Medicare Pays For the AI Eye Scan — Nobody Pays For What Comes Next

MedCity News

Source published: Site published:

Original article

In U.S. programs that screen for diabetic eye problems in primary care clinics, nearly half of the patients who test positive never see an eye doctor within a year. Taking the picture was the easy part.

This problem matters even more now that computers can read the pictures automatically. In 2018, the Food and Drug Administration approved the first AI system that can look at a picture of the back of the eye and decide, without a doctor, whether a person with diabetes needs to see an eye specialist. Four systems have now been approved. Medicare started paying for these AI exams in 2021 and set a national payment rate in 2022. The main quality measure used to grade insurers counts an AI exam. The American Diabetes Association's 2026 Standards of Care also say AI screening is appropriate, as long as patients who need more care can get it.

But very few people are using it. A study in JAMA Ophthalmology looked at 62 health systems and found that fewer than 1 in 1,000 patients with diabetes had an AI eye exam between 2021 and 2023. Use barely increased during that time. Medicare's own data show that use grew from almost nothing: about 5,000 people in traditional Medicare had an AI eye exam billed by a doctor's office in 2024, compared with about 400 in 2021. Even so, that is fewer than 1 in 1,500 Medicare patients with diabetes.

These numbers do not include exams at hospitals, community health centers, or private Medicare Advantage plans. They also leave out programs that screen patients but do not bill Medicare. The numbers also show where the technology is being used: in 2023, optometrists billed Medicare for most AI eye exams, rather than the primary care clinics the tools were designed to help.

Low use of AI does not mean that few people are being screened. Many patients still get regular eye exams. The bigger problem is that some people are never screened, while others test positive but do not get further care. Diabetic retinopathy is damage to the retina caused by diabetes. It affects an estimated 9.6 million Americans and is the leading cause of blindness among working-age adults.

Medicare pays for the picture, not the outcome

In 2026, Medicare pays $46.76 for an AI eye exam in a doctor's office and $60.27 in a hospital outpatient clinic. These rates are listed in Medicare's national payment schedule. The payment does not include a doctor's time, which makes sense because a doctor does not read the picture. But it also leaves out several other important tasks.

It pays for taking the picture. It does not pay for the medical assistant who takes it, helping a patient whose picture is too blurry to read, adding the result to the medical record, or calling patients until those with positive results get an appointment at an eye clinic. Medicare does pay for some care coordination, but none of that payment is linked to making sure patients with positive results actually see an eye doctor. People in traditional Medicare also usually pay 20% of the cost after meeting their yearly deductible. That works out to about $9.

Payment is not the only problem, but it affects all the other problems.

Taking the picture is the easy part

Everyday practical problems can stop these programs from working. Five university medical programs reported that only 49% to 75% of patients produced pictures clear enough to read when eye drops were not used to widen their pupils. At one university medical center, only 31% of patients aged 70 and older produced a readable picture. Using the drops raised that number to 85%. Few primary care clinics use these drops. As a result, most send patients with unclear pictures for a full eye exam anyway. For these patients, AI does not save any time or effort. Staff leaving and being replaced, finding an empty room, and entering results into medical records add more work. Many clinics decide that the program is not worth the trouble.

Then there is the next step, which nobody is specifically paid to complete. Two U.S. screening programs that sent eye pictures elsewhere to be read, one in Houston and one in New York, found that only 55% to 59% of patients with positive results went on to see an eye specialist. In a county hospital system, about half of the patients with disease that could threaten their sight saw a specialist within a year. Only 5% were seen within the recommended time. There are not enough specialists, either. Federal estimates show that the number of eye doctors called ophthalmologists will fall by 12% by 2035, while demand will rise by 24%. Rural areas may have less than one-third of the eye doctors they need. Sending patients with unclear pictures or less serious problems to optometrists could reduce the pressure on specialists. But patients with serious eye disease still need an ophthalmologist.

The quality measure does not solve this problem. The main measure used to grade insurers also affects the star ratings Medicare gives private health plans. It gives full credit when a patient completes a screening, no matter what the result is. A patient who tests positive but never sees an eye doctor gets the same score as someone whose result shows no problem.

AI works better when someone is responsible for making sure patients get care afterward. A 2026 review of six studies found that AI screening nearly doubled the share of patients who followed up with an eye doctor. The biggest improvements happened when programs also changed the way referrals were handled.

Three ways to fix the problem

1. Pay for follow-up care, not just the picture. Medicare should allow its existing care-coordination payments to cover the work of contacting patients with positive results until they see an eye doctor. It should also test a payment that is made only after the patient has attended that appointment. Private Medicare Advantage plans can already do this and can make the screening free for patients.

2. Measure whether patients get follow-up care, not just whether they get screened. Insurers are judged on whether patients complete screenings, not on whether patients with positive results receive care. This year, Medicare kept the eye-exam measure and said it would consider adding a measure for follow-up care. It should add one now. Creating new quality measures takes years, so the work should start as soon as possible.

3. Make someone responsible for every positive result. Today, no one is clearly responsible for making sure a patient with a positive result gets care. The insurer or medical group paid to keep that patient healthy should take responsibility until the patient sees an eye doctor. Some changes do not need new rules. Clinics can make the eye appointment before the patient leaves and require the specialist to send the results back. England's national program follows each patient from the first invitation through referral. It screens about 79% of the people it invites. Even this program has trouble getting patients with urgent needs seen on time.

The next wave of AI will face the same problem

AI eye screening for diabetes is the first AI tool that can work without a doctor and has received an official billing code, a quality measure and support in medical guidelines. That makes it an early example of what may happen with other AI tools.

Medicare's proposed 2027 rule for hospital outpatient care would put this tool and dozens of other software tools into a new payment category. Medicare describes this as a temporary step toward payment methods that 'may better align payment with clinical outcomes.' In simple terms, that means paying for results.

That is the right goal. For patients, the important question is not whether software gets paid. It is whether the people the software identifies as needing help actually receive care. If Medicare keeps paying for the picture instead of the result, the next wave of AI will face the same problem: the tools will be approved and paid for, but very few people will use them.

Photo: Jay_Zynism, Getty Images

Dr. Ashkan Eliasy is the Managing Director of Nuvisy Strategic Advisory. He advises medical technology and AI healthcare companies and investors. Before this, he worked as an investment principal at a global company that makes eye care devices. He also helped start medical technology and financial technology companies. He has a PhD and an MBA. Disclosure: He advises companies that work in eye care technology.

This article was published through the MedCity Influencers program. Anyone can share their views on healthcare business and innovation through MedCity News and its MedCity Influencers program. Click here to learn how.