Assessing the risk of diabetic retinopathy

We used multiple logistic regression to analyze cross-sectional data on 1,006 patients with diabetes mellitus who were examined in a rural diabetic screening program. The factors found to have independent and significant associations with the presence of retinopathy were duration of diabetes, albuminuria, serum creatinine, glycosylated hemoglobin, and plasma glucose levels, blood pressure, and percentage of desirable weight. By comparing the contribution of each significant factor to the deviance from the regression models, we estimated the relative importance of each. Duration of diabetes was by far the most important predictive factor. The logistic model provided a good fit to the observed prevalence rates in various risk groups but was less satisfactory for predicting the presence or absence of retinopathy in individual patients. The discovery of presently unknown risk factors may improve this situation.