News release

From: The Lancet

Screening with AI could half unnecessary glaucoma referrals, study suggests

A new AI-based screening tool reduces the number of unnecessary referrals for glaucoma by half, while maintaining a similar level of accuracy to eye doctor diagnoses, according to a paper published in The Lancet Primary Care journal.

Glaucoma is a leading cause of irreversible blindness worldwide and often goes undiagnosed until vision loss is advanced. Population-wide screening has long been considered impractical, but recent advances in AI may provide a more viable option.

The study was carried out at a single screening centre in Lisbon, Portugal in 2023, where 671 adults aged 55-65 were screened for glaucoma via the AI tool analysing images of the eyes. The images were also independently graded by six glaucoma experts. The AI-tool referred 66 people (9.8%), compared with 118 referrals (18.0%) by the eye doctors, with glaucoma being diagnosed in 40 participants (6.4%). The study found that the AI-based tool correctly identified 78% of people who truly had glaucoma (compared to 75% identified by the eye doctors) and correctly ruled out the disease in 95% of people who did not have it (compared to 91% by the eye doctors). The authors say that the high accuracy at excluding people without glaucoma is especially important, as false alarms can lead to unnecessary hospital visits, patient anxiety, and added strain on healthcare services.

The authors say AI-based screening could support earlier detection, reduce unnecessary specialist referrals, and help prevent avoidable vision loss, while being integrated into routine primary care in a cost-effective way. However, they highlight that because the study was conducted through an existing diabetic eye-screening programme, a large proportion of participants had diabetes, meaning the glaucoma rates observed may not reflect those in the general population.