AI screening system increases adherence

(Image Credit: AdobeStock/Kundra)

Reviewed by Steven Yeh, MD, FASRS

Nebraska Medicine has brought artificial intelligence (AI) to the forefront as it uses an AI screening tool (EyeArt; Eyenuk) to provide better care for its patients with diabetes.

For many patients, annual examinations are key for the early detection of diabetic retinopathy and other eye diseases connected to diabetes. However, adherence to appointments proves to be a challenge, especially in the rural expanse of Nebraska, where one-third of its residents live in a rural community.

Annual eye examinations are crucial for early detection, but many patients miss them due to failure to adhere to appointments, lack of symptoms, or difficulty accessing eye care specialists who typically conduct the examinations. Access to ophthalmologists requires longer driving distances for the one-third of Nebraskans who live in rural counties.

The University of Nebraska Medical Center (UNMC) in Omaha and its clinical partner, Nebraska Medicine, provide access to more than 1400 physicians and nearly 70 specialty and primary care health centers, including the Truhlsen Eye Institute (TEI) in Omaha, Nebraska, a state-of-the-art facility that provides the latest in diagnosing and treating all eye care problems. After coming up short with previous efforts to improve access to vision screenings, Nebraska Medicine began to think about offering diabetic retinopathy screenings within its primary care clinics.

Using the AI screening tool, Nebraska Medicine has been able to conduct screenings for diabetic retinopathy in primary care clinics prior to referral to a specialist, thereby sharply increasing adherence to annual diabetic eye examinations and leading to expected increases in value-based reimbursement. In 2022, Nebraska Medicine teamed up with Eyenuk, developer of the EyeArt AI Eye Screening System, an autonomous AI solution cleared by the FDA for detection of more than mild and vision-threatening diabetic retinopathy. The cloud-based AI solution works in tandem with fundus cameras to capture retinal images, which are then automatically analyzed to detect the presence of disease.

Steven Yeh, MD, FASRS, director of retina and uveitis service at Nebraska Medicine, points out that the AI screening tool is easy to use and can turn out a report in less than 30 seconds without eye dilation for many patients. This has made it possible to screen for diabetic eye disease within the primary care clinics.

Yeh worked closely with Lindsie Buchholz, ambulatory clinical informatics lead, to implement the program. Nebraska Medicine purchased 2 devices and identified primary care clinic locations where the need was the greatest to ensure it would meet a need in the community, with the goal to expand to additional clinics in the future.

“Our primary care providers were actively involved and wanted to ensure they took the most appropriate steps when a patient did screen [with a] positive [result] during their visit,” Buchholz points out. “It was an important education point, and we worked with our nursing practice development specialists and TEI to create specific patient education on diabetic retinopathy that could easily be added to the patient instructions area of the EHR [electronic health record].”

Yeh says the system can help ensure that retina specialists provide the best possible outcomes for patients diagnosed with diabetic retinopathy. “This is a multidisciplinary approach, having the retina at the top of mind for screening, ultimately to ensure that we are able to detect disease before the patient loses vision. We have worked very closely with the primary care team and leadership to implement the system,” he explains.

The system includes a Topcon TRC-NW400 robotic camera, which photographs the retina in tandem with the EyeArt system, fully automating the focus and capture of high- resolution true color fundus photographs.

“The camera is so easy to use together with EyeArt that we could very easily onboard and train our clinical staff,” Buchholz says. “The process is so simple that even the person who’s answering the phone would be able to help screen a patient if needed.”

The system was developed using deep learning and uses multiple deep neural networks for specific classification tasks on images; the outcomes of these various networks are combined in a clinically aligned framework. It was trained on 375,000 images and then validated on 250,000 images.¹

The EyeArt system’s AI analysis is cloud based, with a user interface that is installed on the user’s computer. The system requires two 45° field-of-view images, with one centered on the optic disc and one centered on the macula, captured using a digital fundus camera. The images may be taken without dilation. The patient’s fundus photographs are uploaded to the cloud and are interpreted using the AI system within 60 seconds for the determination of more than mild diabetic retinopathy and vision-threatening diabetic retinopathy.

Nebraska Medicine also created new clinical workflows to clarify when screenings should be performed, who is responsible, and next steps following a positive screening result, Buchholz says. Importantly, Buchholz and her team have developed a process for ensuring that patients with positive screening results are scheduled for an ophthalmology visit at TEI.

“We told patients we found something of concern and had our referral staff ensure they got an appointment at TEI,” she says. “By doing the screening while the patient is here in the clinic and then making sure they see an ophthalmologist, we’ve increased [adherence] with the annual examinations and also gotten those at risk the care they needed in a quicker fashion.”

Within the first 3 months of introducing the EyeArt system, the 2 Nebraska Medicine clinics screened more than 350 patients and recorded a notable rise in adherence to annual diabetic eye examinations (a key measure for all value-based reimbursement programs). Adherence to the annual examination during that time increased by 13.6% in the UNMC Bellevue clinic and by 12% in the UNMC Durham clinic, a substantial increase in adherence given the volume of clinics at UNMC.

Moreover, the rise in adherence occurs without disrupting standard clinic procedures. Thanks to the workflow changes implemented by Buchholz’s team, each 6-minute screening integrates smoothly into daily operations. The EyeArt system’s easy-to-use interface and quick and accurate results allow the clinics’ certified nurse assistants to administer screenings while also reducing care bottlenecks by ensuring ophthalmologists see patients in need of further management and care (including treatment).

More than 40% of patients screened by the EyeArt system were identified as needing follow-up with an ophthalmologist. Those with more than mild diabetic retinopathy were referred to TEI, whereas patients with vision- threatening diabetic retinopathy received a priority referral to ensure their care was managed without delay. This proactive approach promises to enhance Nebraska Medicine’s value-driven rewards from health insurers, whose financial incentives are based in part on adherence to annual diabetic eye examinations.

Stephen Mohring, MD, medical director for patient-centered medical home and population health at Nebraska Medicine, says he is excited about the AI tool’s ability to improve care for patients with diabetes as well as improve reimbursement from its value-based contracts.

“Our physicians are wild about [EyeArt] because it removes the No. 1 barrier to treating patients with diabetes: getting them to have their eyes examined every year,” Mohring says. “It’s a prime example of how AI can narrow the referral pathway for patients who need care the most and be extremely helpful to medicine in general.”

Yeh describes the system as “an autonomous artificial intelligence platform for diabetic retinal screening with an algorithm that allows us to not only detect diabetic retinopathy but also stratify patients into what we consider more than mild or referrable diabetic retinopathy and severe or vision-threatening diabetic retinopathy. That allows us to triage patients in a timelier fashion for patients with more severe disease.”

This is where the cross-disciplinary collaboration begins, with Yeh and the team of physicians working with Buchholz and her team to consider the workflow design, data collection and storage, as well as the security aspects during the implementation process.

“We support the effort by ensuring strong imaging through the capabilities on their side and also a strong referral chain so that patients get the care that they need,” Yeh notes.

Yeh says as the system took flight, he was pleased to see how interested the clinical staff were in learning how to capture the image in a meaningful manner so that the AI could evaluate that image and give the appropriate results.

Buchholz notes that initially, the primary care clinical staff were hesitant, as they don’t have in-depth knowledge about eyes and what to look for.

“We worked with Dr Yeh’s team to provide in-depth education to the primary care staff about diabetic retinopathy and what it looks like in the eye,” she says. “Once they were able to complete the hands-on training, they quickly felt at ease with being able to capture quality images due to how simple the device is to operate.”

With the equipment in their hands, the staff response was positive, and by the time it was implemented, they were comfortable with the technology, having gained knowledge about the eye and now able to adjust the camera to obtain a clean picture for the AI to interpret.

Yeh points out that the AI system can also provide information that leads to an ungradable diagnosis, which would trigger a referral to have another ophthalmologist or optometrist look at the patient’s eyes. “One example of this is a dense cataract,” Yeh says. “If a patient has a dense cataract, it may come out as being ungradable. It isn’t a problem with the tool or the person who is doing the imaging. It is another disease process that has been identified, and we have seen this as well.”

The AI Eye Screening System is continuously advancing. Thanks to recent FDA clearances, it boasts enhanced image quality assessment and real-time image quality feedback, heightening its usability and achieving best-in-class image gradability. UNMC and Nebraska Medicine leadership is optimistic about future prospects. The impact it has had on patients has been witnessed by the UNMC staff.

“Many patients don’t want to do an extra test because they think it’s going to take a lot of time or they will have to get drops in their eyes to be dilated,” Buchholz points out. “With this, it is a simple picture of the eye and we get the results right away.”

If a result is ungradable as Yeh notes, a course of action can be determined much more quickly. He adds that it is a tool that helps retinal specialists improve the level of care for patients who may not have followed through with appointments otherwise.

“Preventive care is critical,” he says. “Being able to detect disease very early as a point-of-care diagnostic test is one of the strongest endorsements of this system.”

Mohring notes the vision is to incorporate the system in every Nebraska Medicine primary care clinic and share data statewide. “Witnessing EyeArt’s success in the clinics where it has been implemented so far, I believe other clinics will be inspired to embrace the technology,” he says.

As the collaboration between UNMC/Nebraska Medicine and Eyenuk continues, there are many opportunities to understand implementation of the AI technology, including direct benefits for patients with diabetes. These lessons learned and systems developed are applicable for health systems considering AI technology to enhance vision health and ultimately address disparities in access and care.

Reference

1. Lim JI, Regillo CD, Sadda SR, et al. Artificial intelligence detection of diabetic retinopathy: subgroup comparison of the EyeArt system with ophthalmologists’ dilated examinations. Ophthalmol Sci. 2022;3(1):100228. doi:10.1016/j.xops.2022.100228