Sample
We used a cross-sectional, ecologic design to explore relationships between county-level rates of ACS diagnoses and county level covariates, including the presence of a CHC, RHC, or both. ACS hospitalization rates are calculated based on hospitalizations of persons who reside in the county, regardless of where the hospitalization takes place. The unit of analysis is an age- and county-specific rate, calculated based on discharges of county residents. Counties were used, rather than smaller geographic units, because most of the data elements needed for multivariate analysis are available only at this level. The study was approved by the Institutional Review Board of the University of South Carolina.
Data were drawn from the 2002 State Inpatient Databases (SIDs). The SIDs, compiled at the state level and supported for research use by the Agency for Healthcare Research and Quality, contain discharge records for all hospitalizations in participating states (100 percent). Only fifteen states included information about patients' counties of residence in 2002. Budgetary constraints coupled with the per-state cost for SID files limited the analysis to eight states: Colorado, Florida, Kentucky, Michigan, New York, North Carolina, South Carolina, and Washington. These states were chosen to provide at least one state in each of the four major Census Divisions of the US, and to offer a large number of counties with CHCs, RHCs, or both facilities. The presence of a CHC or RHC in each county was determined using Area Resource File data for the year 2002. Four mutually exclusive categories were created: CHC but no RHC, RHC but no CHC, both (CHC and RHC), and neither facility. Across all counties, 59 (10.2%) had a CHC and not an RHC; 139 (24.0%) had an RHC but not a CHC; 27 had both facilities (4.7%), and 354 (61.1%) counties had neither facility. Counties were distributed by state as follows: Colorado, 63 counties (10.9%); Florida, 66 counties (11.4%); Kentucky, 120 counties (20.7%); Michigan, 83 counties (14.3%), New York, 62 counties (10.7%); North Carolina, 100 counties, 17.3%; South Carolina, 46 counties (7.9%), and Washington, 39 counties (6.7%).
Measurement of ACS Conditions
We used definitions for ACS diagnoses from the Agency for Healthcare Research and Quality. [
31] The ACS conditions for adults are specific diagnoses for asthma, angina (without procedure), congestive heart failure (CHF), bacterial pneumonia, chronic obstructive pulmonary disease (COPD), dehydration, diabetes long-term complications, diabetes short-term complications, hypertension, lower-extremity amputation for individuals with diabetes, perforated appendix, uncontrolled diabetes, and urinary tract infection. For children, the ACS conditions include asthma, bacterial pneumonia, dehydration, perforated appendix, gastroenteritis, and urinary tract infection. The precise definitions used in this research account for a variety of exclusions detailed in technical specifications that are readily available from the AHRQ [
31]. For children, for example, hospitalizations for asthma are excluded if there is evidence of cystic fibrosis or anomalies of the respiratory system. A hospitalization for any of these diagnosis is considered to be a hospitalization for an ACS condition. We did not attempt to study hospitalizations for individual diagnoses because of the instability of rates in counties with very small populations, and also because most research in this area uses the combined indicator.
Analytic approach
We examined adjusted rates of ACS hospitalization in counties with a CHC, an RHC, or both, and compared these to the analogous rates in counties with neither facility. We calculated ACS rates separately for children (0–17), working age adults (18–64), and older individuals (65 and over). To ensure stable rate estimation, we established population-based criteria for county inclusion before conducting data analysis. County-level population estimates for 2002 were drawn from the 2005 Area Resource File (n = 579 counties). We included a county in the rate analysis for children and for working age adults only if it had at least 1,000 persons ages 0 – 17 (children) or ages 18 – 64 (working age adults). For age 65 and over, we included a county only if it had at least 500 persons in that age group; the threshold of inclusion was lower for older persons because of their higher ACS admission rates. These criteria excluded 21/579 counties from the analysis for children (3.6%), 5/579 counties from the analysis for working age adults (0.9%), and 12/579 counties from the analysis for older adults (2.1%). We considered an alternative approach, retaining all counties in the analysis and adjusting standard errors to account for heteroskedasticity. We judged that this approach might not adequately account for unrepresentative high or low ACSH rates that could appear among such small populations. Such unrepresentative rates could introduce bias into the estimations, because they could be attributable to even small random variations in the number of individuals hospitalized for ACSCs in these small populations, rather than to differences in access to primary health care. A comparison of mean county population and mean number of ACS discharges for included and excluded counties is provided in Table
1.
Table 1
Mean Number of Persons in Each Age Range for Included and Excluded Counties, and Mean Number of ACSC Hospitalizations in Each Age Range in these Counties
| Number of Counties | Mean Population | Mean Number of ACSC Discharges |
Included Counties | 559 | 31,535 | 152.4 |
Excluded Counties | 20 | 614 | 1.8 |
County Total | 579 | | |
|
Ages 18–64
|
Included Counties | 574 | 78,778 | 687.1 |
Excluded Counties | 5 | 661 | 4.6 |
County Total | 579 | | |
|
Ages 65+
|
Included Counties | Number of Counties | Mean Population | Mean Number of ACSC Discharges |
Excluded Counties | 567 | 16,912 | 1,111 |
County Total | 12 | 276 | 14.7 |
| 579 | | |
As noted in the introduction, CHCs and RHCs are located only in specific county types and are not randomly distributed across the US. As illustrated in Table
2, counties with CHCs and/or RHCs differ from counties in those same states with neither facility in several characteristics, including HMO penetration and proportion of the population that is uninsured. To adjust for differences between studied counties and counties with neither facility, adjusted analyses controlled for the county characteristics listed in Table
2.
Table 2
Characteristics of counties, by CHCs/RHCs in the county, studied states, 2002.
SID Sample, n = 579 | CHC Only | | RHC Only | Both CHC and RHC | | Neither facility | |
Number of Counties: | 59 | | 139 | 27 | | 354 | 3,168 |
Resources in county:
| | | | | | | |
MD/DO per 10,000 population | 12.9 | | 10.3 | 14.3 | | 12.3 | 12.1 |
Beds per 10,000 population | 3.6 | | 3.2 | 3.9 | | 3.2 | 3.9 |
Number of hospitals with emergency department | 1.7 | | 1.0 | 1.4 | | 1.6 | 1.3 |
HMO penetration rate | 25.3 | *** | 9.4 | 10.6 | | 14.2 | 11.4 |
ED visits per 1,000 population | 337 | | 372 | 382 | | 330 | 351 |
Non-metropolitan county (%) | 23.7 | *** | 79.9 | 66.7 | *** | 58.8 | 65.3 |
Characteristics of county population:
| | | | | | | |
Percent of population that is: | | | | | | | |
African American | 20.3 | | 17.5 | 16.2 | | 16.4 | 9.5 |
Hispanic white | 6.4 | | 6.6 | 7.7 | | 6.4 | 5.3 |
Asian | 1.7 | | 1.6 | 3.5 | | 2.1 | 1.0 |
American Indian/Native American | 2.1 | | 5.1 | 2.1 | | 2.3 | 1.9 |
Population change, 1990 – 2000 (%) | 10.4 | | 12.5 | 8.5 | | 13.0 | 8.1 |
Percent of population with less than a high school education | 24.9 | | 24.3 | 24.0 | | 25.0 | 22.6 |
Population per square mile | 167 | | 141 | 183 | | 219 | 23 |
Percent of population that is unemployed | 7.3 | | 6.8 | 6.3 | | 7.0 | 7.1 |
Percent uninsured, aged 18–64 | 19.0 | | 20.8 | 22.1 | ** | 18.9 | 19.6 |
Percent uninsured, age 17 or less | 12.0 | | 12.9 | 13.3 | * | 11.4 | 12.4 |
Median household income | 35,595 | | 35,179 | 36,835 | | 35,844 | 35,363 |
Death rate per 10,000 due to:
| | | | | | | |
Cardiovascular disease | 18.3 | | 15.6 | 17.1 | | 17.0 | 20.7 |
Chronic obstructive pulmonary disease | 5.1 | | 4.6 | 5.1 | | 4.7 | 5.1 |
Diabetes | 2.6 | | 2.2 | | 2.2 | 2.2 | 2.8 |
Liver disease | 1.0 | | 0.9 | | 0.9 | 0.9 | 0.9 |
The models for this study are based on Andersen's (1995) conceptualization of use of health services as resulting from the multiple influences of the external community and health services environment, population characteristics, health behavior, and outcomes. [
32] Variables representing health system characteristics and use included physician supply, bed supply, number of hospitals with an emergency department, emergency department visit rates, and managed care penetration rates. Physician supply is generally inversely related to ACS hospitalization rates [
17,
20,
33], but a positive relationship [
34] and no relationship [
35,
36] have also been found. Managed care penetration has been found to be inversely related to ACS hospitalization rates. [
37,
38] County characteristics measured included racial/ethnic composition of the population (proportions that are non-Hispanic black, Hispanic, Asian American, and American Indian/Native American), population change 1990 – 2000; the percent of the population with less than a high school education, the unemployment rate, population per square mile, and whether the county was classified as metropolitan (urban) or non-metropolitan (rural). [
39] The racial/ethnic composition of the population is included to adjust for differing patterns of health and health care use among minorities. [
17,
40‐
42] Population change, education levels, and unemployment are used as measures of the financial and economic status of the county as a whole. Population density is used, in addition to rural status, to adjust for differences within rural counties. Including a rural/urban variable in the model does not introduce unacceptable colinearity with the covariate representing RHCs, because a notable proportion of counties with RHCs are classified as metropolitan (Table
2). Resource characteristics included median household income and the percent of the population estimated to lack health insurance. Estimates of the uninsured population in each county were obtained from the U.S. Census. [
43] Consistent with previous research, we included four covariates to control for county health burdens: unadjusted death rates from cardiovascular disease, chronic obstructive pulmonary disease, diabetes, and liver disease. [
32] Table
2 provides a full description of these parameters across county types. With the exception of county-level estimates of the uninsured population, all variables are drawn from the Area Resource File.
Multivariate Poisson analysis was used to calculate adjusted rate ratios comparing counties with one or more CHCs, one or more RHCs, or at least one CHC plus at least one RHC, to counties having none of these facility types, while holding other county characteristics equal. The rate ratio is the ratio of the mean value of ACS hospital admission rates across counties of a given type, separately estimated for each age group, where the mean rate for a county type of interest (such as counties with both a CHC and an RHC) is the numerator. The denominator is the corresponding rate for counties having neither a CHC nor an RHC, the reference category. The rate ratio is obtained by exponentiating the estimate of interest from the Poisson analysis. Rate ratios less than 1.00 suggest that the hospitalization rate in the county type of interest was lower than the rate in the reference category.
For calculating rates among uninsured adults, we used Census estimates of the number of uninsured adults in each county as the denominator. We made the assumption that nearly all such persons are younger than 65, as most older people are covered by Medicare. For the separate analysis of children, the denominator was the Census estimate of uninsured children. The numerator specific to each age group in each county was the number of ACS admissions for which the payment source was identified as "self pay" in the discharge record. This value may not precisely equal the uninsured population, as some self-pay admissions may later have been converted to an insurer; however, it is reasonable to assume that the number of cases in which this occurred is relatively small. Measurement errors, if present, might have the greatest effect on ACS admission rates among children, which are generally quite low and thus could be affected by small changes.