Longitudinal Studies of Incidence and Progression of Diabetic Retinopathy Assessed by Retinal Photography in Pima Indians

Data for the present analysis come from a longitudinal study among the Pima Indians of the Gila River Community. Community residents were invited to undergo examinations on a biennial basis from the age of 5 years onward. These examinations included measures of BMI, blood pressure, plasma glucose, HbA₁, cholesterol, and albuminuria. Details of medication use were collected. Diabetes was diagnosed on the basis of a 75-g glucose tolerance test according to World Health Organization guidelines (5) or a previous clinical diagnosis. All participants ≥15 years old had direct ophthalmoscopy performed after pupil dilation by a physician who was not aware of other clinical details. None of the physicians were ophthalmologists by training, but all underwent an initial training period before undertaking fundoscopic examinations in the clinic. Retinal photographs (two standard fields for each eye) were taken at the same visit using a Cannon CR4–45NM fundus camera. For the period 1 April 1982 to 31 December 1990, these photographs were graded in a standardized manner without knowledge of clinical details according to a modification of the Airlie House classification system and a modification of the Early Treatment Diabetic Retinopathy Study severity system of diabetic retinopathy (6), which assigns a score to each eye (Table 1). Severity of retinopathy was assessed with an 11-level scale (10/10, 21/10, 21/21, 31/<31, 31/31, 41/<41, 41/41, 51/<51, 51/51, 60+/<60+, and 60+/60+, where the first number is the severity level of retinopathy in the worst eye and the second number is the equal or lesser score in the other eye). We divided retinopathy severity derived by fundoscopy findings into three groups: no retinopathy (including isolated hard exudates with no other sign of retinopathy, because they may be misclassified as drusen and are a common reported finding in nondiabetic Pima Indians [7]), NPDR in which at least one eye had one or more microaneurysm, and proliferative retinopathy in which at least one eye had signs of proliferative retinopathy.

Gradable photographs and fundoscopy data were available for both eyes in 1,585 examinations of 931 people. This occurrence represents 71% of the 2,247 examinations performed on diabetic participants during the study period. Photographs were not taken in 398 examinations (e.g., because of equipment failure or lack of available personnel), 240 examinations were excluded because there was one or more ungradeable photograph (e.g., because of cataracts or technical problems), and a further 24 examinations lacked fundoscopy data (e.g., because of cataracts). The prevalence of retinopathy by fundoscopy in those examinations that did not have retinal photography data was 16.9%. At baseline, the mean age, duration of diabetes, and HbA₁ levels for the 931 people included in our study were 43.8 years, 7.3 years, and 9.8%, respectively, compared with 51.8 years, 11.1 years, and 9.8%, respectively, for those not included. Figure 1 outlines the study population at baseline.

We examined the relation between diabetes duration and retinopathy grade assessed by retinal photography. Duration was categorized into 5-year intervals. For individuals with two or more examinations within a duration interval, only one examination was included. Thus, there were 1,230 examinations included in that analysis. For individuals with multiple examinations during the study, we examined how photographic findings changed from one examination to the next.

Participants with diabetes but without retinopathy at their baseline examinations were followed until retinopathy developed or until the final examination in the follow-up period if retinopathy did not develop. Cumulative incidence was calculated by the product limit method. We used proportional hazards regression to calculate the hazard rate ratios (HRRs) associated with individual factors, with control for potentially confounding variables. The validity of the proportionality assumption was tested for each variable by inclusion of a time-dependent term (8). If the time-dependent term was statistically significant (P < 0.05), the model was repeated using truncated follow-up.

We also evaluated the factors associated with progression of baseline NPDR by looking for factors that were associated with a worsening of two steps or more (e.g., level 21/21 to ≥31/31) or an improvement of two steps or more in retinal photography grade compared with subjects whose grade remained stable between examinations. Analysis was restricted to people with a baseline grade of 21/21 to 51/51 and one or more follow-up examination; when more than one follow-up examination was available, we used data from the last examination for the outcome grade. ANOVA was used to compare levels of continuous variables for subjects whose retinopathy progressed, regressed, or remained stable; analysis of contingency tables was used to compare dichotomous variables. P values were calculated for the comparison of all three groups and for an ordered categorical variable to enhance power to detect differences that follow a linear pattern among groups (P trend).

Baseline variables analyzed included age, sex, duration of diabetes, fasting and 2-h glucose, HbA₁, BMI, cholesterol, hypertension, macroalbuminuria, and use of insulin or oral hypoglycemic therapies. Macroalbuminuria was defined as an albumin-to-creatinine ratio in a spot urine sample of ≥0.3 g/g. Hypertension was defined as a systolic blood pressure ≥140 mmHg, a diastolic blood pressure ≥90 mmHg, or use of antihypertensive medication.

Based on the photographic assessment, the most common finding in the first decade after diagnosis of diabetes was the absence of retinopathy (Fig. 2). NPDR became more prevalent after 10 years’ duration of diabetes, with moderately severe NPDR being most frequent in subjects with 10–25 years’ duration of diabetes. Similar results were obtained using fundoscopy data (data not shown).

When longitudinal changes in the severity of retinopathy were examined, subjects with minimal NPDR at baseline (levels 21/10 and 21/21) were more likely to improve at the subsequent examination rather than worsen (Fig. 3). Although progression and regression of retinopathy was seen in eyes with more severe NPDR at baseline, the severest NPDR or proliferative retinopathy rarely regressed.

There were 280 individuals (77 men, 203 women) without retinopathy on photography at the baseline examination. These individuals had a mean age of 41.9 years and a mean duration of diabetes of 4.6 years. There were 75 incident cases of retinopathy over a mean follow-up of 4.1 years. The 4-year cumulative incidence was 16.8%. Using univariate analysis, duration of diabetes, fasting and 2-h glucose, HbA₁, albuminuria status, and type of hypoglycemic therapy were statistically significantly associated with incident retinopathy. These factors remained significantly associated with retinopathy incidence when adjusted for age and sex in multivariate models. When duration, HbA₁, macroalbuminuria, and hypoglycemic therapy were included in a model together, only HbA₁ and the use of oral hypoglycemic agents remained statistically significantly associated with incidence of retinopathy (Table 2).

Analysis of all data (both at baseline and at follow-up examinations) showed that fundoscopy identified fewer eyes with retinopathy than did grading of retinal photographs (Fig. 1). This was primarily due to more cases of NPDR being identified by retinal photography than by fundoscopy. Comparing the two approaches resulted in an unweighted κ statistic for the two methods of assessment of 0.50 (95% CI 0.45–0.54); this result represents fair to good agreement between the diagnostic methods.

When fundoscopy was used to assess incidence of any retinopathy, the baseline group consisted of 337 people (91 men, 246 women) with a mean duration of diabetes of 6 years and a mean age of 42.6 years. A total of 55 people developed retinopathy over a mean follow-up of 4.2 years. The 4-year cumulative incidence was 10.9%. HRRs were generally similar to those found using the retinal photography data, although older age and lower BMI at baseline were also statistically significant risk factors (Table 2).

We examined the subset of individuals (n = 119) with NPDR at baseline (levels 21/21 to 51/51) to determine factors associated with progression or regression of retinopathy. A total of 67 (58.3%) individuals had retinopathy that did not change by two or more grades between examinations, 27 (23.5%) had an improvement of two or more grades, and 21 (18.3%) had a worsening of two or more grades. Baseline macroalbuminuria and hyperglycemia were statistically significantly associated with progression of retinopathy after adjustment for age and sex (Table 3).

This study examined the changes in retinopathy as determined by grading of retinal photographs and fundoscopy through a dilated pupil. We found that the prevalence of more severe grades of retinopathy is higher in people with longer durations of diabetes. However, for those with mild NPDR, there was a greater probability of improvement than worsening of retinopathy at a subsequent examination. Once lesions associated with more severe NPDR were present, there was less regression.

Earlier studies examining the relationship of risk factors with the incidence of retinopathy in the Pimas have used fundoscopy for the assessment of retinopathy (9,10), except for a study of retinopathy prevalence (11) and a comparative study of various glycemic measures that was based on a subset of the present data within a relatively short follow-up (12). Thus, the present study represents the first longer-term follow-up study in Pimas to use retinal photography to assess a broad range of potential risk factors for retinopathy. In addition, prevalence according to severity of NPDR has not been presented previously in this population.

Risk factors for incidence of retinopathy identified in previous studies that used fundoscopy include longer duration of diabetes, hyperglycemia, presence of proteinuria, and insulin treatment for retinopathy (7,13). The present analysis using retinal photography to assess incidence of retinopathy identified these same factors. Retinal photography identified more subjects with retinopathy than fundoscopy, a finding reported in other studies (14). In general, however, the present study shows that HRRs were similar regardless of whether retinopathy was detected by retinal photography or fundoscopy through a dilated pupil. In fact, in many cases, a stronger HRR was seen with fundoscopy than with retinal photography. Whereas many studies have examined risk factors for the incidence and progression of retinopathy using either photography or fundoscopy, we are not aware of a study that has used both to assess risk factors in the same subjects. These findings suggest that conclusions regarding risk factors for incident retinopathy in epidemiologic studies may be similar regardless of whether retinal photography or fundoscopy is used to assess retinopathy, although the methods produce differing incidence and prevalence estimates. The lower incidence seen with fundoscopy suggests that a study that uses this method to assess retinopathy may need larger numbers of participants than a study using retinal photography to have comparable power.

The retinal photography grading system puts isolated hard exudates into the mild NPDR group, whereas our fundoscopy classification included them in the no retinopathy group. If we include isolated hard exudates on fundoscopy in the NPDR group, an additional 143 examinations are classified as NPDR and the recalculated κ statistic is 0.49 (95% CI 0.45–0.54). However, the HRRs for risk factors for the development of retinopathy were essentially unchanged after this adjustment (data not shown).

Our 4-year cumulative incidence rates of retinopathy based on grading of fundus photographs are lower than rates published in other studies and remain so even after adjusting for the variable durations of diabetes in the studies (2,15). This may be due to the regular screening for diabetes of the Pima population; therefore, Pimas are more likely to be diagnosed earlier than in other studies. Additionally, individuals with more severe diabetes who are most likely to develop retinopathy (or to progress) may have been less likely to be assessed in this study for retinopathy, either because they return for examinations less frequently because of death or disability or because their photographs are more likely to be ungradeable (e.g., because of cataracts). Although there was no significant difference in mean age, duration of diabetes, measures of glycemia, or baseline retinopathy grades between subjects who had one and subjects who had multiple examinations, the group who attended only once had a greater prevalence of macroalbuminuria and hypertension (data not shown) and there were more people with proliferative retinopathy at baseline. Comparing the baseline data for the 931 people included in our study to the baseline examinations of the 227 who were not included because of missing data showed that the people in our analysis were younger at baseline, had shorter mean duration of diabetes, and had less macroalbuminuria and hypertension, and fewer were taking insulin. Examinations were scheduled at 2-year intervals, so subjects with baseline examinations after January 1988 (189 people) were not expected to attend follow-up. This suggests that if retinopathy data had been available on the whole population, we may have found higher incidence and more progression of retinopathy than we report here.

Because many individuals attend examinations only intermittently, a possible concern is that proportional hazards models may produce biased results if the length of follow-up is associated with any of the variables considered as risk factors. However, an analysis of which baseline variables predicted prevalence of retinopathy at the second examination (logistic regression) produced similar conclusions to those presented here. This suggests that any bias in the proportional hazards analysis produced by intermittent follow-up is likely to be minimal.

Many of the factors found to be associated with incidence of retinopathy in Pimas have been found in other studies. Hyperglycemia (16–19) and longer duration of diabetes (3,16,18,19) are probably the most commonly reported associations with incidence of retinopathy. The presence of macroalbuminuria has been reported as a risk factor for the incidence of retinopathy (16,18), although not in all studies (20). We did not find a statistically significant association between hypertension and incidence or progression of retinopathy. This is in keeping with studies in some other populations (19,21) but not all (3,17). It is likely that the association with hypoglycemic treatment we found reflects its role as a marker for severity of diabetes.

When using the cutoff of a change of two or more grades of retinopathy for subjects with baseline NPDR, we found that baseline fasting and 2-h glucose and nephropathy were significantly different for subjects who had progressed two or more grades compared with subjects who had regressed two or more grades, although change in HbA₁ between examinations was not statistically significantly different between the three groups. Progression of two or more grades in a 4-year period is a strong predictor for the later development of proliferative retinopathy (22). The U.K. Prospective Diabetes Study used similar methodology to quantify progression of existing retinopathy and reported associations with baseline and mean follow-up HbA_1c, older age, and male sex for progression of two or more grades (3). We report a relatively high level of regression of NPDR, which could be due to missing data for subjects with progression either through lack of follow-up examinations or an inability to grade photographs because of increasing media problems, such as cataracts. It could also be that retinopathy detected at an early stage is more likely to regress in the short term than it is at longer durations of diabetes. The Wisconsin Epidemiologic Study of Diabetic Retinopathy reported regression rates of between 10.5 and 25% at 4 years (23) and between 0 and 26.4% at 10 years (24) for subjects with minimal or moderate NPDR at baseline.

The present study shows that, among Pima Indians, the presence of mild NPDR is more likely to be followed by findings of no retinopathy at a subsequent examination than progression. However, more severe NPDR is unlikely to regress. Duration of diabetes, hyperglycemia, pharmacological treatment for diabetes, and macroalbuminuria appear to be the major factors associated with the development of any level of retinopathy, as diagnosed by either retinal photographs or fundoscopy. Once retinopathy has been established, hyperglycemia and macroalbuminuria remain risk factors for its progression.

We thank the members of the Gila River Indian Community and everyone at the National Institutes of Health clinic in Sacaton, AZ. We also thank Dr. Robert Nelson for his advice during the preparation of this manuscript.