One photograph was taken of each eye with a Canon CR4-45° nonmydriatic camera. Photographs were taken in a dark room to facilitate dilatation of the pupils and improve the quality of the photographs. Additionally, at the Los Angeles site, pupils were dilated before taking the photos. The retinal field photographed was identical at both sites and consisted of the area nasal to the disc and temporal to the macula and the superior and inferior arcades. All photographs were labeled with only the patient’s identification number and were sent for reading in Santa Barbara. Polaroid prints from the Canon camera were examined and graded by an experienced endocrinologist (L.J.) who, before this study, had readings verified by an ophthalmologist until agreement was virtually 100%. An overall grading was assigned for each eye at each examination using the Wisconsin Epidemiologic Study of Diabetic Retinopathy II/III–modified diabetic retinopathy levels, which used a modification of the Airlie House Criteria. This scale has nine levels per eye, ranging from no retinopathy to total vitreous hemorrhage. The scale was used as follows: no retinopathy (grade 10), very-mild nonproliferative diabetic retinopathy (NPDR) (grade 20), mild NPDR (grade 35), moderate NPDR (grade 43), severe NPDR (grade 53), mild proliferative diabetic retinopathy (PDR) (grade 61), moderate PDR (grade 65), high-risk PDR (grade 71), and advanced PDR and/or fundus partially obscured by disease (grade 85). Photograph quality was deemed adequate for accurate assignment of retinopathy grade in all of the graded photographs included for analysis. The primary outcome measures are the development of retinopathy of any degree in subjects without retinopathy at baseline and the progression of retinopathy in subjects with nonproliferative retinopathy at baseline.

Statistical analyses were performed using SAS statistical software for Windows version 9.1 (SAS Institute, Cary, NC). Continuous data were compared between groups with an unpaired t test, and categorical data were compared with a ?2 test. Odds ratio (OR) for progression of retinopathy in each group was estimated using the PROC GENMOD procedure while accounting for duration of follow-up and controlling for confounders. A repeated statement was used to specify within-subject effect. The first-order Taylor expansion approximation was used to estimate the variance of difference in the ORs between the two groups. Logistic regression with the development of any degree of retinopathy as the binary outcome variable was used to control for covariates in the subset of subjects who had no retinopathy in the baseline photograph.

There were 200 subjects (56 male and 144 female subjects) from the two centers who were included in the retinal photographic studies. Subjects randomized to the intervention (n = 102) and control (n = 98) groups were of similar age and diabetes duration, had similar A1C concentrations, had similar follow-up periods during this study, and had similar drop-out rates.

The progression or development of retinopathy was noted in both the intervention and control group, and the difference in the ORs for progression between the two groups (?0.65) was not statistically significant (P = 0.226). Subjects without evidence of retinal disease at baseline were evaluated separately. As with the total sample, the intervention and control groups were similar, with the exception of follow-up time. However, among 82 subjects who remained in the study through at least a second photograph and could therefore be used to assess the development of retinopathy, the follow-up was similar in the intervention (22.0 months) and the control groups (24.6 months, P = 0.136, data not shown). Those in the intervention group who had no evidence of retinopathy at baseline (Airlie House score = 10 in each eye) were less likely to develop diabetic retinal changes during a mean of 23.1 months of follow-up (5/48) than were those in the control group (10/34, ?2 = 4.805, df = 1, P = 0.028). This difference remained significant in a logistic regression model that controlled for age at diagnosis, duration of diabetes at baseline, duration of follow-up, A1C, ethnicity, and sex (OR 5.35 [95% CI 1.14–2.12], P = 0.034). In a stepwise logistic model, only randomization group, baseline A1C concentration, and age were significant predictors of developing retinopathy (data not shown).