D
Commissioned Papers



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance D Commissioned Papers

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance This page in the original is blank.

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance The Impact of Uninsured Populations on the Availability of Hospital Services and Financial Status of Hospitals in Urban Areas Darrell J. Gaskin and Jack Needleman ABSTRACT Objective: To identify the effects of the percentage of uninsured persons in a community on the availability of hospital services for the entire community. Data and Study Design: Our analysis focuses on the 85 largest metropolitan statistical areas (MSAs) during the 1990s and relies on data from the March Current Population Survey, the American Hospital Association’s (AHA) Survey of Hospitals and Medicare Cost Reports. We estimate the impact of the uninsured rate on hospital margins and four measures of hospital service availability, i.e., capacity, services to vulnerable populations, community services, and high-tech services. We estimate two sets of regression models, MSA-level and hospital-level models. Findings: We find that as the uninsured rate increased the availability of some hospital services declined. The results of the MSA and hospital level analyses aare consistent for the measures of capacity, services to vulnerable populations and community services. The uninsured rate was negatively related to beds per capita in the MSA and the average hospital size. The availability of services for vulnerable populations and community services and the propensity for hospitals to offer these services is negatively associated with the percentage of uninsured residents. The results for hi-tech services for the MSA- and hospital-level analyses are not congruent. The results from the MSA-level analysis suggest that the uninsured rate is negatively

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance associated with the availability of some high-tech services. However, the results of the hospital-level analysis suggest that as the uninsured rate increases, hospitals are more likely to offer some high-tech services, specifically extracorporeal shock-wave lithotripsy (ESWL), angioplasty, and magnetic resonance imaging (MRI). The results of the MSA- and hospital-level analyses also differ for hospitals’ financial health. The MSA level results suggest that hospitals are negatively impacted by the rate of uninsurance. The hospital-level results suggest that there is no association. Conclusion: Our findings suggest that the lack of health insurance not only creates an access to care problem for uninsured individuals but also reduces the availability of hospital services to the entire community. INTRODUCTION An estimated 41 million Americans or 16.5 percent of the population under age 65 lacked health insurance in 2001 (Mills, 2002). The lack of health insurance has a significant impact on the health and well-being of uninsured persons. A number of studies have shown that uninsured persons have less access to health care services and as a result have worse health outcomes and lower overall health status (IOM, 2002a; 2002b). The uninsured are less likely to receive preventive and screening services compared to persons with health insurance. Uninsured persons with chronic conditions are less likely to receive appropriate care to manage their health conditions (Mandelblatt et al., 1999; Powell-Griner et al., 1999; Zambrana et al, 1999; Ayanian et al., 2000; Cummings et al., 2000; Breen et al., 2001). Compared to insured persons, uninsured persons are less likely to have a usual source of care and less likely to seek care when they felt they needed it (IOM, 2001a). When hospitalized, the uninsured receive fewer services, are more likely to receive substandard care than insured patients, and are at greater risk of dying during the hospital stay or soon after discharge (Hadley et al., 1991; Burstin et al., 1992; Haas and Goldman, 1994; Blustein et al., 1995). Persons without health insurance have poorer heath outcomes for an episode of illness and higher overall mortality rates (Ayanian et al., 1993; Blustein et al., 1995; Canto et al., 2000; Roetzheim et al., 2000). While research has shown the persons who are uninsured face significant barriers to health care, little is known about how the overall percentage of those without health insurance, i.e., the uninsured rate, affects access to care for their community. Theoretically, the size and scope of the health care delivery system in a community is determined by the intersection of the demand for health care in a community with health care providers’ ability and willingness to offer services. Because insurance pays for a large proportion of health care services, the distribution of community residents by source of payment will partially determine their demand for health care services. In particular, health insurance coverage does tend

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance to increase individuals’ demand for health care services. Uninsured persons use fewer health care services than do similar insured persons. Also, hospitals, physicians and other health care providers receive substantially less reimbursement for the care provided to uninsured patients compared to similar insured patients. Consequently, a high uninsured rate should reduce overall demand for health services in a community. We are concerned particularly about the impact of uninsurance on the availability of hospital services. Several studies have shown that market forces that have depressed demand for hospital services have had an impact on the size of the hospital delivery system. For example, Medicare’s transition from a cost-based reimbursement system to prospective payment and its subsequent reduction in growth of hospital payment rates resulted in lower hospital utilization, a reduction in the intensity of hospital services, and encouraged a reduction in hospital size (Coulam and Gaumer, 1991; Hodgkin and McGuire, 1994). The introduction of managed care also reduced demand for hospital services. Several studies have demonstrated that increased HMO penetration is associated with reduction in hospital utilization, hospital beds, slower hospital cost inflation, and slower revenue growth (Miller and Luft, 1994; Chernew, 1995; Robinson, 1996; Gaskin and Hadley, 1997). Dranove and colleagues (1986) have modeled the impact of managed care penetration on hospitals, and they conclude that downward pressure on hospital prices would result in a reduction in hospital capacity. Given these major changes in the nature of demand for hospital services, hospital behavior in less concentrated markets has transformed from non-price competition to price competition. Prior to the implementation of Medicare prospective payment and the growth of managed care, the empirical evidence indicated that hospitals in less concentrated markets competed on the basis of technology, amenities, and services, e.g., Luft’s so called “medical arms race” (Robinson and Luft, 1985; Luft et al., 1986; Noether, 1988; Frech, 1996). However, studies based on data from the late 1980s and early 1990s present strong evidence of price competition in hospital markets (Robinson and Luft, 1988; Zwanziger and Melnick, 1988; Melnick et al., 1992; Keeler et al., 1999). Hospitals in less concentrated areas charged significantly lower prices compared with those in more concentrated areas. We postulate that similar to other market forces that reduce the demand for hospital services, high uninsured rates will be associated with reduction in hospital capacity. This potentially can have a negative spillover effect on insured patients. For example, if a hospital is unable to maintain its trauma center because of a high uninsured rate, an ambulance may have to carry insured patients further to obtain trauma care. In this study we attempt to identify the effects of the percentage of uninsured persons in a community on the availability of hospital services for the entire community. In particular, we address the following questions: (1) Are hospitals smaller in communities with high uninsured rates? (2) Are hospitals less likely to offer particular types of services in communities with uninsured rates? (3) Does the availability of hospital services for the entire community decline as the

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance percentage of uninsured persons on a community increases? (4) Does uninsurance negatively affect the financial status of hospitals? CONCEPTUAL FRAMEWORK As stated above, we hypothesize that the uninsured rate is negatively related to the demand for hospital services. Specifically, as the uninsured rate increases, the overall demand for hospital services by insured patients decreases while the overall demand for hospital services by uninsured patients (i.e., charity or discount care) increases. Uninsured patients, however, tend to use less hospital care than do insured patients for similar health care needs because of their limited financial resources. As a result, an increase in the uninsured rate should lower and flatten the demand for hospital services in a geographic market. For simplification, assume that hospitals serve two types of patients: insured and uninsured. Payment rates for treating insured patients are typically greater than average costs. Reimbursements for treating uninsured patients are typically less than average cost. Assume hospitals have a cost structure that exhibits increasing or constant returns to scale, that is, total cost increases with the volume of services at an increasing rate. Empirically, hospitals exhibit scale economies up to a moderate size, 150 to 250 beds (Folland et al., 2001). Beyond this size the evidence is mixed regarding whether hospitals experience constant or decreasing returns to scale. Assume that the typical urban hospital operates at volume levels where marginal cost equals average cost. Hospitals use their marginal cost to determine their supply of hospital services. They equate marginal costs with either marginal revenues or average revenues depending upon their market structure (Tirole, 1998). Hospitals that are monopolies or part of oligopolies will set marginal costs equal to marginal revenues. Hospitals that are in monopolistic competitive markets will set their marginal costs equal to average revenues (Chamberlin, 1962). For simplification, suppose that the distribution of each hospital’s patients by source of payment is equal to the distribution of persons in the market by insurance status. In such a market, as the uninsured rate increases, the percentage of uninsured patients treated by each hospital increases and therefore their average and marginal revenues decline. Consequently, each hospital will reduce its supply of services. Obviously, this is a simplification that does not reflect most hospital markets because hospitals differ with respect to serving the uninsured in their geographic markets. However, the general notion that an increase in the uninsured rate places downward pressure on the supply of hospital services is correct. Now relax the assumption that hospitals care for uninsured patients proportionally to their presence in the market area. Suppose in some markets there are safety-net hospitals that provide a disproportionate share of the care to the uninsured while other hospitals provide less than a proportionate share of uninsured patients. In such a market, an increase in the uninsured rate would have a larger effect on safety-net hospitals than on other hospitals. However, if as the uninsured rate increases, non-safety-net hospitals in the community become proportionately

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance more involved in the care for the uninsured, then these hospitals will be more affected by the change in demand. Another assumption implicit in our model is that uninsured and insured patients use hospitals for the same mix of services. However, suppose there is a set of hospital services that the uninsured are more likely to use than insured patients. For these services, hospitals’ profits will tend to be lower than profits associated with other hospital services because of their payer mix. Therefore, we expect that hospitals will cut back on services that the uninsured are more likely to use as the uninsured rate increases. The discussion above assumed that each hospital bases its decisions on its own revenue and cost structure. The framework can also be extended to consider strategic interactions among hospitals in multi-hospital markets. Hospitals facing demand for services from uninsured or nonpaying patients for specific services may prefer to shift these patients to other hospitals. Where this cannot be accomplished directly, an option is to not provide the service or restrict the size of the service by reducing beds available for the service. This will shift uninsured patients to other hospitals but can also increase the travel time and inconvenience for insured patients to obtain the service. In markets where care for the uninsured is concentrated in a few safety net hospitals, this effect may be small. In markets where care for the uninsured is more widely shared or where there are fewer hospitals, these strategic interactions may reduce the availability of services beyond the level needed to adjust for the lower demand from the uninsured. DATA This study relies primarily on three databases: The March Current Population Survey (CPS) the AHA Annual Survey of Hospitals and the Medicare Cost Reports We used data from the March CPS from 1990 to 2000 to calculate the percentage of uninsured residents in the 85 largest MSAs. See the addendum for the list of MSAs. We focus on the 85 largest MSAs because they had large enough subsamples in the CPS data to yield reliable estimates of the uninsured rate. The size of the subsamples range from 143 to 6,148 with a mean of 900, standard deviation of 145, and a median of 623. To improve the accuracy of our estimates, we calculated the percentage of uninsured persons in an MSA by combining data for 3 years, i.e., the current year and the years before and after. The analysis focuses on the years 1991, 1994, 1996, and 1999. In addition, we calculated the percentage of the MSA’s population covered by Medicare and Medicaid using the CPS data. To measure the availability of hospital services, we used data from the AHA Survey of Hospitals for the years 1991, 1994, 1996, and 1999. In particular, we

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance used four measures of hospital service availability: hospital capacity, services to vulnerable populations, high tech services, and community services. To measure capacity we used the number of hospital beds, medical/surgical beds, psychiatric beds, intensive care unit (ICU) beds, and beds devoted to patients diagnosed with alcoholism, drug abuse, or chemical dependency. Five services for vulnerable populations were examined: psychiatric outpatient services, psychiatric emergency room services, psychiatric inpatient services, outpatient, and rehabilitation services for persons diagnosed with alcoholism, drug abuse, or chemical dependency, and services for patients diagnosed with HIV-AIDS. Nine high-tech services were examined. Four require investments in beds as well as in equipment and personnel: trauma center, neonatal intensive care unit, transplant services, and burn units. Five involve investments in equipment and personnel: magnetic resonance imaging (MRI), radiation therapy, angioplasty, single photo emission computerized tomography (SPECT), and extracorporeal shock-wave lithotripsy (ESWL). Three community services were examined: community outreach centers, transportation services, and Meals on Wheels. We have data on these three services only for 1994, 1996, and 1999 because AHA did not collect information on them in 1991. For each of the 17 services, we created a variable that indicates whether the hospital or one of its subsidiaries provided the service. If the hospital reported that it provided the service locally through a partner in its health system, network, or a joint venture, we did not designate this hospital as a provider of the service. This eliminated some double counting. Consider a local health system that consists of an academic health center (AHC) hospital and three community hospitals, a configuration common in New York City (Salit et al., 2002). Suppose the AHC hospital has a burn unit but the community hospitals do not have one. The AHC hospital would report that it provided the service at the hospital. The community hospitals would report that they provided the service locally through the health system. To avoid counting this burn unit four times, we only recognized those that are provided at the hospital or a subsidiary. Ideally, we would use the number of beds provided or volume of visits to measure the magnitude of these services. However, this information is not available for all of the services on the AHA survey. To measure hospital financial status, we used hospital margins calculated from the Medicare Cost Reports. Other hospital characteristics used in the analysis were ownership status, teaching status based on the residents-to-bed ratio, and the percentage of the hospital inpatient days that were covered by Medicaid. When modeling the hospital service availability, we obtained the hospital characteristics from the AHA data and when modeling hospital financial status, we obtained these hospital characteristics from the Medicare Cost Reports. In these analyses we controlled for four health care market factors: hospital market competition, HMO penetration, local costs, and overall demand. To control for the level of hospital competition, we calculated a Herfindahl index, which is equal to the sum of the squares of the market shares, using staffed hospital beds. We obtained HMO penetration data from InterStudy. We used the HCFA

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance wage index as a measure of cost. To control for overall demand adjusting for geographic convenience, we used population density (Porell and Adams, 1995). METHODOLOGY We conducted MSA- and hospital-level analysis on three sets of variables: hospital beds, hospital services, and hospital margins. Beds In the MSA level analysis, we regressed beds per capita for total beds, medical-surgical beds, beds in intensive care units, psychiatric inpatient beds, and beds for treatment of alcohol and chemical dependency on the percent of uninsured residents in the MSA, the other MSA characteristics, (i.e., percent Medicaid, percent Medicare, percent HMO enrollment, level of hospital competition, hospital wage index, and population density) and year categorical variables to control for fixed time effects. We estimated these models using generalized least squares with robust standard errors, controlling for clustering at the MSA level. We gave greater weight to larger MSAs by using the average MSA population as a weight in the regression analysis. We conducted hospital-level regressions of the natural log of beds in each of the five bed categories on the MSA-level variables, plus controls for hospital characteristics such as ownership, teaching status, size, and percent Medicaid patients. We used a general estimating equations framework to construct random effects regression models controlling for unobserved hospital and time effects. We prefer random effects models because the uninsured rate does not vary substantially over time within an MSA, and within MSA variation over the time period we examine is unlikely to drive hospital behavior. For psychiatric beds and alcohol/chemical dependency beds, because a high proportion of hospitals did not provide beds for the service, we used random effect negative binominal count models. For each model, we calculated standard errors using the Huber-White correction and weighted each hospital by its average number of beds during the study period. Since hospitals are nested within MSAs, we adjusted the estimated standard errors to reflect MSA clustering. We also verified that the models were homoscledastic with respect to the uninsured rate using the Breusch-Pagan Test (Greene, 2000). Services For the MSA analysis, we regressed the proportion of hospitals offering specific services on the percent of uninsured in the MSA and the other MSA-level variables described above. We also conducted this regression weighting the proportion of hospitals offering the service by total hospital beds in these hospitals. This weighted analysis gives larger hospitals offering the service more importance

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance and implicitly uses as the dependent variable the proportion of hospital beds in the MSA that are in hospitals offering the service. In both regressions, we estimated these models with robust standard errors and weighted each MSA by its population size. For the hospital-level analysis, we estimated random effect logistic regression models that controlled for hospital-specific effects for each service. The MSA- and hospital-level variables were those described above, with adjustments for clustering at the MSA level. Margin To determine whether the uninsured rate affected the financial status of hospitals, we estimated MSA-level and hospital-level models. For the MSA-level model, the aggregate margin was the dependent variable. This is a measure used by the Medicare Payment Advisory Commission (MedPAC) to characterize the financial health of a category of hospitals. Aggregate margin equals total hospital revenues in the MSA, minus total hospital expenses in the MSA, divided by total revenues. The independent variables were the percent of uninsured residents in the MSA, the percent Medicaid and Medicare enrollees, the percent of public and for-profit hospitals in the MSA, hospital wage index, population density, and year categorical variables to control for fixed time effects. For the hospital-level models, we used the total hospital margin as the dependent variable. Because of outliers, we excluded observations with margins below the third percentile and above 97th percentile. The independent variables in this model were the uninsured rate, the other MSA characteristics, and the hospital characteristics. Similar to the other hospital-level models, we estimated random effects controlling for unobserved hospital and time effects, weighting by hospital beds and calculating robust standard errors using the Huber-White correction. RESULTS Means and standard deviations for the percent uninsured at the MSA level, beds per 100,000 population, the proportion of hospitals offering each service (unweighted), and percentage of beds in hospitals with each service (weighted), and MSA-level margin are presented in Table D.1. The regression coefficients on the proportion of uninsured in the MSA for regressions on beds, services, and margins are presented in Table D.2. Results from regressions of services using the unweighted and weighted availability measure are both presented. Hospital-level regressions are presented in Table D.3. Proportion of Uninsured On average 12.2 percent of the total population in the 85 MSAs studied was uninsured in 1999. The range varied from 6.2 percent uninsured in Allentown-

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance TABLE D.1 Means and Standard Deviations for Percent Uninsured in MSA and Dependent Variables   Unweighted Weighted Variable N Mean SD N Mean SD Percent Uninsured in MSA 340 14.6 5.3   Beds per 100,000 Population Total 340 316.2 106.6 Medical-Surgical 340 173.2 78.0 ICU 340 29.7 10.3 Psychiatric 340 18.5 12.1 Alcohol and chemical dependence 340 3.9 4.3   % %   % % Services for Vulnerable Populations Psychiatric inpatient 340 39.6 18.1 340 59.2 21.6 Psychiatric emergency 340 40.5 19.4 340 58.4 22.1 Psychiatric outpatient 340 32.5 16.7 340 49.3 21.8 Alcohol and chemical dependence 340 25.8 16.7 340 37.8 22.4 AIDS 340 54.1 21.0 340 73.7 18.9 High Technology Services Trauma 340 22.3 13.8 340 38.5 19.8 NICU 340 25.8 12.6 340 42.9 17.9 Transplant 340 16.5 11.2 340 32.2 17.3 Burn 340 5.2 4.9 340 11.1 11.0 MRI 340 47.2 20.6 340 66.2 23.3 Radiation therapy 340 37.2 15.3 340 61.0 18.5 Angioplasty 340 34.2 14.9 340 60.4 18.8 SPECT 340 41.2 16.8 340 57.3 19.4 ESWL 340 14.5 10.7 340 26.0 19.0 Community Services Community outreach 255 56.9 17.8 255 75.4 17.5 Transportation 255 24.8 15.6 255 33.7 22.0 Meals on Wheels 255 12.3 12.7 255 14.2 15.7 Margin 340 3.4 4.8  

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance FIGURE D.1 The impact of uninsurance rate on the demand and supply of hospital services. DATA Data for this study come from four sources: state hospital discharge data from seven states the AHA Annual Survey of Hospitals the Medicare Cost Reports and the Area Resource File compiled by the Bureau of Health Professions We used hospital discharge data from California, Massachusetts, New Jersey, New York, Pennsylvania, Washington, and Wisconsin for 1991, 1994, and 1996. These states were chosen because they distinguished whether patients’ expected source of payment was self pay or charity care. Our analysis focuses on the availability of hospital services in 168 rural counties in our seven states: 23 counties in California, 25 in Florida, 2 in Massachusetts, 21 in New York, 29 in Pennsylvania and 23 in Washington and 45 in Wisconsin. We defined rural counties as those that were not located within an MSA. Of the 168 counties, 56 percent have only one hospital and 69 percent were located

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance TABLE D.4 Comparison of Counties Included in Analysis to All U.S. Rural Counties   Sample U.S. Number 168 2,369 Average County Population, 1997 46,229 22,250 Population Density (people per square mile) 55.9 44.1 Per Capita Income, 1998 $20,748 $18,917 Poverty Rate (percent), 1997 13.5 16.0 Percent Adjacent to Metropolitan Area 69.6 43.4 adjacent to an MSA. As shown in Table D.4, compared to U.S. rural counties as a whole, the counties included in this analysis have larger populations, are more densely populated, have somewhat higher per capita income, and are more likely to be adjacent to metropolitan areas. We identified patients from rural counties using their county of residence information in the discharge database. For each rural county, we calculated the proportion of discharges originating in that county where the expected sources of payment were self-pay or charity, Medicaid, Medicare, or HMO. For ease of exposition, we refer to these proportions as the county’s percent of uninsured discharges, Medicaid discharges, Medicare discharges, and HMO discharges, respectively. We used these measures as proxies for the rate of insurance coverage for hospital care in the county. The percentage of uninsured discharges indicates the amount of inpatient services provided to uninsured persons relative to the entire inpatient market. This is a reasonable proxy for uninsured persons’ share of the demand for hospital services. To measure the availability of hospital services, we used data from the AHA Survey of Hospitals for the years 1991, 1994, and 1996. We used four measures of hospital service availability: hospital capacity, services to vulnerable populations, community services, and high-tech services. To measure capacity we used the number of hospital beds, medical-surgical beds, psychiatric beds, ICU beds, and beds devoted to patients diagnosed with alcoholism, drug abuse, or chemical dependency. Five services for vulnerable populations were examined: psychiatric outpatient services, psychiatric emergency room services, psychiatric inpatient services, outpatient and rehabilitation services for persons diagnosed with alcoholism, drug abuse, or chemical dependency, and services for patients diagnosed with HIV-AIDS. Eight high-tech services were examined. Three require investments in beds, in addition to equipment and personnel: trauma center, neonatal intensive care unit (NICU), and transplant services. Five involve investments in equipment and personnel: magnetic resonance imaging (MRI), radiation therapy, angioplasty, single photo emission computerized tomography (SPECT), and extracorporeal shock-wave lithotripsy (ESWL). Three community services were examined: community outreach centers, transportation services, and Meals on Wheels. We have

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance data on these services for 1994 and 1996 only because the AHA did not collect this information in 1991. For each of the 16 services, we create a variable that indicates whether the hospital or one of its subsidiaries provided the service. If the hospital indicated that it provided the service locally through a partner in its health system, network, or a joint venture, we did not designate this hospital as a provider of the service. This eliminated some double counting. To measure hospital financial status, we used hospital margins calculated from the Medicare Cost Reports. In our analyses, we are specifically interested in how the relative concentration of uninsured discharges affects the availability of hospital services. To measure the relative concentration of uninsured discharges we calculated two Herfindahl indexes, which equal the sum of the squares of the market shares. The first measures the concentration of all hospital discharges in the county and the second measures the concentration of uninsured discharges in the county. We then divided the Herfindahl index for uninsured discharges by the Herfindahl index for all discharges. In counties where this ratio equals one, the uninsured discharges are no more concentrated within a subset of hospitals than are all discharges. In counties where this ratio exceeds one, the uninsured discharges are more concentrated than all discharges. This is an indication that one or a few hospitals in area have assumed a disproportionate role in providing these counties’ safety-net services. We include in our regression the main effect of percent uninsured, main effect of the concentration ratio, and the interaction of the concentration ratio with percent uninsured. Our hypothesis is that the county-level effect of the uninsured percentage will be smaller where the uninsured are concentrated. We examine the interaction of this ratio with the county’s percentage uninsured discharges to determine whether relative concentration dampens the effect of uninsured discharges on the availability of hospital services. To facilitate the interpretation of the coefficients on the percentage of uninsured discharges, the main effect of the relative concentration ratio, and the interaction term, we centered the relative concentration ratio on its mean. We examine the overall effect of percent of uninsured discharges in counties with low and high concentration of uninsured discharges by summing the coefficient on the percent uninsured discharges with the product of the coefficient on the interaction term and relative concentration ratio evaluated at the three distinct points. Specifically, we interpreted the effects of uninsured discharges at the 25th percentile (counties where the uninsured discharges are dispersed relative to all discharges), at the mean (which is equivalent to the main effect of the percent uninsured measure), and the 75th percentile (counties where the uninsured are concentrated relative to all discharges). In addition, we controlled for three other health care market factors. To control for the level of hospital competition, we used the Herfindahl index for all discharges. We use the HCFA wage index as a measure of cost and population density as a measure of overall demand adjusted for geographic convenience (Porell and Adams, 1995).

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance METHODS To address our research questions, we analyzed three sets of variables: hospital beds, hospital services, and hospital margins. Beds We regressed beds per capita in each county in the following categories: total beds, medical–surgical beds, beds in ICUs, psychiatric inpatient beds, and beds for treatment of alcohol and chemical dependency, on the percent of uninsured discharges in the county, relative concentration of uninsured, other county characteristics (i.e., percent Medicaid discharges, percent Medicare discharges, percent HMO discharges, level of hospital competition, hospital wage index, and population density), and year categorical variables to control for fixed time effects. We estimated these models using generalized least squares with robust standard errors, controlling for clustering at the county level. We gave greater weight to larger counties by using the average county population as a weight in the regression analysis. Services We regressed the proportion of hospitals offering specific services on the percent of uninsured discharges in the county, relative concentration of uninsured, and the other county- level variables described above. Because this formulation of the dependent variable does not distinguish the size of the service, we estimated a second set of models using a weighted proportion of hospitals offering specific services. To give larger hospitals offering the specific services more importance, we recalculated this dependent variable using hospital beds as weights. In both regressions, we estimated these models with robust standard errors, and controlled for clustering at the county level. Results from both models are comparable, and we report results only from the weighted regression. Margin To determine whether the percentage of uninsured discharges negatively affects the financial status of hospitals, we estimated a model with aggregate margin as the dependent variable. The aggregate margin is a measure used by MedPAC to characterize the financial health of a category of hospitals. It equals total hospital revenues in the county minus total hospital expenses in the county divided by total revenues. The independent variables were the percent of uninsured discharges in the county, the percents Medicaid and Medicare discharges, the percents of public and for-profit hospitals in the county, hospital wage index, population density, and year categorical variables to control for fixed time effects.

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance RESULTS Means and standard deviations for the percent of discharges for uninsured patients, beds per 100,000 population, the proportion of hospitals offering each service, (unweighted) percentage of beds in hospitals with each service (weighted), and county-level hospital margins for the rural counties in this analysis are presented in Table D.5. The regression coefficients on the proportion of discharges for the uninsured and the interaction of the percentage of uninsured and uninsured concentration ratios for the regressions on beds, services, and margins are presented in Table D.6. Table D.6 also includes the coefficient on percent uninsured at the 25th and 75th percentile of the relative concentration index and the statistical significance of these coefficients. Proportion of Uninsured The proportion of discharges for patients without insurance in the rural counties studied averaged 4.4 percent. This is substantially lower than the percent of uninsured in rural counties, and reflects the fact that the uninsured are younger than the average person in the population and therefore less likely to be hospitalized. (Persons over age 65, virtually all of whom have coverage through Medicare, are disproportionately represented among hospital discharges.) The variation in the proportion of discharges from the uninsured is very large. The standard deviation for the proportion of uninsured is 4.4, as large as the mean, with a range from 0 to 8.8 percent. Beds In the 168 counties in this study, there are an average of 423.3 hospital beds per 100,000 population, with wide variation across the counties (standard deviation = 392.3). The average number of beds per capita and variation in beds is larger among these rural counties than was observed in metropolitan areas (Gaskin and Needleman, 2003). Forty-four percent are medical or surgical beds. ICU beds represent only 4 percent of the beds. There are fewer psychiatric and alcohol and chemical dependence beds (10.0 and 4.2 beds per 100,000, respectively) with wide variation across the counties. There is some evidence that beds per capita are influenced by the percent uninsured in the county. The coefficients on the percent uninsured are consistently negative, although statistically significant only for ICU beds. There is a statistically significant association of ICU beds with percent uninsured at all three levels of uninsured concentration tested. In addition, for psychiatric beds there is a statistically significant association of percent uninsured and bed supply at low levels of concentration of the uninsured (see Table D.6).

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance TABLE D.5 Means and Standard Deviations for Percent Uninsured Discharges and Dependent Variables   Unweighted Weighted Variable N Mean SD N Mean SD Percent Uninsured Discharges 426 4.4 4.4   Beds per 100,000 Population Total 422 423.3 392.3 Medical-Surgical 396 186.8 146.5 ICU 396 17.8 22.1 Psychiatric 396 10.0 19.1 Alcohol and chemical dependence 396 4.2 16.0   % %   % % Services for Vulnerable Populations   Psychiatric inpatient 396 21.7 37.2 396 23.7 39.0 Psychiatric emergency 396 35.1 43.7 396 37.4 44.9 Psychiatric outpatient 396 15.7 33.5 396 16.5 34.5 Alcohol and chemical dependence 396 17.3 33.1 396 17.7 34.1 AIDS 400 39.6 44.1 396 42.4 45.9 High-Technology Services Trauma 396 16.1 33.4 396 16.6 34.1 NICU 426 2.7 12.9 422 3.7 16.7 Transplant 400 6.0 21.8 396 6.4 22.8 MRI 396 20.1 36.0 396 21.3 37.4 Radiation therapy 400 13.4 29.9 396 15.5 32.6 Angioplasty 396 3.6 16.0 396 4.5 18.5 SPECT 396 20.6 36.6 396 21.6 37.8 ESWL 396 6.6 21.9 396 7.4 23.8 Community Services Community outreach 258 51.8 44.4 258 52.5 45.1 Transportation 258 9.9 25.0 258 10.5 27.1 Meals on Wheels 258 18.9 35.2 258 18.5 35.8 Margin 414 1.8 5.9 410 2.0 5.8 Services for Vulnerable Populations With respect to services for vulnerable populations, AIDS services are the most common, with nearly 40 percent of hospitals (with 42.4 percent of the beds) offering these services. Psychiatric emergency services are available in approximately 35 percent of hospitals (with 37 percent of beds), while psychiatric inpa-

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance tient, outpatient, and alcohol and chemical dependence treatment are available in approximately one-in-five to one-in-six hospitals. For all services for vulnerable populations except AIDS services, the coefficient on percent uninsured, although not statistically significant, is negative. For both psychiatric inpatient and psychiatric emergency services there is a statistically significant association of the interaction of the uninsured concentration and percent uninsured on the proportion of hospitals offering this service. At the lower level of concentration, there is a statistically significant association of percent uninsured with availability of inpatient psychiatric services. High-Technology Services Bed-based high-technology services (trauma, NICU, and transplant) are much less commonly available than services for vulnerable populations, with the proportion of hospitals offering such services varying from 16.1 percent for trauma to 2.7 percent for NICU. Among these three services, transplant services are less likely to be available in communities with lower concentrations of uninsured and higher uninsured rates. For trauma and NICU services, while the coefficients on percent uninsured are negative, they are not statistically significant. High-technology services that do not have dedicated beds associated with them vary in their availability in these rural counties. The two imaging services studied—MRI and SPECT—are available in approximately 20 percent of hospitals. Radiation therapy is available in one out of seven hospitals. Lithotripsy and angioplasty are available in relatively few hospitals (6.6 and 3.6 percent of hospitals, respectively). With the exception of SPECT, there is a negative association between these services and the percent uninsured in the county. The association with percent uninsured is statistically significant for MRI and lithotripsy services across the range of uninsured concentrations and statistically significant for radiation therapy when the concentration of the uninsured is low. Community Services Community outreach services were available from approximately half the hospitals in our sample. By contrast, Meals on Wheels services were only available from 19 percent and transportation services from only 10 percent. Provision of Meals on Wheels services was negatively associated with percent uninsured in counties where the concentration of uninsured were low. Hospital Margins The average county hospital margin in our sample is 1.8 percent, with a standard deviation of 5.9 percentage points, reflecting the substantial variation across counties in margins. We observe statistically significant associations between

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance TABLE D.6 Regression Coefficients of Beds, Services, and Margin on Percent Uninsured and Interaction of Percent Uninsured With Uninsured Concentration, County-Level Regressions (Weighted) Variable N Percent Uninsured SE Beds per Capita Total 411 –2.31 (3.08) Medical-Surgical 411 –1.80 (1.18) ICU 386 –0.45a (0.22) Psychiatric 386 –0.37 (0.21) Alcohol and chemical dependence 386 –0.17 (0.12) Services for Vulnerable Populations Psychiatric inpatient 386 –0.91 (0.59) Psychiatric emergency 386 –0.77 (0.63) Psychiatric outpatient 386 –0.67 (0.46) Alcohol and chemical dependence 386 –0.16 (0.54) AIDS 389 0.71 (0.90) High-Technology Services Trauma 386 –0.85 (0.57) NICU 411 –0.35 (0.30) Transplant 386 –0.49 (0.28) MRI 386 –1.67c (0.51) Radiation therapy 389 –0.90 (0.51) Angioplasty 386 –0.43 (0.27) SPECT 386 0.61 (0.84) ESWL 386 –0.91a (10.36) Community Services Community outreach 255 –0.48 (0.70) Transportation 255 0.22 (0.56) Meals on Wheels 255 –0.58 (0.51) Margin 402 –0.17b (0.06) a p<0.05 b p<0.01 c p<0.001

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance Percent Uninsured × Uninsured Concentration SE 25th Percentile of Uninsured Concentration 75th Percentile of Uninsured Concentration 0.56 (1.07) –2.75 –2.15 1.10 (0.50)a –2.66 –1.48 0.11 (0.07) –0.54a –0.42a 0.34 (0.07)c –0.64a –0.28 0.10 (0.05) –0.25 –0.14 0.60 (0.18)c –1.38a –0.73 0.64 (0.22)b –1.28 –0.59 0.03 (0.14) –0.69 –0.66 0.00 (0.20) –0.16 –0.16 0.04 (0.30) 0.68 0.73 0.18 0.19 –1.00 –0.80 0.15 (0.12) –0.46 –0.30 0.60 (0.08)c –0.96b –0.32 0.41 (0.17)a –1.99c –1.56b 0.66 (0.18)c –1.41a –0.71 –0.10 (0.11) –0.35 –0.46 –0.11 (0.30) 0.69 0.57 0.18 (0.12) –1.04a –0.86a 1.13 0.85 –1.37 –0.16 –0.19 (0.37) 0.37 0.16 0.88 (0.67) –1.27a –0.33 0.07 (0.02)b –0.22b –0.15a NOTES: Regressions of beds and services include percent county Medicaid, percent county Medicare, percent county HMO, Herfindahl index, ratio of uninsured Herfindahl index to total Herfindahl index, population density, CMS wage index, and dummies for years 1994 and 1996. Regressions of margin include percent county Medicaid, percent county Medicare, percent county HMO, percent hospitals in county for-profit, percent hospitals in county public, Herfindahl index, ratio of uninsured Herfindahl index to total Herfindahl index (the uninsured concentration), county population density, and dummies for years 1994, 1996, and 1999.

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance higher proportions of uninsured admissions and lower margins across the range of uninsured concentrations. DISCUSSION In this analysis of hospitals in rural counties, we find some evidence that in counties with higher proportions of uninsured patients and low concentration of these patients across hospitals, hospitals are less likely to offer psychiatric inpatient services. We also find some evidence that there are fewer psychiatric and ICU beds. There is a consistent association of higher uninsured admissions with a lower likelihood of hospitals offering high-technology services, although the association is statistically significant only for transplant, MRI, radiation therapy, and lithotripsy services. There is strong evidence that hospital margins are lower in counties with higher proportions of uninsured admissions. The findings on beds and services for vulnerable populations generally parallel those in our analysis of metropolitan areas, although in this analysis we find fewer statistically significant associations with bed supply and services for vulnerable populations. Specifically, in the metropolitan analysis, we find statistically significant associations with psychiatric outpatient services, alcohol and chemical dependence services, and beds that we do not observe here. Likewise, the findings on high technology services are similar, with most services in both samples displaying a negative association with uninsured admissions. However, for only a few services are the associations statistically significant and, with the exception of transplant services, the services for which the associations are statistically significant differ across the two samples. One feature shared by several of the services for which we observe statistically significant associations with uninsured admissions—inpatient psychiatric services and transplant—is that they have the potential to bring into the hospital uninsured patients whose care may be expensive. The presence of higher numbers of uninsured in a market may discourage hospitals from offering services for fear that they will not be reimbursed for a significant portion of the care they provide. For some of these services, a statistically significant association is only observed at low levels of concentration of the uninsured among the hospitals serving the county. The fact that we observe higher average rates of offering beds or services when care for the uninsured is more concentrated in counties may reflect strategic interactions among hospitals. This illustrates the variability among rural counties in their capacity to maintain hospital services and the impact of higher uninsured rates on hospitals in these counties. One limitation of this study is that it includes principally rural counties in the north and Pacific west. It should be replicated in a broader cross-section of rural hospitals, especially those in the south and southwest. In sharp contrast to the findings in metropolitan areas, for rural counties and hospitals we find a consistent association of lower margins to higher uninsured admissions. This may reflect rural hospital administrators’ preference to maintain

OCR for page 203
A Shared Destiny: Community Effects of Uninsurance services when faced with low (or even negative) margins. This strategy, while preserving a rural hospital’s ability to serve its community in the short-term, has implications for hospital’s ability to provide, maintain, and improve services in the future. Because of lower margins, rural hospitals in areas with high uninsured rates may have difficulty maintaining and replacing their physical plant, investing in new technologies, and expanding their scope of services to meet new community health needs. The lower margins may account in part for the lower likelihood of offering specific costly, high-technology services. Lower margins may also affect quality or staffing in ways that are not observable in this analysis.