A Prospective Study of Serum Lipids and Risk of Diabetic Macular Edema in Type 1 Diabetes

The DCCT was a multicenter, randomized, controlled clinical trial that tested whether an intensive treatment regimen aimed at achieving blood glucose levels as close to the nondiabetic range as possible would affect the onset and progression of retinal, renal, and neurological complications in type 1 diabetes when compared with conventional treatment (12). The DCCT population consisted of 1,441 subjects aged 13–39 years at study entry, who had a duration of diabetes between 1 and 15 years at randomization. The trial included two subcohorts, and within each subcohort, subjects were randomly assigned to intensive or conventional glycemic control. Inclusion criteria for the primary prevention subcohort were diabetes duration of 1–5 years, no retinopathy by seven-field stereoscopic fundus photography, and albuminuria <28 μg/min (<40 mg/24 h). These criteria were met by 726 subjects. The secondary intervention subcohort included 715 subjects with 1–15 years’ duration of diabetes, mild-to-moderate nonproliferative retinopathy, and albuminuria <139 μg/min (<200 mg/24 h). All subjects were recruited from 1983 to 1989 and were followed for an average of 6.5 years (range 3–9). Exclusion criteria included total cholesterol >3 SD above the mean for age and sex, LDL cholesterol >190 mg/dl, body weight >30% above ideal, and major electrocardiographic abnormalities, chronic heart disease, or symptoms of peripheral vascular disease. A total of 711 subjects were randomly assigned to receive intensive treatment, which consisted of insulin injections three or more times a day or external insulin pump therapy guided by self-monitoring of blood glucose at least four times per day. Seven hundred thirty participants were assigned to conventional treatment, which included one to two insulin injections a day and daily urine or blood glucose self-monitoring.

Fasting plasma samples after at least an 8-h overnight fast were collected at baseline and annually and sent for lipid analysis on dry ice to the DCCT Central Biochemistry Laboratory, Department of Laboratory Medicine and Pathology, University of Minnesota, Minneapolis. Serum total and HDL cholesterol and triglyceride concentrations were measured as previously described (4). LDL cholesterol was calculated using the Friedewald equation with VLDL cholesterol concentration assumed to be 20% of the triglyceride concentration (13). The laboratory was certified through the Centers for Disease Control and Prevention Lipid Standardization Program and participated in the quality control program of the College of American Pathologists (4).

HbA_1c levels were determined from whole blood using high-performance liquid chromatography in the DCCT Central HbA_1c Laboratory (14). Whole-blood samples were collected every 3 months and shipped on wet ice to the lab. For the present analysis, we calculated the mean of all previous quarterly HbA_1c measurements for each subject for each 6-month follow-up interval (synchronous with the retinal photography grading, see below). Smoking history was assessed at baseline and then annually using a standardized questionnaire, and we defined smoking status using categories of never, past, and current smoker. Clinical proteinuria was defined as urinary albumin excretion of >139 μg/min using a 4-h standardized urine collection taken annually, and we excluded person-time after that diagnosis was made.

To assess retinopathy, standardized seven-field stereoscopic retinal color photographs were taken by certified photographers at baseline and every 6 months during follow-up. All photographs were mailed to the DCCT Central Ophthalmologic Reading Unit located at the University of Wisconsin, where they were assessed by masked graders in a standardized procedure using the Early Treatment Diabetic Retinopathy Study (ETDRS) protocol (15). Distinct retinal lesions such as microaneurysms, hard exudates, and macular edema were quantified and recorded separately. CSME was defined as in the DCCT, using as a standard definition retinal thickening (or hard exudates adjacent to retinal thickening) located within 500 μm of the center of the macula or an area of retinal thickening at least one disc area in size, some of which was within one disc diameter of the center of the macula (16). The grading of the extent of hard exudates was based on comparison with ETDRS standard photographs 3, 4, and 5 according to the following scale: 0, no hard exudate; 1, questionable hard exudate; 2, definite hard exudate but less than that in standard photograph 3; 3, obvious hard exudate equal to or more than that in standard photograph 3 but less than that in standard photograph 5; 4, moderate hard exudate equal to or more than that in standard photograph 5 but less than that in standard photograph 4; 5, severe hard exudate equal to or more than that in standard photograph 4; and 8, unable to grade. For the present analysis, we considered a subject to have developed hard exudate if the grade was “obvious” or higher (levels 3–5) (5). If grading was not technically possible, that eye was excluded from analysis for that follow-up visit. We defined progression of DR as in previous DCCT reports, i.e., as an increase of at least three steps from the ETDRS level at baseline that was sustained for ≥6 months. The time of occurrence was defined as the first of the two consecutive visits at which the progression by three or more steps was observed.

For this analysis, we created five categories of each lipid parameter based on quintiles of the distribution in the total study population at baseline and examined their relationship with incidence of CSME, hard exudate, and progression of DR and PDR. We used proportional hazards regression to estimate the rate ratios (RRs) and 95% CIs for the relationships of each lipid variable with each DR end point with stratification on randomization assignment and baseline retinopathy subcohort. Follow-up was defined as beginning at the date of randomization and continuing until an end point was reached or the last scheduled follow-up visit was concluded, whichever came first. Follow-up in the DCCT was extremely high, with subjects completing 99% of all scheduled follow-up visits (12). As in previous analyses of DCCT data, we used individuals rather than eyes as the unit of analysis. Thus, individuals who developed CSME or hard exudate in only one eye were classified in the same manner as those who may have developed CSME or hard exudate in both eyes. We used cumulatively updated mean lipid measures as a time-varying covariate in models adjusted for the design variables. We then extended these models first to control for HbA_1c levels and then additionally for other possible confounders, including age, sex, duration of diabetes, and smoking status. Finally, because the cumulative mean of lipid levels may not represent the most appropriate way to express lipid levels as they relate to retinopathy, we also examined associations of baseline and the most recent lipid values in models for CSME and other DR outcomes.

The mean baseline total and LDL cholesterol levels were 176.4 mg/dl (range 73–312) and 109.7 mg/dl (range 17–242), respectively. These values were obviously influenced by the exclusion of potential volunteers with very elevated cholesterol levels. Mean HDL cholesterol was 50.6 mg/dl (range 21–102), and the mean triglyceride level was 81.3 mg/dl (range 20–701). Relationships of lipids with other demographic and clinical characteristics were generally similar to those previously reported. For example, older subjects tended to have higher total and LDL cholesterol levels (Table 1). Also, mean HbA_1c was highest in the top quintile of all lipid measures except for HDL cholesterol. Men tended to have lower total and HDL cholesterol but higher triglycerides and total–to–HDL cholesterol ratio. Both current smokers as well as those with prevalent retinopathy at baseline were disproportionately represented in the top quintile of triglycerides, total cholesterol, LDL cholesterol, and total–to–HDL cholesterol ratio and the bottom quintile of HDL cholesterol.

For the CSME outcome, models controlling for randomized treatment assignment and primary prevention versus secondary intervention cohorts revealed that compared with those in the lowest quintile of total cholesterol, subjects in the highest quintile had an RR 3.35 (P for trend = 0.0005). Similarly, those in the highest quintile of LDL cholesterol had a significant threefold higher risk of CSME (3.03, P for trend = 0.0003). HDL cholesterol was not significantly associated with development of CSME (Table 2). In contrast, higher total–to–HDL cholesterol ratio (4.38, P for trend = 0.002) and triglycerides (3.51, P for trend = 0.03) were associated with an increased risk of CSME in these models. In a second set of models in which we additionally controlled for HbA_1c, the associations between lipids and CSME were attenuated; however, significant trends over quintiles of lipids persisted for total cholesterol (top versus bottom quintile, 2.35, P for trend = 0.02), LDL cholesterol (2.14, P for trend = 0.008), and total–to–HDL cholesterol ratio (3.24, P for trend = 0.03). With further adjustment for other possible risk factors for retinopathy, including age, sex, smoking, duration of diabetes, and censoring person-time after diagnosis of proteinuria, LDL cholesterol (1.95, P for trend = 0.03) (Fig. 1) and total–to–HDL cholesterol ratio (3.84, P for trend = 0.03) (Fig. 2) remained significant predictors of incident CSME.

Models examining lipids and hard exudate are presented in Table 3. In the models controlling for randomized treatment assignment and primary prevention versus secondary intervention subgroup, total cholesterol (RR for top versus bottom quintile, 2.46, P for trend = 0.0008), LDL cholesterol (2.93, P for trend = 0.001), total–to–HDL cholesterol ratio (2.73, P for trend = 0.0003), and triglycerides (3.28, P for trend = 0.003) were each significantly associated with hard exudate. Additional control for HbA_1c level again attenuated the magnitude of these relationships, but all remained statistically significant. We observed similar findings after adjusting for other potential risk factors (total cholesterol: 2.37, P for trend = 0.001; LDL cholesterol: 2.77, P for trend = 0.002 [Fig. 1]; total–to–HDL cholesterol ratio: 2.44, P for trend = 0.0004 [Fig. 2]; and triglycerides: 3.20, P for trend = 0.006).

For the outcome of progression of DR, after controlling for randomized treatment assignment and primary prevention versus secondary intervention cohorts, relationships with total, LDL, and HDL cholesterol were not statistically significant (Table 4). In contrast, there were significant associations with total–to–HDL cholesterol ratio (RR for top versus bottom quintile, 2.38, P for trend = 0.004, and triglycerides, 2.64, P for trend = 0.0001). However, in models that additionally adjusted for HbA_1c, relationships of triglycerides and total–to–HDL cholesterol ratio were no longer significant.

In examining relationships of lipids with PDR, our initial models showed no association with total, LDL, or HDL cholesterol, but significant associations with total–to–HDL cholesterol ratio (RR for top versus bottom quintile, 2.78, P for trend = 0.03) and triglycerides (3.36, P for trend = 0.02) (Table 5). With adjustment for HbA_1c, these associations were no longer significant (total–to–HDL cholesterol ratio: 1.91, P for trend = 0.16, and triglycerides: 1.88, P for trend = 0.23). Adjustment for additional factors further diminished these associations.

In general, results from models of baseline and most recent lipid values were very similar to those from the cumulative average models reported above. For example, in models adjusting for treatment assignment, HbA_1c, and other risk factors, there was a significant association between baseline total cholesterol level (RR for top versus bottom quintile, 3.75, P for trend = 0.04), baseline LDL (2.37, P for trend = 0.03), and baseline total–to–HDL cholesterol ratio (2.61, P for trend = 0.03) and CSME. Similarly, the most recent levels of total and LDL cholesterol and total–to–HDL cholesterol ratio were also associated with CSME (data not shown).

Finally, we examined whether any of the lipid parameters was associated with loss of vision. Specifically, in models with an outcome of doubling of the visual angle (equivalent to a decrease of ≥15 letters read correctly on a standard visual acuity chart between baseline and follow-up examinations) and adjusting for treatment assignment, HbA_1c, and other risk factors, we found no significant relationship with the cumulative average of any lipid parameter: total cholesterol level (RR for top versus bottom quintile, 1.08, P for trend = 0.88), LDL cholesterol (1.08, P for trend = 0.32), and total–to–HDL cholesterol ratio (0.94, P for trend = 0.30). Of note, the low frequency of this outcome probably limited these analyses.

In this, the largest prospective study to date of the relationship of lipids with DR among patients with type 1 diabetes, there was a twofold increased risk of CSME in the highest versus lowest quintile of LDL cholesterol and a fourfold increased risk of CSME in the highest versus lowest quintile of total–to–HDL cholesterol ratio. Similarly, the risk of hard exudate increased more than twofold for subjects in the highest quintile of total cholesterol, LDL cholesterol, or total–to–HDL cholesterol ratio and more than threefold for subjects in the highest quintile of triglycerides. In contrast, no lipid parameters were associated with progression of DR or with PDR after adjusting for HbA_1c and other risk factors.

Several previous studies of lipids and DR did not control for HbA_1c (7, 17–19). In the DCCT (20), as well as in other epidemiological studies, HbA_1c was a strong predictor of diabetic macular edema (21). Hyperglycemia is also associated with dyslipidemia, specifically increased levels of total cholesterol and triglycerides, a slight elevation of LDL, but generally little if any change in HDL, resulting in increased total–to–HDL cholesterol ratio (22). Consequently, we think that the potential for confounding demands adjusting for HbA_1c to assure that any observed association between lipids and retinopathy is not a spurious finding (23).

Although we observed that adjustment for HbA_1c considerably attenuated many of the relationships between lipids and DR end points, several associations persisted. Among previous studies that controlled for HbA_1c, higher total cholesterol was positively associated with presence of hard exudate in a cross-sectional analysis of participants with type 1 diabetes in the Wisconsin Epidemiological Study of Diabetic Retinopathy (WESDR) (8). In contrast, in a prospective analysis of type 1 diabetic subjects in the WESDR (24), there was no association of lipids with macular edema after controlling for HbA_1c and other risk factors. In a cross-sectional analysis from the ETDRS (6), higher total and LDL cholesterol levels were each associated with severity of hard exudate. Moreover, and similar to our findings, in a prospective analysis of ETDRS data (5), the development of hard exudate was ∼50% faster among subjects with elevated baseline levels of total cholesterol and triglycerides and 36% faster among participants with higher baseline levels of LDL compared with participants with normal lipid levels at baseline. In another smaller study (25) of type 1 diabetes, those who had higher cholesterol levels had a 46% higher incidence of macular edema.

In the ETDRS, baseline total cholesterol >240 vs. <200 mg/dl increased the risk of doubling of the visual angle by 50% after 5 years of follow-up (5). Although we did not observe any relationship between lipids and vision loss, which was almost nonexistent in the DCCT, results of the present study are similar to those from the ETDRS with regard to hard exudate. This is interesting since ETDRS study participants had mostly type 2 diabetes, were older, and had a duration of diabetes >15 years. More than two-thirds of participants had macular edema at baseline. Moreover, in the ETDRS, the median cholesterol value when compared with the age-stratified distribution of the patients screened by the Lipids Research Clinics Program was at the 75th percentile, compared with approximately the 50th percentile in the DCCT (26). More specifically, ETDRS participants’ total cholesterol ranged from 106 to 852 mg/dl, with the mean of 229 mg/dl (27), compared with total cholesterol ranging from 73 to 312 mg/dl, with a mean of 176 mg/dl in the DCCT. Additionally, in the ETDRS, 36% of participants had total cholesterol levels >240 mg/dl compared with <5% in DCCT subjects (28).

Although we did not detect any significant findings for PDR, associations between lipids and PDR have been identified in some previous studies (29–31). It is possible that we failed to identify any associations with PDR because the truncated distribution of lipids in DCCT eliminated subjects with the highest levels that, in light of these other studies, may be associated with increased risk. Moreover, the relatively small number of cases of advanced retinopathy during the DCCT almost certainly reduced our capacity to examine this association. Longer-term follow-up of the DCCT cohort during the Epidemiology of Diabetes Interventions and Complications (EDIC) study should afford the opportunity (32).

The literature regarding effects of lipids on severity or progression of DR overall is less consistent (8, 24, 29, 33–35). Triglycerides were a significant risk factor for presence of DR in two previous cross-sectional studies (29, 34). Others failed to find any association with triglycerides, but reported inverse associations between HDL cholesterol and DR (8, 35, 36). For example, in a recent cross-sectional analysis (36) of EDIC data, consisting of about two-thirds of the initial DCCT cohort, an inverse association between HDL and the severity of DR was detected ∼2–4 years after the end of the DCCT trial. Similarly, in a cross-sectional analysis (8) of participants with type 1 diabetes from the WESDR study, the severity of DR was inversely related to HDL. In contrast to these cross-sectional findings, in 5 years of longitudinal data from the WESDR study, no relationship of HDL or total–to–HDL cholesterol ratio with incidence or progression of any of retinal outcome was shown (24).

Although the current prospective analysis of the DCCT data provide the most systematic and comprehensive analysis of relationships of conventional lipid parameters with DR to date among people with type 1 diabetes, the findings must be interpreted in the context of the study design. Because the study was conducted some time ago, newer more sensitive methods of assessing retinal thickening such as with optical coherence tomography were not available. Future studies using such methods would be of interest (37). Nondifferential misclassification of lipid levels is possible, though this would tend to bias the results toward unity. This problem is also minimized by multiple measurements of lipids. Our analysis of the cumulative average of the lipid measurements may not be the correct specification of the relationship between lipids and DR. However, we also looked at baseline lipid measurement as well as the most recent lipid measurements, and results were very similar to those presented. Residual confounding is always a possibility, but it is unlikely for HbA_1c (the most important confounder) in the present study because this was measured repeatedly. In addition, as we made a number of comparisons in the present study, it is possible that some findings were due to chance. The DCCT study population of relatively healthy diabetic volunteers is not representative of all people with type 1 diabetes, so it is uncertain whether these findings are generalizable. In particular, subjects were excluded from the DCCT if they had elevated cholesterol, which may have resulted in some relationships being underestimated or undetected. Finally, we were limited in our analysis to conventional lipid parameters, though recent data from the EDIC suggest that lipoprotein particle size and density may also be risk factors (36).

A relationship between lipid levels and macular edema appears to be biologically plausible. High lipid levels are known to cause endothelial dysfunction (38) via a local inflammatory response, with consequent release of cytokines and growth factors, activation of oxygen-sensitive biological changes in vessel walls, increases in LDL oxidation, and quenching of nitric oxide. In turn, endothelial dysfunction in the diabetic vasculature results in blood-retinal barrier breakdown in animal models of DR (9–11), though data are lacking in humans. Moreover, elevated levels of LDL and triglycerides in type 1 diabetes have been linked with higher levels of fluorescent advanced glycation end products, which are hypothesized to play an important role in the pathogenesis of diabetes complications (39).

Diabetic macular edema is an important cause of central vision loss in patients with diabetes (40). Previous studies in the U.S. have estimated the 10-year cumulative incidence of macular edema to be 20.1% and that of CSME to be 13.6% among those with type 1 diabetes and 25.4 and 17.6%, respectively, among people with type 2 diabetes treated with insulin (20). In the ETDRS, macular edema was the second most common cause of severe visual loss (41). The clinical significance of macular edema (4, 42) highlights the importance of investigating potentially modifiable risk factors for this condition. These prospective data indicate that elevated serum lipids, particularly total–to–HDL cholesterol ratio and triglycerides, are independent risk factors for both CSME and retinal hard exudate. If indicative of a causal relationship, these data lend additional support to current treatment guidelines recommending aggressive lowering of elevated lipids among diabetic patients. Rigorous lipid control, in addition to its known health benefits in preventing cardiovascular disease, may also lessen ocular morbidity and associated health care costs, thereby potentially improving quality of life and vision among people with type 1 diabetes.

This work was supported in part by grant 1-2000-646 from the Juvenile Diabetes Research Foundation (to D.A.S.), by the Harvard Medical School Friends of the 50th Anniversary Scholars in Medicine (to D.A.S.), by the Earle P. Charlton, Jr., Charitable Foundation (to D.M.N.), and by a contract with the Division of Diabetes, Endocrinology, and Metabolic Diseases of the National Institute of Diabetes and Digestive and Kidney Diseases (to D.M.N.).