Kim HM (Center for Practice Management & Outcomes Research, VA Medical Center, Ann Arbor, MI)
Lowery JC (Center for Practice Management & Outcomes Research, VA Medical Center, Ann Arbor, MI)
The objective of this study was to evaluate the accuracy of using digital fundus photographs to diagnose retinopathy in patients with diabetes.
Diagnostic evaluations from a trained retinal grader using mydriatics (pupil dilation) and seven standard field 35-mm stereoscopic slides (the gold standard for diagnostic assessments) were compared with the grader’s evaluations from digital photographs, with and without mydriatics. Additionally, two ophthalmology fellows each evaluated subsets of the digital images, for comparison with the gold standard.
249 patients from the Ann Arbor VAMC were enrolled with photos taken. Using Early Treatment Diabetic Retinopathy (EDTRS) grading criteria, 38% of patients met referral criterion (retinal level >= 35 for one or both eyes) based on the 35-mm slides. Agreement on this criterion between 35-mm slide and digital assessments was higher for the trained grader (kappa of 0.55 for mydriatic assessments) than for the two fellows (kappas of 0.45 and 0.51). Compared with mydratic assessments, non-mydriatic digital assessments had essentially the same or slightly lower kappas, but the proportion of ungradeable images was significantly greater (8% versus 25% ungradeable as determined by the trained grader). Specificity was high (0.93, 0.95, 0.98) for all three readers for mydriatic digital assessments; but sensitivities were low (0.56 by the trained grader, 0.41 and 0.41 by the two fellows). Both specificity and sensitivity were generally low for non-mydriatic assessments by all three graders (0.78, 0.86, 0.97 for specificity and 0.66, 0.51, 0.37 for sensitivity).
The low sensitivity of the digital assessments indicates that a significant proportion of patients in need of referral would not be referred according to the criteria used. The question remains whether the ultimate outcome of blindness incidence could be reduced through the use of telemedicine in populations currently without access to screening services.
VA is expending considerable sums of money to implement telemedicine initiatives, including screening for retinopathy. These findings suggest that implementation of such initiatives should take place with a clear understanding of potential limitations. Findings from this study and other rigorous evaluations of the diagnostic accuracy of telemedicine technology need to guide its implementation and use.