Economic Costs of Diabetes in the U.S. in 2002

American Diabetes Association

Diabetes Care. 2003;26(3) 

In This Article

Research Design and Methods

The approach used to estimate the cost of diabetes follows, to the extent possible, the approach used in previous studies of the cost of diabetes and, in particular, ADA's previous cost estimate[2]. This approach has found acceptance in the general cost-of-disease literature. Deviations from the approach used previously by ADA[2] are noted and occur in some instances when the exact approach used in the earlier study could not be determined or when new data sources and analytical tools enable improvements to past approaches. Below is a summary of the approach used to estimate 1) the size of the population with diabetes, 2) health care use and total health care expenditures attributable to diabetes, and 3) the value of lost productivity attributable to diabetes.

This national cost estimate is based on an estimate of 12.1 million people in the U.S. in 2002 who have been diagnosed with diabetes. This estimate of the magnitude of the diabetic population represents an increase of 1 million (9%) from year 2000 estimates and an increase of 1.8 million (17%) from year 1997 estimates. Based on results from the 2000 Census, it appears that during the period of 1990-2000, the U.S. population grew faster than projected by the Census Bureau. (The actual U.S. population in 2000, based on the 2000 Census, exceeded the Census Bureau's pre-2000 projections of the U.S. population in 2000 by ~6.8 million individuals [or 2.4% of the total population]. One implication is that pre-2000 estimates of the number of people with diabetes in the U.S. were biased downward because the sample weights used in surveys such as the National Health Interview Survey (NHIS) were based on Census Bureau population estimates.) It is based on self-reported prevalence of diabetes only; therefore, it does not account for the considerable number of people with diabetes who are unaware that they have the disease or do not report it. Indeed, the ADA estimates that as many as one-third of people with diabetes are unaware that they have the disease. Further, this estimate excludes women with gestational diabetes.

This cost estimate is based on prevalence rates derived from the combined 1998, 1999, and 2000 files of the NHIS. Combining 3 years' worth of NHIS files created larger samples with which to estimate separate prevalence rates for each of 12 age-groups by sex and by four race/ethnicity designations. (The 12 age categories are 0-17, 18-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54, 55-59, 60-64, 65-69, and ≥70 years. The four race/ethnicity categories are Hispanic, non-Hispanic white, non-Hispanic black, and non-Hispanic other.)

The NHIS collects data on ~43,000 households of more than 106,000 people annually. The combined files for 1998-2000 create a sample of more than 320,000 people. People with diabetes are identified using the survey question that asks whether the survey participant has been told by a doctor that he or she has diabetes (other than gestational diabetes). Responses to the question are coded as "yes," "no," "borderline," and "no response." People responding "yes" are coded as having diabetes. People responding "borderline" are not counted as having diabetes in this analysis.

As shown in Figs. 1 and 2, diabetes prevalence rates increase with age. Prevalence rates vary substantially by race and ethnicity. They are higher for Hispanics and non-Hispanic blacks than for non-Hispanic whites. Furthermore, the rates for other populations (i.e., Asian Americans, American Indians, Pacific Islanders, etc.) are similar to those of non-Hispanic whites among females but are higher than the rates for non-Hispanic whites among males.

Proportion of female population with confirmed diabetes in 2002.

Proportion of male population with confirmed diabetes in 2002.

Applying these prevalence rates to the size of the U.S. population in each demographic group, as determined by the 2000 Census and projected to 2002 using Census Bureau estimates, produced the estimate of 12.1 million people diagnosed with diabetes.

If diabetes prevalence rates within a demographic group remained constant over time, then, based on Census Bureau population projections[1], the size of the population with diabetes will grow to ~14.5 million by 2010 and to 17.4 million by 2020 ( Table 1 ). Whereas the U.S. population is projected to increase by ~17% between 2002 and 2020, the size of the population diagnosed with diabetes is projected to increase by 44% due, in large part, to the increase in the size of the elderly population and the increasing racial and ethnic diversity of the U.S. population. Changing demographic characteristics will contribute to an increase in the overall prevalence rate for diagnosed cases of diabetes from 4.2% in 2002 to a projected 5.2% in 2020.

The number of Hispanics and other minority populations diagnosed with diabetes is projected to double between 2002 and 2020, whereas the number of non-Hispanic blacks and non-Hispanic whites diagnosed with diabetes is projected to increase by 50 and 27%, respectively.

Although there is no projected increase in the total number of people under age 45 years diagnosed with diabetes between 2002 and 2020, the projected increases for populations aged 45-64 and ≥65 years are 48 and 56%, respectively.

In addition to receiving health care services for medical conditions directly related to diabetes, people with diabetes are at greater risk for neurological disease, peripheral vascular disease, cardiovascular disease, renal disease, endocrine/metabolic complications, ophthalmic disease, and other chronic complications compared with individuals without diabetes. A portion of health care use associated with these medical conditions is attributable to diabetes.

The general principle for estimating the cost of diabetes in this analysis is straightforward. Health care use attributable to diabetes is determined by a comparison of the health care use patterns of individuals with and without diabetes, controlling for differences between the two populations in demographic characteristics that are potentially correlated with the use of health care services (e.g., age, sex, and race/ethnicity).

Three limitations of the source data, however, increase the complexity of the analysis design and calculations. These limitations are 1) absence of a single data source for all estimates, 2) small sample sizes for some items of interest, and 3) underreporting of diabetes as a comorbidity. The implications of these limitations and how we have addressed these limitations are summarized below.

  • No single source of data. Because no single data source representative of the U.S. population contains all of the information necessary to estimate the health care cost of diabetes, it is necessary to draw upon multiple data sources. Among some of these sources are differences in definitions for identifying people with diabetes and differences in levels of detail to categorize types of patient visits. One source of complete data required to estimate direct medical expenditures attributable to diabetes is claims from Group Health of Puget Sound (GHPS). This data source contains a diabetic flag in a disease registry, but the GHPS sample might not be representative of health care use patterns and costs for the entire U.S. population. The Medical Expenditure Panel Survey (MEPS) is closest to a single, nationally representative source of data in that it 1) identifies people with diabetes-related conditions, 2) measures health care use, and 3) provides cost information. However, MEPS is limited by the small sample size.

  • Small sample sizes for some items of interests. Disaggregating the U.S. population by age, sex, race, and ethnicity requires relatively large sample sizes to obtain reliable estimates of differences in use patterns by diabetes status when analyzing specific medical conditions associated with diabetes. The number of identified people with diabetes who participated in the most recent MEPS is insufficient to obtain reliable estimates of health care use for some chronic complications associated with diabetes and in some health care settings. The use of alternative data sources, such as the National Ambulatory Medical Care Survey, increases sample size but is hindered by the third major limitation -- underreporting of diabetes as a comorbidity.

  • Underreporting of diabetes as a comorbidity. The literature reports that there is significant underreporting of diabetes as a comorbidity in health care databases. Unless the attending physician lists diabetes as a comorbidity on the patient's medical record, the health care services provided to that patient are not linked to diabetes. Sources such as GHPS and MEPS allow one to identify whether a person has been diagnosed with diabetes but, as discussed above, their representativeness is often questioned (in the case of GHPS) or they have insufficient sample size.

These data limitations are addressed as follows. First, the study uses an eclectic approach that combines findings from empirical analysis of multiple data sources with findings reported in the literature. Second, for several national surveys completed annually, multiple years' worth of data are pooled to increase sample size. Third, similar to previous studies, this study uses an attributable risk methodology to estimate use of health care services that can be attributed to diabetes.

The attributable risk methodology estimates the odds of having a particular medical condition by diabetes status, then combines these odds with estimates of the proportion of the population with diabetes to calculate an etiological fraction. The etiological fraction represents an estimate of the proportion of health care services for a particular medical condition that is attributable to diabetes. The etiological fraction is calculated based on the following:

where E i is the fraction of health care use for medical condition "i" that is attributable to diabetes, P represents the diabetes prevalence rate, and R i is the relative risk of disease i (i.e., the odds ratio) for people with diabetes compared with people without diabetes.

Combining odds ratios estimated using the MEPS with diabetes prevalence rates estimated using the NHIS creates separate etiological fractions for the medical conditions listed in Fig. 3 for each demographic group modeled. This figure combines etiological fractions across the 12 age-groups by race/ethnicity and sex to present etiological fractions for the population aged <45, 45-64, and ≥65 years. The etiological fractions vary substantially by age to reflect the changing prevalence of diabetes and differences in the prevalence of specific medical conditions by age. For example, for the populations aged <45, 45-64, and ≥65 years, the proportions of all health care use associated with neurological disease that is attributable to diabetes are 6, 10, and 5%, respectively. The medical condition with the highest etiological fractions is cardiovascular disease, where the proportions of all health care visits attributable to diabetes for individuals aged <45, 45-64, and ≥65 years are 16, 20, and 17%, respectively.

Etiological fractions, adjusted for race/ethnicity, sex, and finer age-groupings. CVD, cardiovascular disease; GMC, general medical conditions; PVD, peripheral vascular disease; OCC, other chronic complications.

Although not reported here, the etiological fractions vary substantially by race and ethnicity, with the fractions generally higher for Hispanics and non-whites compared with non-Hispanic whites. This finding is consistent with past research that shows ethnic disparities in both diabetes prevalence rates and the rates of diabetic complications[4].

Table 2 summarizes the data sources used to analyze each component of the cost analysis and summarizes the unit cost estimates. Sources of health care use data include the 1998-2000 files of the National Ambulatory Medical Care Survey, the 1998-2000 files of the National Hospital Ambulatory Medical Care Survey, the 1999 National Inpatient Sample, the 1999 National Nursing Home Survey, and the 1998 and 2000 files of the National Home and Hospice Care Survey.

For each of these files, the primary diagnosis is used to classify the health care visit (or inpatient day) into one of nine medical condition classifications: 1) diabetes without complications, 2) one of the seven chronic medical conditions above (i.e., neurological disease, peripheral vascular disease, cardiovascular disease, renal disease, endocrine/metabolic complications, ophthalmic disease, and other chronic complications), or 3) neither 1) nor 2), in which case the visit is classified as a "general medical condition." (See the appendix for a list of diagnosis codes used to categorize visits and hospital inpatient days by medical condition.)

Health care use rates for each of the nine conditions in each health delivery setting are estimated by patient age, sex, and race/ethnicity. Combining these use rates with etiological fractions and estimates of population size for each demographic group produces national estimates of health care use attributable to diabetes for each medical condition.

The 1998 MEPS is the primary source for most estimates of the per-unit price of health care services. Price estimates are based on actual payment for services, not charges. Price estimates from other sources are used when such information is readily available for more recent years or when price estimates from the MEPS appear unreliable (e.g., because of small sample sizes in the MEPS). All price estimates for health care services are adjusted to 2002 dollars using the medical component of the consumer price index. The unit prices represent averages across all patients irrespective of diabetes status or reason for visit (neonatal inpatient stays were omitted from the calculation of average cost per day). To the extent that the unit price is higher when diabetes is a comorbidity, or that inpatient days and outpatient visits tend to be more expensive for the medical complications associated with diabetes, the average unit cost might underrepresent the true unit cost for services attributable to diabetes.

Estimates of the average annual cost of supplies for people using insulin and oral agents were calculated using cost data from The Source Prescription Audit, the 2002 Red Book[13], pharmaceutical companies, and suppliers of devices used by people taking insulin. Based on prevalence rates computed using the combined 1998-2000 files of NHIS and estimates of the population in 2002, the estimated number of people using oral agents for diabetes and the estimated number of people using insulin are 7.5 and 3.9 million, respectively. The percentage of people using insulin and oral agents varies substantially by age, reflecting the increasing proportion of cases involving type 2 diabetes among the population with diabetes in older age brackets. Not all people with diabetes use either insulin or oral agents, especially among the younger age brackets.

People with diabetes are at greater risk of temporary incapacity (defined as lost workdays and bed days), permanent disability, and premature mortality. The pecuniary value of lost productivity is calculated based on the average earnings of the person whose productivity is foregone. Bureau of Labor Statistics (BLS) estimates of year 2001 annual earnings by age and sex for the civilian noninstitutional population are used to estimate the average cost per day of missing work, the average cost per year of permanent disability, and the loss of expected lifetime earnings resulting from premature mortality[19,20]. Earnings estimates for 2001 are inflated to 2002 dollars using the overall consumer price index.

The economic impact of temporary incapacity due to diabetes can be measured by both workdays lost and number of bed days, because both capture physical limitation due to diabetes that results in lost productivity. These data are obtained from the NHIS, in which respondents report workdays lost and bed days during the previous year due to illness. Lost workdays are defined as days in which a person misses work at a job or business because of diabetes or diabetes-related injury (excluding maternity leave). Bed days are defined as days in which a person is kept in bed more than half of the day because of diabetes or diabetes-related injury (including days while an overnight patient at a hospital). Lost workdays are subtracted from bed days to prevent overcounting if a person has both a lost workday and a bed day.

An estimate of workdays lost due to diabetes is found by comparing average days lost by diabetes status for each age-group and by sex. Controlling for age, men with diabetes have 3.1 more lost workdays and 7.9 more bed days per year, on average, than men without diabetes. Women with diabetes had 0.6 more lost workdays and 8.1 more bed days, on average, than women without diabetes. However, these estimates likely underestimate lost workdays to the extent that men and women with diabetes are less likely to be in the labor force than men and women without diabetes.

The pecuniary value of a workday is defined as average earnings for the person incurring the lost workday. Average earnings differ by age-group and sex, but the average earnings for people with diabetes who are between the ages of 18 and 64 is estimated at $168 per day. Following the approach used by Yassin et al.[21], the cost per bed day is defined as 40% of the cost of a lost workday.

People with diabetes are at greater risk for amputations, loss of vision, and other physical problems that can limit their earning potential or preclude them from gainful employment. Ideally, estimating lost earnings would entail comparing the average earnings of all people with diabetes to the average earnings of people without diabetes, controlling for differences in demographic characteristics and other factors that affect earning potential but that are unrelated to diabetes. A comparison of gross average earnings would capture both differences in labor force participation patterns and the possibility that an individual with diabetes will be in a lower-paying job. Unfortunately, there are no recent data that provide reliable information with which to estimate average earnings by diabetes status while controlling for demographic and other factors affecting earning potential.

Consequently, following the approach previously used by the ADA[2], data from the Social Security Administration are used to estimate the prevalence of total number of permanently disabled workers attributable to diabetes. The Social Security Disability Insurance (SSDI) program provides benefits to disabled workers and their spouses or children (whether or not they are disabled), retired workers and their dependent family members, and survivors of deceased workers. Individuals aged 18-64 years who receive SSDI benefits are included in the estimate of lost productivity attributed to diabetes-related disability. The Social Security Administration Office of Research, Evaluation, and Statistics compiles information on the total number of people with disabilities by specified condition. Therefore, using information on the number of disabled workers as a percentage of the total number of beneficiaries from Table 1 in the Annual Statistical Report on the Social Security Disability Insurance Program, 2000, we adjusted the Social Security Administration data to reflect the number of disabled workers by specified condition.

As of January 2002, there were an estimated 122,000 people aged 18-64 years receiving SSDI benefits where diabetes is listed as the primary basis of disability and another 109,000 people aged 18-64 years receiving SSDI benefits where diabetes is listed as the secondary basis of disability. This study attributes to diabetes 100% of the cases where diabetes is the primary basis of disability and 50% of the cases where diabetes is the secondary basis of disability. The number of cases where diabetes is a contributing factor to the disability, but where diabetes is not listed as the primary or secondary diagnosis, was unavailable. Also, the number of unemployed people with diabetes who are not receiving SSDI but who would be employed in the absence of diabetes is unknown. An estimated 176,475 person-years of permanent disability in 2002 are attributable to diabetes. Each case of permanent disability results in an average lost earnings of $42,462 per year. The national cost estimate excludes the cost to family and friends caring for a person with permanent disabilities attributable to diabetes.

Data from the 1998 Multiple Cause of Death File[22] were used to determine the total number of deaths attributable to diabetes. The file reports causes of death, along with economic, geographic, and demographic information for deaths of all U.S. citizens occurring within the U.S.

Mortality-related productivity costs are the estimated value of lost future earnings from paid market and unpaid household labor resulting from premature death due to diabetes or diabetes-related diseases. The estimated loss in annual earnings is based on estimates of the proportion of the population in the labor force, estimates of annual mean earnings from the BLS, and estimates of the mean value of housekeeping services. The estimated value of lost housekeeping services for individuals not in the labor force is 40% of average earnings for people of similar age and sex who are in the work force and 20% of annual earnings for individuals in the labor force. Estimates of the present value of lifetime future earnings are based on human capital methodology, which assumes that earnings reflect the contribution workers make to the value of goods and services and that the present value of expected future earnings is an estimate of the value of human capital[23]. The mortality-related productivity loss estimate incorporates both the number and timing of premature deaths attributable to diabetes.

Using 2001 earnings estimates from the Bureau of Labor Statistics, we updated the present value of future earnings (PVFE) estimates from Haddix et al.[24]. The PVFE for 2002, including unpaid household work, was estimated assuming a 4% real discount rate. The average PVFE estimate for all diabetes-attributed mortality cases is $116,928, although the actual cost estimate differs by age and sex.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....