Racial and ethnic patterns and differences in health care expenditures among Medicare beneficiaries with and without cognitive limitation or Alzheimer's disease and related dementias: a retrospective cohort study

Background : Numerous studies have documented racial and ethnic differences in the prevalence and incidence of Alzheimer's disease and related dementias (ADRD). Less is known, however, about racial and ethnic differences in health care expenditures among older adults at risk for ADRD (cognitive limitation without ADRD) or with ADRD. In particular, there is limited evidence that racial and ethnic differences in health care expenditures change over the trajectory of ADRD or differ by types of service. Methods : We examined racial and ethnic patterns and differences in health care expenditures (total health care expenditures, out-of-pocket expenditures, and six service-specific expenditures) among Medicare beneficiaries without cognitive limitation, those with cognitive limitation without ADRD, and those with ADRD. Using the 1996-2017 Medical Expenditure Panel Survey, we performed multivariable regression models to estimate expenditure differences among racial and ethnic groups without cognitive limitation, those with cognitive limitation without ADRD and those with ADRD. Models accounted for survey weights and adjusted for various demographic, socioeconomic, and health characteristics. Results : Asians, and Latinos without cognitive limitation had lower total health care expenditures than whites without cognitive limitation ($10236, $9497, $9597, and $11541, respectively), but there were no racial and ethnic differences in total health care expenditures among those with cognitive limitation without ADRD and those with ADRD. In all populations, however, blacks, Asians, and Latinos tended to have lower out-of-pocket expenditures than whites (except for Asians with cognitive limitation without ADRD). Furthermore, service-specific health care expenditures varied by racial and ethnic groups. Conclusions : Our findings may suggest that racial and ethnic minority groups did not experience limited access to care before and after ADRD diagnosis. Differences in out-of-pocket expenditures and service-specific expenditures may be attributable to racial and ethnic differences in care access and/or care preference based on family structure and cultural/economic factors. Particularly, heterogeneous patterns of service-specific expenditures by racial and ethnic groups underscore the

importance of future research in identifying determinants leading to variations in service-specific expenditures among racial and ethnic groups.

Background
The prevalence of Alzheimer's disease and related dementias (ADRD) is a growing crisis in the United States (US) that is estimated to increase substantially over the next several decades. In 2010, approximately 3.6 million Americans had been diagnosed with ADRD (1). The number of Americans with ADRD is projected to be 13.8 million in 2050, a 283% increase (1). Furthermore, mean perperson costs for Medicare beneficiaries with ADRD were estimated to be $49,126 in 2016, more than triple the average $15,550 costs for Medicare beneficiaries without ADRD (2). Aggregate costs for Medicare beneficiaries with ADRD are expected to increase from $181 billion in 2010 to $1.1 trillion in 2050 (2). Such a dramatic increase in the costs of ADRD would lead to a substantial burden on the Medicare program.
Numerous studies have documented racial and ethnic differences in the prevalence and incidence of ADRD. Specifically, compared to non-Latino whites (whites), non-Latino blacks (blacks) are approximately two times more likely to have ADRD (3,4) and Latinos are approximately 1.5 times more likely to have ADRD (3,5,6). Recent research found that differences among racial and ethnic groups in the prevalence of ADRD decreased between 2000 and 2012 (7). However, the prevalence rates of ADRD were still found to be higher among blacks and Latinos than among whites (19.3%, 16.3%, and 7.4% for blacks, Latinos, and whites, respectively). Incidence rates of ADRD were also higher among blacks and Latinos than among whites (13.8%, 12.2%, and 10.3% for blacks, Latinos, and whites, respectively) (8).
Less is known, however, about racial and ethnic differences in health care expenditures. To the best of our knowledge, only a few studies have examined how health care expenditures related to ADRD vary among racial and ethnic groups (8)(9)(10). One study used Medicare fee-for-service claims data for 2014 and found that compared to whites with ADRD, blacks, Latinos, and "others" with ADRD had higher Medicare expenditures ($27,315, $26,280, $21,649, and $20,199 for blacks, others, Latinos, and whites, respectively) (8). Higher expenditures among racial and ethnic minority groups with ADRD might be attributable to limited access to care in the early stages of ADRD, which could lead to delays in treatment and diagnosis and greater morbidity from these diseases, incurring higher health care expenditures after receiving a diagnosis of ADRD. Although a few studies have examined health care expenditures among Medicare beneficiaries with mild cognitive limitation, these studies did not analyze differences based on racial and ethnic groups (11,12).
While it is important to examine differences in total health care expenditures among racial and ethnic groups to ensure equal access to ADRD care, it is also important to understand patterns of typespecific health care expenditures; focusing only on total health care expenditures might lead to overlooking mechanisms that contribute to health care expenditures among members of racial and ethnic minority groups with ADRD. This is more likely to be relevant to patients with ADRD because cultural preferences can affect the optimal clinical setting for individuals with ADRD and their families.
Prior research found that caregivers of black patients were less satisfied with hospital discharge planning than caregivers of white patients were, and that caregivers of black patients used formal home care more than caregivers of white patients did (13,14). Furthermore, there were substantial racial and ethnic differences in the number of individuals who chose to be admitted to nursing homes; usage of nursing homes was particularly low among Latinos (15). However, it is worth noting that these findings may be also attributable to structural barriers. Additionally, choice of care setting for patients with ADRD is critical because evidence suggests that these patients experience inefficient care delivery and health care utilization. A significant factor in health care utilization among those with ADRD is due to transitions to high-cost settings such as an inpatient setting or skilled nursing facility (16)(17)(18); some of these transitions have been shown to be unnecessary or preventable (19)(20)(21)(22). This suggests that higher expenditures among patients with ADRD might result from inefficient use of care.
To address this gap, we examined racial and ethnic patterns and differences in health care expenditures among Medicare beneficiaries. We estimated such expenditures among Medicare beneficiaries with cognitive limitation without a diagnosis of ADRD and those diagnosed with ADRD. In addition, we examined various types of health care expenditures: total health care expenditures, out-of-pocket (OOP) expenditures, and six service-specific expenditures.
Our study makes several key contributions to the literature. First, we used data that collects information on race and ethnicity via population survey. Prior research has instead relied on the Medicare claims data. A common concern about the claims data is a lack of in-depth measures of socioeconomic factors that may influence the health care expenditures and racial and ethnic disparities. Using the nationally representative survey data enables us to account for comprehensive measures of demographic and socioeconomic factors. Hence, our finding should be more robust and more accurately predict the racial and ethnic disparities in the amount and pattern of health care expenditures. In addition, we examined racial and ethnic disparities along the trajectory of ADRD (i.e., among Medicare beneficiaries without cognitive limitation, those with cognitive limitation without ADRD, and those with ADRD, respectively). We expect that our findings can identify disparities in the early stages of ADRD and advocate for appropriate prevention screenings or treatment to delay the onset of ADRD among racial and ethnic minority beneficiaries. We included Medicare beneficiaries (aged 65 and above) who were non-Latino white, non-Latino black, non-Latino Asian, or Latino. Then, we identified the following three populations: 1) those who reported as not having cognitive limitation, 2) those who reported as having cognitive limitation without a diagnosis of ADRD, and 3) those diagnosed with ADRD. MEPS measured cognitive limitation based on the household respondent's answers from the individuals in the sample. Cognitive limitation was defined as "confusion or memory loss, had problems making decisions, or required supervision for their own safety." ADRD cases were identified through three-digit ICD-9-CM diagnostic codes (290, 294, 331, or 797) (12,23) or three-digit ICD-10 diagnostic codes (F01, F03, G30, and G31) (24). The key independent variables were the participant's race (white, black, Asian, or Latino), presence of cognitive limitation or ADRD, and its interaction terms. To control for differences in sample characteristics among racial and ethnic groups, we included the following variables: age (65-69, 70-74, 75-79, 80-84, or ≥ 85 years old); sex; marital status (married or unmarried); education (less than high school degree, high school degree, some college, or more than college degree); family income as a share of the federal poverty level (FPL; 0-99%, 100-124%, 125-199%, 200-399%, or ≥ 400%); family size (one, two, three, or more than four); private insurance; eligibility for Medicare and Medicaid; area of residence (Northeast, Midwest, South, or West); medical conditions (myocardial infarction, congestive heart failure, diabetes, hypertension, diabetes, renal disease, cancer, and psychiatric disorder); limitations at school, work, or housework; functional limitations; and a proxy response to an interview (proxy response or self-response).

Statistical Analysis
We first estimated weighted sample characteristics among racial and ethnic groups without cognitive limitation, those with cognitive limitation without ADRD, and those with ADRD and tested differences using chi-squared tests. Then, we examined unadjusted weighted outcomes among racial and ethnic groups without cognitive limitation, those with cognitive limitation without ADRD, and those with ADRD and tested differences using an analysis of variance. Finally, we performed multivariable regression models to estimate expenditure differences among racial and ethnic groups without cognitive limitation, those with cognitive limitation without ADRD and those with ADRD. Specifically, because we did not observe Medicare beneficiaries with zero total health care expenditures, we ran generalized linear models with gamma family and log link to estimate differences in total health care expenditures. For other types of health care expenditures, there were those with zero expenditures and thus we ran two-part models to handle zero expenditures. Using marginal effects at representative values, we produced findings that can be interpreted as dollar values (25,26).
Specifically, we estimated the predicted mean values of the outcomes for each of the racial and ethnic group without cognitive limitation, those with cognitive limitation without ADRD, and those with ADRD. Then, we conducted post-estimation tests to examine statistical significance in the differences in the adjusted outcomes among racial and ethnic minority groups relative to non-Latino white. All models account for survey weights and adjusted for the variables described above as well as yearfixed effects. All analyses were conducted using Stata 15.

Results
Our sample consisted of 57057 Medicare beneficiaries without cognitive limitation (39767 whites, 7974 blacks, 2551 Asians, and 6765 Latinos), 10088 Medicare beneficiaries with cognitive limitation without ADRD (5947 whites, 1933 blacks, 523 Asians, and 1685 Latinos), and 3420 Medicare beneficiaries with ADRD (2028 whites, 693 blacks, 120 Asians, and 579 Latinos) ( Table 1). There were significant differences in sample characteristics among racial and ethnic groups without cognitive limitation, those with cognitive limitation without ADRD, and those with ADRD. For all populations, blacks, Asians, and Latinos were more likely than whites to have less than a high school degree, more likely to have a family income higher lower than 200% of the FPL, more likely to have a family with more than three members, and more likely to have private health insurance or Medicaid. For those without cognitive limitation, there were differences in health status. However, differences were marginal among racial and ethnic groups those with cognitive limitation without ADRD and those with ADRD. Table 1 Sample characteristics.  There were significant differences in unadjusted health care expenditures among racial and ethnic groups without cognitive limitation, those with cognitive limitation without ADRD, and those with ADRD (Table 2). Blacks, Asians, and Latinos without cognitive limitation had significantly lower total health care expenditures than the equivalent whites. Asians with cognitive limitation without ADRD had significantly lower expenditures than the equivalent whites. However, no significant differences were detected among blacks and Latinos with cognitive limitation without ADRD and blacks, Asians, and Latinos with ADRD. For OOP expenditures, blacks, Asians, and Latinos in all groups had significantly lower expenditures than the equivalent whites. However, there was no significant difference in OOP expenditures between whites and Asians with ADRD. For service-specific expenditures, blacks, Asians, and Latinos without cognitive limitation tended to have lower inpatient, outpatient, office-based, home health, and prescription drug expenditures than the equivalent whites.
However, significant differences were detected in a few types of expenditures among racial and ethnic groups with cognitive limitation without ADRD and those with ADRD (home health expenditures among blacks with cognitive limitation without ADRD, inpatient expenditures among Asians with cognitive limitation without ADRD, outpatient and office-based expenditures among Latinos with cognitive limitation without ADRD, and inpatient and office-based expenditures among Latinos with ADRD). Our adjusted analysis showed that blacks, Asians, and Latinos without cognitive limitation had lower total health care expenditures than whites without cognitive limitation ($10236, $9497, $9597, and $11541, respectively), but there were no racial and ethnic differences in total health care expenditures among those with cognitive limitation without ADRD and those with ADRD (Table 3). In all populations, however, blacks, Asians, and Latinos tended to have lower OOP expenditures than whites (except for Asians with cognitive limitation without ADRD). The magnitude of the differences in OOP expenditures was most pronounced for those with ADRD.

Discussion
Our study documented that there were significant differences in total health care expenditures among racial and ethnic groups without cognitive limitation, but racial and ethnic differences in total health care expenditures were insignificant among those with cognitive limitation without ADRD and those with ADRD. This suggests that racial and ethnic minority groups may not experience limited access to care before and after a diagnosis of ADRD. Meanwhile, results showed the discrepancies between unadjusted summary of expenditures and predicted expenditures after adjusting individuals' demographic and socioeconomic characteristics. First, our unadjusted analysis showed that Asians with cognitive limitation without ADRD had lower total health care expenditures than other racial and ethnic groups. However, this was not observed among Asians with ADRD. These indicate that Asians with cognitive decline but no ADRD may receive fewer health care services, possibly leading to late detection and diagnosis of ADRD. This explanation is likely plausible because Asians are more likely to lack a usual source of care (27,28), leading to relatively lower health care utilization (29), especially for primary care, and preventive services (30). However, a significant difference was not detected in our adjusted analysis, probably due to a small sample size. Second, our unadjusted analysis showed that consistent with findings from previous studies (8)(9)(10), blacks and Latinos with ADRD had higher total health care expenditures than the equivalent whites. However, significant differences were not observed after adjusting for demographic and socioeconomic status and health status. This phenomenon was found in incidence (31) and Asians were shown to experience more delayed or forgone care than whites (32). Third, the racial and ethnic minority groups may replace high-costs services with informal care by family caregivers.
One study found that Latinos and Asians were more likely to use informal home care and less likely to use formal care compared to whites (33). This could be feasible because of a relatively large family size of the racial and ethnic minority groups.
Our findings showed that service-specific expenditures varied by racial and ethnic groups, but similar trends were observed in both populations. First, blacks and Latinos had higher home health expenditures than whites in both populations. This may be attributable to the fact that they prefer home health care due to the presence of family members who can provide informal care (34) or cultural reasons (35). However, blacks and Latinos had lower prescription drug expenditures than whites. This is likely to be explained by less contact with physicians, possibly resulting in fewer prescriptions being written (36). Research found that blacks and Latinos were more likely to have mental health visits to primary care providers rather than to specialists, leading to fewer prescriptions for psychotropics (37). On the other hand, Asians with cognitive limitation without ADRD had lower inpatient and outpatient expenditures than the equivalent whites. This may raise concern of delayed detection or diagnosis of ADRD as diagnostic services for disease detection are usually provided in inpatient or outpatient settings (8). However, Asians with ADRD had lower outpatient and ER expenditures than the equivalent whites. This may indicate that Asians manage health better as research showed that ADRD patients had rehospitalization or ER visits mainly due to poor care management such as injuries from falls (38). Further investigation is warranted to understand and identify determinants leading to variations in service-specific expenditures among racial and ethnic groups.
Our study has several limitations. First, MEPS surveys the civilian non-institutionalized US population, and thus our estimates did not account for patterns of health care expenditures for the civilian institutionalized US population. Second, MEPS does not include health care expenditures for SNF services. As racial and ethnic minority groups were shown to have lower expenditures for SNF services than whites, this is unlikely to reverse our findings. Third, MEPS reported Asian Americans and Pacific Islanders as a single group during 1999-2002, and thus we could not distinguish each other during this period. However, this is unlikely to affect our findings due to small sample size of Pacific Islanders. Fourth, MEPS provides limited information on ADRD severity, and thus we could not completely control for this factor. Fifth, we controlled for a range of demographic and socioeconomic characteristics, but we could not adjust for all other potential confounding factors. Sixth, the observed prevalence of ADRD may be inaccurate because we were limited to 3-digit ICD-9-CM or ICD-10-CM codes. Finally, our findings should be interpreted with caution as we did not examine whether whites have appropriate health care expenditures. Thus, we cannot rule out the possibility that whites may overutilize health care.

Conclusions
Our study documented that there were significant differences in total health care expenditures among racial and ethnic groups without cognitive limitation, but no significant differences were detected in total health care expenditures among racial and ethnic groups with cognitive limitation without ADRD and those with ADRD. However, there were substantial differences in OOP expenditures and servicespecific expenditures among racial and ethnic groups with cognitive limitation without ADRD and those with ADRD. These findings have implications for future research. First, this work emphasizes that service-specific expenditures varied by racial and ethnic groups. Second, heterogeneous patterns of service-specific expenditures by racial and ethnic groups underscore the importance of future research in identifying determinants leading to variations in service-specific expenditures among racial and ethnic groups.

Declaractions
Ethics Approval and Consent to Participate: Not applicable.

Availability of Data and Materials:
The datasets analyzed during the current study are available at https://meps.ahrq.gov/mepsweb/.

Competing Interests:
The authors declare that they have no competing interests.
Funding: This study is supported by the National Institute on Aging (1R56AG62315) and the National Institute on Minority Health and Health Disparities (R01MD011523). The sponsors had no role in design and conduct of the study, collection, management, analysis, interpretation of data, review, or approval of the manuscript.
Author's Contributions: SP was responsible for conceptualization, methodology, statistical analysis, and writing. JC was responsible for conceptualization, methodology, and writing. All authors read and approved the final manuscript.