New Diabetes Diagnostic Threshold of Hemoglobin A_1c and the 3-Year Incidence of Retinopathy

We analyzed data from a cohort of 21,137 Japanese adults aged ≥21 years who underwent a health checkup between January 1 and 31 December 2006 and were followed up 3 years later at the Center for Preventive Medicine, St. Luke’s International Hospital (Tokyo, Japan). We used 3-year incidence of retinopathy as the outcome measure in our study based on evidence that retinopathy develops within 5 years of diagnosis of diabetes in substantial proportion of people with diabetes (18). The participants underwent self-administered clinical surveys (e.g., use of the diabetes medication, diagnosis of diabetes, family history of diabetes, smoking status, alcohol consumption), physical examination, laboratory testing, imaging studies, and retinal photographs.

For analysis of prevalent retinopathy, we excluded individuals without information on baseline HbA_1c (n = 375), those who did not undergo baseline retinal exams (n = 303), and those with ungradable retinal images (n = 26), resulting in a sample size of 20,433. For our longitudinal analysis of incident retinopathy, we further excluded those with retinopathy at baseline (n = 245), those missing information on follow-up retinal exams (n = 223), and those with ungradable photographs at follow-up visits (n = 68), resulting in a final sample size of 19,897. Institutional review board approvals were obtained from Beth Israel Deaconess Medical Center (Boston, MA) and St. Luke’s International Hospital (Tokyo, Japan).

Retinal digital photos were taken for both eyes for the participants at both baseline and follow-up visits. The Canon CR-DG10 digital camera (Canon, Tokyo, Japan) was used to take one digital image per eye (total two images per participant) through a nonpharmacologically dilated pupil. Participants were tested in a dark room to allow the pupils to dilate naturally in preparation for the retinal imaging examination. After dark adaptation, 45° retinal photographs centered on the macula were taken of both eyes. The digital images were graded by trained ophthalmologists. Retinopathy was defined as the presence of hard exudates, cotton wool spots, retinal hemorrhage, or more severe forms of retinopathy, the Fukuda standard (19) A2 or higher.

Whole-blood samples collected from the participants were assayed for HbA_1c immediately after the they were obtained with high-performance liquid chromatography (Tosoh G7; Tosoh Corporation, Tokyo, Japan). We categorized HbA_1c levels as follows: <5.0, 5.0–5.4, 5.5–5.9, 6.0–6.4, 6.5–6.9, and ≥7.0%. Those with HbA_1c levels of 5.0–5.4% comprised the largest sample and were used as the reference category in our logistic regression models.

We also extracted the potential predictors of retinopathy in those with diabetes: age (20), systolic and diastolic blood pressures (21–26), LDL cholesterol (27), HDL cholesterol, triglycerides (27,28), BMI, self-reported family history of diabetes, smoking status (29,30), and alcohol consumption (31). Triglyceride levels were right-skewed; therefore, we used a logarithmic transformation of the variable to improve the fit for the model. The diagnosis of diabetes was defined on based on self-report of a diagnosis of diabetes or of taking medication for diabetes. Hypertension was defined as systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg, or the use of hypertension medication. All variables, including HbA_1c and retinal images, were measured and recorded both at baseline and at follow-up visits.

We constructed a series of logistic regression models to investigate the association between HbA_1c levels and the risk of retinopathy. Model 1 was adjusted for age and sex. Model 2 was adjusted for age, sex, hypertension, diagnosis of diabetes, HDL and LDL cholesterol, log-transformed triglyceride, BMI (<25, 25.0–29.9, or ≥30 kg/m²), alcohol consumption (regular, occasional, or never drinker), smoking status (current, former, or never smoker), and family history of diabetes (yes or no). We evaluated these models to estimate multivariable-adjusted odds ratios (ORs) and 95% CIs for prevalent retinopathy (models 1a and 2a) and incident retinopathy (models 1b and 2b). To examine the impact of integrating known cases of diabetes in our analysis, we excluded those with a diagnosis of diabetes and developed additional models 3a and 3b, in which we adjusted for all the variables in models 2, with the exception of the diagnosis of diabetes.

To assess whether the threshold where the risk begins to increase (inflection point) exists, we examined the continuous associations between HbA_1c level and the prevalence or the incidence of retinopathy without assuming linearity by fitting restricted cubic spline models (32–33) with the knots corresponding to the HbA_1c cutoff points of interest: 5.0, 5.5, 6.0, 6.5, 7.0, and 7.5%. The ORs were adjusted for all the variables included in model 2. All statistical analyses were conducted with SAS statistical software version 9.2 (SAS Institute Inc., Cary, NC).

Table 1 shows the baseline characteristics of the study population overall and according to HbA_1c levels. Approximately 49% were men. The mean age was 51.0 years, the mean HbA_1c was 5.6%, and the mean (±SD) follow-up period was 3.0 ± 0.29 years. Participants with higher HbA_1c levels were more likely to be older, to be men, to smoke, to take medication for hypertension, and to have several clinical risk factors, including elevated blood pressure, higher BMI, lower HDL cholesterol, higher triglyceride, and a family history of diabetes.

Among the 20,433 participants, the crude prevalence of retinopathy was 1.2% (245/20,433). The adjusted ORs and 95% CIs for prevalent retinopathy are shown in Table 2. After initial adjustment for age and sex (model 1a), the prevalence of retinopathy was significantly higher at HbA_1c levels ≥6.5% than in the reference category. After further adjustments (model 2a), this estimate was attenuated, and only HbA_1c levels ≥7.0% achieved statistical significance.

The crude cumulative incidence of retinopathy at 3 years was 0.85% (170/19,897). After initial adjustment for age and sex, there was no significant association between HbA_1c value and the incidence of retinopathy at HbA_1c <6.5%; compared with the reference category, however, HbA_1c levels of 6.5–6.9% were associated with significantly higher risk of developing retinopathy at 3 years (OR 2.35 [95% CI 1.08–5.11]; P = 0.031) (Table 3). The risk remained significantly higher after further adjusting for the confounders and other independent risk factors for retinopathy (model 2b). Our results did not alter substantially after the exclusion of those with diagnosis of diabetes at baseline in model 3b.

In our analysis to evaluate whether the relationship between baseline HbA_1c levels and prevalence of retinopathy is nonlinear, we found that the nonlinear relationship was borderline significant (P for curve = 0.08) (Fig. 1). For the outcome of incident retinopathy, however, the nonlinear relationship was statistically significant (P = 0.001), suggesting a possible threshold at HbA_1c levels between 6.0 and 7.0%. We observed a dose-response relationship between higher HbA_1c levels and increased risk of incident retinopathy at HbA_1c ≥6.5%. The restricted cubic spline analysis for the nondiabetic subpopulation yielded similar findings (data not shown).

We found that Japanese adults with HbA_1c levels of 6.5–6.9% were at significantly higher risk of developing retinopathy at 3 years than were those with HbA_1c levels of 5.0–5.4%, whereas the risks did not increase among those with HbA_1c levels <6.5%. To the best of our knowledge, this is the first longitudinal study supporting the validity of the new diagnostic threshold of HbA_1c recommended by the International Expert Committee (5,11). In contrast, we did not observe an explicit threshold effect of HbA_1c in the analyses of the prevalent retinopathy.

Although there have been several longitudinal studies examining the association between HbA_1c levels and the risk of retinopathy, most of those studies were limited to samples treated for diabetes (29,34–38). There have been only a few small longitudinal studies of general nondiabetic populations (6,15–17). Possibly because of small sample sizes, results from these previous studies did not support the current HbA_1c threshold for diagnosing diabetes (6,15–17). In the Pima Indian study (N = 927), investigators found that the risk of incident retinopathy begin to increase only at HbA₁ levels of 9.1% (6). Only HbA₁ data were available at that time (HbA_1c measurements were unavailable), and no adjustments were made for hypertension and other independent risk factors for retinopathy in that study (6). The longitudinal study (N = 233) by van Leiden et al. (15) found that the risk of developing retinopathy was significantly higher in the highest tertile of HbA_1c (HbA_1c 5.8–13.1%) relative to the lowest tertile (HbA_1c 4.3–5.2%). Because of small sample size, these investigators collapsed HbA_1c levels from 5.8 to 13.1% into a single category. Selvin et al. (16) examined the risk of retinopathy among participants from the Atherosclerosis Risk in Communities study. The overall sample size of this study was large (N = 11,357), but repeated retinal examinations were performed in only 767 people. These investigators were unable to detect a statistically significant threshold in the association of HbA_1c with the incidence of retinopathy, possibly because of the lack of power. Recently, Massin et al. (17) studied 700 participants from the Data from an Epidemiological Study on the Insulin Resistance Syndrome study, in which participants were followed up for 10 years, and proposed an HbA_1c threshold of 6.0%. Because retinopathy was not evaluated at baseline in this study, however, they were not able to examine the incidence of retinopathy (17).

Our study has several limitations. First, the retinal images were graded in single-field photographs per eye in this study. Multiple photographic fields per eye would have improved the sensitivity of the funduscopic examinations. Second, our study sample was composed exclusively of native Japanese, so whether our results generalize to other populations is unclear. It is noteworthy that in the DETECT-2 project, which pooled studies from the U.S., Australia, India, Japan, and Singapore, they found no racial difference in optimal HbA_1c threshold (12). Third, we did not take into account the possible effect of hemoglobinopathies on HbA_1c values. However, the prevalence of hemoglobinopathies in Japan is reported to be as low as 0.04% (39) and is therefore likely to have little impact on our overall findings. Finally, the detection of an inflection point in the relation between HbA_1c and retinopathy may not in itself establish the optimal threshold for clinicians to use in the diagnosis and treatment of diabetes. The optimal threshold for any patient is the level at which the benefits of diagnosis and treatment exceed harms for that patient. If there were only benefits and no harm in diagnosing and treating diabetes, the inflection point would represent the level of HbA_1c. However, when the benefits of diagnosis and treatment of diabetes are small, the optimal diagnostic threshold may be higher than the inflection point we observed.

Our longitudinal study is the first to date to suggest a threshold of risk for incident retinopathy at a 6.5% HbA_1c level. These findings support the validity of the new diagnostic HbA_1c threshold for diabetes recently adopted by the American Diabetes Association, the European Association for the Study of Diabetes, and the World Health Organization (5,12). Additional longitudinal studies are needed to validate these findings in other populations.

Y.T. is supported by Shigeaki Hinohara, MD, Primary Care Fellowship at Beth Israel Deaconess Medical Center (Boston, MA), Joint Japan/World Bank Graduate Scholarship Program, and St. Luke’s Life Science Institute (Tokyo, Japan). C.C.W. is supported by a midcareer mentorship award from the National Institutes of Health (Grant K24-DK-087932). J.B.M. is supported by National Institute for Diabetes and Digestive and Kidney Diseases (Grant K24-DK-080140). R.B.D. is supported by Harvard Catalyst, The Harvard Clinical and Translational Science Center (NIH Award UL1-RR-025758 and financial contributions from Harvard University and its affiliated academic health care centers).

No potential conflicts of interest relevant to this article were reported.

Y.T., O.T., T.F., W.C.T., and C.C.W. developed the study concept, developed the design, and interpreted the data. Y.T., R.B.D., and F.I. conducted statistical analyses. Y.T., J.B.M., W.C.T., and C.C.W. drafted the manuscript. All the authors revised the manuscript critically for important intellectual content and approved the final manuscript. Y.T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Parts of this study were presented as an oral presentation at the 34th Annual Meeting of the Society of General Internal Medicine, Phoenix, Arizona, 4–7 May 2011.

The authors thank Donald Halstead, Harvard School of Public Health, for help with preparing the manuscript. The authors also thank Toshiko Kawakita and Sachiko Ohde, St. Luke’s International Hospital, Tokyo, Japan, for collecting the data.