Global Prevalence and Major Risk Factors of Diabetic Retinopathy

We first performed a systematic literature review to identify all population-based studies that had ascertained DR from fundus (retinal) photographs. English-language articles were retrieved using Medline, EMBASE, Current Contents, EBSCO, JSTOR, and Science Direct using the following search terms: “diabetes” and “retinopathy” or “diabetic macular edema” and “population.” We identified 3,539 citations identified to 10 February 2010. Irrelevant and duplicate citations were excluded after a review of the titles and abstracts. The full texts of the remaining articles were reviewed to ensure studies met inclusion and exclusion criteria. In addition, we manually reviewed bibliographies of included articles and consulted with colleagues to identify other potentially relevant population-based studies that had assessed DR from fundus photographs but which may not have published results or in which grading for DR was still ongoing.

Studies were excluded if they were not population-based and/or if fundus photographs were not undertaken to ascertain DR. Two investigators (J.Y., R.Kaw.) independently selected the studies for inclusion. Disagreements between the two were resolved by adjudication with two additional reviewers (S.R., T.Y.W.).

We identified 58 population-based studies in which fundus photographs were potentially assessed for DR. Principal investigators of these identified studies were then invited for collaboration in this individual participant meta-analysis. We requested individual participant data regarding presence and severity of DR, DME status, age, sex, ethnicity, diabetes type and duration, hemoglobin A_1c (HbA_1c), systolic and diastolic blood pressure, lipid profile, cigarette smoking status, BMI, and current use of diabetes, antihypertensive, and lipid-lowering medications.

Investigators from 35 of the 58 identified studies provided data for this analysis (Table 1). Investigators of the remaining 23 studies could not or did not want to participate, or did not respond to repeated invitations. All studies had institutional board review approval and provided appropriately deidentified data for analysis.

Retinal photography was performed in all 35 studies according to standardized protocols. Most of the studies graded for DR using the Early Treatment Diabetic Retinopathy Scale (ETDRS) and its modification or the American Academy of Ophthalmology (AAO) International Clinical Diabetic Retinopathy Disease Severity Scale (Table 1).

DR severity was categorized as non-PDR (NPDR; level 20 through level 53) and PDR (level ≥60). DME was defined as absent or present. The four primary outcomes for this study were based on the severity in the worse eye or of the single eye that was photographed. Any DR was defined as the presence of NPDR, PDR, DME, or any combination thereof; and VTDR was defined as the presence of PDR and/or DME. These composite outcomes serve as the primary outcomes for this report, which respectively, indicate presence of any DR and severe DR likely to result in vision loss if left untreated.

Not all studies reported information on diabetes type. If data on age at diagnosis of diabetes were available in these studies, participants were classified as type 1 if they were diagnosed before age 30 years and as type 2 if they were diagnosed with diabetes after age 30 years, as previously used in one study (7). Hypertension was defined in subjects with a blood pressure >140/90 or who reported being on treatment for hypertension. Serum cholesterol was categorized into levels <4.0 or ≥4.0 mmol/L.

Study methodology and heterogeneity were assessed independently by two investigators (J.Y., R.Kaw.). Any disagreement was settled by consensus or adjudication with a third reviewer (S.R.). Studies were assessed for a list of attributes as defined in Supplementary Table 1. Studies with similar methodologies and rigorous ophthalmologic definitions were defined as those with a score of ≥9 (maximum, 11).

Data from each study were checked for consistency in variable definition before pooling, and where appropriate, data were recategorized according to a common definition. Race/ethnicity was categorized as Caucasian (Europeans and those of European origin), Asian (Chinese, Chinese American, Japanese, Malay, Indian, or people of Asian origin), African American, and Hispanic (Mexican Americans). Asians were further subdivided into Chinese or Japanese origin, and South Asian (Indian, Malay, South Indian, Thai, etc). Study-specific and pooled-data estimates of the prevalence of any DR, PDR, DME, and VTDR were directly age-standardized to the 2010 world diabetes population aged 20–79 years (8) using age strata 20–39, 40–59, and 60–79 years. We calculated 95% CIs for standardized prevalence rates using a normal approximation and Breslow-Day standard errors, after being modified to use a binomial assumption for the variance of the crude stratum-specific rates (9).

Initial analyses included data from all 35 studies, and subsequent analyses were performed using only data from studies with similar methodologies and outcome definitions (i.e., studies with a score of ≥9). Results from the latter analyses are presented throughout this report because of their similar methodologies.

Poisson regression models with robust error variance were used to estimate relative risks for DR, PDR, DME, and VTDR by categories of risk factors (e.g., hypertension, duration), adjusting for age (continuous, from 20–79 years), race (five categories), hypertension (yes/no), HbA_1c (four categories) and study, as appropriate. We also performed supplementary analyses on the interaction between diabetes type and duration, using people with type 2 diabetes for <10 years as the reference group. Including sex in regression models generally did not improve the model fit and did not appreciably alter the results.

The total number of patients with diabetes with DR aged between 20 and 79 years was estimated by multiplying the 2010 country-specific totals of people with diabetes (sourced from Diabetes Atlas) by our pooled racial group–specific rates of DR using the most predominant racial group per country; for example, in Brazil, where 53.7% of country is “white” (according to 2000 census results listed in Central Intelligence Agency, The World Factbook) (10), our pooled Caucasian rate was applied, and in countries where the predominant racial group did not easily align with our limited pooled racial groups (e.g., Melanesians in Papua New Guinea), the overall pooled world rate was applied.

All analyses were undertaken using Stata Intercooled 11.1 software (StataCorp LP, College Station, TX).

Data were collated from 22,896 individuals from 35 studies in the U.S., Australia, Europe, and Asia. Of these, 52% were female, 44.4% were Caucasian, 30.9% were Asian, 13.9% were Hispanic, and 8.9% were African American. The mean age was 58.1 years (range 3–97), median diabetes duration was 7.9 years (interquartile range [IQR] 3–16), and median HbA_1c was 8.0% (6.7–9.9%). Summary characteristics of the diabetic participants from each of the included studies are presented in Table 1 and Supplementary Table 2.

Analyses of these 35 studies showed that the overall age-standardized prevalence of any DR was 34.6% (95% CI 34.5–34.8), PDR was 6.96% (6.87–7.04), DME was 6.81% (6.74–6.89), and VTDR was 10.2% (10.1–10.3; data not shown). Analyses confined to studies with similar methodologies and rigorous outcome definitions showed that the age-standardized prevalence was 35.4% (35.2–35.6) for any DR, 7.24% (7.15–7.33) for PDR, 7.48% (7.39–7.57) for DME, and 11.7% (11.6–11.8) for VTDR (Table 2). There was no discernible sex difference in the prevalence of any DR or for PDR, DME, or VTDR. Extrapolating these prevalence rates to the 2010 world diabetes population, we estimate that 92.6 million (91.2–94.0) adults had any DR, 17.2 million (16.6–17.7) had PDR, 20.6 million (19.6–21.6) had DME, and 28.4 million (27.6–29.2) had VTDR.

Table 3 reports the age-standardized prevalence of any DR by retinopathy risk factors and other subgroups of interest. The prevalence of any DR varied across ethnic groups and was highest among African Americans and lowest among Asians. The prevalence of any DR increased with diabetes duration (21.1 vs. 76.3%, comparing <10 with ≥20 years), HbA_1c (18.0 vs. 51.2%, comparing levels ≤7.0 with >9.0%), and blood pressure (30.8 vs. 39.6%, comparing blood pressure ≤140/90 or >140/90), and was higher in people with type 1 than type 2 diabetes (77.3 vs. 25.2%). Similar relationships were also evident in the prevalence patterns of PDR, DME, and VTDR. There was a trend toward a higher prevalence of VTDR stages, but not any DR, in people with cholesterol levels ≥4.0 mmol/L. Analysis by year/period of fundus photography suggests a decline in the prevalence of any DR in the post-2000 era (Table 3).

After adjusting for known risk factors, individuals with type 1 diabetes for ≥20 years were 2.7 times more likely to have any DR (relative risk 2.69 [96% CI 2.47–2.93]), 15 times more likely to have PDR (15.3 [11.3–20.8]), 5 times more likely to have DME (4.83 [3.71–6.30]), and 8.7 times more likely to have VTDR (8.69 [7.10–10.63]) compared with those with type 2 diabetes for <10 years (Table 4).

This study provides a global estimate of the prevalence of DR and the severe stages of DR (PDR, DME) using individual-level data from population-based studies worldwide. On the basis of the data from all 35 studies on more than 20,000 participants with diabetes, we estimated that among individuals with diabetes, the overall prevalence of any DR was 34.6%, PDR was 7.0%, DME was 6.8%, and VTDR was 10.2%. Analyses confined only to studies with similar methodologies and ophthalmologic definitions showed that the age-standardized prevalence of any DR was 35.4%, PDR was 7.2%, DME was 7.4%, and VTDR was 11.7%, among individuals with diabetes. The prevalence estimates of any DR and VTDR were similar in men and women and were highest in African Americans and lowest in Asians. Prevalence rates were substantially higher in those with type 1 diabetes and increased with duration of diabetes, and values for HbA_1c, blood pressure, and cholesterol. Extrapolated to the world diabetes population in 2010, we estimate that approximately 93 million may have some DR, and 28 million may have sight-threatening stages of DR.

The prevalence of DR has been previously reported in a number of population-based samples (11–16). However, prevalence estimates varied considerably across some studies, depending on the population and study methodology. For example, variable prevalence rates were reported between populations of different ethnicities (e.g., 32.4% in an Australian Caucasian cohort (14) vs. 48.0% in a Mexican American cohort (15)) as well as between different populations of the same ethnicity (e.g., 35% in a U.S. Caucasian cohort (13) and 15.3% in a more recent Australian Caucasian cohort). More important, prevalence estimates for the more severe and vision-threatening end points, such as PDR and DME, are scarce, due to the small numbers of these cases from individual population-based studies. Published estimates for VTDR prevalence (17–20), for example, ranges widely, from 1.2 (17) to 32.2% (18). Our study provides the first precise estimates for these important clinical subgroups of DR.

The most comparable study to ours is the pooled analysis for prevalence of DR in the U.S. (6). On the basis of eight population studies derived from the U.S. and Australia, an overall prevalence of 40% for any DR and 8% for VTDR was reported (6). These estimates, however, represented findings limited to individuals aged older than 40 years and only with type 2 diabetes, were largely derived from individuals of Caucasian background, did not evaluate PDR and DME separately, and did not include studies from Asia. Ours is the first synthesis of individual-level data from all eligible population-based studies worldwide with a sufficiently large sample to allow a more precise estimation of the prevalence of PDR and DME.

Some of the differences in DR prevalence between individual studies may be partly attributed to the differing periods of the studies (Table 1 and Supplementary Table 3). Improvements in the management of DR and diabetes, and increased screening for diabetes, may have led to lower DR incidence and prevalence over time (21). Furthermore, DR susceptibility may also vary among ethnic groups. In support of the latter hypothesis, a number of multiethnic cohort studies have reported a higher DR prevalence among Mexican Americans than in non-Hispanic whites (5,22,23). Others, however, showed a similar or lower prevalence of DR in African Americans (18) and Mexican Americans (24) than in non-Hispanic whites. In some studies (5), after adjusting for putative DR risk factors, racial differences in the prevalence of DR was attributed to differing levels of risk factors for DR, but in others, the excess risk was unexplained (22,23,25). Differences in socioeconomic factors, including access to and the level of diabetes care, and possibly genetic susceptibility (26), may also possibly explain some of the disparities in rates and severity of DR in the different ethnic groups. In addition, racial differences in the effect of DR risk factors could also have accounted for some of these variations (23,27). Population-based studies incorporating host and environmental data are needed to further clarify the effect of race and ethnicity on DR prevalence.

We highlight several key points regarding the major risk factors for DR: First, we confirm the importance of the three major risk factors for DR—diabetes duration (17,19,28), HbA_1c (17,28–32), and blood pressure (17,28,33)—and suggest that they apply broadly across the mild to vision-threatening stages of DR.

Second, we establish that higher total serum cholesterol was associated with a higher prevalence of DME, bringing clarity to previously conflicting reports about this risk factor (19). This is particularly relevant to recent reports from trials suggesting that fenofibrate, a lipid-altering agent, may slow the development and progression of DR (34). Fenofibrate, however, acts mostly on triglycerides, and its effects on retinopathy in those trials were independent of lipid levels achieved. Statins, however, did not affect DR severity in the few studies in which this was evaluated, although not as a primary outcome (35,36).

Third, we provide estimates of risk of DR by diabetes type, in which studies in individuals with type 1 diabetes are currently scarce. We showed that the prevalence of DR is substantially higher in type 1 than in type 2 diabetes (11,37), an outcome independent of diabetes duration. However, because we classified type of diabetes by age of onset (younger or older than age 30 years), in some studies there may be potential misclassification (e.g., some people with type 2 diabetes will be younger than 30 years).

The strengths of our study include a large sample size to determine prevalence and risk factor associations for sight-threatening end points (PDR, DME), the inclusion of diverse ethnic population samples from around the world, and studies that had used photographic documentation of DR.

Our study has limitations. Pooling of data from various sources introduces many potential sources of heterogeneity that could influence accuracy; thus, although our estimates are highly precise, their accuracy is unknown. Samples of different study designs could have considerably different inclusion criteria, sample selection, and study protocols. For example, population samples could have varied considerably between a cardiovascular disease study and an eye survey, or a study on diabetes complications.

There was also a range of methods used in ascertaining diabetes status. Studies in which diagnosis of diabetes was based on self-report, without confirmation from blood tests, could have resulted in an overestimate of DR prevalence rates because those with undiagnosed diabetes might have been erroneously excluded from the sample denominator.

Furthermore, there were differences in the methodologies used to detect and diagnose DR, such as the number of eyes photographed per subject, number of retinal fields examined per eye, and the grading protocols and definitions used. In studies that did not collect data on diabetes type, this information was defined on the basis of age of diagnosis, with a cutoff at age 30 years to use as many studies with detailed information other than types of diabetes. Misclassification could have occurred as a result of this assumption. This, however, would not have affected the overall prevalence estimates but could have had a small effect of attenuating the comparative estimates between the type 1 and type 2 diabetes groups. A few studies with large numbers of participants could have influenced our results. Finally, the absence of studies from the Middle East, Africa, or South America could also affect the accuracy of our findings.

In conclusion, our current study provides the first global estimate of DR and, more important, the two sight-threatening end points (PDR and DME), based on a pooled individual participant analysis of more than 20,000 participants from 35 studies around the world. Our study shows that 35% of people with diabetes had some form of DR, and that 7% had PDR, 7% had DME, and 10% were affected by these vision-threatening stages. We estimate that in 2010, approximately 93 million were affected by DR, and 28 million by VTDR. This suggests that DR has the potential to be the leading cause of visual impairment and blindness worldwide. We confirmed the importance and impact of three major modifiable risk factors—hyperglycemia, hypertension, and dyslipidemia—on the risk of all DR end points, including for the first time, PDR and DME. These results highlight the substantial public health effect of diabetes, and thus, the need for effective screening and management of DR risk factors.

Funding for the data pooling analysis was provided by Global Health Outcomes Research, Allergan, Inc. (Irvine, CA). The National Institutes of Health grant EY-06594 (to B.E.K.K., R.Kle.) and, in part, by the Research to Prevent Blindness (to B.E.K.K., R.Kle., Senior Scientific Investigator Awards), New York, New York, provided funding for the entire study, including collection and analyses and of data.

Andhra Pradesh Eye Disease Study: Supported by grants from the Hyderabad Eye Research Foundation, Hyderabad, India and Christoffel-Blindenmission, Bensheim, Germany. Funagata Study: Supported by Grant-in-Aid Global COE program of the Japanese Society for the Promotion of Science, Japan. Blue Mountains Eye Study: Supported by grants from the Australian National Health & Medical Research Council. New Jersey 725 Study: Supported by grant R01-EY-09860 from the National Eye Institute. Samutsakon Study: Supported by grants provided by the Samutsakhon General Hospital. Rotterdam Study Ophthalmology: Swart van Essen, Rotterdam; Blindenpenning, Amsterdam; Blindenhulp, The Hague; Algemene Nederlandse Vereniging ter Voorkoming van Blindheid (ANVVB), Doorn; Stichting Oogfonds Nederland, Utrecht; Stichting Lijf en Leven, Krimpen aan de Lek; Rotterdamse Vereniging Blndenbelangen, Rotterdam; MD Fonds, Utrecht; Oogfonds Nederland, Utrecht; Laméris Ootech BV, Nieuwegein; Medical Workshop, de Meern; Topcon Europe BV, Capelle aan de IJssel, all in the Netherlands. No other potential conflicts of interest relevant to this article were reported.

J.W.Y.Y. researched the data and wrote and edited the manuscript. S.L.R. analyzed the data and reviewed and edited the manuscript. R.Kaw., E.L.L., J.W.K., T.B., S.-J.C., J.M.D., A.F., J.G., S.H., R.F.H., M.K.I., T.K., B.E.K.K., R.Kle., S.K., K.M., J.P.O., T.J.O., M.P., M.R., M.S.R., T.S., J.S., H.T., J.M.T., R.V., J.J.W., N.W., S.W., L.X., M.Y., X.Z., P.M., and T.Y.W. contributed to discussion and reviewed and edited the manuscript. The sponsors or funding organizations had no role in the design, conduct, analysis, or publication of this research. T.Y.W. 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.