Klein BEK, Knudtson MD, Tsai MY, Klein R. The Relation of Markers of Inflammation and Endothelial Dysfunction to the Prevalence and Progression of Diabetic RetinopathyWisconsin Epidemiologic Study of Diabetic Retinopathy. Arch Ophthalmol. 2009;127(9):1175-1182. doi:10.1001/archophthalmol.2009.172
To determine the relation of glycemia, blood pressure, and serum total cholesterol level as systemic markers of inflammation and endothelial dysfunction to the prevalence and incidence of diabetic retinal outcomes in persons with long-duration type 1 diabetes mellitus.
Longitudinal population-based study of persons with type 1 diabetes mellitus who received care for their diabetes in south central Wisconsin from July 1, 1979, to June 30, 1980. Data for this investigation were obtained from the 1990-1992 through the 2005-2007 follow-up examinations. Main outcome measures included the severity of diabetic retinopathy (DR) and macular edema (ME).
In the 1990-1992 prevalence data, soluble vascular cell adhesion molecule, tumor necrosis factor, and homocysteine levels were associated with increased odds of more severe DR (odds ratios [highest vs lowest quartile], 3.95 [95% confidence interval, 1.66-9.39], 5.46 [2.38-12.52], and 7.46 [2.91-19.16], respectively) in those with kidney disease while controlling for relevant confounders. Similar odds were found for proliferative DR. Only total homocysteine level was associated with increased odds of ME (3.80 [95% confidence interval, 1.91-7.54]), irrespective of kidney disease. None of the markers were associated with incidence of proliferative DR, ME, or progression of DR 15 years later.
A limited number of markers are associated with increased odds of prevalent retinal outcomes in persons with type 1 diabetes mellitus and kidney disease. Only homocysteine level is associated with ME in those with and without kidney disease. In the absence of kidney disease, the markers do not add to the more conventional descriptors and predictors of DR in persons with type 1 diabetes mellitus. This may reflect the close association of DR and kidney disease in diabetic persons.
Data from epidemiological studies and clinical trials have shown associations of hemoglobin A1c level, blood pressure, and serum total cholesterol level with the prevalence, incidence, and progression of retinopathy.1- 6 However, these factors only explain a small proportion of the presence and 14-year cumulative progression of retinopathy (R2 = 9%) and incidence of proliferative retinopathy in persons with type 1 diabetes mellitus (R2 = 10%) (R.K., unpublished data from the 14-year follow-up of the Wisconsin Epidemiologic Study of Diabetic Retinopathy cohort, February 1995 to June 1996).
Increased levels of systemic inflammatory markers have been found in persons with diabetes.7,8 However, few data have examined the role of systemic inflammation in diabetic retinopathy (DR) in persons with type 1 diabetes mellitus.8- 12 In 2 case-control studies, diabetic subjects with macular edema (ME)12 or proliferative DR (PDR)11 had higher levels of vascular endothelial growth factors and cytokines in their vitreous than those without ME or PDR. In a cross-sectional study of normotensive persons with type 1 diabetes mellitus, serum C-reactive protein (CRP) and fibrinogen levels were positively associated with the severity and progression of DR.8
Endothelial dysfunction may result in increased vascular permeability, alteration of blood flow, oxidative stress, angiogenesis, and DR.13- 19 Endothelial dysfunction is characterized by elevated levels of soluble vascular cell adhesion molecule (sVCAM-1) and soluble intercellular adhesion molecule (sICAM-1). In addition, homocysteine (Hcy) levels, which have been found to be elevated in persons with type 1 diabetes mellitus, have been shown to damage endothelial cells via generation of hydrogen peroxide.20,21 Homocysteine and other markers of endothelial dysfunction have been shown to predict macrovascular diseases in diabetic and nondiabetic persons, but they have not been uniformly associated with DR.8,22- 29 Leukocyte adherence to retinal endothelium may be a cause of capillary occlusion and an important factor in the pathogenesis of DR.30 Adherence of leukocytes to capillary and arteriolar endothelium occurs as a result of a process involving expression of adhesion molecules (eg, selectins and sICAM-1) and sVCAM-1.31- 38 Levels of sVCAM-1 have been shown to be elevated in patients with type 1 diabetes mellitus who have more severe retinopathy.30,39 However, no prospective or population-based data have shown that elevated concentrations of these molecules precede the progression of DR or incidence of ME or whether these associations remain while controlling for hemoglobin A1c level, blood pressure, and signs of diabetic nephropathy. In addition, aspirin therapy has been shown to prevent some histopathological signs of DR (eg, acellular capillary formation and retinal hemorrhage) in dogs with experimentally induced diabetes.40 Zheng et al41 reported a beneficial effect of salicylates on retinopathy in a rat model of diabetes presumably related to inhibition of nuclear factor κβ. We investigated associations of selected markers of inflammation that have been commonly associated with systemic disease (high-sensitivity CRP [hsCRP], interleukin 6 [IL-6], and tumor necrosis factor [TNF]) and with endothelial dysfunction (sVCAM-1, sICAM-1, and Hcy) with diabetic retinal outcomes in a well-characterized large cohort of persons with type 1 diabetes mellitus who have been followed up for a long period and report our findings herein.
The population consisted of a sample selected from 10 135 diabetic patients who received primary care in an 11-county area in southern Wisconsin from July 1, 1979, to June 30, 1980.1,42- 47 The analyses in this report are limited to the group of patients who had onset of the diseases at a younger age, all of whom were taking insulin and had been diagnosed as having type 1 diabetes mellitus before 30 years of age. Of the 1210 such persons, 996 participated in the baseline examination (1980-1982),43 903 in the 4-year follow-up examination (1984-1986),46 816 in the 10-year follow-up examination (1990-1992),47 667 in the 14-year follow-up examination (1994-1996),1 567 in the 20-year follow-up examination (2000-2002),45 and 520 in the 25-year follow-up examination (2005-2007).48 The reasons for nonparticipation and comparisons between participants and nonparticipants at the various examinations have been presented elsewhere.1,43,45- 48 Frozen serum samples were available from the time of the 10-year examination. Analyses in this report are limited to patients with disease onset at a younger age who had frozen serum samples available from that examination and had information about prevalent retinopathy or progression of retinopathy at the 14-year and/or the 25-year follow-up examination. Because of funding constraints, there were no photographs taken at the 20-year examination.
The examinations were performed in a mobile examination van in or near the city where the participants resided. All examinations followed a similar protocol that was approved by the institutional human subjects committee of the University of Wisconsin and conformed to the tenets of the Declaration of Helsinki. The pertinent parts of the ocular and physical examinations included measuring weight, height, and blood pressure49; dilating the pupils; taking stereoscopic color fundus photographs of 7 standard fields50 (not performed at the 20-year follow-up examination); and performing a semiquantitative determination of protein levels in the urine with the use of reagent strips (Labstix; Ames, Elkhart, Indiana).
A structured interview was conducted that included questions about the use of specific medications for control of hyperglycemia and blood pressure, regular use of aspirin or other nonsteroidal anti-inflammatory drugs (NSAIDs), and smoking history. If there was any question about medication use, the patient report was verified by a physician's report. Examinations and administration of the interview were performed by trained examiners. Quality control was monitored throughout each study phase.
An aliquot of whole blood was used for determination of the hemoglobin A1c level by means of affinity chromatography (Isolab, Incorporated, Akron, Ohio).51- 53 The normal range for hemoglobin A1c level was 4.6% to 7.9% (to convert to the proportion of total hemoglobin, multiply by 0.01), with an intra-assay coefficient of variation of 2.4%. We measured serum levels of total and high-density lipoprotein cholesterol.54,55 The remaining serum samples were stored without preservative at −80°C in cryogenic vials with O-rings for up to 16 years, until the vials were shipped on dry ice to the University of Minnesota laboratory for the analyses reported herein. From the frozen specimens obtained at the 1990-1992 examination, an aliquot of serum was used for the determination of serum levels of hsCRP, IL-6, sVCAM-1, sICAM-1, TNF, serum total Hcy (tHcy), and cystatin C.
The level of hsCRP was measured on a chemical analyzer (Hitachi 911; Roche Diagnostics, Indianapolis, Indiana) using a particle-enhanced immunonephelometric assay (CRP  K-ASSAY; Kamiya Biomedical Company, Seattle, Washington). Expected values for hsCRP in normal, healthy individuals are 5 mg/L or less (to convert to nanomoles per liter, multiply by 9.524), with interassay coefficients of variation ranging from 2.1% to 4.5%. We used enzyme-linked immunosorbent assays with quantitative sandwich enzyme immunoassay techniques (R&D Systems Inc, Minneapolis, Minnesota) to measure levels of IL-6, sVCAM-1, sICAM-1, and TNF. The IL-6 level was measured using the Quantikine HS human IL-6 immunoassay kit (R&D Systems Inc). The laboratory coefficient of variation for this assay was 6.1%. The expected normal range per the manufacturer is 0.24 to 12.5 pg/mL. The TNF level was measured using the QuantiGlo Chemiluminescent enzyme-linked immunosorbent assay kit (R&D Systems Inc). The interassay coefficient of variation was 9.1%, with an expected normal range of 0 to 9.0 pg/mL. The sICAM-1 level was measured using the Quantikine human sICAM-1 immunoassay kit (R&D Systems Inc). The laboratory coefficient of variation was 6.7%, with an expected normal range of 115 to 306 ng/mL. The sVCAM-1 level was measured using the Quantikine human sVCAM-1 immunoassay kit (R&D Systems Inc). The laboratory coefficient of variation was 4.5%, with an expected normal range of 349 to 991 ng/mL. Serum tHcy level was measured by a fluorescence polarization immunoassay (IMx homocysteine assay; Axis Biochemicals ASA, Oslo, Norway) using an automated analyzer (IMx analyzer; Abbott Diagnostics, Abbott Park, Illinois). The laboratory coefficient of variation was 6.7%, with an expected normal range of 5 to 12 μmol/L (to convert to milligrams per liter, divide by 7.397).
The cystatin C level was determined on a nephelometer using a cystatin C kit (BN100 nephelometer and N Latex kit; Dade Behring GmbH, Marburg, Germany). The laboratory coefficient of variation was 4.8%, with an expected normal range of 0.53 to 0.95 mg/L.
Photographs for DR were graded by trained individuals using modifications of the Early Treatment Diabetic Retinopathy Study adaptation of the modified Airlie House classification of DR.46,56- 58 Ongoing quality control of grading was performed for each examination (except for the 2000-2002 examination).46,56,58,59
For each eye, the maximum grade in any of the 7 standard photographic fields was determined for each of the lesions and used in defining the Early Treatment Diabetic Retinopathy Study retinopathy levels.47,58
The severity for a participant was derived by concatenating the levels for the 2 eyes, giving the eye with the higher level greater weight. This scheme provided a 15-step scale (retinopathy levels: 10/10, 21/<21, 21/21, 31/<31, 31/31, 37/<37, 37/37, 43/<43, 43/43, 47/<47, 47/47, 53/<53, 53/53, ≥60/<60, and ≥60/≥60) when all levels of PDR are grouped as a single level.
Macular edema was defined as retinal thickening in the macular area according to the Early Treatment Diabetic Retinopathy Study classification protocol. The 15-year cumulative incidence of ME was estimated from all persons who had no ME at the 1990-1992 examination and who participated in the follow-up examinations.
Progression to PDR was estimated from all persons who were free of this complication at the baseline examination. For persons with no PDR or only nonproliferative DR, progression was defined as the first instance of an increase in the severity of DR by 2 steps or more from the baseline level at any of the follow-up examinations.
Age was defined as the age at the time of the 1990-1992 examination. Age at diagnosis of diabetes was defined as the age at the time the diagnosis was first recorded by a physician on the patient's medical chart or in a hospital record. The duration of diabetes was the difference between the age at the 1990-1992 examination and the age at diagnosis. Systolic and diastolic blood pressures were the average of the last 2 of 3 measurements taken according to the protocol of the Hypertension Detection and Follow-up Program.50 Body mass index was defined as the weight in kilograms divided by the height in meters squared. Proteinuria was defined as a urine protein concentration of 0.03 g/dL or greater (to convert to grams per liter, multiply by 10) as measured with the reagent strips. Kidney disease at the 1990-1992 examination was defined as undergoing dialysis or receipt of a kidney transplant, proteinuria of 0.03 g/dL or greater, or a cystatin C level in the fourth quartile. Only 43 (17%) cases of kidney disease as defined herein were included solely on the basis of cystatin C level in the highest quartile (0.97-9.18 mg/L).
In these analyses, we considered the retinal outcomes and their associations with the inflammatory markers in cross-sectional and incidence analyses. Statistical methods included calculations of means, frequencies, correlations, and multivariate models. We used commercially available statistical software (SAS, version 9.1; SAS Institute Inc, Cary, North Carolina) for all analyses.
In cross-sectional analyses, we considered 3 outcomes. Severity of DR was categorized into 5 groups based on the level in the worse eye (10, 21, 31-37, 43-53, and ≥60). This outcome was analyzed with proportional odds models. The other 2 outcomes, prevalence of PDR and ME, were dichotomous and analyzed with logistic regression.
In incidence analyses, we considered the following 3 outcomes: the incidence of PDR, a 2-step or greater progression of DR, and the incidence of ME. We used generalized linear models for these binary outcomes (incidence during the examination interval) with the complementary log-log link function to estimate an underlying continuous-time proportional hazards model while accounting for the varying follow-up times between examinations (4 years from the 1990-1992 to the 1994-1996 examination and 11 years from the 1994-1996 examination to the 2005-2007 examination). For these analyses, duration of diabetes was the time variable, and the baseline hazard was assumed to be piecewise constant within 5-year bands of diabetes duration starting at 20 years and continuing to longer than 40 years. Hazard ratio estimates were calculated by exponentiation of estimated coefficients. We used statistical software (PROC NLMIXED of SAS, version 9.1; SAS Institute Inc) for these analyses.
The distributions of the markers were highly skewed (data not shown). Therefore, we divided the distributions into quartiles (Table 1). For the sake of space, we contrasted quartile 4 (the highest) with quartile 1 (the lowest). In nearly all cases, the contrasts for the intermediate quartiles were not significant. We included only 1 marker in each model because of our interest in evaluating each separately. Also, many of the markers were highly correlated (eg, for serum hsCRP and IL-6 levels, r = 0.56 [P < .001]) and, therefore, the effect of one might obscure the relative effect of another. Because kidney disease in those with diabetes is strongly associated with severity of DR and because it, too, is associated with inflammatory and endothelial dysfunction markers, we analyzed our data dichotomizing by kidney disease status.
Because of their association with retinopathy in previous publications, we considered the hemoglobin A1c level, body mass index, systolic blood pressure, ratio of serum total to high-density lipoprotein cholesterol level, and duration of diabetes at baseline as potential confounders. Additional models controlled for the use of NSAIDs. For ease of space, we provide only the multivariate-adjusted models.
Patient characteristics at the 1990-1992 examinations are presented in Table 2. The high prevalence of kidney disease reflects the long duration of diabetes (mean, 22.9 years) in this cohort, as does the high prevalence of DR (96.1%) and ME (22.0%) and the incidence of ME (16.0%) and progression of retinopathy (57.3%) by the 15-year follow-up.
We examined the cross-sectional associations between the markers and the retinal outcomes, including duration of diabetes, hemoglobin A1c level, body mass index, ratio of serum total to high-density lipoprotein cholesterol level, and systolic blood pressure at the 1990-1992 examination (Table 3). We tested the associations by contrasting values in quartile 4 compared with those in quartile 1. Results showed an association of sVCAM-1, TNF, and tHcy levels with an increased risk of more severe DR in those with kidney disease but not in those without it. An association of sICAM-1 level with decreased risk was also found. A similar pattern was seen for PDR, with small differences from the previous analysis in the odds ratio for each marker. For ME, only tHcy level was associated with increased risk in those with or without kidney disease.
Similar analyses were performed for incidence of PDR, progression of DR, and incidence of ME at the 2005-2007 examination (Table 4). There was no evidence of increased risk of these end points in association with any of these markers for any outcome. There was decreased risk of progression of DR associated with increased serum IL-6, TNF, and tHcy levels in those with kidney disease and for increased sICAM-1 level in those without kidney disease.
At the 1990-1992 examination, 27.5% of persons with kidney disease were taking NSAIDs compared with 18.1% of persons without kidney disease (P = .005). We repeated the multivariate analyses including use of NSAIDs (only 41 NSAID users were not taking aspirin) with the other variables. There were minimal changes in the odds ratio (or in the hazard ratio) for the incidence data and no changes in the statistical significance compared with the previous models (data not shown).
In persons with long-term type 1 diabetes mellitus and evidence of kidney disease, we found that serum sVCAM-1, TNF, and tHcy levels had a strong cross-sectional association with increasing severity of DR, after controlling for important previously known risk factors for that end point. This finding was also reflected in the cross-sectional association with PDR. These results are important in that they indicate that, even in those with kidney disease, a complication that itself is associated with elevated markers of systemic inflammation and endothelial dysfunction,60 relatively high levels of serum sVCAM-1, TNF, and tHcy are associated with markedly elevated odds ratios for increasing severity of DR (and PDR). We found a significant association of tHcy level with ME in those with and without kidney disease. This suggests that there is a different pathophysiological process associated with retinal thickening compared with specific lesions of DR that is operative in persons with type 1 diabetes mellitus, irrespective of kidney function.
The findings for serum sICAM-1 appear to be paradoxical in light of the findings for serum levels of sVCAM-1, TNF, and tHcy. Other studies have suggested that sVCAM-1 plays a dominant role in the initiation of atherosclerosis.61 In our population, the sICAM-1 values are moderately elevated, whereas the sVCAM-1 values are more markedly elevated. Also, the tHcy values are high in our population. These findings are in keeping with recent studies that demonstrated sVCAM-1 but not sICAM-1 was expressed in TNF-stimulated smooth muscle cells62 and that Hcy stimulated sVCAM-1 but not sICAM-1 expression.63 Thus, our finding supports the earlier findings that, although sVCAM-1 and sICAM-1 are both endothelial adhesion molecules of the IgG gene superfamily, sVCAM-1 is more responsive to stimulation by TNF and may play a more important role in atherosclerosis. Similar physiologic characteristics may apply to microvascular disease as well.
There were no significant associations of the markers we measured in the hypothesized direction on incidence of retinal outcomes 15 years later. We did find decreased odds of progression of retinopathy for IL-6, TNF, and tHcy levels in those with kidney disease and for sICAM-1 levels in those without kidney disease. These results may be related to the striking mortality rates in the whole cohort overall and especially in those with kidney disease (45% in those with kidney disease and 12% in those without kidney disease). Those who survive are likely to differ from those who die with regard to the level of the markers and the relationship of those markers to incident retinal outcomes. Alternatively, this may indicate that these markers are not involved in the progression of DR, or that they might, through some unknown mechanism, protect against progression.
Thus, our findings reinforce the notion that the systemic markers sVCAM-1, TNF, and tHcy levels are strongly associated with the prevalence and severity of DR in those with kidney disease, but that not all the markers we tested were similarly associated. Although we could speculate as to why this occurs, more research is needed to understand whether the markers themselves or, perhaps more likely, the systemic causes of these elevated marker levels, are causally associated with the severity of DR. For example, although kidney disease is characterized by proteinuria or elevated cystatin C levels, it is also associated with systemic acidosis, azotemia, anemia, etc. These may be involved in the pathogenesis of DR rather than the markers we measured.
Despite our expectation of such effects based on work by others,40,41 we found trivial or no changes on the risk of outcomes related to use of NSAIDs. Our data, however, are not ideal to test such effects in that we lack information on the duration of NSAID therapy or the dose of the NSAID. Such data would further our understanding of a potentially modifying effect of such preparations on retinopathy.
Although there are many strengths of our study, there are limitations as well. The cohort was identified in 1979 to 1980, when glycemic and blood pressure controls were less stringent compared with current management practices. The changes occurred early during the course of our follow-up (from the 1980-1982 baseline examination) and began to affect our cohort before the 1990-1992 examination. This is likely to have affected the presence of kidney disease and the level of markers we measured. We cannot assess possible effects of changes in the management of diabetes on the relationships we found, although we did control for current glycemia and blood pressure in our analyses. In addition, our cohort had had diabetes for, on average, 22.9 years by the time of the 1990-1992 examination. Thus, kidney disease was quite common and the duration of that complication is not accounted for in these analyses. We do not know whether this would influence our findings. The high mortality associated with kidney disease in this group (and the known relationship of the markers of interest to mortality) limited our ability to examine effects of the range of the markers on retinal outcomes. Nevertheless, few other data describe these relationships in persons with long-term type 1 diabetes mellitus.
Finally, our specimens were frozen at −80°C for about 15 years before the laboratory assays. Although there may have been some deterioration during that interval, we found significant associations in the prevalence data. We would have therefore expected to find important relationships in the incidence data if they existed. Thus, we doubt that the interval from drawing to assay materially affected our findings, but we cannot be certain of this.
In summary, we found cross-sectional associations of some inflammatory and endothelial markers with severity of DR and ME in persons with long-term type 1 diabetes mellitus. The positive associations of the markers with DR found in the prevalence data were confined to those with kidney disease. There was no evidence that these markers are associated with increased risk of long-term progression of DR or with increased incidence of PDR or ME.
Correspondence: Barbara E. K. Klein, MD, MPH, Department of Ophthalmology and Visual Sciences, University of Wisconsin School of Medicine and Public Health, 610 N Walnut St, Fourth Floor Wisconsin Alumni Research Foundation, Madison, WI 53726 (email@example.com).
Submitted for Publication: October 20, 2008; final revision received January 8, 2009; accepted January 30, 2009.
Author Contributions: Dr B. E. K. Klein 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.
Financial Disclosure: None reported.
Funding/Support: This study was supported by grants EY03083 and EY016379 from the National Institutes of Health (Drs B. E. K. Klein and R. Klein) for the entire study, including collection and analyses of data, and in part by Senior Scientific Investigator Awards from Research to Prevent Blindness (Drs B. E. K. Klein and R. Klein) for data analyses.
Additional Contributions: Naomi Hanson, MS, CLS(NCA), MT(ASCP) (University of Minnesota), provided excellent technical assistance and assistance with manuscript revision. Mary Kay Aprison, BS, assisted in manuscript preparation. Neither of these individuals received any compensation for their contributions.