https://www.ajmc.com/newsroom/artificial-intelligence-for-the-diagnosis-of-diabetic-retinopathy-an-interview-with-michael-d-abramoff-md-phd
Artificial Intelligence for the Diagnosis of Diabetic Retinopathy: An Interview With Michael D. Abramoff, MD, PhD


Michael D. Abramoff, MD, PhD, is the Robert C. Watzke, MD Professor of ophthalmology and visual sciences at the University of Iowa Carver College of Medicine. He is also a professor of electrical and computer engineering and a professor of biomedical engineering at the University of Iowa. Abramoff’s research interests include how autonomous artificial intelligence (AI) can be used to improve the lives of patients, with a focus on autonomous AI-based diagnostic and quantification algorithms for retinal disease. He is the founder and executive chairman of IDx Technologies, which developed IDx-DR, an FDA-authorized autonomous AI diagnostic system for the detection of diabetic retinopathy and macular edema.

An editor from The American Journal of Managed Care® recently conducted a question-and-answer session with Abramoff regarding the use of AI in the diagnosis of diabetic retinopathy.


The American Journal of Managed Care (AJMC®): What is diabetic retinopathy? How has it been traditionally diagnosed? At what point do patients usually present or get referred for evaluation, and is it too early or too late?
Michael D. Abramoff, MD, PhD: Diabetic retinopathy is a complication of diabetes and also the most important cause of blindness overall among all working-age Americans, not just individuals with diabetes. Traditionally, healthcare providers who take care of people who have diabetes are supposed to make sure that these [patients] get an eye exam every year. This requires a referral to an eye care specialist. Unfortunately, most patients do not follow through on these referrals and never see an eye care specialist: Just 15% of insured patients 10 to 64 years of age completed an annual eye exam, according to the results of a large, recently published study.1 Many of the noncompliant patients do not have diabetic retinopathy or diabetic macular edema and so do not need additional treatment. However, in some patients, diabetic retinopathy and/or diabetic macular edema is missed [because they do not get an annual eye exam]. They will make an appointment with the eye care specialist only when they have symptoms of vision loss, when it is usually too late to prevent permanent visual loss.

AJMC®: How important is timing of diagnosis of diabetic retinopathy? How is the condition managed or treated? Do delays in diagnosis affect management, prognosis, and/or outcomes?
Abramoff: Diagnosing diabetic retinopathy in its early stages is crucial. A wealth of scientific studies shows that early detection decreases costs and improves patient outcomes.2,3 Patients with early-stage diabetic retinopathy may not have symptoms, so they are often unaware that damage is being done to their eyes.

If diabetic retinopathy and/or diabetic macular edema is diagnosed, it can be managed with injections, implants, laser surgery, and microsurgery. Delays in diagnosis make the expected outcome much worse and management more expensive.

AJMC®: What prompted you to work on applying artificial intelligence (AI) to the diagnosis of diabetic retinopathy? What unmet need does IDx-DR address?
Abramoff: More than 20 years ago, when I was in training to become an ophthalmologist and later, a retina specialist, I noticed how much time I spent examining people with diabetes who did not have any signs of disease. At the same time, the people who had severe disease often arrived at my office far too late, resulting in needless vision loss. I had a background in neuroscience and computer engineering, so I decided to use AI to improve this process and make things more efficient. My early research focused on creating the computer algorithms to automate the process of diagnosing diabetic retinopathy, [to allow diagnosis] without human involvement. My goal was to mimic the cognitive processes of clinicians like me with these computer algorithms.

After many years of research, I founded my company in 2010 to translate this scientific research into clinical practice to benefit patients. I cannot think of anything more gratifying in my career than knowing this is in clinical deployment today, helping to prevent blindness in patients.

The primary need that IDx-DR addresses is better patient access to care, in addition to reducing cost and improving quality. Patients who face barriers to access can now be tested in convenient locations much more quickly than they could before. We now have a way to make sure patients who need to see retina specialists like me are identified earlier and referred on appropriately. 

AJMC®: How does IDx-DR work? What are its strengths and weaknesses? How can its clinical utility be increased?
Abramoff: The exam typically takes 5 to 15 minutes. Minimally trained operators use an easy-to-use robotic fundus camera to take pictures of the patient’s retinas, which are then analyzed by the autonomous AI’s algorithms for signs of diabetic retinopathy and diabetic macular edema. An immediate diagnostic report is produced at the point of care, allowing the physician to discuss the results with the patient while they are still in the office.

A significant strength of IDx-DR is its rigorous validation for safety, efficacy, and equity. IDx-DR is the only diagnostic technique for diabetic retinopathy and macular edema that has been validated in a preregistered clinical trial against clinical outcome. Preregistration is important because it eliminates many biases that easily creep in when doing studies of new technology and maximizes the validity and replicability of the study. We use surrogate outcome markers because diabetic retinopathy is a chronic disease and outcome matters to the patient and payor, rather than comparing the AI to clinicians, who have not been validated against patient outcome.
 
Another strength is that IDx-DR was designed using explicit detectors for biomarkers, which increases robustness and minimizes racial and ethnic biases. We built the algorithms to search for the same markers of disease that clinicians look for, such as hemorrhages, exudates, and neovascularization. The output of the AI is aligned with the clinical standard for diagnosis and treatment of diabetic retinopathy.

Additionally, with the use of AI, we are able to provide an instant diagnosis at the point of care. Clinical staff and patients do not have to wait for results, enabling more comprehensive care plan discussions while the patient is in the office and eliminating the need for callbacks. IDx-DR can be implemented at scale in existing clinical environments without needing additional staff or resources. The platforms provide a highly accurate diagnosis without human input and are easy enough to use that any office staff member can be trained.

IDx-DR does not replace a comprehensive eye exam; however, it is a powerful diagnostic tool that reaches patients who are not regularly seeing an eye care specialist. The system is currently limited to the detection of diabetic retinopathy, and we are actively working to add additional autonomous AI applications to the platform.

AJMC®: In what clinical settings and patient types is IDx-DR being used? What is the clinical impact of the introduction of IDx-DR?
Abramoff: IDx-DR is currently used to examine individuals with diabetes for diabetic retinopathy in a number of clinical settings: To date, many thousands of patients have been tested with IDx-DR, and hundreds who previously were undiagnosed have been identified with diabetic retinopathy and diabetic macular edema, at lower cost. Health systems are reporting greater efficiency for their screening programs and significant increases in compliance for the diabetic retinopathy exam, which is an important part of comprehensive diabetes care measures.

AJMC®: How has IDx-DR been received by health plans?
Abramoff: The majority of health plans have actively embraced IDx-DR as they see its potential to reduce costs through early detection. Customers are actively being reimbursed.

AJMC®: How and where do you see IDx-DR being used in the next 5 to 10 years?
Abramoff: IDx-DR has been embraced by the medical community and patient organizations, and is now part of the standard of care for patients with diabetes—the American Diabetes Association included FDA-authorized AI in its 2020 Standards of Medical Care in Diabetes.4 [Moving forward,] I see wider expansion of the use of IDx-DR in primary care clinics, as well as retail health, pharmacies, and lab settings. Because of their accessibility, these settings can provide added convenience for patients—when they go pick up their prescriptions or have blood drawn, they can also complete their diabetic retinopathy exam. Increase in access in these types of settings could have an enormous impact on population health.

AJMC®: Is your company working on other applications of AI?
Abramoff: We are developing additional autonomous AI diagnostic platforms for macular degeneration, glaucoma, and diseases outside the eye, such as otitis media. We are also exploring partnerships with other autonomous AI companies that use an ethics-based approach to AI algorithm development.

References
  1. Benoit SR, Swenor B, Geiss LS, Gregg EW, Saaddine JB. Eye care utilization among insured people with diabetes in the US, 2010-2014. Diabetes Care. 2019;42(3):427-433. doi:10.2337/dc18-0828.
  2. Bresnick GH, Mukamel DB, Dickinson JC, Cole DR. A screening approach to the surveillance of patients with diabetes for the presence of vision-threatening retinopathy. Ophthalmology. 2000;107(1):19-24. doi:10.1016/s0161-6420(99)00010-x.
  3. Early Treatment Diabetic Retinopathy Study Research Group. Early photocoagulation for diabetic retinopathy. ETDRS report number 9. Ophthalmology. 1991;98(5 suppl):766-785.
  4. American Diabetes Association. 11. Microvascular complications and foot care: Standards of Medical Care in Diabetes-2020. Diabetes Care. 2020;43(suppl 1):S135-S151. doi:10.2337/dc20-S011.
Print | AJMC Printing...