Diabetes-Related Morbidity and Mortality in a National Sample of U.S. Elders

We conducted a national nonconcurrent cohort study of U.S. elders with diabetes by using data from the 5% Standard Analytical File, which contained Medicare claims for 1,941,453 individuals in 1994. Details regarding inclusion/exclusion criteria and data collection have been published (9,10). We selected 1,284,653 beneficiaries aged ≥65 years from the 1994 Medicare Denominator file (containing demographic and eligibility data) who lived in the U.S., who were enrolled in both Medicare Part A (inpatient) and Part B (outpatient), who did not die in 1994, and who were not enrolled in a health maintenance organization (HMO) at any time in 1994 because individual claims for disease episodes would not have been generated in this setting. Individuals were classified as having diabetes if, in 1994, there were at least two Medicare Part B claims with a diabetes-related ICD-9 diagnosis code (250.xx) or at least one hospitalization claim with a diabetes-related diagnosis code; this yielded 178,792 patients with diabetes. Merging the claims-identified diabetes patients with the post exclusion denominator file yielded a final sample of 148,562 elders with diabetes alive as of 1 January 1995.

Two national data sets were used for comparison purposes. First, to confirm the national representativeness of the Medicare sample, we compared the demographic characteristics of this study cohort with the 1990 Census (≥65 years of age). Second, to determine the standardized mortality ratio (SMR) related to diabetes, we compared the mortality rates of this cohort with the 1996 vital statistics for individuals aged ≥65 years from the National Center for Health Statistics.

We selected diabetes-related complications to investigate based on several considerations, including relatedness to diabetes, typical severity, likelihood of requiring hospitalization or outpatient visit, and/or potential for prevention. Deaths were ascertained using the Medicare Denominator File, which includes data on mortality, including date of death (11). Complications were classified using claims in the 5% Standard Analytical File from 1994 to 1996. A complication was considered to have occurred if at least one claim was made for the outcome. These included amputation (ICD-9 codes 84.10–84.19 and Current Procedural Terminology codes 28810, 28820, 28825), gangrene (ICD-9 codes 785.4, 040.0, and 440.24), lower extremity infection (a composite of lymphangitis [ICD-9 code 457.2] and cellulitis [ICD-9 codes 680.6, 680.7, 682.1, 682.6, and 682.7]), blindness (ICD-9 codes 369.0–369.9), acute myocardial infarction (ICD-9 code 410.x), ischemic heart disease (ICD-9 codes 410–414), stroke (ICD-9 codes 430–438), and the diabetes-related metabolic disorders, i.e., hypoglycemia (ICD-9 codes 251.1 and 251.2), hyperosmolar syndrome (ICD-9 code 250.2), and ketoacidosis (ICD-9 code 250.1).

We identified the date of death for deaths occurring in 1995 and 1996. We used survival analysis to calculate event rates, assuming person-time accrued from 1 January 1995 to the date of complication, death, or censoring (12). For determination of complication rates, censoring occurred due to withdrawal to HMO care in 1995 or 1996 (censoring date 31 December 1994 or 1995) or the end of the observation period (31 December 1996). For mortality rates, we did not censor for withdrawal to HMO because Medicare records the date of death for all beneficiaries. We identified claims for diabetes-related complications from 1 January 1994 to 31 December 1996. In all complications analyses, we excluded only individuals with the specific complication of interest in 1994 in order to calculate incidence in 1995–1996. Because discrete episodes of care are difficult to determine for events that can plausibly occur more than once in a year (e.g., ketoacidosis) or even more than once in a hospitalization (e.g., amputation), we only counted the first appearance of the diagnostic code in a given year but allowed for up to one occurrence per year in most analyses. Due to the chronic nature of blindness, ischemic heart disease, and stroke, we identified incident cases of these diagnoses as the first occurring episode after 1 January 1995. All complications were deemed to have occurred at midyear or on the date of death if the patient died before midyear. Based on the age in 1994, we created the following age strata, taking into account aging from 1994 to 1996: 65–69, 70–74, 75–79, 80–84, and ≥85 years. We used these strata to calculate rates by dividing the number of individuals with a complication by the total number of individuals at risk in each strata. Mortality rates were similarly calculated. Where appropriate, we carried out direct age-adjustment using the 1990 Census population aged ≥65 years as the standard. We compared death rates for this cohort to mortality rates drawn from vital statistics for 1996 to calculate a standardized mortality ratio (SMR; ratio of observed number of deaths to expected number of deaths). Confidence intervals for rates and SMRs were calculated assuming that parameters followed a Poisson distribution (12). Cox proportional hazards regression was used to assess the relative hazards associated with age and sex. All significance tests were two-tailed.

Compared with all U.S. residents aged ≥65 years enumerated in the 1990 census, elders with diabetes in our cohort were similar in terms of sex distribution (39.9 vs. 40% male) and ethnicity (11.9 vs. 9.9% African-American). However, this cohort was substantially older than the general elderly population: 22 vs. 33% aged 65–69 years and 48 vs. 42% aged ≥75 years.

During 24 months of follow-up, 22,044 patients died. The unadjusted all-cause mortality rate was 79.6 per 1,000 person-years (95% CI 78.6–80.7). The rate was 69.9 per 1,000 person-years after adjustment to the 1990 census population. Of note, the mortality rate in the 7,066 persons (4.5% of the 1994 cohort) who enrolled in an HMO in 1995 or 1996 was 30.3 per 1,000 person-years, compared with 82.3 per 1,000 person-years in those who remained enrolled in fee-for-service Medicare. The incidence of the selected diabetes-related complications in 1996 is presented in Table 1. Ischemic heart disease was prevalent in 38% of elders with diabetes in 1994; over the ensuing 2 years, 30% of the remaining cohort recorded a diagnosis of ischemic heart disease. Stroke was the second most common complication; in those without a diagnosis of stroke in 1994 (83% of the 1994 cohort), 22% recorded a diagnosis of stroke in 1996. Infection of the lower extremity and gangrene occurred more frequently than amputation of the lower extremity. The mortality rate was much higher in individuals with incident ischemic heart disease (102 per 1,000 person-years, 95% CI 100.3–103.8) compared with those without ischemic heart disease (59 per 1,000 person-years, 95% CI 57.8–60.4). The highest death rates occurred in individuals with incident acute myocardial infarction (183 per 1,000 person-years), amputation (190 per 1,000 person-years), and gangrene (216 per 1,000 person-years). Of the metabolic complications, hypoglycemia occurred most frequently, whereas hyperosmolar syndrome occurred infrequently.

After age-adjustment using the age distribution of the U.S. population aged ≥65 years in 1990, the mortality rate was 62.3 per 1,000 in women and 81.8 per 1,000 in men. Mortality rates increased with age similarly in men and women with diabetes (Fig. 1). The overall age-adjusted relative risk of mortality for men with diabetes was 1.34 (95% CI 1.31–1.38) compared with women with diabetes.

At every age, mortality in elders with diabetes was significantly greater than mortality in the general population. The SMR for the entire cohort was 1.41 (95% CI 1.39–1.43); the SMR was 1.44 (1.41–1.46) in women and 1.37 (1.34–1.40) in men. Excess mortality declined with age in men and women (Fig. 2). Although the mortality rate was higher in diabetic men than diabetic women, in individuals aged <85 years, the excess mortality was greater in women. Of note, even in individuals aged ≥85 years, excess mortality was still significantly elevated in both men and women with diabetes (overall SMR 1.15, 95% CI 1.12–1.18).

Rates of selected complications are listed, by age group, in Fig. 3. All complication rates increased with age (all P values for trend <0.01). After adjusting for age, women had a higher risk of hypoglycemia (relative hazard [RH] 1.27), hyperosmolar syndrome (RH 1.19, P < 0.001), ketoacidosis (RH 1.17), and blindness (1.22). Women had a lower risk of amputation (0.70), gangrene (0.78), acute myocardial infarction (0.80), and ischemic heart disease (0.88) (all P < 0.001). There was no gender difference in the occurrence of infection of the lower extremity and stroke.

The results of this study support the following conclusions. First, diabetes is associated with excess mortality in U.S. elders, even in those aged ≥85 years. Second, the leading causes of diabetes-related morbidity in elderly individuals are ischemic heart disease and stroke. Third, all complications assessed increase with age. Finally, this work suggests that it is feasible to use claims data for national surveillance of diabetes-related morbidity and mortality. Strengths of our study include a large, nationally representative sample of patients with diabetes, the use of survival time analysis, and a study design that both allowed for the temporal separation of exposure from outcomes and minimized the possibility of detecting preexisting complications.

The major limitation of this study is the exclusive reliance on claims data, which raises concerns about possible misclassification bias. Individuals not identified as having diabetes in 1994 may lack Medicare Part B coverage, may be enrolled in an HMO, may have undiagnosed diabetes, or may be nonusers of care. Individuals without Medicare Part B coverage may not be able to afford it, thus rendering our identified cohort healthier due to a higher socioeconomic status, whereas those individuals who were enrolled in an HMO in 1994 are likely to be healthier than this cohort. In fact, individuals who were censored due to HMO enrollment in 1995 and 1996, after adjustment for age and sex, were less likely to have had prevalent ischemic heart disease, stroke, amputation, or gangrene in 1994. Individuals with undiagnosed diabetes have an elevated all-cause and cardiovascular mortality risk, but not as high as those diagnosed with diabetes (13). It is also possible that individuals simply did not list diabetes as one of the diagnoses supporting a claim. However, our scheme for defining diabetes has been reported to have a sensitivity of 63% and a specificity of 98% (14). Furthermore, of our initial sample (after general exclusions), 14% had diabetes, which is consistent with national estimates of the prevalence of diabetes in the elderly U.S. population. Moreover, when clinical and claims data were compared, high specificity was demonstrated for diagnoses such as congestive heart failure, hyperlipidemia, angina, and stroke (15). Similar data for many other complications are not available, although the specificity for death is likely to be high because it results in the termination of benefits.

Methodological issues have impeded the determination of accurate morbidity and mortality rates in elders with diabetes, whether based on vital statistics/death certificate data or cohort studies (8,16). Previously published mortality rates in elderly diabetic individuals range from 51.4 to 89.7 per 1,000 person-years at ≥65 years of age (7,17,18) and up to 136 per 1,000 person-years at ≥75 years of age (16). In comparison, our mortality rates are somewhat lower. Several possible explanations may account for this difference. The previously published estimates are chiefly from the 1980s, and death rates may have declined in the following 10 years. Alternatively, eliminating from our population in 1994 individuals without Medicare Part B coverage may have resulted in a population with diabetes comprising individuals with a better socioeconomic status and/or better access to care who are less likely to die. Possibly counteracting this influence, however, is the observation that the HMO enrollees in 1995/1996 had lower death rates than those who remained, suggesting that those excluded in 1994 would have also had lower death rates. Finally, there may be underascertainment of death by Medicare, compared with cohort studies, but without personal identifiers, vital status cannot be validated against the national death index.

Because we compared mortality rates with those of the general population, which includes individuals with diabetes, the excess risk of death due to diabetes is probably underestimated in this study. However, the inverse association of excess mortality (as indicated by the SMR) with age is consistent with previous studies of mortality related to diabetes, whether reporting SMR or relative risk (7,8,19–21). Similarly, the greater excess of death in elderly diabetic women than diabetic men, compared with their nondiabetic counterparts, has also been reported elsewhere (16).

The complication rates we report may be conservative because for complication-specific analyses, we chose to exclude elders with that specific complication in 1994. However, we may have overestimated rates of complications by excluding Medicare HMO enrollees from our cohort, which may have resulted in a less healthy population. Nonetheless, our complication rates are similar to previous reports using cohort-derived data or national surveillance data with regard to amputation (6.2–11.9 per 1,000 person-years) (22–24) and hypoglycemia (18.1 per 1,000 person-years) (25). Published U.S. surveillance data for 1994 showed a rate of 10.2 fewer extremity amputations per 1,000 diabetic elders aged 65–74 years and 11.9 per 1,000 in those aged ≥75 years (26). The rate of ketoacidosis is higher in this cohort than that reported by the Centers for Disease Control and Prevention in 1994: 4.6 discharges per 1,000 diabetic patients aged ≥65 years (27). Data from the Wisconsin Epidemiologic Study of Diabetic Retinopathy demonstrated an incidence of blindness of 4 per 1,000 person-years in insulin users aged 65–74 years (28). Our estimate of 10.2 per 1,000 person-years may be higher because we were unable to determine whether the cause of blindness was related to diabetes or other aging-related causes, such as macular degeneration.

There are several implications of our study. First, Medicare data may be useful as a surveillance tool for monitoring trends in diabetes-related morbidity and mortality in elderly individuals over time. Second, greater efforts to prevent cardiovascular disease and lower extremity complications are needed in elderly individuals with diabetes. Finally, the observation that diabetes is associated with excess mortality across age groups suggests that the treatment of diabetes and the conditions that commonly accompany it may be desirable even in the oldest patients.

This study was supported, in part, by NHLBI Training Grant 5-T32 HL07180–24 (to A.G.B.) and by an Established Investigator grant from the American Heart Association (to F.L.B.).