Population-wide screening for primary open-angle glaucoma (glaucoma) is typically not cost-effective because of low prevalence and high costs. We evaluated the cost-effectiveness of repeated artificial intelligence (AI)-based glaucoma screening using fundus photos in the Dutch population aged 50 to 75, compared with opportunistic case finding.
We developed a health-economic model consisting of a decision tree for screening simulation and a Markov model for disease progression and treatment effects. In AI screening, the population is invited every 5 years for fundus photography in primary care labs, followed by AI triage. Positive cases undergo ophthalmological evaluation. Model inputs included screening repetitions, AI sensitivity (85%), specificity (95%), screening compliance (50%), referral compliance (60%), and glaucoma progression probabilities, based on AI development data, a Dutch cohort, expert opinions, and literature. We performed analyses from societal and healthcare perspectives over a lifetime.
AI screening identified glaucoma earlier than opportunistic methods, detecting 1.60 times more cases and reducing visual impairment by 0.8 months per invited individual. Societal perspective analysis indicated a gain of 0.014 quality-adjusted life-years and an incremental cost of €284 per individual, resulting in an incremental cost-effectiveness ratio of €19 311, with a 51.2% probability of being below the €20 000 threshold. Outcomes were most sensitive to glaucoma progression rates, utilities, and visual impairment costs.
AI-based glaucoma screening in The Netherlands could cost-effectively improve early detection and reduce visual impairment burden. Further research should focus on reducing uncertainty regarding disease progression rates and impairment-related costs.