Effects of artificial intelligence-based diabetic retinopathy screening on access to timely treatment, in people with diabetes mellitus

• Grant number: 24/09572-1
• Support Opportunities: Regular Research Grants
• Start date: November 01, 2024
• End date: October 31, 2026
• Field of knowledge: Health Sciences - Medicine - Medical Clinics
• Principal Investigator: Fernando Korn Malerbi
• Grantee: Fernando Korn Malerbi
• Host Institution: Escola Paulista de Medicina (EPM). Universidade Federal de São Paulo (UNIFESP). Campus São Paulo. São Paulo , SP, Brazil
• Associated researchers: Caio Vinicius Saito Regatieri ; Dimitris Rucks Varvaki Rados ; Greice Caletti ; Lucas Zago Ribeiro ; Luis Filipe Nakayama ; Marcelo Rodrigues Gonçalves ; Mateus Augusto dos Reis ; Philippe Olivier Alexandre Navaux ; Roberto Nunes Umpierre ; Thiago da Silva Araújo

Abstract

Diabetic retinopathy (DR) is a complication of diabetes that can lead to amaurosis, screening for which is essential to prevent permanent visual damage, but in Brazil, coverage is low due to limited access to ophthalmologists and financial resources. This is a partnership previously established between the coordinators on this topic, configuring continuity of research in a line of common interest, seeking a pragmatic approach to the problem and a feasible solution that can be immediately applied by the SUS. This project's general objective is to evaluate whether universal screening for DR and macular edema in people with diabetes through retinography performed using a portable retinography camera and a computer program (artificial intelligence algorithm, AI, patented by the authors) is non-inferior to screening for these conditions. through retinography performed by a portable retinography camera and reported by ophthalmologists to qualify referrals to ophthalmologists and increase the number of treatments in an appropriate time. The specific objectives are: 1. Carry out modeling to estimate the impact on the service queue of implementing the proposed tracking system, seeking to simulate the prevalence of DR that would determine system saturation (consultations with an ophthalmologist, laser treatment and pharmacological intraocular treatment) and the cost-effectiveness of the proposed intervention before it begins. Subsequently assess whether this estimated impact matches the real impact after implementing the system; 2. Validate a computer program already registered to identify referenceable RD for use in a portable fundus camera, with performance comparison with evaluation by ophthalmologists; 3. Evaluate the safety of AI-based screening for DR in a randomized clinical trial (RCT) comparing mobile retinography associated with a computer program (RDIA) vs retinography performed by a portable retinography camera approved by ophthalmologists (RDOF). 4. Evaluate the cost-effectiveness of tracking the proposed strategies. The RCT proposes to evaluate whether universal screening for DR and macular edema in people with diabetes using retinography (portable fundus camera) obtained by a trained nursing technician and analyzed by a registered computer program (AI algorithm) is non-inferior to screening using fundus scans reported by ophthalmologists. The primary outcome will be the number of appropriate referrals to ophthalmologists and the secondary outcome will be the number of patients referred for laser, pharmacological intraocular treatment, and cataract surgery. This is an RCT to be conducted in a city in RS, where retinography images of adults with diabetes will be randomized for interpretation by AI (RDIA, n=1029) vs. reported by ophthalmologists (RDOF, n=1029). Once the operational tracking model is organized, it can be applied to other regions in Brazil, optimizing the use of public resources and improving the prevention of amaurosis in these patients. By proposing this DR tracking, we will enable an innovative strategy to include greater access and management of this resource for patients with diabetes, with the potential to prevent amaurosis, according to data published in countries with higher income. Through validation of a portable fundus camera and the use of a recently patented computer program (AI algorithm) that used data from the Brazilian population, we will allow more efficient use of less available resources in the diagnosis and management of DR, including in areas of limited healthcare access. . This tool can be used in other assistance, teaching and research institutions, bringing innovation and better resource management to its users. Cost-effectiveness analyzes will support the justification for the implementation and dissemination of knowledge generated after the results are completed. (AU)