Wide-scale implementation of telemedicine has become increasingly important in a variety of medical fields, including ophthalmology, as the COVID-19 pandemic continues to impact healthcare delivery. It is well known that age-related macular degeneration (AMD) is a major cause of blindness among adults aged 50 years and older in developed countries. If AMD can be diagnosed with screenings in an effective telemedicine platform, clinicians may be able to treat the disease early and potentially reduce the number of patients who become blind from it.
A New Platform
Theodore Smith, MD, from the Icahn School of Medicine of Mount Sinai was part of a research team that co-authored a study presented at the 2020 American Academy of Ophthalmology Annual Meeting assessing an artificial intelligence (AI)-based telemedicine platform for screening referable patients with AMD. His co-authors included Sharmina Alauddin, MD; Alauddin Bhuiyan, PhD; Arun Govindaiah; Raphael Gildengorn; Jake E. Radell; Catherine Ye; and Oscar Otero-Marquez, MD. “Our prospective study of this AI-based telemedicine platform was conducted to test its accuracy in screening AMD in a ‘real world’ clinic setting,” says Dr. Smith. “We wanted to validate its widespread use in primary care.”
Earlier, the deep learning algorithm was originally developed by iHealthScreen Inc. and tested using images from the Age-Related Eye Disease Study. This deep learning algorithm teaches itself from massive datasets through artificial and neural networks. Importantly, each image was independently reviewed and graded by 3 expert human ophthalmologists to validate comparisons.
For the study, researchers imaged 299 nondilated subjects aged 50 years and older at the New York Eye and Ear faculty retina practices with a color fundus camera. In total, 266 patients with 503 gradable images were analyzed by the deep learning AI algorithm. Based on the worst eye, AI-graded patients as referable if they had intermediate AMD (drusen >125 μm and/or pigmentary abnormalities) or late AMD (geographic atrophy or neovascularization). Patients were AI-graded as non-referable if they had normal macula or early AMD (drusen >63 to ≤125 μm).
In total, 161 eyes were graded as referable by the AI platform whereas 342 were graded as non-referable. “The key finding from our study that the AI system achieved 88.7% accuracy in identifying subjects who had AMD of intermediate or advanced severity for which referral to an ophthalmologist was warranted,” Dr. Smith says. The AI platform also demonstrated sensitivity of 86.3%, specificity of 89.8%, and a kappa score of 0.75.
“The results from our study show that early detection of referral level AMD in primary care settings will now be possible with telemedicine,” says Dr. Smith. “This will prevent many cases of vision loss, and also the ensuing downward spiral in overall health, such as depression and the need for nursing home care due to loss of independence, falls, fractures, and death. These adverse outcomes are also extremely costly. The total economic burden—direct and indirect—of vision loss and blindness from all causes is now $145 billion, but this figure is expected to triple by the year 2050 in real dollars. Screening patients for AMD will optimize the use of ophthalmology services for those who need it most.”
Alauddin S, at al. An Automated AI-Based Telemedicine Platform for Screening Patients With Referable AMD: A Prospective Trial. Presented at: American Academy of Ophthalmology 2020 Virtual Congress. Session: PA057. November 2020. Available at: https://aaommg.apprisor.org/apsSession.cfm?id=PA057.
Li B, Powell AM, Hooper PL, Sheidow TG. Prospective evaluation of teleophthalmology in screening and recurrence monitoring of neovascular age-related macular degeneration: a randomized clinical trial. JAMA Ophthalmol. 2015;133(3):276-282.
Sommer AC, Blumenthal EZ. Telemedicine in ophthalmology in view of the emerging COVID-19 outbreak. Graefes Arch Clin Exp Ophthalmol. 2020:1-12.