Accountable Care Organizations and Health Disparities of Rural Latinos: A Longitudinal Analysis

Abstract

The purpose of this study was 2-fold: (1) to analyze the change in diabetes-related hospitalization rates of rural Latino older adult patients as compared with their White counterparts and (2) to determine what factors, including rural health clinic (RHC) participation in accountable care organizations (ACOs), are related to reduced disparities in diabetes-related hospitalization rates. Data for Latino Medicare beneficiaries who were served by RHCs over an 8-year period were analyzed. First, a difference-of-means test was conducted to determine whether there was a change in disparity from the pre-ACO period (2008–2011) to the post-ACO period (2012–2015). A statistically significant decrease in disparity over time was found (t = −7.6899, df = 115, P < 0.001.) Second, multiple regression analyses of 3 separate models were conducted to determine whether ACO participation contributed to reducing disparities in diabetes-related hospitalization rates between Latinos and Whites. The analyses indicated moderate evidence that consistent ACO participation is associated with lower health disparities (t = −1.947, P = 0.0525). However, this association is not significant after balancing covariates, and no causal relationship can be established. Latinos compose one of the fastest growing groups in rural as well as urban areas of the United States. It is critical that ACOs, with their emphasis on care coordination, health care quality, and value, monitor their provision of services to Latinos, rural, and other vulnerable populations.

Introduction

Adeeper understanding of health disparities has grown from our shared experiences of living through a pandemic. “Health disparities” may be defined as “…a health difference that adversely affects disadvantaged populations, based on categories of health outcomes.”¹ The general population has become more aware of how racial and ethnic minorities, rural populations, and other underserved groups are disproportionately affected by COVID-19, as well as some chronic and mental health conditions.

The complementary goals of reducing health disparities and achieving “health equity” among all populations within the United States are being addressed on several levels. On the federal level, the Affordable Care Act (ACA) facilitates access to health care by expanding health insurance coverage and promoting health care delivery models that focus on coordinated care.² On the state level, Medicaid expansion has increased access to health care services and facilitated the reduction of disparities in many states.

On the health care system level, new models of health care delivery seek to minimize poor health outcomes and reduce health care costs. One of these models is the accountable care organization (ACO). Although ACOs exist in both the public and private sectors, the current study focused on ACOs designed for Medicare beneficiaries. Medicare ACOs are defined as “…groups of doctors, hospitals, and other health care providers, who come together voluntarily to give coordinated high-quality care to their Medicare patients.”³ Although ACOs have no explicit goal of reducing health disparities, they have a goal of improving quality of care. When primary care organizations join ACOs, the coordination of care among member health care partners may contribute to improved health outcomes in the long term.

The focus of this study was on a potentially doubly disadvantaged population: Latinos who reside in rural areas of the United States. Compared with other groups, US Latinos have high percentages of overweight or obesity, which are among the risk factors associated with type 2 diabetes.⁴ Compared with Blacks/African Americans and Whites, they have lower percentages of insured persons,⁵ which complicate their financial access to diabetes prevention and management services. In addition, Latinos in rural areas share the same geographic access barriers and other disadvantages long experienced by rural populations as a whole. Compared with urban residents, rural residents in the United States are older, experience higher rates of chronic disease, and have some of the highest rates of adverse health conditions.⁶

The researchers analyzed data for Latino and White Medicare beneficiaries served by rural health clinics (RHCs) located throughout the United States. Today's ∼4500 RHCs nationwide offer primary care services in underserved rural areas.⁷ In recent years, more RHC patients are becoming participants of ACOs. Of the many types of ACOs, RHCs are more likely to participate in Medicare Shared Savings Program ACOs. As of 2021, 1397 RHCs (or ∼31%) were participating in ACOs of this type.³

The purpose of this study was to ascertain the contribution of prevention and management services provided by RHCs in Medicare ACOs on health care outcomes of patients diagnosed with diabetes. To assess the relationship between ACO participation and health disparities of rural Latinos, the following research question was posed:

What factors, including Rural Health Clinic (RHC) participation in Medicare ACOs, are related to reduced disparities in diabetes-related hospitalization rates of rural Latino older adult patients?

By means of quantitative analyses, the researchers assessed the impact of several factors on disparities in diabetes-related hospitalization rates between rural Latino and rural White older adults while controlling for sociodemographic factors such as the percentage of persons of age 65 years and over and the percentage in poverty. Data were analyzed for Latino Medicare beneficiaries who were served by RHCs during an 8-year period.

Related literature

Diabetes-related hospitalizations in rural areas

Although the prevention, diagnosis, and management of diabetes have improved over time in the United States, diabetes mortality remained steady in rural areas for almost 2 decades before 2016.⁸ Over the 7-year period ending 2015, Ferdinand et al⁹ found the highest odds of diabetes-related hospital deaths to have occurred in micropolitan and noncore or rural-remote areas.*

In a study of hospitalizations for ambulatory care sensitive conditions (ACSCs), including diabetes-related conditions, adjusted ACSC rates were 45% greater for rural counties than for urban counties in 8 states throughout the United States.¹⁰ Other studies indicate that Medicare beneficiaries in rural areas were less likely to have a readmission, although necessary readmissions may not occur because of lack of access to follow-up health care services.¹¹

RHCs and community health centers (CHCs) are valuable providers of primary care in the US health care system. In rural counties, they may contribute to reducing the rate of avoidable hospitalizations. In 1 examination of counties where both RHCs and CHCs were located, the avoidable hospitalization rate was 88% that of counties where neither was present.¹² In their longitudinal analysis of rural patients served by RHCs, Wan et al¹³ analyzed diabetes-related hospitalization rates for a 7-year period ending 2013, and compared rates before and after the implementation of the ACA. Overall, hospitalization rates declined slightly during the study period.

ACO performance and impact on disparities

A few studies examine the relationship between ACOs' patient base composition and quality performance. Early in the history of ACOs, RHC participation in Shared Savings Program ACOs was not found to contribute to reducing diabetes-related hospitalizations of older adult patients served by RHCs.¹⁴ Elsewhere, using Medicare ACO quality performance measures, Lewis et al¹⁵ found that the higher the ACO's percentage of patients in racial/ethnic minority groups, the lower the Medicare Shared Savings Program ACO performed on quality measures during the first 2 years of contracts.

Of the few studies and reports on the topic of ACOs as related to disparities, several are descriptive of disadvantaged or vulnerable populations in general,¹⁶ or descriptive of the disparities between vulnerable populations and the general population.¹⁷ To address the needs of vulnerable populations and reduce disparities, suggestions for policy changes to make ACOs more effective have also been described.¹⁶

Anderson et al¹⁸ analyzed the disparities in ACO quality on 2 dimensions. First, they analyzed the relationship between ACO size (as measured by the number of provider group members) and racial disparities in quality. Second, using ACOs of larger size, the relationship between quality and disparities was analyzed. Among ACOs of larger size, better quality for White beneficiaries (as measured by diabetes- or cardiovascular disease-related ACSC hospitalizations) was associated with smaller racial disparities, although no correlation with disparities was found for other measures.

Methods

The study was approved by the University of Central Florida's Institutional Review Board (SBE-17-13475).

Design

This longitudinal study was designed to (1) compare health disparities and patient outcomes of rural Latino older adult patients diagnosed with diabetes with their non-Latino White counterparts and (2) ascertain the impact of ACO participation by rural primary care providers (along with other factors) on rural Latino older adult patients.

The investigation was the second of 2 phases of a multiyear project. An adaptation of the Donabedian¹⁹ structure-process-outcome framework guided the development of the study variables. The variables were organized into the following categories: “context” (sociodemographic and other variables that describe the environment that affects the RHC patients); “structure” (human and material resources that describe the attributes of the health care setting); “process” (preventive and management services that describe the primary care services provided); and “outcomes” (the result or effect of the health care provided).

The framework for the study is illustrated in Figure 1.

FIG. 1.

Framework for assessing diabetes-related disparities of rural Latinos.

The unit of analysis was the individual RHC (or “clinic”). The researchers created a large data set containing data on the RHC organizational characteristics, their service area, and their patients. All data were aggregated to the individual clinic level.

Samples

Using the provider of services data files,²⁰ a panel of 2683 clinics continuously certified as RHCs throughout the study period of 2008–2015 was created. During this period, some of the RHCs were Medicare ACO participants; others were not. From this panel, a subset of 516 RHCs located in 3 states—California, Florida, and Texas—was extracted. These study states were selected based on the following criteria: (1) each state represents a different health and human services region, (2) each had higher number of RHCs than other states, and (3) each had higher number of Hispanic/Latinos in their rural areas than many other states.

Data sources and variables

In this section, data sources and construction of the variables used in this investigation are described. First, the researchers' existing data set on RHC characteristics, service area, and patients was expanded. The resulting data set contains data for 536 variables. The data sources were as follows: for context variables—the Area Health Resource Files²¹; for structure variables—the Medicare Cost Reports^22–25; and for the process and outcome variables—the Centers for Medicare and Medicaid Services' Chronic Condition Data Warehouse (CCW).⁷

“Structure” variable of interest: ACO

The structure variables measure organizational structure and operational practices that are (theoretically) under management control. Using dummy variables, each RHC in the panel was categorized as a participant or nonparticipant in either 1 of 2 types of Medicare ACOs: Pioneer ACOs or Medicare Shared Savings Program ACOs (where for each ACO type, 1 = ACO participant and 0 = nonparticipant). The data for these variables were combined into 1 variable, “ACO,” for each study year.

“Outcome” variable: disparities in diabetes-related hospitalization rates

To determine the hospitalization rates for Latinos and Whites, the researchers used data from the CMs' CCW. Although data from this and other sources use the term “Hispanics,” the current study used the term “Latinos” to describe this ethnic group.

In constructing the race variable for the data files, the CCW team uses the Research Triangle Institute (RTI) race categories “…by taking the beneficiary race code that has historically been used by the Social Security Administration (and is in turn used in CMs' enrollment data base) and applying an algorithm (the RTI race code) that identifies more beneficiaries as Hispanic or Asian.”²⁶ The RTI race variable has been found to be very accurate for identifying Hispanics and non-Hispanic Whites in studies of health disparities.²⁷

The following definition guided the measurement of disparities for the current study: “the quantity that separates a group from a specified reference point on a particular measure of health that is expressed in terms of a rate, percentage, mean, or some other quantitative measure.”^28(p.3) The first step in creating the disparities variables was to construct a set of patient-related outcome variables called “ACSC—Diabetes.” This set of variables describes the ACSC rate for diabetes-related admissions.

The variables were first calculated as the risk-adjusted expected number of hospital discharges with ICD-9-CM principal diagnosis code for diabetes of the RHC's beneficiaries, divided by the number of the RHC's beneficiaries with outpatient diagnosis of diabetes. The final rate was computed for each year by dividing an expected number of admissions by the actual number of admissions for each RHC.

The second step in creating the disparities variables was to calculate the “absolute” and “relative” measures of disparities as recommended by current literature.^28,29 Whereas the “absolute” measure of disparity is the arithmetic difference between a group rate and a specified reference point, the “relative” measure is the difference between rates in terms of the reference point. For purposes of this investigation, the absolute disparity was used as the dependent variable. It was calculated as the difference in diabetes-related hospitalization rates of older Latino beneficiaries served by RHCs as compared with their White counterparts.

Analytical approaches

Two statistical methods were used with the pooled data for the years 2008–2015 to analyze the absolute disparity in diabetes-related hospitalization rates between rural Latino older adult Medicare beneficiaries and their White counterparts. First, a difference-of-means test was conducted to determine whether there was a change in disparity from the Pre-ACO period (2008–2011) to the Post-ACO period (2012–2015). For this method, the disparities between Latino and White patients served by RHCs that became participants in either Medicare's Pioneer or Shared Savings Program ACOs were included in the analysis.

Second, regression analyses were conducted using the disparity in diabetes-related hospitalization rates as the dependent variable. Several covariates were included in the models as measures of context, structure, and process. The key covariate was a binary variable indicating participation in a Medicare ACO. A significant negative coefficient on the ACO variable would indicate that participation in a Medicare ACO decreased the disparity in diabetes-related hospitalizations of Latinos as compared with Whites.

In developing the final regression models, bivariate analyses were conducted to determine the relationship between the outcome variable and each of the predictor variables. Individual variables and the overall relationship between the variables were analyzed to ensure that the basic assumptions of regression analysis were met. Individual variables were transformed when necessary before regression analysis was conducted.

Multiple regression analysis of 3 separate models was conducted. The first model analyzed data for 2 groups of RHCs: those with ACO participation in 1 or more of the study years, and those with no ACO participation during any of the study years. The second model analyzed data for 2 groups of RHCs: those that continuously participated in ACOs during the 3 consecutive years 2013–2015 and those that did not continuously participate in ACOs during that same 3-year period. The third model included entropy-balanced covariates for the 2 groups.

A list of the study variables in the final model and their definitions is given in Table 1.

Table 1.

Study Variables and Definitions

Variable name	Definition
ACOSSPa	RHC that continuously participated in a SSP ACO during 2013–2015 (dummy variable where 1 = continuous SSP ACO participant; 0 = not a continuous SSP ACO)
Provbsd	Dummy variable (1 = provider-based RHC; 0 = independent RHC)
AgeRHC	Years Medicare certified for participation in the RHC program
RUCACat	Four RUCA categories: urban, large rural, small rural, isolated. 1 = urban; 2 = large rural; 3 = small rural; 4 = isolated. Based on WWAMI's 4-tiered classification.
Older08b	% of county population that is Medicare eligible 2008
Pov08b	% of county population that is at 200% of poverty level 2008
White08b	% of county population that is White 2008
Hispanic08b	% of county population that is Hispanic 2008
PhysPop08b	No. of active GP+FP+DO physicians per 1000 population in county 2008
PerDiab08b	No. of RHC's beneficiaries with HCPCS_CD for diabetes screening divided by total No. of RHC's beneficiaries

^a ACOSSP is constructed using SSP ACO beneficiary file and outpatient claims file. An RHC is an SSP ACO participant (ACOSSP = 1), for a given year, when a beneficiary (65+) participating in an SSP ACO receives services in that same year from the RHC.

^b The “08” extensions on many of the variables refer to 2008. These variables appear in the data set for each of the years 2008–2015 with extensions for the corresponding year.

ACO, accountable care organization; RHC, rural health clinic; RUCA, Rural-Urban Commuting Area Code; SSP, shared savings program; WWAMI, Washington, Wyoming, Alaska, Montana, Idaho Rural Health Research Center.

Results

Change in disparity over time

To determine whether the disparity in diabetes-related hospitalization rates between rural older Latinos and Whites had changed over time, the pre-ACO period (2008–2011) was compared with the post-ACO period (2013–2015). A statistically significant decrease in disparity over time was found (t = −7.6899, df = 115, P < 0.001).

Impact of ACO participation versus nonparticipation on disparities

During the study period, 420 RHCs had participated in a Medicare ACO for 1 or more years, whereas 96 RHCs had not participated in these ACOs during any of the study years (2008–2015). The 2 groups of RHCs were compared to determine whether ACO participation contributed to reducing disparities in diabetes-related hospitalization rates between Latinos and Whites. Based on the multiple regression analysis, ACO participation was not a significant factor in reducing disparities.

Impact of continuous ACO participation on disparities

Since much of the data on disparities for the non-ACO participant RHCs were missing, 2 additional groups of RHCs were formed for further analysis: those that consistently participated in Medicare ACOs for 3 consecutive years and those that did not consistently participate in these ACOs. During the study period, 222 RHCs had continuously participated in an ACO for 3 consecutive years and 94 had not continuously participated in an ACO.

To identify the determinants of disparities in diabetes-related hospitalizations, the following independent variables were used in a multiple regression model: ACO (dummy variable where 1 = consistently in an ACO and 0 = not consistently in an ACO), older, poverty, White, Hispanic, “physpop,” “perDiab,” Provider-based, Rural-Urban Commuting Area Code (RUCA), and age of RHC (See Table 2). Of these, the following independent variables were significant: ACO (t = −1.947, P < 0.10) and poverty (t = −1.732, P < 0.10). Thus, there was moderate evidence that consistent ACO participation is associated with lower disparity.

Table 2.

Multiple Linear Regression Analysis

Variables	Unstandardized coefficients		t	P
	B	Standard error
Constant	−0.009	0.017	−0.568	0.570
ACO	−0.005	0.003	−1.947	0.053
Older	−0.014	0.030	−0.467	0.641
Poverty	−0.046	0.026	−1.732	0.084
White	0.021	0.020	1.067	0.287
Hispanic	0.005	0.008	0.632	0.528
PhysPop	−0.006	0.008	−0.776	0.438
PerDiab	0.006	0.055	0.111	0.912
Provbsd	−0.003	0.002	−1.090	0.277
RUCACat2	0.002	0.004	0.616	0.538
RUCACat3	−0.0002	0.003	−0.075	0.941
RUCACat4	0.0005	0.005	0.105	0.917
AgeRHC	−0.0002	0.0002	−0.879	0.380

Multiple R-square: 4.43%; adjusted R square: 0.64%.

ACO, accountable care organization.

Entropy balancing on the ACO variable was also conducted. Entropy balancing is a data preprocessing method used to achieve covariate balance with a binary treatment of the ACO variable (dummy variable where 1 = consistently in an ACO and 0 = not consistently in an ACO). This preprocessing method adjusts for inequalities in representation of the moments of covariate distributions.³⁰ The reweighted data were analyzed in a regression. The entropy-balanced regression showed no significant evidence of a causal relationship between consistent ACO participation and lower disparity.

It is important to note that the majority of RHCs that were not ACO participants were also those that had no data for the outcome variable “diabetes-related hospitalization rates.” The lack of data to construct this variable resulted from 1 of the following conditions: (1) no diabetes-related hospitalizations for the RHC's Latino patients during 1 or more of the study years or (2) the RHC's not having treated Latino patients during 1 or more of the study years.

Discussion

The analysis of 8 years of data for 516 RHCs enabled the researchers to compare the diabetes-related hospitalizations for Latino patients with those of White patients served by RHCs for the years before and after participating in an ACO. In addition, the contribution of ACO participation to reducing health disparities between Latino and White older adult patients in rural areas was analyzed.

First, the analysis of the number of diabetes-related hospitalizations pre- and post-ACO participation by RHCs revealed a decrease in disparities between Latino and White hospitalization rates. This trend runs parallel to the overall decline in diabetes-related hospitalizations in the United States during roughly the same time period.⁹ This decline is also consistent with the decline in diabetes incidence (rate of new cases of diagnosed diabetes) after 2009.³¹

There are several possible explanations for the decline in disparities between Latinos and Whites as demonstrated by this investigation. The focus of this study was on rural Latino residents served by RHCs. Latinos are among the ethnic groups at greatest risk for developing diabetes and its complications. Diabetes management requires geographic access to health care services, knowledge of and adherence to preferred diets and exercise, and sufficient financial resources. The Medicare Latino patients served by RHCs may have been able to circumvent some of these risk factors and achieve better health outcomes, including those related to diabetes.

RHCs promote diabetes education and the self-management practices that may contribute to the decline in health disparities of their Latino patients. They must meet quality guidelines according to federal regulations governing their certification and recertification, and maintain quality standards to be in good standing with the Medicare/Medicaid programs. Many RHCs have employed the services of community health workers to educate and monitor patients on diabetes self-management practices.

Many have become patient-centered medical homes (PCMHs) that seek to improve the quality of services, access to those services, efficiency, and patient satisfaction. Whether formally designated as PCMHs or not, RHCs facilitate the physician–patient (or nurse practitioner/physician assistant-patient) trust needed to reinforce adherence to diabetes management recommendations and minimize diabetes-related hospitalizations.

Diabetes incidence and some diabetes outcomes may be more favorable for rural residents, including rural Latinos. Some studies have shown that Hispanics living in rural areas have lower diabetes rates than their urban counterparts.³² Elsewhere it was found that although overall hospitalization rates of adults with diabetes were lower in rural counties, the ACSC rates were similar in mostly rural as compared with mostly urban counties.³³

Other possible explanations for the decline of disparities relate to undiagnosed diabetes. Although Hispanics have among the highest rates of diagnosed diabetes among ethnic groups, there are some indications that a greater proportion of Latinos remain undiagnosed with diabetes. Of the 7.3 million adults with undiagnosed diabetes in 2018, 1.5 million were Hispanics—making up the second highest number after Whites (with 4.1 million).³⁴ For 2018, the ratio of Hispanics with undiagnosed diabetes to those with diagnosed diabetes was slightly over 1 in 3, whereas for Whites closer to just 1 in 4 remained undiagnosed.^† This possible explanation seems unlikely as applied to the findings of this investigation, however, as data were analyzed for patients under the care of RHCs that routinely provide diabetes prevention and management services.

The second part of the study—the analysis of the contribution of ACO participation to reducing health disparities between Latino and White older adult patients in rural areas—led to mixed results. When health disparities of ACO-participant RHCs were compared with non-ACO participants, ACO participation was not found to be a significant factor in reducing disparities. This is consistent with previous research that has found the risk-standardized acute admission rates of patients with diabetes for almost 45% of ACOs were no different from the national rate.³⁵

However, continuous participation in an ACO for a period of 3 years was associated with lower disparity. ACOs have complex organizational structures. They are often composed of providers from different levels of the health care system, such as hospitals, primary care organizations, and skilled nursing facilities. Many serve multiple states. Collaboration among the ACO's members, and meeting federal standards on measures of cost, quality care, and the patient experience, requires time for modifying clinical protocols and communication practices, training staff, building infrastructure, and the like.

Previous research indicates that much of the immediate benefit of participating in a collaborative organizational model is the shared learning among member organizations rather than on patient and financial outcomes.³⁶ It is possible that 3 or more years must elapse before ACOs have measurable impact on health outcomes and health disparities.

Strengths and limitations

Statistical analyses in previous studies of health disparity trends have been complicated by the transition from a few to multiple race categories in government data collection tools. A strength of this study analysis was its use of CCW data with its self-reported descriptors for race and ethnicity and implementation of the RTI race code.

The statistical analysis was hindered, however, by missing data for the dependent variable (disparities) for the vast majority of non-ACO participant clinics. That is, either the clinic had no Latino beneficiary patients and/or no White beneficiary patients with diabetes-related hospitalizations during the study year. One explanation for the lack of patients of either or both ethnicities could be that the clinic had no beneficiaries who were diagnosed with diabetes during 1 or more of the study years.

Given the prevalence of diabetes among older persons, and those of Latino ethnicity in particular, this explanation seems unlikely. More data on factors common to non-ACO RHCs might yield further insights. As ACO participation by RHCs increases, the volume of data will increase to further clarify whether ACO participation contributes to decreasing health disparities in rural populations.

Finally, this investigation was limited to diabetes-related health outcomes of Medicare beneficiaries. Uninsured Latinos and those having Medicaid only were not included in the data analyzed. It is possible that the decline of disparities did not occur during the study years for Latinos in these insurance categories.

Conclusions

This longitudinal study compared the diabetes-related hospitalization rates of rural Latino older adult patients served by RHCs with their White counterparts over an 8-year period. In addition, the impact of RHC participation in ACOs on reducing diabetes-related disparities between Latino and White patients was examined. The analysis led to 2 broad conclusions. First, the disparities between Latino and White diabetes-related hospitalizations did decrease over the 8-year study period. Second, there was no statistically significant evidence that ACO participation was causal in the decrease in disparities. These findings are based on data from the very first years of Medicare ACOs. Future studies will strengthen the knowledge about the contribution of ACOs to reducing disparities.

Latinos compose one of the fastest growing groups in rural as well as urban areas of the United States. Organizational structure of health care provider organizations, health care delivery practices, and other factors that may improve their health care and health outcomes must continue to be evaluated and monitored. It is critical that ACOs, with their emphasis on care coordination, health care quality, and value, continually improve their provision of services to Latinos, rural, and other vulnerable populations.

Authors' Contributions

Dr. Ortiz was responsible for the conceptualization, funding acquisition, data curation, project administration, and supervision for the study. Drs. Ortiz and Hofler were responsible for the methodology of the study. Dr. Ortiz was responsible for the original draft of the published work. All authors were responsible for the formal analysis of the study data and the review of the published work.

Disclaimer

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

The analysis for this article was supported by the National Institute on Minority Health and Health Disparities of the National Institutes of Health under Award No. R15MD011663.