Every year, the NHS takes millions of retinal photographs from people with diabetes to catch the earliest signs of eye disease—but each image must be checked by a human grader, a slow and expensive process. This review asks whether artificial intelligence can safely take over that first look, flagging only the suspicious images for a human to examine. The current screening pathway is labour-intensive and costly, and as diabetes rates rise, the pressure on the system will only grow. If automated grading proves accurate enough, it could triage the majority of normal images without human involvement, freeing up specialist graders to focus on the cases that need urgent attention. The researchers will assess not just diagnostic accuracy but also the wider clinical impact and cost-effectiveness for the UK screening programme. This is a practical, policy-focused review—success would mean a faster, cheaper screening pathway that maintains safety while reducing strain on NHS eye services.
View original technical description
Background: Diabetic retinopathy (DR) is one of the most common complications of diabetes. Screening for DR aims to detect DR during the asymptomatic early stages, enabling timely treatment to reduce the risk of vision loss. However, diabetic eye screening programmes (DESPs) are costly and labour-intensive. Objectives: To update the previous UK National Screening Committee’s external review regarding the accuracy of automated retinal image grading systems (ARIASs) at detecting DR in patients with diabetes as well as the wider clinical impact and cost-effectiveness of replacing the primary human grader in the current screening pathway with ARIAS to triage patients prior to human grading. Design: Enhanced rapid evidence review. Data sources: MEDLINE, EMBASE, the Cochrane Library and ICTRP databases from 1st June 2020 to 3rd October 2025. Eligibility criteria: Accuracy studies will be included if they reported external validation of ARIAS in diabetes patients =12 years of age who underwent standard fundus photography to detect DR. For the wider clinical impact, we will include studies that evaluated DESPs with the use of ARIAS for primary grading compared with DESPs with fully manual grading and reported the wider effects for patients, health professionals, or the screening programme. For cost-effectiveness, we will include UK-based economic evaluations and reviews of these, comparing DESPs with the use of ARIAS for primary grading to DESPs with fully manual grading. For all three key questions, we will exclude articles not available in the English language, published before June 2020 as well as conference abstracts. Review methods: Title and abstract screening, full-text assessment and data extraction will be undertaken by a single reviewer, with 20% checked by a second reviewer. Quality appraisal will be conducted by one reviewer, with a random 20% independently assessed by a second reviewer. Studies on accuracy and the wider clinical impact will be prioritised for data extraction, quality appraisal and evidence synthesis based on pre-specified criteria.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know