ARTIFICIAL INTELLIGENCE IN DIABETIC RETINOPATHY SCREENING IN PRIMARY CARE: DIAGNOSTIC ACCURACY AND COST-EFFECTIVENESS

Abstract

Diabetic retinopathy (DR) is one of the leading causes of preventable blindness worldwide, and its early detection depends on systematic screenings, traditionally performed by ophthalmologists. However, the limitation of human resources and the overload of specialized services make this screening insufficient in primary care. Artificial intelligence (AI), especially through deep learning algorithms, has proven to be a promising alternative for the automated diagnosis of DR from fundus images. This article aims to evaluate the diagnostic accuracy and cost-effectiveness of these systems when applied to DR screening in the context of primary health care. A narrative literature review was conducted between 2018 and 2025 in PubMed, SciELO, and Google Scholar databases, covering clinical trials, multicenter studies, and economic analyses. The results show that AI achieves accuracy greater than 90%, sensitivity and specificity close to 95%, with the potential to reduce costs associated with manual screening by up to 45%. In addition to expanding access, the technology optimizes referral flows, allowing severe cases to be prioritized. It is concluded that the integration of AI into primary care is a cost-effective and high-impact strategy for the population, provided that it is accompanied by local clinical validation, adequate technological infrastructure, and ethical and legal regulatory policies.