TOPLINE:

AI-based glaucoma screening integrated into primary care within an existing screening programme for diabetic retinopathy demonstrated a sensitivity of 78%, a specificity of 95%, and a positive predictive value of 53% in real-world settings. The AI-based approach led to 10% referrals vs 18% by experts.

METHODOLOGY:

• Researchers conducted a cross-sectional study to evaluate the feasibility, diagnostic performance, and cost-effectiveness of integrating AI-based glaucoma screening into an existing screening programme for diabetic retinopathy within primary care in Lisbon, Portugal.
• They used baseline data from a single primary care screening facility (March to December 2023) and included 671 adults aged 55-65 years (median age, 61.3 years) with and without diabetes.
• Screening procedures used non-mydriatic fundus photography, with the images analysed using an AI glaucoma risk model (MONA GLC), and intraocular pressure (IOP) was measured through rebound tonometry; referrals were triggered by an AI risk score threshold of 0.73 or an IOP ≥ 24 mm Hg.
• Referred participants underwent specialist evaluation with slit lamp exam, standard automated perimetry, and optical coherence tomography and diagnostic adjudication using adapted Thessaloniki Eye Study criteria; six glaucoma experts independently graded images with adjudication.
• The primary outcome was the glaucoma referral rate in the screened population by the AI-based pathway and the adjudicated expert grading pathway.

TAKEAWAY:

• The AI pathway referred 66 (10%) participants, of whom 88% attended visual field testing and 53% met the diagnostic criteria for glaucoma. Overall, 118 (18%) participants were identified by expert review who were not referred by either the AI-based pathway or IOP criteria, thereby referring 52 more participants than the AI-based pathway.
• Diagnostic performance achieved a sensitivity of 0.78 (95% CI, 0.62-0.89) and a specificity of 0.95 (95% CI, 0.93-0.97); the positive predictive value was 0.53 (95% CI, 0.40-0.67), and the negative predictive value was 0.98 (95% CI, 0.97-0.99).
• Using the same reference standard, 31/40 glaucoma cases were detected by the AI-based pathway compared with 30/40 glaucoma cases detected by experts.

IN PRACTICE:

"The high specificity observed in this study therefore supports the suitability of this approach for targeted, primary-care-based screening strategies, particularly in settings where specialist access is low and efficient referral pathways are essential," the authors of the study wrote.

SOURCE:

The study was led by Afonso Lima-Cabrita, MD, of the Department of Ophthalmology at Unidade Local de Saúde Santa Maria in Lisbon, Portugal. It was published online on March 12, 2026, in The Lancet Primary Care.

LIMITATIONS:

The proportion of participants with diabetes was high due to integration within a diabetic retinopathy screening pathway. The requirement for a separate hospital visit for confirmatory testing may have contributed to attrition. Applicability to other primary care screening settings might vary according to access to screening devices, trained staff, and referral pathways.

DISCLOSURES:

The authors stated that they received no funding for the study. One author disclosed being a co-founder, shareholder, and consultant for MONA.health, a spin-off from KU Leuven and VITO, to which the AI model was transferred; many other authors reported participation in clinical trials; receiving grant support, honoraria, or consulting fees; and having stock options or other ties with various companies, including AbbVie and Santen.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.