Performance characteristics of digital fundus photography as a screening test for diabetic retinopathy in a low-risk...

The purpose of the present study was to estimate the sensitivity and specificity of reviewers' assessments of digital fundus photography (DFP) images, using the findings from comprehensive ophthalmologic examination as the criterion standard. One hundred and fifty-two patients with diabetes underwent comprehensive ophthalmologic examination by an active duty U.S. Air Force (ADAF) staff ophthalmologist, and the examination findings were used as the criterion standard for the present study. Eight other ophthalmologists (M.D.'s) and 10 optometrists (O.D.'s), all ADAF providers, each evaluated seven nonstereo standard field DFP images from the left and right eyes (14 images per patient) of the 152 diabetic patients, assessing each patient for evidence of diabetic retinopathy (yes, no, not sure). The sensitivity, specificity, and proportion of "not sure" responses were computed separately for MD and OD reviewers and then compared. O.D. reviewers gave "not sure" responses more frequently than M.D.'s. With "not sure" responses treated as correct (i.e., identical to the results of ophthalmologic examination), both types of providers had 100% sensitivity for cases that are usually treated and 83% sensitivity for cases that are not usually treated but require close follow-up. Specificity (for any diabetic retinopathy) was 92% for M.D.'s and 93% for O.D.'s. With M.D.'s as photographic reviewers, DFP-based screening required only 27 ophthalmologic examinations per 100 patients screened; with O.D.'s as photographic reviewers, 31 ophthalmologic examinations were required per 100 patients screened. Using either ophthalmologists or optometrists as photographic reviewers, DFP-based screening for diabetic retinopathy has very good sensitivity and excellent specificity.