End-Stage Renal Disease Attributable to Diabetes Mellitus

Methods

This study had a population-based, case–control design [15]. Study procedures were tested in a pilot study and approved by institutional review boards at the Johns Hopkins University and the Health Care Financing Administration.

Patients and Controls

The study population included residents of Maryland, Virginia, West Virginia, and Washington, D.C., aged 20 to 64 years who had working telephones at their homes. Persons who lived in institutions (such as nursing homes, prisons, and military facilities), had been absent from their homes for more than 2 weeks at the time of contact, were mentally incompetent, were physically unable to complete a telephone interview, or did not speak English were excluded.

Patients who started renal replacement therapy between January and July 1991 and who satisfied the criteria cited above were identified by the Mid-Atlantic Renal Coalition, which is part of the Medicare End-Stage Renal Disease Program. Patient identifiers (including telephone numbers) were transmitted to the study investigators. The sample was not restricted to Medicare registrants; it also included patients whose care for end-stage renal disease was being paid for by private insurance policies, Medicaid, and Veterans Affairs benefits. During the study period, 978 potentially eligible patients with end-stage renal disease were identified, of whom 885 (90%) had use of a private telephone line. Of these, 133 (15%) were ineligible for participation in the study: Sixty-five had died, 19 were institutionalized, 14 had moved out of the study area, 8 had renal function that had returned to normal, 8 were too sick to be interviewed, 7 had a hearing problem, 5 did not speak English, 5 were hospitalized for more than 2 weeks, and 2 older than 64 years had been preselected by mistake. Of the 752 eligible patients, 716 (95%) were interviewed, 16 refused to participate, 5 interrupted the interview, and 15 could not be reached. The median delay between the onset of therapy for end-stage renal disease and the interview was 5 months. Patients were further divided in two subgroups: those who had diabetic end-stage renal disease according to the patient's nephrologist (n = 239) and those with other underlying causes of renal failure (n = 477).

Control persons from the general population, frequency-matched with case-patients within 5-year age groups, were identified by random-digit dialing [16]. Of 1311 residences reached during enrollment, 1259 (96%) were screened for eligibility, and 402 were found to have one or more eligible residents. When several persons in the same household were eligible, one was chosen randomly. Reasons for ineligibility in the remaining 857 households were the following: no members within eligible age groups (846 households), English language not spoken (7 persons), hearing problems (3 persons), and end-stage renal disease in respondent (1 person). Of those eligible, 361 (90%) completed the interview.

Data Collection

Trained, certified interviewers contacted potential participants by telephone (persons with end-stage renal disease, but not controls, received an explanatory letter before the telephone call), informed them about the purpose of the study, provided them with a telephone number at which additional information could be obtained, and asked for consent to participate. Whenever possible, the interview was done immediately. The average duration of an interview was 24 minutes. Those who refused to participate or who were not sure whether they should take part were contacted again 2 weeks later; about 40% of them then agreed to participate.

Exposure Variables

Diabetes-related variables included the following: previous diagnosis of diabetes mellitus; month and year of diagnosis; use of insulin therapy, diabetes medications, and diabetic diet; month and year of first insulin treatment; and history of retinal laser photocoagulation treatment, heart disease, and numbness or amputations of toes or feet.

Type of diabetes was assigned using an algorithm adapted from Cowie and colleagues [13] (Figure 1). This algorithm was developed to select the type of diabetes in patients with diabetes-related end-stage renal disease using data routinely available in medical records. The algorithm predicted assessments of diabetes type by attending nephrologists in 84% of patients with insulin-dependent diabetes and in 92% of patients with non–insulin-dependent diabetes [13]. Potential confounders included sex, race, and history of hypertension (time since diagnosis and hospitalization because of hypertension).

Statistical Analysis

All-cause end-stage renal disease was the primary outcome variable. Persons with end-stage renal disease and controls were compared by cross-tabulations, race-stratified analyses, and logistic regression [17]. Odds ratios were used to estimate relative risks. Successive regression models were compared using the likelihood ratio test. Population-attributable risks were computed by means of a procedure developed for case–control studies [18]. Polychotomous logistic regression [17], using a three-level outcome variable (kidney failure attributed to diabetes, kidney failure attributed to other causes, and control), was used to assess risk factors for diabetic and other nephropathy. Analyses were done using SYSTAT software (Evanston, Illinois) [19]; the ATFRAC program was used to estimate population-attributable risks [20].

Results

Descriptive Data

Patients with end-stage renal disease were more likely to be men and less likely to be white than were controls (Table 1); both differences, P < 0.01). Mean age was similar in all three groups of participants (P > 0.05).

Self-reported prevalence of diabetes was about as expected (7.2%) among controls, was almost three times higher (19.3%) among patients with kidney failure attributed to causes other than diabetes, and was 100% in patients with kidney failure attributed to diabetes. Patients with end-stage renal disease were more likely than controls to have insulin-dependent diabetes. The median age at onset of diabetes was 30 years in patients with diabetic end-stage renal disease, 42.5 years in patients with nondiabetic end-stage renal disease, and 49 years in controls with diabetes. Half of the patients with end-stage renal disease attributed to diabetes had had diabetes for 15 to 24.9 years before the onset of renal failure; the median duration was 19 years. The duration of diabetes was shorter in patients who had diabetes and end-stage renal disease ascribed to another cause. Among all participants with diabetes, the median age at onset of diabetes was 19 years for patients with insulin-dependent diabetes and 36 years for patients with non–insulin-dependent diabetes.

Among patients with diabetes, those who had end-stage renal disease reported a higher lifetime prevalence of complications than did controls (all differences, P < 0.05). Controls with diabetes frequently (38%) reported only numbness of feet or toes (indicating neuropathy). Among case-patients with diabetes, 63% reported numbness of lower extremities, 68% had a history of laser photocoagulation treatment (indicating proliferative retinopathy), and 37% had heart disease from diabetes. The prevalence of retinopathy was 75% for case-patients whose renal failure was attributed to diabetes compared with 50% for those whose failure was ascribed to another cause (P > 0.01).

Odds Ratios of End-Stage Renal Disease

Overall, having diabetes mellitus of any type was associated with an odds ratio for end-stage renal disease of 11.1 (95% CI, 7.2 to 16.9). For insulin-dependent diabetes, the odds ratio was 33.7 (CI, 12.4 to 91.9), and in non–insulin-dependent diabetes, the odds ratio was 7.0 (CI, 4.4 to 11.1) when compared with the absence of diabetes. The excess risk for end-stage renal disease strongly depended on the duration of diabetes: Estimates of relative risk were only moderately increased for diabetes of less than 15 years but were much higher for diabetes of 15 to 25 years and for diabetes of more than 25 years (crude odds ratios; (Figure 2).

When type and duration of diabetes were adjusted for each other Table 2, the odds ratios for duration of diabetes continued to show the same pattern. Compared with having non–insulin-dependent diabetes, having insulin-dependent diabetes increased the odds of end-stage renal disease approximately fourfold. We could not stratify the effects of diabetes duration by type of diabetes because too few controls had insulin-dependent diabetes. To examine whether hypertension could be either a confounding or an intervening variable in the causal pathway between diabetes and end-stage renal disease, we added duration and severity of hypertension to the regression model (Table 2). Adjustment for hypertension decreased the odds ratios for diabetes of any duration to about half of the previous values, suggesting that hypertension could be an intervening variable between diabetes and end-stage renal disease.

Race-Specific Odds Ratios

Black race was associated with a 7.4-fold increase in odds for end-stage renal disease (CI, 5.3 to 10.3) from any cause compared with white race [21]. The patterns of risk by type of diabetes differed between blacks and whites. In whites, the odds ratio for insulin-dependent diabetes was 49.0 (CI, 15.2 to 157.5); for non–insulin-dependent diabetes, the odds ratio was 6.0 (CI, 3.4 to 10.4) compared with the absence of diabetes. In blacks, the corresponding odds ratios were 15.7 (CI, 2.1 to 117) for insulin-dependent diabetes and 7.1 (CI, 2.5 to 20.3) for non–insulin-dependent diabetes. The racial difference, tested in a regression model with an interaction term between black race and type of diabetes, was significant (P < 0.01). Blacks and whites had a similar pattern of relative odds for end-stage renal disease associated with longer duration of diabetes (Table 2).

End-Stage Renal Disease Attributed to Diabetes and to Other Causes

The previous analyses were for end-stage renal disease regardless of the underlying cause assigned by the patient's nephrologist. Using polychotomous logistic regression, we repeated the odds ratio computations while distinguishing between end-stage renal disease attributed to diabetes and to any other cause. Patients with end-stage renal disease who had insulin-dependent diabetes and those who had had diabetes for 15 years or longer (whether insulin-dependent or non–insulin-dependent), tended to be labeled as having “diabetic” nephropathy. The associations between type and duration of diabetes and end-stage renal disease were strong and statistically significant only for end-stage renal disease attributed to diabetes but were much weaker and nonsignificant for end-stage renal disease attributed to other causes (results not shown).

Population-attributable Risks for End-Stage Renal Disease

Population-attributable risk estimates provide a measure of maximum potential benefit, in terms of the reduction in incidence of end-stage renal disease, from primary prevention of diabetes. For diabetes mellitus, without distinction of type, the estimate of population attributable risk was 42% (Table 3). Half of the population-attributable risk was associated with insulin-dependent diabetes (21%) and the other half with non–insulin-dependent diabetes (21%). Although the overall effect of diabetes was only slightly less in black persons (41%) than in white persons (44%), the relative importance of each type of diabetes varied considerably: In whites, insulin-dependent diabetes accounted for more patients with end-stage renal disease than did non–insulin-dependent diabetes, whereas in blacks, non–insulin-dependent diabetes was the more important contributor.

Estimates of population-attributable risk were then compared with the proportion of patients with end-stage renal disease who were identified by their nephrologists as having diabetic nephropathy, in the study sample and in the overall U.S. population of patients aged 20 to 64 years with end-stage renal disease (Table 3). In whites, the proportion of end-stage renal disease attributable to diabetes was similar (40% to 44%), whether based on population-attributable risks or on clinical diagnoses. In blacks, however, the population-attributable risk estimate (41%) was considerably higher than were estimates based on clinical diagnoses of underlying disease (29% in the study sample, 33% in the U.S. population).

Discussion

The Effect of Diabetes on End-Stage Renal Disease

In this case–control study, we calculated what proportion of the incidence of end-stage renal disease in an unselected population of nonelderly adults can be attributed to diabetes mellitus. We found that 42% of all new cases of treated end-stage renal disease in this population could be ascribed to diabetes. This estimate was based on a statistical comparison of the risk for end-stage renal disease in persons with and without diabetes. Thus, this result was conceptually independent of the traditional approach for assessing the effect of diabetes on the incidence of kidney failure, which relies on clinical diagnoses of the underlying causes of renal failure. Both methods indicate that diabetes mellitus is the predominant cause of renal failure in nonelderly U.S. adults. Our analysis suggests that the incidence of diabetes-related end-stage renal disease is greater than that estimated from clinical data.

Clinical diagnoses and population-attributable risks yield different estimates because their assessment is based on different premises. Clinical diagnoses fully use relevant information available to the attending physician. Unfortunately, their validity and reliability are unknown. Selecting the underlying cause of kidney disease may be a challenge, especially when several possible causes of kidney failure coexist in a given patient. Diabetes may render the kidney more vulnerable to other disease processes, but nephropathy in patients with diabetes may also have clinical forms that can be confused with other kidney diseases. For instance, patients with diabetes have various histologic patterns of renal disease [22].

The approach used in our study is advantageous because we avoided the issue of assigning a single cause of end-stage renal disease to each patient, which may be impossible if the disease process is multicausal. On the other hand, the method we used to compute population-attributable risks [18] requires valid estimates of the relative risk for kidney failure and of diabetes prevalence in patients with end-stage renal disease. The prevalence of diabetes in patients with end-stage renal disease who are under close medical scrutiny is likely to be accurate. In contrast, some controls may have been unaware of having diabetes, particularly non–insulin-dependent diabetes. This would have led us to overestimate the odds ratio of kidney failure associated with diabetes. If present, this bias must have been minor because diabetes prevalence in controls (7.2%) was close to what was expected in this population. The use of odds ratios to approximate relative risks caused only a negligible distortion of population-attributable risk estimates given the low incidence (in absolute terms) of end-stage renal disease in the general population. Our high participation rates (95% in cases, 86% in controls) ensured that study participants were representative of the general population, which is another necessary condition for the correct interpretation of population-attributable risks. Reporting of the duration and treatment modality for diabetes was probably accurate because medical history reports from patients with diabetes have been shown to have high validity [23].

Racial Differences

Race-stratified analyses provided new insights into the mechanisms of the excessive amount of end-stage renal disease in blacks. Diabetic nephropathy is currently diagnosed in 43% of white patients with end-stage renal disease aged 20 to 64 years but is diagnosed only in 33% of black patients with end-stage renal disease who are in the same age group [1]. The relative importance of diabetes is not only higher when population-attributable risks are considered, but the racial difference is also smaller (44% in whites and 41% in blacks). Thus, the role of diabetes in the excess incidence of end-stage renal disease in blacks may be underestimated by national statistics based on the presumed underlying cause of kidney disease. This underestimate may be caused by the tendency of nephrologists to diagnose a renal disease other than diabetic nephropathy, such as hypertensive end-stage renal disease [24], more readily in black than in white patients. A competing cause of kidney disease may also be more readily selected in patients who have non–insulin-dependent diabetes as opposed to insulin-dependent diabetes; this situation is also more frequent among blacks than among whites.

Although the proportion of patients with end-stage renal disease attributable to diabetes appears to be similar in blacks and whites, this cannot be said of the incidence of end-stage renal disease (expressed as the annual number of new cases of end-stage renal disease per population at risk). The reason is that the overall incidence of end-stage renal disease is about four times greater in blacks than in whites. In 1988 to 1990, the unadjusted annual incidence of end-stage renal disease among U.S. blacks aged 20 to 64 years was 335 cases per million [1]. Of these, 41% (or 137 cases per million) could be attributed to diabetes according to our analysis. In whites, the incidence of end-stage renal disease was 86 cases per million per year, of which 44% (or 38 cases per million) could be attributed to diabetes. Measured in absolute terms, the effect of diabetes was much greater among blacks than among whites.

Black and white persons also differed in terms of the relative importance of insulin-dependent and non–insulin-dependent diabetes. The former was the predominant cause of end-stage renal disease in whites; the latter was the predominant cause in blacks. Reasons for this difference, which were also suggested by two other studies from Michigan [13] and Maryland [14], are unclear. This difference may be due to a biological interaction between hypertension and non–insulin-dependent diabetes, which occur together more frequently in blacks than in whites, or to a racial discrepancy in the quality of care (glycemic control, concurrent control of blood pressure) in patients with non–insulin-dependent diabetes.

Comparison with Other Studies

Results of our case–control study are consistent with previous observations. Insulin-dependent diabetes was associated with a fivefold increase in the risk for end-stage renal disease compared with non–insulin-dependent diabetes. This is close to the value of 4.7 obtained in the Rochester cohort study [6] but is lower than the value of 11.6 obtained from a combination of various data sources in the Michigan study [13]. The latency period for diabetic renal failure estimated in this study was also similar to that reported in other studies [6-11]. The odds ratio for end-stage renal disease was not substantially increased for diabetes of less than 15 years duration, but the odds ratio increased sharply for diabetes of 15 to 25 years duration. In the Rochester study [6], the cumulative incidence of chronic renal failure took an upward turn after about 15 to 20 years of follow-up. Data from the Joslin Clinic [7] also indicate that in patients with type I diabetes, end-stage renal disease first occurs about 13 years after diagnosis of diabetes; after that, the cumulative incidence of end-stage renal disease increases steadily without evidence of a plateau.

Our results indirectly support the suggestion that hypertension may be an important cofactor in diabetic nephropathy [25, 26]. When adjusted for the duration and severity of hypertension, the odds ratios associated with longer duration of diabetes decreased considerably, and the odds ratios associated with hypertension also decreased. This finding indicates only that diabetes and hypertension are associated at the population level; whether diabetes and hypertension are related by a causal biological mechanism remains uncertain.

Recommendations

Using a methodologic approach that differs from the traditional assessment of cause of renal failure, our study supports the hypothesis that diabetes is the leading cause of end-stage renal disease in nonelderly adults. Prevention of diabetes-associated kidney failure should have a high priority because more than 4 out of 10 cases of end-stage renal disease can be attributed to diabetes. Unfortunately, determinants of kidney failure in patients with diabetes are not sufficiently known. Our study suggests that the case–control design may be a valuable approach to exploring the possible causes of kidney failure in diabetes. In addition, other study designs, such as nonconcurrent follow-up of persons screened in large cardiovascular disease prevention trials [27], could also yield useful information.

Kidney failure associated with non–insulin-dependent diabetes should receive increased attention from laboratory and clinical researchers. From the point of view of the community burden of disease, this entity deserves at least as much research effort as does insulin-dependent diabetes-associated kidney failure. Kidney failure associated with non–insulin-dependent diabetes should perhaps be given priority because an understanding of it would explain a large part of the impressive but poorly understood excessive amount of end-stage renal disease in blacks.