Lifetime Medical Spending Attributed to Incident Type 2 Diabetes in Medicare Beneficiaries: A Longitudinal Study Using 1999–2019 National Medicare Claims

Diabetes affected 38.4 million people, or 11.6% of the U.S. population, in 2021 and imposed a considerable economic burden on households, the health care system, and society (1–6). More than 90% of diabetes cases are type 2 diabetes (T2D), which is largely preventable. Lifestyle intervention is effective in preventing T2D among people who are at high risk of developing T2D (7–11).

The prevalence of diabetes in older adults is both high and growing (6). In 2021, 13.8 million adults aged ≥65 years (24.4%) had diagnosed diabetes (6). Medicare spent $42 billion more in 2016 for beneficiaries with diabetes than it would have spent if those beneficiaries did not have diabetes (12). To reduce the health and economic burden of diabetes among Medicare beneficiaries, the Centers for Medicare & Medicaid Services established the Medicare Diabetes Prevention Program (MDPP), which offers a structured behavior change intervention to Medicare beneficiaries who are at high risk of developing T2D (13). Medicare began to cover MDPP services in 2018, and an estimated 16 million Americans aged ≥65 years are eligible for the MDPP (14). Medicare covered up to $705 per person for the MDPP in 2022, with the amount depending on the number of sessions attended and weight loss achieved by the participant (15).

Whether the MDPP as implemented in the real world is cost-effective remains unclear. Accurately assessing the cost-effectiveness of the MDPP requires information on the potential economic benefits of the program. A key component for this calculation is lifetime incremental medical spending for T2D from the time of diagnosis to death, which measures the additional medical care resources consumed due to treating T2D and its related morbidities. This measure considers the extra medical spending while a person is alive and the reduction in spending caused by premature death due to diabetes. This measure can serve as a proxy for the economic benefit of preventing a case of T2D.

Lifetime medical spending for T2D in U.S. adults has been estimated using various methodologies and data sources (16–18). These studies relied on simulation models or survey data that may limit the accuracy of their estimates due to assumptions and varying data sources. Our study addresses these issues by using real-world evidence and a longitudinal approach to provide more reliable estimates of lifetime medical spending for T2D. The objective of our study was to estimate the national lifetime medical spending attributed to incident T2D in Medicare beneficiaries, segmented by age at diagnosis, sex, and race/ethnicity and leveraging the extensive multiyear 100% Medicare fee-for-service claims database.

We used the 1999–2019 100% Medicare fee-for-service claims data from the Centers for Medicare & Medicaid Services Chronic Conditions Data Warehouse (19). Various parts of Medicare cover different types of health services, including Part A for inpatient hospital stays, skilled nursing facilities, hospice care, and some home health care; Part B for outpatient care, physician services, medical supplies, and preventive services; and Part D (optional) for prescription drugs. Adding these parts of Medicare claims data gives the total spending for a beneficiary. We included only beneficiaries with fee-for-service plans with all parts (A, B, and D), as cost data for those enrolled in Medicare Advantage plans and prescription cost data for those who are not enrolled in Part D are not available.

Medicare beneficiaries with T2D who were newly diagnosed during 2001–2003 were selected as the incident T2D cohort and followed until disenrollment, death, or the end of 2019. We used a validated algorithm to identify people with diabetes using ICD codes based on inpatient and outpatient claims (1,20). To be diagnosed with diabetes, patients must have had at least one inpatient claim for diabetes or an outpatient claim for diabetes plus another inpatient/outpatient claim for diabetes within 2 years following the year of the first claim. In addition, beneficiaries should have at least 2 years without claims records of diabetes after they enrolled in Medicare to make sure they had incident diabetes. The year of the first claim of diabetes was then recognized as the diagnosis year or index year. Beneficiaries with type 1 diabetes were excluded. Details of the construction of our incident T2D cohort are provided in the Supplementary Material and have been described previously (1).

A nondiabetes comparison cohort was matched in a 1:1 ratio to the incident T2D cohort using propensity score methods to ensure comparability between groups (21). To qualify for inclusion in the nondiabetes cohort, beneficiaries were required to enroll in the Medicare all-parts fee-for-service plan in the same year the incident beneficiaries in the T2D group were initially diagnosed. These individuals had no diagnosed diabetes from their enrollment until the end of the study period. As we required two separate outpatient records within 3 years of the first record for included beneficiaries with T2D, those who only had one outpatient claim and died soon after this time were considered as not having diabetes and, thus, were included in the nondiabetes cohort (∼10.6% of the nondiabetes cohort). The propensity score was calculated to balance both cohorts in terms of baseline demographic and health status characteristics, thereby minimizing confounding effects. The variables included in the propensity score calculation were age, sex, race/ethnicity, geographical region at baseline, and nondiabetes-related chronic disease comorbidities (Alzheimer disease, asthma, chronic obstructive pulmonary disease, anemia, arthritis, AIDS, liver disease, hypothyroidism, and cancer). Additionally, we incorporated health care utilization and spending variables from the year preceding the baseline to further refine the matching process. These variables included the number of inpatient visits, outpatient visits, emergency department visits, and total health care costs, providing a comprehensive assessment of each beneficiary’s overall health status and health care needs. We included health condition and utilization variables to ensure that members of the nondiabetes cohort were as sick as the T2D cohort, excluding the presence of diabetes. This approach allowed for a more accurate comparison of medical spending attributable solely to the incidence of T2D. Each person in both cohorts was followed until death or reaching the age of 100 years to capture their complete medical spending trajectories.

Lifetime incremental medical spending for diabetes was calculated as the total medical spending of those in the T2D cohort over their lifetime minus the total medical spending of those in the nondiabetes comparison cohort over their lifetime, including all medical spending claimed by fee-for-service plans with all three parts (A, B, and D). The total medical spending of a person with T2D was the aggregated annual spending from the year of T2D diagnosis until death or reaching the age of 100 years. The corresponding spending for a person without diabetes was the aggregated annual spending from the initial index year until death or until reaching the age of 100 years.

We estimated this lifetime spending by several subgroups: age at diagnosis (70, 75, 80, and 85 years), sex (female, male), and race/ethnicity identified by Medicare enrollment (non-Hispanic White, non-Hispanic Black, Hispanic, Asian/Pacific Islander, and American Indian/Alaska Native). We obtained the lifetime spending for each subgroup in four steps. First, we estimated annual medical spending by age and diabetes status, including end-of-life care for those who did not survive the year. Second, we derived age-specific annual survival rates by diabetes status and then multiplied surviving probability by the estimated annual spending from the first step to obtain survival-adjusted annual spending. Third, we aggregated survival-adjusted annual spending over the predicted life expectancy by diabetes status. Fourth, we subtracted the lifetime medical spending estimate for the nondiabetes cohort from the T2D cohort. The calculation details are described in Supplementary Material.

All spending was adjusted for inflation and expressed in 2019 U.S. dollars using the personal health care expenditures deflator (22). Future spending was discounted at 3% per year. SDs for predicted mean annual medical spending were estimated using 100 nonparametric bootstrap iterations. All statistical analyses were conducted using SAS Enterprise Guide software, and statistical significance comparing groups was set at P < 0.05.

We conducted two sensitivity analyses using two different nondiabetes comparison cohorts. First, we created a new nondiabetes comparison cohort by matching only age, race/ethnicity, sex, region, and comorbidities. Second, we created a nondiabetes comparison cohort that included all beneficiaries who never developed diabetes during the follow-up period without matching. Estimates derived from these analyses could provide clearer insight into how different variables contribute to the lifetime incremental medical spending for T2D.

We included 1,057,676 beneficiaries with newly diagnosed T2D and an equal number of beneficiaries without diabetes from the propensity score matching (Supplementary Fig. 1). Table 1 presents the characteristics of study cohorts by diabetes status. The mean age at diagnosis of T2D was 78.9 years, and the mean length of follow-up was 7.7 years. The mean length of follow-up for the matched nondiabetes cohort was 8.1 years. Other characteristics were similar between the T2D and nondiabetes cohorts after propensity score matching. Mortality was higher in the T2D than in the nondiabetes cohort (89.6% vs. 87.0%, respectively).

Figure 1 shows survival-adjusted annual medical spending for the T2D cohort and matched nondiabetes cohort at four ages at diagnosis. For each age at T2D diagnosis, the average annual spending was high in the year of diagnosis and decreased rapidly for the next 2 years. The survival-adjusted annual medical spending decreased gradually year by year. For the matched nondiabetes cohort, the survival-adjusted annual medical spending also decreased with time but at a slower rate than the T2D cohort. For example, the expected survival-adjusted annual spending for a person diagnosed with T2D at age 70 years decreased from $23,152 per year during the year of diagnosis to $10,813 10 years later, whereas the expenses for a comparable person without diabetes declined from $14,826 to $6,495 over that same period.

Table 2 summarizes life expectancy after a diagnosis of T2D, life years lost due to T2D, lifetime medical spending with T2D, and lifetime incremental medical spending attributed to diabetes. Overall, those with T2D had 0.06 (if diagnosed at age 85) to 1.37 (if diagnosed at age 70) years shorter life expectancy compared with those without diabetes. This pattern holds for most subgroups, though in a few subgroups of non-Hispanic Black female, Hispanic, Asian/Pacific Islander, and American Indian/Alaska Native beneficiaries, those with T2D had slightly longer life expectancy than those without T2D.

Regardless of age at diagnosis, the T2D cohort spent considerably more on health care than the matched nondiabetes cohort (Table 2). Those diagnosed with T2D at age 70 years spent $65,587 (36%) more than their counterparts without diabetes over their remaining lifetime. This amount gradually decreased as age at diagnosis increased to $36,200 (39%) for those diagnosed at age 85 years. By subgroup, lifetime incremental medical spending due to T2D ranged from $16,115 among American Indian/Alaska Native males diagnosed at age 80 years to $122,146 among Asian/Pacific Islander males diagnosed at age 70 years. Lifetime incremental medical spending due to T2D was not significantly different between females and males (Fig. 2) (P > 0.05 for each age at diagnosis). Lifetime incremental medical spending was highest for Asian/Pacific Islander beneficiaries, followed by non-Hispanic Black and Hispanic beneficiaries and lowest for American Indian/Alaska Native and non-Hispanic White beneficiaries (Fig. 2).

Removing health care utilization and health care spending during the 2 years preceding the first follow-up year to the propensity score matching increased lifetime excess medical spending attributable to T2D (Supplementary Table 1). It also increased the relative difference to 44–49%. We observed the largest increase among non-Hispanic White women diagnosed at age 70 years ($54,301 in the main analysis vs. $70,277 with fewer variables used for matching).

Comparison of the T2D cohort with the nondiabetes cohort showed that the incremental lifetime spending and life years lost associated with T2D was generally larger ($41,973 for those diagnosed at age 85 years to $95,563 for those diagnosed at age 70 years), with a greater relative difference (49–65%) (Supplementary Table 2).

Our study is the first to estimate the incremental lifetime medical spending among Medicare beneficiaries with incident T2D, stratified by age at diagnosis, sex, and race/ethnicity, using real-world claims data and a longitudinal study design. We found that beneficiaries with T2D had much higher annual medical spending after diagnosis than those without diabetes in the first several years, and this difference then narrowed. Medicare beneficiaries with incident T2D, despite having a shorter life expectancy, accumulated 36–40% more health care spending in their lifetime compared with those without diabetes. The incremental lifetime medical spending attributed to T2D varied by race/ethnicity but little by sex. Our study provides a comprehensive picture of medical spending and survival for Medicare beneficiaries with newly diagnosed T2D and their counterparts without diabetes over the remainder of their lifetime. Our findings of a large incremental lifetime medical spending due to T2D imply that T2D imposes a large economic burden on Medicare and that preventing beneficiaries from getting T2D could result in substantial economic benefits.

Our estimated lifetime incremental medical spending of $65,587 for T2D diagnosed at age 70 years is much higher than that of Zhuo et al. (16), which estimated a lifetime incremental medical spending of $39,975 for a person diagnosed at 65 years (after inflating from 2012 U.S. dollars to 2019 U.S. dollars). There are several differences between the two studies. First, Zhuo et al. used cross-sectional survey data, while we used medical claims data. Second, the sample size in our study was much larger and included >1 million people, while Zhuo et al. included only a few hundred individuals diagnosed at age 65 years. Third, Zhuo et al. did not include certain medical expenses that we captured, such as medical spending at the age of death and expenses of long-term care. Finally, the survival functions differed due to different data sources and study sample sizes. The estimated lifetime incremental health care spending by Leung et al. (18) was $10,488–$138,406, depending on race, sex, and BMI. However, this study used a simulation model and prevalent cases, while we used actual spending from claims data for incident cases. In another study by Zhuo et al. (17), the estimated lifetime spending ranged from $60,909 to $145,648, depending on the age at diagnosis. However, these estimates are not comparable with ours because they represent total lifetime medical spending for a new diabetes cohort, not additional spending due to diabetes.

We examined results across five racial/ethnic groups in detail and found that lifetime incremental medical spending was highest for Asian/Pacific Islander beneficiaries, followed by non-Hispanic Black, Hispanic, American Indian/Alaska Native, and non-Hispanic White beneficiaries. Leung et al. (19) reported that lifetime incremental medical spending was greater for White adults than Black adults for all age-groups. Different ranking by race/ethnicity between our study and the study by Leung et al. could be due to different study methods and data sources used. Leung et al. used a simulation model and data from the Medical Expenditure Panel Survey to project the lifetime expenditure. In comparison, we used real-world data from >1 million people with T2D with a longitudinal follow-up to estimate the actual medical spending of Medicare enrollees.

We observed more life-years for some subgroups of Hispanic, Asian/Pacific Islander, and American Indian/Alaska Native beneficiaries diagnosed after age 75 years, suggesting that T2D may not be a risk factor for death in these subgroups. This observation aligns with the findings reported in a population-based cohort study that some older Asian people with T2D had longer life expectancy than those without diabetes and that the difference increased with age at diagnosis (23). Another study found a significant attenuated risk of death in people diagnosed with diabetes at age ≥75 years (25). It should be noted, however, that this effect largely disappeared in our unmatched sensitivity analysis (Supplementary Table 2), and future studies may further explore possible reasons for these unexpected findings.

Estimates from this study could be used for the planning and prioritization of health care resource allocations, particularly with respect to diabetes management and economic evaluation of programs and policies for T2D prevention. The economic burden of T2D among Medicare beneficiaries is high (4,20), although few studies have examined it. Our estimates suggest that ∼$16,115–$122,146 in medical spending could be averted over the remaining lifetime for every case of T2D prevented among Medicare beneficiaries, depending on the age, sex, and race/ethnicity. The high spending of T2D could be reduced by targeted interventions, such as the MDPP, which helps beneficiaries at high risk avoid or delay T2D.

Our estimated racial/ethnic differences in incremental lifetime medical spending imply that T2D impacts people differently in terms of survival and medical spending. These disparities may be explained by a variety of factors, including differences in progression rates of diabetes-related complications, variations in access to quality health care, disparities in socioeconomic status, and cultural differences in health-seeking behaviors and lifestyle choices (25–28). Understanding these disparities is crucial as it can inform targeted T2D prevention programs, interventions, policies, and treatments aimed at reducing the economic burden of diabetes in the most affected groups and groups with the most economic burden. Our race/sex-specific lifetime medical spending estimates could be used for assessing the cost-effectiveness of diabetes prevention targeted at different racial groups and sexes. Future studies can examine how risk factors such as socioeconomic status, access to health care, and quality of care received and their prevalence contribute to racial/ethnic differences.

This study has several strengths. We constructed a cohort with >1 million people with incident T2D and a comparable cohort without diabetes, with a compatible follow-up period of up to 19 years. Using real-world national data from Medicare fee-for-service claims allowed us to mediate limitations in previous studies. We constructed survival curves and calculated lifetime spending using a single source of national claims data to reduce bias caused by using varied sample populations in previous studies and to represent lifetime medical spending more accurately. Furthermore, while previous studies reported disparities in medical spending across age-groups and sexes, our study additionally highlighted disparities across racial and ethnic groups.

Our study also has several limitations. First, we used the propensity score matching method to select a comparable cohort without diabetes. While propensity score matching helps to mitigate selection bias by balancing known confounders between the T2D and nondiabetes cohorts, it does not account for unmeasured variables that could influence lifetime medical spending. The potential impact of such unmeasured confounders remains a limitation of this study. Second, we focused exclusively on the Medicare population aged >65 years. While this is a critical group at high risk for T2D, lifetime spending for people diagnosed with T2D at younger ages would likely be quite different. The cohort with newly diagnosed T2D included in the current study was likely healthier than those diagnosed at a younger age. Therefore, this exclusion also limits the generalizability of our findings to individuals with diabetes diagnosed at an earlier age. In addition, while nearly all the population aged >65 years in the U.S. is enrolled in Medicare, we excluded those beneficiaries without Medicare Part D enrolled in a fee-for-service plan and those enrolled in managed care plan or Part C. Our estimated incremental lifetime medical spending may not be generalizable to these excluded beneficiaries. Third, we were unable to follow everyone until death; beyond 19 years, where existing data were not available, we projected future annual spending based on previous spending, which could have introduced bias. Fourth, to ensure enough follow-up time, our T2D cohort was limited to incident cases in 2001–2003; results may not be applied to cases diagnosed before 2001 or after 2003. Although we accounted for inflation, the cost of treating diabetes and its complications might change over time because of factors such as newer medications, technological advancement, or payment policy changes (4,29,30). Our results may therefore underestimate lifetime incremental medical spending of diabetes for which treatments have become more complicated and expensive in recent years. Fifth, some groups, such as American Indians/Alaska Natives, had very small populations aged >80 years, so findings should be interpreted with caution. Finally, the algorithm we used to identify people with incident T2D could have misclassified some of the study sample. While it is unlikely that many of our cases were misclassified as T2D versus type 1 diabetes due to the age of the study sample, some of the people in the nondiabetes cohorts may have had diabetes, since 4.7% of people aged ≥65 years have undiagnosed diabetes (6). Such misclassification would tend to bias the results toward an underestimation of incremental spending.

We estimated the race/ethnicity- and sex-specific incremental lifetime medical spending among Medicare beneficiaries with incident T2D by age at diagnosis using real-world claims data and a longitudinal study design. Our findings of the large incremental lifetime medical spending due to T2D across all racial/ethnic groups underscore the need for preventing T2D in Medicare beneficiaries. Our estimates can be used to measure the potential medical spending saved from T2D prevention programs, such as the MDPP, including those targeted to different races and ethnicities and sexes.

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

Author Contributions. Y.S. and Y.W. drafted the manuscript. Y.S., Y.W., E.B., G.I., C.H., and P.Z. contributed to the critical review and editing of the manuscript. Y.S., Y.W., and P.Z. planned and conducted data analyses. P.Z. conceived the study idea. Y.W. 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.

Prior Presentation. Parts of this study were presented in abstract form at the 83rd Scientific Sessions of the American Diabetes Association, San Diego, CA, 23–26 June 2023.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Alka M. Kanaya.