The UK’s Evidence Framework Sets a Global Example
Artificial intelligence (AI) in diabetic eye screening is gaining momentum, promising to boost efficiency and reduce preventable sight loss. But how can we ensure that the technology delivers not just clinical promise, but population-level impact?
A recent Health Policy article published in The Lancet Digital Health by Macdonald et al. presents a robust roadmap developed by the UK National Screening Committee (NSC). This framework sets the standards for integrating machine learning-based automated retinal imaging analysis software (ML-ARIAS) into England’s diabetic eye screening programme (DESP).
The UK DESP is among the most advanced globally and has significantly reduced diabetes-related vision loss. However, with diabetes cases rising and workforce capacity strained, there’s increasing interest in using AI to streamline or augment the screening process.
AI in Diabetic Eye Screening: Two Promising Use Cases and a Core Principle
ML-ARIAS could support diabetic eye screening in two main ways: first, by automatically filtering out low-risk cases before human review; and second, by replacing the primary human grader. While both approaches offer efficiency gains, the NSC emphasizes that any implementation must be grounded in high-quality evidence, covering accuracy, safety, practical feasibility, and ethical integrity.
Building the Evidence: From Lab Success to Real-World Impact
Although many AI systems show strong performance in lab settings, the NSC stresses that these tools must be evaluated within the full screening pathway. It’s not enough to demonstrate standalone accuracy; real-world implementation must not compromise screening specificity or lead to excessive false positives that burden hospital eye services.
To address this, the authors recommend a phased strategy for generating evidence on AI in diabetic eye screening:
- Retrospective accuracy studies using UK-relevant screening data
- Prospective non-interventional trials to test feasibility within workflows
- Randomized interventional trials to evaluate real-world impact on clinical outcomes and resource use
So far, researchers have rigorously evaluated only three systems in UK settings—EyeArt v2.1, RetmarkerSR, and iGradingM. Independent, head-to-head comparisons across vendors are encouraged to establish fair benchmarks and ensure objectivity.
Equity, Trust, and Ethics: Critical to AI Screening Success
The NSC’s framework goes beyond performance metrics. Social and ethical implications—including algorithmic bias, population equity, and public trust—are central to the evaluation process. Ongoing UK-based studies are testing ML-ARIAS across diverse ethnic groups, addressing historical underrepresentation in ophthalmic datasets.
The authors also highlight transparency, interoperability, and the adoption of open APIs as vital to successful AI deployment. Vendor-neutral standards will help minimize integration barriers and enable scalable, flexible adoption across different health systems.
What’s Next for AI in Diabetic Eye Screening?
The UK NSC isn’t pushing for fast adoption—it’s advocating for thoughtful, evidence-based transformation. This measured approach ensures that any AI integration into diabetic eye screening maximizes benefit, minimizes harm, and maintains public trust.
As AI technologies become more embedded in healthcare, frameworks like this one are essential. They shift the conversation from “Can it work?” to “Will it work—safely, fairly, and at scale?” For digital health innovators and policymakers worldwide, the UK’s approach offers a valuable model for responsible AI integration in population health programmes.
Full source: Macdonald T, Zhelev Z, Liu X, et al. Generating evidence to support the role of AI in diabetic eye screening: considerations from the UK National Screening Committee. Lancet Digit Health. 2025;7:100840. https://doi.org/10.1016/j.landig.2024.12.004
