Abstract
Purpose :
To identify the most common areas for discrepancy in retinopathy of prematurity (ROP) classification between experts and between image-based and ophthalmoscopic classification
Methods :
325 infants were recruited as part of a multi-center ROP cohort study from 8 participating centers. Each site had participating ophthalmologists who provided the clinical classification after routine ophthalmoscopic examination, and obtained wide-angle retinal images (RetCam; Clarity Medical Systems, Pleasanton, CA). Images were independently classified by two study experts using a secure web-based module. Image-based classifications (zone, stage, plus disease, overall disease category) were compared between the two experts, and to the clinical classification by the examining ophthalmologist. Inter-expert image-based agreement and image-based vs. clinical diagnostic agreement were determined using absolute agreement and weighted kappa statistic.
Results :
1774 study eye examinations from 325 infants were included in the study. Experts disagreed with each other on the stage classification in 721/1774 (41%) of comparisons, plus disease classification (including pre-plus) in 334/1774 (19%), and zone in 147/1774 (8%). Disagreement in classifying presence of type 1 disease was between 3 and 5% for all comparisons. There were systematic differences between image-based classification clinical (ophthalmoscopic) classification for zone, but not for stage, plus, type 1 or type 2 disease. Among discrepancies in the diagnosis of type 1 disease, the majority of disagreements involved classification discrepancies in both stage and plus disease.
Conclusions :
The most common area of disagreement in ROP classification between experts in this study is in diagnosis of stage. However, agreement about presence of type 1 and type 2 disease is high. There were no systematic differences between image-based classification and the clinical exam in detecting type 1 or type 2 disease. These findings are important in identifying potential areas for error during ROP diagnosis and education, and support efforts to incorporate telemedicine approaches using image-based classification either using human graders or automated computer based image analysis.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.