The Illinois Eye and Ear/
UIC Health Experience
The University of Illinois Hospital & Health System (UI Health) is a safety net serving a community that primarily identifies as racial and ethnic minorities and has greater rates of unemployment, uninsured individuals, and poverty than the average in Illinois and the U.S. Diabetes affects 12.4% of adult Chicagoans, mirroring the national prevalence of 11.3%. In the areas surrounding UI Health, however, the prevalence of diabetes has been found to be over 30% higher compared to citywide estimates. To address the growing need for diabetic screening and identifying patients who are at risk of developing diabetic retinopathy, we developed a partnership between the Department of Family & Community Medicine (FM), the Division of Endocrinology, the Department of Ophthalmology and Visual Sciences (OVS), and the Institute for Healthcare Delivery Design at UI Health to establish a care delivery model to integrate an AI-based screening program, using an autonomous AI system, into the primary care setting.9 To further validate results and identify pathology beyond diabetic retinopathy, we also have an eye care clinician grade images through a store-and-forward telemedicine approach.
Illinois Eye and Ear/UIC Health:
Screening Program
Goal – The primary goal is to prevent vision loss from diabetic retinopathy and increase the number of patients with diabetes who are screened for DR. We also want to develop better processes for making appointments for patients who screen positive for DR into the ophthalmology clinical practice.
Equipment – We are using the EyeArt AI (EyeNuk, Inc., Woodland Hills, CA) system with the Topcon camera. We are also implementing additional units with the Canon imaging hardware. The cameras were purchased using support provided by the Department of Ophthalmology and Visual Sciences. In our model, the Department of Ophthalmology and Visual Sciences incurred the upfront cost of the hardware.
Screening Exam – Our initial pilot of using Eyenuk’s EyeArt and EyeScreen software to perform AI diabetic retinopathy screening was with our partners in the Department of Family and Community Medicine. We employed Human-Centered Design (HCD), or “design thinking,” to help us understand real-world context and behaviors of individuals, engage stakeholders, and rapidly prototype and test solutions to any issues of implementation.11 We felt that it would be extremely important to identify all stakeholders in the implementation of the DR AI screening workflow, and human-centered design offers a methodology that allows us to potentially better explore solutions that account for stakeholder concerns. Figure 5 shows our key stakeholder map and what each group felt defined the success of the program.
Imaging is typically performed by a trained medical assistant during intake before seeing the clinician in the Family Medicine clinic. Imaging is occasionally postponed to after the visit with the clinician, pending time constraints. Of note, it was a critical discussion with the primary care team to explain that the screening performed with the AI system does not substitute for a
complete eye exam. After obtaining input from stakeholders, handouts were also created to provide additional information to patients during their visit (Figure 6).
How Patients Access the Camera – Patients are identified by the primary care physician and clinicians as having diabetes, and whether or not they have had an eye exam performed either at UI Health or outside the UI Health system within the prior 12 months. If patients are due for a screening, they have testing performed during their routine primary care exam. To help identify patients who need screening, we have worked with the information systems team to create a best practice advisory (BPA) that would appear for patients due for screening during rooming. This would prompt the medical assistant (MA) to confirm that the patient hasn't had an outside dilated eye exam in the past year. Once confirmed, the orders for both the image and referral to ophthalmology appear for the MA to send to the clinician so that they can then order.
Interpreting Screening Results – The results of the AI screening are reviewed by the primary care physician, and the results are provided to the patient at the time of their visit. In addition, an eye care clinician reviews the images and the results of the screening. Diagnosis is based on image review by the eye care clinician, and a workflow was established on how to communicate these results as described below.
Communicating Results to Patients and Clinicians – The results from the AI system are communicated to the patient on the day of the exam. If DR is detected, an appointment for a comprehensive eye examination is made by the MA for that patient prior to leaving the primary care office. The images are also reviewed by an eye care clinician in the Department of Ophthalmology and Visual Sciences within a week of the exam. If any additional findings are noted, then the patient is contacted by the Department of Ophthalmology and Visual Sciences, and the primary care physician is also notified. In addition to being provided with the results of the diabetic screening exam, patients are also provided with educational materials (Figure 6) regarding diabetic retinopathy.
This material is given to patients whether or not they undergo screening. We felt that the education of patients was paramount for helping them understand why comprehensive eye exams are required to maintain their vision health.
Billing – In our current model, billing is performed by the primary care department, and we, the Department of Ophthalmology and Visual Sciences, do not bill for the readings performed by our eye care clinicians.
Scheduling Follow-up Care – Scheduling and ensuring follow-up for care was one of the most significant changes implemented in our program. Prior to the implementation of AI screening, appointments for an ophthalmology exam were the responsibility of the patient. We felt it was critically important that patients leave FM with an ophthalmology appointment. Therefore, we agreed to shift the responsibility of scheduling from the patient to the medical assistant in FM. We believe that tying the scheduling process to Ophthalmology with the check-out process in the primary care clinic not only removes potential barriers to setting up an appointment by the patient but also allows patients to manage their health at a time when they are most prepared to do so. Through a collaborative effort, we developed the workflow seen in Figure 7.
This workflow shows additional details based on when the photo is taken, who can provide results to the patient (based on changes made in the design phase), and shows prioritization of patients with positive results to ensure timely follow-up within the ophthalmology clinic.
Measuring Screening Outcomes – We are still in the process of evaluating and fine-tuning the process of our AI screening program. We continue to get feedback from end-users of the program and will measure the effectiveness of our implementation through surveys to capture the following implementation outcomes: acceptability, adoption, appropriateness, feasibility, and adaptation. We will also look at the number of post-implementation referrals, appointments scheduled, and appointments completed. Continuing education for imagers is also an important consideration, as infrastructure for training medical assistants or other staff needs to be in place. We are also routinely reviewing the image quality of the images taken to identify opportunities for improvement. Additional data may be collected from surveys of patients, staff, and clinicians. We are also exploring the cost-effectiveness of this program and working with the health system on developing a financially sustainable model that will enhance the care of our patients and improve access to care for the most underserved.