To evaluate the feasibility of provision of a reading center for tele-screening of diabetic retinopathy (DR) in our region.

Fundus photographs of 134 patients recruited in a project for screening of DR were rated by 2 readers. Images were taken 10 minutes after instillation of one drop of tropicamide 1%, including macula centered and disk centered 45 degrees fundus photographs. Photographs were first presented individually to 2 readers on a computer screen by pre-designed software. Reader 1 was the same person who had examined patients clinically. Both images of each eye were then presented to the readers simultaneously. In each step random re-presentation of about 10% of images by the software was designed to reader 2. Readers rated the photos (patients) as referral warranted if there were more than 3 microaneurysms, diabetic macular edema (DME), or other fundus pathologies in the image. Sensitivity, specificity and accuracy of referral in reading images (single, dual, and patient based according to the dual images of both eyes of a patient) of readers compared to clinical examination along with Kappa statistics for correlation of the 2 readers were calculated.

Sensitivity, specificity and accuracy were 0.83, 0.86, and 0.85 for reader 1 and 0.76, 0.76, and 0.76 for reader 2 in reading single images. The same parameters in reading dual images were 0.88, 0.9 and 0.9 for reader 1 and 0.71, 0.92, and 0.87 for reader 2. Patient based analysis revealed sensitivity, specificity and accuracy of 0.97, 0.67, and 0.75 for reader 1, and 0.94, 0.4, and 0.54 for reader 2 respectively. The positive and negative predictive values for readers 1 and 2 were (0.67, 0.99) and (0.68 and 0.96) respectively in patient based analysis. In assessing the correlation of the 2 readers, Kappa statistics was 0.64 for single images, 0.71 for dual images, and 0.72 for patient based analysis. Reader 2 had reproducibility of 0.70 reading single images, and 0.40 reading dual images.

Presentation of both images of both eyes to readers provides good sensitivity, specificity and negative predictive values and is acceptable for a screening project.