Prospective Study of Inflammatory Biomarkers and Risk of Diabetic Retinopathy in the Diabetes Control and Complications Trial

Importance This study demonstrates that increasing quintiles of baseline high-sensitivity C-reactive protein (hsCRP) level may be associated with higher risk of incident clinically significant macular edema, the leading cause of vision loss in working-aged individuals in North America.

Objective To determine whether baseline levels of hsCRP and intercellular adhesion molecule 1 (ICAM-1) predict development and progression of diabetic retinopathy (DR), clinically significant macular edema (CSME), retinal hard exudates, and proliferative DR in the Diabetes Control and Complications Trial (DCCT) cohort.

Design The DCCT was a large multicenter randomized controlled clinical trial.

Setting Twenty-nine medical centers in the United States and Canada.

Participants The DCCT population consisted of 1441 subjects with type 1 diabetes mellitus aged 13 to 39 years at study entry.

Intervention We measured levels of hsCRP, ICAM-1, vascular cell adhesion molecule 1, and tumor necrosis factor α receptor 1 in stored baseline blood samples.

Main Outcome Measures We assessed the association of levels of hsCRP, ICAM-1, vascular cell adhesion molecule 1, and tumor necrosis factor α receptor 1 with incident DR end points ascertained from grading of standardized 7-field stereoscopic retinal color photographs taken at baseline and every 6 months during follow-up.

Results After adjustment for randomized treatment assignment and other factors, we observed a statistically significant association between hsCRP and risk of CSME, with a relative risk (RR) for the top vs bottom quintile of 1.83 (95% CI, 0.94-3.55; P for trend = .01). Similarly, for the development of retinal hard exudates, the RR for the top vs bottom quintile of hsCRP level was 1.78 (95% CI, 0.98-3.25; P for trend = .004), whereas for ICAM-1 level, the RR comparing the top vs bottom quintiles was 1.50 (95% CI, 0.84-2.68; P for trend = .05). There were no statistically significant associations between baseline VCAM-1 or tumor necrosis factor α receptor 1 levels and risk of any of the DR end points.

Conclusions and Relevance After adjusting for known risk factors, increasing quintiles of baseline hsCRP level may be associated with higher risk of incident CSME and macular hard exudate in the DCCT cohort. Circulating levels of ICAM-1 may also be associated with the development of retinal hard exudates.

Diabetic retinopathy is the leading cause of vision loss in working-aged individuals in North America, with most vision loss being attributable to diabetic macular edema.¹ Several studies have suggested that chronic low-grade inflammation may be involved in the pathogenesis of diabetic retinopathy (DR).^2,3 The benefits of intravitreal steroids and anti–vascular endothelial growth factor agents such as ranibizumab (Genentech) in the treatment of diabetic macular edema, as shown in recent randomized trials, support this theory.⁴ Moreover, some studies have found significant associations of inflammatory biomarkers with DR, including associations with high-sensitivity C-reactive protein (hsCRP),⁵ intercellular adhesion molecule 1 (ICAM-1), and vascular cell adhesion molecule 1 (VCAM-1)⁶ and tumor necrosis factor α (TNF-α).⁷ However, conflicting evidence has also been published.^8,9 To our knowledge, however, there have been no prospective studies.

We therefore set out as our primary aim to prospectively examine whether baseline levels of hsCRP and ICAM-1 predict future development and/or progression of DR, including the development of clinically significant macular edema (CSME), retinal hard exudates, and proliferative DR. Of secondary interest, we additionally examined associations with TNF-α receptor 1 (TNFR1) and VCAM-1. We measured serum levels of hsCRP, ICAM-1, VCAM-1, and TNFR1 from stored baseline blood specimens among the 1441 patients from the Diabetes Control and Complications Trial (DCCT)¹⁰ and studied their association with development of retinopathy during an average of 6 years of follow-up.

The DCCT was a large multicenter randomized controlled clinical trial that compared an intensive treatment regimen directed at achieving blood glucose levels as close to normal as possible with conventional treatment as practiced at that time (1980s-1990s). The DCCT population consisted of 1441 subjects aged 13 to 39 years at study entry.¹⁰

The trial included 2 subcohorts. Participants in the primary prevention subcohort had a diabetes duration of 1 to 5 years, no retinopathy by 7-field stereoscopic fundus photography, and no evidence of microalbuminuria at baseline (726 subjects). The secondary intervention subcohort included 715 subjects with 1 to 15 years of diabetes, mild to moderate nonproliferative DR, and an albumin level less than 140 μg/min.

After a mean follow-up of 6.5 years, the DCCT reported a statistically significant reduction in microvascular end points in the intensive compared with conventional therapy group.¹⁰ Follow-up was excellent in the DCCT with subjects attending 99% of scheduled follow-up visits. Subjects were followed up for an average of 6.5 years (range, 3-9 years).

To assess various DR end points, standardized 7-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.¹¹

This study was approved by the Partners' Human Research Committee Institutional Review Board at the Brigham and Women's Hospital.

Fasting serum samples were obtained from DCCT participants at baseline and each annual visit. Blood was drawn into a red-topped tube, allowed to clot for at least 20 minutes, and then spun in a centrifuge at room temperature for 10 minutes at 3000 rpm. Serum was then divided into 1.8-mL cryotubes and promptly frozen. Samples were maintained at −70°C at the DCCT Central Biochemistry Laboratory, Department of Laboratory Medicine and Pathology, University of Minnesota, until preparation for the present study.

For this study, we measured baseline levels of hsCRP, ICAM-1, VCAM-1, and TNFR1 in baseline blood samples from the 1441 participants in the DCCT. Serum levels of hsCRP were determined by a latex-enhanced immunonephelometric assay on a BN II analyzer (Dade Behring). Serum levels of ICAM-1, VCAM-1, and TNFR1 were determined by enzyme-linked immunosorbent assays (R&D Systems). The day-to-day variabilities of each biomarker assay were less than 10%. Hemoglobin A_1c (HbA_1c) levels were determined from whole blood at the time of collection using high-performance liquid chromatography in the DCCT Central HbA_1c Laboratory.^12-14

We constructed a Kaplan-Meier survival curve for quintiles of the 4 biomarkers adjusted for randomized treatment group and then used Cox proportional hazards models to estimate the relative risks (RRs) and 95% confidence intervals over quintiles of hsCRP level and the other markers for the following DR outcomes: (1) a 3-step or more progression of DR along the ETDRS scale sustained for at least 6 months, (2) incident CSME, (3) development of obvious (or higher-grade) retinal hard exudates,¹⁵ and (4) incident proliferative DR. Patients were used as the unit of analysis rather than eyes. Therefore, patients were considered to have developed 1 of the DR outcomes if they developed the outcome in either eye. Patients with any of the outcomes at baseline were excluded from the analysis for that specific outcome.

We categorized hsCRP level and the other biomarkers into quintiles to assess the effect of high levels on the time to development of the various DR end points while reducing the influence of extreme levels. We defined the period of follow-up as beginning at the date of randomization and continuing until the end point was reached or the last scheduled follow-up visit was concluded, whichever came first. The Cox models were adjusted for baseline HbA_1c level, randomized treatment group, age, sex, duration of diabetes, body mass index, smoking status, and total cholesterol to high-density lipoprotein cholesterol ratio. We categorized smoking status into never smoked, past smoker, or current smoker. We assessed our models for collinearity and effect modification and tested proportional hazard assumptions and made the appropriate adjustments to the model. We examined collinearity of the variables by comparing the bivariate models to ensure that the standard errors in the model did not increase by 15% or more. Although we observed collinearity between HbA_1c level and duration of diabetes, we elected to keep both variables in the model to control for the possibility of a large amount of confounding because of the strong associations of these variables with DR. We tested for effect modification by comparing the −2 log likelihood for the Cox models with and without interaction terms for each biomarker and age, sex, duration of diabetes, baseline HgbA_1c level, baseline low-density lipoprotein cholesterol level, smoking status, stratification, and treatment group. All statistical analyses were performed with SAS version 9.2 (SAS Institute).

The characteristics of the DCCT study population have been previously described in detail.⁴ Mean (SD) baseline hsCRP level was 2.09 (3.93) mg/L in the entire cohort, with a mean (SD) of 0.178 (0.796) mg/L in the bottom quintile vs a mean (SD) of 7.168 (6.577) mg/L in the top quintile. The baseline characteristics of all study patients by quintile of hsCRP level are shown in Table 1. Age and race were similar across quintiles of hsCRP level. The proportion of participants in the intensive insulin group was similar across quintiles of hsCRP level, with 48.5% of patients in the bottom vs 53.1% in the top quintile. The proportion of patients in the primary prevention group was also similar across quintiles, with 52.3% in the bottom vs 44.3% in the top quintile. There were fewer males with higher levels of hsCRP, with 77.3% in the bottom quintile in comparison with 32.8% in the top quintile. Duration of type 1 diabetes, baseline HbA_1c level, and baseline body mass index slightly increased with increasing quintile of hsCRP level, whereas proportion of participants who never smoked decreased from the bottom to top quintiles.

There were no significant associations of hsCRP level with 3-step or more progression of DR. In the multivariable model, the RR for the top vs bottom quintile was 1.18 (95% CI, 0.76-1.81) with a P value for trend of .25. For ICAM-1 level, the RR for the top vs bottom quintile was 1.06 (95% CI, 0.69-1.62; P for trend = .51). There were no statistically significant associations between increasing quintiles of baseline VCAM-1 and TNFR1 levels and risk of 3-step or more progression of DR (Table 2).

We observed significant associations between increasing quintiles of baseline ICAM-1 level and incidence of CSME in models adjusting for DCCT randomized treatment assignment. The RR for the top vs bottom quintile of ICAM-1 level was 1.67 (95% CI, 0.95-2.97) with a P value for trend of .03. This trend did not remain statistically significant when adjusting for additional covariates (including age, sex, duration of diabetes, body mass index, smoking status [never, past, or current], and total cholesterol to high-density lipoprotein cholesterol ratio), with a top vs bottom quintile RR of 1.41 (95% CI, 0.76-2.61) and a P value for trend of .19. We did observe a statistically significant association between hsCRP level and incidence of CSME after adjustment for these additional factors, with an RR for the top vs bottom quintile of 1.83 (95% CI, 0.94-3.55; P for trend = .01). There were no statistically significant associations between increasing quintiles of baseline VCAM-1 and TNFR1 levels and risk of CSME (Table 3).

In models adjusted for randomized treatment assignment, there were significant trends of increasing risk of retinal hard exudate across quintiles of baseline hsCRP and ICAM-1 levels. For hsCRP, the RR for the top vs bottom quintile was 1.65 (95% CI, 0.96-2.84) with a P value for trend of .02. For ICAM-1, the top vs bottom quintile RR was 1.51 (95% CI, 0.88-2.59; P value for trend = .03). These associations remained statistically significant after adjusting for multiple covariates. In the fully adjusted models, RR for the top vs bottom quintile of hsCRP level was 1.78 (95% CI, 0.98-3.25; P value for trend = .004), whereas for ICAM-1 level, the RR comparing the top vs bottom quintiles was 1.50 (95% CI, 0.84-2.68; P value for trend .05). There were no statistically significant associations between increasing quintiles of baseline VCAM-1 and TNFR1 levels and risk of retinal hard exudates (Table 4).

There was a nonstatistically significant trend for increasing risk of proliferative DR with increasing baseline hsCRP levels when adjusting for DCCT randomized treatment assignment, with an RR for the top vs bottom quintile of 1.80 (95% CI, 0.81-4.02) and a P value for trend of .12 across quintiles. In the full model with multiple covariates, the RR comparing the top vs bottom quintiles was 1.49 (95% CI, 0.61-3.66) with a P value for trend of .38. In the model adjusted for randomized treatment group, the RR for the top vs bottom quintile of ICAM-1 level was 2.24 (95% CI, 0.99-5.06) with a P value for trend across quintiles of .02. However, this association was no longer statistically significant in the multivariable model (P value for trend across quintiles = .41). There were no statistically significant associations between increasing quintiles of baseline VCAM-1 and TNFR1 levels and risk of proliferative DR (Table 5).

In exploratory analyses to look at potential associations between extreme levels of hsCRP and DR, we recategorized hsCRP to levels at or above the 95th percentile vs below the 95th percentile and observed a statistically significant increased risk of retinal hard exudates and proliferative DR with nonstatistically significant results for CSME and 3-step or more progression. For retinal hard exudates, the RR was 2.79 (95% CI, 1.64-4.74) when adjusting for randomized treatment group and 2.38 (95% CI, 1.35-4.19) in the full model with multiple covariates. For proliferative DR, the RR was 3.67 (95% CI, 1.90-7.06) when adjusting only for randomized treatment group and 2.91 (95% CI, 1.39-6.08) in the full model (Figure).

Finally, in analyses to test for possible interactions, there were no statistically significant differences between the −2 log likelihood with and without the interaction terms in any models (data not shown).

In the present study, we examined data from the DCCT to address the hypothesis of whether chronic subclinical inflammation, as measured by elevated levels of hsCRP and ICAM-1, is associated with the development and progression of DR. We also conducted secondary analyses to look at the association between VCAM-1 and TNFR1 levels and both CSME and development of retinal hard exudates. None of the markers were found to consistently predict all of the DR outcomes. Instead, the findings of this study suggested that inflammation as measured by hsCRP and ICAM-1 levels may be more relevant to the development of CSME and retinal hard exudates than to progression of retinopathy per se, as measured by changes along the ETDRS retinopathy grading scale. There were no statistically significant associations between increasing quintiles of baseline VCAM-1 and TNFR1 levels and any of the DR end points.

One of the limitations of this study was that only baseline biomarker levels were measured for all DCCT participants and findings are consequently restricted to a single measurement of the biomarkers. Although the markers we measured have been shown to remain stable in stored specimens over long periods, it is always possible that some degradation may have occurred in 1 or more of the markers, which would tend to result in findings that are biased toward no association. An analysis of change in biomarker levels over time might also be clinically relevant, because levels can vary and risk of retinopathy may depend on the cumulative impact of such variation or on average levels over time. Another issue that may impact interpretation is that following the DCCT, care of patients with type 1 diabetes has changed considerably with increased attention to tight regulation of glycemia. We showed previously that tight glycemic control tended to increase levels of hsCRP among individuals who gained weight during intensive control therapy.¹⁴ It is unknown whether associations of hsCRP or other inflammatory markers with DR might be different in the current clinical climate of intensive glycemic control. Given the uniqueness of the DCCT population, generalizability of these findings to individuals with type 2 diabetes mellitus and to ethnic minority groups is uncertain.

Systemic inflammation increases with the onset of clinical diabetes and is thought to contribute to the development of complications including nephropathy and retinopathy.¹⁶ Diabetic macular edema is believed to occur because of a breakdown of the blood-retinal barrier that allows fluid to accumulate within the retina. Factors contributing to the blood-retinal barrier breakdown include inflammatory processes. The observation that hsCRP predicts the development of retinal hard exudates and CSME suggests that systemic inflammatory activity may contribute directly to these local retinal changes (eg, through changes in the retinal vasculature) or at least that the local inflammatory activity in the retina appears to mirror the overall level of systemic inflammatory activity.

The association between quintiles of hsCRP level and development of CSME is interesting and carries potential clinical relevance. Clinically significant macular edema is the most common cause of vision loss in patients with diabetes and the risk of incident CSME was increased by 83% among those with hsCRP levels in the highest vs lowest quintile. Although further prospective studies are required to corroborate our results, these findings suggest that hsCRP could be a useful adjunct to other clinical information such as HbA_1c levels and serum lipid levels to predict the likelihood of development of CSME and perhaps identify a subgroup of patients for whom more frequent follow-up and/or more intensive management is needed.

We previously identified a strong association between serum lipid levels and the development of CSME.¹⁵ Lipid levels were predictive of CSME and retinal hard exudate formation but were not associated with 3-step or more DR progression by ETDRS grading, or development of proliferative DR, similar to the present findings for hsCRP. These findings taken together suggest the possibility of a particular pathogenic mechanism or pathway for the development of CSME and retinal hard exudates associated with serum lipid levels and inflammatory activity that may be distinct from other mechanisms involved in the development of other DR lesions and progression of DR more generally.

In models adjusted for randomized treatment assignment in the present study, there were significant trends of increasing risk of retinal hard exudates across quintiles of baseline ICAM-1 level. Spijkerman et al³ suggested that the loss of retinal capillary pericytes that has been observed histologically in DR in humans may be secondary to damage to retinal vascular endothelial cells, perhaps involving an inflammatory response. Pericyte loss may be indirectly related to leukocyte adhesion to the vasculature and accumulation of advanced glycation end products seen in early diabetes.¹⁷ This leukocyte adhesion could be mediated by ICAM-1.

There has been significant debate in the medical literature about the association of hsCRP and DR, including CSME; however, this debate has occurred in the absence of large prospective studies. Streja et al¹⁸ investigated hsCRP and fibrinogen levels in 202 patients in a cross-sectional study and found that hsCRP was not associated with DR. Kang et al⁸ found higher levels of hsCRP in 269 patients with type 2 diabetes compared with nondiabetic individuals; however, they found no significant difference in hsCRP levels in those patients with and without retinopathy in their cross-sectional study. On the other hand, van Hecke et al⁹ found in a cross-sectional study that the prevalence of retinopathy was positively associated with tertiles of hsCRP and soluble ICAM-1 levels in a study of prevalent retinopathy in individuals with and without type 2 diabetes. Loukovaara et al¹⁹ in a prospective study found that hsCRP levels were higher in women with type 1 diabetes during pregnancy and post partum with progression of retinopathy and in those with worse glycemic control.

In conclusion, we found that after adjusting for known risk factors, increasing quintiles of baseline hsCRP level may be associated with higher risks of incident CSME and the development of macular hard exudate. Circulating levels of ICAM-1 may also be associated with the development of retinal hard exudates. With further research, these findings may lead to a better understanding of the mechanisms underlying the development of CSME and retinal hard exudates and may lead to more effective strategies for retinopathy prevention and management.

Correspondence: Rajeev H. Muni, MD, MSc, FRCSC, Department of Ophthalmology and Vision Sciences, University of Toronto, St Michael's Hospital, 61 Queen St E, Ste 801, Toronto, ON M5C2T2, Canada (rajeev.muni@utoronto.ca).

Submitted for Publication: May 23, 2012; final revision received October 28, 2012; accepted November 12, 2012.

Published Online: February 7, 2013. doi:10.1001/jamaophthalmol.2013.2299

Author Contributions: Drs Muni and Schaumberg had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Funding/Support: This work was supported by Juvenile Diabetes Research Foundation grant 1-2000-646 (Dr Schaumberg). The DCCT and its follow-up Epidemiology of Diabetes Interventions and Complications (EDIC) study were conducted by the DCCT/EDIC Research Group and supported by National Institute of Health grants and contracts from the National Institute of Diabetes and Digestive and Kidney Diseases and by the General Clinical Research Center Program, National Center for Research Resources. Dr Muni was supported by an E. A. Baker grant from the Canadian National Institute of the Blind.