National racial/ethnic and geographic disparities in experiences with health care among adult Medicaid beneficiaries

Objectives To investigate whether health care experiences of adult Medicaid beneficiaries differ by race/ethnicity and rural/urban status. Data Sources A total of 270 243 respondents to the 2014‐2015 Nationwide Adult Medicaid Consumer Assessment of Healthcare Providers and Systems Survey. Study Design Linear regression was used to estimate case mix adjusted differences in patient experience between racial/ethnic minority and non‐Hispanic white Medicaid beneficiaries, and between beneficiaries residing in small urban areas, small towns, and rural areas vs large urban areas. Dependent measures included getting needed care, getting care quickly, doctor communication, and customer service. Principal Findings Compared with white beneficiaries, American Indian/Alaska Native (AIAN) and Asian/Pacific Islander (API) beneficiaries reported worse experiences, while black beneficiaries reported better experiences. Deficits for AIAN beneficiaries were 6‐8 points on a 0‐100 scale; deficits for API beneficiaries were 13‐22 points (P's < 0.001); advantages for black beneficiaries were 3‐5 points (P's < 0.001). Hispanic white differences were mixed. Beneficiaries in small urban areas, small towns, and isolated rural areas reported significantly better experiences (2‐3 points) than beneficiaries in large urban areas (P's < 0.05), particularly regarding access to care. Racial/ethnic differences typically did not vary by geography. Conclusions Improving experiences for racial/ethnic minorities and individuals living in large urban areas should be high priorities for policy makers exploring approaches to improve the value and delivery of care to Medicaid beneficiaries.

state income eligibility cutoff was 61% of the federal poverty level (FPL) and nonelderly adults were categorically ineligible in most states. The ACA fundamentally changed Medicaid by establishing eligibility for nonelderly adults and enacting a uniform, national income eligibility threshold for this group, 138% of the FPL, 3 thereby increasing Medicaid enrollment by ~11 million people. 4 State-level funding to meet the needs of the Medicaid population has not kept pace with increasing enrollment rates, 5 prompting policy makers to explore approaches to improve the value and delivery of care to Medicaid beneficiaries. 6 A key component of providing high-quality care involves the identification and elimination of disparities based on sociodemographic characteristics. 7,8 The current study takes a step in that direction by using a newly avail- A variety of factors tied to geography may contribute to health care, including the health care delivery environment, transportation infrastructure, and the distribution of health care resources. [17][18][19] Although little is known about geographic disparities in care faced by Medicaid beneficiaries, studies of the general U.S. adult population have identified significant barriers to accessing care for rural residents, including provider shortages, recent hospital closures, and long travel distances to see providers. [20][21][22] In the primarily low-income Medicaid population, geographic barriers to care may exist in large urban areas as well as rural areas.
Recent research has demonstrated substantial growth in highpoverty urban neighborhoods that are disproportionately occupied by communities of color. [23][24][25][26] In much the same way that geographic isolation can limit rural residents' access to high-quality care, 20,27 economic and racial segregation can lead to "health care deserts" in urban areas that effectively cut poor people off from high-quality care that is nearby, but still inaccessible. [28][29][30] Given multiple disincentives for primary care physicians to live and work in economically depressed urban areas 31 and some residents' inability or disinclination to seek care outside of their own neighborhoods, the urban poor may end up disproportionality receiving care from poorerperforming providers.
There is evidence that geographic factors may exacerbate racial/ ethnic disparities in health care. 30,[32][33][34][35] For example, studies of the general adult population have found that racial/ethnic disparities in health care access are generally greater in rural than in urban settings. 33 Within the Medicaid population, similar synergistic effects of race/ethnicity and geography may be observed among the urban poor. For example, combined racial and economic segregation in large metropolitan areas may make it especially difficult for poor, racial/ethnic minorities to access high-quality care in these areas. 26 The current study used data from the NAM CAHPS survey to investigate differences in adult Medicaid beneficiaries' experiences of care based on (a) race/ethnicity, (b) rural/urban residency, and (c) the combination of these two sociodemographic characteristics. We expected to find racial/ethnic differences that are comparable to ones found in other populations, [36][37][38] that is, that white beneficiaries would, for the most part, report more favorable experiences with care than racial/ethnic minorities. For the reasons outlined above, we also expected to find a disadvantage for beneficiaries living in rural and large urban areas compared with those living in intermediate areas, and we expected a synergistic effect of race/ethnicity and rural/urban residence on beneficiaries' experiences with care.

| Dependent variables
The dependent variables were four multi-item composites constructed from 10 NAM CAHPS report items. These composites measured beneficiaries' own experiences of care in the prior 6 months in the following areas: Getting Needed Care (two items; Cronbach's alpha [α] = 0.64), Getting Care Quickly (two items; α = 0.68), How Well Doctors Communicate (four items; α = 0.90), and Health Plan Information and Customer Service (two items; α = 0.77). Items comprising the composites (see Appendix Table S1 for detail * ) were asked of the subset of beneficiaries to whom they were applicable (screener items were used to assess applicability). All items had the following response options: never, sometimes, usually, and always.
Top-box scoring was used for the four composite measures for ease of interpretation. In this scoring approach, the most favorable response option (i.e, always) is coded 100 and all other options (i.e, never, sometimes, and usually) are coded 0 prior to averaging nonmissing items to create composite scores. For example, the score for a respondent who answered "always," "always," "never," and "sometimes" scale. Previous analyses of CAHPS scores have suggested that statistically significant differences of 1 point on a 0-100 scale can be considered small; differences of three points can be considered medium; and differences of five points can be considered large. 39 For instance, a three-point increase in some CAHPS measures has been associated with a 30% reduction in disenrollment from health plans, which suggests that even "medium" differences in CAHPS scores may indicate substantially different care experiences. 40 In describing results below, we will refer to nonsignificant differences on the dependent measures as "similar" scores.

| Race/Ethnicity
Information on race/ethnicity was primarily collected via self-report on the NAM CAHPS survey, although in some instances state Medicaid personnel assisted in the response or included administrative, rather than self-reported values. The cases that were not fully self-reported cannot be identified. In the survey, respondents were asked to indicate whether they were of Hispanic, Latino, or Spanish origin. Those who responded affirmatively were asked to indicate whether they were Mexican, Puerto Rican, Cuban, or of other Hispanic, Latino, or Spanish origin. Race was measured on the survey using an item with fifteen response options: white; black or African American; American Indian or Alaska Native (AIAN); Asian Indian; Chinese; Filipino; Japanese; Korean; Vietnamese; other Asian; Native Hawaiian; Guamanian or Chamorro; Samoan; other Pacific Islander; or some other race. Our primary set of analyses focused on five mutually exclusive racial/ethnic groups: Hispanic (any beneficiary of Hispanic, Latino, or Spanish origin regardless of race), non-Hispanic white, non-Hispanic black, non-Hispanic AIAN, and non-Hispanic Asian or Pacific Islander (API; a combination of all Asian and Pacific Islander categories). We refer to beneficiaries in these five categories as Hispanic, white, black, AIAN, and API.
Although they are included in our analysis, we do not report effects for multiracial beneficiaries (n = 19 842) because they are a heterogeneous and therefore hard-to-interpret group. Focusing our primary analyses on these five categories gave us the statistical power needed to investigate how racial/ethnic differences in experiences with care vary by categories of rurality/urbanicity. To understand variation within these larger racial/ethnic categories, we additionally

| Control variables
Because other beneficiary characteristics are also known to be associated with response tendencies and may differ by race/ethnicity or rural/urban status, we also included age (18-24, 25-
Information from these respondents was excluded from the analysis. Missing values on control variables (2%-3%) were imputed using within-state and enrollment-sampling-stratum means for all nonmissing respondents. Respondents with missing outcome variables were excluded from analyses of those outcomes.

| Statistical analysis
The SURVEYREG procedure within the SAS 9.4 software package (SAS Institute Inc., Cary, NC, USA) was used to apply weights that accounted for the complex survey design and adjusted for postsampling-ineligible adults and nonresponse at the level of the enrollment stratum within states. We used linear regression with weighted least-squares estimation to model the association between race/ethnicity and Medicaid beneficiaries' experiences with care. Each of our primary models (one for each of the four dependent variables) of the association between race/ethnicity and patient experience included four indicators of race/ethnicity (white was the reference group) and the previously described control variables. A second set of models focused on API and Hispanic subgroups. These models were identical to the primary models except that they included two indicators of API subgroup membership (Asian and NHOPI) in place of the single API indicator, and four indicators of Hispanic subgroup membership (Mexican, Puerto Rican, Cuban, and other) in place of the single indicator of Hispanic ethnicity. We modeled the association between rural/ urban status and experiences with care using a similar set of analyses.
These models included three indicators of rural/urban status (large urban was the reference group) and the previously described covariates as predictors. A final series of models predicted each outcome variable from race/ethnicity (Hispanic, black, AIAN, and API vs white), rural/urban status, the interaction of race/ethnicity and rural/urban status, and the covariates. We conducted joint tests of the interaction coefficients from these final models to assess whether differences in care between white and racial/ethnic minority beneficiaries varied significantly by rural/urban strata.

| RE SULTS
Excluding those of unknown race/ethnicity, 53% of beneficiaries were white, 19% were black, 13% were Hispanic, 7% were multiracial, 5% were API, and 2% were AIAN. Table 1 shows how beneficiaries were distributed across the four rural/urban categories.
Approximately 70% of beneficiaries resided in large urban areas, 14% in small urban areas, 8% in small towns, and 7% in rural areas; rural residence differed strongly by race/ethnicity. More than a third of AIAN beneficiaries lived in rural areas, compared with approximately 1% of API beneficiaries, 2% of black beneficiaries, 4% of Hispanic beneficiaries, and 10% of white beneficiaries.
An appendix table (Table S2)  There were no small urban respondents from DC or RI, no small town respondents from CT, DC, NJ, or RI, and no rural respondents from DC, DE, or NJ.
in small towns and rural areas (P's < 0.001). Within racial/ethnic groups, there were three notable exceptions to these overall associations. Among AIAN beneficiaries, small town and rural residents were younger than urban residents (P = 0.03). Among API beneficiaries, residents of large urban areas were older than residents of other areas (P = 0.002), and enrollment status (P = 0.51) and overall health (P = 0.65) did not differ by rural/urban categories.    Notes: A total of 2302 respondents were missing information on race/ethnicity and thus were excluded from this analysis. Estimates are from weighted linear regression models predicting each top-box-scored measure from race/ethnicity, age, education, overall health, and state of residence. Multiracial beneficiaries were included in the analysis but coefficients comparing this group to white beneficiaries are not shown. AIAN, American Indian or Alaska Native; API, Asian or Pacific Islander. *0.01 ≤ P < 0.05; **0.001 ≤ P < 0.01; ***P < 0.001. a Each group compared with white.
than for the other three measures, which had means in the range of 55-71 out of 100. Controlling for differences in age, education, overall health, and state of residence, beneficiaries in large urban areas reported worse experiences with getting needed care and getting care quickly than did beneficiaries in small urban areas, small towns, and rural areas (deficits ranged from 2 to 3 points on the 0-100 scale; P's < 0.05). Similarly, beneficiaries in large urban areas reported worse experience getting information and customer service from their plans than did beneficiaries in small towns (P < 0.01). across rural/urban categories. These tests revealed that racial/ethnic differences were generally similar across rural/urban categories (see Table   S3 for details). The only exception was for AIAN-white differences in getting needed care, which varied significantly by rural/urban category (P = 0.002). Specifically, the disadvantage of AIAN beneficiaries relative to white beneficiaries was only observed in small urban and rural areas (where deficits on getting need care were 9 and 11 points, respectively).

| D ISCUSS I ON
Our study contributes substantially to the literature by taking a highresolution look at racial/ethnic differences among beneficiaries (including national-origin subgroups), providing the first examination of urban/rural differences in Medicaid beneficiaries' experiences with care, and examining how these two factors intersect in predicting experiences with care. We found considerable racial/ethnic differences in Medicaid beneficiaries' care experiences that broadly mirror differences seen in other populations. [36][37][38] That is, compared with white beneficiaries, AIAN, API, and Hispanic beneficiaries tended to report worse experiences with access to care, provider communication, and customer service received from plans, though there were some exceptions (e.g, Hispanic beneficiaries reported better experiences with customer service than white beneficiaries).
In contrast, black beneficiaries tended to report better experiences than white beneficiaries. This, too, is consistent with findings from prior studies. 16 barriers are an even bigger impediment to patient experience for lowerincome minority patients than for higher-income patients. 16,43 Thus, black beneficiaries may not show the same deficit in care compared to white beneficiaries in part because they do not face the same language difficulties that may be especially acute in this low-income population.
Though not a new finding, the reason why black beneficiaries score higher than white beneficiaries on some CAHPS measures is not known.
TA B L E 5 Weighted racial/ethnic differences in experiences of care by categories of rural/urban residence  Access to nonemergent specialty care services is notably poor in these areas. 55 When specialty care is required, the IHS often refers patients to larger, contracted hospitals in urban areas, far from the patient's home and community. 56 Despite their need for specialty care, many of these referred patients cannot find a way to reach that care 56 Thus, it is perhaps unsurprising that we found AIAN-white disparities in getting needed care to be especially pronounced in rural areas.  58,59 it is possible that nonresponse bias influenced our findings. Sixth, this study provides limited insight into the causes of the racial/ethnic and rural/urban differences that were uncovered. Additional research is needed to understand the degree to which these differences reflect geographic isolation, unequal treatment by providers based on patient race or ethnicity, or an increased tendency on the part of certain groups to receive care from providers who are poorer performing overall. 60 It is possible, in principle, that expectations based on health care received prior to Medicaid might influence one's reference point for an "acceptable" health care experience and result in differences in evaluations associated with race, ethnicity, or geography. In practice, it has been difficult to find evidence that health care expectations differ by race or ethnicity. For example, Weinick and colleagues 61 found that black, Hispanic, and white respondents reported generally similar health care expectations and provided similar mean responses to CAHPS composites in response to standardized encounters. We are not aware of investigations of possible differences in health care expectations by geography, which may be an important topic for future research.
These limitations notwithstanding, our study suggests improving care for racial/ethnic minorities and those living in urban areas as top priorities for policy makers seeking to improve the quality of care delivered to Medicaid beneficiaries. Our study also suggests a need for improved access to care (i.e, the ability to get care that is needed and to get it in a timely way) for AIAN beneficiaries in small urban and rural areas. Which policy options are likely to be most effective at meeting these needs depends on the reasons that underlie the observed disparities. For example, if transportation barriers among impoverished urban populations drive the geographic differences that we observed, it may be that better urban transportation planning is needed 49 or that providers need to be further incentivized to live and work in impoverished urban neighborhoods. If, on the other hand, rural/urban differences in care are due to a Medicaid payment system that creates incentives for Medicaid managed care plans to undertreat patients, then raising capitations rates may help to eliminate some of the geographic differences that we observed. 62 Other strategies are likely to be needed to address racial/ethnic differences in care, including cultural competency training and the provision of language-appropriate services. Although prior approval and notification by CMS is not required, CMS was provided with an advanced copy of the manuscript as a courtesy.

E N D N OTE
* Categories for the control variables listed here are the ones used for case mix adjustment. For simplicity, coarser categories were used in describing the sample.