Purpose
To compare the agreement of clinical diabetic retinopathy (DR) severity between fundus images captured using a low-cost refurbished non-mydriatic retina camera at University Medical Associates (UMA), a primary care clinic within the University of Virginia (UVA), and the 7-standard field color 35mm stereoscopic retinal photographs obtained at the time of dilated eye exam at the UVA Eye Clinic.
Methods
Over 16 months, 794 adults with type 1 and 2 diabetes mellitus with no eye exam in the past year underwent tele-ophthalmology screening for DR during their visit to UMA. A trained nurse obtained a single 45° macula-centered image of each eye using a refurbished Topcon TRC-45N ($2100) with a Canon T2i ($300) DSLR-back. A qualified ophthalmologist reviewed the photos which were graded for presence of DR. Each eye was assigned a clinical level of DR severity. All patients with DR received a f/u visit to the Eye Clinic. During the f/u exam, the ophthalmologist graded the level of DR severity after a dilated eye exam and ETDRS protocol fundus photography using a Topcon 50EX with OIS cooled CCD camera back ($75k). Retinopathy severity was graded in all images as: no DR, mild DR, moderate DR, severe DR, or proliferative DR. Agreement between gradings was assessed by calculating the percentage of agreement and k statistics.
Results
Upon screening, 137 (17%) of the 794 total patients were found to have DR in at least one eye. 5 (4%) were ungradable and 22 (16%) did not complete a f/u dilated eye exam. Analysis was done on 220 eyes of 110 patients. Table 1 compares the DR severity levels between the screening photos and dilated eye exam. Exact agreement for level of DR was observed in 150 (68.2%) eyes with a k of 0.55+/-0.04 and weighted k of 0.62 (95% CI 0.46-0.63). This is considered fair to good agreement. Agreement within one level of DR severity was 92%. The screening photos had a positive predictive value of 85% for determining the presence of DR and a sensitivity of 96.5%.
Conclusions
There was good agreement between the non-mydriatic screening photos and dilated eye exams for clinical DR severity. The lack of full agreement can be related to the limitations of the non-mydriatic camera in capturing all retinal fields causing key DR lesions to be missed. Our results show the current screening program using a refurbished camera is both cost-effective and sensitive in detecting presence and severity of DR.
Keywords: 499 diabetic retinopathy •
498 diabetes •
688 retina