S C I E N T I F I C
`
`S T A T E M E N T
`
`Economic Costs of Diabetes in the U.S.
`in 2012
`
`AMERICAN DIABETES ASSOCIATION
`
`OBJECTIVEdThis study updates previous estimates of the economic burden of diagnosed
`diabetes and quantifies the increased health resource use and lost productivity associated with
`diabetes in 2012.
`RESEARCH DESIGN AND METHODSdThe study uses a prevalence-based approach
`that combines the demographics of the U.S. population in 2012 with diabetes prevalence, ep-
`idemiological data, health care cost, and economic data into a Cost of Diabetes Model. Health
`resource use and associated medical costs are analyzed by age, sex, race/ethnicity, insurance
`coverage, medical condition, and health service category. Data sources include national surveys,
`Medicare standard analytical files, and one of the largest claims databases for the commercially
`insured population in the U.S.
`RESULTSdThe total estimated cost of diagnosed diabetes in 2012 is $245 billion, including
`$176 billion in direct medical costs and $69 billion in reduced productivity. The largest com-
`ponents of medical expenditures are hospital inpatient care (43% of the total medical cost),
`prescription medications to treat the complications of diabetes (18%), antidiabetic agents and
`diabetes supplies (12%), physician office visits (9%), and nursing/residential facility stays (8%).
`People with diagnosed diabetes incur average medical expenditures of about $13,700 per year, of
`which about $7,900 is attributed to diabetes. People with diagnosed diabetes, on average, have
`medical expenditures approximately 2.3 times higher than what expenditures would be in the
`absence of diabetes. For the cost categories analyzed, care for people with diagnosed diabetes
`accounts for more than 1 in 5 health care dollars in the U.S., and more than half of that expen-
`diture is directly attributable to diabetes. Indirect costs include increased absenteeism ($5 billion)
`and reduced productivity while at work ($20.8 billion) for the employed population, reduced
`productivity for those not in the labor force ($2.7 billion), inability to work as a result of disease-
`related disability ($21.6 billion), and lost productive capacity due to early mortality ($18.5
`billion).
`CONCLUSIONSdThe estimated total economic cost of diagnosed diabetes in 2012 is $245
`billion, a 41% increase from our previous estimate of $174 billion (in 2007 dollars). This
`estimate highlights the substantial burden that diabetes imposes on society. Additional compo-
`nents of societal burden omitted from our study include intangibles from pain and suffering, re-
`sources from care provided by nonpaid caregivers, and the burden associated with undiagnosed
`diabetes.
`
`D iabetes imposes a substantial bur-
`
`den on the economy of the U.S. in
`the form of increased medical costs
`and indirect costs from work-related ab-
`senteeism, reduced productivity at work
`
`Diabetes Care 36:1033–1046, 2013
`
`and at home, reduced labor force partic-
`ipation from chronic disability, and pre-
`mature mortality (1,2). In addition to the
`economic burden that has been quanti-
`fied, diabetes imposes high intangible
`
`c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c
`
`This report was prepared under the direction of the American Diabetes Association by Wenya Yang (The Lewin
`Group, Inc., Falls Church, Virginia); Timothy M. Dall (IHS Global Inc., Washington, DC); Pragna Halder
`(The Lewin Group, Inc.); Paul Gallo (IHS Global Inc.); Stacey L. Kowal (IHS Global Inc.); and Paul F. Hogan
`(The Lewin Group, Inc.).
`Address correspondence to Matt Petersen, American Diabetes Association, 1701 N. Beauregard Street, Alex-
`andria, VA 22311. E-mail: mpetersen@diabetes.org.
`DOI: 10.2337/dc12-2625
`This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/suppl/doi:10
`.2337/dc12-2625/-/DC1.
`© 2013 by the American Diabetes Association. Readers may use this article as long as the work is properly
`cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.org/
`licenses/by-nc-nd/3.0/ for details.
`See accompanying commentary, p. 775.
`
`costs on society in terms of reduced qual-
`ity of life and pain and suffering of people
`with diabetes, their families, and friends.
`Improved understanding of the eco-
`nomic cost of diabetes and its major
`determinants helps to inform policymakers
`and to motivate decisions to reduce di-
`abetes prevalence and burden. The pre-
`vious cost of diabetes study by the
`American Diabetes Association (ADA) esti-
`mated that there were nearly 17.5 million
`people living in the U.S. with diagnosed
`type 1 or type 2 diabetes in 2007, at an
`estimated cost of $174 billion in higher
`medical costs and lost productivity (2).
`The percentage of the population
`with diagnosed diabetes continues to
`rise, with one study projecting that as
`many as one in three U.S. adults could
`have diabetes by 2050 if current trends
`continue (3). In this updated cost of di-
`abetes study, we estimate the total na-
`tional economic burden of diagnosed
`diabetes in 2012 reflecting continued
`growth in prevalence of diabetes and its
`complications; changing health care prac-
`tices, technology, and cost of treatment;
`and changing economic conditions.
`
`RESEARCH DESIGN AND
`METHODSdThis study follows the
`methodology used in the 2002 and 2007
`costs of diabetes studies by the ADA, with
`modifications to refine the analyses
`where appropriate (1,2). A prevalence-
`based approach is used to estimate the
`medical costs by demographic group,
`health service category, and medical con-
`dition. One difference from earlier studies
`is that for some analyses we now include
`race/ethnicity as a demographic dimen-
`sion. We analyze the prevalence of diag-
`nosed diabetes, utilization and costs
`attributable to diabetes by age-group (un-
`der 18, 18–34, 35–44, 45–54, 55–59,
`60–64, 65–69, and over 70 years of age),
`sex, race/ethnicity (non-Hispanic white,
`non-Hispanic black, non-Hispanic other,
`and Hispanic), and insurance status (pri-
`vate; government including Medicare,
`Medicaid, Children’s Health Insurance
`Program, and other government-sponsored
`coverage; and uninsured). State-specific es-
`timates of prevalence and costs are pro-
`vided in Supplementary Table 11.
`
`care.diabetesjournals.org
`
`DIABETES CARE, VOLUME 36, APRIL 2013
`
`1033
`
`PROTECTIVE ORDER MATERIAL
`
`Page 1
`
`

`
`Scientific Statement
`
`Major data sources analyzed include
`National Health Interview Survey (NHIS),
`American Community Survey (ACS), Be-
`havioral Risk Factor Surveillance System
`(BRFSS), Medical Expenditure Panel Sur-
`vey (MEPS), OptumInsight’s de-identified
`Normative Health Information database
`(dNHI), the Medicare 5% sample Standard
`Analytical Files (SAFs), Nationwide Inpa-
`tient Sample (NIS), National Ambulatory
`Medical Care Survey (NAMCS), National
`Hospital Ambulatory Medical Care Survey
`(NHAMCS), National Nursing Home Sur-
`vey (NNHS), National Home and Hospice
`Care Survey (NHHCS), and Current Pop-
`ulation Survey (CPS). We use the most re-
`cent year’s data available for each of these
`data sources, though for certain analyses
`we combine 3 years of data to achieve suf-
`ficient sample size. To estimate medical
`costs for less common health service cate-
`gories such as hospital inpatient care, emer-
`gency care, home health, and podiatry, we
`combine 5 years of MEPS data to reduce
`variance in utilization and cost. The demo-
`graphics of the U.S. population in 2012
`with diabetes prevalence, epidemiological
`data, health care cost, and economic data
`are then combined into a Cost of Diabetes
`Model. Supplementary Table 1 describes
`how these data sources are used, along
`with their respective strengths and limita-
`tions, pertinent to this study. All cost and
`utilization estimates are extrapolated to the
`projected U.S. population in 2012 (4), with
`cost estimates calculated in 2012 dollars
`using the appropriate components of the
`medical consumer price index or total con-
`sumer price index (5).
`
`Estimating the size of the population
`with diabetes
`To estimate the number of people with
`diagnosed diabetes in 2012 we combined
`U.S. Census Bureau population numbers
`with estimated prevalence of diabetes by
`age-group, sex, race/ethnicity, insurance
`coverage, and whether residing in a nurs-
`ing home.
`Combining the 2009, 2010, and
`2011 NHIS data produced a sample
`sufficient to estimate diabetes prevalence
`by demographic and insurance coverage
`(n 5 123,185). Prevalence is based on re-
`spondents answering “yes” to the ques-
`tion, “Have you EVER been told by a
`doctor or health professional that you
`have diabetes or sugar diabetes?” We ex-
`clude gestational diabetes mellitus from
`the prevalence estimates. Previous re-
`search finds that self-report of a physi-
`cian’s diagnosis of diabetes is accurate in
`
`estimating prevalence of diagnosed diabe-
`tes (6).
`For the 2007 cost study, the esti-
`mated prevalence of diagnosed diabetes
`among the institutionalized population
`(24%) came from an analysis of the 2004
`NNHS. There has been no update of the
`NNHS since 2004. Nearly one in three
`(32.8%) nursing home residents has di-
`agnosed diabetes based on a nationally
`representative study that analyzed medi-
`cal charts, minimum dataset records, and
`prescription claims files to identify people
`with diabetes (7). On the basis of this up-
`dated information on diabetes prevalence
`among nursing home residents, we estimate
`age-group–, sex-, and race/ethnicity–
`specific prevalence using the same distri-
`bution of the population demographic
`variables as shown in the 2004 NNHS
`survey data among the 1.6 million nursing
`home residents in 2012. Few data exist
`regarding the prevalence of diabetes
`among the noncivilian population or the
`institutionalized populations other than
`those in nursing homes (e.g., in prisons).
`We assume that the noncivilian popula-
`tion and the institutionalized populations
`other than those in nursing homes have
`diabetes prevalence similar to the nonin-
`stitutionalized population, controlling for
`demographics, based on the limited evi-
`dence available (8,9).
`Combining the NHIS and NNHS
`data, we estimate the prevalence of di-
`agnosed diabetes among population sub-
`groups (by age-group, sex, race/ethnicity,
`and insurance coverage). Supplementary
`Table 3 shows that prevalence of diabetes
`increases with age, is somewhat higher for
`males than for females, and is highest
`among non-Hispanic blacks. Reflecting
`the high prevalence among the elderly
`population, 13.4% of the population
`with government-sponsored medical in-
`surance (e.g., Medicare, Medicaid) has di-
`agnosed diabetes as compared with 4.6%
`among the privately insured and 3.7%
`among the uninsured populations.
`State-specific estimates of diabetes
`prevalence (Supplementary Table 11)
`come from combing the 2010 ACS, the
`2009 and 2010 BRFSS, and the 2004
`NNHS. We applied a statistical matching
`procedure that randomly matches each
`person in the 2010 ACS with a similar
`person either in the BRFSS (if not living
`in a nursing home) or in the NNHS (if
`living in a nursing home). Each noninsti-
`tutionalized person in the ACS is matched
`with a person in the BRFSS in the same
`state, sex, age-group (15 age-groups),
`
`race/ethnicity, household income level
`(eight levels), and insured/uninsured sta-
`tus. Each person in the ACS in a nursing/
`residential facility is matched with a person
`in the NNHS in the same sex, age-group,
`and race/ethnicity. Our state prevalence es-
`timates are slightly different from those re-
`ported by the U.S. Centers for Disease
`Control and Prevention (CDC) for 2010,
`which are based solely on the BRFSS (10).
`
`Estimating the direct medical cost
`attributed to diabetes
`We estimate health resource use among
`the population with diabetes in excess of
`resource use that would be expected in
`the absence of diabetes. Diabetes increases
`the risk of developing neurological, periph-
`eral vascular, cardiovascular, renal, endo-
`crine/metabolic, ophthalmic, and other
`complications (see Supplementary Table
`2 for a more comprehensive list of comor-
`bidities) (2). Diabetes also increases the
`cost of treating general conditions that
`are not directly related to diabetes (2,11–
`13). Therefore, a portion of health care
`expenditures for these medical conditions
`is attributed to diabetes.
`As elaborated in the 2007 study, the
`approach used to quantify the increase in
`health resource use associated with di-
`abetes was influenced by four data limi-
`tations: 1) absence of a single data source
`for all estimates, 2) small sample size in
`some data sources, 3) correlation of both
`diabetes and its comorbidities with other
`factors such as age and obesity, and 4)
`under-reporting of diabetes and its co-
`morbidities in certain data sources. Be-
`cause of these limitations we estimate
`diabetes-attributed costs using one of
`two approaches for each cost component.
`For cost components estimated solely
`from the MEPS (ambulance services,
`home health, podiatry, diabetic supplies,
`and other equipment and supplies), we
`use a simple comparison of annual per
`capita health resource use for people with
`and without diabetes controlling for
`age, sex, and race/ethnicity. For nursing/
`residential facility use (which is not cap-
`tured in the MEPS) and for cost compo-
`nents that rely on analysis of medical
`encounter data (hospital inpatient, emer-
`gency care, and ambulatory visits), we use
`an attributed risk methodology often
`used in disease-burden studies that relies
`on population etiological fractions (2,14).
`Etiological fractions estimate the excess
`use of health care services among the di-
`abetic population relative to a similar
`population that does not have diabetes.
`
`1034
`
`DIABETES CARE, VOLUME 36, APRIL 2013
`
`care.diabetesjournals.org
`
`PROTECTIVE ORDER MATERIAL
`
`Page 2
`
`

`
`contain claims data filed on behalf of
`Medicare beneficiaries under both Part
`A and Part B, and like the dNHI we iden-
`tify people with diabetes based on dia-
`betes ICD-9 diagnosis codes. The large
`size of these two claims databases enables
`the generation of age/sex/setting–specific
`rate ratios for each medical condition,
`which are more stable than rates estimated
`using the MEPS.
`Unlike the MEPS, the dNHI data and
`Medicare 5% claims data do not contain
`race/ethnicity and select patient charac-
`teristics that could affect both patients’
`health status and health seeking behav-
`iors. For the 10 medical conditionsd
`cataract, cellulitis, conduction disorders
`and cardiac dysrhythmias, general medi-
`cal condition, heart failure, hypertension,
`myocardial infarction, other chronic is-
`chemic heart disease, renal failure and
`its sequelae, and urinary tract infectiond
`which are the largest contributors to the
`overall cost of diabetes, we estimated two
`multivariate Poisson regressions, using
`data from the MEPS, to determine the ex-
`tent to which controlling only for age and
`sex might bias the rate ratios. First, we
`estimated a naïve model that produces
`diabetes-related rate ratios for hospital in-
`patient days, emergency visits, and ambu-
`latory visits controlling for age and sex
`only. Then, we estimated a full model
`that includes diabetes status as the main
`explanatory variable and various known
`predictors of health service utilization in-
`cluding age, sex, education level, income,
`marital status, medical insurance status,
`and race/ethnicity as covariates. For the
`full model our focus is not on the relation-
`ship between health care use and the co-
`variates (other than diabetes), but rather
`these covariates are included to control
`for patient characteristics not available
`in medical claims data that could be cor-
`related with both medical conditions and
`health-seeking behavior. The full model
`omits indicators for the presence of co-
`existing conditions or complications of
`diabetes (e.g., hypertension), since in-
`cluding such variables could bias low
`the estimated relationship between diabe-
`tes and health care use for each of the 10
`medical conditions. The rate ratio coeffi-
`cients for the diabetes flag variable in the
`naïve and full models are then compared.
`The findings suggest statistically signifi-
`cant overestimates of the rate ratios for
`emergency visits when using the naïve
`model for five condition categories. For
`inpatient days, we found significant over-
`estimates in the rate ratios for three
`
`American Diabetes Association
`
`condition categories. For ambulatory vis-
`its, only hypertension was found to have a
`significantly higher rate ratio by compar-
`ing the MEPS-based naïve model and the
`full model.
`To remedy the relative risk overesti-
`mation for these condition categories, we
`scaled the rate ratios estimated from
`dNHI and Medicare 5% sample using
`the regression results from the MEPS
`analysis by applying a scalar (with the
`scalar calculated as the full model rate
`ratio divided by the naïve model rate ra-
`tio) (2). For emergency department visits,
`claims-based rate ratios were scaled down
`for myocardial infarction (scale 5 0.94),
`other chronic ischemic heart disease
`(0.93), hypertension (0.71), cellulitis
`(0.72), and renal failure (0.95). For inpa-
`tient days, claims-based rate ratios were
`scaled down for hypertension (0.62), cel-
`lulitis (0.93), and renal failure (0.90).
`Physician office visits were scaled down
`for hypertension (0.89). We did not
`find a significant overestimate of the rate
`ratios for general medical conditions for
`any of the three health service delivery
`settings comparing the MEPS-based naïve
`model and the full model. However, a
`comparison of the claims-based rate ratios
`with the rate ratios calculated from the
`MEPS-based naïve model found that the
`claims-based rate ratios for general condi-
`tions were significantly higher than the
`MEPS-based rate ratios for emergency
`department visits, hospital
`inpatient
`days, and ambulatory visits, respectively.
`Therefore, to be conservative in our cost
`estimates, we downward adjusted claims-
`based rate ratios for emergency department
`visits (0.70), hospital inpatient days (0.68),
`and ambulatory visits (0.66) for the general
`condition group by applying a scalar calcu-
`lated as the MEPS-based naïve model rate
`ratio divided by the claims-based rate ratio.
`Estimates of health resource use at-
`tributed to diabetes were combined with
`estimates of the average medical cost per
`event, in 2012 dollars, to compute total
`medical costs attributed to diabetes. For
`hospital inpatient days, office visits, emer-
`gency visits, and outpatient visits, we use
`average cost per visit/day specific to the
`medical conditions modeled. We com-
`bined the 2008–2010 MEPS files to esti-
`mate the average cost per event, except
`that for less common conditions or cost
`categories we combined the 2006–2010
`MEPS files to obtain a larger sample and
`thereby produce more precise cost esti-
`mates. Although the MEPS contains
`both inpatient facility and professional
`
`Both approaches are equivalent under a
`reasonable set of assumptions, but the
`first approach cannot be used with some
`national data sources analyzed (e.g., NIS)
`that are visit/hospital discharge level files,
`which might or might not identify the pa-
`tient as having diabetes even if the patient
`does indeed have diabetes (2,14).
`The attributable fraction approach
`combines etiological fractions («) with to-
`tal projected U.S. health service use (U) in
`2012 for each age-group (a), sex (s), med-
`ical condition (c), and care delivery set-
`ting (H)dhospital inpatient, emergency
`departments, and ambulatory visits (phy-
`sician office visits combined with hospital
`outpatient/clinic visits):
`
`Attributed health resource useH
`5 ∑
`∑
`«H;a;s;c 3 UH;a;s;c
`age
`medical
`condition
`
`∑ s
`
`ex
`
`The etiological fraction is calculated
`using the diagnosed diabetes prevalence
`(P) and the relative rate ratio (R):
`
`
`RH;a;s;c 2 1
`«H;a;s;c5 Pa;s 3
`
`
`1 1
`Pa;s 3
`RH;a;s;c 2 1
`
`The rate ratio for hospital inpatient
`days, emergency visits, and ambulatory
`visits represents how annual per capita
`health service use for the population
`with diabetes compares to the population
`without diabetes:
`
`RH;a;s;c
`annual per capita use for people with diabetesa;s;c
`5
`annual per capita use for people without diabetesa;s;c
`
`Diabetes and its comorbidities are cor-
`related with other patient characteristics
`(e.g., demographics and body weight). To
`mitigate bias caused by correlation, we
`estimate age/sex/setting–specific etiologi-
`cal fractions for each medical condition.
`The primary data sources for calculating
`etiological fractions are OptumInsight’s
`dNHI data (a consolidation of the Ingenix
`Research Data Mart and MCURE databases
`used in the 2007 study) and the 2010 5%
`sample Medicare SAFs. The dNHI data
`contains a complete set of medical
`claims for over 23 million commercially
`insured beneficiaries in 2011 and allows
`patient records to be linked during the
`year and across health delivery settings.
`This allows us to identify people with a
`diabetes ICD-9 diagnosis code (250.xx)
`in any of their medical claims during the
`year. The Medicare 5% sample SAFs
`
`care.diabetesjournals.org
`
`DIABETES CARE, VOLUME 36, APRIL 2013
`
`1035
`
`PROTECTIVE ORDER MATERIAL
`
`Page 3
`
`

`
`Scientific Statement
`
`expenditures and the NIS contains only
`facility charges (which are converted to
`costs using hospital-specific cost-to-charge
`ratios), the NIS has a much larger sample
`(n 5 ;8 million discharges in 2010) and
`also contains 5-digit diagnosis codes.
`Therefore, we use the 2010 NIS to esti-
`mate inpatient facility costs and the com-
`bined 2008–2010 MEPS to estimate
`the cost for professional services. The av-
`erage costs per event or day by medical
`condition are shown in Supplementary
`Table 4.
`Utilization of prescription medication
`(excluding insulin and other antidiabetic
`agents) for each medical condition is
`estimated from medications prescribed
`during physician’s office, emergency de-
`partment, and outpatient visits attributed
`to diabetes. The average number of med-
`ications prescribed during a visit for each
`age-sex-race stratum was estimated from
`2008–2010 NAMCS and 2007–2009
`NHAMCS data. We calculated the total
`number of people with diabetes that use
`insulin and other antidiabetic agents by
`combining diabetes prevalence and rate
`of use for these antidiabetic agents ob-
`tained from the 2009–2011 NHIS. The
`average cost per prescription filled, insu-
`lin, and oral and other antidiabetic agents
`were obtained from the combined MEPS
`2008–2010. We combined the utilization
`of these medications with the average cost
`per prescription to estimate the cost by
`age, sex, race/ethnicity, and insurance sta-
`tus. The average per capita cost for dia-
`betic supplies by age-sex-race stratum
`was calculated from the MEPS 2008–
`2010. Over-the-counter medications
`were not included owing to the lack of
`data on whether diabetes increases the
`use of such medications.
`Consistent with the 2007 study, total
`nursing/residential facility days attributed
`to diabetes were estimated by combining
`the average length of stay and the nursing/
`residential
`facility population. Using
`2004 NNHS, we calculated the number
`of residents with diabetes in each age-sex
`stratum, which was adjusted using the
`32.8% diabetes prevalence estimate
`among nursing home residents, obtained
`from literature (7). Nursing/residential fa-
`cility use attributed to diabetes was esti-
`mated using an attributable risk approach
`where the prevalence of diabetes among
`residents was compared with the preva-
`lence of diabetes among the overall pop-
`ulation in the same age-sex stratum. The
`analyses were conducted separately for
`short-stay, long-stay, and residential
`
`facility residents to estimate total days of
`care. Similar to the 2007 study, cost per
`day was obtained from a geographically
`representative cost of care survey for
`2012 (15).
`Hospice days attributed to diabetes
`represents a combination of length of stay
`and diabetes prevalence among hospice
`residents. The 2007 NHHCS was used to
`calculate the number of hospice residents
`with diabetes and those that have a pri-
`mary diagnosis of diabetes along with the
`average length of stay for each age-sex-race
`stratum. Cost per resident per day obtained
`from the Hospice Association of America
`was combined with hospice days attributed
`to diabetes to estimate the total cost of
`hospice care attributed to diabetes.
`The 2006–2010 MEPS files were
`combined to increase the sample size to
`analyze the use of home health, podiatry,
`ambulance services, and other equipment
`and supplies. These cost components are
`estimated by comparing annual per capita
`cost for people with and without diabetes,
`controlling for age. Due to small sample
`size, sex and race/ethnicity were not in-
`cluded as a stratum when calculating
`costs per capita.
`
`Estimating the indirect cost
`attributed to diabetes
`The indirect costs associated with diabe-
`tes include workdays missed due to
`health conditions (absenteeism), re-
`duced work productivity while working
`due to health conditions (presenteeism),
`reduced workforce participation due to
`disability, and productivity lost due to
`premature mortality (16–18). Produc-
`tivity loss occurs among those in the
`labor force as well as among the nonem-
`ployed population. To estimate the
`value of lost productivity, we calculate
`the number of missed workdays result-
`ing from absenteeism, reduced work
`productivity due to presenteeism, work-
`force participation reductions associated
`with chronic disability, and work years
`lost resulting from premature mortality
`associated with diabetes. This approach
`mirrors the one used in the 2007 study,
`with the exception of adding race/ethnicity
`as a dimension. More recent data sources
`were used with per capita productivity
`loss calculated by combining the estimates
`derived from the 2009–2011 NHIS and
`the average annual earnings from the
`2011 CPS. Earnings were inflated to
`2012 dollars using the overall consumer
`price index, and per capita estimates
`were applied to the number of people
`
`with diabetes by age-group, sex, and
`race/ethnicity.
`
`c Absenteeism is defined as the number
`of workdays missed due to poor health,
`and prior research finds that people
`with diabetes have higher rates of ab-
`senteeism than the population without
`diabetes (16–18). Estimates of excess
`absenteeism associated with diabetes
`range from 1.8 to 7% of total workdays
`(17,19–22). Ordinary least squares re-
`gression with the 2009–2011 NHIS
`shows that self-reported annual missed
`workdays are statistically higher for
`people with diabetes. Control variables
`include age-group, sex, race/ethnicity,
`diagnosed hypertension status (yes/no),
`and body weight status (normal, over-
`weight, obese, unknown). Diabetes is
`entered as a dichotomous variable (di-
`agnosed diabetes 5 1; otherwise 0), as
`well as an interaction term with age-
`group. Controlling for hypertension and
`body weight produces more conserva-
`tive estimates of the diabetes impact on
`absenteeism as comorbidities of diabetes
`are correlated with body weight status
`and a portion of hypertension is attrib-
`uted to diabetes. Workers with diabetes
`average three more missed workdays
`than their peers without diabetes, with
`excess missed workdays varying by de-
`mographic group.
`c Presenteeism is defined as reduced
`productivity while at work, and is
`generally measured through worker
`responses to surveys. These surveys rely
`on the self-reported inputs on the
`number of reduced productivity hours
`incurred over a given time frame. Mul-
`tiple recent studies report
`that
`in-
`dividuals with diabetes display higher
`rates of presenteeism than their peers
`without diabetes (19,21,22). The rate
`of presenteeism among the population
`with diabetes exceeds rates for their
`colleagues without diabetesdwith the
`excess rates ranging from 1.8 to 38%
`of annual productivity (17,19–22).
`These estimates comparing presen-
`teeism for employees with diabetes
`versus those without diabetes, how-
`ever, fail to control for other factors
`that may be correlated with diabetes
`(e.g., age and weight status). Conse-
`quently, we model productivity loss
`associated with diabetes-attributed pre-
`senteeism using the estimate (6.6%)
`from the 2007 study that controls for
`the impact of factors correlated with
`diabetes (2).
`
`1036
`
`DIABETES CARE, VOLUME 36, APRIL 2013
`
`care.diabetesjournals.org
`
`PROTECTIVE ORDER MATERIAL
`
`Page 4
`
`

`
`c Inability to work associated with di-
`abetes is estimated using a conservative
`approach that focuses on unemploy-
`ment related to long-term disability.
`The CDC estimates
`that
`roughly
`65,700 lower-limb amputations are
`performed each year on people with
`diabetes (23). These amputations and
`other comorbidities of diabetes can
`make it difficult
`for some people
`with diabetes to remain in the work-
`force or to find employment in their
`chosen profession (22,24). To quantify
`diabetes-related disability, we identify
`people in the 2009–2011 NHIS be-
`tween ages 18 and 65 years who receive
`Supplemental Security Income (SSI)
`payments for disability. Using logistic
`regression, we estimate the relationship
`between diabetes and the receipt of
`SSI payments controlling for age-group,
`sex, race/ethnicity, hypertension, and
`weight. The results of
`this analysis
`suggest that people with diabetes have a
`2.4 percentage point higher rate of be-
`ing out of the workforce and receiving
`disability payments compared with
`their peers without diabetes. The di-
`abetes effect increases with age and
`varies by demographicdranging from
`0.7 percentage points for non-Hispanic
`white males aged 65–69 years to 7.4
`percentage points
`for non-Hispanic
`black females aged 55–59 years. Mod-
`eling disability-related unemployment
`is a conservative approach to modeling
`the employment effect of diabetes; re-
`gression analysis of the NHIS suggests
`that people with diabetes have actual
`labor force participation rates averag-
`ing approximately 10 percentage points
`lower than their peers without diabetes.
`The average daily earnings for those in
`the workforce are used as a proxy for
`the economic impact of reduced em-
`ployment due to chronic disability. SSI
`payments are considered transfer pay-
`ments and therefore are not included in
`the social cost of not working due to
`disability.
`c Reduced productivity for those not
`in the workforce is included in our
`estimate of the national burden. This
`population includes all adults under 65
`years of age who are not employed
`(including those voluntarily or
`in-
`voluntarily not in the workforce). The
`contribution of people not
`in the
`workforce to national productivity in-
`cludes time spent providing child care,
`household activities, and other ac-
`tivities such as volunteering in the
`
`Table 1dHealth resource use in the U.S. by diabetes status and cost component, 2012
`(in millions of units)
`
`American Diabetes Association
`
`Population with diabetes
`
`Attributed
`to diabetes
`
`Incurred by
`people with
`diabetes
`
`% of U.S.
`total
`
`% of U.S.
`total
`
`Units
`
`Units
`
`Incurred by
`population
`without
`diabetes
`
`U.S.
`total*
`
`26.4
`
`15.7%
`
`43.1
`
`25.7%
`
`124.9
`
`168.0
`
`101.3
`0.2
`
`16.4% 198.4
`0.3%
`9.3
`
`32.2%
`12.8%
`
`85.7
`7.3
`7.8
`25.7
`361.4
`
`8.3% 174.0
`5.7%
`15.3
`7.8%
`15.0
`9.2%
`64.9
`11.8% 673.1
`
`16.9%
`11.9%
`14.9%
`23.2%
`22.1%
`
`418.0
`63.1
`
`852.8
`113.5
`85.6
`214.7
`2,377.9
`
`616.4
`d
`1,026.7
`128.7
`100.7
`279.7
`72.4
`3,051.1
`
`Health resource
`
`Institutional care
`Hospital inpatient days
`Nursing/residential facility
`days
`Hospice days
`Outpatient care
`Physician office visits
`Emergency department visits
`Hospital outpatient visits
`Home health visits
`Medication prescriptions
`
`Data sources: NIS (2010), NNHS (2004), NAMCS (2008–2010), NHAMCS (2007–2009), MEPS (2006–
`2010), and NHHCS (2007). *Numbers do not necessarily sum to totals because of rounding.
`
`community. Prior estimates of reduced
`productivity for those not
`in the
`workforce were based on estimates of
`“bed days” (which is defined as a day
`spent in bed because of poor health).
`The NHIS no longer collects data on
`bed days. Therefore, we use per capita
`absenteeism estimates for the working
`population as a proxy for reduced
`productivity days among the non-
`employed population in a similar
`demographic. Whereas each work-
`day lost due to absenteeism is based
`on estimated average daily earnings,
`there is no readily available measure
`of the value of a day lost for those not
`
`in the workforce. Studies often use
`minimum wage as a proxy for the
`value of time lost, but this will un-
`derestimate the value of time. Using
`average earnings for their employed
`counterparts will overestimate the
`value of time. Similar to the 2007
`study, we use 75% of the average
`earnings for people in the workforce
`as a productivity proxy for those un-
`der 65 years of age not in the labor
`force (which is close to the midpoint
`between minimum wage and the av-
`erage hourly wage earned by a de-
`mographic similar to the unemployed
`under 65 years of age).
`
`Table 2dHealth resource use attributed to diabetes in the