Purchase this article with an account.
S. L. Williams, L. Wang, S. A. Kane, T. C. Lee, D. J. Weissgold, A. M. Berrocal, J. Starren, J. T. Flynn, M. F. Chiang; Telemedical Diagnosis of Retinopathy of Prematurity: Accuracy of Expert vs. Nonexpert Graders. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1397.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To assess the accuracy of telemedical retinopathy of prematurity (ROP) diagnosis by expert graders compared to trained nonexpert graders.
Two-hundred forty-eight eyes from 67 consecutive infants underwent wide-angle retinal photography using a standard imaging protocol, with a commercially available camera (RetCam-II; Clarity Medical Systems, Pleasanton, CA). Data were uploaded to a web-based telemedicine system. A group of nonexpert graders consisting of ophthalmology residents and medical students attended 2 hour-long training sessions on image-based ROP diagnosis. Using the telemedicine system, 3 experienced pediatric retinal specialists (expert graders), 4 ophthalmology residents, and 8 medical students provided a diagnosis for each eye (no ROP, mild ROP, type-2 prethreshold ROP, treatment-requiring ROP). Accuracy of telemedical diagnosis by each grader was determined, based on a reference standard of indirect ophthalmoscopy by an experienced pediatric ophthalmologist.
For detection of mild or worse ROP, the mean (range) sensitivities and specificities were 0.896 (0.839-0.929) and 0.958 (0.936-0.987) for experts, 0.868 (0.811-0.952) and 0.831 (0.41-0.991) for residents, and 0.770 (0.699-0.854) and 0.975 (0.957-1.00) for medical students. For detection of type-2 or worse ROP, the mean (range) sensitivities and specificities were 0.949 (0.939-0.970) and 0.933 (0.912-0.957) for experts, 0.872 (0.710-0.970) and 0.727 (0.387-0.947) for residents, and 0.775 (0.516-0.879) and 0.916 (0.873-0.961) for medical students. For detection of treatment-requiring ROP, the mean (range) sensitivities and specificities were 1.00 (1.00-1.00) and 0.933 (0.877-0.961) for experts, 0.875 (0.500-1.00) and 0.844 (0.708-0.982) for residents, and 0.872 (0.600-1.00) and 0.921 (0.845-0.965) for medical students.
It is possible to train nonexpert graders to perform telemedical ROP evaluations with very high diagnostic accuracy, but the mean sensitivity and specificity of trained nonexperts in this study were lower than that of experts. Development of methods for training and education of graders may be warranted to support telemedical ROP evaluation.
This PDF is available to Subscribers Only