Risk Factors for Diabetic Retinopathy: A Populatioii - Diabetes Care

R isk Factors for Diabetic Retinopathy:A Populatioii'Based Studyin Rochester, Minnesota

D. J. BALLARD, MD, L. J. MELTON III, MD, M. S. DWYER, MD, J. C. TRAUTMANN, MD, C.-P. CHU, MS,W. M. O'FALLON, PhD, AND P. J. PALUMBO, MD

Retinopathy is an important sequela of diabetes mellitus, but clinical risk factors for this condition haverarely been assessed in a geographically defined population. In this population-based study, the 1135Rochester, Minnesota, residents with diabetes mellitus initially diagnosed between 1945 and 1969(incidence cohort) were followed through their complete medical records in the community to January1, 1982. Because most of the cases of diabetic retinopathy in Rochester residents developed in patientswith non-insulin-dependent diabetes mellitus (NIDDM), risk factors for diabetic retinopathy were ex-amined in this group (N = 1031). A proportional hazards model identified the following risk factorsfor diabetic retinopathy in NIDDM: elevated initial fasting blood glucose level, marked obesity, andearlier age at onset of diabetes. Stratified analyses indicated that duration of diabetes was also significantlyassociated with an increased risk of retinopathy. Two secular trends, increasing detection of "mild"NIDDM and decreasing risk of diabetic retinopathy, had a major effect on retinopathy risk assessment.These data also suggest that insulin therapy is not an independent risk factor for diabetic retinopathy.DIABETES CARE 1986; 9:334-42.

In an earlier report we1 documented the incidence andestimated the prevalence of retinopathy in a large co-hort of Rochester, Minnesota, residents initially diag-nosed with diabetes mellitus in the period 1945-1969.

We noted that the incidence rates for any diabetic retinopathyand for proliferative retinopathy alone were greater amongpatients with insulin-dependent diabetes mellitus (IDDM)than with non-insulin-dependent diabetes mellitus (NIDDM),although the latter group accounted for a larger actual numberof cases of diabetic retinopathy. The influence of other factorson the risk of retinopathy was not addressed. The objectiveof our investigation was to examine in a population-basedincidence cohort of diabetic individuals the relationship be-tween various clinical characteristics at the time of initialdiagnosis of diabetes and the risk of subsequent diabetic ret-inopathy.

METHODS

The population of Rochester, Minnesota, is well suited forinvestigations into the natural history of diabetes mellitusbecause comprehensive unit medical records for the residentsare available and because these records are accessible through

a centralized index of diagnoses made by essentially all med-ical care providers utilized by the local population. This indexincludes the diagnoses made among outpatients seen in clinicand office consultations, emergency room visits, house calls,and nursing home care, as well as diagnoses recorded amonghospital inpatients and at death. The potential of this datasystem for population-based studies has been described pre-viously.2'3 The original medical records of the patients iden-tified through the index were retrieved and reviewed for aninitial diagnosis of diabetes in the 25-yr period 1945-1969.The diagnostic criteria used for diabetes mellitus have beenreported in detail previously4 and required fasting hypergly-cemia on two deteminations >120 mg/dl (Folin-Wu method)for 1945-1958 or > 110 mg/dl (autoanalyzer ferrocyanide re-ductase technique) for 1959-1969 on whole venous blood.Blood glucose values from other institutions where Rochesterpatients sought medical care were interpreted in light of themethod being employed. An oral glucose tolerance test wasgenerally carried out when an equivocal fasting or postpran-dial blood glucose determination was obtained. At the MayoClinic the oral glucose tolerance test was performed by ad-ministering 1 g glucose/k body wt and determining bloodglucose concentrations at 0, 1, 2, and 3 h after the loading

334 DIABETES CARE, VOL. 9 NO. 4, JULY-AUGUST 1986

Dow

nloaded from http://diabetesjournals.org/care/article-pdf/9/4/334/497783/9-4-334.pdf by guest on 15 February 2022

RISK FACTORS FOR DIABETIC RETINOPATHY/D. J. BALLARD AND ASSOCIATES

TABLE 1Clinical characteristics of Rochester, Minnesota, residents with* and with-outt retinopathy at time of diagnosis of NIDDM, 1945-1969

Characteristic

Retinopathyat diagnosis

% N

Noretinopathyat diagnosis

Yo N

SexMen 37 10Women 63 17

Age (yr)<30 0 030-49 11 350-59 11 360-69 33 9>70 44 12

Fasting blood glucose (mg/dl)>300 8 2200-299 41 11<200 52 14

Glycosuria 89 24Relative weight

<1.00 15 41.00-1.19 19 51.20-1.39 33 9si.40 33 9

Macrovascular disease 37 10Persistent proteinuria 4 1Hypertension 59 16Smoker 15 4Initial therapy

Insulin 37 10Oral agents 30 8Diet alone 33 9

50

50

318252826

9256678

501503

26180253282263

85253666785

92833302684530

141473

8628333530025783453302

139136729

gN = 27.tN = 1004.

dose. Only the 1- and 2-h values were used for interpretationof the oral glucose tolerance test, and both values had to beelevated in comparison to age-specific standards4 for the di-agnosis of diabetes mellitus to be made. For the other insti-tutions providing data for this study, glucose tolerance testswere interpreted by established standards for the method used.These criteria resulted in an incidence cohort of 1135 Roch-ester residents newly diagnosed with diabetes between 1945and 1969. Although the diagnostic criteria used here weresomewhat more inclusive than those proposed by the NationalDiabetes Data Group,5 we have shown that the differenceshave little practical effect on the resulting clinical spectrumof diabetes or on the risk of microvascular complications.6

Each of the 1135 incidence cases was followed through thelinked medical records until death, emigration from the com-munity, or through December 31, 1981. A diagnosis of di-abetic retinopathy was based on eye examinations made byMayo Clinic staff ophthalmologists throughout each patient'sclinical history. The retinopathy recorded was divided into

nonproliferative, maculopathy, or proliferative types. Non-proliferative retinopathy included microaneurysms, dot andflame hemorrhages, and soft and hard exudates. Maculopathywas defined as nonproliferative retinopathy causing a markedloss of visual acuity due to edema of the macula and wasrecorded only if specifically diagnosed. Proliferative retinop-athy included neovascularization of the disc and elsewhere,and vitreous hemorrhage believed due to diabetic neovas-cularization. Most eye examinations included a specific di-agnosis of the type of retinopathy, if present, that was found.In instances where no actual diagnosis was made by the ex-

TABLE 2Incidence density of retinopathy by clinical characteristics of Rochester,Minnesota, residents at time of diagnosis of NIDDM, 1945-1969, followedfor retinopathy to 1982

Anyretinopathy

Characteristic Rate* N

Proliferativeretinopathy

Rate* N

SexMen 15.6 83 0.8 5Women 15.6 87 1.4 9

Age (yr)<30 18.4 7 0 030-49 17.2 47 1.3 450-59 22.0 68 2.2 860-69 10.3 30 0.3 1>70 10.2 18 0.5 1

Fasting blood glucose (mg/dl)>300 32.9 22 2.4 2200-299 33.5 78 3.3 10<200 8.9 70 0.2 2

GlycosuriaYes 15.1 131 1.4 13No 16.1 39 0.4 1

Relative weight<1.00 14.1 10 1.3 11.00-1.19 15.4 45 0.9 31.20-1.39 13.3 51 0.5 2>1.40 18.6 64 2.0 8

Macrovascular diseaseYes 11.7 20 0 0No 16.4 150 1.3 14

Persistent proteinuriaYes 7.7 5 0 0No 16.1 165 1.2 14

HypertensionYes 14.7 63 1.3 6No 16.2 107 1.1 8

SmokerYes 15.1 55 0.7 3No 15.9 115 1.3 11

Initial therapyInsulin 42.6 52 1.2 2Oral agents 19.1 23 2.2 3Diet alone 11.2 95 1.0 9

' Incidence per 1000 person-years.

DIABETES CARE, VOL. 9 NO. 4, JULY-AUGUST 1986 335

Dow



TABLE 3Incidence density of any retinopathy by age at diagnosis of retinopathy and duration of diabetes for Rochester, Minnesota, residents with NIDDM, 1945—1969, followed for retinopathy to 1982

Follow-upperiod (yr)

0-45-9

10-1415-1920 +

TotalATLTt

<30

Rate*

019.9000

6.0

p-yt

99501800

167

30-49

Rate'

6.911.728.933.6

146.8

13.3

p-yt

728426208597

1428X2. = 8.99,

P < .01

Age group (yr)

50-59

Rate*

10.815.529.739.759.6

17.8X 2 . =

P =

p-yt

102064337012634

21933.33,

= .07

60-69

Rate*

8.923.219.531.165.7

18.4X 2 . =

P =

p-yt

123894656422576

30499.72,.01

Rate*

4.110.920.027.623.0

13.6

X2,P

70 +

p-yt

12191193952508174

4046= 1.20,= .25

Rate

7.415.622.230.541.3

15.6

x2.

Total

p-yt

430432592112918290

10,883= 12.86,

P < .01

'Incidence per 1000 person-years (p-y).tPerson-years of observation.tArmitage test for linear trend.

amining ophthalmologist, a diagnosis was inferred, if appro-priate, according to the above criteria. The data of diagnosisof retinopathy was established when the first eye was detectedto have retinopathy, even though the other eye might nothave had retinopathy at that time.

The fasting blood glucose value used in this report was thefirst for each patient that met the diagnostic criteria. We usedthe method recommended by West7 for standardization ofblood glucose values measured by the Folin-Wu and autoan-alyzer ferrocyanide reductase techniques to the glucose oxi-dase method, which has been employed at the Mayo Clinicsince 1972, and all values were expressed in terms of thelatter. Relative weight at diagnosis of diabetes was calculatedwith recommended height-weight tables.5 Therapeutic regi-mens were classified as insulin (with or without other ther-apy), oral agent (with or without diet but without insulin),or diet alone (no insulin or oral agents), as of the time ofdismissal after the initial diagnosis and work-up. The char-acterization of specific clinical type of diabetes generally fol-lowed National Diabetes Data Group recommendations, al-though, as explained in detail in a separate report, somemodifications were required in the context of a retrospectivestudy with extant medical records.8 The risk factor analysesdescribed in this article pertain only to the patients withNIDDM (N = 1031). Insulin-dependent (N = 75) and sec-ondary (N = 29) diabetes incidence cases were excluded fromthis report because of small numbers and consequent inade-quate statistical power. Diabetic complications were classifiedat time of diagnosis of diabetes as a history of macrovasculardisease (angina pectoris, myocardial infarction, stroke, tran-sient ischemic attack, or peripheral vascular disease) or mi-crovascular disease other than retinopathy (persistent pro-teinuria). A diagnosis of persistent proteinuria was based ontwo consecutive random urine samples with at least grade 1protein. The determination of glycosuria was based on thedetection of glucose in a urine sample at diagnosis of diabetes.

A diagnosis of hypertension was made if the subject had twoconsecutive blood pressure readings > 160/95 mmHg or wason antihypertensive drug therapy at diagnosis of diabetes.Patients were also classified as smokers or nonsmokers as ofthe date of diagnosis of diabetes.

The cumulative incidence of retinopathy was calculatedwith the Kaplan-Meier product limit method.9 Cumulativeincidence curves were compared with the Peto-Peto Wil-coxon test statistic.10 Retinopathy incidence density was de-termined by person-years analysis." The statistical signifi-cance of differences in cumulative incidence and in incidencerates was assessed with the method of Fleiss12 and with theArmitage Test for Linear Trend.13 The relative influence ofvarious clinical characteristics on the risk of subsequent ret'inopathy was evaluated with proportional hazards models,1415

with x2-tests for the coefficients of the proportional hazardsmodels and for the model likelihood ratio.

RESULTS

Clinical characteristics at diagnosis of diabetes. The clinical char-acteristics at the time of initial diagnosis of NIDDM are shownin Table 1. Only 27 of 1031 (2.6%) NIDDM patients hadretinopathy diagnosed before or within 30 days of initial di-agnosis of diabetes. Although the low prevalence of diabeticretinopathy at diagnosis of diabetes precludes a rigorous com-parison, NIDDM patients with retinopathy at diagnosis ofdiabetes were: more likely to be female (P = .19), older(P = .13), more like to have higher fasting blood glucoselevels (P = .19), more likely to have glycosuria (P = .19),more likely to have macrovascular disease (P = . 18) or hy-pertension (P = .16), less likely to be smokers (P = .22),and more likely to be placed on insulin as initial therapy(P < .01).

Univariate analyses. To determine the influence of thesevarious factors on the risk of subsequent retinopathy, we first


Dow



TABLE 4Cumulative incidence (%) of retinopathy at various periods of follow-up by clinical characteristics of Rochester, Minnesota, residents at time of diagnosisof NIDDM, 1945-1969, followed for retinopathy to 1982

Characteristic

SexMenWomen

Age (yr)<3030-4950-5960-692=70

Fasting blood glucose (mg/dl)>300200-299<199

GlycosuriaYesNo

Relative weight<1.001.00-1.191.20-1.39>1.40

Macrovascular diseaseYesNo

Persistent proteinuriaYesNo

HypertensionYesNo

SmokerYesNo

Initial therapyInsulinOral agentsDiet alone

Total

5

4.23.1

02.85.83.32.7

7.36.92.0

3.25.2

3.04.23.33.6

2.83.9

1.53.8

3.43.9

4.33.3

8.44.52.7

3.6

Any retinopathy

10

9.412.4

4.08.2

18.36.6

10.1

29.922.25.2

11.110.4

7.610.69.9

13.0

11.410.9

6.311.3

12.210.1

9.611.6

28.39.78.2

11.0

15

18.522.0

23.019.626.216.815.4

48.142.210.5

21.418.1

23.517.517.226.4

17.420.9

9.521.1

19.620.9

16.122.9

52.533.513.3

20.6

Duration of follow-up (yr)

20

31.832.0

•30.741.521.5

•

•58.320.9

32.131.0

•35.927.332.9

•32.6

22.422.4

31.232.2

29.333.2

68.1•

23.6

31.9

5

0.20.2

000.400.6

1.500.2

0.30

00.400.4

00.3

00.3

0.30.2

00.3

0.90.90

0.2

Proliferative retinopathy

10

0.20.5

000.40.50.6

1.50.50.2

0.50

00.400.8

00.4

00.4

0.60.2

00.5

0.91.90

0.4

15

0.61.0

0.801.00.50.6

1.52.30.2

1.00

01.101.6

00.9

00.8

1.80.2

0.50.9

0.91.90.6

0.8

20

2.14.2

1.71.77.20.50.6

9.78.50.9

4.10

7.71.11.36.4

03.5

03.4

3.63.0

1.54.0

3.87.12.6

3.2

< 10.

evaluated each characteristic individually. The incidence ofsubsequent retinopathy in each clinical group is shown inTable 2. Factors significantly associated with an increased risk(P < .01) of any retinopathy among patients with NIDDMwho were free of retinopathy at diagnosis (N = 1004) wereas follows: younger age at diagnosis of diabetes mellitus, higherfasting blood glucose levels at diagnosis, and initial insulintherapy. Glycosuria, relative weight, macrovascular disease,persistent proteinuria, hypertension, and smoking were notsignificantly associated with the development of diabetic ret-

inopathy in the initial univariate analysis. Because none hadproliferative retinopathy at initial diagnosis of diabetes, all1031 NIDDM patients were at risk for this complication.Essentially the same qualitative results were observed regard-ing univariate predictors for the development of proliferativeretinopathy among Rochester residents with NIDDM (Table2). With the small numbers involved, the only findings thatwere statistically significant were younger age and initial hy-perglycemia.

The effect of age was also assessed in the context of aging,


Dow



TABLE 5Proportional hazards models for the development of any retinopathy by clinical characteristics of Rochester, Minnesota, residents at time of diagnosis ofNIDDM, 1945-1969, followed for retinopathy to 1982

Characteristic

SexAgeFasting blood glucoseGlycosuriaRelative weightMacrovascular diseasePersistent proteinuriaHypertensionSmokingInitial therapyYear of diagnosis

Model: (x2,, = 89.93), P < .01,

Coefficient

0.00320.00910.0057

-0.03100.0882

-0.1226-0.6141

0.0287-0.1048

1.3257-0.0506

r = 0.176.

Univarjate

x2

0.002.59

78.820.031.060.261.830.030l41

63.0716.65

P value

.98

.11<.01

.87

.30

.61

.18

.86

.52<.01<.01

Coefficient

0.1559-0.0098

0.0045-0.0698

0.18520.0987

-0.59750.1606

-0.01000.7268

-0.0260

Full multivariate

x2

0.872.41

29.220.144.040.151.670.810.00

11.533.94

P value

.30

.12<.01

.71

.04

.70

.20

.37

.96<.01

.05

where the effect of duration of diabetes also had to be takeninto account (Table 3). The incidence of any retinopathyincreased with duration of follow-up within each age groupexcept those <30 yr old for whom there were little data.There was no entirely consistent pattern within the duration-specific strata but retinopathy risk generally peaked between50 and 60 yr of age and declined thereafter. Because relativelyfew events were observed, a comparable analysis could notbe completed for proliferative retinopathy.

The cumulative incidence of retinopathy among patientswith various characteristics (Table 4) reflected the findingsdescribed above. Proliferative retinopathy was extremely un-common until 20 yr duration of NIDDM, but the generalassociations were similar to those for nonproliferative reti-nopathy.

Multivariate analyses. A proportional hazards analysis wasused to identify factors at time of diagnosis of diabetes thatmight be predictive of the development of subsequent reti-nopathy. The variables entered as linear main effect termswere: sex (0 = female, 1 = male), age at diagnosis of dia-betes (yr), fasting blood glucose (mg/dl), glycosuria (0 = no,1 = yes), relative weight (0 = <1.00, 1 = 1.00-1.19,2 = 1.20-1.39,3 = ^1.40), macrovascular disease (0 = no,1 = yes), persistent proteinuria (0 = no, 1 = yes), hyper-tension (0 = no, 1 = yes), smoking (0 = no, 1 = yes),initial therapy (0 = oral agents or diet, 1 = insulin), andyear of diagnosis of diabetes. Because the Kaplan-Meier uni-variate analysis suggested a nonlinear parabolic relationshipbetween age and retinopathy risk, we also examined age + age2

in the models and found no appreciable change in the coef-ficients for the other variables in the model and no changein the model R statistic.15 Separate consideration of oral agentsand diet in the form of dummy variables11 indicated that itwas appropriate to combine the two therapy categories toform a dichotomous variable and simplify the presentation ofthe proportional hazards analysis. In addition to presenting

the univariate and full multivariate models (Table 5), themost parsimonious multivariate model (based on considera-tion of validity,u precision," and model R statistic15) is shownin Table 6 with the corresponding hazard ratios and 95%confidence intervals. Based on the proportional hazards modelanalysis alone, fasting blood glucose level was the most im-portant independent risk factor for the development of dia-betic retinopathy among patients with NIDDM. Initial ther-apy, relative weight, year of diagnosis, and age also appearedto be important independent risk factors.

From previous reports concerning the study population,8

we know that an increasing proportion of clinically recognizeddiabetes over this time period (1945-1969) was comprised ofmild cases. We therefore used stratified analysis to explorethe relationships between fasting blood glucose level, initial

TABLE 6Proportional hazards model for the development of any retinopathy by clin-ical characteristics of Rochester, Minnesota, residents at time of diagnosisof NIDDM, 1945-1969, followed for retinopathy to 1982

Characteristic

AgeFasting blood glucoseRelative weightInitial therapyYear of diagnosis

Model: (x25 = 81.85),

Reduced multivariate

Coefficient

-0.00930.00400.17430.4312

-0.0288

P < .01, r

X2 F

2.3429.134.07

11.863.74

= 0.187.

' value

.13<.01

.04<.01

.05

ratio

1,2,2,2,1

.14*

.451

.01*

.06§•46||

95%

interval

0.96, 1.351.76, 3.421.00, 3.281.35,3.131.00, 2.14

'Difference of 15 yr (e.g., 50 vs. 65 yr).tDifference of 200 mg/dl (e.g., 350 vs. 150 mg/dl).tTwo-level difference (e.g., 1.00-1.19 vs. >1.40).§ Insulin vs. noninsulin.HDifference of 15 yr (e.g., 1950 vs. 1965).


Dow



TABLE 7Ten-year cumulative incidence (C.I.) of any retinopathy by year of diagnosis, initial therapy, and fasting blood glucose for Rochester, Minnesota, residentswith NIDDM diagnosed 1945-1969 and followed for retinopathy to 1982

Year ofdiagnosis

of diabetes

1945-19541955-19641965-1969

Total

Insulin

C.I.

042.9

22.1

N*

880

16

<200

Oral agents

C.I. 1ST

00 183.2 35

2.3 53

Fasting blood glucose

C.I

8.54.45.2

5.0

Diet

N*

64349184

597

C.

37257

28

(mg/dl)

insulin

I.

.5

.1

.1

9

N*

555414

123

>200

Oral agents

C.I. N '

019.0 4210.1 41

14.8 83

Diet

C.I.

38.219.510.5

24.0

N*

406725

132

C.

33277

28

Insulin

I.

.1

.9

.1

.3

N*

636214

139

Total

Oral agents

C.I.

13.56.6

9.7

N'

06076

136

Diet

C.I.

14.36.65.8

8.2

N*

104416209

729

'Number of patients with diagnosis of NIDDM at beginning of 10-yr interval.

therapy, year of diagnosis, and retinopathy risk in more detail,as shown in Table 7. Over the three secular cohorts, thepercentage of patients diagnosed with NIDDM having aninitial fasting blood glucose level <200 mg/dl increased from43 to 70 to 73%. This level (<200 mg/dl) consistently defineda group with a low risk of retinopathy across the three timeperiods. The 10-yr cumulative incidence of retinopathy wasapproximately six times greater in the ^200-mg/dl groupthan in the <200-mg/dl group for patients on oral agentsand five times greater for patients on dietary therapy alone.Because there were only 16 patients with an initial fastingblood glucose level <200 mg/dl who were initially placed oninsulin, the above comparison was not possible for patientstreated with insulin. Although Table 7 demonstrates the pre-dominant effect of initial fasting blood glucose level on ret-inopathy risk, a marked secular trend of decreasing retinop-athy risk is also apparent. As shown in Table 7, the seculartrend in retinopathy risk was present in all three initial treat-ment groups with initial fasting blood glucose levels ^200mg/dl. After stratification by year of diagnosis and bloodglucose level (^200 mg/dl), retinopathy risk did not vary byinitial treatment assignment. Thus, the detailed stratified

analysis strongly suggests that insulin is not an independentrisk factor for retinopathy.

Figure 1 demonstrates that the secular trend in retinopathyrisk did not begin until after 5 yr duration of diabetes andbecame more prominent with increasing duration of NIDDM.These curves were significantly different (P = .04) with thePeto-Peto Wilcoxon test.10

DISCUSSION

On a population basis and from a public healthperspective, the relative contribution of NIDDMto the problem of diabetic retinopathy is impres-sive. Although the incidence rates of any reti-

nopathy and of proliferative retinopathy alone were three andsix times greater, respectively, in IDDM, 82% of cases of allretinopathy and 64% of cases of proliferative retinopathy inRochester diabetic patients developed in NIDDM diabeticpatients.' This confirms the importance of ophthalmologicsurveillance in NIDDM that was recently demonstrated bythe Wisconsin diabetic retinopathy prevalence studies.1617

Klein et al.1617 found that the prevalence of proliferative

FJG. 1. Cumulative incidence of any retinopathyby duration of diabetes mellitus and year of diag-nosis ofMDDMfor J 945-1969 Rochester, Min-nesota, residents with initial fasting blood glucose^200 mg/dl. (Curves terminate athl < 10.)

70

60

50

Cumulative 4oincidence

. 30

20

10

o

Year of DX DM1945-1954

1955-1964

1965-1969I I I l I l I l l l l l l I I I I

6 8 10 12 14 16

Years after DX DM18 20


Dow



retinopathy was more than three times greater in younger-onset than in older-onset diabetic individuals (22 vs. 6%)but that 58% (326/566) of the cases of proliferative retinop-athy in that population-based prevalence study were foundin older-onset diabetic patients. Our incidence data indicatethat the relative contribution of NIDDM to diabetic reti-nopathy is even greater than has been suggested by the Wis-consin studies. This is due in part to the influence of selectivesurvival (impaired survival of NIDDM patients with diabeticretinopathy18).

In this population-based study of risk factors for diabeticretinopathy in NIDDM we have identified a subset of patientswho are at significantly increased risk. They are markedlyobese NIDDM individuals who are <60 yr of age at diagnosisof diabetes and who have initial fasting blood glucose levels>200 mg/dl. The retinopathy risk for these diabetic patientsincreased with duration of disease across all age groups.

The only other population-based incidence study of reti-nopathy in individuals with NIDDM was conducted in theGila River Indian Community of Arizona. 19l2° The incidenceof retinopathy in the Pima Indians was associated with dia-betes duration, elevated blood pressure between diagnosis ofdiabetes and the 6-yr follow-up examination in individualswho were not taking antihypertensive medications, and thefollowing factors at the initial examination: insulin treatment,plasma glucose concentration, peripheral neuropathy, pro-teinuria, older age, and lower body mass index. The signif-icance of the disparity in the findings of the Pima Indianstudy compared with the Rochester study is unclear given thevastly different ethnic and sociodemographic characteristicsof the two communities. Moreover, many of the Pima Indiansactually had a previous diagnosis of diabetes at initial studyexamination (i.e., they were prevalence cases, not incidencecases).

In a setting similar to Rochester, the recent population-based prevalence study of retinopathy risk factors in southernWisconsin patients ^30 yr of age at diagnosis of diabetesindicated that the prevalence of retinopathy was related tolonger duration, younger age at diagnosis, higher levels ofglycosylated hemoglobin, higher systolic blood pressure, pres-ence of proteinuria, lower body mass index, and use of in-sulin.17 We have observed similar effects for duration and ageat diagnosis of diabetes. Although we do not have glycosylatedhemoglobin levels at diagnosis of diabetes for our 1945-1969incidence cohort, our findings with respect to fasting bloodglucose level are in agreement with the glycosylated hemo-globin level results of Klein et al.17

We found a nonsignificant (P = .37) relationship betweenthe presence of hypertension and retinopathy risk in theRochester incidence study. The study of Klein et al.17 ex-amined factors at the time of diagnosis of retinopathy, whereaswe were concerned with factors at the time of diagnosis ofdiabetes that might be associated with subsequent retinopa-thy. Thus, the influence of antihypertensive therapy couldaccount for the disparity with respect to the relationship be-tween hypertension and retinopathy in our incidence studycompared with the Wisconsin prevalence study. Evidence to

support this notion is that the decline in the incidence ofstroke in Rochester appears attributable in part to improvedblood pressure control due to antihypertensive medication.21

Retinopathy appears to predict proteinuria in our protein-uria risk factor study.22 A prevalence study cannot determinewhich is the antecedent factor in this relationship. Furtherresearch is clearly indicated to establish the temporal rela-tionships between various manifestations of microvascular dis-ease in NIDDM.

Our incidence data are also at variance with the prevalencefindings of Klein et al.17 and others23"25 with respect to theassociation between body mass index and retinopathy risk inNIDDM. Obese patients with NIDDM had a higher risk forsubsequent retinopathy in our incidence study. We have doc-umented relatively impaired survival in the obese NIDDMpatient with retinopathy contrasted with nonobese NIDDMpatients with retinopathy,18 which may account for the dif-ferences observed between our incidence findings and theprevious prevalence results.

Perhaps the most important finding of this study concernsthe relationship between insulin use and retinopathy risk.Recent work by Seghieri et al.26 indicates that overweightNIDDM individuals may be at increased risk for developingretinopathy if they are treated with insulin. However, thedata from that referral center prevalence study involving only23 patients must be interpreted with caution. Klein et al.17

also found that the prevalence of retinopathy was associatedwith insulin use. Note that neither of these prevalence studieswas able to definitively address the antecedent-consequentrelationship between retinopathy and any potential risk fac-tors, including insulin therapy.27 In our relatively large pop-ulation-based incidence study we found that retinopathy riskdid not vary by initial treatment assignment after controllingfor the important confounding influence of fasting blood glu-cose level and the secular trend of decreasing retinopathyrisk. This assessment is probably not significantly contami-nated by changes in therapy over time, because over 75% ofpatients remained in the same treatment group (insulin vs.noninsulin) from diagnosis of diabetes to final follow-up. Theretinopathy risk attributable to insulin therapy will not bedefinitively resolved in the absence of prospective randomizedclinical trials involving adequate numbers of patients withlong-term follow-up. However, our data do not support thesuggestion of Seghieri et al.26 that withdrawal or avoidanceof the use of insulin in these patients promotes the preventionof diabetic retinopathy.

We have previously discussed other aspects of the strengthsand weaknesses of the retrospective cohort study design thatwe have used.1 Several factors could have influenced theassociations we observed. It is possible that younger obesepatients with more highly elevated blood glucose levels weresimply more likely to have regular, thorough examinationsfor retinopathy. We do not have information regarding thenumber and type of eye examinations performed for eachpatient. However, indirect evidence indicates that detectionbias of this type was unlikely to have been of sufficient mag-nitude to account for the observed associations.{ A random


Dow



sample of Rochester diabetic subjects without retinopathy hada median of seven examinations by an ophthalmologist afterdiagnosis of diabetes. Eighty-five percent had three or more,and > 9 5 % had at least one such funduscopic examinationby a consultant ophthalmologist in addition to examinationsby their primary care physicians. Additionally, if detectionbias was a major factor in this study, one would expect anincreasing retinopathy risk due to increasing awareness onthe part of ophthalmologists of an elevated risk of retinopathyin diabetic individuals.27 There are, however, many potentialconfounding variables that could influence the secular trendof decreasing retinopathy risk that we cannot assess, such asthe influence of improved blood glucose and blood pressurecontrol and possible, though unlikely, decreasing sensitivityof ophthalmologic examinations. Although it is not possibleto definitively address the actual magnitude of potential de-tection bias in this retrospective cohort study, it seems im-probable that any such bias totally accounted for the rela-tionships that we observed between five factors (fasting bloodglucose level, age at diagnosis, relative weight at diagnosis,year of diagnosis, duration of diabetes) and retinopathy risk.Finally, loss to follow-up and emigration are extremely un-likely to have distorted our findings, because 704 of the 1004NIDDM patients in the cohort were followed until death,whereas the median length of follow-up for the 241 survivorswho remained in the community was 18 yr and for the 59survivors who emigrated from Rochester was 16 yr.

Our findings indicate that major advances in the preven-tion of diabetic retinopathy in the population will have tobe concerned primarily with NIDDM individuals. Patientswho appear to be at greatest risk for this untoward outcomeare those < 6 0 yr of age at diagnosis of diabetes who havefasting blood glucose levels > 2 0 0 mg/dl and who are mark-edly obese. Whereas we cannot recommend a specific timeinterval for screening NIDDM individuals for retinopathy,newly diagnosed patients without the risk factors identifiedin this study have a 10-yr cumulative retinopathy risk of< 1 0 % , compared with 40% for high-risk patients. However,by 20 yr duration, cumulative retinopathy risk is > 2 0 % , evenin the low-risk groups and > 6 0 % in the high-risk groups, andapproaches 40% overall.

As we have discussed, the diabetes literature is comprisedalmost exclusively of prevalence studies, in which antecedentcharacteristics cannot be distinguished from consequent find-ings.27 To more accurately interpret the results of these prev-alence studies, we are now conducting a study of the riskfactors for the prevalence of diabetic retinopathy, which wewill compare with the results from this incidence study.

ACKNOWLEDGMENTS: We thank Kay Hurless for computer pro-gramming and Mary Ramaker for assistance in preparing themanuscript.

This investigation was supported in part by research grantsfrom the American Diabetes Association, the National In-stitutes of Heal th (AM-30582), and the Juvenile DiabetesFoundation International (JDF-#5) .

From the Department of Medical Statistics and Epidemiology(D.J.B., L.J.M., M.S.D., C.-P.C, W.M.O.), Department of Oph-thalmology (J.C.T.), and Division of Endocrinology, Departmentof Internal Medicine (P.J.P.), Mayo Clinic and Mayo Foundation,Rochester, Minnesota 55905.

Send reprint requests to Dr. Melton at the above address.

REFERENCES1 Dwyer, M. S., Melton, L. J., Ill, Ballard, D. J., Palumbo,

P. J., Trautmann, J. C , and Chu, C.-P.: Incidence of diabeticretinopathy and blindness: a population-based study in Rochester,Minnesota. Diabetes Care 1985; 8:316-22.

2 Kurland, L. T., Elveback, L. R., andNobrega, E T.: Populationstudies in Rochester and Olmsted County, Minnesota, 1900-1968.In The Community as an Epidemiologic Laboratory: A Casebook ofCommunity Studies. Kessler, I. I., and Levin, M. L., Eds. Baltimore,The Johns Hopkins Univ. Press, 1970:47-70.

3 Kurland, L. T , and Molgaard, C. A.: The patient record inepidemiology. Sci. Am. 1981; 245:54-63.

4 Palumbo, P. J., Elveback, L. R., Chu, C.-P, Connolly, D. C ,and Kurland, L. T : Diabetes mellitus: incidence prevalence, sur-vivorship, and causes of death in Rochester, Minnesota, 1945-1970.Diabetes 1976; 25:566-73.

5 National Diabetes Data Group: Classification and diagnosis ofdiabetes mellitus and other categories of glucose intolerance. Dia-betes 1979; 28:1039-57.

6 Melton L. J., Ill, Palumbo, P. J., Dwyer, M. W., and Chu,C. -P.: Impact of recent changes in diagnostic criteria on the apparentnatural history of diabetes mellitus. Am. J. Epidemiol. 1983; 117:559-65.

7 West, K. M.: Epidemiology of Diabetes and Its Vascular Lesions.New York, Elsevier/North-Holland, 1978:403-28.

8 Melton, L. J., Ill, Palumbo, P. J., and Chu, C.-P: Inci-dence of diabetes mellitus by clinical type. Diabetes Care 1983;6:75-86.

9 Kaplan, E. L., and Meier, P.: Nonparametric estimationfrom incomplete observations. J. Am. Stat. Assoc. 1958; 53:457—81.

10 Peto, R., and Peto, J.: Asymptotically efficient rank invarianttest procedures (with discussion). J. R. Stat. Soc. Ser. A 1972;135:185-206.

11 Kleinbaum, D. G., Kupper, L. L , and Morgenstern, H.: Ep-idemiologic research: principles and quantitative methods. Belmont,CA, Lifetime Learning Publications, 1982:96-116.

12 Fleiss, J. L.: Statistical Methods for Rates and Proportions, 2nded. New York, Wiley, 1981.

13 Armitage, P.: Tests for linear trends in proportions and fre-quencies. Biometrics 1955; 11:375-86.

14 Cox, D. R.: Regression models and life-tables (with discussion).J. R. Stat. Soc. Ser. B 1972; 34:187-202.

15Harrell, F. E.: The PHGLM Procedure. SAS SupplementalLibrary. Cary, NC, SAS Inst., 1983:267-94-

16 Klein, R., Klein, B. E. K., Moss, S. E., Davis, M. D., andDeMets, D. L.: The Wisconsin epidemiology study of diabetic ret-inopathy. II. Prevalence and risk of diabetic retinopathy when ageat diagnosis is less than 30 years. Arch. Ophthalmol. 1984; 102:520—26.

17 Klein, R., Klein, B. E. K., Moss, S. E., Davis, M. D., andDeMets, D. L.: The Wisconsin epidemiologic study of diabetic ret-inopathy. III. Prevalence and risk of diabetic retinopathy when ageat diagnosis is 30 or more years. Arch. Ophthalmol. 1984; 102:527—32.


Dow



18 Ballard, D. J., and Melton, L. J.: Sources of disparity in in-cidence and prevalence studies of diabetic retinopathy: the influenceof selective survival on risk factor assessment. Letter to the Editor.Diabetes Care 1986; 9:313-15.

19 Dorf, A., Ballintine, E. J., Bennett, P. H., and Miller, M.:Retinopathy in Pima Indians: relationships to glucose level, durationof diabetes, age at diagnosis of diabetes, and age at examination ina population with a high prevalence of diabetes mellitus. Diabetes1976; 25:554-60.

20 Knowler, W. C , Bennett, P. H., and Ballintine, E. J.: Increasedincidence of retinopathy in diabetics with elevated blood pressure:a six-year follow-up study in Pima Indians. N. Engl. J. Med. 1980;302:645-50.

21 Garraway, W. M., Whisnant, J. P., and Drury, I.: The con-tinuing decline in the incidence of stroke. Mayo Clin. Proc. 1983;58:520-23.

22 Ballard, D. J., Palumbo, P. J., Melton, L. J., Frohnert, P. P.,and O'Fallon, W. M.: The natural history of renal disease in type

II diabetes mellitus: a population-based study in Rochester, Min-nesota. Abstract 17. Diabetes 1985; 34 (Suppl. 1):5A.

23 West, K. M., Erdreich, L. J., and Stober, J. A.: A detailedstudy of risk factors for retinopathy and nephropathy in diabetes.Diabetes 1980; 29:501-508.

24 Kornerup, T.: Studies in diabetic retinopathy: an investigationof 1,000 cases of diabetes. Acta Med. Scand. 1955; 153:81-101.

25 Nilsson, S. E., Nilsson, J. E., Frostberg, N., and Emilsson, T.:The Kristianstad Survey II. Studies in a representative adult diabeticpopulation with special reference to comparison with an adequatecontrol group. Acta Med. Scand. Suppl. 1967; 469:1-42.

26 Seghieri, G., De Giorgio, L. A., Bartolomei, G., Alessandri,G.D., Venturino, G., Gironi, A., and Mammini, P.: Insulin therapyin overweight type 2 (non-insulin-dependent) diabetic patients: anassociation with higher prevalence of retinopathy. Abstract. Dia-betologia 1984; 27:331 A.

27 Sackett, D. L : Bias in analytic research. J. Chron. Dis. 1979;32:51-63.


Dow


Documents

Risk Factors for Diabetic Retinopathy: A Populatioii - Diabetes Care