Abstract
Purpose :
The Early Treatment Diabetic Retinopathy Study (ETDRS) study established a 12 step scale to document the severity of diabetic retinopathy (DR) using 7 field stereoscopic color fundus photographs (CFP). The ETDRS scale is used as an outcome in clinical trials. We compared reproducibility in the ETDRS Scale using three methods of evaluation workflow – dual read with adjudication, majority rule and single reader scenario.
Methods :
Certified readers with > 10 years experience in evaluating DR using CFP at a central reading center participated in the study. CFPs from 129 eyes, stratified across the ETDRS scale were graded using 3 methods: (1) dual read with adjudication, where two independent readers evaluate images with adjudication by a third senior reader in cases of discrepancy (2) majority rule, where three or more readers document the ETDRS scale and the majority grade is identified as the final score (3) Single reader system, where each image is evaluated by an individual reader and monitored using inter and intra-grader agreement. Each method was repeated in its entire sequence to assess reproducibility. For example two readers evaluated an image with an adjudicator’s review and two additional readers evaluated the same image with an alternate adjudicator’s review. The reproducibility of the adjudicated grade was analyzed. Both percentage agreements and kappa statistics within each method were compared.
Results :
Agreement rates and weighted kappa are shown in the table below. Weighted kappa was > 0.8 for all three methods and highest with the majority rule system. Ungradable rate was highest with the majority rule system (11%) compared to 1.5 % with the other two methods.
Conclusions :
With experienced readers, all three methods of evaluating ETDRS scale are reproducible. Dual read with adjudication provides high agreement rates with least compromise on data loss due to ungradable images and should be considered the gold standard method for evaluating DR.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.