine that

this little niche of clinical trial research is immune to that sort of thing. And then, going back to this particular case, as I mentioned, there's many things that I love about clinical trials. There's many, otherwise, why would you bother doing it? There's also many things that are quite painful. are pain points. I'm very lucky. have dedicated staff and they're amazing and I wouldn't be able to do what I do without them.

Um, but I also know that what they're doing is not very, very efficient. I mean, you have primary source and you have to document that. And then from the primary source, you have to put it in an electronic data capture. then, then afterwards you get, you know, 15,000 queries and you have to resolve those. It's, you know, we all know the principle of touch at once. This is, this is

touch it 15 times before any one particular data point is finally accepted. And if there is a possibility of reducing that, why wouldn't I try it? granted, there'll be learnings along the way. So there'll probably be some bumps in the road. But I'm not afraid of a few bumps in the road if I can be one of the first people to actually learn how to introduce that into clinical trials.

Ram Yalamanchili (13:14.882)

Right. And what was staff reaction like? I'm curious because you're obviously bringing something which is new and it's something people may or may not be okay with in terms of like the change, right? So how do you handle that and how do you handle that?

Mark Barakat (13:31.855)

I mean, it's a learning curve, right? So, you you have a conversation with a team saying, you know, initially, it may seem to you like it's more work, because what we're trying to do is we're trying to lay the foundation for something. But basically, we all need to get on board with here's the reason. I mean, do you love spending this much time sitting in front of your screen doing manual data entry?

Whereas you could be following up with patients or maybe setting them under the screen or doing something that's actually, essence, I feel like more productive or more valuable use of your time. And so as long as you sort of lay out the vision, I think we can all be on the same team.

Ram Yalamanchili (14:17.646)

sense. And in terms of just coming back to the type of AI or the type of technology we're like looking at, right? So I've sort of thought about it as there are certain types of problems you can solve today with AI and it's fairly high accuracy, high consistency. And then there's some which are, I would say, much higher value, but it might be a little bit further away because you might need more validation, you might need regulation, regulatory approvals, things like that.

So are you seeing things in the similar review? And what would some of these examples be if I were to ask specifically in the optimology space?

Mark Barakat (14:55.011)

Yeah, I mean, you're right. You have to learn how to walk before you can run. And so right now you look for the low hanging fruit. You look for things that are more actionable. So some of the things that I mentioned right now, right? So the filling out the forms, the 1572s, a killer is informed consent. The informed consents, they go through several iterations throughout the trial. And even though the subject may already be in the trial for months.

All of a they need to be reconsented and it's not rocket science, but you need to keep track of which form it is, because otherwise there's a deviation and the less deviations there are, the better trial we conduct and hopefully the cleaner our data is. there's, you know, DOA logs, for example, all those things. It basically boils down to, there's many actionable things that AI can help us with that are the important

but still minutia of the day-to-day aspects of the clinical trial. The other thing is, I cannot tell you how many times I've run back to find those little folders they have that print the protocol and find fonts that you need a magnifier to even be able to see because, well, what's the inclusion exclusion again? Or what's the exact rescue treatment criteria? And hey, listen, it works. Is it efficient though?

No, of course it's not efficient. all those things, I think would be squarely within the belly wig of something that we can do now. I think that's something that AI can address now and make things run more smoothly. So rather than me hitting the pause button saying, I don't know if I'm supposed to treat you or not, and then running to look at the form or frankly picking up the phone and calling the CRO or calling the sponsor, this is something that can be answered now. Now, you're absolutely right.

Ram Yalamanchili (16:30.542)

Mm-hmm.

Mark Barakat (16:51.855)

I mean, the sky's the limit. So would it be great to have AI search and find patients for me or be part of discussion about informed consent or all those things that we may have thought was science fiction just a couple of decades ago? I think they're feasible, but it's harder to do. let's, for the time being,

Ram Yalamanchili (17:18.242)

Mm-hmm.

Mark Barakat (17:20.889)

So the concept works. Let's see AI is a safe partner in the clinical trial space. And then once you've established that, I think you can slowly expand the functionality and usefulness.

Ram Yalamanchili (17:37.048)

So it's interesting how you stratified it, right? There is a certain set of tasks and workflows which are, I would say not entirely patient facing in how you describe it. It could be preparing your documentation, basically managing your regulatory overhead. Finance is another one we see quite often, We're just managing all the invoice bills, all the billing. There's a certain amount of like overhead there which requires a lot of time to do well.

Mark Barakat (17:56.067)

That's fair.

Ram Yalamanchili (18:06.766)

And then of course all the manual chores like data entry and double data capture, things like that. I certainly think there is a taller ask in terms of putting an AI in front of a patient and having it interact with the patient. I think that we're entering a very different territory if we go there. But somewhere in between, I don't know, is it? I'd love to know your...

Mark Barakat (18:26.799)

But wouldn't that be awesome? Wouldn't that be awesome? Well, I mean, that's thinking way, way ahead. But I mean, you're right. But right now you focus on the tasks that I would love to outsource if I could, but I can't, right? And then slowly, incrementally increase.

Ram Yalamanchili (18:50.126)

That's right, yeah. And what about things where, because especially retina tends to be very image heavy in terms of just the modalities involved. I think there's some exciting areas where AI can take us potentially. I'm just curious, have you seen anything interesting, anything you're excited about, and where are we on terms of timeline to be actually adopting these things?

Mark Barakat (19:10.255)

for sure yeah

I mean, no, no, absolutely. I mean, when it comes to that, when it looks to biomarkers on OCTs, for example, they've done some really good work on AI segmenting, like subartinal fluid versus interartinal fluid, or the presence or absence thereof. It's already being used in some trials, notably with Notilivision, have their own way of segmenting as well. And this is why it's used in the TRUKi trial, for example, following patients that are used first in the real world.

I know there's many, many other imaging centers across the country that are looking at this also with their own proprietary artificial intelligence modules. yeah, I mean, again, the sky's the limit. I'm sure as we speak, labs, multiple teams are looking at the use of AI. And for example, measurement, measuring geographic atrophy or predicting the progression thereof. And that's something that's really useful because we, you know, as a field,

we can tell you the averages and can tell you certain risk factors for progression. But, you know, when you see that one patient, that one subject in front of you, do you really know what the rate is going to be? No, that's why many times you see them back and even in some of these trials and you look for the progression, it sure would be nice if you had greater confidence, like an AM module telling you, well, this was most likely the rate of progression. And these are

the patients that are most likely to benefit from treatment versus not. So yes, I mean, we're definitely knocking on that door. It is very imaging heavy. so,

Ram Yalamanchili (20:45.89)

Mm-mm.

Ram Yalamanchili (20:54.146)

Got it. And are you seeing this in clinic as well? I mean, I know we spoke about in the clinical trial context, but have you adopted or have you like started to see practitioners bring this into your regular, know, standard of care practice?

Mark Barakat (21:08.441)

You know, I haven't seen too much of it in the clinic yet, at least AI in terms of imaging, but I wouldn't be surprised if it was slowly becoming introduced, right? So, I mean, for that matter, it is being introduced. Again, with the recent FDA approval of the Home OCT device, Home OCT device, cannot separate that from the AI module that's running in the background. So...

I guess I'm behind at times when I say, I haven't seen it in the clinic. It will be in the clinic.

Ram Yalamanchili (21:41.762)

So as making inroads, you think that'll happen, I guess it's happening right now in some ways in that case. Interesting. Great. So in terms of like, just fast forwarding this, Specifically talking about clinical trials. One area which I find myself thinking quite a bit is when we speak to sponsors, I mean, you're on the trial side, but speaking to sponsors, some of the common things they talk about is, hey,

Mark Barakat (21:46.095)

It is happening as I speak.

Ram Yalamanchili (22:10.95)

I have a tough time staying on track with my program. I can't find enough sites who are able to recruit, able to do the work. And it always seems very similar. All the problems seem very similar in terms of how they describe these challenges with the trials. Do you think that changes at some point? Where could this go and how could this change?

Mark Barakat (22:38.415)

You know, some of it will never change. If sponsors saying, boy, I wish we had more sites or more activity or faster recruitment, I don't think will ever change. A clinical trialist saying, boy, I wish my staff had more or less turnover. Some of that will never change, right? But I think when it comes to this in particular, what

what AI is uniquely positioned to do is help level the playing field. So trying to put on my sponsor hat right now, you've never been one, so I don't know if this is accurate or not, but I would assume that you have, let's say, 100 sites for trial of those 100 sites. I would say maybe 50 of them are really active. The other 50 may not be. So that's number one. Can AI help those other 50 sites be more productive?

The question is why are they not productive? Is it something that is manageable? Is it something that they're spending so much time on these other tasks that AI could take care of them? Who knows? That's one of the pain points. Then the other question is, of those sites that are active, what's the quality of data that you're getting? Because I can tell you this right now, it is going to be a wide range. You'll get some sites that

are very proficient and you get some sites that provide high quality data. And sometimes those two come in combinations and sometimes they don't. And so I would say it would behoove me as a sponsor to know throughout the trial that we're doing these quality assessments and quality assurances rather than at the tail end when let's say some of these patients at these sites are already finished and well,

what happened here, where's the documentation or these things are missing or that's missing, right? So I think to answer your question, if you ever hear a sponsor or clinical trial has tell you that something's perfect, they're not looking hard enough. There will always be some issue that needs solving, but I think some of those are solved.

Ram Yalamanchili (25:01.046)

Yeah. And you bring up an interesting point about sort of discrete versus continuous management rate of your site quality or data. I do see the argument for you have to do more, maybe like more frequent checks on what's happening in terms of your data coming in. But then right now that also directly correlates to the cost you have to invest into the trial. So you've got to make a balanced choice, I suppose. And I think that's one of the challenges for me.

many organizations, right? And especially if you're in some form of a scaled mode with lots of sites, lots of patients, then it becomes even more challenging. So these are like practical issues, I think, where I can see the scale which AI brings in could be another accelerant in terms of like what you can do in more real time than where we are today, right? So that's another exciting part of it.