A new Johns Hopkins Medicine study is adding evidence to a growing belief in healthcare technology: artificial intelligence can help identify patients who might otherwise fall through the cracks. But the findings also sharpen a harder question for employers, insurers and care leaders, which is whether better screening alone is enough to improve outcomes when follow-up care remains uneven.
Researchers at the Wilmer Eye Institute at Johns Hopkins Medicine reported this month that African American adults with diabetes were more likely to receive referrals for diabetic eye exams when an FDA-approved AI-assisted screening tool was used in primary care settings. The study, published April 13, 2026 in npj Digital Medicine, analyzed 3,745 adult patients with diabetes who underwent diabetic retinopathy evaluation between August 2020 and September 2022.
The referral difference was notable. According to the study, 64.9% of African American patients received an eye exam referral when the AI tool was used, compared with 44.4% through standard primary care provider referrals. The study also found that patients who chose the AI-assisted screening tool and attended their diabetic retinopathy evaluation were 15% more likely to be African American.
Those numbers matter because diabetic retinopathy remains one of the most serious and often silent complications of diabetes. Johns Hopkins describes it as the most common diabetes-associated eye disease and the leading cause of blindness globally, noting that patients may not notice symptoms early, which makes annual eye exams critical for timely diagnosis and treatment.
Still, the study stops short of declaring AI a complete answer to care inequity. Johns Hopkins explicitly noted that more work is needed to determine whether better access to AI screening ultimately leads to improved long-term vision outcomes. In other words, the technology may be improving the front end of the care pathway while leaving the back end (i.e., scheduling, transportation, insurance status, provider availability and out-of-pocket cost) largely unchanged.
That distinction is likely to matter as employers and benefit decision-makers weigh how AI should fit into preventive care strategies. Screening can create urgency and surface previously unmet need, but it does not automatically guarantee treatment. A patient may leave a primary care visit with clearer evidence that follow-up is necessary, yet still face practical barriers to completing an eye exam or accessing specialty care. The Johns Hopkins findings suggest AI can improve the precision and immediacy of referrals; they do not suggest that technology alone resolves the structural barriers surrounding care access.
That is where vision benefit design becomes part of the story. Vision Care Direct is a solution that positions itself as a doctor-led vision benefits alternative built around simple, flexible and affordable access to eye care. The company says its model serves individuals, employers, brokers and providers, and describes its plans as designed to reduce confusion and minimize surprise out-of-pocket expenses. Its employer and broker materials also frame vision coverage as a preventive-care benefit that can be bundled into broader workforce offerings.
In practical terms, that kind of structure points to the next phase of the AI-access debate. If AI screening identifies more at-risk patients, then networks, benefit design and affordability determine whether that added demand translates into treatment or simply a longer referral list. Vision Care Direct has also highlighted diabetic screenings as part of its preventive approach for employees, underscoring the idea that early detection is only as useful as the care pathway attached to it.
The equity concern is especially relevant for uninsured and underinsured patients. The Johns Hopkins study found that Medicaid coverage did not materially change attendance patterns between referrals generated by primary care providers and those generated by the AI tool. At the same time, the institution’s reporting makes clear that improved referral rates do not erase the “other obstacles” that can still prevent patients from completing screenings or specialist visits.
For employers and brokers, the broader takeaway may be that preventive-care access cannot be measured only by whether a screening tool works. It also has to be measured by what the patient can do next. If AI makes disease risk more visible, benefit leaders may face increased pressure to ensure provider access, manageable cost-sharing and easier navigation to follow-up care. Otherwise, technology could reveal disparities faster than the system is prepared to close them. That challenge becomes more urgent in eye care, where delayed intervention can mean irreversible vision loss.
The Johns Hopkins study offers a meaningful sign of progress: AI-assisted tools may help reduce a referral gap that has disproportionately affected African American adults with diabetes. But the deeper policy and business question is no longer just whether AI can find more patients in need. It is whether the healthcare system, including employer-sponsored vision benefits, is ready to meet them once it does.
