The overall study population, weighted to be representative of the U.S. noninstitutionalized population aged ≥40 years, had a mean age of 56 years, 47% were male, 79% were non-Hispanic white, 9% were non-Hispanic black, and 12% were of “other” race and ethnicity. Mean A1C and FPG were 5.7% and 5.9 mmol/l (106 mg/dl), respectively. Among the participants with diabetes, 88% were using hypoglycemic medication and 40% of those taking hypoglycemic medication had FPG <7.0 mmol/l. In this study population, the prevalence of any retinopathy was 11% and was appreciably higher in those with (36%) than in those without diagnosed diabetes (8%).
shows that people with retinopathy were older, more likely to be men, had a higher prevalence of diagnosed diabetes and diabetes treatment, and were more likely to have hypertension. The coefficients of variation (CVs) (%), calculated as (100 × standard deviation)/mean, where standard deviation = [(sample size × standard error2)/design effect)0.5] are unitless and therefore can be compared between datasets having different units; the CVs were 15.6 and 28.8 for A1C and FPG (P < 0.001), respectively.
Characteristics of analytic population by diabetic retinopathy status
shows retinopathy prevalence by A1C deciles (5.0, 5.2, 5.3, 5.5, 5.6, 5.7, 5.8, 6.0, and 6.7%), cut points of the Pima Indian study (4.8, 5.0, 5.1, 5.3, 5.5, 5.7, 5.9, 6.6, and 9.4%) (2
), and a 0.1% increment with a fixed 0.5% width window. Regardless of the approach used, there was a sharp increase in retinopathy prevalence above an A1C of 5.5% (95% CI 5.3–5.6 for the decile approach, 5.1–6.1 for the Pima Indian approach, and 5.3–5.8 by the moving average approach) (all P
values <0.05). The linear regression coefficients of retinopathy prevalence by the 1% increment approach were 0.7 (P
= 0.756) and 12.7 (P
< 0.001) for before and after the change point of 5.5%, respectively. That is, above an A1C of 5.5%, the prevalence of retinopathy rose 12.7% for each 1% A1C increment.
Relation between prevalence of retinopathy and A1C (A) and FPG (B).
describes retinopathy prevalence by FPG deciles (4.8, 5.0, 5.2, 5.4, 5.5, 5.7, 6.0, 6.4, and 7.7 mmol/l), cut points of the Pima Indian study (4.9, 5.2, 5.3, 5.5, 5.8, 6.0, 6.4, 7.5, and 12.4 mmol/l) (2
), and a 0.1-mmol/l increment with a fixed 0.5-mmol/l–width window. Regardless of the approach to categorize FPG, there was a sharp increase in retinopathy prevalence after 5.8 mmol/l (95% CI 5.1–6.1 for the decile approach, 5.2–6.8 for the Pima Indian approach, and 5.3–6.3 by the moving average approach) (all P
values <0.05). The linear regression coefficients of retinopathy prevalence by the FPG 1% increment approach were 0.8 (P
= 0.476) and 3.9 (P
< 0.001) for before and after the change point of 5.8 mmol/l, respectively.
After excluding participants taking hypoglycemic medications, by using the moving average approach, the change point for A1C (%) remained at 5.5 (95% CI 5.2–5.7); the regression coefficients were 0.8 (P = 0.409) and 10.5 (P < 0.001) for before and after the change point of 5.5, respectively. However, for FPG, the change point increased from 5.8 to 7.0 mmol/l (6.8–7.2). The regression coefficients for retinopathy prevalence were 1.1 (P = 0.079) and 4.3 (P < 0.001) for before and after the change point, respectively.
Based on the total study population, AUC indicated that A1C was more accurate than FPG in discriminating retinopathy cases from noncases: AUC 0.71 (95% CI 0.66–0.76) for A1C and 0.65 (0.60–0.70) for FPG, P for difference = 0.009 (). At A1C cut points of 5.5, 6.0, and 6.5%, the sensitivities and specificities were 80 and 37%, 55 and 79%, and 38 and 92%, respectively. At an FPG cut point of 5.8, 6.5, 7.0, and 7.5 mmol/l, sensitivities and specificities were 58 and 64%, 43 and 84%, 35 and 89%, and 30 and 92%, respectively.
Receiver operating characteristic curves for A1C (%) and FPG (mmol/l) and prevalent retinopathy.
We reran the models using multivariate logistic regression on the total study population that included selected covariates of glycemia and retinopathy risk (age, sex, race/ethnicity, BMI, waist circumference, time since diagnoses of diabetes, diabetes treatment, and hypertension status). The AUC of A1C and prevalence of retinopathy increased from 0.71 to 0.75 (P = 0.060), and the AUC of FPG and prevalence of retinopathy increased from 0.65 to 0.74 (P < 0.001). For both A1C and FPG, only time since diagnosis of diabetes and sex were significantly related to prevalence of retinopathy, and their inclusion improved the discrimination of prevalence of retinopathy in these two full models (both P values <0.05).