OBJECTIVE—The degree to which chronic glycemic exposure (CGE) (fasting plasma glucose [FPG], HbA1c [A1C], duration of diabetes, age at onset of diabetes, or combinations of these) is associated with or predicts the severity of microvessel complications is unsettled. Specifically, we test whether combinations of components correlate and predict complications better than individual components.

RESEARCH DESIGN AND METHODS—Correlations and predictions of CGE and complications were assessed in the Rochester Diabetic Neuropathy Study, a population-based, cross-sectional, and longitudinal epidemiologic survey of 504 patients with diabetes followed for up to 20 years.

RESULTS—In multivariate analysis, A1C and duration of diabetes (and to a lesser degree age at onset of diabetes but not FPG) were the main significant CGE risk covariates for complications. A derived glycemic exposure index (GEi) correlated with and predicted complications better than did individual components. Composite or staged measures of polyneuropathy provided higher correlations and better predictions than did dichotomous measures of whether polyneuropathy was present or not. Generally, the mean GEi was significantly higher with increasing stages of severity of complications.

CONCLUSIONS—A combination of A1C, duration of diabetes, and age at onset of diabetes (a mathematical index, GEi) correlates significantly with complications and predicts later complications better than single components of CGE. Serial measures of A1C improved the correlations and predictions. For polyneuropathy, continuous or staged measurements performed better than dichotomous judgments. Even with intensive assessment of CGE and complications over long times, only about one-third of the variability of the severity of complications is explained, emphasizing the role of other putative risk covariates.

Chronic glycemic exposure (CGE) (the degree and duration of plasma hyperglycemia) is thought to be the important modifiable risk covariate for the complications of diabetes (13). This view comes from cohort studies (48) and from trials with clamping hyperglycemia at two levels of glycemic control (912). Although it has been debated whether a CGE threshold exists (13,14), such a threshold is assumed in estimating the lowest level of CGE that induces complications. This level of CGE may then be used to set minimal criteria for the diagnosis of diabetes itself (1,2,13). At issue, however, is how CGE should be estimated. Orchard et al. (13,15) evaluated the severity and duration of hyperglycemia and complications in young people with type 1 diabetes, but the CGE variable that was preset (the percentage of HbA1c [A1C] was more than or equal to the minimal criteria for diabetes times duration of diabetes in months) did not predict complications any better than its components.

Here, we study CGE and complications evaluated intensively and comprehensively in the Rochester Diabetic Neuropathy Study (RDNS). The following are specific questions addressed: Does a combination of significant CGE variables correlate and predict microvessel complications better than individual components? Are the correlations and predictors dependent on how frequently fasting plasma glucose (FPG) and A1C are measured? Is these a dependence on how severity of polyneuropathy is assessed? Is a continuous quantitative measurement of polyneuropathy better than a dichotomous judgment of whether polyneuropathy is present or not?

Included in the RDNS cohort are all consenting individuals in Rochester (later Olmsted County), Minnesota, with diabetes by the National Diabetes Data Group (and later by the American Diabetes Association) criteria as of 1 July 1986. This prospective cross-sectional and longitudinal study assesses the prevalence and incidence of complications and their risk covariates. The cohort of 504 subjects (327 subjects seen on two or more occasions) is mainly of northern European extraction. By the criteria of comorbidity, there were no significant differences between consenting and nonconsenting patients <70 years old (16). The RDNS Normal Subject Cohort of 430 subjects, of whom 330 did not have neurologic disease or disease predisposing to polyneuropathy, was described previously (17,18).

Diabetes complication end points evaluated

All neuropathic end points assessed were described previously (1921). Neurologic signs and symptoms were entered into the Clinical Neuropathy Assessment, allowing entry and interactive surveillance before entry into the database. Severity of neurologic signs was calculated using the Neuropathy Impairment Score (NIS); severity of symptoms was calculated by Neuropathy Symptoms and Change (NSC) (22,23). Neuropathic signs, symptoms, and test abnormalities were scored independently on each occasion without reference to previous or present other examination or test results. Nerve conduction results were expressed in measured units and as normal deviates (from percentiles) having corrected values for applicable variables (e.g., age, height, and weight) (18). Vibration, cooling, and heat pain (5, 0.5, and 5–0.5) thresholds were assessed on the left great toe or foot using CASE IV (WR Medical Electronics, Stillwater, MN) with thresholds expressed as just noticeable difference steps, measured units, percentiles, and normal deviates, with corrections being made for applicable variables (18). Assessment of patients by the Diabetes Control and Complications Trial (DCCT) criteria for polyneuropathy (two of three of decreased or absent ankle reflexes or vibration sensation of the great toes or symptoms of polyneuropathy not attributable to any other condition than diabetes) was based on information recorded in the Clinical Neuropathy Assessment. Complications of polyneuropathy, retinopathy, and nephropathy were staged into three categories of severity (none, mild or intermediate, and severe) as shown in Table 2 and Fig. 2. For polyneuropathy, five other measures of severity were also assessed (Table 2).

Severity of retinopathy was staged on the basis of masked grading of seven 30-degree color stereoscopic fundus photographs of each eye using the modified Airlie House classification and the Early Treatment Diabetic Retinopathy Study severity scale at the University of Wisconsin Ocular Epidemiology Reading Center (R.K.). Nephropathy was staged as N0 = microalbuminuria <30 mg/24-h urine collection, N1 = microalbuminuria ≥30 to <300 mg/24-h urine collection, and ≥N2 = macroalbuminuria ≥300 mg/24-h urine collection (or a previous history of end-stage kidney disease).

Measures of components of CGE

The duration of diabetes was ascertained by questioning of patients and review of their medical records. Type of diabetes was determined by the C-peptide response to glucagon stimulation (16,24). Assessed at first examination and at 3-month intervals (and beginning in 2003 at 6-month intervals) were FPG and A1C (as calculated from GHbA1) (using DCCT standards).


In testing for associations between putative risk covariates (e.g., FPG or A1C) and various complications, both baseline and averaged (calculating the mean value per year and then the mean of the annual values) values were considered independent variables; severity of complications (threshold or a continuous or staged level of abnormality at onset and at last evaluation) were the dependent variables. Associations between severity and risk covariates were evaluated univariately with Spearman rank correlations for quantitative risk factors and with rank-sum tests for dichotomous risk factors. Variables were then inspected for departures from a normal distribution, and appropriate transformations were made as needed for a multivariate assessment. Stepwise regression (stepping up) was used for the multivariate analysis, and the criterion for inclusion of a variable in the model was P < 0.05.

Demographic and disease characteristics

These are summarized in Tables 1 and 2. Of 504 subjects with diabetes entered into study, 327 had subsequent serial evaluations. By the criteria of CGE (duration of diabetes, FPG, and A1C) or staged severity of complications (retinopathy, polyneuropathy, or nephropathy), a significant difference was not found among patients who left the study after one examination and patients who continued in the study (Table A available from the authors on written request). FPG and A1C had been measured frequently (median 27 times). Using A1C as a measure of CGE, our cohort did not have the degree of control recommended by the American Diabetes Association (A1C <7.0%); however, the mean of the A1C values averaged over time to the last examination was slightly and significantly better than at baseline. The prevalence of complications was highest for retinopathy, followed by polyneuropathy, and then by nephropathy (Table 2). The prevalence of polyneuropathy, however, depended critically on which end point criterion was used for diagnosis; the order from highest to lowest frequencies was the sum score of five attributes of nerve conduction (Σ 5 NC nds), NIS (lower limb [LL]), Σ quantitative sensation test (QST) nds, neuropathy symptoms (NSC or Neuropathy Symptoms Score), the DCCT criteria, and HP-DB (Table 2). For all measures of complications, the frequency was higher at the last examination. The comparable frequency of retinopathy increased from 54 to 74% and nephropathy increased from 27 to 38%.

Modeling of components of CGE and complications

In univariate analysis, duration of diabetes (year)1/4, and A1C (percent)1/4 at baseline were positively and significantly associated with most measures of complications at baseline (correlations) (Table B available from the authors on written request). Age at onset was negatively and less frequently also significantly correlated with complications. By contrast, FPG was seldom correlated with complications. Duration of diabetes and A1C at baseline significantly predicted complications at the last examination. When A1C was averaged over time, the prediction of complications was strengthened. By contrast FPG, even when averaged over time, seldom predicted complications at the last examination (Table A available from the authors on written request).

Derivation of a glycemic exposure equation

To develop a glycemic exposure regression equation, we included the statistically significant components of A1C, duration of diabetes, and age at onset of diabetes and plotted these values for type 1 or type 1 and 2 diabetic patients on Σ 5 NC nds to obtain a common regression equation. We chose the composite score Σ 5 NC nds as the best measure for this purpose because it is an objective and continuous measure without an obvious basement or ceiling effect and reflects differences within the range of normal values.

In linear regressions of the glycemic exposure equation (using combinations of A1C1/4, duration of diabetes [years]1/4, and age at onset [years]1/4), by plotting the GE index (GEi) of each patient on their Σ 5 NC nds (Fig. 1), values appear to be normally distributed around a common regression line. The three components of CGE were all included in the regression model, even though not all components were significant for all combinations of CGE and complications. The regression equation for type 1 diabetic patients at the last visit was GEi averaged over time = 4.70 + 0.043 × Σ 5 NC nds at the last visit. For all diabetic patients, the GEi averaged over time = 4.85 + 0.026 × Σ 5 NC nds at the last visit. Using the values for type 1 diabetic patients (because age of onset is more certain), we derived a general GE equation: −2.79 + 3.8 · A1C1/4 + 0.75 × duration of diabetes (years)1/4 + −0.07 · onset of diabetes (year)1/4. The correlation coefficient (R2) was only slightly better when A1C was averaged many times.

Multivariate analysis of components and the index of CGE and complications

To test whether the GEi correlates and predicts microvessel complications better than individual components, we performed multivariate analysis of significant CGE components and the index (calculated at baseline or as averaged over time) on complications at baseline and at last examination. With few exceptions, the GEi was the significant covariate for complications (Table 3). This was especially true when A1C was averaged over the duration of the study, and complications were assessed at the last evaluation. In only three cases was A1C1/4 the sole significant multivariate variable predicting complications, whereas GEi was the sole covariate 18 times. FPG averaged over time was the significant covariate only once (for NIS [LL]) (Table 3). The table also provides information about the degree of the variability explained by the CGE variables. Considering all patients with diabetes and Σ 5 NC nds as complications, 28% of the variability is explained by the GEi. The comparable figure for retinopathy is 31%. For type 1 diabetes, the percentages are higher, 42 and 43%, respectively. For nephropathy GEi, was a significant factor for type 2 diabetes but explained only 8% of the variability of the data.

In Fig. 2, we provide the GEi (25th, 50th, and 75th percentiles and ranges) by staged severity of complications. With a few exceptions, the GEi was significantly greater with increasingly higher stages.

The RDNS data are suited for modeling CGE and complications because 1) risk covariates and complications are prospectively and quantitatively studied for this purpose at regular and frequent intervals over many years and not at times of intercurrent illness, which if done, might have affected results; 2) the cohort is representative of diabetic patients of northern European extraction and includes both type 1 and 2 diabetic patients of both sexes and all ages, allowing inferences to be applied broadly; and 3) bias was minimized by use of independent assessment of complications.

Considering both correlations and predictions of complications, a combination of the three significant components of CGE (expressed as GEi) performed better than any one component alone. This result is different from that of Orchard et al. (13) but may be explained by differences in the choice of patients (in our studies all patients with diabetes and all degrees of severity) and differences in assessment of CGE and complications. In their study, CGE was predetermined as a variable emphasizing duration (in months) of A1C percent above a diagnostic level; in our studies, the actual contribution of the components was calculated from regression equations. The complication of polyneuropathy was also more comprehensively evaluated in our study in which the use of a continuous measure or staged severity of polyneuropathy provided greater power in assessing correlations and predictions.

The important insight that correlations and predictions depend not only on how well CGE is estimated but also on how complications are assessed needs emphasis. All complications in our studies were assessed using standard quantitative measures using reference values and staging. For diabetic sensory polyneuropathy (DSPN), severity was expressed as continuous measures of abnormality of nerve conduction (Σ 5 NC nds), summated neurological signs of the lower limbs (NIS [LL]), symptoms (NSC [LL]), and abnormality of QST (Σ QST). To avoid bias, all complications were independently assessed without reference to previous or concurrent other evaluations.

The use of continuous quantitative measures of DSPN provided stronger associations and predictions with CGE than did use of a dichotomous judgment of the presence or absence of polyneuropathy (e.g., the DCCT criterion). However, the Σ 5 NC nds performed only slightly better than did NIS (LL).

Considering any diabetic patients and all microvessel complications, we suggest that the GEi might be used to express the CGE of any patient. The GEi may be calculated from values that pertain at present. In this case, a single measure (or the mean of several measures) of A1C, duration of diabetes (at the present time), and age at onset of diabetes are used in the equation. The GEi can also be calculated for a future time (as a prediction) knowing age at onset of diabetes, duration of diabetes at a future time, and assuming that the A1C remains unchanged. The GEi can also be calculated, assuming better or worse A1C values.

What are the implications of the present studies for an understanding of microvessel complication and their management? First, glycemic exposure is an important correlate and predictor of microvessel complications. This fact is clinically relevant because A1C and weight are potentially modifiable. However, the fact that the A1C of our patients remained relatively unchanged over the years, despite the emphasis of physicians on weight loss and improved glycemic control during the period of study, may be discouraging. The data obtained here also provides some information, albeit incomplete, about the equivalency of the degree of hyperglycemia and the duration of diabetes in development of complications. It would be of interest to know whether very high A1C levels for short times induce the same complications as mildly elevated A1C levels for long times. Our data suggest a rough equivalency, but further studies focused on this issue are needed. In studies done on cats, we found that severe hyperglycemia for short durations can cause severe nerve injury (25).

How well does the GEi correlate with and predict complications? As shown in Fig. 2, the GEi is usually significantly higher with increasing stages of severity of complications. The overlap of values makes clear that it does not correlate and predict complications exactly. However, the correlative and predictive information is sufficient to be used to encourage patients to lower their A1C levels by loss of weight or treatment. Several reasons might be given for the fact that our correlations and predictions were not higher or better: inaccurate measurement of CGE and complications, nonlinear effects of CGE on complications, and the putative role of other mechanisms for complications (7). For example, it is likely that genetic mechanisms modulate the adverse effects of CGE by various metabolic pathways, and there is recent evidence suggesting such mechanisms for type 2 diabetes (26,27). Although we tried to exclude such cases, immune or mechanical events could also be implicated in complications (28).

This work was supported in part by a grant obtained from the National Institute of Neurological Disease and Stroke (NINDS 36797).

We gratefully acknowledge the help of Mary Lou Hunziker in preparation of the manuscript.

The Expert Committee on the Diagnosis and Classification of Diabetes Mellitus: Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus.
Diabetes Care
The Expert Committee on the Diagnosis and Classification of Diabetes Mellitus: Follow-up report on the diagnosis of diabetes mellitus.
Diabetes Care
Brownlee M: Biochemistry and molecular cell biology of diabetic complications.
Pirart J: Diabetes mellitus and its degenerative complications: a prospective study of 4,400 patients observed between 1947 and
. Part 2.
Diabetes Care
Porte DJ, Graf RJ, Halter JB, Pfeifer MA, Halar E: Diabetic neuropathy and plasma glucose control.
Am J Med
Maser RE, Steenkiste AR, Dorman JS, Nielsen VK, Bass EB, Manjoo Q, Drash AL, Becker DJ, Kuller LH, Greene DA, Orchard TJ: Epidemiological correlates of diabetic neuropathy: report from Pittsburgh Epidemiology of Diabetes Complications Study.
Dyck PJ, Davies JL, Wilson DM, Service FJ, Melton LJ, O’Brien PC: Risk factors for severity of diabetic polyneuropathy: intensive longitudinal assessment of the Rochester Diabetic Neuropathy Study cohort.
Diabetes Care
Klein R, Klein BE, Moss SE, Wong TY, Hubbard L, Cruickshanks KJ, Palta M: Retinal vascular abnormalities in persons with type 1 diabetes: the Wisconsin Epidemiologic Study of Diabetic Retinopathy: XVIII.
Service FJ, Rizza RA, Daube JR, O’Brien PC, Dyck PJ: Near normoglycaemia improved nerve conduction and vibration sensation in diabetic neuropathy.
Amthor K-F, Dahl-Jorgensen K, Berg TJ, Skard Heier M, Sandvik L, Aagenaes O, Hanssen KF: The effect of 8 years of strict glycaemic control on peripheral nerve function in IDDM patients: the Oslo Study.
DCCT Research Group: The Diabetes Control and Complications Trial (DCCT): design and methodologic considerations for the feasibility phase.
DCCT Research Group: The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus.
N Engl J Med
Orchard TJ, Forrest KY, Ellis D, Becker DJ: Cumulative glycemic exposure and microvascular complications in insulin-dependent diabetes mellitus: the glycemic threshold revisited.
Arch Intern Med
Service FJ, O’Brien PC: The relation of glycaemia to the risk of development and progression of retinopathy in the Diabetic Control and Complications Trial.
Forrest KY, Becker DJ, Kuller LH, Wolfson SK, Orchard TJ: Are predictors of coronary heart disease and lower-extremity arterial disease in type 1 diabetes the same? A prospective study.
Dyck PJ, Kratz KM, Lehman KA, Karnes JL, Melton LJ III, O’Brien PC, Litchy WJ, Windebank AJ, Smith BE, Low PA, Service FJ, Rizza RA, Zimmerman BR: The Rochester Diabetic Neuropathy Study: design, criteria for types of neuropathy, selection bias, and reproducibility of neuropathic tests.
O’Brien PC, Dyck PJ: Procedures for setting normal values.
Dyck PJ, Litchy WJ, Lehman KA, Hokanson JL, Low PA, O’Brien PC: Variables influencing neuropathic endpoints: the Rochester Diabetic Neuropathy Study of healthy subjects (RDNS-HS).
Dyck PJ, Litchy WJ, Daube JR, Harper CM, Dyck PJB, Davies J, O’Brien PC: Individual attributes versus composite scores of nerve conduction abnormality: sensitivity, reproducibility, and concordance with impairment.
Muscle Nerve
Dyck PJ, O’Brien PC, Litchy WJ, Harper CM, Klein CJ, Dyck PJB: Monotonicity of nerve tests in diabetes: subclinical nerve dysfunction precedes diagnosis of polyneuropathy.
Diabetes Care
Dyck PJ, Hughes RAC, O’Brien PC: Quantitating overall neuropathic symptoms, impairments, and outcomes. In
Peripheral Neuropathy
. 4th ed. Dyck PJ, Thomas PK, Eds. Philadelphia, Elsevier,
, p.
Dyck PJ, Turner DW, Davies JL, O’Brien PC, Dyck PJB, Rask CA, the rhNGF Group: Electronic case-report forms of symptoms and impairments of peripheral neuropathy.
Can J Neurol Sci
Dyck PJ, Boes CJ, Mulder D, Millikan C, Windebank AJ, Dyck PJB, Espinosa R: History of standard scoring, notation, and summation of neuromuscular signs: a current survey and recommendation.
J Peripher Nerv Syst
Dyck PJ, Karnes JL, Daube J, O’Brien P, Service FJ: Clinical and neuropathological criteria for the diagnosis and staging of diabetic neuropathy.
Dyck PJ, Lambert EH, Windebank AJ, Lais AC, Sparks MF, Karnes J, Sherman WR, Hallcher LM, Low PA, Service FJ: Acute hyperosmolar hyperglycemia causes axonal shrinkage and reduced nerve conduction velocity.
Exp Neurol
Grant SF, Thorleifsson G, Reynisdottir I, Benediktsson R, Manolescu A, Sainz J, Helgason A, Stefansson H, Emilsson V, Helgadottir A, Styrkarsdottir U, Magnusson KP, Walters GB, Palsdottir E, Jonsdottir T, Gudmundsdottir T, Gylfason A, Saemundsdottir J, Wilensky RL, Reilly MP, Rader DJ, Bagger Y, Christiansen C, Gudnason V, Sigurdsson G, Thorsteinsdottir U, Gulcher JR, Kong A, Stefansson K: Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes.
Nat Genet
Raeder H, Johansson S, Holm PI, Haldorsen IS, Mas E, Sbarra V, Nermoen I, Eide SA, Grevle L, Bjorkhaug L, Sagen JV, Aksnes L, Sovik O, Lombardo D, Molven A, Njolstad PR: Mutations in the CEL VNTR cause a syndrome of diabetes and pancreatic exocrine dysfunction.
Nat Genet
Younger DS, Rosoklija G, Hays AP, Trojaborg W, Latov N: Diabetic peripheral neuropathy: a clinicopathologic and immunohistochemical analysis of sural nerve biopsies.
Muscle Nerve

P.C.O.B. receives royalties from WR Medical Electronics.

A table elsewhere in this issue shows conventional and Système International (SI) units and conversion factors for many substances.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.