Abstract
Purpose: :
To assess the sensitivity and specificity of 1-, 2- and 3-field retinal photographs in the screening and grading of diabetic retinopathy (DR) by readers at various levels of ophthalmology training.
Methods: :
A prospective study with a blind comparison of the index and reference test. Type II diabetes mellitus patients without prior retinal laser photocoagulation were recruited from an ophthalmology clinic. Dilated fundal photographs were taken in 1-, 2-, 3- and 7-field stereoscopic views. All photographs were randomized and graded by 5 blinded readers at various levels of ophthalmological training (retinal specialist, non-retinal specialist, registrar, junior resident, optometrist) according to the International Clinical Diabetic Retinopathy Disease Severity Scale. The sensitivity and specificity for each view was calculated and inter-rater agreement measured by Cohen’s Kappa coefficient.
Results: :
30 eyes were analysed. Using the retinal specialist 7-field view reading as reference, 13 (43.3%) eyes had no DR, 10 (33.3%) had mild non-proliferative DR (NPDR), 5 (16.7%) eyes had moderate NPDR and 2 (6.7%) eyes had proliferative DR. For detection of any DR (‘screening’), the ranges of sensitivity and specificity for all readers were: 1-field sensitivity 59-100%, specificity 92-100%; 2-field sensitivity 82-100%, specificity 92-100%; 3-field sensitivity 82-100%, specificity 77-92%. For detection of moderate or worse NPDR (‘grading’), the ranges were: 1-field sensitivity 22-89%, specificity 67-100%; 2-field sensitivity 56-89%, specificity 67-100%; 3-field sensitivity 56-89%, specificity 62-100%. Inter-rater agreement Kappa value ranges for DR screening ranged from: 1-field 0.54-0.87, 2-field 0.80-0.93 and 3-field 0.79-0.93. Agreement for DR grading was poorer and ranged from: 1-field 0.33-0.57, 2-field 0.33-0.59 and 3-field 0.30-0.71.
Conclusions: :
Sensitivity increased and specificity decreased with the number of fields read. The 2-field view may be most optimal for DR screening. There was good inter-rater agreement for all views for DR screening. However, agreement was poorer for detecting moderate or worse NPDR, varying from fair to good agreement. Results suggest that screening of DR may be done by readers at various levels of training. However, more training may be required in readers to detect and differentiate severe retinopathy.
Keywords: diabetic retinopathy • detection • clinical (human) or epidemiologic studies: systems/equipment/techniques