Data from 70 type I diabetic patients with nonproliferative retinopathy participating in a multicenter clinical trial of control and complications were analyzed to test for associations of clinical variables with baseline levels and 8-mo changes in retinopathy. Predictor variables included age, duration of diabetes, systolic blood pressure, inpatient and outpatient plasma glucose levels, glycosylated hemoglobin (HbA1), M-values, serum cholesterol, serum triglycerides, and creatinine clearance. Retinopathy was assessed by fundus photography and graded at the Fundus Photograph Reading Center according to a detailed protocol. For the entire group, baseline retinopathy was positively correlated (P less than 0.05) with baseline systolic blood pressure, plasma glucose, HbA1, serum cholesterol, and duration of disease and negatively correlated with creatinine clearance. Conversely, during treatment, progression of retinopathy was negatively correlated (P less than 0.05) with mean levels during treatment of plasma glucose, HbA1, M-values, serum cholesterol, and with changes during treatment in plasma glucose and serum triglycerides. Two-group and three-group multivariate classification analysis of progression of retinopathy (improved or unchanged versus worsening--mild or moderate) indicated lower plasma glucose as the single best predictor of worsening of retinopathy (P less than 0.05), correctly classifying 71% of patients with positive progression. Decreased creatinine clearance during therapy was found to be the best discriminator between mild and moderate progression. Other multivariate models yielded specificity values of up to 71% and sensitivity values of up to 92%. We conclude that associations among clinical predictors and retinopathy during short-term glycemic control differ strikingly from those at baseline.(ABSTRACT TRUNCATED AT 250 WORDS)