Change in geographic access to community health centers after Health Center Program expansion.
Community health centers (CHCs) are vital in the provision of primary care for the poor and underserved in the United States, serving 25 million patients at over 11 000 CHC sites. (1,2) Patients who use CHCs are disproportionately poor and underserved--92 percent have incomes at or below 200 percent of the federal poverty level, 62 percent are minorities, and 73 percent have Medicaid or are uninsured. (2) CHCs are patient-directed, community-based clinics that offer primary care and ancillary services (eg, transportation, health education, and language translation) to all individuals regardless of ability to pay, (3) and work to address social determinants of health locally. (4,5) Increased access to CHCs is associated with favorable health outcomes including reductions in preventable hospitalizations (6,8) and age-adjusted mortality. (9)
The Health Resources and Services Administration's (HRSA) Health Center Program (HCP) funds and monitors CHCs. One way the HCP increases primary care access in underserved areas is through funding the establishment of new, permanent, full-time CHC service delivery sites. (10) In 2009, the Obama Administration substantially invested in new CHC sites through the American Recovery and Reinvestment Act (ARRA), (11) allocating $500 million to establish 126 new sites. (12) Subsequently, the Affordable Care Act (ACA) of 2010 allocated $9.5 billion between 2011 and 2015 to new sites and service expansions, (13) and hundreds more CHC sites were established as a result.
There are important limitations on where CHC sites (subsequently referred to as "CHCs") are placed. Currently, HRSA mandates that they be established in medically underserved areas, a designation given primarily to counties based on the index of medical un-derservice (IMU), which considers the proportion of the population in poverty, primary care physician-to-population ratio, proportion of the population over 65, and infant mortality rate. (14) Areas with IMU values at or below 62 are eligible for CHC placement. This designation system may not best represent areas in need of CHC services for several reasons. First, the IMU threshold value of 62 was computed in the 1970s and has not been nationally updated. (15) Second, the designation is a dichotomous categorization that does not distinguish different levels of medical underservice. In theory, all areas within a medically underserved county are equally eligible for CHC placement. Additionally, using counties as the geographic unit for designation masks within-county variation in underservice, which could be substantial. Counties contain an average of 100 000 people. (16) Census tracts are geographic subunits of counties, approximately equivalent to neighborhoods in urban areas, that contain an average of 4000 people. (17) Research by Wright et al (18) showed that the IMU value for Denver County, Colorado, which comprises 181 census tracts, was 79.8, compared to a portion of the county comprising 54 census tracts, which had an IMU value of 55. Another limitation is that the IMU does not account for sociodemographic factors, beyond poverty and age, known to be associated with disparities and therefore indicative of increased potential need. For example, area-level factors, including racial and ethnic composition, educational attainment, unemployment rate, and insurance rate, have been associated with poor health care access, utilization, and outcomes. (19-23)
Establishment of new CHCs in underserved areas is intended to improve access by easing geographic barriers to affordable primary care, that is, improving the geographic accessibility of CHCs within these areas. Geographic accessibility refers to the geographic alignment of supply and distribution of health services in relation to the distribution of people with potential need for those services. (24) Prior studies have associated geographic accessibility of health services with improved health care quality, (25,26) perceived access, (27) and utilization, (28,29) as well as favorable health outcomes. (6-8,30)
Commonly used CHC accessibility measures include the county-level CHC-to-population ratio and distance to the nearest CHC. These measures are easily interpreted, but each has major weaknesses. The county-level CHC-to-population ratio conceals potentially meaningful variation in CHC accessibility within counties and assumes patients living in the county only seek services in that county, even if they reside closer to a CHC in an adjacent county, for example. Distance to nearest CHC disregards the number of patients potentially demanding care from that CHC, which impacts the accessibility level of that particular CHC. The two-step floating catchment area (2SFCA) method (31) overcomes these weaknesses by estimating the supply-to-demand (CHC-to-population) ratio within a user-specified catchment area. The method incorporates the number of potential patients in each CHC's service area, can be estimated at the census tract level, and considers all CHCs within a predefined travel distance as accessible, even if located in an adjacent county. Since its development, researchers have widely applied the method to study a variety of health services topics, including the effect of primary care physician, clinic, and hospital accessibility on preventable hospitalizations (30); disparities in primary care physician accessibility among immigrant populations and linguistic minorities (32); and the effect of primary care and mammography accessibility on late stage breast cancer diagnosis and treatment. (26) To our knowledge, however, it has not been used to examine CHC accessibility although, as we argue in the methods, it is a stronger measure than those used previously that tend to underestimate CHC accessibility.
In this study, we used the 2SFCA method to calculate CHC accessibility at the census tract level before and after Obama-era policies that funded the establishment of many new CHCs in underserved areas. The two time points used were 2008, before ARRA funding, and 2016, after ACA funding. The objectives of this study were to 1) describe the extent of CHC accessibility across census tracts in 2008 and 2016 and 2) assess the extent to which tract-level indicators of potential need were associated with the extent of change in CHC accessibility from 2008 to 2016. We hypothesized that extent of change in CHC accessibility would be significantly associated with indicators of potential need.
2 | METHODS
2.1 | Study area
We focused this study on Arkansas, Louisiana, and Mississippi, states that are contiguous and have similar demographic makeup, including a higher proportion of African Americans, lower proportion of Hispanics, lower median income, and higher poverty rates than the national average. (33) A high proportion of low-income residents meant that a large share of the population comprised potential CHC patients. All three states ranked poorly on measures of health and health disparities (34,35); therefore, information gleaned on improvements or gaps in CHC accessibility could inform efforts to reduce disparities in the region. Almost all counties (90 percent) in these states were classified as medically underserved areas under current HRSA criteria, (36) so examining CHC accessibility by census tract could yield valuable insight of within-county variation. Finally, the three states' Medicaid expansion status varied. Arkansas was one of the few Southern states to adopt the ACA's Medicaid expansion, Louisiana expanded at the end of the study period in 2016, and Mississippi did not expand.
2.2 | Data sources
All data were obtained from publicly available sources. CHC location was obtained from HRSA, and sociodemographic data from American Community Survey (ACS) 5-year estimates. (17) We also used data from the Dartmouth Health Atlas Primary Care Service Area (PCSA) Project to construct a provider supply variable. Census tracts were the unit of analysis, and all census tracts in the states of Arkansas (n = 686), Louisiana (n = 1148), and Mississippi (n = 664) were eligible for inclusion. We excluded from our analysis census tracts with an estimated population of 0 from 2010 to 2014, for example, census tracts that comprised bodies of water or airports (n = 30; 1.2 percent).
2.3 | CHC accessibility
Community health center accessibility was estimated using the 2SFCA method, (31) which has been extensively used in studies of health services. (22,26,37-39)
The 2SFCA method produces a stronger measure of CHC accessibility and captures accessibility not detected by cruder measures. Figure 1 shows CHC-to-population ratio levels in Mississippi in 2008 comparing the 2SFCA measure to two cruder measures. The crudest measure, Measure A, shows the CHC-to-population ratio (number of CHCs per 10 000 people) in each county; Measure B shows the CHC-to-population ratio in a 30-minute drive time catchment area from the population-weighted center (centroid) of each census tract, a standard used in other accessibility studies (39-41); and Measure C shows the 2SFCA measure. Measure B unmasks within-county variation not detected by Measure A and uncovers accessibility in Newton County, located at the bottom right of the area contained in the box in Figure 1, that Measure A misses. Measure C reveals even higher levels of CHC accessibility than Measure B in all four counties contained in the box in Figure 1 and shows substantial accessibility in many counties determined by Measure A as lacking CHC accessibility (ratio of 0 CHCs per 10 000 people).
Community health center accessibility was calculated as a supply-to-demand (CHC-to-population) ratio within the census tract's catchment area. Higher values indicate better accessibility, and values of 0 indicate no accessible CHCs within the census tract's catchment area. Details can be found in the Appendix S1. Briefly, for the numerator, we used a dataset of all CHC service delivery sites overseen by the HCP from the beginning of the program through 2016 that included the geocoded location of each site, site type (eg, homeless, mobile clinic, school-based health center), and date the site opened. We included only sites that (a) provided primary care (eg, excluded dental and behavioral health clinics), (b) served the general population (eg, excluded women only clinics and school-based health centers), (c) were brick and mortar clinics (eg, excluded mobile sites), and (d) were open year-round (eg, excluded seasonal sites). We used attributes in the dataset to code each site according to inclusion criteria, then had CHC advocacy organizations in each state verify that the sites met our criteria, any site closures were captured, and addresses were current. We used sites that opened in or before 2008 to calculate CHC accessibility in 2008 and sites that opened in or before 2016 to calculate CHC accessibility in 2016.
For the denominator, we restricted the population to those most likely to use CHCs for care: low-income nonelderly adults. Low-income nonelderly adults, defined as those between the ages of 18 and 64 with incomes below 200 percent of the federal poverty level, constitute the majority of CHC patients. (2) We took these estimates from ACS 5-year periods that aligned as closely as possible to 2008 and 2016, respectively, to account for population shifts and ensured both periods used geographic boundaries from the 2010 Census. We used 2006-2010 ACS estimates for the 2008 denominator and 2011-2015 estimates for the 2016 denominator. The census tract centroid was used to represent the location of all low-income nonelderly adults in the tract.
We calculated accessibility estimates within catchment areas around the census tract equivalent to the drive time distance covered in average work commuting time (in minutes). Work commuting patterns may be similar to commuting patterns to obtain health care (42); therefore, average commute time may be a stronger proxy for defining catchment areas around census tracts than the 30-minute standard. We sorted tracts into four groups using commuting pattern codes from the Rural-Urban Commuting Area (RUCA) classification system. (43) RUCA classifications categorize tracts according to population density and commuting destinations of the tracts' residents, for example, if most residents commute locally or to a larger neighboring area for work. Then, using ACS data, we calculated the average commute time for tracts in each group, which were 23.5, 29.0, 29.4, and 32.3 minutes, depending on the group. We allowed catchment areas to vary by group, so a tract's catchment was defined to be the area contained in either a 23.5, 29.0, 29.4, or 32.3-minute drive from its centroid. Additional details can be found in the Appendix S1. We used drive time distances because driving is the most ubiquitous mode of travel in the study area--91 percent of the population in these states have access to a vehicle (33) and 93 percent commute to work by car, truck, or van. (44)
As a sensitivity analysis, we estimated CHC accessibility in Mississippi using the 30-minute standard for census tract catchment areas as an alternative to average commute time. Additionally, we calculated change in CHC accessibility in Mississippi using CHC-to-population ratio in the county and within a 30-minute standard catchment area around the census tract to determine to what extent the results differed from our 2SFCA approach.
2.4 | Indictors of potential need
Using 2010-2014 ACS data, we computed the proportion of low-income nonelderly adults in the tract who were (a) below poverty, (b) unemployed, (c) uninsured, (d) black or African American, (e) Hispanic, and (f) did not graduate high school (see Table S2 in Appendix S1 for more information). We chose a 5-year period between pre and post time points because we expected these characteristics to inform CHC site placement at both times. Primary care provider (PCP) supply was defined as PCP-to-population ratio--a criterion HRSA uses to indicate areas in greater need of CHCs (14)-within the census tract's catchment area, such as in other studies. (45) We obtained the number of clinically active primary care physicians, physician assistants, and nurse practitioners in each census tract catchment area using 2009-2012 provider data from the Dartmouth Health Atlas PCSA Project version 3.1.
2.5 | Analysis
Community health center accessibility values were mapped in 2008 and 2016 using ArcGIS Pro version 2.1.1 (ESRI, Redlands, CA). We used multivariable regression to assess to what extent indicators of potential need were associated with change in CHC accessibility from 2008 to 2016. Our outcome was change in CHC accessibility from 2008 to 2016, and the main predictors were indicators of potential need: the six sociodemographic characteristics and PCP supply. We included baseline CHC accessibility (value in 2008) as a covariate, reasoning that census tracts with greater baseline accessibility would change less from pre to post. We examined descriptive statistics and bivariate associations and tested for collinearity among predictors by examining condition index and variance inflation factor values. After ruling collinearity out, we tested for spatial autocorrelation using the Global Moran's I statistic, specifying a connectivity matrix that classified census tracts as neighbors if their geographic boundaries shared edges or corners. Our final model was a spatial error model, which accounts for spatial autocorrelation and adds a parameter, lambda, to indicate the extent to which the spatial component of neighboring observations' error terms are correlated. (46) A comparison of Akaike information criterion (AIC) values for the spatial error model and an ordinary least-squares regression model confirmed that the spatial error model was the best fit. Analyses were performed with GeoDa version 1.12 (Center for Spatial Data Science, Chicago, IL) and SAS version 9.4 (SAS Institute, Cary, NC).
For sensitivity analyses, we assessed model parameters with and without baseline accessibility included as a predictor to see the potential influence it had on other predictors in the model. We also examined the results of our model using indicators of potential need calculated as the proportion of all nonelderly adults--not just restricted to low-income nonelderly--in the denominator to ensure cross-tract variation in indicators of potential need was adequately captured. Finally, we tested our model with census tracts aggregated to Primary Care Service Areas, (47) which are geographic units much larger than census tracts but smaller than counties. None of these analyses substantially changed our results.
3 | RESULTS
Our analytical sample consisted of 2468 census tracts. Sample characteristics appear in Table 1. The average proportion of low-income nonelderly adults below poverty was slightly lower in census tracts in Arkansas compared to Louisiana and Mississippi. Mississippi tracts had a higher average proportion of low-income nonelderly adults who were unemployed, did not graduate high school, and identified as black or African American. Arkansas tracts had a higher average proportion of low-income nonelderly adults who identified as Hispanic or Latino. The average proportion of low-income nonelderly adults who lacked health insurance was slightly higher in Louisiana tracts.
Community health center accessibility values, indicating the number of CHCs per 10 000 low-income nonelderly adults within the tract's catchment area, ranged from 0 to 12.65. The interquartile range of values fell between 0 and 0.63 in 2008 and between 0.37 and 1.25 in 2016.
3.1 | Change in CHC Accessibility from 2008 to 2016
Overall, median CHC accessibility increased by 192 percent, from 0.26 CHCs per 10 000 low-income nonelderly adults in 2008 to 0.76 CHCs per 10 000 low-income nonelderly adults in 2016 (Figure 2).
Median CHC accessibility grew the most from 2008 to 2016 in Arkansas--from 0 to 0.47 CHCs per 10 000 low-income nonelderly adults--followed by Louisiana and then Mississippi. Mississippi census tracts, however, started with the highest median CHC accessibility in 2008 (0.59 CHCs per 10 000 low-income nonelderly adults) and ended with the highest median CHC accessibility in 2016 (1.02 CHCs per 10 000 low-income nonelderly adults).
A comparison of change in CHC accessibility in Mississippi from 2008 to 2016 using our 2SFCA approach and the two cruder accessibility measures showed that mean change in accessibility was substantially less using the cruder measures compared to the 2SFCA measure. It was 28 percent less using CHC-to-population ratio at the county level and 62 percent less using CHC-to-population ratio within a 30-minute standard catchment from the census tract (see Table S3 in Appendix S1). Also we found that mean and median CHC accessibility values were 24 and 29 percent greater, respectively, in Mississippi in 2008 when using a 30-minute standard catchment area instead of average commute time: mean and median values increased from 0.74 to 0.92 and from 0.59 to 0.76 CHCs per 10 000 low-income nonelderly adults (see Table S4 in Appendix S1).
Currently, we do not know what CHC accessibility levels are optimal for community health. But, we found substantially fewer census tracts in the study area had 0 CHCs per 10 000 low-income nonelderly adults in 2016 compared to 2008. Specifically, the number of census tracts with a CHC accessibility value of 0 dropped by about 20 percentage points, from approximately 33 percent in 2008 to 12 percent in 2016 (Figure 3). Arkansas had the largest decrease, from 51 to 21 percent, and Mississippi had the lowest proportion of tracts with a CHC accessibility value of 0 in both 2008 and 2016 (24 and 7 percent).
3.2 | Census tract-level variation in CHC accessibility by state from 2008 to 2016
Figure 4 shows the extent of CHC accessibility in 2008 and 2016 by census tract in Arkansas. Similar maps of Louisiana and Mississippi appear in Figure S1A,B in the Appendix S1.
In 2008, higher CHC accessibility ratios were concentrated in the eastern-central region of the state, from east of Little Rock to the state's eastern border along the Mississippi River. CHC accessibility values increased in the northeastern region of the state from 2008 to 2016, while values in many areas in the western half of the state remained at 0 in the same time period.
3.3 | Association between change in CHC accessibility and sociodemographic and PCP supply indicators of potential need
Results of collinearity tests showed all variance inflation factor and condition index values were below 2, indicating predictors in the multivariate model were not collinear. However, a high and statistically significant Moran's / value indicated the presence of spatial autocorrelation (coefficient = 48.39; P < 0.0001). The spatial error model had a better fit statistic (lower AIC value) than the OLS model assuming independent observations, and it adjusted for spatial autocorrelation (spatial error AIC = 3808; OLS AIC = 5644). Results of the spatial error model showed that, out of all predictors, baseline accessibility was the only predictor significantly associated with change in CHC accessibility from 2008 to 2016. Each increase of 1 CHC per 10 000 low-income nonelderly adults at baseline was associated with a 0.09 smaller change in CHC accessibility from 2008 to 2016 (estimate - -0.09; SE = 0.01; P < 0.0001).
We tested models for Arkansas, Louisiana, and Mississippi separately, allowing the relationships between predictors and outcome to vary by state. Our findings were consistent. In Louisiana, for example, baseline accessibility remained the only significant predictor, being negatively associated with change in CHC accessibility from 2008 to 2016 (estimate = -0.06; SE = 0.02; P = 0.01). Table S4 in Appendix S1 contains full regression results.
4 | DISCUSSION
This study assessed change in CHC accessibility from 2008 to 2016, after Obama-era policies that substantially increased funding for new CHCs, and documented within-county heterogeneity in accessibility by census tract in Arkansas, Louisiana, and Mississippi. CHC accessibility practically tripled across the study area, and the number of tracts with no accessibility in 2008 substantially dropped by 2016, indicating substantial improvements in geographic access to CHCs across the study area during this time. We further explored to what extent characteristics of census tracts indicating greater potential need for CHC services aligned with change in accessibility. We expected sociodemographic characteristics, such as a higher proportion of people in poverty, who are uninsured, and who are unemployed, and lower PCP supply to be largely associated with greater gains, but did not find evidence of such relationships. On the other hand, census tracts with higher levels of accessibility at baseline (in 2008) saw smaller increases, suggesting broad distribution of new CHCs rather than targeting specific areas according to potential need-based variables.
Across the three states, Arkansas showed the largest overall gain in CHC accessibility from 2008 to 2016. These gains in Arkansas could have been driven by efforts to bolster CHC capacity in preparation for a flood of new patients into the health care system following Medicaid expansion (11) and by the financial benefits to CHCs as the proportion of CHC patients with Medicaid increased, (48) as was shown to occur in Medicaid expansion states. (49) Expected increases in patient volume and proportion of Medicaid patients may have spurred the establishment of CHCs in new areas across the state, thereby increasing CHC accessibility for low-income nonelderly adults.
FIGURE 3 Proportion of census tracts with 0 community health centers (CHCs) within average commuting distance in Arkansas, Louisiana, and Mississippi in 2008 and 2016 Proportion in 2008 Proportion in 2016 3 States Overall (n=2,468) 33.27% 11.55% Arkansas (n=684) 50.88% 21.35% Louisiana (n=l,127) 28.13% 8.43% Mississippi (n=657) 23.74% 6.70% Note: CHC accessibility represents the number of CHCs per 10000 low-income nonelderly adults within the area covered within average commuting distance from the population-weighted centroid of the census tract. Average commuting distance was determined by using the Rural-Urban Commuting Area classification system to group tracts into four groups by commuting pattern codes and calculating the average commute time (in min) for tracts in each group. Commute time data were from American Community Survey 2010-2014 5-y estimates, and average commuting times were found to be 23.5, 29.4, 32.3, or 29.0 min depending on the group. Drive time distances were assumed. Note: Table made from bar graph.
On the other hand, Arkansas also had the highest proportion of census tracts with no accessibility in the post period--20 percent had no CHC accessibility in 2016, defined as having no CHCs within average commuting distance from the census tract. This may not be surprising, or inefficient, if tracts with no accessibility also have no or few potential CHC patients residing there. However, we found that tracts with no accessibility had an average of 1007 low-income nonelderly adults residing in them. Since census tracts contain about 4000 people on average, (17) this is no small share. Two of HRSA's goals through the HCP are to improve access to care in underserved areas and improve health equity. (13) Our findings show that disparities exist in geographic access to CHCs for low-income nonelderly adults in certain tracts. CHCs in general provide care to one in four rural residents, (2) and lack of access is thought to explain part of the urban/rural disparity in age-adjusted mortality, with rural residents faring worse. (50) Improving CHC accessibility in areas with none, therefore, could contribute to efforts to reduce these access disparities by improving access to affordable care provided by CHCs, and CHC establishment is associated with reductions in age-adjusted mortality. (9)
Median accessibility value was markedly higher in Mississippi's census tracts compared to those in the two other states in 2008 and 2016. Mississippi also had the smallest proportion of tracts with no accessibility at both time points. By 2016, only 6.7 percent of census tracts in the state had no CHC accessibility compared to 21.35 percent in Arkansas and 8.4 percent in Louisiana. Interestingly, Mississippi had the most and widest-spread CHC accessibility throughout the state, yet it was the only state included that did not expand Medicaid. A plausible explanation for this seeming contradiction--CHC expansion was intended to complement Medicaid expansion by building capacity to care for more Medicaid-covered patients--is that Mississippi focused attention on establishing CHCs, which provide care regardless of ability to pay, in a way that improved geographic access to care to compensate for not raising Medicaid eligibility thresholds. Access still expanded, just in a different way. In fact, Mississippi has long history of supporting CHCs-one of the first sites in the country was located in the Mississippi town of Mound Bayou. (51) Meanwhile, the state's legislators remain unsupportive of Medicaid expansion. (52) Future research exploring changes in CHC accessibility by states' Medicaid expansion status could inform whether Mississippi is a unique case.
Our finding that sociodemographic characteristics and PCP supply, used to indicate potential CHC need at the census tract level, were not significantly associated with accessibility gains suggests that examining heterogeneity in potential need within counties, many of which are designated medically underserved areas, could aid planning. HRSA has begun to embed some tract-level measures into mapping tools available to CHCs, (53) but incorporation of more tract-level indicators of potential need and accessibility metrics that use more sophisticated methods, like the 2SFCA method, would greatly strengthen identification of areas requiring additional CHC sites and services. This is particularly important because conventional county-level CHC-to-population ratios mask variation and substantially underestimate CHC accessibility and change in accessibility compared to the 2SFCA approach we used.
Another explanation for why we did not find an expected association between the indicators of potential need and gain in CHC accessibility is because the criteria and data HRSA uses to designate areas in need, that is, medically underserved areas, differs from those we used. HRSA uses the primary care physician-to-population ratio; proportion below poverty; proportion 65 or older; and infant mortality rate. (14) We included primary care physician assistants and nurse practitioners in our provider supply measure and did not include proportion of the population 65 and older, who are covered by Medicare. We examined additional sociodemographic characteristics of racial and ethnic composition, educational attainment, unemployment rate, and uninsurance rate. Also, we used recent data to calculate these measures. Substantial opportunities exist to incorporate new criteria and current data into the medically underserved area designation process to achieve HCP goals of improving health care access and equity.
Our analysis has several limitations. First, we assumed CHC site capacity was uniform. CHC capacity data, that is, number of providers, were not available by site. Research suggests, however, that CHC establishment regardless of capacity is associated with positive health effects in the surrounding area. (9) Second, we used ACS estimates as point estimates and did not account for margins of error. This may have over- or underestimated the number of adults used for accessibility calculations and proportions with certain characteristics. ACS 5-year period estimates, however, have the highest statistical reliability of all available estimates. (17) Third, we used HRSA's criteria to define CHC service areas, but CHCs may serve people living in a larger catchment area than that allows for. We recognize a larger catchment could be appropriate in this region with many sparsely populated areas. We also defined tract catchment areas based on average commute time, but cannot be sure if this is a strong proxy. Our results were fairly sensitive to these catchment area thresholds, and future research using this method should pay careful attention to selecting thresholds according to local context. In our study, the thresholds we selected and the decision to use drive times were informed by local commute and vehicle ownership data, but may not be appropriate to adopt for other areas, for example, areas with a larger share of public transportation users and less access to vehicles. Fourth, we included a measure of PCP supply to capture availability of primary care, but could not distinguish between CHC and non-CHC care providers. Some areas with low CHC accessibility, therefore, may have high non-CHC accessibility. However, CHCs are different in that they provide care regardless of ability to pay, offer robust ancillary services, and are entrenched in local culture, so CHC accessibility indicates access to a unique type of primary care. Finally, our analysis examined three states, so our results are not generalizable to the South or the nation. However, our findings could prove useful for state and local decision makers in these states, and our methodological approach could be replicated in other areas.
5 | CONCLUSION
Community health center accessibility is an understudied facet of health care access. We found that CHC accessibility substantially improved after ARRA and ACA funding that established many new CHCs. By using the census tract as the unit of analysis, our findings call attention to smaller pockets within larger areas HRSA defines as medically underserved that may be targeted for future CHC planning. Our results also urge the reevaluation of criteria, data, and methods currently used to designate geographic areas as eligible for CHC services to improve access to affordable care for those most in need.
Joint Acknowledgment/Disclosure Statement: The authors would like to acknowledge and thank Deborah Gurewich, Thomas Stopka, Jenna Sirkin, and Lauren Olsho for reviewing a draft of the manuscript, and Anamika Sinha for her data collection support. A revision of this manuscript was partially supported by Abt Associates internal funds. The revision was improved by presentation at and feedback from Abt's Work in Progress Seminar. Work by Dr. Fabian was supported by the National Institutes on Minority Health and Health Disparities, National Institutes of Health [grant number P50 MD010428]; and U.S. Environmental Protection Agency [grant number 83615601], It has not been formally reviewed by EPA. The views expressed in this document are solely those of the authors and do not necessarily reflect those of the Agency. EPA does not endorse any products or commercial services mentioned in this publication.
Leigh Evans https://orcid.org/0000-0001-5922-4381
(1.) Health Resources and Services Administration. Health center program factsheet. August 2017. https://www.bphc.hrsa.gov/about/health-centerfactsheet.pdf. Accessed May 22, 2018.
(2.) National Association for Community Health Centers. Community health center chartbook 2017. http://www.nachc.org/wp-con-tent/uploads/2017/06/Chartbook2017.pdf. Published June 2017. Accessed May 22, 2018.
(3.) Health Resources and Services Administration. What is a health cen-ter?https://bphc.hrsa.gov/about/what-is-a-health-center/index.html. Published 2018. Accessed May 22, 2018.
(4.) Adashi EY, Geiger HJ, Fine MD. Health care reform and primary care-the growing importance of the community health center. N Engl J Med. 2010;362(22):2047-2050.
(5.) Geiger HJ. Community-oriented primary care: a path to community development. Am J Public Health. 2002;92(11):1713-1716.
(6.) Epstein AJ. The role of public clinics in preventable hospitalizations among vulnerable populations. Health Serv Res. 2001;36(2):405-420.
(7.) Evans CS, Smith S, Kobayashi L, Chang DC. The effect of community health center (CHC) density on preventable hospital admissions in Medicaid and uninsured patients. J Health Care Poor Underserved. 2015;26(3):839-851.
(8.) Probst JC, Laditka JN, Laditka SB. Association between community health center and rural health clinic presence and county-level hospitalization rates for ambulatory care sensitive conditions: an analysis across eight US states. BMC Health Serv Res. 2009;9(1):134.
(9.) Bailey MJ, Goodman-Bacon A. The War on Poverty's experiment in public medicine: community health centers and the mortality of older Americans. Am Econ Rev. 2015;105(3):1067-1104.
(10.) Health Resources and Services Administration. Health Center Program New Access Points (HRSA 17-009) Technical Assistance. https://bphc.hrsa.gov/programopportunities/fundingopportunities/NAP/index.html. Published 2017. Accessed September 23, 2018.
(11.) Rosenbaum S. Will health centers go over the "Funding Cliff"? Milbank Q. 2015;93(1):32-35.
(12.) Department of Health and Human Services. Recovery Act (ARRA): Health Centers: New Access Point Grant Awards, https://www.dem-ocraticwhip.gov/sites/democraticwhip.house.gov/files/documents/communityhealthcentersreceivinggrants030209.pdf. Published n.d. Accessed September 23, 2018.
(13.) Department of Health and Human Services, Health Resources and Services Administration. Justification of Estimates for Appropriations Committees FY 2012. https://www.hrsa.gov/sites/default/files/about/budget/budgetjustification2012.pdf. Published 2012. Accessed September 23, 2018.
(14.) Health Resources and Services Administration. Medically underserved areas and populations (MUA/Ps). https://bhw.hrsa.gov/shortage-desig-nation/muap. Published October 2016. Accessed December 28, 2017.
(15.) Salinsky E. Health Care Shortage Designations: HPSA, MUA, and TBD. National Health Policy Forum; 2010. http://www.nhpf.org/library/background-papers/BP75_HPSA-MUA_6-04-2010.pdf. Accessed July 31, 2018.
(16.) U.S. Census Bureau. Data from American Community Survey 5-year estimates 2012-2016: Table B01003 total population by county. 2016. https://factfinder.census.gov/. Accessed September 21, 2018.
(17.) U.S. Census Bureau. Understanding and Using American Community Survey Data: What All Data Users Need to Know. Washington, DC: U.S. Government Printing Office; 2018. https://www.census.gov/con-tent/dam/Census/library/publications/2018/acs/acs_general_hand-book_2018.pdf. Accessed September 23, 2018.
(18.) Wright RA, Andres TL, Davidson AJ. Finding the medically under-served: a need to revise the federal definition. J Health Care Poor Underserved. 1996;7(4):296-307.
(19.) Jones MN, Brown CM, Widener MJ, Sucharew HJ, Beck AF. Area-level socioeconomic factors are associated with noncompletion of pediatric preventive services. J Prim Care Community Health. 2016;7(3):143-148.
(20.) Lin Y, Wimberly MC. Geographic variations of colorectal and breast cancer late-stage diagnosis and the effects of neighborhood-level factors. J Rural Health. 2017;33(2):146-157.
(21.) Prentice JC. Neighborhood effects on primary care access in Los Angeles. Soc Sci Med. 2006;62(5):1291-1303.
(22.) Wang F, Luo W. Assessing spatial and nonspatial factors for healthcare access: towards an integrated approach to defining health professional shortage areas. Health Place. 2005;11(2):131-146.
(23.) Davidson PL, Andersen RM, Wyn R, Brown ER. A framework for evaluating safety-net and other community-level factors on access for low-income populations. Ina J Health Care Organ Provis Financ. 2004;41(1):21-38.
(24.) Wang F. Quantitative Methods and Socioeconomic Applications in GIS, 2nd edn. Boca Raton, FL: CRC Press; 2015.
(25.) Fortney J, Rost K, Zhang M, Warren J. The impact of geographic accessibility on the intensity and quality of depression treatment. Med Care. 1999;37(9):884-893.
(26.) Donohoe J, Marshall V, Tan X, Camacho FT, Anderson R, Balkrishnan R. Predicting late-stage breast cancer diagnosis and receipt of adjuvant therapy: applying current spatial access to care methods in Appalachia. Med Care. 2015;53(11):980-988.
(27.) Comber AJ, Brunsdon C, Radburn R. A spatial analysis of variations in health access: linking geography, socio-economic status and access perceptions. Int J Health Geogr. 2011;10(1):44.
(28.) Hadley J, Cunningham P. Availability of safety net providers and access to care of uninsured persons. Health Serv Res. 2004;39(5):1527-1546.
(29.) Wehby GL, Shane DM, Joshi A, et al. The effects of distance to dentists and dentist supply on children's use of dental care. Health Serv Res. 2017;52(5):1817-1834.
(30.) Fishman J, McLafferty S, Galanter W. Does spatial access to primary care affect emergency department utilization for nonemergent conditions? Health Serv Res. 2018;53(1):489-508.
(31.) Luo W, Wang F. Measures of spatial accessibility to health care in a GIS environment: synthesis and a case study in the Chicago region. Environ Plan B Plan Des. 2003;30(6):865-884.
(32.) Bissonnette L, Wilson K, Bell S, Shah TI. Neighbourhoods and potential access to health care: the role of spatial and aspatial factors. Health Place. 2012;18(4):841-853.
(33.) U.S. Census Bureau. Data from American Community Survey 5-year estimates 2012-2016: Table S0501 by state. 2016. https://factfinder.census.gov/. Accessed September 23, 2018.
(34.) United Health Foundation. America's Health Rankings Annual Report. 2016. https://assets.americashealthrankings.org/app/uploads/ahrl6-complete-v2.pdf. Accessed October 9, 2017.
(35.) Gennuso KP, Jovaag A, Catlin BB, Rodock M, Park H. Assessment of factors contributing to health outcomes in the eight states of the Mississippi Delta Region. Prev Chronic Dis. 2016;13:E33.
(36.) Health Resources and Services Administration. Data from HRSA Data Warehouse: Medically Underserved Areas. 2017. https://datawarehouse.hrsa.gov/tools/analyzers/muafind.aspx. Accessed November 2, 2017.
(37.) Eberth JM, Eschbach K, Morris JS, Nguyen HT, Hossain MM, Elting LS. Geographic disparities in mammography capacity in the South: a longitudinal assessment of supply and demand. Health Serv Res. 2014;49(1):171-185.
(38.) McGrail MR, Humphreys JS. The index of rural access: an innovative integrated approach for measuring primary care access. BMC Health Serv Res. 2009;9:124.
(39.) Luo W, Wang F, Douglass C. Temporal changes of access to primary health care in Illinois (1990-2000) and policy implications. J Med Syst. 2004;28(3):287-299.
(40.) Daly MR, Mellor JM, Millones M. Do avoidable hospitalization rates among older adults differ by geographic access to primary care physicians? Health Serv Res. 2018;53(Suppl 1):3245-3264.
(41.) Luo W, Qi Y. An enhanced two-step floating catchment area (E2SFCA) method for measuring spatial accessibility to primary care physicians. Health Place. 2009;15(4):1100-1107.
(42.) Hall SA, Kaufman JS, Ricketts TC. Defining urban and rural areas in US epidemiologic studies. J Urban Health. 2006;83(2):162-175.
(43.) Morrill R, Cromartie J, Hart G. Metropolitan, urban, and rural commuting areas: toward a better depiction of the United States settlement system. Urban Geogr. 1999;20(8):727-748.
(44.) U.S. Census Bureau. Data from American Community Survey 5-year estimates 2012-2016: Table S0801 by state. 2016. https://factfinder.census.gov/. Accessed December 30, 2018.
(45.) Continelli T, McGinnis S, Holmes T. The effect of local primary care physician supply on the utilization of preventive health services in the United States. Health Place. 2010;16(5):942-951.
(46.) Ward MD, Gleditsch KS. Spatial Regression Models, Vol. 155. Thousand Oaks, CA: Sage; 2008.
(47.) Goodman DC, Mick SS, Bott D, et al. Primary care service areas: a new tool for the evaluation of primary care services. Health Serv Res. 2003;38(1 Pt 1):287-309.
(48.) Ku LC, Richard P, Dor A, Tan E, Shin P, Rosenbaum SJ. Strengthening Primary Care to Bend the Cost Curve: The Expansion of Community Health Centers Through Health Reform (Geiger Gibson/RCHN Community Health Foundation Research Collaborative). Washington, DC: George Washington University, School of Public Health and Health Services, Department of Health Policy; 2010.
(49.) Cole MB, Galarraga O, Wilson IB, Wright B, Trivedi AN. At federally funded health centers, Medicaid expansion was associated with improved quality of care. Health Aff (Millwood). 2017;36(1):40-48.
(50.) Moy E, Garcia M, Bastian B, et al. Leading causes of death in non-metropolitan and metropolitan areas--United States, 1999-2014. MMWR Surveill Summ. 2017;66(No.SS-1):1-8.
(51.) Ward TJ. Out in the Rural: A Mississippi Health Center and Its War on Poverty. New York, NY: Oxford University Press; 2017.
(52.) Dreher A. Poll: majority in state support Medicaid expansion, but the legislature doesn't. Jackson Free Press. April 2016. http://www.jacksonfreepress.com/news/2016/apr/ll/poll-ma-jority-state-support-medicaid-expansion-leg/.Accessed December 27, 2017.
(53.) American Academy of Family Physicians. UDS Mapper, https://www.udsmapper.org/index.cfm. Published 2017. Accessed Decern ber 14, 2017.
Additional supporting information may be found online in the Supporting Information section at the end of the article.
Leigh Evans PhD (1,[dagger]) | Martin P. Charns DBA (2,3) | Howard J. Cabral PhD (4) | M. Patricia Fabian ScD (5)
(1) Division of Health and Environment, Abt Associates, Inc., Cambridge, Massachusetts
(2)Department of Health Law, Policy, and Management, Boston University School of Public Health, Boston, Massachusetts
(3)Center for Healthcare Organization and Implementation Research (CHOIR), VA Boston Healthcare System, Boston, Massachusetts
(4) Department of Biostatistics, Boston University School of Public Health, Boston, Massachusetts
(5)department of Environmental Health, Boston University School of Public Health, Boston, Massachusetts
Leigh Evans, PhD, Division of Health and Environment, Abt Associates, Inc., Cambridge, MA.
Email: Leigh Evans@abtassoc.com
([dagger]) Author was with the Department of Health Law, Policy, and Management, Boston University School of Public Health, at the time the research was conducted.
TABLE 1 Census tract characteristics National Three states overall Sample size (b) (n) 72 291 2468 Proportion of low-income 44.58(15.03) 47.32 (13.78) nonelderly adults below poverty, % Mean (SD) Proportion of low-income 33.94(15.56) 38.93(11.80) nonelderly adults who are uninsured, % Mean (SD) Proportion of low-income 29.85 (21.09) 28.97 (18.73) nonelderly adults who are unemployed, % Mean(SD) Proportion of low-income 17.32 (23.60) 5.65(10.57) nonelderly adults who are Hispanic, % Mean (SD) Proportion of low-income 15.34(23.27) 37.51 (31.25) nonelderly adults who are black or African American, % Mean (SD) Proportion of low-income 25.25 (17.73) 30.74(15.41) nonelderly adults who did not graduate high school, % Mean (SD) Primary care n/a 9.63 (3.89) provider-to-population ratio (providers per 10 000 people) (c) Baseline CHC n/a 0.52 (0.95) accessibility (2008) Mean (SD) Arkansas Louisiana Sample size (b) (n) 684 1127 Proportion of low-income 44.48 (12.47) 48.40 (14.33) nonelderly adults below poverty, % Mean (SD) Proportion of low-income 38.51 (11.00) 39.64(12.69) nonelderly adults who are uninsured, % Mean (SD) Proportion of low-income 27.85 (18.79) 27.72 (18.98) nonelderly adults who are unemployed, % Mean(SD) Proportion of low-income 7.77 (12.89) 5.45 (10.13) nonelderly adults who are Hispanic, % Mean (SD) Proportion of low-income 22.22 (27.80) 40.61(30.60) nonelderly adults who are black or African American, % Mean (SD) Proportion of low-income 28.13(14.36) 31.21 (16.35) nonelderly adults who did not graduate high school, % Mean (SD) Primary care 9.63 (4.19) 9.78 (3.68) provider-to-population ratio (providers per 10 000 people) (c) Baseline CHC 0.57(1.32) 0.35 (0.65) accessibility (2008) Mean (SD) Mississippi P value (a) Sample size (b) (n) 657 n/a Proportion of low-income 48.44 (13.69) <0.0001 nonelderly adults below poverty, % Mean (SD) Proportion of low-income 38.15(10.93) 0.02 nonelderly adults who are uninsured, % Mean (SD) Proportion of low-income 32.27(17.86) <0.0001 nonelderly adults who are unemployed, % Mean(SD) Proportion of low-income 3.78 (7.92) <0.0001 nonelderly adults who are Hispanic, % Mean (SD) Proportion of low-income 48.11 (29.77) <0.0001 nonelderly adults who are black or African American, % Mean (SD) Proportion of low-income 32.66 (14.46) <0.0001 nonelderly adults who did not graduate high school, % Mean (SD) Primary care 9.34 (3.91) 0.06 provider-to-population ratio (providers per 10 000 people) (c) Baseline CHC 0.75 (0.87) <0.0001 accessibility (2008) Mean (SD) (a) P values represent ANOVA tests for significant differences in each predictor across the three states of interest: Arkansas, Louisiana, and Mississippi. We used P < 0.05 as a threshold for statistically significant differences between states. (b) Sample includes only census tracts where the total number of persons was greater than 0 in American Community Survey 2010-2014 5-y estimates. (c) Ratio defined as the number of primary care physicians, physician assistants, and nurse practitioners per 10 000 people within a defined travel distance from the census tract's location.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||RESEARCH ARTICLE|
|Author:||Evans, Leigh; Charns, Martin P.; Cabral, Howard J.; Fabian, M. Patricia|
|Publication:||Health Services Research|
|Date:||Aug 1, 2019|
|Previous Article:||The impact of tort reform on defensive medicine, quality of care, and physician supply: A systematic review.|
|Next Article:||Ambulance diversions following public hospital emergency department closures.|