- Email: [email protected]
The availability of community health center services and access to medical care
Healthcare (xxxx) xxxx–xxxx
Contents lists available at ScienceDirect
Healthcare journal homepage: www.elsevier.com/locate/hjdsi
The availability of community health center services and access to medical care☆ ⁎
James B. Kirbya, , Ravi Sharmab a
AHRQ, Center for Financing, Access and Cost Trends, 5600 Fishers Ln Rockville, MD 20852, United States Office of Quality Improvement Bureau of Primary Care Health Resources and Services Administration, 5600 Fishers Lane Rockville, MD 20852, United States
b
A R T I C L E I N F O
A BS T RAC T
Keywords: Access Community health centers Medicaid Uninsured
Background: Community Health Centers (CHCs) funded by Section 330 of the Public Health Service Act are an essential part of the health care safety net in the US. The Patient Protection and Affordable Care Act expanded the program significantly, but the extent to which the availability of CHCs improve access to care in general is not clear. In this paper, we examine the associations between the availability of CHC services in communities and two key measures of ambulatory care access – having a usual source of care and having any office-based medical visits over a one year period. Methods: We pooled six years of data from the Medical Expenditure Panel Survey (2008–2013) and linked it to geographic data on CHCs from Health Resources and Services Administration's Health Center Program Uniform Data System. We also link other community characteristics from the Area Health Resource File and the Dartmouth Institute's data files. The associations between CHC availability and our access measures are estimated with logistic regression models stratified by insurance status. Results: The availability of CHC services was positively associated with both measures of access among those with no insurance coverage. Additionally, it was positively associated with having a usual source of care among those with Medicaid and private insurance. These findings persist after controlling for key individual- and community-level characteristics. Conclusions: Our findings suggest that an enhanced CHC program could be an important resource for supporting the efficacy of expanded Medicaid coverage under the Affordable Care Act and, ultimately, improving access to quality primary care for underserved Americans.
1. Introduction Community Health Centers (CHCs) funded by Section 330 of the Public Health Service Act are an essential part of the health care safety net in the US, currently providing primary care to more than 21 million individuals.1 With the expansion of Medicaid under the Patient Protection and Affordable Care Act, the CHC program may become even more critical. The Affordable Care Act's Medicaid expansion, currently implemented by 27 States and the District of Columbia, will result in millions of individuals gaining coverage.2,3 Health insurance coverage, however, does not guarantee access to medical care; the success of the Affordable Care Act's Medicaid expansion depends, in part, on whether the capacity of local medical service markets is adequate to serve the influx of new enrollees. Previous research
suggests that many providers may choose not to participate in the Medicaid program 4,5 and that those who do may curtail the provision of low cost or charity care to the uninsured.6 Moreover, even after the Affordable Care Act is fully implemented, the Congressional Budget Office estimates that between 20 and 30 million people will still be uninsured.3 The CHC program is therefore essential both for the success of the Affordable Care Act's Medicaid expansion and for providing access to the people who remain uninsured or underinsured. Previous research documents the positive impacts that safety net providers have on the communities they serve. One study, for example, found that uninsured people living in close proximity to a safety net provider are modestly less likely to report unmet need and less likely to have an emergency department visit.7 Conversely, another study found that reductions in safety net capacity were associated with increases in
☆ The views in this article are those of the authors and no official endorsement by the Agency for Healthcare Research and Quality, the Health Resources and Services Administration or the Department of Health and Human Services is intended or should be inferred. ⁎ Corresponding author. E-mail addresses: [email protected] (J.B. Kirby), [email protected] (R. Sharma).
http://dx.doi.org/10.1016/j.hjdsi.2016.12.006 Received 11 November 2015; Received in revised form 21 June 2016; Accepted 22 December 2016 2213-0764/ Published by Elsevier Inc.
Please cite this article as: Kirby, J.B., Healthcare (2017), http://dx.doi.org/10.1016/j.hjdsi.2016.12.006
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
emergency department visits.8 In this paper, we update and extend previous research by describing the associations between the availability of CHC services in geographic areas and two key indicators of ambulatory care access – having a usual source of care and having at least one office-based medical visits over a one year period. Unlike most previous research, we do not focus entirely on the safety-net population but, instead, examine a nationally representative one. This approach allows for the possibility that the availability of CHC services may benefit entire communities, including those with private insurance, not just those at which CHC services are primarily aimed. While a disproportionate share of people served at CHCs are on Medicaid or uninsured, CHCs will treat anyone and bill on a sliding scale according to one's ability to pay. Those with private insurance could choose to go to a CHC because it is particularly convenient, because there are no other sources of primary care nearby (as is the case in some rural areas), or because CHC care is more affordable for those with high cost-sharing plans. Alternatively, privately insured individuals could benefit from the availability of CHC services indirectly because CHCs provide care to vulnerable populations that would otherwise strain the primary care market in a community. The possibility of such a positive “spillover” effect has not been examined. We further extend previous research by using an alternative measure of the availability of CHC services. Most previous research measures the availability of services based on the presence or number of CHCs in a community, though some studies have used distance to a CHC and funding levels.7,9,10 However, CHCs vary widely in their service capacity and the geographic dispersion of patient care sites. In this study, we investigate an alternative measure of CHC service availability— the “low-income penetration rate”, or the number of people in an area who used CHC services at least once divided by the number of residents with incomes below 200% of the federal poverty line. We suggest that this measure is one way to gauge the overall capacity of CHCs relative to the size of the populations they typically serve. Results from this research will be of interest to policy makers deciding on how the substantial gains in insurance coverage under the Affordable Care Act can best be translated into real improvements in access to care and health outcomes. Two possible approaches to increasing the availability of services to the newly insured are: 1) increase the number of medical providers willing to accept Medicaid payments by incentivizing program participation and 2) increasing funding directly to safety net providers such as CHCs so that they can expand their service capacity. Results from this paper should shed light on the latter of these approaches by estimating the association between the availability of CHC services in an area and individuals’ access to care.
Health Policy and Clinical Practice's data files. A Primary Care Service Area is the smallest geographic unit that can be considered a discrete service area for primary care. Each Primary Care Service Area consists of a zip code tabulation area with at least one primary care provider, and all contiguous zip code tabulation areas in which the population therein obtains a plurality of their care from the same providers.13 Currently, Primary Care Service Areas are constructed based on Medicare claims data. 2.2. Variables The main outcome variables in this study are dichotomous variables indicating whether individuals had a usual source of care and whether they had at least one office-based visit to a medical provider during the year. Individuals were asked, “Is there a particular person or place to which you go when you are sick or have a question about your health?” Those who answer in the affirmative to this question are coded as having a usual source of care. Individuals were coded as having at least one office-based provider visit if they reported having a visit to any type of medical provider in an office setting (i.e. non-Hospital) during the year. These measures are widely used as benchmarks for access to ambulatory care services.14 Our main independent variable is the number of unique patients reported by CHCs that reside in a particular Primary Care Service Area divided by the total number of residents with incomes below 200% of the federal poverty line (hereafter referred to as the CHC “penetration rate”). We consider this a proxy for the availability of CHC services relative to the population size of an area. A variety of individual characteristics may be related to both CHC penetration and our outcome measures and therefore should be included in the analysis. One important characteristic is insurance coverage. In this study, individuals are classified into one of the following seven insurance categories: covered by a private plan all year, by Medicaid all year, by Medicare all year, by both Medicaid and Medicare all year, by both Medicare and supplemental private insurance all year, uninsured all year, or some other combination of insurance coverage. The “other insurance” category consists mostly of individuals who changed insurance status across these categories during the year. Income is measured with dichotomous variables capturing household income relative to the federal poverty line. We also control for age, sex, race, and ethnicity, all potentially associated with living in underserved areas and with our access to care measures. Health status is the main driver of medical need and could be related to both access and residing in an area with high CHC penetration. Unhealthy people are more likely to have at least one office-based visit during a year and frequent contact with the health care system may make it more likely that they have a usual source of care. Further, CHCs are more likely to be placed in geographic areas that are underserved and, consequently, have poor access to needed care, which in turn may lead to worse health on aggregate. In our analysis, we control for subjective health (excellent, very good, good, fair or poor), the presence of the most expensive chronic conditions (angina, asthma, congestive heart failure, diabetes, chronic obstructive pulmonary disease, hypertension, myocardial infarction, other heart disease, and stroke), and whether a person has an activity or functional limitation (ADL or IADL). Finally, we control for a variety of Primary Care Service Area characteristics that are related to CHC placement. These include the percent of residents in poverty, the percent of a population that is black and the percent Hispanic, all of which are important factors to guide the placement of CHCs. We also accounted for whether individuals lived in a Medically Underserved Area or a Health Professional Shortage Area. By program design, a CHC must be located in a Medically Underserved Area and be designated as a Health Professional Shortage Area. Together, these variables were included to examine whether penetration had an impact on our outcome variables independent of the site selection process of CHCs.
2. Methods 2.1. Data Sources Our analysis uses data from the Medical Expenditure Panel Survey, a nationally representative household survey collected by the Agency for Healthcare Research and Quality since 1996. The Medical Expenditure Panel Survey collects information on health, health care use and expenditures, experiences with the US health care system and basic sociodemographic characteristics and is representative of the US non-institutionalized population.11,12 In this study, we created a large cross-section by pooling six years of data, 2008–2013. Our findings therefore pertain to the average associations between CHC penetration and our access measures over the study period. We link these data to characteristics of the Primary Care Service Areas in which individuals live using data from the Health Resources and Services Administration's Uniform Data System (2008–2013), the Bureau of Primary Health Care's Management Information System (2008–2013), the latest Area Health Resource File, and the Dartmouth Institute for 2
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Descriptive statistics for all variables in our analysis are shown in Table 1.
Table 1 Weighted means, weighted proportions and unweighted frequencies for all variables. Variables
Weighted means & proportions
Unweighted frequency
0.78 0.73
146,498 131,398
18.46%
–
Full-Year Insurance Categories Private Only Medicaid Only Medicare Only Medicare and Medicaid Medicare and Private Uninsured Other Insurance Combination Proportion Male Mean Age
0.49 0.10 0.06 0.02 0.06 0.13 0.14 0.49 37.46
77,455 33,578 10,812 3971 7777 33,754 29,065 93,615 –
Race and Ethnicity Non-Hispanic, White Non-Hispanic, Black Hispanic of Any Race Non-Hispanic, Asian Non-Hispanic, Other Race
0.64 0.12 0.16 0.05 0.02
79,851 39,661 57,898 14,262 4740
Self-Reported Health Excellent Very Good Good Fair Poor
0.31 0.33 0.25 0.08 0.03
59,502 61,201 52,494 18,107 5108
Household Income Relative to Poverty < =100% 100%−125% 125%−200% 200%−400% > 400%
0.15 0.05 0.14 0.30 0.36
44,265 13,014 33,502 56,126 49,505
0.02 0.10 0.04 0.07 0.02
3524 19,082 7481 14,253 2796
0.25 0.03 0.08 0.03
45,929 4958 13,177 5088
0.03 0.05
5075 8587
30.18% 12.20% 15.59% 0.35
– – – 77,827
0.17
46,113
Outcome Variables Has a Usual Source of Care Has One or More Office Visits CHC Penetration (Primary Care Service Area level)
The Presence of Select Chronic Conditions Angina Asthma Coronary Heart Disease Diabetes Chronic Obstructive Pulmonary Disease Hypertension Miocardial Infarction Other Heart Disease Stroke Activity and Functional Limitations At Least One ADL At Least One IADL Primary Care Service Area Characteristics Poverty Rate Percent Black Percent Minority In a Medically Underserved Area In a Health Professional Shortage Area N
2.3. Analytic approach We estimate logistic regression models on our two measures of access, controlling for the individual and geographic variables described above. Because CHCs tend to serve populations that are predominantly uninsured or who are on Medicaid, and because we found differences in many of the coefficients of the control variables across insurance groups, we stratify our models by insurance status. We report odds ratios for CHC penetration obtained from estimating separate regressions. To aid in interpretation, we then graph mean predicted probabilities by insurance status over the entire range of values for CHC penetration. Mean predicted probabilities are calculated by fixing CHC penetration to particular values, leaving all other variables at their sample values, and calculating the mean of the resultant predictions. As recommended by the Agency for Healthcare Research and Quality,15 all point estimates are calculated using survey weights and their standard errors are adjusted for the complex sample design of the Medical Expenditure Panel Survey using Taylor series linearization procedures available in Stata 14.16 3. Results Results suggest that among those on Medicaid, there is a positive association between the CHC penetration rate and having a usual source of care (Table 2). Specifically, among individuals on Medicaid, an increase of one percentage point in the CHC penetration rate is associated with an increase in the odds of having a usual source of care by 0.8%. We also find an association between the CHC penetration rate and the likelihood of having a usual source of care among the uninsured. For those without health insurance, an increase of one percentage point in the CHC penetration rate is associated with an increase in the odds of having a usual source of care of 0.7%. (Table 3). Given the target population of the CHC program, a positive association between CHC penetration rate and having a usual source of care among the uninsured and Medicaid populations is not too surprising. However, we also find a positive association between CHC penetration and having a usual source of care among those with private insurance. A one percentage point increase in the CHC penetration rate is associated with an increase in the odds of having a usual source of care of 0.3% among those with private insurance. These results provide tentative evidence of a positive “spillover” effect— the availability of CHC services in an area may benefit individuals who are not the primary target of the CHC program. The association between CHC penetration rate and having one or more office-based visits is not as consistent across insurance status as the association between CHC penetration and having a usual source of care. We find that a one percentage point increase in the CHC penetration rate is associated with a 0.5% increase in the odds of having at least one office visit but only among those with no insurance. To obtain a better sense of the magnitude of the associations estimated in our models, we show the marginal impact of a change in CHC penetration on the expected percent with a usual source of care (Fig. 1) and with at least one office-based provider visit (Fig. 2). In these figures, the penetration rate is along the x-axis and ranges from zero to 100%, values that are all represented in our data (i.e. there are no out-of-sample predictions). The y axes show the mean predicted percentages of individuals with a usual source of care or an office-based visit obtained from the results of the logistic regression models. In each figure, there is one line for each of the four largest insurance groups (except “Other insurance combination”): private only throughout the
196,412
3
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 2 Odds Ratios and P-values from Logistic Regression Models on Having a Usual Source of Care Stratified by Full-Year Insurance Status. Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
77,455 1.003*
10,812 0.997
33,578 1.008***
3971 1.007
7777 0.997
33,754 1.007***
29,065 1.004
(0.029)
(0.274)
(0.000)
(0.075)
(0.345)
(0.000)
(0.117)
Men
1.628*** (0.000)
1.130 (0.130)
1.051 (0.242)
1.116 (0.411)
1.121 (0.334)
1.845*** (0.000)
1.427*** (0.000)
Age
0.991*** (0.000)
1.043*** (0.000)
0.981*** (0.000)
1.032*** (0.000)
1.051*** (0.000)
0.997* (0.048)
0.997* (0.020)
0.777*** (0.000)
0.730* (0.036)
0.699** (0.002)
0.917 (0.651)
0.717 (0.261)
0.848* (0.016)
0.778*** (0.000)
Hispanic of Any Race
0.899 (0.061)
0.749 (0.091)
0.855 (0.153)
0.755 (0.309)
0.835 (0.493)
0.709*** (0.000)
0.944 (0.282)
Non-Hispanic, Asian
0.623*** (0.000)
0.640* (0.037)
0.969 (0.839)
0.864 (0.540)
0.784 (0.377)
0.507*** (0.000)
0.564*** (0.000)
Non-Hispanic, Other Race
1.134 (0.269)
0.873 (0.664)
0.784 (0.272)
1.285 (0.696)
1.010 (0.986)
1.554* (0.029)
1.238 (0.132)
1.082* (0.012)
1.342 (0.062)
1.157* (0.024)
0.969 (0.926)
0.805 (0.192)
1.120* (0.033)
0.890* (0.025)
Good
1.118** (0.006)
1.416* (0.032)
1.094 (0.168)
0.804 (0.528)
1.015 (0.938)
1.104 (0.060)
0.879* (0.019)
Fair
1.136 (0.088)
1.287 (0.135)
1.190 (0.150)
0.744 (0.399)
1.143 (0.542)
1.209** (0.005)
0.884 (0.175)
Poor
0.865 (0.278)
0.822 (0.331)
1.037 (0.856)
0.590 (0.099)
0.942 (0.820)
1.218 (0.187)
0.700* (0.015)
0.580*** (0.000)
0.676** (0.007)
1.308 (0.133)
1.663 (0.144)
0.703 (0.089)
0.603*** (0.000)
0.651*** (0.000)
100%−125%
0.705*** (0.001)
0.646* (0.011)
1.332 (0.115)
1.662 (0.164)
0.583 (0.052)
0.619*** (0.000)
0.632*** (0.000)
125%−200%
0.692*** (0.000)
0.688** (0.004)
1.641** (0.004)
1.637 (0.181)
0.974 (0.899)
0.680*** (0.000)
0.751*** (0.000)
200%−400%
0.813*** (0.000)
0.924 (0.533)
1.522* (0.025)
1.038 (0.910)
0.927 (0.606)
0.797** (0.001)
0.780*** (0.000)
1.139 (0.517)
1.425 (0.169)
1.827 (0.054)
1.596 (0.257)
1.535 (0.252)
1.074 (0.724)
0.907 (0.605)
Asthma
1.576*** (0.000)
1.601** (0.008)
1.661*** (0.000)
1.988** (0.004)
1.182 (0.516)
1.233** (0.002)
1.201* (0.013)
Coronary Heart Disease
1.363 (0.057)
1.373 (0.059)
1.408 (0.141)
0.874 (0.630)
1.258 (0.300)
1.544** (0.008)
1.370* (0.044)
Diabetes
2.498*** (0.000)
1.887*** (0.000)
1.730** (0.006)
1.328 (0.144)
1.820** (0.002)
2.222*** (0.000)
2.714*** (0.000)
Emphasema
1.635 (0.051)
1.039 (0.826)
1.537 (0.146)
1.927 (0.110)
1.190 (0.600)
1.538* (0.031)
1.672* (0.015)
Hypertension
1.885*** (0.000)
1.868*** (0.000)
1.874*** (0.000)
1.536* (0.032)
1.629*** (0.000)
2.007*** (0.000)
2.249*** (0.000)
Miocardial Infarction
0.943
0.795
0.591
0.688
0.934
0.887 1.018 (continued on next page)
N CHC penetration rate (with number of residents with income below 200% of poverty as the denominator)
Race and Ethnicity (reference: Non-Hispanic White) Non-Hispanic, Black
Self-rated Health (reference: Excellent) Very Good
Household Income Relative to Federal Poverty (reference: > 400%) < =100%
Presence of Chronic Conditions (reference: none) Angina
4
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 2 (continued) Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
(0.743)
(0.181)
(0.146)
(0.210)
(0.773)
(0.519)
(0.926)
Other Heart Disease
**
1.287 (0.002)
**
1.520 (0.006)
0.963 (0.821)
1.078 (0.754)
1.198 (0.194)
**
1.324 (0.008)
1.450*** (0.000)
Stroke
1.478* (0.048)
1.400* (0.030)
1.421 (0.107)
1.896* (0.014)
1.108 (0.647)
1.464** (0.008)
1.637** (0.003)
0.881 (0.527)
0.617** (0.001)
1.067 (0.737)
0.568* (0.015)
0.701 (0.099)
0.971 (0.901)
1.000 (0.999)
1.327 (0.064)
1.032 (0.840)
1.044 (0.800)
1.044 (0.853)
0.545** (0.004)
1.544* (0.018)
1.434** (0.003)
0.994 (0.054)
0.995 (0.439)
1.000 (0.920)
0.982* (0.017)
0.985 (0.068)
1.003 (0.311)
1.004 (0.200)
Percent Black
0.996 (0.052)
1.004 (0.324)
1.000 (0.945)
1.003 (0.624)
0.999 (0.946)
0.995* (0.013)
0.996* (0.032)
Percent Hispanic
0.994*** (0.001)
1.002 (0.574)
1.000 (0.926)
1.003 (0.556)
1.002 (0.763)
0.991*** (0.000)
0.993*** (0.000)
In a Medically Underserved Area
0.917 (0.182)
0.964 (0.803)
1.085 (0.311)
0.989 (0.944)
1.367* (0.048)
1.022 (0.701)
0.974 (0.611)
In a Health Professional Shortage Area
0.979 (0.757)
0.931 (0.602)
0.909 (0.220)
1.003 (0.988)
1.072 (0.761)
0.981 (0.755)
0.992 (0.892)
Constant
4.109*** (0.000)
0.267*** (0.000)
4.500*** (0.000)
1.289 (0.566)
0.442* (0.015)
0.430*** (0.000)
1.951*** (0.000)
Activity Limitations (reference: none) Has at Least One ADL
Has at Least One IADL
Area Characteristics Poverty Rate
* **
Statistically Significant, p < 0.05 Statistically Significant, p < 0.01 Statistically Significant, p < 0.00
***
percentage points less likely to have had an office visit during the year when compared to their counterparts in Primary Care Services Areas with a CHC penetration rate of 100%. The odds ratios underlying the other lines in Fig. 2 are not statistically significant.
year, Medicaid only throughout the year, Medicare only throughout the year and Uninsured throughout the year. Among the uninsured, individuals living in a Primary Care Service Area with a penetration rate of zero are, on average, sixteen percentage points less likely to have a usual source of care than those living in counties with a penetration rate of 100% (41–57%). From the mean, a 21% point increase in the penetration rate, approximately one standard deviation, is associated with a three percentage point increase in the likelihood of having a usual source of care (44–47%). The association between CHC penetration rate and the likelihood of having a usual source of care among those with Medicaid is somewhat weaker than that among those with no insurance. Among Medicaid enrollees, individuals living in a Primary Care Service Area with a penetration rate of zero are, on average, eight percentage points less likely to have a usual source of care than those living in a Primary Care Service Area with a penetration rate of 100% (84% and 92% respectively). For the same group, a 21% point increase in the penetration rate from the mean is associated with a two percentage point increase in the likelihood of having a usual source of care (86–88%). Individuals with private insurance living in a Primary Care Service Area with a CHC penetration rate of zero are four percentage points less likely to have a usual source of care than their counterparts in Primary Care Service Areas with a penetration rate of 100%. As in Fig. 1, the line representing individuals with no insurance in Fig. 2 has an upward slope, suggesting a positive association between the CHC penetration rate and the likelihood of having an office-based visit during the year. Among the uninsured, those living in a Primary Care Service Area with a CHC penetration rate of zero are ten
4. Discussion We find that the availability of CHC services, as measured by the penetration rate, is positively associated with the likelihood of having a usual source of care not only among individuals on Medicaid and the uninsured, but also among those with private insurance. The penetration rate is positively associated with having at least one office based visit, but only for individuals who are uninsured. These associations persist after controlling for a variety of individual- and communitylevel variables. In supplemental analyses, we investigated other measures of CHC availability, including whether an area had any CHCs and the count of CHCs, and found a similar pattern of results, though standard errors were larger resulting in fewer significant results. We suspect that this is because the penetration rate captures the availability of CHC services with less measurement error than the other measures examined. Several limitations should be considered when interpreting these findings. First, results from past research suggest that findings can be sensitive to the geographic unit at which area characteristics are operationalized. While Primary Care Service Areas are designed to approximate primary care markets, they do so imperfectly. For example, they are constructed entirely with Medicare claims data and therefore do not necessarily reflect the entire primary care market. We 5
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 3 Odds ratios and p-values from logistic regression models on having one or more office-based provider visits stratified by full-year insurance status. Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
77,455 1.001
10,812 1.001
33,578 1.003
3971 0.998
7777 0.995
33,754 1.005***
29,065 1.002
(0.413)
(0.644)
(0.088)
(0.565)
(0.076)
(0.000)
(0.224)
Men
2.026*** (0.000)
1.453*** (0.000)
1.250*** (0.000)
1.308 (0.071)
1.572*** (0.000)
2.295*** (0.000)
1.992*** (0.000)
Age
0.994*** (0.000)
1.011*** (0.000)
0.982*** (0.000)
1.014*** (0.001)
1.019*** (0.000)
1.000 (0.889)
1.000 (0.818)
0.556*** (0.000)
0.524*** (0.000)
0.578*** (0.000)
0.858 (0.496)
0.359*** (0.000)
0.659*** (0.000)
0.569*** (0.000)
Hispanic of Any Race
0.671*** (0.000)
0.737 (0.053)
0.636*** (0.000)
0.703 (0.189)
0.396*** (0.000)
0.653*** (0.000)
0.719*** (0.000)
Non-Hispanic, Asian
0.526*** (0.000)
0.530** (0.002)
0.527*** (0.000)
0.866 (0.660)
0.465** (0.001)
0.496*** (0.000)
0.581*** (0.000)
Non-Hispanic, Other Race
0.878 (0.115)
0.449** (0.003)
0.720** (0.008)
1.797 (0.287)
0.815 (0.746)
0.918 (0.497)
0.674*** (0.001)
1.267*** (0.000)
1.305* (0.048)
1.099 (0.069)
1.113 (0.672)
1.159 (0.369)
1.156** (0.004)
1.175** (0.001)
Good
1.472*** (0.000)
1.457* (0.013)
1.321*** (0.000)
1.414 (0.186)
1.540* (0.014)
1.480*** (0.000)
1.298*** (0.000)
Fair
2.855*** (0.000)
1.517* (0.012)
2.903*** (0.000)
2.068* (0.013)
2.458*** (0.001)
2.116*** (0.000)
1.697*** (0.000)
Poor
3.881*** (0.000)
1.303 (0.203)
2.240*** (0.000)
1.741 (0.135)
2.141* (0.026)
3.131*** (0.000)
2.596*** (0.000)
0.608*** (0.000)
0.446*** (0.000)
0.724* (0.023)
0.899 (0.798)
0.542* (0.023)
0.588*** (0.000)
0.728*** (0.000)
100%−125%
0.568*** (0.000)
0.622* (0.017)
0.735* (0.042)
0.925 (0.871)
0.483* (0.017)
0.570*** (0.000)
0.746*** (0.001)
125%−200%
0.641*** (0.000)
0.574*** (0.000)
0.744* (0.049)
0.858 (0.725)
0.719 (0.084)
0.652*** (0.000)
0.753*** (0.000)
200%−400%
0.743*** (0.000)
0.693** (0.007)
0.817 (0.167)
0.635 (0.303)
0.850 (0.231)
0.720*** (0.000)
0.787*** (0.000)
1.292 (0.217)
1.590 (0.089)
1.083 (0.775)
0.861 (0.615)
0.828 (0.571)
0.977 (0.896)
0.888 (0.501)
Asthma
1.969*** (0.000)
2.089*** (0.000)
1.602*** (0.000)
1.988** (0.009)
1.762* (0.044)
1.388*** (0.000)
1.601*** (0.000)
Coronary Heart Disease
1.333 (0.054)
1.528* (0.012)
1.736* (0.028)
1.124 (0.637)
2.127* (0.020)
1.422* (0.030)
1.240 (0.169)
Diabetes
2.785*** (0.000)
1.602** (0.002)
3.007*** (0.000)
2.722*** (0.000)
1.534** (0.008)
2.221*** (0.000)
2.362*** (0.000)
Emphasema
1.267 (0.298)
1.276 (0.213)
1.875* (0.042)
0.824 (0.621)
1.233 (0.503)
1.599* (0.036)
1.105 (0.659)
Hypertension
1.918*** (0.000)
2.430*** (0.000)
2.463*** (0.000)
1.698** (0.003)
2.475*** (0.000)
2.185*** (0.000)
2.091*** (0.000)
Miocardial Infarction
0.843
0.965
1.130
0.942
0.789
0.894 1.572* (continued on next page)
N CHC penetration rate (with number of residents with income below 200% of poverty as the denominator)
Race and Ethnicity (reference: Non-Hispanic White) Non-Hispanic, Black
Self-rated Health (reference: Excellent) Very Good
Household Income Relative to Federal Poverty (reference: > 400%) < =100%
Presence of Chronic Conditions (reference: none) Angina
6
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 3 (continued) Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
(0.307)
(0.863)
(0.658)
(0.842)
(0.422)
(0.606)
(0.024)
***
Other Heart Disease
1.826 (0.000)
2.047 (0.000)
1.098 (0.567)
1.514 (0.122)
2.171 (0.001)
1.405 (0.000)
1.745*** (0.000)
Stroke
1.710** (0.003)
0.999 (0.995)
0.939 (0.809)
0.645 (0.083)
1.142 (0.577)
1.304 (0.056)
1.724** (0.002)
1.456 (0.174)
0.826 (0.330)
1.376 (0.075)
0.871 (0.565)
1.132 (0.702)
1.190 (0.594)
1.419 (0.074)
2.509*** (0.000)
1.095 (0.637)
1.809*** (0.000)
1.446 (0.065)
0.705 (0.102)
2.269*** (0.000)
1.604*** (0.000)
0.988*** (0.000)
0.994 (0.295)
0.998 (0.535)
1.004 (0.696)
0.972*** (0.000)
0.998 (0.316)
0.995 (0.072)
Percent Black
1.000 (0.818)
1.001 (0.840)
0.996* (0.020)
0.990 (0.053)
1.006 (0.343)
0.996** (0.002)
0.997 (0.148)
Percent Hispanic
0.997** (0.002)
1.001 (0.676)
1.000 (0.990)
0.999 (0.777)
1.004 (0.384)
0.996** (0.008)
0.998 (0.238)
In a Medically Underserved Area
0.970 (0.441)
0.977 (0.856)
1.021 (0.700)
0.840 (0.305)
1.810*** (0.000)
0.912* (0.039)
0.985 (0.756)
In a Health Professional Shortage Area
0.950 (0.275)
0.865 (0.267)
0.907 (0.126)
1.428 (0.060)
1.005 (0.974)
0.944 (0.252)
0.921 (0.222)
Constant
1.959*** (0.000)
1.644 (0.083)
3.781*** (0.000)
1.432 (0.450)
2.090 (0.050)
0.279*** (0.000)
0.976 (0.817)
Activity Limitations (reference: none) Has at Least One ADL
Has at Least One IADL
Area Characteristics Poverty Rate
* **
***
***
Statistically Significant, p < 0.05 Statistically Significant, p < 0.01 Statistically Significant, p < 0.00
***
Fig. 1. Mean predicted percent with a usual source of care by insurance status and CHC penetration rate.
7
***
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Fig. 2. Mean predicted percent with at least one office visit by insurance status and CHC penetration.
particular CHC program that result in an “exogenous” increase or decrease in CHC service availability, under some circumstances prepost differences in access can be used to estimate a causal effect. We consider this paper a first step towards more rigorous causal analysis of the relationship between CHC availability and access to health care.
tested the sensitivity of our results by estimating our models again with CHC penetration calculated at the county-level and find that our findings changed little. We continue to believe that Primary Care Service Areas are the most appropriate geographic unit at which to calculate CHC penetration, but acknowledge that measurement error may attenuate our results. Though a variety of individual and community characteristics are included in our analysis, there still may be unobserved characteristics that are related both to access and to the penetration rate in a particular area. Unobserved health differences in an area, for example, could drive both our measures of access (especially having an office visit) and the number of unique CHC patients in a community, the numerator in the penetration rate. Our results are not sensitive, however, to the exclusion of variables measuring health status. A similar limitation is that CHCs are not placed randomly throughout the country but, rather, placed where they are most needed. Medical need, in turn, could be related to our measures of access. Individuals with high levels of medical need are more likely to have medical use, including office-based visits, and could be more likely to have a usual source of care due to frequent contact with the health care system. To ameliorate this concern, we included dichotomous variables indicating whether a geographic area was in a Medically Underserved Area (MUA) or a Health Professional Shortage Area (HPSA). These designations are extremely influential in determining CHC locations. When these variables were excluded from the model, results changed very little, suggesting that our results are not entirely due to the selection of CHCs into needy areas. Despite these sensitivity checks, the problem of endogeneity of CHC penetration is a serious limitation in this study and results should not be interpreted as causal. Future research should investigate the effects of the availability of CHC services on access by using data and analytic methods that can control for the non-random selection of CHCs into primary care markets and of people into residential areas. One approach is to use longitudinal data to estimate a person-level fixed effects model, which would remove the influence of all characteristics that do not change within person. In effect, individuals act as their own controls. Future research might also consider a quasi-experimental approach to estimating the effect of the availability of CHC services on access. For example, if a state or municipality makes changes to a
5. Conclusion One of the primary goals of the Affordable Care Act is to ensure access to affordable medical care for all Americans and it aims to achieve this largely by increasing the number of people with insurance coverage, via Medicaid or private insurance. The success of this approach hinges, at least in part, on the extent to which the supply of primary care meets the increase in demand driven by health insurance expansions. Our findings are consistent with the idea that an expanded CHC program may help to ensure that Affordable Care Act's health insurance expansions translate into improvements in access to affordable medical care, especially in states that opt to expand their Medicaid programs and therefore have a large number of new enrollees seeking primary care services. In all states, however, CHC service availability may improve access to care for those who remain uninsured or underinsured, and perhaps even improve access to individuals who have private insurance. Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.hjdsi.2016.12.006. References 1 Health Resources and Services Administration. 2013 Health Center Data, 2013 National Data, Table 3A – Patients by Age and Gender. 2013; 〈http://bphc.hrsa.gov/ uds/datacenter.aspx?Q=tall & year=2013 & state=〉. 2 Haberlein M, Brooks T, Artiga S, Stephens J. Getting into Gear for 2014: shifting New Medicaid Eligibility and Enrollment Policies into Drive. Menlo Park, CA: The Henry J. Kaiser Family Foundation; 2013. 3 Congressional Budget Office. Insurance coverage provisions of the affordable care Act—CBO’s April 2014 baseline. 〈http://www.cbo.gov/sites/default/files/cbofiles/ attachments/43900-2014-04-ACAtables2.pdf〉; 2014. 4 Bisgaier J, Rhodes K. Auditing access to specialty care for children with public
8
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
2009;25(1):8–16. 11 Cohen J. The medical expenditure panel survey: a national health information resource. Inquiry. 1996;33(4):373–389. 12 Cohen J. Design and Methods of the Medical Expenditures Panel Survey Household Component. Rockville, MD: Agency for Health Care Policy and Research; 1997. MEPS Methodology Report No. 1, AHCPR Pub. No. 97-0026 13 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;38:287–309. 14 Agency for Healthcare Research and Quality. National Healthcare Disparities Report. Rockville, MD: Agency for Healthcare Research and Quality; 2013 15 Agency for Healthcare Research and Quality. Documentation for MEPS Full-Year 2013 Consolidated Data File. 2015; 〈https://meps.ahrq.gov/data_stats/download_ data/pufs/h163/h163doc.shtml〉 Accessed 6/17/2016, 2016; 2016 Accessed 6.17. 2016, . 16 StataCorp . Stata: release 14 Software. Survey Data Rerence Manual, College Station, TX: StataCorp, LP; 2015.
insurance. N Engl J Med. 2011;364:2324–2333. 5 Boukus E, Cassil A, O'Malley A. A snapshot of US physicians: Key Findings from the 2008 Health Tracking Physician Survey, Washington, DC: Center for Studying Health System Change; 2009. 6 Sabik LM, Gandhi SO. Impact of changes in medicaid coverage on physician provision of safety net care. Med Care. 2013;51(11):978–984. 7 Hadley J, Cunningham P. Availability of safety net providers and access to care of uninsured persons. Health Serv Res. 2004;39(5):1527–1546. 8 Cunningham P. What accounts for differences in the use of hospital emergency departments across US communities?. Health Aff (Millwood). 2006;25:W324–W336. 9 McMorrow S, Zuckerman S. Expanding federal funding to community health centers slows decline in access for low‐income adults. Health Serv Res. 2014;49(3):992–1010. 10 Rust G, Baltrus P, Ye J, et al. Presence of a community health center and uninsured emergency department visit rates in rural counties. J Rural Health.
9
Contents lists available at ScienceDirect
Healthcare journal homepage: www.elsevier.com/locate/hjdsi
The availability of community health center services and access to medical care☆ ⁎
James B. Kirbya, , Ravi Sharmab a
AHRQ, Center for Financing, Access and Cost Trends, 5600 Fishers Ln Rockville, MD 20852, United States Office of Quality Improvement Bureau of Primary Care Health Resources and Services Administration, 5600 Fishers Lane Rockville, MD 20852, United States
b
A R T I C L E I N F O
A BS T RAC T
Keywords: Access Community health centers Medicaid Uninsured
Background: Community Health Centers (CHCs) funded by Section 330 of the Public Health Service Act are an essential part of the health care safety net in the US. The Patient Protection and Affordable Care Act expanded the program significantly, but the extent to which the availability of CHCs improve access to care in general is not clear. In this paper, we examine the associations between the availability of CHC services in communities and two key measures of ambulatory care access – having a usual source of care and having any office-based medical visits over a one year period. Methods: We pooled six years of data from the Medical Expenditure Panel Survey (2008–2013) and linked it to geographic data on CHCs from Health Resources and Services Administration's Health Center Program Uniform Data System. We also link other community characteristics from the Area Health Resource File and the Dartmouth Institute's data files. The associations between CHC availability and our access measures are estimated with logistic regression models stratified by insurance status. Results: The availability of CHC services was positively associated with both measures of access among those with no insurance coverage. Additionally, it was positively associated with having a usual source of care among those with Medicaid and private insurance. These findings persist after controlling for key individual- and community-level characteristics. Conclusions: Our findings suggest that an enhanced CHC program could be an important resource for supporting the efficacy of expanded Medicaid coverage under the Affordable Care Act and, ultimately, improving access to quality primary care for underserved Americans.
1. Introduction Community Health Centers (CHCs) funded by Section 330 of the Public Health Service Act are an essential part of the health care safety net in the US, currently providing primary care to more than 21 million individuals.1 With the expansion of Medicaid under the Patient Protection and Affordable Care Act, the CHC program may become even more critical. The Affordable Care Act's Medicaid expansion, currently implemented by 27 States and the District of Columbia, will result in millions of individuals gaining coverage.2,3 Health insurance coverage, however, does not guarantee access to medical care; the success of the Affordable Care Act's Medicaid expansion depends, in part, on whether the capacity of local medical service markets is adequate to serve the influx of new enrollees. Previous research
suggests that many providers may choose not to participate in the Medicaid program 4,5 and that those who do may curtail the provision of low cost or charity care to the uninsured.6 Moreover, even after the Affordable Care Act is fully implemented, the Congressional Budget Office estimates that between 20 and 30 million people will still be uninsured.3 The CHC program is therefore essential both for the success of the Affordable Care Act's Medicaid expansion and for providing access to the people who remain uninsured or underinsured. Previous research documents the positive impacts that safety net providers have on the communities they serve. One study, for example, found that uninsured people living in close proximity to a safety net provider are modestly less likely to report unmet need and less likely to have an emergency department visit.7 Conversely, another study found that reductions in safety net capacity were associated with increases in
☆ The views in this article are those of the authors and no official endorsement by the Agency for Healthcare Research and Quality, the Health Resources and Services Administration or the Department of Health and Human Services is intended or should be inferred. ⁎ Corresponding author. E-mail addresses: [email protected] (J.B. Kirby), [email protected] (R. Sharma).
http://dx.doi.org/10.1016/j.hjdsi.2016.12.006 Received 11 November 2015; Received in revised form 21 June 2016; Accepted 22 December 2016 2213-0764/ Published by Elsevier Inc.
Please cite this article as: Kirby, J.B., Healthcare (2017), http://dx.doi.org/10.1016/j.hjdsi.2016.12.006
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
emergency department visits.8 In this paper, we update and extend previous research by describing the associations between the availability of CHC services in geographic areas and two key indicators of ambulatory care access – having a usual source of care and having at least one office-based medical visits over a one year period. Unlike most previous research, we do not focus entirely on the safety-net population but, instead, examine a nationally representative one. This approach allows for the possibility that the availability of CHC services may benefit entire communities, including those with private insurance, not just those at which CHC services are primarily aimed. While a disproportionate share of people served at CHCs are on Medicaid or uninsured, CHCs will treat anyone and bill on a sliding scale according to one's ability to pay. Those with private insurance could choose to go to a CHC because it is particularly convenient, because there are no other sources of primary care nearby (as is the case in some rural areas), or because CHC care is more affordable for those with high cost-sharing plans. Alternatively, privately insured individuals could benefit from the availability of CHC services indirectly because CHCs provide care to vulnerable populations that would otherwise strain the primary care market in a community. The possibility of such a positive “spillover” effect has not been examined. We further extend previous research by using an alternative measure of the availability of CHC services. Most previous research measures the availability of services based on the presence or number of CHCs in a community, though some studies have used distance to a CHC and funding levels.7,9,10 However, CHCs vary widely in their service capacity and the geographic dispersion of patient care sites. In this study, we investigate an alternative measure of CHC service availability— the “low-income penetration rate”, or the number of people in an area who used CHC services at least once divided by the number of residents with incomes below 200% of the federal poverty line. We suggest that this measure is one way to gauge the overall capacity of CHCs relative to the size of the populations they typically serve. Results from this research will be of interest to policy makers deciding on how the substantial gains in insurance coverage under the Affordable Care Act can best be translated into real improvements in access to care and health outcomes. Two possible approaches to increasing the availability of services to the newly insured are: 1) increase the number of medical providers willing to accept Medicaid payments by incentivizing program participation and 2) increasing funding directly to safety net providers such as CHCs so that they can expand their service capacity. Results from this paper should shed light on the latter of these approaches by estimating the association between the availability of CHC services in an area and individuals’ access to care.
Health Policy and Clinical Practice's data files. A Primary Care Service Area is the smallest geographic unit that can be considered a discrete service area for primary care. Each Primary Care Service Area consists of a zip code tabulation area with at least one primary care provider, and all contiguous zip code tabulation areas in which the population therein obtains a plurality of their care from the same providers.13 Currently, Primary Care Service Areas are constructed based on Medicare claims data. 2.2. Variables The main outcome variables in this study are dichotomous variables indicating whether individuals had a usual source of care and whether they had at least one office-based visit to a medical provider during the year. Individuals were asked, “Is there a particular person or place to which you go when you are sick or have a question about your health?” Those who answer in the affirmative to this question are coded as having a usual source of care. Individuals were coded as having at least one office-based provider visit if they reported having a visit to any type of medical provider in an office setting (i.e. non-Hospital) during the year. These measures are widely used as benchmarks for access to ambulatory care services.14 Our main independent variable is the number of unique patients reported by CHCs that reside in a particular Primary Care Service Area divided by the total number of residents with incomes below 200% of the federal poverty line (hereafter referred to as the CHC “penetration rate”). We consider this a proxy for the availability of CHC services relative to the population size of an area. A variety of individual characteristics may be related to both CHC penetration and our outcome measures and therefore should be included in the analysis. One important characteristic is insurance coverage. In this study, individuals are classified into one of the following seven insurance categories: covered by a private plan all year, by Medicaid all year, by Medicare all year, by both Medicaid and Medicare all year, by both Medicare and supplemental private insurance all year, uninsured all year, or some other combination of insurance coverage. The “other insurance” category consists mostly of individuals who changed insurance status across these categories during the year. Income is measured with dichotomous variables capturing household income relative to the federal poverty line. We also control for age, sex, race, and ethnicity, all potentially associated with living in underserved areas and with our access to care measures. Health status is the main driver of medical need and could be related to both access and residing in an area with high CHC penetration. Unhealthy people are more likely to have at least one office-based visit during a year and frequent contact with the health care system may make it more likely that they have a usual source of care. Further, CHCs are more likely to be placed in geographic areas that are underserved and, consequently, have poor access to needed care, which in turn may lead to worse health on aggregate. In our analysis, we control for subjective health (excellent, very good, good, fair or poor), the presence of the most expensive chronic conditions (angina, asthma, congestive heart failure, diabetes, chronic obstructive pulmonary disease, hypertension, myocardial infarction, other heart disease, and stroke), and whether a person has an activity or functional limitation (ADL or IADL). Finally, we control for a variety of Primary Care Service Area characteristics that are related to CHC placement. These include the percent of residents in poverty, the percent of a population that is black and the percent Hispanic, all of which are important factors to guide the placement of CHCs. We also accounted for whether individuals lived in a Medically Underserved Area or a Health Professional Shortage Area. By program design, a CHC must be located in a Medically Underserved Area and be designated as a Health Professional Shortage Area. Together, these variables were included to examine whether penetration had an impact on our outcome variables independent of the site selection process of CHCs.
2. Methods 2.1. Data Sources Our analysis uses data from the Medical Expenditure Panel Survey, a nationally representative household survey collected by the Agency for Healthcare Research and Quality since 1996. The Medical Expenditure Panel Survey collects information on health, health care use and expenditures, experiences with the US health care system and basic sociodemographic characteristics and is representative of the US non-institutionalized population.11,12 In this study, we created a large cross-section by pooling six years of data, 2008–2013. Our findings therefore pertain to the average associations between CHC penetration and our access measures over the study period. We link these data to characteristics of the Primary Care Service Areas in which individuals live using data from the Health Resources and Services Administration's Uniform Data System (2008–2013), the Bureau of Primary Health Care's Management Information System (2008–2013), the latest Area Health Resource File, and the Dartmouth Institute for 2
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Descriptive statistics for all variables in our analysis are shown in Table 1.
Table 1 Weighted means, weighted proportions and unweighted frequencies for all variables. Variables
Weighted means & proportions
Unweighted frequency
0.78 0.73
146,498 131,398
18.46%
–
Full-Year Insurance Categories Private Only Medicaid Only Medicare Only Medicare and Medicaid Medicare and Private Uninsured Other Insurance Combination Proportion Male Mean Age
0.49 0.10 0.06 0.02 0.06 0.13 0.14 0.49 37.46
77,455 33,578 10,812 3971 7777 33,754 29,065 93,615 –
Race and Ethnicity Non-Hispanic, White Non-Hispanic, Black Hispanic of Any Race Non-Hispanic, Asian Non-Hispanic, Other Race
0.64 0.12 0.16 0.05 0.02
79,851 39,661 57,898 14,262 4740
Self-Reported Health Excellent Very Good Good Fair Poor
0.31 0.33 0.25 0.08 0.03
59,502 61,201 52,494 18,107 5108
Household Income Relative to Poverty < =100% 100%−125% 125%−200% 200%−400% > 400%
0.15 0.05 0.14 0.30 0.36
44,265 13,014 33,502 56,126 49,505
0.02 0.10 0.04 0.07 0.02
3524 19,082 7481 14,253 2796
0.25 0.03 0.08 0.03
45,929 4958 13,177 5088
0.03 0.05
5075 8587
30.18% 12.20% 15.59% 0.35
– – – 77,827
0.17
46,113
Outcome Variables Has a Usual Source of Care Has One or More Office Visits CHC Penetration (Primary Care Service Area level)
The Presence of Select Chronic Conditions Angina Asthma Coronary Heart Disease Diabetes Chronic Obstructive Pulmonary Disease Hypertension Miocardial Infarction Other Heart Disease Stroke Activity and Functional Limitations At Least One ADL At Least One IADL Primary Care Service Area Characteristics Poverty Rate Percent Black Percent Minority In a Medically Underserved Area In a Health Professional Shortage Area N
2.3. Analytic approach We estimate logistic regression models on our two measures of access, controlling for the individual and geographic variables described above. Because CHCs tend to serve populations that are predominantly uninsured or who are on Medicaid, and because we found differences in many of the coefficients of the control variables across insurance groups, we stratify our models by insurance status. We report odds ratios for CHC penetration obtained from estimating separate regressions. To aid in interpretation, we then graph mean predicted probabilities by insurance status over the entire range of values for CHC penetration. Mean predicted probabilities are calculated by fixing CHC penetration to particular values, leaving all other variables at their sample values, and calculating the mean of the resultant predictions. As recommended by the Agency for Healthcare Research and Quality,15 all point estimates are calculated using survey weights and their standard errors are adjusted for the complex sample design of the Medical Expenditure Panel Survey using Taylor series linearization procedures available in Stata 14.16 3. Results Results suggest that among those on Medicaid, there is a positive association between the CHC penetration rate and having a usual source of care (Table 2). Specifically, among individuals on Medicaid, an increase of one percentage point in the CHC penetration rate is associated with an increase in the odds of having a usual source of care by 0.8%. We also find an association between the CHC penetration rate and the likelihood of having a usual source of care among the uninsured. For those without health insurance, an increase of one percentage point in the CHC penetration rate is associated with an increase in the odds of having a usual source of care of 0.7%. (Table 3). Given the target population of the CHC program, a positive association between CHC penetration rate and having a usual source of care among the uninsured and Medicaid populations is not too surprising. However, we also find a positive association between CHC penetration and having a usual source of care among those with private insurance. A one percentage point increase in the CHC penetration rate is associated with an increase in the odds of having a usual source of care of 0.3% among those with private insurance. These results provide tentative evidence of a positive “spillover” effect— the availability of CHC services in an area may benefit individuals who are not the primary target of the CHC program. The association between CHC penetration rate and having one or more office-based visits is not as consistent across insurance status as the association between CHC penetration and having a usual source of care. We find that a one percentage point increase in the CHC penetration rate is associated with a 0.5% increase in the odds of having at least one office visit but only among those with no insurance. To obtain a better sense of the magnitude of the associations estimated in our models, we show the marginal impact of a change in CHC penetration on the expected percent with a usual source of care (Fig. 1) and with at least one office-based provider visit (Fig. 2). In these figures, the penetration rate is along the x-axis and ranges from zero to 100%, values that are all represented in our data (i.e. there are no out-of-sample predictions). The y axes show the mean predicted percentages of individuals with a usual source of care or an office-based visit obtained from the results of the logistic regression models. In each figure, there is one line for each of the four largest insurance groups (except “Other insurance combination”): private only throughout the
196,412
3
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 2 Odds Ratios and P-values from Logistic Regression Models on Having a Usual Source of Care Stratified by Full-Year Insurance Status. Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
77,455 1.003*
10,812 0.997
33,578 1.008***
3971 1.007
7777 0.997
33,754 1.007***
29,065 1.004
(0.029)
(0.274)
(0.000)
(0.075)
(0.345)
(0.000)
(0.117)
Men
1.628*** (0.000)
1.130 (0.130)
1.051 (0.242)
1.116 (0.411)
1.121 (0.334)
1.845*** (0.000)
1.427*** (0.000)
Age
0.991*** (0.000)
1.043*** (0.000)
0.981*** (0.000)
1.032*** (0.000)
1.051*** (0.000)
0.997* (0.048)
0.997* (0.020)
0.777*** (0.000)
0.730* (0.036)
0.699** (0.002)
0.917 (0.651)
0.717 (0.261)
0.848* (0.016)
0.778*** (0.000)
Hispanic of Any Race
0.899 (0.061)
0.749 (0.091)
0.855 (0.153)
0.755 (0.309)
0.835 (0.493)
0.709*** (0.000)
0.944 (0.282)
Non-Hispanic, Asian
0.623*** (0.000)
0.640* (0.037)
0.969 (0.839)
0.864 (0.540)
0.784 (0.377)
0.507*** (0.000)
0.564*** (0.000)
Non-Hispanic, Other Race
1.134 (0.269)
0.873 (0.664)
0.784 (0.272)
1.285 (0.696)
1.010 (0.986)
1.554* (0.029)
1.238 (0.132)
1.082* (0.012)
1.342 (0.062)
1.157* (0.024)
0.969 (0.926)
0.805 (0.192)
1.120* (0.033)
0.890* (0.025)
Good
1.118** (0.006)
1.416* (0.032)
1.094 (0.168)
0.804 (0.528)
1.015 (0.938)
1.104 (0.060)
0.879* (0.019)
Fair
1.136 (0.088)
1.287 (0.135)
1.190 (0.150)
0.744 (0.399)
1.143 (0.542)
1.209** (0.005)
0.884 (0.175)
Poor
0.865 (0.278)
0.822 (0.331)
1.037 (0.856)
0.590 (0.099)
0.942 (0.820)
1.218 (0.187)
0.700* (0.015)
0.580*** (0.000)
0.676** (0.007)
1.308 (0.133)
1.663 (0.144)
0.703 (0.089)
0.603*** (0.000)
0.651*** (0.000)
100%−125%
0.705*** (0.001)
0.646* (0.011)
1.332 (0.115)
1.662 (0.164)
0.583 (0.052)
0.619*** (0.000)
0.632*** (0.000)
125%−200%
0.692*** (0.000)
0.688** (0.004)
1.641** (0.004)
1.637 (0.181)
0.974 (0.899)
0.680*** (0.000)
0.751*** (0.000)
200%−400%
0.813*** (0.000)
0.924 (0.533)
1.522* (0.025)
1.038 (0.910)
0.927 (0.606)
0.797** (0.001)
0.780*** (0.000)
1.139 (0.517)
1.425 (0.169)
1.827 (0.054)
1.596 (0.257)
1.535 (0.252)
1.074 (0.724)
0.907 (0.605)
Asthma
1.576*** (0.000)
1.601** (0.008)
1.661*** (0.000)
1.988** (0.004)
1.182 (0.516)
1.233** (0.002)
1.201* (0.013)
Coronary Heart Disease
1.363 (0.057)
1.373 (0.059)
1.408 (0.141)
0.874 (0.630)
1.258 (0.300)
1.544** (0.008)
1.370* (0.044)
Diabetes
2.498*** (0.000)
1.887*** (0.000)
1.730** (0.006)
1.328 (0.144)
1.820** (0.002)
2.222*** (0.000)
2.714*** (0.000)
Emphasema
1.635 (0.051)
1.039 (0.826)
1.537 (0.146)
1.927 (0.110)
1.190 (0.600)
1.538* (0.031)
1.672* (0.015)
Hypertension
1.885*** (0.000)
1.868*** (0.000)
1.874*** (0.000)
1.536* (0.032)
1.629*** (0.000)
2.007*** (0.000)
2.249*** (0.000)
Miocardial Infarction
0.943
0.795
0.591
0.688
0.934
0.887 1.018 (continued on next page)
N CHC penetration rate (with number of residents with income below 200% of poverty as the denominator)
Race and Ethnicity (reference: Non-Hispanic White) Non-Hispanic, Black
Self-rated Health (reference: Excellent) Very Good
Household Income Relative to Federal Poverty (reference: > 400%) < =100%
Presence of Chronic Conditions (reference: none) Angina
4
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 2 (continued) Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
(0.743)
(0.181)
(0.146)
(0.210)
(0.773)
(0.519)
(0.926)
Other Heart Disease
**
1.287 (0.002)
**
1.520 (0.006)
0.963 (0.821)
1.078 (0.754)
1.198 (0.194)
**
1.324 (0.008)
1.450*** (0.000)
Stroke
1.478* (0.048)
1.400* (0.030)
1.421 (0.107)
1.896* (0.014)
1.108 (0.647)
1.464** (0.008)
1.637** (0.003)
0.881 (0.527)
0.617** (0.001)
1.067 (0.737)
0.568* (0.015)
0.701 (0.099)
0.971 (0.901)
1.000 (0.999)
1.327 (0.064)
1.032 (0.840)
1.044 (0.800)
1.044 (0.853)
0.545** (0.004)
1.544* (0.018)
1.434** (0.003)
0.994 (0.054)
0.995 (0.439)
1.000 (0.920)
0.982* (0.017)
0.985 (0.068)
1.003 (0.311)
1.004 (0.200)
Percent Black
0.996 (0.052)
1.004 (0.324)
1.000 (0.945)
1.003 (0.624)
0.999 (0.946)
0.995* (0.013)
0.996* (0.032)
Percent Hispanic
0.994*** (0.001)
1.002 (0.574)
1.000 (0.926)
1.003 (0.556)
1.002 (0.763)
0.991*** (0.000)
0.993*** (0.000)
In a Medically Underserved Area
0.917 (0.182)
0.964 (0.803)
1.085 (0.311)
0.989 (0.944)
1.367* (0.048)
1.022 (0.701)
0.974 (0.611)
In a Health Professional Shortage Area
0.979 (0.757)
0.931 (0.602)
0.909 (0.220)
1.003 (0.988)
1.072 (0.761)
0.981 (0.755)
0.992 (0.892)
Constant
4.109*** (0.000)
0.267*** (0.000)
4.500*** (0.000)
1.289 (0.566)
0.442* (0.015)
0.430*** (0.000)
1.951*** (0.000)
Activity Limitations (reference: none) Has at Least One ADL
Has at Least One IADL
Area Characteristics Poverty Rate
* **
Statistically Significant, p < 0.05 Statistically Significant, p < 0.01 Statistically Significant, p < 0.00
***
percentage points less likely to have had an office visit during the year when compared to their counterparts in Primary Care Services Areas with a CHC penetration rate of 100%. The odds ratios underlying the other lines in Fig. 2 are not statistically significant.
year, Medicaid only throughout the year, Medicare only throughout the year and Uninsured throughout the year. Among the uninsured, individuals living in a Primary Care Service Area with a penetration rate of zero are, on average, sixteen percentage points less likely to have a usual source of care than those living in counties with a penetration rate of 100% (41–57%). From the mean, a 21% point increase in the penetration rate, approximately one standard deviation, is associated with a three percentage point increase in the likelihood of having a usual source of care (44–47%). The association between CHC penetration rate and the likelihood of having a usual source of care among those with Medicaid is somewhat weaker than that among those with no insurance. Among Medicaid enrollees, individuals living in a Primary Care Service Area with a penetration rate of zero are, on average, eight percentage points less likely to have a usual source of care than those living in a Primary Care Service Area with a penetration rate of 100% (84% and 92% respectively). For the same group, a 21% point increase in the penetration rate from the mean is associated with a two percentage point increase in the likelihood of having a usual source of care (86–88%). Individuals with private insurance living in a Primary Care Service Area with a CHC penetration rate of zero are four percentage points less likely to have a usual source of care than their counterparts in Primary Care Service Areas with a penetration rate of 100%. As in Fig. 1, the line representing individuals with no insurance in Fig. 2 has an upward slope, suggesting a positive association between the CHC penetration rate and the likelihood of having an office-based visit during the year. Among the uninsured, those living in a Primary Care Service Area with a CHC penetration rate of zero are ten
4. Discussion We find that the availability of CHC services, as measured by the penetration rate, is positively associated with the likelihood of having a usual source of care not only among individuals on Medicaid and the uninsured, but also among those with private insurance. The penetration rate is positively associated with having at least one office based visit, but only for individuals who are uninsured. These associations persist after controlling for a variety of individual- and communitylevel variables. In supplemental analyses, we investigated other measures of CHC availability, including whether an area had any CHCs and the count of CHCs, and found a similar pattern of results, though standard errors were larger resulting in fewer significant results. We suspect that this is because the penetration rate captures the availability of CHC services with less measurement error than the other measures examined. Several limitations should be considered when interpreting these findings. First, results from past research suggest that findings can be sensitive to the geographic unit at which area characteristics are operationalized. While Primary Care Service Areas are designed to approximate primary care markets, they do so imperfectly. For example, they are constructed entirely with Medicare claims data and therefore do not necessarily reflect the entire primary care market. We 5
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 3 Odds ratios and p-values from logistic regression models on having one or more office-based provider visits stratified by full-year insurance status. Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
77,455 1.001
10,812 1.001
33,578 1.003
3971 0.998
7777 0.995
33,754 1.005***
29,065 1.002
(0.413)
(0.644)
(0.088)
(0.565)
(0.076)
(0.000)
(0.224)
Men
2.026*** (0.000)
1.453*** (0.000)
1.250*** (0.000)
1.308 (0.071)
1.572*** (0.000)
2.295*** (0.000)
1.992*** (0.000)
Age
0.994*** (0.000)
1.011*** (0.000)
0.982*** (0.000)
1.014*** (0.001)
1.019*** (0.000)
1.000 (0.889)
1.000 (0.818)
0.556*** (0.000)
0.524*** (0.000)
0.578*** (0.000)
0.858 (0.496)
0.359*** (0.000)
0.659*** (0.000)
0.569*** (0.000)
Hispanic of Any Race
0.671*** (0.000)
0.737 (0.053)
0.636*** (0.000)
0.703 (0.189)
0.396*** (0.000)
0.653*** (0.000)
0.719*** (0.000)
Non-Hispanic, Asian
0.526*** (0.000)
0.530** (0.002)
0.527*** (0.000)
0.866 (0.660)
0.465** (0.001)
0.496*** (0.000)
0.581*** (0.000)
Non-Hispanic, Other Race
0.878 (0.115)
0.449** (0.003)
0.720** (0.008)
1.797 (0.287)
0.815 (0.746)
0.918 (0.497)
0.674*** (0.001)
1.267*** (0.000)
1.305* (0.048)
1.099 (0.069)
1.113 (0.672)
1.159 (0.369)
1.156** (0.004)
1.175** (0.001)
Good
1.472*** (0.000)
1.457* (0.013)
1.321*** (0.000)
1.414 (0.186)
1.540* (0.014)
1.480*** (0.000)
1.298*** (0.000)
Fair
2.855*** (0.000)
1.517* (0.012)
2.903*** (0.000)
2.068* (0.013)
2.458*** (0.001)
2.116*** (0.000)
1.697*** (0.000)
Poor
3.881*** (0.000)
1.303 (0.203)
2.240*** (0.000)
1.741 (0.135)
2.141* (0.026)
3.131*** (0.000)
2.596*** (0.000)
0.608*** (0.000)
0.446*** (0.000)
0.724* (0.023)
0.899 (0.798)
0.542* (0.023)
0.588*** (0.000)
0.728*** (0.000)
100%−125%
0.568*** (0.000)
0.622* (0.017)
0.735* (0.042)
0.925 (0.871)
0.483* (0.017)
0.570*** (0.000)
0.746*** (0.001)
125%−200%
0.641*** (0.000)
0.574*** (0.000)
0.744* (0.049)
0.858 (0.725)
0.719 (0.084)
0.652*** (0.000)
0.753*** (0.000)
200%−400%
0.743*** (0.000)
0.693** (0.007)
0.817 (0.167)
0.635 (0.303)
0.850 (0.231)
0.720*** (0.000)
0.787*** (0.000)
1.292 (0.217)
1.590 (0.089)
1.083 (0.775)
0.861 (0.615)
0.828 (0.571)
0.977 (0.896)
0.888 (0.501)
Asthma
1.969*** (0.000)
2.089*** (0.000)
1.602*** (0.000)
1.988** (0.009)
1.762* (0.044)
1.388*** (0.000)
1.601*** (0.000)
Coronary Heart Disease
1.333 (0.054)
1.528* (0.012)
1.736* (0.028)
1.124 (0.637)
2.127* (0.020)
1.422* (0.030)
1.240 (0.169)
Diabetes
2.785*** (0.000)
1.602** (0.002)
3.007*** (0.000)
2.722*** (0.000)
1.534** (0.008)
2.221*** (0.000)
2.362*** (0.000)
Emphasema
1.267 (0.298)
1.276 (0.213)
1.875* (0.042)
0.824 (0.621)
1.233 (0.503)
1.599* (0.036)
1.105 (0.659)
Hypertension
1.918*** (0.000)
2.430*** (0.000)
2.463*** (0.000)
1.698** (0.003)
2.475*** (0.000)
2.185*** (0.000)
2.091*** (0.000)
Miocardial Infarction
0.843
0.965
1.130
0.942
0.789
0.894 1.572* (continued on next page)
N CHC penetration rate (with number of residents with income below 200% of poverty as the denominator)
Race and Ethnicity (reference: Non-Hispanic White) Non-Hispanic, Black
Self-rated Health (reference: Excellent) Very Good
Household Income Relative to Federal Poverty (reference: > 400%) < =100%
Presence of Chronic Conditions (reference: none) Angina
6
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Table 3 (continued) Full year insurance status Private only
Medicare only
Medicaid only
Medicare & medcaid
Medicare & private
Uninsured
Other insurance
(0.307)
(0.863)
(0.658)
(0.842)
(0.422)
(0.606)
(0.024)
***
Other Heart Disease
1.826 (0.000)
2.047 (0.000)
1.098 (0.567)
1.514 (0.122)
2.171 (0.001)
1.405 (0.000)
1.745*** (0.000)
Stroke
1.710** (0.003)
0.999 (0.995)
0.939 (0.809)
0.645 (0.083)
1.142 (0.577)
1.304 (0.056)
1.724** (0.002)
1.456 (0.174)
0.826 (0.330)
1.376 (0.075)
0.871 (0.565)
1.132 (0.702)
1.190 (0.594)
1.419 (0.074)
2.509*** (0.000)
1.095 (0.637)
1.809*** (0.000)
1.446 (0.065)
0.705 (0.102)
2.269*** (0.000)
1.604*** (0.000)
0.988*** (0.000)
0.994 (0.295)
0.998 (0.535)
1.004 (0.696)
0.972*** (0.000)
0.998 (0.316)
0.995 (0.072)
Percent Black
1.000 (0.818)
1.001 (0.840)
0.996* (0.020)
0.990 (0.053)
1.006 (0.343)
0.996** (0.002)
0.997 (0.148)
Percent Hispanic
0.997** (0.002)
1.001 (0.676)
1.000 (0.990)
0.999 (0.777)
1.004 (0.384)
0.996** (0.008)
0.998 (0.238)
In a Medically Underserved Area
0.970 (0.441)
0.977 (0.856)
1.021 (0.700)
0.840 (0.305)
1.810*** (0.000)
0.912* (0.039)
0.985 (0.756)
In a Health Professional Shortage Area
0.950 (0.275)
0.865 (0.267)
0.907 (0.126)
1.428 (0.060)
1.005 (0.974)
0.944 (0.252)
0.921 (0.222)
Constant
1.959*** (0.000)
1.644 (0.083)
3.781*** (0.000)
1.432 (0.450)
2.090 (0.050)
0.279*** (0.000)
0.976 (0.817)
Activity Limitations (reference: none) Has at Least One ADL
Has at Least One IADL
Area Characteristics Poverty Rate
* **
***
***
Statistically Significant, p < 0.05 Statistically Significant, p < 0.01 Statistically Significant, p < 0.00
***
Fig. 1. Mean predicted percent with a usual source of care by insurance status and CHC penetration rate.
7
***
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
Fig. 2. Mean predicted percent with at least one office visit by insurance status and CHC penetration.
particular CHC program that result in an “exogenous” increase or decrease in CHC service availability, under some circumstances prepost differences in access can be used to estimate a causal effect. We consider this paper a first step towards more rigorous causal analysis of the relationship between CHC availability and access to health care.
tested the sensitivity of our results by estimating our models again with CHC penetration calculated at the county-level and find that our findings changed little. We continue to believe that Primary Care Service Areas are the most appropriate geographic unit at which to calculate CHC penetration, but acknowledge that measurement error may attenuate our results. Though a variety of individual and community characteristics are included in our analysis, there still may be unobserved characteristics that are related both to access and to the penetration rate in a particular area. Unobserved health differences in an area, for example, could drive both our measures of access (especially having an office visit) and the number of unique CHC patients in a community, the numerator in the penetration rate. Our results are not sensitive, however, to the exclusion of variables measuring health status. A similar limitation is that CHCs are not placed randomly throughout the country but, rather, placed where they are most needed. Medical need, in turn, could be related to our measures of access. Individuals with high levels of medical need are more likely to have medical use, including office-based visits, and could be more likely to have a usual source of care due to frequent contact with the health care system. To ameliorate this concern, we included dichotomous variables indicating whether a geographic area was in a Medically Underserved Area (MUA) or a Health Professional Shortage Area (HPSA). These designations are extremely influential in determining CHC locations. When these variables were excluded from the model, results changed very little, suggesting that our results are not entirely due to the selection of CHCs into needy areas. Despite these sensitivity checks, the problem of endogeneity of CHC penetration is a serious limitation in this study and results should not be interpreted as causal. Future research should investigate the effects of the availability of CHC services on access by using data and analytic methods that can control for the non-random selection of CHCs into primary care markets and of people into residential areas. One approach is to use longitudinal data to estimate a person-level fixed effects model, which would remove the influence of all characteristics that do not change within person. In effect, individuals act as their own controls. Future research might also consider a quasi-experimental approach to estimating the effect of the availability of CHC services on access. For example, if a state or municipality makes changes to a
5. Conclusion One of the primary goals of the Affordable Care Act is to ensure access to affordable medical care for all Americans and it aims to achieve this largely by increasing the number of people with insurance coverage, via Medicaid or private insurance. The success of this approach hinges, at least in part, on the extent to which the supply of primary care meets the increase in demand driven by health insurance expansions. Our findings are consistent with the idea that an expanded CHC program may help to ensure that Affordable Care Act's health insurance expansions translate into improvements in access to affordable medical care, especially in states that opt to expand their Medicaid programs and therefore have a large number of new enrollees seeking primary care services. In all states, however, CHC service availability may improve access to care for those who remain uninsured or underinsured, and perhaps even improve access to individuals who have private insurance. Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.hjdsi.2016.12.006. References 1 Health Resources and Services Administration. 2013 Health Center Data, 2013 National Data, Table 3A – Patients by Age and Gender. 2013; 〈http://bphc.hrsa.gov/ uds/datacenter.aspx?Q=tall & year=2013 & state=〉. 2 Haberlein M, Brooks T, Artiga S, Stephens J. Getting into Gear for 2014: shifting New Medicaid Eligibility and Enrollment Policies into Drive. Menlo Park, CA: The Henry J. Kaiser Family Foundation; 2013. 3 Congressional Budget Office. Insurance coverage provisions of the affordable care Act—CBO’s April 2014 baseline. 〈http://www.cbo.gov/sites/default/files/cbofiles/ attachments/43900-2014-04-ACAtables2.pdf〉; 2014. 4 Bisgaier J, Rhodes K. Auditing access to specialty care for children with public
8
Healthcare (xxxx) xxxx–xxxx
J.B. Kirby, R. Sharma
2009;25(1):8–16. 11 Cohen J. The medical expenditure panel survey: a national health information resource. Inquiry. 1996;33(4):373–389. 12 Cohen J. Design and Methods of the Medical Expenditures Panel Survey Household Component. Rockville, MD: Agency for Health Care Policy and Research; 1997. MEPS Methodology Report No. 1, AHCPR Pub. No. 97-0026 13 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;38:287–309. 14 Agency for Healthcare Research and Quality. National Healthcare Disparities Report. Rockville, MD: Agency for Healthcare Research and Quality; 2013 15 Agency for Healthcare Research and Quality. Documentation for MEPS Full-Year 2013 Consolidated Data File. 2015; 〈https://meps.ahrq.gov/data_stats/download_ data/pufs/h163/h163doc.shtml〉 Accessed 6/17/2016, 2016; 2016 Accessed 6.17. 2016, . 16 StataCorp . Stata: release 14 Software. Survey Data Rerence Manual, College Station, TX: StataCorp, LP; 2015.
insurance. N Engl J Med. 2011;364:2324–2333. 5 Boukus E, Cassil A, O'Malley A. A snapshot of US physicians: Key Findings from the 2008 Health Tracking Physician Survey, Washington, DC: Center for Studying Health System Change; 2009. 6 Sabik LM, Gandhi SO. Impact of changes in medicaid coverage on physician provision of safety net care. Med Care. 2013;51(11):978–984. 7 Hadley J, Cunningham P. Availability of safety net providers and access to care of uninsured persons. Health Serv Res. 2004;39(5):1527–1546. 8 Cunningham P. What accounts for differences in the use of hospital emergency departments across US communities?. Health Aff (Millwood). 2006;25:W324–W336. 9 McMorrow S, Zuckerman S. Expanding federal funding to community health centers slows decline in access for low‐income adults. Health Serv Res. 2014;49(3):992–1010. 10 Rust G, Baltrus P, Ye J, et al. Presence of a community health center and uninsured emergency department visit rates in rural counties. J Rural Health.
9