Podcast

Autonomous AI Increases Diabetic Retinopathy Screening Rates, Access For Young Patients

Risa M. Wolf MD 

Risa M. Wolf, MD, and her colleagues at Johns Hopkins University hypothesized that autonomous artificial intelligence diabetic eye exams at the point of care would increase screening completion rates and would help close the access gap in under-resourced communities for young patients compared with the current standard of care. While the results confirm this hypothesis, in this podcast, Dr Wolf discusses why the differences between the two study groups were still surprising, the next steps for future research, and more.

Additional Resource:
Wolf RM, Channa R, Liu TYA, et al. Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial. Nat Commun. 2024;15(1):421. doi:10.1038/s41467-023-44676-z


 

TRANSCRIPTION:

Anthony Calabro, MA: Hello everyone, and welcome to another installment of Podcast360, your go-to resource for medical news and clinical updates. I'm your moderator, Anthony Calabro with Consultant360, a multidisciplinary medical information network.

Researchers at Johns Hopkins University hypothesized that autonomous artificial intelligence diabetic eye exams at the point of care would increase screening completion rates and would help close the access gap in under-resourced communities for pediatric and adolescent patients compared with the current standard of care. While the results confirm this hypothesis, the differences between the two study groups were still surprising. How so?

Dr Risa M. Wolf is here to speak with us today about that, as well as the practical differences between autonomous AI diabetes eye exams at the point of care, and the current standard of care for pediatric and adolescent patients. Dr Wolf is a pediatric endocrinologist and the director of pediatric diabetes program at the Johns Hopkins Children's Center in Baltimore, Maryland. Thank you for joining us today, Dr. Wolf. I want to start with a simple question. What prompted this study?

Risa M. Wolf, MD: So, it's a great question, and we've actually done some work in in the area of diabetic retinopathy screening for several years. And some of our early data really from back in 2017, demonstrated to us that while it was a recommendation that children get screened for diabetic retinopathy, only about 50% of our patients were actually doing so. And it's really because in pediatric diabetes care, children are expected to see their endocrinologist every three months, so that's four times a year.

And the visits are not short because they see a nurse, diabetes educator, they see an endocrinologist, they often will see a nutritionist and a psychologist. So, the visits are long. Then asking them to also go see an ophthalmologist or optometrist for their diabetic eye exam was a little bit burdensome, and a lot of people were not actually doing it. So, we recognize that we had a care gap. And that's really what prompted us to move forward with this work, and using an artificial intelligence camera or a fundus photography camera with automated artificial intelligence on it to try to help improve that screening completion.

Anthony Calabro: What do we know about diagnostic AI systems? And what is it that we still need to find out?

Dr Wolf: So prior to using autonomous AI for diagnosing diabetic retinopathy, really for the last two decades, people have been doing point of care fundus photography, meaning that they're just taking a few pictures of the back of the eye of the retina, and then having them read in a deferred manner, usually in a remote manner, by an ophthalmologist or an optometrist, and then the patient gets the results in a deferred manner. So, really within like 3 days to 2 weeks. So that was being done, and there's some variability in the readings and where this is happening.

With the first approval by the FDA of an autonomous AI system, which was in April of 2018, interestingly at the time, we were actually getting ready to deploy a camera in our clinic because we had recognized this care gap for our patients. And I'll just mention that you know in Baltimore we do serve a very large minority and underserved community, and we had recognized that there were disparities also in who was getting screened, and we really wanted to close those care gaps. So when this Autonomous AI camera got FDA approved, and it's FDA approved for adult 22 and up, we thought this would be a really great innovative intervention to implement to help our patients get screened, and at the same time, to make sure that it works well in the pediatric population.

So at this time there actually are three autonomous AI fundus cameras that are FDA approved and that actually means that you have an operator who takes generally four pictures of the eye and the the AI algorithm directs you how to take them. It tells you if the images are sufficient for interpretation, and then you run it through the algorithm to get an autonomous or an automated response. So, there's really no oversight, but there does not need to be oversight by a physician or an ophthalmologist because the camera gives you a result in 30 seconds, and it says there is diabetic retinopathy or there is not. And then you can give that result immediately to the patient.

Anthony Calabro: That leads into my next question. Can you summarize the results of your study?

Dr Wolf: Yeah, so before I give the results of this study, I'll just mention that prior to doing this randomized control trial, we actually had done a prospective observational study in the first year that that we rolled out this camera. We enrolled 310 children and adolescents to do their eye imaging with the camera, and then we had the results read by ophthalmologists just to make sure that the sensitivity and specificity was adequate, and it was very similar to the adult data, so we felt very comfortable in using it, and we did use it at that time under our research program protocol. So, in that study, we actually found that we increased our screening completion rates from 49% to 95% in our population.

So, we knew that prospectively, you could deploy it in a sort of quality improvement way and actually improve screening. But, you know, we wanted to actually look further and say, does this really close care gaps, and does this improve screening as compared to the standard of care practice, which for forever has been you have diabetes, you have a risk for diabetic retinopathy, and can you please go make an appointment with an ophthalmologist or optometrist for a diabetic eye exam? And that's a dilated diabetic eye exam. So again, for our families, that's another visit to another doctor.

Usually your eyes are dilated, so it's a few hours out of school or work, and if you're a child, it's school for the child, work for the parent. So, the actual act of going to the extra visit is burdensome, costly, time-consuming.

So, being able to do it at the point of care in our first study, we knew that it increased it but then we really wanted to say, does this actually improve screening rates compared to the standard of care? We wanted to use a rigorous study design, which is why we did a randomized control trial.

Anthony Calabro: What were the main takeaways from your study results?

Dr Wolf: So, this was a randomized control trial. So, patients who enrolled had to meet inclusion criteria, which meant that they needed to have a diabetic eye exam and had not had one in more than 6 months.

And we enrolled 164 individuals and they got randomized to either the AI arm, which was to just have your point of care autonomous AI diabetic eye exam done right after your regular clinic visit, which took about 10 minutes total to do the consenting and the imaging, and you got a result right away. Or, if you got the standard of care group, if you are randomized to that arm, we gave you the specific recommendation that you should go see an eye doctor, being optometrist or an ophthalmologist. This is important because there's a risk of retinopathy, and then we would give them some local contact information of how to actually call and reach and make an appointment.

And what we found was that of all the people who went into the AI group, 100% of them completed their diabetic exam that day in clinic at the point of care. But in the standard of care group, by six months at the clinic, they were able to get an eye doctor, and they were able to get an eye and they were able to get an eye doctor, and they were able to get an eye doctor, follow-up, only 22% had actually gone to complete that diabetic eye exam. And we actually gave them a full 6 months and, you know, other studies had given 3 months and we felt, you know, let's give 6 months because they do go so infrequently and we always remind them at every visit. But even by 6 months, only 22% had actually gone for that diabetic eye exam.

Anthony Calabro: Even though your hypothesis was confirmed, did the study results still surprise you?

Dr Wolf: It was a bigger difference than we expected. We had based our sort of a priori sample size and power analysis off of demonstrating a smaller difference because we had taken the baseline at closer to 40% to 50%. So, it was a bigger difference than we expected to see.

We also looked at people who had not had eye exam results before versus the people that did, and we found disparities there as well and that people who had not ever had a prior diabetic eye exam were more likely to be minorities, have lower household income, lower parental education levels. So again, we knew that by doing this also we would be hopefully mitigating some disparities there as well.

Anthony Calabro: After the results of your study, what are the gaps that still remain?

Dr Wolf: So you know, I think we definitely, this was the first randomized control trial using an autonomous AI for diabetic eye exams. So you know, I think as the first study has shown that it does close care gaps, it does help people complete their diabetic eye exam at the point of care, and you know, a very large issue in the country, in the world is really access to actually doing that diabetic eye exam.

So, this really felt like we were bringing it you know to our patients, but I think some of the gaps that still exist and that we're actually looking at right now is you know we know to close this care gap. We know it helps people get their eye exam done, but can it actually mitigate the disparities that we know exist in the care that our patients receive and in their outcomes? And we based on a lot of data in the adult population of diabetes, we know that minority and underserved populations are less likely to get their diabetic eye exams done and more likely have diabetic retinopathy, which is an unfortunate outcome of having diabetes. And we recognize those same disparities in our earlier studies when we first started deploying the camera. So actually, we're conducting a prospective healthcare delivery trial right now where patients are, we're actually looking to demonstrate whether the AI camera can create equivalence in who gets screened and actually mitigate disparities in screening. And we actually are going to be completing that trial in the next 2 months. And then we should have results by the summer.

But you know, our hypothesis is that by doing this at the point of care and really helping make this more accessible and easier for people to complete that we can mitigate disparities in who gets screened, hopefully have earlier detection of any concerns regarding diabetic retinopathy and help those patients get to follow-up care and to an ophthalmologist for any treatment that's required.

Anthony Calabro: Is there anything about your study that we didn't get a chance to discuss?

Dr Wolf: As a pediatric endocrinologist, I think it is important for us to know that while we take care of children and adolescents and there may be low rates and prevalence of diabetic retinopathy in this population, recent large studies have shown that the prevalence is actually very significant after having a diabetes duration of 10 to 12 years, and we see that both for type 1 and type 2 diabetes. And I think, you know, while we may not see it in those pediatric age ranges, we must have it front of mind in that we don't want these children to grow up into young adults who have diabetic retinopathy and screening now is really important. Also, to get them sort of into the routine of doing this on a yearly or every 2-year basis as is needed and recommended.

Anthony Calabro: Thank you again for speaking with us today, Dr Wolf. For more digital health solutions and diabetes content, visit consultant360.com.


©2024 HMP Global. All Rights Reserved.
Any views and opinions expressed are those of the author(s) and/or participants and do not necessarily reflect the views, policy, or position of Consultant360 or HMP Global, their employees, and affiliates.