Risk Factors for Retinopathy in Type 1 Diabetes: The DCCT/EDIC Study

Retinopathy is a major complication of type 1 diabetes. The Diabetes Control and Complications Trial (DCCT) demonstrated that a mean of 6.5 years of intensive diabetes therapy with a mean HbA_1c of ∼7% substantially reduced microvascular complications, including retinopathy, compared with conventional therapy with a mean HbA_1c of ∼9% (1). Thereafter, observational follow-up of DCCT participants continued in the Epidemiology of Diabetes Interventions and Complications (EDIC) study (1994 to present) (2). Over the first 4 years of EDIC, the former DCCT intensive therapy group experienced a lower incidence of further progression of retinopathy than did the former conventional group, despite similar HbA_1c levels in both groups (3). This “metabolic memory” persisted after 10 years of EDIC, albeit to a lesser degree (4). By year 18 of EDIC, the effect of “metabolic memory” had largely faded, with no further divergence of retinopathy rates, but the former intensive group continued to have fewer ocular complications, including a substantially lower risk of advanced retinopathy outcomes (5).

HbA_1c was the primary variable for risk of progression of retinopathy studied in the DCCT. In the current analyses, we examined multiple risk factors (modifiable and nonmodifiable) to determine if additional modifiable factors could broaden preventive interventions. The goal of these analyses was to identify risk factors for retinopathy outcomes in the context of a multicenter study with a well-characterized cohort of individuals with type 1 diabetes (T1D) after 30 years of follow-up. Risks for proliferative diabetic retinopathy (PDR), clinically significant macular edema (CSME), and ocular surgery were examined.

The DCCT and EDIC protocols were approved by the institutional review boards of all participating centers, and all participants provided written informed consent. The methods of the DCCT and EDIC study have been described in detail (2,6). In brief, the DCCT (1983–1993) was a controlled clinical trial that evaluated whether intensive versus conventional diabetes therapy reduced the risk of complications of diabetes. A total of 1,441 patients with T1D were randomly assigned to receive either intensive (n = 711) or conventional diabetes therapy (n = 730). Intensive therapy was aimed at achieving glycemic control as close to the nondiabetic range as safely possible, whereas conventional therapy was aimed at preventing symptoms of hypo- or hyperglycemia but with no specific glucose targets.

At DCCT baseline, the study cohort consisted of a primary prevention cohort (n = 726) with 1–5 years diabetes duration, no retinopathy based on fundus photography, and <40 mg of albuminuria per 24 h; and a secondary intervention cohort (n = 715) with 1–15 years duration, minimal to moderate nonproliferative retinopathy, and <200 mg of albuminuria per 24 h (6). Exclusion criteria included neuropathy sufficiently severe to require therapy, hypertension (≥140/90 mmHg or medication), and hyperlipidemia (LDL cholesterol [LDLc] >130 mg/dL or medication).

At the end of the DCCT, after an average follow-up of 6.5 years, all participants were taught intensive therapy and were referred to their health care providers for subsequent diabetes care. In 1994, EDIC enrolled 98% of the surviving DCCT cohort, and 94% of the surviving cohort still actively participates after more than 20 years of additional follow-up.

The potential risk factors were assessed by standardized methods at periodic visits during DCCT and EDIC (6,7). HbA_1c was measured quarterly during DCCT and annually during EDIC (8). Fasting lipid levels (triglycerides and total and HDL cholesterol [HDLc]) were measured annually during DCCT and every other year during EDIC. LDLc was calculated using the Friedwald equation. Albuminuria was measured annually during DCCT and every other year during EDIC, alternating with the fasting lipids, and serum creatinine was measured annually. Albumin-to-creatinine ratio values obtained after 1 August 2012 were converted to albumin excretion rate (AER) levels as previously described (9). Estimated glomerular filtration rate (eGFR) was calculated with the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation using serum creatinine levels, age, sex, and race (10). All laboratory measurements were performed in the DCCT/EDIC central biochemistry laboratory, with standardized methods and controls in place to guard against long-term drift.

As previously described (11), candidate risk factors were grouped into the following 11 blocks (described in detail in Supplementary Table 1): design (treatment group and cohort), physical (sex, age, weight, and BMI), behavioral (smoking, drinking, and exercise), family history (family history of hypertension, myocardial infarction, and type 1 or type 2 diabetes), traditional blood pressure/pulse (systolic blood pressure [SBP] and diastolic blood pressure [DBP] and pulse rate), medication use (ACE inhibitors, angiotensin receptor blockers, β-adrenergic blockers, lipid-lowering agents, and calcium channel blockers), traditional lipid levels (total cholesterol, triglycerides, LDLc, and HDLc), diabetes specific (duration of diabetes at enrollment, stimulated C-peptide, insulin dose, and estimated glucose disposal rate), renal complications (eGFR <60 mL/min and AER ≥30 mg/24 h), hypoglycemia events (coma/seizure and/or required assistance), and glycemia (HbA_1c at eligibility and HbA_1c during follow-up).

A risk factor could be included in the model as a fixed or baseline covariate (B), or as a time-dependent covariate, using either the current (most recent) measurement (C) or as the updated mean of all follow-up values since randomization (M) (Supplementary Table 1). The updated mean accounts for the different measurement frequencies during DCCT and EDIC by weighting each value by the time interval since the last measurement. For example, the mean updated HbA_1c value of a participant at year 10 is the weighted average of the HbA_1c values up to and including year 10 for that particular individual.

Standardized stereoscopic seven-field fundus photographs were obtained every 6 months during the DCCT, and every 4th year (staggered from the start of the EDIC follow-up period) during EDIC. In addition, photographs were obtained in the full cohort at EDIC study years 4 and 10. The photographs were graded centrally using the final Early Treatment of Diabetic Retinopathy Study (ETDRS) severity grading scale (12), and graders were masked to treatment assignment and other risk factors.

PDR was defined by neovascularization observed on fundus photograph grading or evidence of scatter photocoagulation. CSME was defined using fundus photography grading or the presence of focal photocoagulation scars. Ocular surgical interventions were self-reported at annual visits and were captured based on structured interviews conducted by study staff. Ocular surgery is a composite outcome that included cataract extraction, vitrectomy and/or retinal detachment surgery, glaucoma-related surgery (including laser treatment, filtering surgery, cyclocryotherapy, or other operative procedures to lower intraocular pressure), cornea or lens-related surgery (including corneal transplant or yttrium aluminum garnet [YAG] posterior capsulotomy), or enucleation (13).

Risk factor modeling in a Cox proportional hazards (PH) model using 100 conventional group outcome case subjects and ∼150 case subjects in the full cohort (i.e., intensive and conventional combined), allowing for an R² = 0.35 for the association of up to 10 adjusting covariates with a given risk factor of interest, and using a test at the 0.01 level (two sided), provided 97% power to detect a 30% risk reduction per SD change in a factor (14).

For each outcome (PDR, CSME, and ocular surgery), the analysis was based on the time to that outcome. The survival (event-free) probability was estimated using the Kaplan-Meier method. Semiparametric Cox PH models assessed the association between fixed and time-dependent covariates and the risk of an outcome (15). For quantitative risk factors, hazard ratios (HRs) are reported per 1 unit change in the risk factor. If HR denotes the HR per 1 unit change in a quantitative risk factor, then the HR per x units change in that risk factor is calculated as HR^x (HR to the power x). The functional form for the effect of the weighted mean HbA_1c on the empirical log hazards of each outcome was investigated using smoothing splines (16).

Continuous variables were described using medians and first and third quartiles, and discrete variables were described using counts and percentages. The risk factor variable selection approach has been previously described (11). In brief, both forward- and backward-selection approaches were used. The forward-selection procedure added variables into the Cox PH model one block at a time, and at each step, factors were eliminated to yield the best subset model based on statistical significance (i.e., P values), the minimum (best) Akaike information criterion (17), and a penalized likelihood (lasso method) (18). The next block of variables was then entered, and the process continued until a final model was reached. The backward-selection approach used the lasso and, separately, selected the model with the best (smallest) Akaike information criterion, both starting with all the variables included in the model. Interaction terms with sex were used to investigate sex differences in the effect of covariates on risk of retinopathy.

Because any z statistic with absolute value of 3.89 or larger has P < 0.0001, and z values as high as ∼16 are observed, the z value better represents the significance of the covariate effect in the model than does the designation “P < 0.0001.” Moreover, the z statistic is independent of the measurement unit used and therefore allows comparisons across risk factors.

A total of 379 participants reached PDR over an average follow-up of 22 years (rate of 12 events per 1,000 person-years), 431 participants reached CSME over an average follow-up of 21.2 years (rate of 14.5 events per 1,000 person-years), and 280 participants had ocular surgeries over an average follow-up of 25.7 years (rate of 7.6 cases per 1,000 person-years). Moreover, 261 participants reached both PDR and CSME during the follow-up, 156 participants reached CSME and had ocular surgery, and 178 participants reached PDR and had ocular surgery, whereas 127 participants reached both PDR and CSME and had ocular surgery over the follow-up period (Fig. 1). Approximately 65% of the cohort remained free of PDR, ∼60% remained free of CSME, and ∼70% remained free of ocular surgery after 30 years of follow-up (Supplementary Fig. 1).

At DCCT baseline, 53% of the participants were male, 51% were assigned to the conventional therapy, and 50% were in the primary prevention cohort; the participants had a median (quartiles) age of 27 years (22, 32), duration of diabetes of 4.3 years (2.3, 9.1), HbA_1c of 8.7% (7.8, 9.9), and 18% were smokers. Compared with subjects without PDR, those with PDR were more likely to be in the conventional group and in the original secondary intervention cohort (Table 1). They also had higher baseline BMI, DBP, pulse rate, total cholesterol, triglycerides, LDLc, AER, and HbA_1c levels and longer duration of T1D. Higher risk of CSME was associated with conventional therapy and secondary intervention cohort; male sex; older age; family history of T1D; higher BMI, DBP, pulse rate, total cholesterol, triglycerides, LDLc, AER, and HbA_1c levels; and longer duration of T1D (Supplementary Table 2). Higher risk of ocular surgery was associated with conventional treatment group and secondary intervention cohort; older age; currently smoking; higher weight, BMI, pulse rate, triglycerides, and HbA_1c levels; and longer duration of T1D (Supplementary Table 3).

Similar patterns were observed when these risk factors were considered over the entire follow-up period as time-dependent variables (Table 2 and Supplementary Tables 4–6). When adjusted for age and mean updated HbA_1c, higher risk of PDR was associated with original DCCT cohort (secondary intervention vs. primary prevention), adult versus adolescent status at baseline, lower weight gain in females, higher SBP and DBP, higher pulse rate, presence of hypertension, higher lipids, higher AER, any macroalbuminuria, and higher HbA_1c at baseline. Similarly, higher risk of CSME was observed in participants in the secondary intervention cohort and in males and was associated with higher weight, mean BMI, blood pressure, pulse rate, total cholesterol, LDLc, and triglyceride levels; alcohol use; lack of lipid-lowering medication use; either higher or lower HDLc; longer duration of T1D; any eGFR <60 mL/min; and the presence of macroalbuminuria. Higher risk of ocular surgery was observed in participants in the secondary intervention cohort, in males and in participants enrolled as adults, and was associated with higher pulse rate, mean SBP, pulse pressure, mean triglycerides, eGFR, AER, HbA_1c at baseline, and current HbA_1c value; history of hypertension; longer duration of T1D; use of ACE inhibitors; and use of any β and calcium channel blockers. The association between use of ACE inhibitors and use of any β blockers and calcium channel blockers and higher risk of ocular surgery likely represents an indication bias due to hypertension.

Table 3 reports the final multivariable Cox models for PDR, CSME, and ocular surgery, with variables presented in order of significance (i.e., larger absolute z values first). Using HbA_1c as an example of a time-dependent covariate, we used “baseline HbA_1c” to denote the HbA_1c at baseline and “mean HbA_1c” to denote the mean updated HbA_1c value, whereas the current (most recent) HbA_1c value was referred to as “HbA_1c.” In the model for PDR (Table 3A), a higher mean HbA_1c (HR = 2.1539 per 1% higher [95% CI 1.9597, 2.3673], z = 15.9155, P < 0.0001), longer duration of T1D (HR = 1.1135 per year [95% CI 1.0757, 1.1525], z = 6.1194, P < 0.0001), AER (HR = 1.0310 per 20% increase [95% CI 1.0177, 1.0445], z = 4.6223, P < 0.0001), and higher mean DBP (HR = 1.0448 per 1 mmHg [95% CI 1.0255, 1.0645], z = 4.6141, P < 0.0001) were the most significant factors. Other significant factors associated with higher risk of PDR were higher pulse rate (z = 2.9705, P = 0.0029), being in the secondary intervention versus primary prevention cohort (z = 2.3797, P = 0.0173), higher HbA_1c at baseline (z = 2.2629, P = 0.0236), and older age (z = 1.9611, P = 0.0498), whereas use of lipid-lowering medication was protective (HR 0.6786, z = −2.3842, P = 0.0171).

In the model for CSME (Table 3B), mean HbA_1c (HR = 1.8257 per 1% increase [95% CI 1.6840, 1.9795], z = 14.5977, P < 0.0001) was again the strongest risk factor, followed by duration of T1D (HR = 1.0912 per 1 year [95% CI 1.0654, 1.1161], z = 7.4741, P < 0.0001), age (HR = 1.0562 per 1 year [95% CI 1.0399, 1.0728], z = 6.9031, P < 0.0001), and DBP (HR = 1.0260 per 1 mmHg [95% CI 1.0150, 1.0371], z = 4.6653, P < 0.0001). Other significant risk factors associated with higher risk of CSME were higher AER (z = 3.6024, P = 0.0003), higher eGFR (z = 3.5133, P = 0.0004), higher mean triglycerides (z = 3.0883, P = 0.0020), and higher LDLc (z = 1.9995, P = 0.0455), whereas use of lipid-lowering medication was again protective (z = −2.5272, P = 0.0114).

In the model for ocular surgery (Table 3C), mean HbA_1c (HR = 1.8065 per 1% increase [95% CI 1.6011, 2.0383], z = 9.6047, P < 0.0001) was the strongest risk factor, followed by age (HR = 1.0660 per 1 year [95% CI 1.0466, 1.0858], z = 6.8259, P < 0.0001) and duration of T1D (HR = 1.0886 per 1 year [95% CI 1.0591, 1.1188], z = 5.9957, P < 0.0001). Other significant risk factors associated with higher risk of ocular surgery were higher AER (z = 3.6706, P = 0.0002), higher mean pulse (z = 2.6776, P = 0.0074), and higher mean SBP (z = 2.0856, P = 0.0370), whereas males had lower risk than females (z = −2.6079, P = 0.0091).

Interaction terms with sex in the final multivariable models for PDR and ocular surgery were not significant. For CSME, there were significant interactions between sex and age, and sex and DBP. The association between age and CSME was higher in females (HR = 1.07 per 1 year, z = 5.85) than in males (HR = 1.04 per 1 year, z = 3.64), whereas the association between DBP and CSME was only significant in males (HR = 1.04 per 1 mmHg, z = 5.36) but not in females (HR = 0.003 per 1 mmHg, z = 0.33).

The total exposure to glycemia as captured by the mean updated HbA_1c was by far the strongest risk factor for the three outcomes of PDR (z = 15.9155), CSME (z = 14.5977), and ocular surgery (z = 9.6047). The next most significant factors associated with PDR were duration of T1D (z = 6.1194), AER (z = 4.6223), and mean DBP (z = 4.6141); with CSME were duration of T1D (z = 7.4741), age (z = 6.9031), and DBP (z = 4.6653); and with ocular surgery were age (z = 6.8264) and duration of T1D (z = 5.9957). As explained in Research Design and Methods, we emphasized z values over P values since they better describe strong associations, such as those observed in our analyses, and, similarly to the P values, they are independent of the measurement units used.

Note that other than T1D duration, which was the second strongest risk factor for PDR and CSME and third strongest for ocular surgery, other nonmodifiable risk factors included age for all three outcomes, treatment cohort for PDR, and sex for ocular surgery. Most risk factors for retinopathy disease in T1D identified in our work are modifiable. Intensive glycemia control is of overwhelming importance in decreasing retinopathy progression. However, the additional modifiable risk factors examined suggest that early, vigorous, non–glycemia-related interventions, in addition to controlling glycemia, might further mitigate retinopathy progression and vision loss.

Although we reported these as risk factors, the inverse of the modifiable factors identified can be thought of as preventative strategies. Certainly improved HbA_1c control reduces the risk of retinopathy. Our analyses support that blood pressure control and lipid-lowering medication (for PDR and CSME) could also reduce risk of retinopathy progression. Whether optimal control of all modifiable risk factors examined would result in a substantial further decrease of retinopathy progression independent of glycemic control alone would require prospective clinical trials.

The goal of this analysis was to identify risk factors for the three clinically relevant retinopathy outcomes considered in a well-characterized cohort of individuals with T1D. A limitation of our study is the relatively small number of DCCT/EDIC participants with an outcome event (379 PDR, 431 CSME, and 280 ocular surgery); therefore, we did not attempt to develop prediction models. Such prediction models would require additional external cohort(s) with T1D for validation and calibration. Similarly, the prevalence of blindness (visual acuity <20/100 in either eye) remains extremely low in both original treatment groups (<2%), precluding an analysis (5).

In the forward variable selection approach for PDR, the model included the initial DCCT randomization group, mean SBP, LDLc, mean triglycerides, and eGFR, all significant at a P value threshold of 0.05 before adding the glycemia block. However, none of those five variables remained significant after adjustment for glycemia, which is likely explained by mediation (e.g., the DCCT group effect on PDR is mediated by its effect on glycemic levels) or causal (e.g., glycemia lowers eGFR) mechanisms. For CSME, the addition of the glycemia block in the forward selection for CSME only resulted in elimination of the initial randomization group and eGFR, and in a reduction of the association between AER and the risk of CSME (from z = 7.11 without mean HbA_1c to z = 3.60 with HbA_1c). For ocular surgery, the model before adding the glycemia block included the initial randomization group, smoking, DBP, mean insulin, and use of β blockers, all significant at level 0.05, but none of which remained significant after further adjustment for glycemia (data not shown).

Cox PH models assessed the association between potential risk factors and the risk of retinopathy outcomes. Since these risk factors (such as HbA_1c) were measured longitudinally during the follow-up, they were included as time-dependent covariates in the Cox models. There are no established R² or area under the curve measures for Cox models with time-dependent covariates. Instead, the Akaike information criterion and the model χ² values were used for comparing models. Moreover, since the R² measures in other models are directly proportional to the test statistic value (the z value), the strength of association between risk factors and outcomes was described using the corresponding z values.

Most of the ocular surgeries were cataract extractions (89 in the intensive group and 125 in the conventional group), followed by vitrectomy or retinal detachment (41 and 66, in intensive and conventional, respectively), glaucoma-related surgeries (14 and 19), corneal-related surgeries (5 and 7), YAG capsulotomy (2 and 4), and enucleation (2 and 2) in the intensive and conventional groups, respectively. The small number of individual types of surgeries other than cataract extraction precluded us from investigating them individually, and instead we used a composite outcome defined as any ocular surgery.

In conclusion, long-term exposure to hyperglycemia (as captured by high levels of mean updated HbA_1c) was the strongest risk factor for the progression of retinopathy. We found that most risk factors identified were modifiable, with the exception of duration of T1D, age, study cohort (for PDR), and sex. These findings suggest that aggressive glycemic management is key but should also be coupled with aggressive management of several other non–glycemia-related risk factors, such as blood pressure control and control of lipids to reduce the burden of retinopathy in individuals with T1D. The general principles derived from the DCCT/EDIC study most likely apply to current patients with T1D, but the higher rate of overweight and obesity in current patients (19) may lead to additional risk factors.

Funding. The DCCT/EDIC study has been supported by cooperative agreement grants (1982–1993 and 2012–2017) and contracts (1982–2012) with the Division of Diabetes Endocrinology and Metabolic Diseases of the National Institute of Diabetes and Digestive and Kidney Diseases (current grant numbers U01-DK-094176 and U01-DK-094157) and through support by the National Eye Institute, the National Institute of Neurological Disorders and Stroke, the General Clinical Research Centers Program (1993–2007), and the Clinical Translational Science Center Program (2006 to present), Bethesda, MD. Industry contributors have had no role in the DCCT/EDIC study but have provided free or discounted supplies or equipment to support participants’ adherence to the study: Abbott Diabetes Care (Alameda, CA), Animas (West Chester, PA), Bayer Diabetes Care (North America Headquarters, Tarrytown, NY), Becton Dickinson (Franklin Lakes, NJ), Eli Lilly and Company (Indianapolis, IN), Extend Nutrition (St. Louis, MO), Insulet Corporation (Bedford, MA), LifeScan (Milpitas, CA), Medtronic Diabetes (Minneapolis, MN), Nipro Home Diagnostics (Fort Lauderdale, FL), Nova Diabetes Care (Billerica, MA), Omron (Shelton, CT), Perrigo Diabetes Care (Allegan, MI), Roche Diabetes Care (Indianapolis, IN), and Sanofi (Bridgewater, NJ).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. D.P.H. wrote the initial draft of the manuscript. I.B. conducted statistical analyses and wrote the initial draft of the manuscript. L.P.A., W.S., R.G.-K., J.M., N.H.W., R.D., A.W., A.J.B., A.D., S.P., T.W.G., and J.M.L. contributed revisions to the manuscript. X.G. assisted with statistical analyses and contributed revisions to the manuscript. All authors approved the final content. I.B. is the guarantor of this work and, as such, 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.