The Prevalence of Diabetic Retinopathy Among Adults in the United States FREE

Diabetic retinopathy, a retinal vascular disorder that occurs as a complicationof diabetes mellitus (DM), is a leading cause of blindness in the United States,often affecting working-aged adults.¹ It ischaracterized by signs of retinal ischemia (microaneurysms, hemorrhages, cotton-woolspots, intraretinal microvascular abnormalities, venous caliber abnormalities,and neovascularization) and/or signs of increased retinal vascular permeability.Vision loss can result from several mechanisms, including neovascularizationleading to vitreous hemorrhage and/or retinal detachment, macular edema, andretinal capillary nonperfusion.¹ Retinopathyoccurs in most persons with long-standing DM,² butits incidence rate can be reduced by aggressive control of hyperglycemia^3- 4 and hypertension.^3- 6

The Eye Diseases Prevalence Research Group, a consortium of population-basedeye studies and ophthalmic epidemiologists, was charged with generating thebest possible estimates of the prevalence rates for major eye diseases anddisorders, including diabetic retinopathy. Since 1980, several population-basedeye surveys have assessed the prevalence of diabetic retinopathy using the"gold standard" Early Treatment Diabetic Retinopathy Study (ETDRS) interimor final fundus photograph grading protocol for diabetic retinopathy.^7- 8 The National Health Interview Survey(NHIS) provides annual estimates of the self-reported prevalence of DM inthe United States, and the 2000 US Census population data were recently madeavailable for the populations of interest. In this article, we report prevalenceestimates for diabetic retinopathy in persons 40 years and older, based oncombined data from these sources.

To estimate the prevalence of diabetic retinopathy among persons withDM, data on participants 40 years and older were requested from population-based,cross-sectional eye diseases prevalence studies^9- 16 thathad ascertained diabetic retinopathy by grading color fundus photographs (Table 1).

Severity of diabetic retinopathy was graded in all studies by a colorfundus photograph reading center. The reading center also indicated whethermacular edema was observed.

Grading was performed using the ETDRS interim or final scale,^7- 8 except in the Barbados Eye Study, Barbados,West Indies (see below). Use of the ETDRS final scale to evaluate the severityof diabetic retinopathy is summarized in Table 2 and Figure 1.With the ETDRS interim scale, level 40/41 was approximately equivalent tolevel 43 of the final scale, and level 50/51 was approximately equivalentto level 53 of the final scale. For our evaluation of the prevalence of diabeticretinopathy, these severity scales were collapsed into the following categories,based on the consensus of the Eye Diseases Prevalence Research Group:

In natural history observations from the ETDRS, among eyes with level35 retinopathy (at the highest end of our mild level), 1.2%, 6.5%, and 15.2%progressed to high-risk proliferative retinopathy (warranting panretinal photocoagulation)¹⁷ within 1, 3, and 5 years' follow-up, respectively.Eyes with level 43 and 47 retinopathy (corresponding to our moderate level)progressed to high-risk proliferative retinopathy in 3.6% and 8.1%, 13.3%and 24.7%, and 21.0% and 27.1% of cases by 1, 3, and 5 years' follow-up, respectively.Eyes with level 53 retinopathy, at the low end of the spectrum of severe retinopathyas defined in our study, progressed to high-risk proliferative retinopathyin 17.1%, 44.4%, and 57.8% of cases by 1, 3, and 5 years' follow-up, respectively.⁷

The Barbados Eye Study used a valid, simplified grading scheme, in whichfundus photographs were graded as no retinopathy, mild retinopathy (3 microaneurysms,hard or soft exudates, or retinal hemorrhages), moderate retinopathy (intraretinalmicrovascular abnormalities or venous beading), or severe retinopathy (anyneovascularization).¹⁸ For purposes of pooledanalysis, severe retinopathy as defined in the Barbados Eye Study was usedto correspond to the severe category (level ≥50) used for the remainingstudies, recognizing that some cases of severe nonproliferative retinopathymay have been included in the Barbados Eye Study's moderate category. Mildretinopathy from the Barbados Eye Study was used to correspond to mild nonproliferativeretinopathy as defined earlier. The extent to which the differences in outcomedefinitions in the Barbados Eye Study affected results was explored by sensitivityanalysis.

We derived meaningful composite outcomes from these results as follows:(1) DR (any diabetic retinopathy [technically defined as level 14 or higherretinopathy and/or macular edema]), consisting of mild, moderate, or severeretinopathy, diabetic macular edema, or any combination thereof; and (2) vision-threateningdiabetic retinopathy (VTDR), consisting of severe retinopathy, diabetic macularedema, or both. These composite outcomes serve as the primary outcomes forthis article, respectively indicating (1) the presence of any diabetic retinopathy;and (2) a level of diabetic retinopathy likely to result in vision loss inthe short run, absent treatment with laser photocoagulation.

The contributing studies also provided data on age, gender, and race/ethnicity.Diagnosis of DM included laboratory verification for persons with DM, exceptin the Melbourne Visual Impairment Project, Melbourne, Victoria, Australia.Unfortunately, data on the duration of DM, type of DM, and severity of hyperglycemiaover time were not available from all studies in a manner that could be combined,so these factors could not be evaluated in the pooled analysis. For the southernWisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR), and San AntonioHeart Study, San Antonio, Tex, we elected to use only data for subjects withtype 2 DM, on the grounds that only a few persons 40 years and older wouldhave type 1 DM, so that the persons with type 2 DM would be approximatelyrepresentative of the US general population with DM in this age group. Forthe other studies, data for all persons 40 years and older identified as havingDM were used.

Data analysis was conducted by pooling the age-, gender-, and race/ethnicity-specificprevalence rates for DR and for VTDR among persons with DM from each of theparticipating studies. In accord with the established age groupings used forreporting the prevalence of DM from the NHIS,¹⁹ thefollowing age categories were used: 40 through 49, 50 through 64, 65 through74 years, and 75 years and older. Pooled prevalence estimates for each ofthese age groupings, by race/ethnicity and gender, were calculated using minimumvariance linear estimation logarithm odds transformation, of the prevalenceproportion. As in the other Eye Diseases Prevalence Research Group articles,racial/ethnic groups were defined as white, non-Hispanic; black, non-Hispanic;Hispanic; and other. Variances were computed assuming a binomial distribution.Prevalence rates were derived from the pooled data of all relevant studiesto obtain estimates for retinopathy among white and Hispanic persons withDM. Prevalence rates for retinopathy among black persons with DM were takendirectly as reported from the Barbados Eye Study. Prevalence rates for otherAmerican race/ethnicity groups that have not been studied were estimated usingan unweighted average of the pooled prevalence estimates for white, black,and Hispanic persons, so that national prevalence estimates could be generated.

To determine the number of individuals with diabetic retinopathy inthe US general population 40 years and older, we first estimated the numberof individuals with DM in strata of age, gender, and race/ethnicity. Theseestimates were based on self-reported answers to the question, "Have you everbeen told by a physician that you have diabetes?" from the most recently available(1999) NHIS. The NHIS, conducted by the National Center for Health Statistics,gathers data from a nationally representative sample of civilian, noninstitutionalizedpersons residing in the United States.¹⁹ Thestratum-specific 1999 prevalence rates for DM were applied to the stratum-specific2000 US Census population results²⁰ to arriveat estimates of the number of persons with DM in each stratum in 2000. Thepooled stratum-specific prevalence estimates for DR and VTDR among personswith DM then were applied to the estimates of the number of persons with DMin each stratum, resulting in stratum-specific estimates of the prevalenceof DR and of VTDR for the US population in 2000. A similar approach was usedto obtain prevalence estimates for mild, moderate, and severe diabetic retinopathy,and also for diabetic macular edema.

For confidence intervals based on the prevalence of DM in the generalpopulation, SUDAAN software (Research Triangle Institute, Research TrianglePark, NC) was used to apply the sampling weights from the complex multistagesampling design of the NHIS appropriately. Confidence intervals for prevalencerates and counts of persons with DR and VTDR in the general population werecalculated from variance values derived using the delta method for the productof 2 dependent binomial proportions (proportion of persons with DM times theproportion of persons with DM having retinopathy [either DR or VTDR]).²¹

Projections for the prevalence rates for DR and for VTDR in 2020 werebased on 2 scenarios, one assuming a constant stratum-specific prevalenceof DM over time, and the other using projected increases in the prevalenceof DM over the 20-year period. For both scenarios, a constant prevalence ofDR and of VTDR among persons with DM was assumed. For the latter scenario,we used the projections of increasing DM prevalence reported by Boyle et al,²² based on NHIS results during the 1980s and 1990s.These projections were based on a linear model (J. P. Boyle, PhD, writtencommunication, September 11, 2002), which allowed us to recalculate the anticipatedprevalence of DM in strata of age group and gender in 2020 by applying theslope to the 2000 rates. The ratio of the stratum-specific prevalence of DMin 2020 to that in 2000 then was calculated for each age and gender stratum.These ratios were applied to our 2000 age-, gender-, and race/ethnicity-specificvalues for DM prevalence to obtain stratum-specific estimates of the anticipatedprevalence of DM in 2020, used in our projections of diabetic retinopathyprevalence under the increasing DM prevalence scenario. The ratio derivedfrom persons aged 45 through 64 years in the model of Boyle et al²² was applied to both our 40- through 49- and 50- through64-year age strata.

Tests for age, gender, and race/ethnicity effects were conducted amongpersons with DM for DR and for VTDR, as well as for mild, moderate, and severediabetic retinopathy and for diabetic macular edema. Tests for gender differences,based on the pooled data from all contributing studies, were done separatelyby race/ethnicity using the Mantel-Haenszel χ² test, controllingboth for age and for the study effect. Wald χ² tests from logisticregression models of the pooled prevalence rates were used to evaluate race/ethnicityand age effects, adjusting for gender. Age, gender, and race/ethnicity effectson the prevalence of DR and of VTDR in the general population were evaluatedwith logistic regression models and Wald χ² tests. For agetests, comparisons were made using ordinal age categories (40-49, 50-64, 65-74,and ≥75 years), and odds ratios (ORs) were reported for the prevalencein 1 category with respect to the category below. Pooled data analysis wasconducted following approval by the institutional review board of The JohnsHopkins School of Medicine, and followed the principles of the Declarationof Helsinki.

Eight studies contributed data on diabetic retinopathy for 4440 personswith DM, of whom 615 were black (all from the Barbados Eye Study) and 1415were Hispanic (100%, 80%, and 65%, respectively, of participants in the ProyectoVER, Nogales and Tucson, Ariz; San Antonio Heart Study, San Antonio, Tex;and San Luis Valley Study, San Luis Valley, Colo). Of 2410 white subjects,1313 (54%) were from the WESDR. Characteristics of the subjects with DM fromparticipating studies are given in Table1. For DR (Figure 2) andVTDR (Figure 3), age-, gender-,and race/ethnicity-specific prevalence rates were overlapping between studies,except that the WESDR, conducted from 1980 through 1982, tended to have higherprevalence rates than did the other studies. Tests for homogeneity for eachof the stratum-specific pooled rates showed no statistically significant differencesin rates across studies when WESDR data were excluded.

Stratum-specific pooled prevalence estimate results for DR and for VTDRare presented below. For the sake of brevity, reporting of prevalence resultsfor the underlying primary outcomes (mild, moderate, and severe diabetic retinopathyand diabetic macular edema) will be deferred to a supplemental Web site at: http://www.nei.nih.gov/eyedata/, at a future date.

Estimated prevalence rates for DR and for VTDR among persons with DMare given in Table 3. The estimatedcrude prevalence of DR among persons with DM was high, with an overall crudeprevalence of 40.3% (95% confidence interval [CI], 38.8%-41.7%). The estimatedoverall crude prevalence of VTDR was 8.2% (95% CI, 7.4%-9.1%). Overall crudeprevalence estimates would have been slightly lower had the WESDR resultsbeen excluded (35.8% and 7.3%, respectively, for DR and VTDR).

Among persons with DM, little difference in the prevalence of retinopathybetween age groups was observed. In logistic models, the odds of DR differedby no more than 6% between successive age groups (40-49, 50-64, 65-74, and≥75 years) in any racial/ethnic group, although in some instances thesedifferences were statistically significant. The strongest association withage was observed for increasing estimated prevalence of diabetic retinopathyamong Hispanic persons with DM (OR per step in age category = 1.06, P <.001). Among black (OR = 0.97, P = .02) and white (OR = 1.00, P = .88) personswith DM, there was little change with age, whether or not WESDR data wereincluded. For VTDR, no substantial differences in estimated prevalence betweenage groups were observed in any of the racial/ethnic groups (OR range, 0.99-1.01).

As with age, no consistent association between gender and the prevalenceof diabetic retinopathy in persons with DM was observed. Although Hispanicwomen with DM had an estimated prevalence rate for DR (OR = 0.72, P = .004) and for VTDR (OR = 0.77, P = .02)approximately three fourths that among males, for black and white personswith DM, neither DR nor VTDR was associated with gender (OR range, 0.97-1.02).

In contrast, the prevalence of DR among persons with DM seemed to varymodestly among racial groups. The prevalence of DR in persons with DM wassubstantially higher among Hispanic subjects (OR = 1.17, P <.001) than among black persons. Comparisons of white personswith Hispanic or black persons differed according to whether data from theWESDR study (a study of white persons in which the rates of diabetic retinopathywere higher than in all of the other studies) were included, making inferencesdifficult. When WESDR data were included, the prevalence of DR was similaramong white and Hispanic persons (OR = 0.99, P =.61), but the prevalence was significantly lower among white than Hispanicpersons when WESDR data were excluded (OR = 0.944, P <.001).For the comparison of white with black persons, with data from WESDR excluded,the prevalence rates were less than 5% different (OR = 1.04, P = .02). For VTDR, the prevalence rates among persons with DM weresimilar in all 3 racial/ethnic groups, whether or not WESDR data were excluded(all ORs were between 0.98 and 1.02, all P valuesnonsignificant).

As mentioned previously, the definition of severe diabetic retinopathyused in the Barbados Eye Study did not include severe nonproliferative diabeticretinopathy (level 50-53), likely resulting in underestimation of the prevalenceof VTDR among black persons in our analysis. To evaluate the extent of underestimationthat may have occurred, the proportion of persons with level 50 or worse retinopathywho had level 50 (severe nonproliferative) retinopathy in the study with highestDR prevalence (WESDR), 40 years or older and diagnosed as having DM at 30years or older, was calculated, and found to be 14.6% (B. E. K. Klein, MD,MPH, and R. K. and S. E. M., unpublished data, 2003). Thus, our estimatesof the prevalence of severe retinopathy among black persons are probably underestimatedby about 15%. The Barbados Eye Study definition of mild retinopathy also differedslightly from that used in the other studies, in that patients with 1 or 2microaneurysms may not have been counted as having DR. However, based on clinicalexperience, the effect of this discrepancy on prevalence estimates was likelytrivial.

The prevalence of diabetic retinopathy in the general population isstrongly dependent on the prevalence of DM itself, because only persons withDM can have diabetic retinopathy. Our estimates of the age-, gender-, andrace/ethnicity-specific prevalence rates for DM in persons 40 years and older,derived from the 1999 NHIS data set,¹⁹ aregiven in Table 4, and the estimatesof the prevalence rates for DR and VTDR are given in Table 5. The estimated crude prevalence of DR in the US population40 years and older was 3.4% (95% CI, 3.2%-3.6%) or 4.1 million persons. Theestimated crude prevalence of VTDR in the same population was 0.75% (95% CI,0.66%-0.85%) or 899 000 persons. Excluding WESDR data, the crude prevalenceestimate for DR would be 2.81% or 3.4 million persons, and the crude prevalenceestimate for VTDR would be 0.57% or 680 000 persons.

Reflecting the increasing prevalence of DM with age, DR and VTDR amongthe US general population 40 years and older tended to increase in prevalencewith age. The prevalence of DR increased on average across successive agegroups among white (OR per step in age category = 1.47, P <.001; OR = 1.48, P <.001 withoutthe WESDR data), black (OR = 1.30, P <.001), andHispanic (OR = 1.58, P <.001) persons. For VTDR,the US general population prevalence also increased on average across successiveage groups among white (OR = 1.34, P = .008; OR =1.15, P = .39 without the WESDR data), Hispanic (OR= 1.56, P <.001), and black persons (OR = 1.31, P = .06), but not all increases were statistically significant.However, lower prevalence rates of DR and of VTDR generally were observedin the oldest (≥75 years) age group with respect to those aged 65 through74 years.

A significant gender difference in the estimated US general populationprevalence rates for DR and VTDR was observed in only 1 of 6 tests for gendereffects, that for the prevalence of VTDR among black persons, wherein femalepersons had higher rates than male persons (OR = 2.53, P = .03). No statistically significant differences were observed betweengender in any racial/ethnic group for DR, nor were substantial gender differencesin the prevalence of VTDR observed in white and Hispanic persons.

Differences in the prevalence of DR between racial/ethnic groups inthe US general population were larger than the differences among persons withDM, because of statistically significant differences between racial/ ethnicgroups in the prevalence of DM. Higher rates of DR in the general populationwere observed for Hispanic compared with white persons (OR = 1.42, P <.001; OR = 1.63, P <.001 excludingthe WESDR data), and for Hispanic compared with black persons (OR = 1.52, P = .01). Black persons also tended to have higher ratesof DR than white persons (OR = 1.32, P = .10; OR= 1.75, P <.001 excluding the WESDR data). ForVTDR in the general population, higher rates were observed in Hispanic comparedwith white persons (OR = 1.75, P = .04; OR = 2.47, P =.004 excluding the WESDR data), and for black compared with white persons(OR = 1.32, P = .02; OR = 1.51, P = .003 excluding the WESDR data). Rates of VTDR for Hispanic andblack persons were similar (OR = 1.00, P = .98).

Applying age-, gender-, and race/ethnicity-specific rates of DR amongpersons with DM 40 years and older to projected changes in the US populationin the future, assuming the prevalence of DM remains constant in the interval,the projected prevalence rates of DR and of VTDR are approximately 3.95% (6.1million persons) and 0.88% (1.4 million persons), respectively, for 2020.If the expected increase in the prevalence of DM during the interval is considered,approximately 4.64% (7.2 million persons) and 1.02% (1.6 million persons)of the US general population 40 years and older, respectively, can be expectedto have DR and VTDR in 2020.

The results of our pooled analysis indicate that diabetic retinopathyaffects approximately two fifths of persons 40 years and older who identifythemselves as having DM. An estimated one twelfth of persons with DM in thisage group have reached the stage of vision-threatening disease. Even thoughdiabetic retinopathy is a disease occurring only among persons with DM, theprevalence of DM in the general population is high enough that diabetic retinopathyis highly prevalent in the general US adult population. Approximately 1 in29 Americans 40 years and older (4.1 million persons) has diabetic retinopathyof any level of severity, and approximately 1 in 132 persons (899 000persons) has VTDR. To the extent that the eyes randomized to deferral of therapyin the Diabetic Retinopathy Study reflect the extent of disease present ineyes of persons with VTDR in the current general population, one third ofthe persons with VTDR (approximately 1 in 400, or 300 000 persons) wouldbe expected to reach the legal blindness level (visual acuity of 20/200 orworse) in the affected eye(s) within 3 years absent appropriate applicationof photocoagulation, but only half to one tenth that many (1 in 800 to 1 in4000 persons) if appropriate photocoagulation treatment is applied.^17,23 The current proportions of visualimpairment and blindness in the general population attributable to diabeticretinopathy are reviewed in our accompanying article on visual impairmentand blindness.²⁴

While the general population prevalence rates for DR and VTDR are likelyto be lower in younger persons, owing to the lower prevalence of DM in youngerage groups,¹⁹ prevalence of type 2 DM is increasingin younger persons,²⁵ so this difference maynarrow in the future. The prevalence of diabetic retinopathy specificallyamong persons with type 1 DM is described in an accompanying article by Royet al²⁶ in this issue. Because the prevalenceof DR was higher in that group than in the current pooled analysis, our assumptionthat the prevalence of DR among persons with type 2 DM in the WESDR and SanAntonio Heart studies was approximately representative of the prevalence ofDR in the general adult population aged 40 years and older probably led tosome degree of underestimation of prevalence values.

Among persons with DM, the prevalence of DR did not vary substantiallyby age group or gender—only small differences were observed, and theirpattern was inconsistent. However, the general population prevalence of DRclearly increased with age, driven by the increasing prevalence of DM withage, but then declined beginning at the age of 75 years. A similar patternof reduced diabetic retinopathy prevalence in the oldest age groups was observedin Model Reporting Area data in 1970²⁷ andin the WESDR.²⁸ It seems likely that the apparentlyreduced prevalence of DR in the elderly could be explained on the basis ofreduced survival among elderly persons with DR (possibly a marker for moresevere DM and its life-threatening consequences). Although this theory cannotbe tested directly in our pooled analysis, evidence supporting it has beenobserved in the WESDR.²⁹

Although the prevalence of retinopathy among persons with DM appearedto vary by racial/ethnic group, comparison of prevalence rates in white personswith DM with respect to Hispanic and black persons with DM was difficult tointerpret, because conclusions differed depending on whether or not the WESDRresults were included. Likewise, comparisons involving black persons werebased on the results of a single study, which used slightly different retinopathyseverity definitions that we could not perfectly adjust for. Considering theselimitations, our data suggest that the prevalence of DR is higher among Hispanicthan black persons with DM, and may be higher among Hispanic than white personswith DM. These interpretations are supported by concordant observations fromthe population-based Third National Health and Nutrition Examination Survey,³⁰ which was excluded from our pooled analysis becausediabetic retinopathy was ascertained based on a single nonmydriatic imageof the posterior pole. However, the prevalence of VTDR (not described in theThird National Health and Nutrition Examination Survey report) was similarbetween racial/ethnic groups. Further research using a nationally representativesample would be useful to clarify whether the risk of VTDR among persons withDM varies by race or ethnicity. Our study did not have appropriate data toevaluate the prevalence of diabetic retinopathy in other racial/ethnic groups,such as Asians, who represent a fast-growing segment of the US population.

From the general population perspective, however, large differencesin the prevalence of diabetic retinopathy between racial/ethnic groups exist,driven by racial/ethnic differences in the underlying prevalence of DM. Hispanicpersons had the highest prevalence of DR—1.4- to 1.6-fold higher oddsthan in white persons, and about 1.5-fold higher than in black persons—andalso had higher rates of VTDR than did white persons. No significant differencein the general population prevalence of VTDR between Hispanic and black personswas observed. The odds of diabetic retinopathy among black persons, both DRand VTDR, was about 1.3-fold higher than among white persons, but this differencewas only statistically significant for VTDR.

Although these results are based on the best available data, our estimatesmust be interpreted with caution for several reasons. Our results attemptto estimate the situation based on the status of a limited number of personswith DM who were identified in population-based samples from specific regions,4440 people in all (including only 615 black subjects from a single Caribbeanlocation). Because differences between geographic locations in diet, use ofmedical care, and socioeconomic factors might affect the prevalence of diabeticretinopathy, a more robust approach would have been to use a larger, nationallyrepresentative sample; but appropriate data from such a group are unavailable.Hispanic rates are based on results from predominantly Mexican American populationsin Arizona, Colorado, and Texas, whereas other Hispanic subgroups might differin their prevalence rates. Because the estimates of the prevalence of DR arebased on a smaller sample than that for the other diseases the Eye DiseasesPrevalence Research Group addressed, confidence intervals are correspondinglywider.

It is known that a substantial number of persons with DM are unawarethat they have DM, and that such persons may have diabetic retinopathy.¹⁰ Our pooled analysis was constrained to use self-reportedDM to obtain DM prevalence values because evidence suggests that the prevalenceof DM is rising over time,²² and no current,nationally representative examination survey results will be available untilthe next National Health and Nutrition Examination Survey report. Becauseof this limitation, it is best to interpret our general population–levelresults as reflecting the prevalence of diabetic retinopathy among persons40 years and older who think they have DM. It is likely that additional persons,unaware of that they have DM, have diabetic retinopathy as well. In the BeaverDam Eye Study, Beaver Dam, Wis,¹⁰ The RanchoBernardo Study, Rancho Bernardo, Calif,³¹ andProyecto VER study,¹³ the proportion of personswith previously undiagnosed DM who had diabetic retinopathy was 10.2%, 0.6%,and about 20%, respectively. Retinopathy meeting our VTDR definition was presentin 2%⁹ and 0%,³¹ respectively,of the Beaver Dam and Rancho Bernardo subjects, while in Proyecto VER, about2%¹³ had proliferative diabetic retinopathyor macular edema. These results suggest that persons not known to have DMhave a low frequency of advanced, vision-threatening retinopathy.³² Therefore, our results may underestimate the overallprevalence of diabetic retinopathy in the general US population in this agegroup to an appreciable degree, but probably only slightly underestimate theprevalence of VTDR.

Another limitation of our study is the absence of data on the type ofDM, the duration of DM, and on the degree of hyperglycemia and hypertensioneach subject experienced. We were unable to include these aspects in the pooledanalysis because this information was not uniformly collected by all the participatingstudies in a way that could be combined. Therefore, we were unable to estimatethe prevalence of diabetic retinopathy conditional on values of these criticalvariables, information that would be useful in clinical practice. However,such information is available from the primary WESDR reports,^{2,9,28,33- 34} theSan Luis Valley Diabetes Study,¹⁵ the San AntonioHeart Study,³⁵ the Third National Health andNutrition Examination Survey,^30,36 andother sources.

The lack of such information also makes it difficult to interpret thehigher prevalence rates for diabetic retinopathy reported by the WESDR withrespect to the other contributing studies. It seems likely, but cannot beproven, that the differences arise from poorer glycemic control, longer durationof DM, differences in blood pressure status, and/or an excess of other riskfactors for diabetic retinopathy among WESDR subjects with respect to subjectsin more recent studies, resulting in higher prevalence values. Supportingthis theory is clinical experience suggesting that the primary care of personswith DM has improved over time since the early 1980s, when the WESDR was conducted,with the incorporation of recommendations based on the observations of theDiabetes Control and Complications Trial,^3,37 theUnited Kingdom Prospective Diabetes Study,^4- 5 andother studies. Improved control of blood glucose levels, blood pressure, andserum lipid levels is likely to reduce the incidence, rate of progression,and/or severity of diabetic retinopathy. On these grounds, an argument canbe made that the WESDR results may be less generalizable to the US populationin 2000 than the results of more recent studies. However, it seems surprisingthat a large change in the population risk of diabetic retinopathy would haveoccurred in the span of the few years that elapsed between the WESDR (1980-1982)and the Beaver Dam Eye Study (1988-1990), conducted in the same region. Becausethe WESDR was the most comprehensive study of diabetic retinopathy conductedto date, we have elected to include its data in developing the prevalenceestimates reported herein. However, we also have provided the overall valuesfor the estimated prevalence of DR and of VTDR that would have been obtainedhad the WESDR data been excluded, which are lower. Under either scenario,the estimated prevalence rates for DR and VTDR are high from a populationperspective.

Unknown future secular trends in the risk of diabetic retinopathy limitour ability to make future diabetic retinopathy prevalence projections. Forinstance, improvements in the effectiveness of primary DM care over time couldreduce the incidence of diabetic retinopathy and its rate of progression (asthe preceding sentence states). However, improved DM care also could resultin improved survival. More time at risk of diabetic retinopathy could potentiallybalance the effects of a lower risk per unit time on prevalence rates. Itis impossible to predict reliably what effects secular trends occurring betweennow and 2020 will have on the prevalence of DR and VTDR. Therefore, our projectionsshould be interpreted with an especially high degree of caution.

A strength of our study is that determinations of the outcomes for allstudies were made using gold standard, highly reproducible techniques anchoredby standard photograph reference points. The outcomes can be compared usingthe ETDRS scale,^7- 8 except forthe Barbados Eye Study, which had slightly different definitions as describedearlier. However, the number of standard fields photographed was not identicalbetween studies, which could have contributed to slightly lower prevalenceestimates from the studies imaging fewer fields. It has been demonstratedthat 2-, 3-, and 4-field protocols have reasonable agreement with standard7-field imaging using the Diabetic Retinopathy Study protocol,³⁸ particularlywhen the categorization of retinopathy was collapsed into fewer categories(85%, 93%, and 95% agreement with 7-field results, respectively)³⁹ ina manner similar to that used in our analysis. Therefore, underascertainmentof diabetic retinopathy in studies using fewer than 7 fields is likely tohave been on the order of 5% to 15%. Inspection of the crude prevalence ratesfor the various studies (Table 2)demonstrates that the studies using larger numbers of photographic fieldsdid not consistently have higher prevalence rates, suggesting that the degreeof underascertainment of DR resulting from imaging fewer fields was probablysmall compared with other factors influencing prevalence rates. Nevertheless,underascertainment of diabetic retinopathy because of a small number of photographicfields is another factor that may have made our estimates of the prevalenceof diabetic retinopathy among persons with DM artificially low.

In this study, we have opted to report actual results of the pooledanalysis, rather than attempting to impute the effects of the various studylimitations on prevalence rates. Although these problems impose limitationson the reliability of our estimates, no better methods for estimation of theburden of diabetic retinopathy in the US population are available. Our resultsprovide an estimate as to what that burden of diabetic retinopathy may be,which should be useful for decisions regarding health policy and researchpriorities. Even if the true burden of diabetic retinopathy were considerablylower than we have estimated, diabetic retinopathy still would be a majorpublic health problem in the United States. Because most of the study limitationswould be expected to lead to underestimation of the prevalence of DR, it ismore likely that our estimates are too low rather than too high.

Unlike other age-related eye diseases, diabetic retinopathy often causesblindness during the working-age years, resulting in a larger number of person-yearsof vision lost per case, more disability during the working years per case,and correspondingly large economic costs.^32,40- 41 Inaddition, most vision loss due to diabetic retinopathy is avoidable, throughprimary prevention (intensive control of hyperglycemia,^3,37,42 hypertension,⁵ and of other risk factors for diabetic retinopathy),and secondary prevention (detecting high-risk diabetic retinopathy in timeto apply palliative laser therapy^17,43- 44).Because diabetic retinopathy is often asymptomatic during the period in whichlaser photocoagulation should be applied, screening of asymptomatic personsis needed to minimize the risk of vision loss.¹ Althoughit is recognized that the benefits of screening and treatment outweigh thecosts, even from a purely financial perspective,^40- 41,45 manypersons with DM do not presently receive such management.⁴⁶ Itis our hope that this description of the extent of the problem of diabeticretinopathy in the United States will stimulate further efforts to preventblindness from this disease.

The prevalence of diabetic retinopathy in the United States is high.An estimated 4.1 million persons age 40 and older in the US general populationhave diabetic retinopathy, 1 in 29 persons. An estimated 899 000 personsin this age range have vision-threatening diabetic retinopathy, 1 in 132 persons.The prevalence of diabetic retinopathy is expected to increase substantiallyby 2020, driven by an increasing prevalence of DM over time with the agingof the US population, in combination with anticipated increases in the age-specificprevalence of DM. Because diabetic retinopathy is a substantial public healthproblem, public and private policy efforts directed toward improving primaryand secondary prevention programs are warranted.

Corresponding author: John H. Kempen, MD, PhD, 550 N Broadway, Suite700, Baltimore, MD 21205 (e-mail: jkempen@jhmi.edu).

Submitted for publication May 8, 2003; final revision received December8, 2003; accepted December 8, 2003.

From the Department of Ophthalmology, The Johns Hopkins University Schoolof Medicine, Baltimore, Md (Drs Kempen, West, Congdon, and Friedman); Departmentsof Epidemiology (Drs Kempen and West) and International Health, The JohnsHopkins University Bloomberg School of Public Health, Baltimore (Drs Friedmanand Congdon); Department of Epidemiology and Biostatistics, School of PublicHealth and Health Services, George Washington University Medical Center, Washington,DC (Ms O'Colmain); Macro International, Inc, Calverton, Md (Ms O'Colmain);Department of Preventive Medicine, School of Medicine, Stony Brook University,Stony Brook, NY (Dr Leske); Department of Medicine, University of Texas HealthSciences Center at San Antonio (Dr Haffner); Department of Ophthalmology andVisual Sciences, University of Wisconsin Medical School, Madison (Dr Kleinand Mr Moss); Centre for Eye Research Australia, University of Melbourne,East Melbourne, Victoria (Dr Taylor); Department of Preventive Medicine andBiometrics, University of Colorado School of Medicine, Denver (Dr Hamman);and the Department of Ophthalmology, University of Sydney Centre for VisionResearch, Westmead Hospital, Westmead, New South Wales, Australia (Dr Wang).

This study was supported by contract NO1-EY-8-2108 and grant EY00386(Dr Kempen) from the National Eye Institute, National Institutes of Health,Bethesda, Md.

Data from the 2000 United States Census and the 1999 NHIS are in thepublic domain. The Research Group gratefully acknowledges the work of themany contributors to these studies. The Eye Diseases Prevalence Research Groupacknowledges the valuable input of Lynn Hutt in preparing the manuscript.

The members of the Eye Diseases Prevalence Research Group, DiabeticRetinopathy subsection are as follows:

Participating Studies

The Barbados Eye Studies, Barbados, West Indies: M.Cristina Leske; Suh-Yuh Wu; Barbara Nemesure; Anselm Hennis; Leslie Hyman;Andrew Schachat. Beaver Dam Eye Study, Beaver Dam, Wis: Barbara E. K. Klein; Ronald Klein; Scot E. Moss. Blue Mountains Eye Study, Sydney, New South Wales, Australia: PaulMitchell; Jie Jin Wang; Elena Rochtchina; Wayne Smith; Robert G. Cumming,Karin Attebo; Jai Panchapakesan; Suriya Foran. MelbourneVisual Impairment Project, Melbourne, Australia: Hugh R. Taylor; CathyMcCarty; Bickol Mukesh; LeAnn M. Weih; Patricia M. Livingston; Mylan Van Newkirk;Cara L. Fu; Peter Dimitrov; Matthew Wensor; Yury Stanislavsky. Proyecto Vision Evaluation Research, Ariz: Sheila K. West; Ronald Klein;Jorge Rodriguez (deceased); Beatriz Muñoz; Aimee T. Broman; RobertSnyder; Harry A. Quigley. San Antonio Heart Study, San Antonio,Tex: Steven M. Haffner; Donald Fong; Michael P. Stern; Jacqueline A.Pugh; Helen P. Hazuda; Judith K. Patterson; Wichard A. van Heuven; RonaldKlein. San Luis Valley Diabetes Study, San Luis Valley,Colo: Richard F. Hamman; Elizabeth Mayer-Davis; Julie A. Marshall;Judy Baxter. Wisconsin Epidemiologic Study of Diabetic Retinopathy, Ronald Klein; Barbara E. K. Klein; Scot E. Moss.

Resource Centers

Meta-analysis Coordinating Center, Baltimore, Md: John H. Kempen, Diabetic Retinopathy Subsection; David S. Friedman;Nathan G. Congdon; Benita J. O'Colmain.

National Eye Institute, Bethesda, Md: FrederickL. Ferris III.