Economic Costs Attributed to Diagnosed Diabetes in Each U.S. State and the District of Columbia: 2021 Free

Diabetes is a chronic disease associated with serious complications, including heart disease, kidney failure, nerve damage, and other health issues (1). In 2021, 11.6% of the U.S. population had diabetes (2). Furthermore, diabetes imposes substantial health care costs, ranking first among 155 health conditions in terms of expenditure in 2013 (3). Indirect costs from lost productivity are also significant factors contributing to the diabetes economic burden. The total cost of diabetes reached $327 billion in the U.S. in 2017, escalating to $413 billion by 2022, showing an upward trend over the last two decades (4,5).

In 2016, the Centers for Disease Control and Prevention (CDC) released the Diabetes State Burden Toolkit reporting the health and economic burdens of diabetes for each state and the District of Columbia (DC; https://nccd.cdc.gov/Toolkit/DiabetesBurden). The toolkit reports state-specific data on the health, economic, and mortality burdens of diabetes. Estimates reported in the prior toolkit version were based on 2013 data, which had become outdated. The state economic cost of diabetes can change over time because of changes in factors such as demographic composition of individuals with diabetes, prevalence of diabetes, and cost of medical care. Between 2012 and 2022, costs of diabetes increased by 35%, even after adjusting for inflation (5), thus justifying the need for the current analysis. Accurate estimates of state-level economic costs of diabetes can help state policymakers in developing evidenced-based public health interventions in their states to prevent and control diabetes prevalence. We present updated estimates of the state-level economic burden of diagnosed diabetes using established methodology and more recent data (6).

This study fills two others gaps in the literature. First, previous studies reported solely national-level estimates or state-level estimates derived from national data, hindering assessment of disparities in the diabetes burden across states (5,7). Second, payer-specific costs were not estimated. We overcome these limitations by using state-specific demographic information in estimating costs of diabetes and by decomposing the medical costs by payer.

State-specific costs by demographics and payers can provide valuable insights for state public health planners, policymakers, and payers. Understanding the disparities in diabetes burden between and within states based on demographics and payers can be essential for effectively monitoring diabetes, determining the appropriate public health interventions, and allocating resources to prevent and manage diabetes. These diabetes economic burden estimates are also available in the updated online version of the CDC Diabetes State Burden toolkit (https://nccd.cdc.gov/Toolkit/DiabetesBurden).

We used data either from 2019 or 2021, depending on data availability and considerations of disruptions in health care access and delivery resulting from the COVID-19 pandemic. Because 2020 did not accurately represent typical health care use, we excluded data from 2020 when estimating diabetes-related medical costs. We used previously developed methodology (6), calculating direct medical and indirect costs of diagnosed diabetes. Direct costs are health expenditures associated with treating diabetes and its complications and comorbidities. Indirect costs reflect the labor and household productivity losses from diabetes-attributable missed workdays (i.e., absenteeism), on-the-job productivity losses (i.e., presenteeism), household productivity losses, disability that prevents people from working, and premature mortality. We estimated costs in total and per person with diagnosed diabetes. Supplementary Table 2 provides definitions of diabetes for each data source. We applied sample weights from the surveys with complex survey design. All costs are expressed in 2021 U.S. dollars. Additional details on the methods are available in the CDC’s diabetes toolkit technical document, section 2.2 (https://nccd.cdc.gov/Toolkit/DiabetesBurden). To compare our 2021 costs with the 2013 estimates (6), we inflated the 2013 costs to 2021 dollars using the Consumer Price Index.

We used an attributable fraction (AF) approach to estimate the diabetes-attributable state health expenditures by age, sex, payer, and service type: AF = (pd_j × [RR_j − 1])/(1 + pd_j × [RR_j − 1]), where pd represents the diabetes prevalence, RR represents the ratio of medical costs for individuals with and without diabetes, and j indicates age, sex, payer, and service type. We estimated state-level diabetes prevalence by age-group, sex, and payer using the 2021 Behavioral Risk Factor Surveillance data. We computed RRs by payer (Medicare, Medicaid, or other), service type, and age-group using the 2015–2019 Medical Expenditure Panel Survey. Because state-level data were not available to estimate RRs, we applied national-level RRs to state-level diabetes prevalence. We estimated medical costs using a two-part regression model, where the first part was a logit model predicting whether respondents had positive medical spending, and the second part was a generalized linear model with a log link and γ distribution predicting costs among those with positive spending (8). We estimated the RR as the ratio of predicted costs for patients with diabetes to predicted costs assuming no diabetes. Models controlled for age, age squared, sex, race/ethnicity, poverty status, education, and census region.

We multiplied the AF by the projected 2021 State Health Expenditure Accounts (SHEA) estimates (9). SHEA provides information on state-level personal health care based on National Health Expenditure Accounts data and state-level data on health care expenditures from the economic census and other sources. It excludes administrative and net costs of private health insurance (administrative costs include the costs of running various government health care programs, whereas net costs of private health insurance represent the differences between premiums earned and claims or losses incurred for which insurers are liable (10)); government public health activities; and investments in research, structures, and equipment (9). To obtain per-person diabetes-attributable medical costs, we divided total cost by the number of individuals with diabetes in each age-sex category. For Medicare and Medicaid, we calculated costs using the number of beneficiaries with diabetes covered by each plan; for other payers, these costs used all individuals with diabetes, because other costs in SHEA represent costs paid by private payers and other third-party payers and patient out-of-pocket payments.

We applied the previously developed methodology (11) to estimate diabetes-attributable costs for the institutionalized population, because Medical Expenditure Panel Survey does not include this population. We calculated the AF for nursing home costs weighted by Centers for Medicare and Medicaid Services (CMS) Resource Utilization Group (RUG) payments. RUGs represent levels of care intensity in nursing homes (12). We calculated RUG-weighted AFs for each state, age-group, and sex category and multiplied by nursing home expenditures from SHEA. The RUG-weighted AF is the weighted excess diabetes prevalence in nursing homes compared with community prevalence: AF = ([N^D × RUG^D]/[N^D × RUG^D + N^N × RUG^N]) − C^D, where, N^D and N^N are the numbers of long-term stays (nursing home episodes of at least 100 days [12]) for nursing home residents with and without diabetes, respectively. RUG^D and RUG^N are average RUG payments for nursing home residents with and without diabetes, respectively. C^D is the diabetes prevalence in the community. We used the CMS 2018–2019 Minimum Data Set to estimate diabetes prevalence in the nursing home and average RUG payments by diabetes status.

We estimated state-level nursing home diabetes-attributable expenses by multiplying AFs by the state nursing home spending estimates by payer, sex, and age-groups. For state-level nursing home expenditures, we used the Nursing Care Facilities and Continuing Care Communities costs from SHEA, subtracting the continuing care retirement communities portion. We calculated per-person costs as total diabetes-attributable nursing home cost divided by the number of individuals with diabetes in the community plus nursing home residents with diabetes.

We used a human capital approach to estimate diabetes-attributable indirect costs, including morbidity and mortality costs. We calculated costs per person with diabetes by dividing total cost by the sum of the noninstitutionalized individuals and nursing home residents with diabetes.

Morbidity-related diabetes-attributable costs included costs derived from absenteeism, presenteeism, household productivity losses, and inability to work.

We calculated absenteeism costs by multiplying the number of diabetes-attributable missed workdays by the average state-, age-, and sex-specific wages and the estimated number of employed individuals with diabetes in each state. We estimated the number of diabetes-attributable workdays missed per person with diabetes stratified by Census region, age, and sex, using the 2016–2021 National Health Interview Survey (NHIS). With each stratified two-part model, we first used logistic regression to predict the probability of missing work and then a generalized linear model with γ distribution and log link to estimate the number of workdays missed for individuals who had missed work. Models controlled for demographic and socioeconomic factors (age, age squared, race/ethnicity, education, family income, health insurance, and occupation) and comorbidities (arthritis, asthma, cancer, depression, chronic bronchitis, back problems, and pregnancy).

We obtained national mean annual earnings for 2021 from the 2022 Current Population Survey and adjusted them to state wages by applying a state-to-wage ratio calculated from the 2021 Bureau of Labor Statistics estimates. We divided annual mean earnings by 250 workdays to compute daily earnings. The state-level number of employed individuals with diabetes was calculated by multiplying the employed percentage of individuals with diabetes by region from the 2016–2021 NHIS by the corresponding number of individuals with diabetes from the 2021 Behavioral Risk Factor Surveillance.

We estimated state-level diabetes-attributable productivity losses for employed individuals with diabetes by multiplying the average number of presenteeism days lost by daily earnings. Using American Diabetes Association (ADA) estimates, we assumed that 6.6% of annual productivity was lost because of diabetes (5). Days worked were calculated as 250 days minus estimated absenteeism days.

Household productivity losses occur when individuals cannot perform household services. We used the number of diabetes-attributable days spent in bed to estimate diabetes-attributable household productivity losses. Per-person household productivity losses were calculated as the number of diabetes-attributable bed days per year, following the same methodology as that used to estimate workdays lost, multiplied by the daily value of household production. Because of data availability, we used 2014–2018 NHIS for this analysis. We estimated average per-capita monetary value of a day of household production by sex and age-groups from the 2020 dollar value of a day (13), inflated to 2021 dollars. We estimated state-level household productivity losses by applying Bureau of Labor Statistics state-to-national wage ratios.

Individuals who are unable to work because of diabetes-related disability lose the full value of their expected annual earnings. Using the 2016–2021 NHIS, we multiplied the probability of being unable to work because of diabetes (by sex and age-groups) by the number of individuals with diabetes (by state, sex, and age-groups) to estimate the total number of individuals unable to work. We calculated inability to work costs by multiplying the number of individuals unable to work by the state-level mean annual earnings (by sex and age-groups) using the same approach as was used for other morbidity costs.

To estimate mortality costs, we valued premature death resulting from a disease as forgone future productivity (14–16). We multiplied the state-level number of diabetes-attributable deaths by age and sex groups by the present value of lifetime earnings and household productivity costs to calculate total diabetes-attributable mortality cost.

We estimated the number of diabetes-attributable deaths by age and sex groups using an AF approach and combined information on the prevalence of diabetes, the relative risk of death for those with and without diabetes, and the total number of deaths in the population. Relative risk of death was estimated using the 2013–2017 NHIS data linked with mortality data through 2019. Death counts were obtained from the 2021 CDC Wide-Ranging Online Data for Epidemiologic Research.

We calculated the present value of future labor earnings and household production using the same earnings data as were used for other indirect costs. Future costs were discounted by the probability of surviving to each year of age at which the expected production occurs. Compounded survival rates for each age-group were calculated using the 2021 U.S. life tables from the National Vital Statistics Report (17). We adjusted present values so that losses were applied only to the populations expected to incur those losses. We applied a 1% annual productivity growth rate and a 3% annual discount rate (14).

In 2021, 30 million (11.6%) U.S. adults were estimated to have diabetes. Total diabetes-attributable cost was $640 billion, with $335 billion in direct medical and $305 billion in indirect costs (Table 1). The total state-level diabetes-attributable cost ranged from $842 million in Wyoming to $80.9 billion in California, with a median cost of $8.2 billion (Fig. 1). Medical costs accounted for the largest portion of total cost in Vermont (61%) and the lowest portion in California (45%). All costs were highest in California; medical and morbidity costs were lowest in Wyoming, and mortality costs were lowest in Vermont (Supplementary Tables 3–5).

The total cost borne by Medicaid ranged from $64 million in Wyoming to $11.4 billion in California, with a median of $885 million (Fig. 2 and Supplementary Table 3). Costs borne by Medicare ranged from $110 million in Alaska to $12.4 billion in California, with a median of $1.6 billion. Costs borne by other payers ranged from $249 million in Wyoming to $16.4 billion in California, with a median of $2 billion. The proportion of direct medical costs borne by different payers varied across states. On average, 23% (range 9–44%) of medical costs were borne by Medicaid, 35% (range 17–46%) by Medicare, and 42% (range 34–62%) by other payers.

Total morbidity cost was $157.6 billion, ranging from $184 million in Wyoming to $23.1 billion in California, with a median of $1.9 billion (Fig. 1 and Supplementary Table 4). Nationally, indirect costs comprised 184 million lost workdays ($58.8 billion) from presenteeism, 35 million lost workdays ($11.0 billion) from absenteeism, 201 million lost days ($11.8 billion) of household production, and 993,000 individuals ($76 billion) unable to work because of diabetes (Supplementary Table 6). Total mortality cost was $147.8 billion (288,000 deaths), ranging from $181 million (381 deaths) in Vermont to $17.6 billion in California (27,000 deaths), with a median of $1.7 billion (Fig. 1 and Supplementary Tables 5 and 6).

Total annual diabetes-attributable cost per person with diabetes ranged from $17,452 in Idaho to $37,090 in DC, with a median of $21,082 (Fig. 3). Per-person costs for all components were the highest in DC ($17,911 for medical costs, $9,201 for morbidity costs, and $9,977 for mortality costs). The lowest per-person cost was in Utah ($8,656) for medical costs, in Mississippi ($3,921) for morbidity, and in Nebraska ($3,742) for mortality. Per-person direct medical costs by payer also varied by state (Supplementary Table 3). Per-person medical costs borne by Medicaid ranged from $6,192 in Tennessee to $15,430 in Minnesota, with a median of $9,119. Per-person medical costs borne by Medicare ranged from $6,464 in Hawaii to $10,588 in DC, with a median of $7,987. Per-person medical costs borne by other payers ranged from $3,433 in Arizona to $8,698 in Alaska, with a median of $5,054.

Between 2013 and 2021, state-level total cost of diabetes increased by a median of 28% across states, ranging from a 13% increase in New Hampshire to a 62% increase in Kentucky (Supplementary Table 7). Total medical cost of diabetes increased by a median of 50%, ranging from a 33% increase in New Hampshire to a 79% increase in Kentucky. Medicaid costs increased by a median of 153%, ranging from 41% in Florida to 483% in Kentucky. Medicare costs increased by a median of 53%, ranging from 28% in DC to 104% in Nevada. Other payers’ costs increased by a median of 20%, ranging from a 14% reduction in DC to a 71% increase in South Dakota. Total indirect cost of diabetes increased by a median of 14%, ranging from a 6% decrease in New Hampshire to a 47% increase in Kentucky.

Between 2013 and 2021, medical cost per person with diabetes increased by a median of 25%, ranging from 15% in Rhode Island to 41% in South Dakota (Supplementary Table 8). Morbidity costs per person with diabetes decreased across the states by a median of 9%, ranging from a 20% decrease in Nevada to a 2% increase in Wisconsin. Mortality costs per person with diabetes increased by a median of 4%, ranging from an 11% reduction in Massachusetts and Rhode Island to a 34% increase in New Mexico.

We provide the latest estimates of diabetes-attributable economic costs in each state and DC. State-level economic costs of diabetes in 2021 were substantial, ranging from $842 million to $81 billion for total economic cost, from $453 million to $40 billion for direct medical costs, and from $390 million to $41 billion for indirect costs. Total diabetes-attributable cost increased by a median of 33%, ranging from 16% to 68% across states between 2013 and 2021.

State differences in the number of individuals with diabetes, which is a product of total population and diabetes prevalence, and per-person diabetes costs explain the variation in total cost of diabetes across states. States with larger populations had a greater number of individuals with diabetes. For example, California is the most populous state; it therefore had the greatest number of individuals with diabetes and the highest diabetes-attributable costs. Diabetes prevalence also varied across states (range 7–14%), which can be explained by state-level differences in demographic composition, socioeconomic status, neighborhood characteristics and physical environment, food environment, prevalence of health behavior risk factors, and chronic prevention efforts (18,19).

Per-person diabetes costs also varied substantially across states, ranging from $17,452 in Idaho to $37,090 in DC. These differences result from differences in direct medical and indirect costs. Per-person medical costs are determined by the prices of individual health care services and the number of services used, both of which vary across states (20). Different sociodemographic compositions of states partially explain the variation in prices and use. For example, older age and higher personal income result in increased demand, higher prices, and substantial use (21). Furthermore, differences in health status, access to and quality of care, health care systems and policies, insurance regulations, and payer mix also contribute to state-level variation in costs (20–23). For example, health care spending is higher in states with significant Medicare and Medicaid enrollment and larger provider supply (24). Similar factors, including socioeconomic characteristics, health profiles, and policy and public health initiatives, drive variation across states in indirect costs of diabetes. States with higher earnings likely have higher indirect costs; states with more comprehensive diabetes prevention and management programs likely have better disease control and potentially fewer missing workdays, with lower indirect costs.

The state-level variation in economic costs of diabetes could reflect disparities in risk factors of diabetes in state populations, diabetes prevention and management practices, access to and quality of diabetes care, insurance coverage policies, and payer mix. Future studies could examine reasons that contributed to the variation in costs across states and identify areas where interventions are needed to address the disparities and improve clinical and economic outcomes of diabetes. Our findings could also assist decision makers in establishing health research and intervention programs to reduce the economic burden of diabetes in their states.

The drivers of cost growth between 2013 and 2021 and its state-level variation are similar to those that explain variation in state-level costs in 2021. By cost component, increases in direct medical costs were the biggest driver of the cost growth increase (range 33–79%). Increasing costs reflect an increasing number of individuals with diabetes, resulting from growing populations and diabetes prevalence and higher per-person diabetes costs.

Total cost increased relatively the most in Kentucky because, even though its per-person costs increased similarly to other states, its diabetes prevalence had the highest relative increase (from 9.7% to 12.3%). Per-person costs increased similarly in New Hampshire, but with the smallest relative increase in total cost, because it had the largest reduction in diabetes prevalence (from 8.1% to 7.3%).

Differences in changes in the demographic and risk factor profiles in states over time as well as differences in income and consumer prices have been reported to partially explain the variation in health care spending growth across states between 2000 and 2019 (25). Furthermore, differences in state-level health care policies, specifically, variation in the expansion of Medicaid (which was implemented in 34 states, including DC, between 2013 and 2019), have affected changes in health spending across the states (25–27). We found that Medicaid experienced the highest cost increase (range 41–483% across states). Medicaid expansions have been found to increase coverage, service use, quality of care, and Medicaid spending, particularly in adults with diabetes (26,28). Variation in adoption of new technologies, which are often more costly, might be another contributing factor to differences in diabetes cost growth across states (29). These factors also affect state-level variation in changes in indirect costs. For example, improved access to and quality of care result in improved morbidity and mortality outcomes, thus reducing productivity losses. We found that in all but two states, per-person morbidity costs decreased between 2013 and 2021, which was a result of a smaller number of individuals reporting being unable to work because of disability.

Our total and per-person medical costs were comparable to recent ADA estimates (adjusted to 2021 U.S. dollars for comparison): $335 billion in our study versus the ADA’s $284 billion for total cost and $11,199 per person with diabetes in our study versus $11,145 per person with diabetes from the ADA (5). However, consistent with the previous analysis (6), our indirect costs were higher than the ADA’s ($305 billion in our study vs. ADA’s $99 billion). Several factors may explain these differences. First, when estimating diabetes-attributable absenteeism, the ADA controlled for hypertension and body weight, but we did not, resulting in potentially higher estimates in our case. Second, in estimating the number of individuals with diabetes unable to work, the ADA used Social Security Income Disability program participation, whereas we relied on self-reported disability status from the NHIS. Consequently, we identified a larger population of individuals with diabetes unable to work. The ADA’s approach was noted to likely underestimate the costs associated with the inability to work because of diabetes (5). Third, our estimated lifetime earnings used for mortality costs were higher than the ADA’s. Finally, our mortality costs included household productivity losses, in addition to labor productivity losses. Increases in the economic burden of diabetes between 2013 and 2021 in our study (31%) were also comparable to the ADA’s (25%). We both reported a decrease in diabetes per-person indirect costs during that period.

Our study is subject to several limitations. First, ideally, we would have both the prevalence of diabetes and RRs at the state level and by payer to reflect the differential burden on different payer types at the state level. However, because of a lack of available state-level data, payer-specific RRs were estimated at the national level. This may have biased our cost estimates, depending on the degree of state variation in the RRs. Similarly, we applied regional estimates for some components, such as morbidity-related work loss. Consequently, these estimates may not fully capture state-specific variation. Second, our findings may have underestimated total cost. We did not estimate nonmedical direct costs, such as patient transportation or caregiver costs. We did not account for the indirect costs associated with the pain and suffering endured by patients and families. Third, our estimates of the population with diabetes were based on individuals with diagnosed diabetes, and ∼23% of diabetes cases are likely to be undiagnosed. Fourth, the lack of information in our data sources to distinguish between diabetes types prevented us from estimating costs for type 1 and type 2 diabetes separately. Finally, our estimates do not reflect the impact of the COVID-19 pandemic on individuals with diabetes. In 2020, COVID-19 increased mortality and severely disrupted health care and society, including shutdowns, stay-at-home orders, and delays in care. COVID-19 disproportionally affected individuals with diabetes, because they have high risks of complications and mortality associated with COVID-19 (30). To some extent, the issues caused by COVID-19 persisted into 2021 as mortality levels remained high but the disruptions alleviated as society and health care systems adjusted. Because 2020–2021 did not accurately represent typical health care use because of disruptions in health care access and delivery, we omitted data from 2020 to 2021 for estimates of diabetes-related health care use and costs.

In conclusion, our results provide updated estimates of economic costs of diagnosed diabetes in each U.S. state and DC. Diabetes carries a substantial economic burden across the nation. Over the last decade, the economic costs of diabetes, specifically direct medical costs, have continued to rise as both the number of individuals with diabetes and the per-person costs of diabetes climb. Our updated estimates may help policy makers, especially those at the state, regional, and local levels, to develop evidenced-based public health interventions to mitigate the health and economic burdens of diabetes.

Funding. This project and publication were supported by CDC contract 75D30122P14019.

The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the CDC.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. O.A.K.contributed to the development of the technical approach for the study and supervised all components of the analysis. M.S. conducted the analysis of direct medical costs of diabetes. S.R.D. conducted the analysis of the indirect costs of diabetes. S.J.N. contributed to the development of the technical approach and guided the analysis of costs of diabetes among nursing home residents. T.J.H. advised on the development of the technical approach for the study and provided subject matter expertise. P.C. contributed to the development of the technical approach and reviewed analysis results. K.M. reviewed analysis results. P.Z. conceived the study and supervised development of the technical approach. All authors contributed to the development and critical review and revision of manuscript content. O.A.K. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Steven E. Kahn and Neda Laiteerapong.