Automated grading in the Diabetic Eye Screening Programme

Diabetes is a health condition which makes a person’s blood sugar level high. This can harm different body parts, including the retina located at the back of the eyes. This eye damage to the retina is called diabetic retinopathy. It can cause loss of eyesight if not found and treated early.

In the UK, people aged 12 and over with diabetes are invited for an eye screening test every 1-2 years. This test takes pictures of the back of the eyes. Trained staff, known as primary (or first) graders, look for changes or signs of damage in the pictures. Pictures that show eye damage or are hard to read go to a second grader for a second opinion. If the 2 graders do not agree, the picture goes to a third grader to make a final decision. Patients without eye damage and most with mild changes can return in 1 – 2 years for screening. Patients with worse damage may need more tests and treatment. The screening process takes a lot of staff time. As the number of people with diabetes gets bigger, so will the workload of staff and the costs.

An Automated Retinal Image Analysis System (ARIAS) is a type of computer programme. It can check eye pictures to find signs of eye disease using artificial intelligence. In some countries, ARIAS is already used to help with diabetic eye screening. In 2021, the UK National Screening Committee (UK NSC) looked at evidence about whether ARIAS could replace some human graders in UK Diabetic Eye Screening Programmes. They found there was not enough good evidence to recommend the use of ARIAS over.

The current review will look at the new evidence on:
• Whether ARIAS can find diabetic eye disease as well as humans
• How ARIAS affect patients, staff, and the screening programme
• Whether replacing human primary graders with ARIAS is good value for money

We will use the UK National Screening Committee’s approach to carrying out Evidence Summaries.