Survival by severity of diabetic retinopathy in 4 age groups. See the "Methods" section for description of retinopathy categories. Numbers in the figure key represent the number at risk at baseline.
Survival by quartile ranges of arteriovenous ratio (AVR) in 4 age groups. Arteriovenous ratio categories were as follows: quartile 1, 0.590 to 0.808; quartile 2, 0.809 to 0.853; quartile 3, 0.854 to 0.896; and quartile 4, greater than 0.896. Numbers in the figure key represent the number at risk at baseline.
Klein BEK, Klein R, McBride PE, Cruickshanks KJ, Palta M, Knudtson MD, Moss SE, Reinke JO. Cardiovascular Disease, Mortality, and Retinal Microvascular Characteristics in Type 1 DiabetesWisconsin Epidemiologic Study of Diabetic Retinopathy. Arch Intern Med. 2004;164(17):1917-1924. doi:10.1001/archinte.164.17.1917
Copyright 2004 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2004
Diabetic retinopathy and proteinuria, manifestations of microvascular abnormalities, occur early in the course of diabetes mellitus; in contrast, macrovascular cardiovascular complications usually occur later. Retinal vessel characteristics may be informative about risk of cardiovascular disease in persons with diabetes. We evaluated this in a longitudinal cohort study of persons with type 1 diabetes.
The population consisted of persons with type 1 diabetes who were receiving care in 11 counties in Wisconsin. Subjects (n = 996) were examined at baseline (1980-1982), and 4, 10, 14, and 20 years later. Evaluations included medical history and measurements of height, weight, blood pressure, and glycosylated hemoglobin. Fundus photographs were graded for diabetic retinopathy at baseline, and the same photographs were graded later for the diameters of retinal blood vessels. At each examination, a history of cardiovascular disease events since the last examination (and prior to baseline) was obtained. Mortality was monitored yearly.
The 20-year age-adjusted cumulative incidences were 18.1% for angina, 14.8% for myocardial infarction, and 5.9% for stroke. Severity of diabetic retinopathy was associated with angina and stroke. Arteriovenous ratio was associated with myocardial infarction. Of 273 deaths, 176 involved heart disease. The severity of retinopathy and arteriovenous ratio was associated with heart disease mortality. Nephropathy was more informative about the cardiovascular end points than were the blood vessel characteristics.
Incidences of cardiovascular disease, including mortality, were common in people with type 1 diabetes during a 20-year interval. Retinal vascular characteristics were associated with these end points, but this association was confounded by nephropathy.
Cardiovascular disease is a well-known complication of diabetes,1- 5 and microvascular characteristics have been hypothesized to contribute to it.6 Some studies have shown that generalized and focal narrowing of retinal arterioles and arteriovenous crossing changes were associated with increased cardiovascular disease in persons with hypertension.7- 9 The odds of having a cerebral infarction was 2.3 (95% confidence interval, 1.2-4.4) for low arteriovenous ratio (AVR).10 The purpose of the current study is to describe the incidence of cardiovascular outcomes and of heart disease mortality as related to diabetic retinopathy and ratio of arteriole to venule diameters as graded from images taken in 1980 to 1982 in persons with type 1 diabetes mellitus.
Of 1210 persons receiving care for type 1 diabetes in a defined area of Wisconsin in 1979 to 1980, 996 participated in a study of diabetic retinopathy.11 The institutional human subjects committee approved the study, and informed consent was obtained from each participant at each study visit. The participants had C-peptide levels less than 0.9 ng/mL (0.3 nmol/L).12 The baseline evaluation included a medical history; measurements of blood pressure, glycosylated hemoglobin,13 height, and weight; dilation of the pupil; and photography of the ocular fundus of each eye. Because we were interested in associations of observations taken at one time to subsequent end points, we used only baseline measures of risk variables.
Fundus photographs were graded for lesions of diabetic retinopathy by means of a modification of the Early Treatment Diabetic Retinopathy Study protocol.14,15 The retinopathy level for a participant was derived by concatenating the levels for the 2 eyes, giving greater weight to the eye with the higher level.16 This scheme provided a 15-step scale,17 which, for most analyses, was categorized into 4 groups: 1, no retinopathy; 2, early nonproliferative; 3, moderate to severe nonproliferative; and 4, proliferative.
Retinal microvascular measurements were taken from digitized reproductions of field 1.14 Computer-assisted measurements of arterioles and venules from a designated area of the fundus were each combined to provide "central equivalents" for the arteriole and venule trunk vessels.18- 20 The central arteriolar and venular equivalents were then expressed as the AVR. An AVR of 1.0 indicates that, on average, retinal arteriole diameters were the same as retinal venule diameters. Smaller values of AVR were thought to represent relative arteriole constriction, although it is possible that at least some venous dilation may play a role in determining the value of the ratio. The distributions of the central retinal artery and vein equivalents and AVR are given in Table 1. The mean value for the 2 eyes for 871 persons was used in the analysis. The mean (SD) values in the cohort were as follows: central retinal artery equivalent, 208.1 µm (22.1 µm); central retinal vein equivalent, 246.3 µm (21.9 µm); and AVR, 0.85 (0.07).
Follow-up evaluations of the cohort occurred 4, 10, 14, and 20 years after the baseline evaluation. At each examination, subjects were asked whether a physician had told them that they had had a myocardial infarction, angina, or stroke. The first report of each event was used in our analysis; medical record confirmation of these events was not obtained at the time that they were reported.
Yearly telephone contact with study subjects or a contact person was made to monitor vital status. Local newspapers were monitored for obituaries of study subjects. We obtained death certificates for those known to have died. When we were unable to contact a subject, and the contact persons were unaware of vital status or last known address or death, we submitted the subject's name to the National Death Index. We ascertained that 273 of the 996 participants had died by 2001. There were 27 subjects lost to follow-up, and an additional 101 who were known to be alive but refused participation in follow-up examinations.
We used death certificate diagnoses for cause of death in our survival analyses. Any mention on the death certificate of diseases of the heart ("any heart") was included as cardiovascular death. The corresponding International Classification of Diseases, Ninth Revision, codes were 402, 404, and 410-429, and International Classification of Diseases, 10th Revision, codes were I20-I51, with the total number of such deaths being 176.
Characteristics at baseline included in our analysis were age, sex, duration of diabetes, glycosylated hemoglobin level, blood pressure, hypertension, smoking status (never, past, or current), pack-years smoked, presence of neuropathy (a positive response to any or all questions on loss of tactile sensation, numbness or tingling in the hands, and loss of temperature sensitivity), daily aspirin use, body mass index (calculated as weight in kilograms divided by the square of height in meters), nephropathy (none, proteinuria, self-reported dialysis, or transplant), pulse rate, pulse pressure (systolic-diastolic blood pressure in millimeters of mercury), and the retinal vascular characteristics.
Analyses were performed with SAS version 8.01 (SAS Institute Inc, Cary, NC).21 Incidence analyses were performed with the Kaplan-Meier life table approach.22 Odds ratios for the multivariable models using the angina, myocardial infarction, and stroke end points were computed by means of discrete linear logistic models.23 Kaplan-Meier survival procedures were used for cardiovascular death.22 Cox proportional hazards regression was used for the multivariable models.24 For all analyses, P values less than .05 were considered significant, while P values of .05 to .10 were considered marginally significant.
Of the 996 subjects who were seen at baseline, 918 provided at least an interview at the 4-year follow-up, 815 at the 10-year follow-up, 699 at the 14-year follow-up, and 652 at the 20-year follow-up. With respect to deaths, 64 died before the second examination, 86 more died before the third examination, and 64 more died before the fourth examination. The remainder of the deaths (59) occurred after the fourth examination. In general, those not providing data after baseline tended to be older, had longer duration of diabetes, had hypertension, and were more likely to be male than those returning for follow-up evaluations; glycosylated hemoglobin levels were not different from those of subjects who continued in the study.25 The prevalence of angina was 3.1%; myocardial infarction, 3.0%; and stroke, 1.0%. Persons with a prevalent condition at baseline were not considered for incidence of that end point. The 20-year age-adjusted cumulative incidence of angina was 18.1%; myocardial infarction, 14.8%; and stroke, 5.9%. Table 1 describes the distributions of characteristics we considered as possible correlates of incidence of the cardiovascular outcomes. Age, duration of diabetes, systolic blood pressure, hypertension, smoking, history of neuropathy, body mass index, pulse pressure, retinopathy, and AVR were significantly related to all end points. Diastolic blood pressure, nephropathy, and daily aspirin use were significant for 2 outcomes. Serum cholesterol levels were not available at the baseline examination.
All significant factors for each outcome were included in developing the multivariable model. A stepwise regression procedure was used to eliminate the nonsignificant factors. When all significant factors except the ocular measures were included in 20-year cumulative incidence multivariable logistic models, the models were similar for myocardial infarction and angina (Table 2, model A). For stroke, smoking was not significant, whereas glycosylated hemoglobin and daily aspirin use were significant.
We added ocular factors to the model; first, severity of diabetic retinopathy (Table 2, model B) and, in a separate analysis, AVR (Table 2, model C). This was done because the Spearman rank correlation coefficient between AVR and severity of retinopathy (full scale) was −0.46 (P<.001). In these multivariable models, the severity of retinopathy was associated with increased odds of all 3 cardiovascular disease outcomes, but was of borderline significance for myocardial infarction. Lower AVR was associated with increased odds of all the outcomes, but was significant only for myocardial infarction.
Survival plots by severity of any retinopathy (Figure 1) indicated that more severe retinopathy was associated with greater mortality in all age groups. This was most apparent when stratum 4 (proliferative retinopathy) was compared with the other groups. Survival plots by quartile of AVR suggested that it was not related to mortality except in the oldest age group, when those with the smallest AVR had the poorest survival experience (Figure 2).
Multivariable analyses were performed for mortality involving heart disease (any heart) including significant systemic variables first, before ocular characteristics were added (Table 3, model A). We did not consider nephropathy in this first analysis because of the possible confounding of the retinal vascular characteristics–heart disease mortality relationship by nephropathy. The significant factors were age, sex, duration of diabetes, glycosylated hemoglobin level, diastolic blood pressure (per 10 mm Hg), neuropathy, hypertension, and history of cardiovascular disease; smoking was of borderline significance. When category of severity of retinopathy was included, there was a significant trend of increased risk with increasing severity (P = .004) (Table 3, model B). When quartile of AVR was added to the model instead of severity of retinopathy, there was a significant trend of increased risk with decreased AVR (P = .01) (Table 3, model C). If renal insufficiency was included as a variable instead of the severity of the ocular variables, it increased mortality risk significantly (P<.001) (data not shown).
Our data are consistent with the possibility that microvascular characteristics in those with type 1 diabetes mellitus precede macrovascular disease, as reflected by cardiovascular disease and mortality involving heart disease. The severity of diabetic retinopathy appeared to be more consistently related to the outcomes we examined than did AVR. Diabetic retinopathy and AVR were highly correlated and may be variant manifestations of the same process. Both were related to duration of diabetes, blood pressure, and diabetic nephropathy.14,25 In those without diabetes, AVR, as well as retinal blot hemorrhages and microaneurysms, were associated with blood pressure,7,9 so our findings in those with diabetes were not novel. It may be that the hyperglycemia of diabetes was another insult to the retina or it exacerbated effects of blood pressure, resulting in hemorrhages, microaneurysms, and more severe lesions of retinopathy; this may also have affected AVR–blood pressure relationships. In those without diabetes in the Atherosclerosis Risk in Communities Study,7- 10,26 AVR and (nondiabetic) retinopathy were related to cardiovascular events. Wong et al8 found that AVR was associated with coronary heart disease in women, but not in men, in the Atherosclerosis Risk in Communities Study. In our population, we found no difference in the relationships of the retinal vascular characteristics to outcomes by sex. The reasons for this are not apparent.
The population enrolled in our study had risk factors for incident cardiovascular events and mortality from heart disease that were similar to those found in other population-based studies of persons with diabetes (eg, age, blood pressure, smoking, and duration of diabetes).27- 32 In the univariable and multivariable models, glycosylated hemoglobin was not significantly associated with angina and myocardial infarction. The Pittsburgh Epidemiology of Diabetes Complications Study also found no significant relationship between glycemia and coronary disease.32 In our analysis, the level of glycemia in those with diabetes was significantly associated with incidence of stroke. Despite the relative infrequency of stroke, the confidence limits were not wide (odds ratio, 1.2; 95% confidence interval, 1.01-1.50). Diabetes has previously been shown to increase the risk of stroke,33- 36 but we are unaware of any previous study that has found the risk in people with diabetes to be related to level of glycemia.
Self-reported symptoms of sensory neuropathy were significantly associated with all 3 cardiovascular events and mortality. We chose to include this variable as a possible risk factor for 2 reasons. One is that it is common; its prevalence tends to increase with age37 and duration of diabetes37,38 in those with type 2 diabetes mellitus, and it may serve as a marker of general complications in diabetes. The other is that we use it as a surrogate for autonomic neuropathy, which may influence our cardiovascular disease end points.
We chose to use "any mention" of heart disease on the death certificate in our survival analyses because a cardiac event is often the terminal event in people who, in addition, have other life-threatening complications of diabetes. The heart disease codes we included reflect atherosclerotic heart disease. Although one may be concerned about possible heterogeneity resulting from our use of several International Classification of Diseases codes, we have found that even when using a broader classification of "all causes [of death]," the risk factors were very similar to those for our grouping of heart disease causes.39 While we cannot be certain that our grouping of causes of death does not obscure a more specific relationship of the microvascular characteristics to specific heart disease mortality codes, our sample would be too small to detect an effect in a subgroup of heart disease classification with great confidence.
We did not consider nephropathy in our early models of risk factors for mortality because that complication is highly correlated with retinopathy and may be a manifestation of the same process that causes retinopathy. When the nephropathy was considered in addition to retinopathy, nephropathy was significant and retinopathy no longer was significant. This is in keeping with findings of other studies that have found that nephropathy in diabetes increases the risk of ischemic heart disease mortality and morbidity.40- 42 It is likely that nephropathy not only is a direct result of diabetic microangiopathy but that, when present, it is associated with elevation in fibrinogen level, blood viscosity, lipoprotein level, and platelet aggregation,4,43- 45 which may increase risk of death. Incidence of all 3 end points was higher in those who took aspirin daily. This was most remarkable for stroke. While this could be due to a causal association, it may instead reflect the clinician's recommendation to subjects who they thought were sicker and were, therefore, at greater risk for stroke. We do not have enough stroke events to evaluate whether the aspirin association is related to an especially high risk of hemorrhagic stroke.
Limitations of our study include the fact that self-report may be inaccurate, leading to underreporting or overreporting. Queries were done at study examinations that were about 5 years apart, and these relatively long intervals may have led to faulty memory of study subjects. It is not clear whether these sorts of errors would be likely to lead to overestimates or underestimates, or whether they were likely to bias our estimates of the relationships of cardiovascular events to microvascular characteristics. Medical records do not always solve the problem because they are not always complete or accurate as a result of transcription errors and incomplete copying of records at the institutions. Another consideration is that this cohort was defined 20 years ago. Interim changes in patterns of care, which have occurred relatively recently, were likely to influence cardiovascular disease and mortality and possibly the relationship of microvascular characteristics to these outcomes. Cohorts defined more recently and followed up more frequently with more precise documentation of cardiovascular disease and its risk factors will afford better opportunities to evaluate the associations of microvascular characteristics to these end points.
In conclusion, we found that age-adjusted cumulative incidence of cardiovascular disease was high in this cohort of persons with type 1 diabetes mellitus, and that these end points were associated with retinal microvascular characteristics 20 years earlier. The pathways linking the microvascular to macrovascular characteristics require further research.
Correspondence: Barbara E. K. Klein, MD, Department of Ophthalmology and Visual Sciences, University of Wisconsin–Madison, 610 N Walnut St, 409 WARF, Madison, WI 53726.
Accepted for publication October 3, 2003.
This study was supported by grant HL59259 from the National Institutes of Health, Bethesda, Md.