Prediction of diabetic retinopathy from clinical variables and color vision data.

Predictions about the onset of retinopathy in 295 diabetic patients, all originally having no evidence of retinopathy, have been made in a longitudinal study over 7 yr. Out of many color vision tests and clinical variables, the best individual predictor was a measure of yellow-blue discrimination, using an anomaloscope. The other predictors of significance were the degree of blood glucose control and the duration of diabetes. Although the predictions from a linear logistic model were significant in classifying the diabetic subjects into those whose fundus will remain normal and those in whom it will develop retinopathy, the number of misclassifications was substantial. An examination of the goodness of fit between the data and the model suggested a criterion value (P) of around P = 0.3 for the probability that a patient develops retinopathy. At this value, the probability of being normal for an individual classed as normal was 0.82, and the probability of developing retinopathy for an individual classed as having retinopathy was 0.54.