Keeping an Eye on Diabetes

When the NHS Diabetic Eye Screening Program (DESP) was established in England in 2003, the ambition was bold: to systematically identify and treat sight-threatening diabetic retinopathy (DR) before irreversible damage occurred. Two decades on, that ambition has largely been realized and the World Health Organization (WHO) formally recognizes DESP as a public health triumph (1). Despite this, the program now faces a new set of challenges that will define its next chapter.

The program offers annual screening to all individuals with diabetes mellitus in England aged 12 years and over, with pathways including routine digital screening (RDS), digital surveillance (DS), and slit lamp bio-microscopy (SLB) (2). Comparable national programs operate across Scotland, Wales, and Northern Ireland, each with their own grading infrastructures but sharing the same overarching public health mission.

The rationale for population-level screening is well-established. Diabetic eye disease remains one of the most common causes of certified blindness in people of working age in the UK, and a significant proportion of vision loss is avoidable through early detection and timely treatment (3).

What the data tell us

The headline outcome of DESP is striking. In the years 2009–2010, for the first time in almost five decades DR was no longer the leading cause of blindness amongst the working age group in England – and DESP is considered a major contributor to this shift (3). In England, despite an overall increase in the total number of blindness certifications, those attributable to diabetic eye disease fell from 1,334 (5.5% of all certifications) in 2009/10 to 840 (3.5%) in 2018/19 (3).

Screening uptake has remained consistently high: in 2017–18, 2.70 million people with diabetes were offered screening, of whom 2.23 million attended (82.7%), generating 8,782 urgent referrals and 54,893 routine referrals to ophthalmology departments (3).

Current challenges: scale, capacity, and equity

Despite boasting high uptake since implementation, a lack of uniformity is present across the population. Of those newly registered in the RDS pathway, 80% attended within the first 12 months – but 20% remained unscreened after three years. Time from registration to first screen was notably longer for people aged 18–34 (3). This persistent gap in younger adults remains an unresolved challenge.

The number of people diagnosed with diabetes, particularly type 2 diabetes, is predicted to reach record levels over the next decade. Coupled with rising costs of the technology and workforce required to deliver screening, this will make sustaining the long-term cost-effectiveness of the DESP increasingly difficult. The English DESP already generates almost 13 million retinal images annually (4), each requiring human grading for diabetic retinopathy. As diabetes prevalence continues to rise, this creates a major challenge for healthcare providers. The pressure also extends beyond screening, with increasing demands on hospital eye services (HES) for intravitreal anti-VEGF injections, steroid therapies, laser treatment, and vitrectomy further adding to the overall burden on the system.

Equity remains a significant concern, particularly with the move to risk-stratified screening intervals. Risk factors for progression of DR include but are not limited to duration of diabetes, diabetes type, suboptimal glycemic control, ethnicity, and elevated blood pressure. Research from North East London found that a blanket two-yearly recall for low-risk patients may not account for the heterogeneity of that population, with some subgroups – particularly those from South Asian and Black ethnic backgrounds – carrying disproportionately higher risk of progression (5).

Recent program reforms

In direct response to capacity and sustainability pressures, the DESP has undergone significant structural changes. Following the most recent upgrade of the diabetic eye screening software in 2024, across England DESP introduced a 2-yearly screening interval for low-risk groups who have had a negative screen (R0M0) on two consecutive occasions. The rationale was based on an audit of nearly 350,000 patients from seven geographically dispersed regions and has been demonstrated to save £404,000 per QALY (quality-adjusted life year) (6,7).

Recently, the implementation of OCT in DS clinics has enabled precise disease stratification, preventing roughly 120,000 unnecessary HES appointments annually and saving £76 per patient (8). October 2024 also brought updated pathway standards from NHS England, including revised performance metrics for referral timescales and new coverage standards within digital surveillance (2). These reforms are intended to improve system-wide efficiency while maintaining safety – though definitive performance thresholds will not be set until sufficient post-reform data has been collected, expected by Q3 2026.

The road ahead: AI and automated grading

The most transformative development on the horizon is the integration of artificial intelligence-based automated retinal image analysis systems (ARIAS). In the UK, Scotland has used a rules-based autograder since 2011, but diabetic eye screening in the rest of the UK still relies solely on human graders [9].

A landmark 2025 evaluation from Moorfields, City St George’s (University of London), and Homerton Healthcare NHS Trust – analyzing 1.2 million retinal images from the ethnically diverse North East London DESP – demonstrated that algorithm performance ranged from 83.7–98.7% for detecting diabetic eye disease potentially requiring clinical intervention, with accuracy exceeding 96% for moderate-to-severe disease and 95% for proliferative disease (5). Crucially, the algorithms performed consistently across ethnic groups – a first in large-scale UK evaluation.

Despite this evidence, formal implementation is pending. The UK NSC concluded in 2021 that further research was needed before ARIAS could be introduced, citing limitations around real-world health outcomes and cost-effectiveness data. That evidence base has now substantially matured. The two use-cases considered most promising are using ARIAS to replace primary graders, or positioning it as a filter before human screening – potentially triaging the roughly 90% of low-risk encounters away from human review entirely [9].

Workforce anxieties are real and should not be underestimated. Healthcare professionals have raised concerns about job security, resource allocation, benchmarking against AI, and the need for further training and career development (10). Both clinicians and patients favor a hybrid model in which AI handles initial image analysis while humans retain responsibility for complex or borderline cases. Building that trust – through transparent validation, clear governance frameworks, and workforce planning – will be as important as the technology itself.

Looking forward

The UK’s DESP stands as one of the most successful public health screening programs implemented. The reduction in diabetic blindness over two decades is a testament to consistent investment, rigorous quality assurance, and – above all – high population uptake. However, the program is at a crossroads. Rising diabetes prevalence, capacity constraints in both screening and HES, and the emergence of AI tools all demand a coherent strategic response with appropriate regulatory measures. The biennial interval for low-risk patients is a pragmatic step but must be implemented with equity safeguards that account for differential risk across demographic groups. The case for ARIAS deployment in England is now compelling – the evidence exists; what remains is appropriate regulatory frameworks coupled with systematic implementation.

For ophthalmologists, the coming years will require active engagement with program governance, AI validation methodology, and the patient communication strategies that will determine whether these advances translate into equitable sight preservation at scale.

Take home messages

• DESP has been a landmark public health success: diabetic retinopathy is no longer the leading cause of blindness in working-age adults in England, a transformation largely attributable to systematic screening since 2003.

• Uptake exceeds 80%, but persistent gaps remain – particularly in adults aged 18–34 and certain ethnic minority groups – and closing these must remain a program priority.

• The 2024 move to biennial screening for low-risk patients is pragmatic but must be implemented with equity safeguards accounting for differential progression risk across demographic subgroups.

• The evidence base for AI-powered automated retinal image analysis is now compelling – validated at scale in diverse UK populations with accuracy exceeding 95% for sight-threatening disease.

• A hybrid human–AI model is both technically feasible and preferred by clinicians and patients however requires clear governance frameworks and proactive workforce planning.

• Capacity pressures extend beyond screening: rising demand for intravitreal therapy, laser treatment, and vitrectomy means Hospital Eye Services must be part of any long-term sustainability strategy.