Abstract
Purpose :
Current diabetic retinopathy (DR) screening setups employ fully manual grading of retinal images and therefore cannot scale to meet the needs of the large, growing diabetic population. Cost savings enabled by EyeArt – a fully-automated, high-throughput, cloud-based DR screening system – are demonstrated for a grading setup similar to the UK NHS Diabetic Eye Screening Program.
Methods :
In the manual grading system, all patient episodes are reviewed by level 1 graders who identify normal patients and recommend re-screening in 12 months. All other episodes are forwarded to level 2 outcome graders who determine referral status or re-screening interval based on the DR severity and surrogate markers for clinically significant macular edema (CSME).
EyeArt automatically analyzes multiple fundus images captured during a patient episode and generates a screening recommendation, while flagging external eye images and poor quality images. We evaluate cost savings enabled by EyeArt when level 1 graders are replaced by EyeArt and patient episodes recommended for referral by EyeArt are forwarded to the level 2 outcome graders.
For this analysis, costs associated with manual grading and quality assurance (QA) grading in UK sterling for the year 2005-06; number of QA reviews per grader per year; and performance of level 1 graders are obtained from Scotland et. al (Br J Ophthalmol 2010; 94:712-719). The DR severity prevalence rates are from Kapetanakis et. al (J Med Screening 2015; 22(3):112-118).
Results :
Assuming the annual number of DR screening episodes to be 2.7 million based on the current diabetic population in the UK, the annual cost of fully manual grading is estimated to be £12,000,709 (level 1 grading cost=£7,992,000; level 2 grading cost=£2,784,409; QA cost=£1,224,300).
EyeArt achieves 85% sensitivity at 86% specificity at detecting episodes with pre-proliferative/proliferative DR or surrogate markers for CSME on a real-world dataset with 2045 episodes. Therefore, with EyeArt replacing level 1 graders, the level 2 grading cost is £3,626,974 and with QA review of 6% of episodes, the QA cost is £528,000. Therefore without considering the cost of EyeArt or the benefits of EyeArt’s higher sensitivity compared to level 1 graders, the economic value added by level 1 automated grading is £7,627,054.
Conclusions :
Automated DR screening using EyeArt can provide significant cost savings in DR screening programs.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.