This article was downloaded by: [The University of Manchester Library] On: 12 October 2014, At: 03:54 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Social Work in Public Health Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/whsp20
The Small Area Predictors of Ambulatory Care Sensitive Hospitalizations: A Comparison of Changes over Time a
b
Jayasree Basu , Lee R. Mobley & Vennela Thumula
c
a
Agency for Healthcare Research and Quality , Rockville , Maryland , USA b
RTI International, Research Triangle Park , North Carolina , USA
c
Department of Pharmacy Administration , University of Mississippi , Oxford , Mississippi , USA Published online: 09 Jan 2014.
To cite this article: Jayasree Basu , Lee R. Mobley & Vennela Thumula (2014) The Small Area Predictors of Ambulatory Care Sensitive Hospitalizations: A Comparison of Changes over Time, Social Work in Public Health, 29:2, 176-188, DOI: 10.1080/19371918.2013.776316 To link to this article: http://dx.doi.org/10.1080/19371918.2013.776316
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/termsand-conditions
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Social Work in Public Health, 29:176–188, 2014 Copyright © Taylor & Francis Group, LLC ISSN: 1937-1918 print/1937-190X online DOI: 10.1080/19371918.2013.776316
The Small Area Predictors of Ambulatory Care Sensitive Hospitalizations: A Comparison of Changes over Time Jayasree Basu Agency for Healthcare Research and Quality, Rockville, Maryland, USA
Lee R. Mobley RTI International, Research Triangle Park, North Carolina, USA
Vennela Thumula Department of Pharmacy Administration, University of Mississippi, Oxford, Mississippi, USA
The hospital admission for ambulatory care sensitive conditions (ACSCs) is a validated indicator of impeded access to good primary and preventive care services. The authors examine the predictors of ACSC admissions in small geographic areas in two cross-sections spanning an 11-year time interval (1995–2005). Using hospital discharge data from the Healthcare Cost and Utilization Project of the Agency for Healthcare Research and Quality for Arizona, California, Massachusetts, Maryland, New Jersey, and New York for the years 1995 and 2005, the study includes a multivariate cross-sectional design, using compositional factors describing the hospitalized populations and the contextual factors, all aggregated at the primary care service area level. The study uses ordinary least squares regressions with and without state fixed effects, adjusting for heteroscedasticity. Data is pooled over 2 years to assess the statistically significant changes in associations over time. ACSC admission rates were inversely related to the availability of local primary care physicians, and managed care was associated with declines in ACSC admissions for the elderly. Minorities, aged elderly, and percent under federal poverty level were found to be associated with higher ACSC rates. The comparative analysis for 2 years highlights significant declines in the association with ACSC rates of several factors including percent minorities and rurality. The two policy-driven factors, primary care physician capacity and Medicaremanaged care penetration, were not found significantly more effective over time. Using small area
This research is funded wholly by the authors’ employers, the Agency for Healthcare Research and Quality (AHRQ) and RTI International (Research Triangle Institute). The views expressed in this article are those of the authors. No official endorsement by any agency of the federal government is intended or should be inferred. The authors would like to acknowledge the state data organizations that participate in the HCUP State Inpatient Databases in Arizona, California, Oregon, Washington, Massachusetts, Maryland, New Jersey, and New York. The authors would also like to acknowledge the data and programming support provided by Social and Scientific Systems, Inc. in Maryland. Address correspondence to Jayasree Basu, PhD, Agency for Healthcare Research and Quality, 540 Gaither Road, Rockville, MD 20850, USA. E-mail: [email protected]
176
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
177
analysis, the study indicates that improvements in socioeconomic conditions and geographic access may have helped improve the quality of primary care received by the elderly over the last decade, particularly among some minority groups. Keywords: Preventable hospitalizations, small area analysis, ambulatory care sensitive conditions, quality of primary care, changes over time
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
BACKGROUND The hospital admission for ambulatory care sensitive conditions (ACSCs) is a validated indicator of impeded access to good primary and preventive care services (Millman, 1993). Several prior studies have explored the quality of primary care by using ambulatory care sensitive (or preventable) hospitalization rates. A few of them looked into elderly hospitalization patterns. Very few studies (e.g., Kozak et al., 2001; McCall et al., 2004) examined the changes in elderly hospitalization patterns over a longer time period for preventable conditions. Such a study provides an overview of how access to quality primary care has changed in the nation. The issue is particularly important for the elderly as there were several federal legislative initiatives during 1995 to 2005 that could have had direct and indirect impacts on the access to care for the elderly. We examine the predictors of ACSC admissions in small geographic areas in two cross-sections spanning an 11-year time interval (1995–2005). Previous studies examining the elderly trends have focused on earlier periods. Kozak et al. (2001) looked at the trends during 1980 to 1995 whereas McCall et al. (2004) used a more recent period (1992–2000). Both studies reported increasing rates of hospitalization for the elderly in general, and for specific ACSC conditions. The period we chose is a more recent one spanning the implementation of the Balanced Budget Act in 1997 (which resulted in substantial Medicare-managed care plan disenrollment), followed by a major push to enroll more elderly in Medicare managed care plans through implementation of the Medicare Modernization Act (MMA) of 2003. The supply of primary care services may have increased over this period due to several funding initiatives implemented by the federal government to increase access in underserved areas (Grumbach et al., 1997). The socioeconomic climate of the small areas may also have changed over the period. This study examines the net result of these policy initiatives and market changes on the ACSC hospitalization rates among the elderly over time. Using a rigorous conceptual framework, the major aim of this article is to identify the factors that predicted ACSC hospitalization rates in two cross sections of time 10 years apart, thereby assessing how or whether the influence of these explanatory variables on ACSC hospitalization rates changed with time. This research is important because it informs policy makers about the small-area variation in quality of primary care for the elderly; and whether access to primary care and preventive care services improved overall, for specific minority groups, or for Medicare-managed care enrollees relative to Medicare Fee-for-Service (FFS) enrollees over time. The association of adverse changes in the spatial patterns of ACSC admission rates over time with various sociodemographic and contextual factors can be used to identify small areas where policy interventions might be most fruitfully targeted to improve the current system.
CONCEPTUAL HYPOTHESIS There is a large body of literature about the geographic distribution of health services and patient access to care (e.g., Chandra & Skinner, 2003; Coughlin et al., 2002; Meade et al., 1988; Ricketts et al., 1994; Virnig et al., 2007). This study builds on the past research concerned with ACSC or preventable hospitalizations (e.g., Friedman & Basu, 2001; Krakauer et al., 1996; Mobley et al.,
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
178
J. BASU ET AL.
2006; Parchman & Culler, 1994). These studies examined variation in local hospitalization rates for ACSC conditions, chiefly in cities or hospital service areas. Studies also compared rates of hospitalization across catchment areas such as Health Service Areas (HSA) that were developed statistically as clusters of counties for patients going to the hospital (Makuc et al., 1991). This study focuses on primary care market areas (to be described below) to study the predictors of ACSC hospitalization rates. The conceptual models for ACSC hospitalizations usually examine supply and demand for outpatient services to explain the variations of admission rates (Basu & Friedman, 2001; Friedman & Basu, 2001). Consumers (patients and families) are one group of actors who seek, use, and pay for services. Primary care suppliers (physicians or physician directed), managed care plans, and hospital managers are independent groups of actors supplying services and sometimes making decisions on behalf of patients. ACSC hospitalization occurs when (a) demand for primary care exceeds its supply; or (b) it is “rational” not to use primary care because hospital care will typically be available and better paid for by the insurance; (c) economic, cultural, and social barriers prevent utilization of primary care; or perhaps (d) primary care physicians (PCPs) are unsuccessful in reducing ACSC admissions. Some of the factors affecting the demand for outpatient care, and hence (inversely) the hospitalization for ACSC conditions, include poverty, education level, and public and private insurance (Billings et al., 1993; Weissman et al., 1992). A number of these determinants could vary among racial and ethnic groups. A growing body of elderly literature has found that Medicare beneficiaries in managed care receive more preventive services and have better outcomes than their FFS counterparts (Riley et al., 1994; Riley, Potosky, Klabunde, Warren, & Ballard-Barbash, 1999; Rizzo, 2005). Although the role of managed care market penetration and the spillover effects on other providers could be mixed, previous research found Medicare managed care enrollment to be associated with less use of hospitals for ACSC conditions (Basu & Mobley, 2007). The supply factors associated with ACSC admissions are inpatient bed capacity (Goodman, Gittelsohn, Chang, & Fleming, 1994; Krakauer et al., 1996), supplies of PCPs, and physician practice patterns (Homer et al., 1996; Parchman & Culler, 1994). A greater availability of PCPs, all else equal, would tend to make primary care more accessible and reduce average prices for primary care for potential patients who could be treated in an ambulatory setting, relative to hospital inpatient stays (Basu et al., 2002). Higher local hospital inpatient capacity would, on the other hand, lower the average cost of inpatient versus primary care and could increase ACSC admissions (Folland et al., 1997). Area characteristics such as elderly in poverty or in rural location are expected to exert important influences on increased rates of ACSC or preventable admissions (Mobley et al., 2006). Studies have found degree of remoteness and rural/urban residence to be positively associated (Ansari et al., 2006) and population density to be negatively associated with ACSC admissions (e.g., Schreiber & Zielinski, 1997). Several of these factors could have played changing roles in influencing ACSC hospitalizations during the period we studied. For example, the Medicare Modernization Act of 2003 may have increased preventive care services availability through increased participation by Medicare managed care plans with prescription drug coverage, which could have enhanced the quality of primary care offered by managed care plans.1 On the other hand, the growth of Private Fee-for-Service (PFFS) managed care plans after 2003 may have resulted in lesser care coordination and provision of preventive care delivery than what health maintenance organization (HMO) plans provided. The improved supply and allocation of PCPs into underserved (Grumbach et al., 1997) areas could have resulted in lower marginal benefit from adding PCPs over time. Several socioeconomic changes, such as increased average income and declining poverty, could have also boosted accessibility of primary care over time. Along with increasing mix of aged elderly and minority population affecting overall ACSC rates, increased socioeconomic conditions could also improve primary care access in some minority groups. In the multivariate analysis that follows, we examine the relative contributions of these factors and how and whether these contributions changed over time.
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
179
METHOD
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Scope The study used hospital discharge data for six states on the East and West Coast: Arizona (AZ), California (CA), Massachusetts (MA), Maryland (MD), New Jersey (NJ), and New York (NY) for the years 1995 and 2005. We first selected all elderly (age 65 and older) patients who were hospitalized for any conditions in these six states. We then selected 20 ACSC conditions based on criteria developed by Billings et al. (1993) and selected all elderly patients who were hospitalized with these conditions in the above states, to use in our analysis of ACSC hospitalization rates over time. The study used descriptive and multivariate analysis (to be described in detail below) to examine the predictors of small area ACSC rates in two cross-sections to assess any apparent changes over time. We classified patients by their area of residence, and used primary care service area (PCSA) as the natural market area in which patients receive preventive care (more below).
Design The unit of analysis is PCSA, which is a natural market defined by FFS Medicare patient flows to physician offices (Goodman et al., 2003). PCSAs were developed using Medicare utilization data to represent geographic approximations of markets for primary care services received by the elderly. We assume that these areas are the best approximation of the small market area in which elderly patients receive their ambulatory care, and these regions have been validated in previous research on elderly access to preventive care services (Mobley et al., 2006; Mobley et al., 2008). PCSAs are generally smaller and more numerous than counties. Multivariate analysis. We constructed two multivariate models, one for each year, to understand the relative importance of predictors of ACSC admission rates for each year. To gauge the changes in relative contributions of explanatory variables between 1995 and 2005, we also used pooled regression models that combined the data for these 2 years and computed coefficients of time interactions. The model was run using with and without state fixed effects. The rationale for fixed effects model lies in the fact that it controls for across the state variations due to other factors and thus partially adjusts for omitted variable bias in the estimation process (Greene, 1993). We used statistical significance tests to identify the predictors that significantly changed its effects (coefficients) over time. We focused on all ACSCs combined for the multivariate models and selected all elderly patients hospitalized with any ACSC in the six sample states. The outcome variable was ACSC hospitalization rates by PCSAs, derived as the number of ACSC admissions (i.e., those admitted with ACSCs) of elderly (65 and older) residents of each PCSA, divided by the number of all hospitalizations of elderly patients (65 and older) living in that PCSA. As control variables, we used compositional factors describing hospitalized patients as well as contextual factors describing socioeconomic characteristics, all aggregated to the PCSA level. The compositional factors represent characteristics of the entire hospitalized population, not the subset admitted for ACSCs. The independent variables were suggested by past literature (as mentioned earlier in the conceptual hypotheses section) and winnowed down to the final set after considering multicollinearity as well as predictive value. The variables include sociodemographic composition, composition of insurance coverage of all hospitalized patients, average disease severity of all hospitalized patients in the area, and health services resource availability (including inpatient hospital capacity and primary care physician supply). Table A1 in the appendix shows a complete list of the independent variables and sources of data used in the regression models.
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
180
J. BASU ET AL.
FIGURE 1 Sources of data and analytical file construction. HCUP SID D Healthcare Cost and Utilization Project; RTI D Research Triangle Institute; PCSA D primary care service area; ARF D area resource files.
We used the ordinary least squares (OLS) method to estimate multivariate regression models. The initial tests showed that ACSC hospitalization rates closely followed a normally distributed (bell-shaped) curve. We conducted additional tests for heteroscedasticity, and corrected standard errors in our models where appropriate. The sources of the data and the data file construction followed the steps outlined in Figure 1.
FINDINGS Multivariate Analysis Table 1 presents the summary statistics, and Table 2 shows the corresponding results from the multivariate models, estimated separately for each year. Table 3 reports selected variables from the interaction model without state fixed effects, and Table 4 reports the same with state fixed effects. Sample statistics. Table 1 reports the mean (unweighted) values of all the variables used in the regression analysis for 1995 and 2005. The table also reports the changes over 1995 to 2005 that were statistically significant, based on results of two sample t tests conducted for each variable. Table 1 indicates that ACSC admission rates for the elderly declined significantly as a whole in the six states. A few major significant changes occurred during the 11-year period in the demographic, contextual, as well as in the policy environment. First, the average number
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
181
TABLE 1 Mean Values for Regression Variables, 1995 and 2005 Total Six States
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Variable ACSC admissions per 1,000 discharges by elderly residents Proportion of elderly age 65–74 Proportion of elderly age 75–84 Proportion of elderly age 85 and older Proportion of elderly: Medicare-managed care plans Proportion of African American discharges Proportion of Hispanic discharges Proportion of other race discharges Average income Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density PCP capacity per 1000 population Proportion of PCSA that is rural Proportion of elderly below the FPL in the PCSA
1995 N D 990
2005 N D 1,010
0.18 0.43 0.39 0.17 0.01 0.04 0.05 0.06 38,579.17 880.28 21.80 3,775.06 0.51 0.34 0.097
0.17* 0.37* 0.41* 0.21* 0.12* 0.05 0.08* 0.06 52,514.8* 3,518.87* 21.61 3,407.99 0.72* 0.29* 0.086*
Note. Mean values reported in this table are unweighted. N refers to the number of PCSAs. Regression analysis and mean values are estimated with data aggregated at the PCSA level. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; PCP D primary care physician; FPL D federal poverty level. *Two-tailed t test indicates statistically significant differences between 1995 and 2005 .p < 0:05/. TABLE 2 Ordinary Least Squares Regression Results, 1995 and 2005: Hospitalization for ACSC Conditions per 1,000 Discharges by Elderly Residents in PCSA of Patient Origin 1995 All States (N D 990) Variable Proportion of elderly age 65–74 Proportion of elderly age 75–84 Proportion of elderly covered by Medicare HMOs Proportion of African American discharges Proportion of Hispanic discharges Proportion of other race discharges Average income (in $1,000s) Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density Primary care physician capacity per 1,000 population Proportion of PCSA that is rural Proportion of elderly below the FPL in the PCSA Intercept R2
ˇ 0.160 0.188 0.068 0.083 0.054 0.025 1.83E-07 1.16E-06 0.00013 2.24E-07 0.036 0.019 0.080 0.334 0.2304
p Value 0.000 0.000 0.025 0.000 0.000 0.126 0.233 0.122 0.220 0.096 0.000 0.000 0.104 0.000
2005 All States (N D 1,010) ˇ 0.128 0.165 0.044 0.052 0.036 0.018 3.43E-07 3.10E-08 0.00025 3.08E-07 0.007 0.001 0.174 0.301 0.3044
p Value 0.000 0.000 0.000 0.000 0.000 0.102 0.000 0.419 0.000 0.001 0.003 0.796 0.000 0.000
Note. Standard errors of the regression estimates are heteroscedasticity consistent. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; HMO D health maintenance organization; FPL D federal poverty level.
182
J. BASU ET AL.
TABLE 3 Ordinary Least Squares Regression Results, 1995 and 2005 Pooled Data: Hospitalization for ACSC Conditions per 1,000 Discharges by Elderly Residents in PCSA, Selected Variables All States (N D 2,000)
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Variable Proportion of elderly age 65–74 * Year Proportion of elderly age 75–84 * Year Proportion of elderly covered by Medicare HMOs * Year Proportion of African American discharges * Year Proportion of Hispanic discharges * Year Proportion of other race discharges * Year Average income (in $1,000s) * Year Hospital inpatient capacity per 1000 population * Year Average distance from patient to hospital ZIP codes * Year Population density * Year Primary care physician capacity per 1000 population * Year Proportion of PCSA that is rural * Year Proportion of elderly below the FPL in the PCSA * Year
ˇ 0.032 0.023 0.024 0.032 0.018 0.045 0.000 0.000 0.000 0.000 0.029 0.020 0.094
p Value 0.480 0.736 0.451 0.090 0.235 0.029 0.349 0.133 0.333 0.001 0.000 0.001 0.094
Note. Standard errors of the regression estimates are heteroscedasticity consistent. The R 2 is 0.2748. Only interaction (with time) variables are reported (1995 D 0, 2005 D 1) in this Table. The full results are available upon request. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; HMO D health maintenance organization; FPL D federal poverty level.
of primary care physicians per 1,000 population increased by 34% from 0.51 to 0.71 in all six states between 1995 and 2005. Also, significant increase in the Medicare-managed care enrollment occurred over the 11-year period. The proportion of the elderly Medicare beneficiaries who were enrolled in managed-care plans increased from 1% in 1995 to 12% overall in these six states. The proportions of elderly under the federal poverty level (FPL) declined. The sociodemographic composition also changed, with greater increase in hospitalizations of minority and aged elderly. Among other changes, average household income increased, and population density declined. In addition, average rurality declined and average inpatient hospital capacity per 1,000 population increased. Multivariate Models Table 2 shows the OLS regression results for 1995 and 2005 using the same set of variables for comparison. Table 2 indicates that younger elderly age groups (65–74 and 75–84) had a negative association with ACSC hospitalization rates (relative to those in 85 and older age group), experiencing also a relative decline in impact over time. The two racial and ethnic groups, African Americans and Hispanics, were associated with increased ACSC hospitalizations, whereas both experiencing relative declines in effects over time. The proportions of enrollment under Medicare managed care plans was a significant predictor in 1995 and 2005, but its marginal contribution toward declines in ACSC rates dropped in 2005 (ˇ D 0:045, p < :05) relative to 1995 (ˇ D 0:068, p < 0:01). Likewise, the PCP density was a significant predictor in 1995 and 2005, being associated with reductions in ACSC rates. However, there was a substantial decline in marginal contributions (from 0.036 to 0.008, p < 0:01) between 1995 and 2005. The area socioeconomic indicators, measured by average area income and proportions below poverty, were found to have
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
TABLE 4 Ordinary Least Squares Regression Results, State Fixed Effects Pooled Regression Models (1995 and 2005): Hospitalization for ACSC Conditions per 1,000 Discharges by Elderly Residents in PCSA
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
All States (N D 2,000) Variable CA (NY D 0) AZ (NY D 0) NJ (NY D 0) MA (NY D 0) MD (NY D 0) Proportion of elderly age 65–74 Proportion of elderly age 75–84 Proportion of elderly covered by Medicare HMOs Proportion of African American discharges Proportion of Hispanic discharges Proportion of other race discharges Average income (in $1,000s) Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density Primary care physician capacity per 1000 population Proportion of PCSA that is rural Proportion of elderly below the FPL in the PCSA Year CA * Year AZ * Year NJ * Year MA * Year MD * Year Proportion of elderly age 65–74 * Year Proportion of elderly age 75–84 * Year Proportion of elderly covered by Medicare HMOs * Year Proportion of African American discharges * Year Proportion of Hispanic discharges * Year Proportion of other race discharges * Year Average income (in $1,000s) * Year Hospital inpatient capacity per 1,000 population * Year Average distance from patient to hospital ZIP codes * Year Population density * Year Primary care physician capacity per 1000 population * Year Proportion of PCSA that is rural * Year Proportion of elderly below the FPL in the PCSA * Year Intercept R2
ˇ 0.037 0.035 0.011 0.005 0.001 0.146 0.158 0.059 0.075 0.115 0.012 9.19E-07 8.28E-08 0.000029 2.17E-07 0.007 0.011 0.047 0.044 0.024 0.003 0.005 0.007 0.004 0.041 0.011 0.044 0.034 0.063 0.044 3.60E-07 5.84E-10 0.00002 2.58E-07 0.002 0.013 0.115 0.347 0.3676
p Value 0.000 0.000 0.000 0.285 0.804 0.000 0.001 0.110 0.000 0.000 0.475 0.000 0.910 0.612 0.089 0.314 0.015 0.235 0.299 0.000 0.748 0.236 0.196 0.574 0.355 0.868 0.239 0.028 0.001 0.029 0.040 0.999 0.787 0.109 0.809 0.027 0.017 0.000
Note. Standard errors of the regression estimates are heteroscedasticity consistent. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; NY D New York; CA D California; AZ D Arizona; NJ D New Jersey; MA D Massachusetts; MD D Maryland; HMO D health maintenance organization; FPL D federal poverty level.
183
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
184
J. BASU ET AL.
played important role in ACSC hospitalizations over time. The proportion of elderly under poverty exerted a stronger influence over time, being positively associated with ACSC hospitalization rates both years. Likewise, higher average area income levels played a significant role in pulling ACSC rates down and this effect became stronger over time. Among other findings, the rural urban continuum (measured by population density and proportions in rural area) changed the signs of coefficients between 1995 and 2005 to have caused declines in ACSC admission rates. The 1995 and 2005 results were further analyzed to examine the factors with significant changes in relative contributions over time (Table 3). The chow test was conducted on 1995 to 2005 pooled data to assess whether there were structural changes in the two elderly regression models over time (Greene, 1993). The chow test result (F D 5:73, prob < 0.001) indicated that, overall, the effects of the different parameters, measured by regression coefficients, changed over time. To assess the changes in relative contributions of individual factors, we used pooled regression models to examine the interaction coefficients of the independent variables with time. The partial results from this model are presented in Table 3. The results of this analysis in Table 3 highlight several interesting findings. The “other races” showed significantly higher association. The marginal contribution of rural population on ACSC admissions declined significantly over time, as was the contribution of the PCP density. On the other hand, population density became a stronger predictor of ACSC rates. None of the other sociodemographic or policy-driven factors was found to be significantly more or less predictive of ACSC admission rates over time. Somewhat different findings were obtained by using a state fixed effects model (Table 4) that shows the multivariate results for pooled 1995 and 2005 data with state fixed effects. In this model, we use state dummies, year and state interactions, and interactions of each explanatory variable with year. The model differs from the one presented before (Table 3) in terms of better controlling for unobserved differences across states. We present the full model with interaction coefficients to isolate the changes in marginal contributions of the explanatory variables over time (please note that separate estimates for 1995 and 2005 can be obtained from this table). The interaction coefficients indicate that several variables had statistically significant changes in marginal contributions over time to the ACSC rates, namely, race, income, and rurality. African Americans and Hispanics had smaller marginal contributions to ACSC rate over time, whereas the contributions of other races to ACSC rate increased significantly. In contrast to previous models (Tables 2 and 3), the negative association of income with ACSC rate was found significantly weaker over time, whereas stronger association of ACSC rate with poverty rate was observed. The marginal contributions of rurality declined significantly over time.
DISCUSSION Using small area analysis, the study identified several predictors from multivariate models for individual years which are consistent with previous research and as hypothesized. For example, the study showed that ACSC admission rates were inversely related to the availability of local PCPs in both years. This is consistent with previous research using larger area level units (HSAs or counties) that reported negative associations between the two variables (Parchman & Culler, 1994), particularly for adults (Friedman & Basu, 2001), although some studies also found weak to no association between primary care supply and ACSC admissions (Grumbach et al., 1997). We also found that managed care was associated with a decline in ACSC admissions for the elderly in both years, which was also consistent with previous research (Basu & Mobley, 2007). We found minorities, aged elderly, and poverty to be associated with higher ACSC rates, all of which are consistent with previous research findings related to socioeconomic status and ACSC
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
185
hospitalizations (e.g., Begley, Slater, Engel, & Reynolds, 1994; Billings et al., 1993; Blustein et al., 1998). ACSC admission rates for the elderly declined significantly as a whole in the six states. By conducting two cross-sectional studies as well as combined analyses for 2 years, we were able to assess how the effects of the explanatory variables on ACSC hospitalization rates changed over time. These findings, along with data presented in Table 1, would suggest some potential forces behind the observed reduction in ACSC admissions over time. Our descriptive statistics and multivariate results indicate that the period 1995 to 2005 was characterized by several changes in the socioeconomic environment. Our findings suggest that, although the percentage of hospitalizations increased among the minority elderly population between 1995 and 2005, this did not appear to have significantly driven the ACSC hospitalization rates upward. This is because the association between minorities and ACSC admission rates did not commensurately increase. The association between the proportions of Hispanics and African Americans with ACSC rate actually dropped between 1995 and 2005. Although the proportions of Hispanics among hospitalized elderly population increased from 4.5% to 6.8%, the noticeably smaller effect of Hispanics may indicate improved access to primary care among this subgroup over time. Although “other races” showed significantly higher marginal contributions over time, the relative size of this minority group did not increase to have a sizable impact on ACSC rates. Overall, our findings indicate that the racial ethnic minorities commonly associated with higher ACSC rates did not play a significant role in driving ACSC rates up in our study states. On the other hand, older age groups had a stronger influence on increasing ACSC rates over time. As the younger elderly group (65–74) became smaller over time, and their marginal contributions declined, those in the age 85 and older group as well as their contribution to ACSC rates were likely to become stronger. These older age groups are also those likely to have higher rates of preventable hospitalizations. The two socioeconomic factors, the poverty rate among the elderly, and average per capita household income in PCSAs, influenced the changes in ACSC rates in different ways. The average area income levels increased over time, although it was found to be a weaker predictor of ACSC rates in 2005 once unobserved variances across the states was controlled (Table 4). On the other hand, poverty levels became a stronger predictor of ACSC rates; though poverty rates in general declined. Because there was a decline over time in the proportions of elderly under the FPL, the growth of ACSC admissions attributable to poverty was somewhat moderated. The geographic location was found to be an important factor explaining reductions in ACSC rates. The proportion of population in rural areas significantly declined, as was the marginal contribution of rural population on ACSC admissions, indicating better geographic access over time. Because rurality was associated with higher ACSC rates, the declining size and influence of rurality jointly could have led to an overall decline in ACSC rates. The two factors, PCP capacity and Medicare managed care penetration rates, were of primary interest to us because they expanded in size and were likely to be related to lower ACSC hospitalization. The mean levels of both increased in size substantially between 1995 and 2005, whereas the relative contributions did not improve. Although experiencing a major spurt of growth, the role of increased penetration of managed care in the Medicare market was not clear. Although associated with reductions in ACSC rates in both years, its relative marginal contribution fell somewhat over time. A likely reason could be the growth in PFFS plans after 2003, with less emphasis than HMO models on care coordination and prevention, which might have made these plans less willing and efficient in providing primary care. On the other hand, the Medicare HMO plans in 1995 were probably more organized and effective in delivering primary care. The decline in the coefficients in primary care density between 1995 and 2005 could be explained by increased supply and distribution of primary care physicians over the last decade and a potential decline in additional benefits from further increase. Another factor could be the increasing role
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
186
J. BASU ET AL.
of nonphysician personnel, international medical graduates, as well as community health centers in primary care delivery, which was not captured in the primary care variable we constructed. Several limitations of the study should be acknowledged. The six states examined are not fully representative of the entire U.S. population and market conditions. They were chosen based on data availability appropriate for multivariate analysis. Another limitation could be that the study examines only one outcome indicator for the quality of primary care. Although our quality indicator is quite important regarding total social cost of care and a signal of effective management of chronic illness, other measures could be examined. This is an important area of ongoing research. In terms of methods, data limitations did not allow us to adjust well for severity of elderly population morbidity in the comparative analysis for 1995 to 2005, which may cause omitted-variables bias in the impact of managed care penetration, to the extent that managed care plans adversely or beneficially select patients by illness severity. However, we used other area variables as proxies for severity, such as poverty level, age composition, racial composition, average income, as well as average distance from patient’s ZIP-code to hospital ZIP-code (Basu & Friedman, 2007).
CONCLUSION The study uses small area analysis to examine the predictors of ACSC hospitalization rates and their potential roles over 1995 to 2005. The period 1995 to 2005 is particularly significant in terms of changes in elderly hospitalization patterns as well as in socioeconomic environments. The study indicates that improvements in socioeconomic conditions and geographic access may have helped improve the quality of primary care received by the elderly over the last decade (and reduced ACSC hospitalization rates). This change was observed among the minorities, particularly among Hispanics and African Americans. Several policy-driven factors, for example, Medicare managed care and PCP density, were not found significantly more effective over time.
NOTES 1. In 2003, a pilot project was implemented which offered Preferred Provider Organizations (PPOs) with drug coverage to the elderly in 21 states, see Pope et al. (2005). 2. There are some gaps in the ZIP code tabulation area (ZCTA) coverage in very rural areas where no ZIP codes existed in 1990 or 2000, and there are no accompanying census data for these areas. These gaps correspond to the addresses for rural patients with no ZIP code (or giving a PO box that has no latitude or longitude). We did not encounter any problems, because all patients kept in the data set had residential ZIP codes. 3. The census data is nested by ZCTA inside the PCSAs, but the county data may span several PCSAs and cross their borders.
REFERENCES Ansari, Z., Laditka, J. N., & Laditka, S. B. (2006). Access to health care and hospitalization for ambulatory care sensitive conditions. Medical Care Research and Review, 63(6), 719–741. Begley, C. E., Slater, C. H., Engel, M. J., & Reynolds, T. F. (1994). Avoidable hospitalizations and socio-economic status in Galveston County, Texas. Journal of Community Health, 19(5), 377–387. Basu, J., & Friedman, B. (2001). Preventable illness and out-of-area travel of children in New York counties. Health Economics, 10, 67–68. Basu, J., Friedman, B., & Burstin, H. (2002). Primary care, HMO enrollment, and hospitalization for ambulatory care sensitive conditions. Medical Care, 40(12), 1260–1269.
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
187
Basu, J., & Mobley, L. (2007). Do HMOs reduce preventable hospitalizations for Medicare beneficiaries? Medical Care Research and Review, 64(5), 544–567. Basu, J., & Friedman, B. (2007). A reexamination of distance as a proxy for severity of illness and implications for differences in utilization by race/ethnicity. Health Economics, 16, 687–701. Billings, J., Zeitel, L., Lukomnik, J., Carey, T. S., Blank, A. E., & Newman, L. (1993). Impact of socioeconomic status on hospital use in New York City. Health Affairs, 12, 162–173. Blustein, J., Hanson, K., & Shea, S. (1998). Preventable hospitalizations and socioeconomic status. Health Affairs, 17(2), 177–189. Chandra, A., & Skinner, J. (2003). Geography and racial health disparities (NBER Working Paper No. W9513). Cambridge, MA: National Bureau of Economic Research. Retrieved from http://www.uber.org/papers/w9513 Coughlin, S., Thompson, T., Seeff, L., Richards, T., & Stallings, F. (2002). Breast, cervical, and colorectal carcinoma screening in a demographically defined region of the southern U.S. Cancer, 95, 2211–2222. Folland, S. A., Goodman, A. C., & Stano, M. (1997). The economics of health care (2nd ed.). Upper Saddle River, NJ: Prentice Hall. Friedman, B., & Basu, J. (2001). Health insurance, primary care, and preventable hospitalization of children in a large state. American Journal of Managed Care, 7(5), 473–481. Goodman, D. C., Fisher, E. S., Gittelsohn, A., Chang, C. H., & Fleming, C. (1994). Why are children hospitalized? The role of non-clinical factors in pediatric hospitalizations. Pediatrics, 93(6, Pt 1), 896–902. Goodman, D. C., Mick, S. S., Bott, D., Stukel, T., Chang, C., Marth, N., : : : Carretta, H. J. (2003). Primary care service areas: A new tool for the evaluation of primary care services. Health Services Research, 38(1), 287–310. Green, W. H. (1993). Econometric analysis (2nd ed.). New York: MacMillan. Grumbach, K., Vranizan, K., & Bindman, A. B. (1997). Physician supply and access to care in urban communities. Health Affairs, 16(1), 71–85. Homer, C. J., Szilagyi, P., Rodewald, L., Bloom, S. R., Greenspan, P., Yazdgerdi, S., : : : Perrin, J. M. (1996). Does quality of care affect rates of hospitalization for childhood asthma? Pediatrics, 98(1), 18–23. Kozak, L., Hall, M. J., & Owings, M. F. (2001). Trends in avoidable hospitalizations, 1980–1998. Health Affairs, 20(2), 225–232. Krakauer, H., Jacoby, I., Millman, M., & Lukomnik, J. E. (1996). Physician impact on hospital admission and on mortality rates in the Medicare population. Health Services Research, 31(2), 191–210. Makuc, D., Hugland, B., Ingram, D. D., Kleinman, J. C., & Feldman, J. J. (1991). Health service areas for the United States. Vital and Health Statistics, 2, Nov.(112), 1–102. McCall, N., Brody, E., Mobley, L., & Subramanian, S. (2004, April). Investigation of increasing rates of hospitalization for ambulatory care sensitive conditions (avoidable hospitalizations) among Medicare beneficiaries. Final report. Retrieved from http://www.cms.hhs.gov/Reports/Downloads/McCall_2004_3.pdf Meade, M. S., Florin, J. W., & Gesler, W. M. (1988). Medical geography. New York, NY: Guilford Press. Millman, M. (Ed.). (1993). Access to health care in America. Washington, DC: National Academy Press. Mobley, L., Root, E., Anselin, L., Gracia, N., & Koschinsky, J. (2006, May 15). Spatial analysis of elderly access to primary care services. International Journal of Health Geographics, 5(1), 19. Retrieved from http://www.ij-healthgeographics. com/content/5/1/19 Mobley, L. R., Kuo, T., & Andrews, L. S. (2008). How sensitive are multilevel regression findings to defined area of context? A case study of mammography use in California. Medical Care Research and Review, 65(3), 315–337. Parchman, M. L., & Culler, S. (1994). Primary care physicians and avoidable hospitalization. Journal of Family Practice, 39(2), 123–128. Pope, G., Olmsted, E. A., Kautter, J., Mobley, L. R., & Greenwald, L. M. (2005, April). The Medicare preferred provider organization demonstration: Plan offerings and enrollment. Retrieved from http://www.cms.hhs.gov/HealthCareFinancing Review/downloads/06Springpg95.pdf Ricketts, T. C., Savitz, L. A., Gesler, W. M., & Osborne, D. N. (1994). Geographic methods for health services research. Lanham, MD: University Press of America. Riley, G. F., Potosky, A. L., Lubitz, J. D., & Brown, M. L. (1994). Stage of cancer at diagnosis for Medicare HMO and fee-for-service enrollees. American Journal Public Health, 84(10), 1598–1604. Rizzo, J. (2005). Are HMOs bad for health maintenance? Health Economics, 14(11), 1117–1131. Schreiber, S., & Zielinsky, T. (1997). The meaning of ambulatory care sensitive admissions: Urban and rural perspectives. Journal of Rural Health, 13(4), 276–284. Virnig, B., Scholle, S., Chou, A., & Shih, S. (2007). Efforts to reduce racial disparities in Medicare managed care must consider the disproportionate effects of geography. American Journal of Managed Care, 13, 51–56. Weissman, J. S., Gatsonis, C., & Epstein, A. M. (1992). Rates of avoidable hospitalization by insurance status in Massachusetts and Maryland. Journal of the American Medical Association, 268(17), 2388–2394.
188
J. BASU ET AL.
APPENDIX Table A1. Independent Variables and Data Sources Variables
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Proportion of elderly age 65 to 74
Proportion of elderly age 75 to 84
Proportion of elderly age 85 and older Proportion of elderly: Medicare FFS plans
Proportion of elderly: Medicare-managed care plans
Proportion of White discharges
Proportion of African American discharges Proportion of Hispanic discharges
Proportion of other race discharges
Average income Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density PCP capacity per 1000 population
Proportion of elderly below the federal poverty level in the PCSA Proportion of rural
Description Number of elderly residents age 65–74 years divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents age 75–84 years divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents age 85 and older divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents with Medicare FFS insurance divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents with Medicare-managed care insurance divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of White elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of African American elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of Hispanics elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of other races elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Average household income of PCSA Total inpatient days in hospitals in the PCSA per 1,000 population of the PCSA Average distance in miles between patient’s ZIP code and hospital ZIP code in each PCSA Thousands of total population per square mile land areaa Family practitioners, internists, and specialists expressed as counts per 1,000 U.S. populationa;b Proportion of elderly below federal poverty levela Proportion of population living in rural census tractsa
Source HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID American Hospital Association HCUP-SID U.S. Census, HRSA AMA/AOA Masterfiles, ARF and HRSA U.S. Census, HRSA US Census, HRSA
HCUP-SID D Healthcare Cost and Utilization Project; PCSA D primary care service area; FFS D Fee for Service; HRSA D Health Resources and Services Administration; AMA/AOA D American Medical Association/American Osteopathic Association; PCP D primary care physician; ARF D Area Resource Files. a For HRSA data and more explanations of the variables, see http://datawarehouse.hrsa.gov/. b 1995 data are from Area Resource Files (ARF), county level, interpolated to represent PCSA.
Social Work in Public Health Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/whsp20
The Small Area Predictors of Ambulatory Care Sensitive Hospitalizations: A Comparison of Changes over Time a
b
Jayasree Basu , Lee R. Mobley & Vennela Thumula
c
a
Agency for Healthcare Research and Quality , Rockville , Maryland , USA b
RTI International, Research Triangle Park , North Carolina , USA
c
Department of Pharmacy Administration , University of Mississippi , Oxford , Mississippi , USA Published online: 09 Jan 2014.
To cite this article: Jayasree Basu , Lee R. Mobley & Vennela Thumula (2014) The Small Area Predictors of Ambulatory Care Sensitive Hospitalizations: A Comparison of Changes over Time, Social Work in Public Health, 29:2, 176-188, DOI: 10.1080/19371918.2013.776316 To link to this article: http://dx.doi.org/10.1080/19371918.2013.776316
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/termsand-conditions
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Social Work in Public Health, 29:176–188, 2014 Copyright © Taylor & Francis Group, LLC ISSN: 1937-1918 print/1937-190X online DOI: 10.1080/19371918.2013.776316
The Small Area Predictors of Ambulatory Care Sensitive Hospitalizations: A Comparison of Changes over Time Jayasree Basu Agency for Healthcare Research and Quality, Rockville, Maryland, USA
Lee R. Mobley RTI International, Research Triangle Park, North Carolina, USA
Vennela Thumula Department of Pharmacy Administration, University of Mississippi, Oxford, Mississippi, USA
The hospital admission for ambulatory care sensitive conditions (ACSCs) is a validated indicator of impeded access to good primary and preventive care services. The authors examine the predictors of ACSC admissions in small geographic areas in two cross-sections spanning an 11-year time interval (1995–2005). Using hospital discharge data from the Healthcare Cost and Utilization Project of the Agency for Healthcare Research and Quality for Arizona, California, Massachusetts, Maryland, New Jersey, and New York for the years 1995 and 2005, the study includes a multivariate cross-sectional design, using compositional factors describing the hospitalized populations and the contextual factors, all aggregated at the primary care service area level. The study uses ordinary least squares regressions with and without state fixed effects, adjusting for heteroscedasticity. Data is pooled over 2 years to assess the statistically significant changes in associations over time. ACSC admission rates were inversely related to the availability of local primary care physicians, and managed care was associated with declines in ACSC admissions for the elderly. Minorities, aged elderly, and percent under federal poverty level were found to be associated with higher ACSC rates. The comparative analysis for 2 years highlights significant declines in the association with ACSC rates of several factors including percent minorities and rurality. The two policy-driven factors, primary care physician capacity and Medicaremanaged care penetration, were not found significantly more effective over time. Using small area
This research is funded wholly by the authors’ employers, the Agency for Healthcare Research and Quality (AHRQ) and RTI International (Research Triangle Institute). The views expressed in this article are those of the authors. No official endorsement by any agency of the federal government is intended or should be inferred. The authors would like to acknowledge the state data organizations that participate in the HCUP State Inpatient Databases in Arizona, California, Oregon, Washington, Massachusetts, Maryland, New Jersey, and New York. The authors would also like to acknowledge the data and programming support provided by Social and Scientific Systems, Inc. in Maryland. Address correspondence to Jayasree Basu, PhD, Agency for Healthcare Research and Quality, 540 Gaither Road, Rockville, MD 20850, USA. E-mail: [email protected]
176
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
177
analysis, the study indicates that improvements in socioeconomic conditions and geographic access may have helped improve the quality of primary care received by the elderly over the last decade, particularly among some minority groups. Keywords: Preventable hospitalizations, small area analysis, ambulatory care sensitive conditions, quality of primary care, changes over time
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
BACKGROUND The hospital admission for ambulatory care sensitive conditions (ACSCs) is a validated indicator of impeded access to good primary and preventive care services (Millman, 1993). Several prior studies have explored the quality of primary care by using ambulatory care sensitive (or preventable) hospitalization rates. A few of them looked into elderly hospitalization patterns. Very few studies (e.g., Kozak et al., 2001; McCall et al., 2004) examined the changes in elderly hospitalization patterns over a longer time period for preventable conditions. Such a study provides an overview of how access to quality primary care has changed in the nation. The issue is particularly important for the elderly as there were several federal legislative initiatives during 1995 to 2005 that could have had direct and indirect impacts on the access to care for the elderly. We examine the predictors of ACSC admissions in small geographic areas in two cross-sections spanning an 11-year time interval (1995–2005). Previous studies examining the elderly trends have focused on earlier periods. Kozak et al. (2001) looked at the trends during 1980 to 1995 whereas McCall et al. (2004) used a more recent period (1992–2000). Both studies reported increasing rates of hospitalization for the elderly in general, and for specific ACSC conditions. The period we chose is a more recent one spanning the implementation of the Balanced Budget Act in 1997 (which resulted in substantial Medicare-managed care plan disenrollment), followed by a major push to enroll more elderly in Medicare managed care plans through implementation of the Medicare Modernization Act (MMA) of 2003. The supply of primary care services may have increased over this period due to several funding initiatives implemented by the federal government to increase access in underserved areas (Grumbach et al., 1997). The socioeconomic climate of the small areas may also have changed over the period. This study examines the net result of these policy initiatives and market changes on the ACSC hospitalization rates among the elderly over time. Using a rigorous conceptual framework, the major aim of this article is to identify the factors that predicted ACSC hospitalization rates in two cross sections of time 10 years apart, thereby assessing how or whether the influence of these explanatory variables on ACSC hospitalization rates changed with time. This research is important because it informs policy makers about the small-area variation in quality of primary care for the elderly; and whether access to primary care and preventive care services improved overall, for specific minority groups, or for Medicare-managed care enrollees relative to Medicare Fee-for-Service (FFS) enrollees over time. The association of adverse changes in the spatial patterns of ACSC admission rates over time with various sociodemographic and contextual factors can be used to identify small areas where policy interventions might be most fruitfully targeted to improve the current system.
CONCEPTUAL HYPOTHESIS There is a large body of literature about the geographic distribution of health services and patient access to care (e.g., Chandra & Skinner, 2003; Coughlin et al., 2002; Meade et al., 1988; Ricketts et al., 1994; Virnig et al., 2007). This study builds on the past research concerned with ACSC or preventable hospitalizations (e.g., Friedman & Basu, 2001; Krakauer et al., 1996; Mobley et al.,
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
178
J. BASU ET AL.
2006; Parchman & Culler, 1994). These studies examined variation in local hospitalization rates for ACSC conditions, chiefly in cities or hospital service areas. Studies also compared rates of hospitalization across catchment areas such as Health Service Areas (HSA) that were developed statistically as clusters of counties for patients going to the hospital (Makuc et al., 1991). This study focuses on primary care market areas (to be described below) to study the predictors of ACSC hospitalization rates. The conceptual models for ACSC hospitalizations usually examine supply and demand for outpatient services to explain the variations of admission rates (Basu & Friedman, 2001; Friedman & Basu, 2001). Consumers (patients and families) are one group of actors who seek, use, and pay for services. Primary care suppliers (physicians or physician directed), managed care plans, and hospital managers are independent groups of actors supplying services and sometimes making decisions on behalf of patients. ACSC hospitalization occurs when (a) demand for primary care exceeds its supply; or (b) it is “rational” not to use primary care because hospital care will typically be available and better paid for by the insurance; (c) economic, cultural, and social barriers prevent utilization of primary care; or perhaps (d) primary care physicians (PCPs) are unsuccessful in reducing ACSC admissions. Some of the factors affecting the demand for outpatient care, and hence (inversely) the hospitalization for ACSC conditions, include poverty, education level, and public and private insurance (Billings et al., 1993; Weissman et al., 1992). A number of these determinants could vary among racial and ethnic groups. A growing body of elderly literature has found that Medicare beneficiaries in managed care receive more preventive services and have better outcomes than their FFS counterparts (Riley et al., 1994; Riley, Potosky, Klabunde, Warren, & Ballard-Barbash, 1999; Rizzo, 2005). Although the role of managed care market penetration and the spillover effects on other providers could be mixed, previous research found Medicare managed care enrollment to be associated with less use of hospitals for ACSC conditions (Basu & Mobley, 2007). The supply factors associated with ACSC admissions are inpatient bed capacity (Goodman, Gittelsohn, Chang, & Fleming, 1994; Krakauer et al., 1996), supplies of PCPs, and physician practice patterns (Homer et al., 1996; Parchman & Culler, 1994). A greater availability of PCPs, all else equal, would tend to make primary care more accessible and reduce average prices for primary care for potential patients who could be treated in an ambulatory setting, relative to hospital inpatient stays (Basu et al., 2002). Higher local hospital inpatient capacity would, on the other hand, lower the average cost of inpatient versus primary care and could increase ACSC admissions (Folland et al., 1997). Area characteristics such as elderly in poverty or in rural location are expected to exert important influences on increased rates of ACSC or preventable admissions (Mobley et al., 2006). Studies have found degree of remoteness and rural/urban residence to be positively associated (Ansari et al., 2006) and population density to be negatively associated with ACSC admissions (e.g., Schreiber & Zielinski, 1997). Several of these factors could have played changing roles in influencing ACSC hospitalizations during the period we studied. For example, the Medicare Modernization Act of 2003 may have increased preventive care services availability through increased participation by Medicare managed care plans with prescription drug coverage, which could have enhanced the quality of primary care offered by managed care plans.1 On the other hand, the growth of Private Fee-for-Service (PFFS) managed care plans after 2003 may have resulted in lesser care coordination and provision of preventive care delivery than what health maintenance organization (HMO) plans provided. The improved supply and allocation of PCPs into underserved (Grumbach et al., 1997) areas could have resulted in lower marginal benefit from adding PCPs over time. Several socioeconomic changes, such as increased average income and declining poverty, could have also boosted accessibility of primary care over time. Along with increasing mix of aged elderly and minority population affecting overall ACSC rates, increased socioeconomic conditions could also improve primary care access in some minority groups. In the multivariate analysis that follows, we examine the relative contributions of these factors and how and whether these contributions changed over time.
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
179
METHOD
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Scope The study used hospital discharge data for six states on the East and West Coast: Arizona (AZ), California (CA), Massachusetts (MA), Maryland (MD), New Jersey (NJ), and New York (NY) for the years 1995 and 2005. We first selected all elderly (age 65 and older) patients who were hospitalized for any conditions in these six states. We then selected 20 ACSC conditions based on criteria developed by Billings et al. (1993) and selected all elderly patients who were hospitalized with these conditions in the above states, to use in our analysis of ACSC hospitalization rates over time. The study used descriptive and multivariate analysis (to be described in detail below) to examine the predictors of small area ACSC rates in two cross-sections to assess any apparent changes over time. We classified patients by their area of residence, and used primary care service area (PCSA) as the natural market area in which patients receive preventive care (more below).
Design The unit of analysis is PCSA, which is a natural market defined by FFS Medicare patient flows to physician offices (Goodman et al., 2003). PCSAs were developed using Medicare utilization data to represent geographic approximations of markets for primary care services received by the elderly. We assume that these areas are the best approximation of the small market area in which elderly patients receive their ambulatory care, and these regions have been validated in previous research on elderly access to preventive care services (Mobley et al., 2006; Mobley et al., 2008). PCSAs are generally smaller and more numerous than counties. Multivariate analysis. We constructed two multivariate models, one for each year, to understand the relative importance of predictors of ACSC admission rates for each year. To gauge the changes in relative contributions of explanatory variables between 1995 and 2005, we also used pooled regression models that combined the data for these 2 years and computed coefficients of time interactions. The model was run using with and without state fixed effects. The rationale for fixed effects model lies in the fact that it controls for across the state variations due to other factors and thus partially adjusts for omitted variable bias in the estimation process (Greene, 1993). We used statistical significance tests to identify the predictors that significantly changed its effects (coefficients) over time. We focused on all ACSCs combined for the multivariate models and selected all elderly patients hospitalized with any ACSC in the six sample states. The outcome variable was ACSC hospitalization rates by PCSAs, derived as the number of ACSC admissions (i.e., those admitted with ACSCs) of elderly (65 and older) residents of each PCSA, divided by the number of all hospitalizations of elderly patients (65 and older) living in that PCSA. As control variables, we used compositional factors describing hospitalized patients as well as contextual factors describing socioeconomic characteristics, all aggregated to the PCSA level. The compositional factors represent characteristics of the entire hospitalized population, not the subset admitted for ACSCs. The independent variables were suggested by past literature (as mentioned earlier in the conceptual hypotheses section) and winnowed down to the final set after considering multicollinearity as well as predictive value. The variables include sociodemographic composition, composition of insurance coverage of all hospitalized patients, average disease severity of all hospitalized patients in the area, and health services resource availability (including inpatient hospital capacity and primary care physician supply). Table A1 in the appendix shows a complete list of the independent variables and sources of data used in the regression models.
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
180
J. BASU ET AL.
FIGURE 1 Sources of data and analytical file construction. HCUP SID D Healthcare Cost and Utilization Project; RTI D Research Triangle Institute; PCSA D primary care service area; ARF D area resource files.
We used the ordinary least squares (OLS) method to estimate multivariate regression models. The initial tests showed that ACSC hospitalization rates closely followed a normally distributed (bell-shaped) curve. We conducted additional tests for heteroscedasticity, and corrected standard errors in our models where appropriate. The sources of the data and the data file construction followed the steps outlined in Figure 1.
FINDINGS Multivariate Analysis Table 1 presents the summary statistics, and Table 2 shows the corresponding results from the multivariate models, estimated separately for each year. Table 3 reports selected variables from the interaction model without state fixed effects, and Table 4 reports the same with state fixed effects. Sample statistics. Table 1 reports the mean (unweighted) values of all the variables used in the regression analysis for 1995 and 2005. The table also reports the changes over 1995 to 2005 that were statistically significant, based on results of two sample t tests conducted for each variable. Table 1 indicates that ACSC admission rates for the elderly declined significantly as a whole in the six states. A few major significant changes occurred during the 11-year period in the demographic, contextual, as well as in the policy environment. First, the average number
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
181
TABLE 1 Mean Values for Regression Variables, 1995 and 2005 Total Six States
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Variable ACSC admissions per 1,000 discharges by elderly residents Proportion of elderly age 65–74 Proportion of elderly age 75–84 Proportion of elderly age 85 and older Proportion of elderly: Medicare-managed care plans Proportion of African American discharges Proportion of Hispanic discharges Proportion of other race discharges Average income Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density PCP capacity per 1000 population Proportion of PCSA that is rural Proportion of elderly below the FPL in the PCSA
1995 N D 990
2005 N D 1,010
0.18 0.43 0.39 0.17 0.01 0.04 0.05 0.06 38,579.17 880.28 21.80 3,775.06 0.51 0.34 0.097
0.17* 0.37* 0.41* 0.21* 0.12* 0.05 0.08* 0.06 52,514.8* 3,518.87* 21.61 3,407.99 0.72* 0.29* 0.086*
Note. Mean values reported in this table are unweighted. N refers to the number of PCSAs. Regression analysis and mean values are estimated with data aggregated at the PCSA level. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; PCP D primary care physician; FPL D federal poverty level. *Two-tailed t test indicates statistically significant differences between 1995 and 2005 .p < 0:05/. TABLE 2 Ordinary Least Squares Regression Results, 1995 and 2005: Hospitalization for ACSC Conditions per 1,000 Discharges by Elderly Residents in PCSA of Patient Origin 1995 All States (N D 990) Variable Proportion of elderly age 65–74 Proportion of elderly age 75–84 Proportion of elderly covered by Medicare HMOs Proportion of African American discharges Proportion of Hispanic discharges Proportion of other race discharges Average income (in $1,000s) Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density Primary care physician capacity per 1,000 population Proportion of PCSA that is rural Proportion of elderly below the FPL in the PCSA Intercept R2
ˇ 0.160 0.188 0.068 0.083 0.054 0.025 1.83E-07 1.16E-06 0.00013 2.24E-07 0.036 0.019 0.080 0.334 0.2304
p Value 0.000 0.000 0.025 0.000 0.000 0.126 0.233 0.122 0.220 0.096 0.000 0.000 0.104 0.000
2005 All States (N D 1,010) ˇ 0.128 0.165 0.044 0.052 0.036 0.018 3.43E-07 3.10E-08 0.00025 3.08E-07 0.007 0.001 0.174 0.301 0.3044
p Value 0.000 0.000 0.000 0.000 0.000 0.102 0.000 0.419 0.000 0.001 0.003 0.796 0.000 0.000
Note. Standard errors of the regression estimates are heteroscedasticity consistent. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; HMO D health maintenance organization; FPL D federal poverty level.
182
J. BASU ET AL.
TABLE 3 Ordinary Least Squares Regression Results, 1995 and 2005 Pooled Data: Hospitalization for ACSC Conditions per 1,000 Discharges by Elderly Residents in PCSA, Selected Variables All States (N D 2,000)
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Variable Proportion of elderly age 65–74 * Year Proportion of elderly age 75–84 * Year Proportion of elderly covered by Medicare HMOs * Year Proportion of African American discharges * Year Proportion of Hispanic discharges * Year Proportion of other race discharges * Year Average income (in $1,000s) * Year Hospital inpatient capacity per 1000 population * Year Average distance from patient to hospital ZIP codes * Year Population density * Year Primary care physician capacity per 1000 population * Year Proportion of PCSA that is rural * Year Proportion of elderly below the FPL in the PCSA * Year
ˇ 0.032 0.023 0.024 0.032 0.018 0.045 0.000 0.000 0.000 0.000 0.029 0.020 0.094
p Value 0.480 0.736 0.451 0.090 0.235 0.029 0.349 0.133 0.333 0.001 0.000 0.001 0.094
Note. Standard errors of the regression estimates are heteroscedasticity consistent. The R 2 is 0.2748. Only interaction (with time) variables are reported (1995 D 0, 2005 D 1) in this Table. The full results are available upon request. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; HMO D health maintenance organization; FPL D federal poverty level.
of primary care physicians per 1,000 population increased by 34% from 0.51 to 0.71 in all six states between 1995 and 2005. Also, significant increase in the Medicare-managed care enrollment occurred over the 11-year period. The proportion of the elderly Medicare beneficiaries who were enrolled in managed-care plans increased from 1% in 1995 to 12% overall in these six states. The proportions of elderly under the federal poverty level (FPL) declined. The sociodemographic composition also changed, with greater increase in hospitalizations of minority and aged elderly. Among other changes, average household income increased, and population density declined. In addition, average rurality declined and average inpatient hospital capacity per 1,000 population increased. Multivariate Models Table 2 shows the OLS regression results for 1995 and 2005 using the same set of variables for comparison. Table 2 indicates that younger elderly age groups (65–74 and 75–84) had a negative association with ACSC hospitalization rates (relative to those in 85 and older age group), experiencing also a relative decline in impact over time. The two racial and ethnic groups, African Americans and Hispanics, were associated with increased ACSC hospitalizations, whereas both experiencing relative declines in effects over time. The proportions of enrollment under Medicare managed care plans was a significant predictor in 1995 and 2005, but its marginal contribution toward declines in ACSC rates dropped in 2005 (ˇ D 0:045, p < :05) relative to 1995 (ˇ D 0:068, p < 0:01). Likewise, the PCP density was a significant predictor in 1995 and 2005, being associated with reductions in ACSC rates. However, there was a substantial decline in marginal contributions (from 0.036 to 0.008, p < 0:01) between 1995 and 2005. The area socioeconomic indicators, measured by average area income and proportions below poverty, were found to have
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
TABLE 4 Ordinary Least Squares Regression Results, State Fixed Effects Pooled Regression Models (1995 and 2005): Hospitalization for ACSC Conditions per 1,000 Discharges by Elderly Residents in PCSA
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
All States (N D 2,000) Variable CA (NY D 0) AZ (NY D 0) NJ (NY D 0) MA (NY D 0) MD (NY D 0) Proportion of elderly age 65–74 Proportion of elderly age 75–84 Proportion of elderly covered by Medicare HMOs Proportion of African American discharges Proportion of Hispanic discharges Proportion of other race discharges Average income (in $1,000s) Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density Primary care physician capacity per 1000 population Proportion of PCSA that is rural Proportion of elderly below the FPL in the PCSA Year CA * Year AZ * Year NJ * Year MA * Year MD * Year Proportion of elderly age 65–74 * Year Proportion of elderly age 75–84 * Year Proportion of elderly covered by Medicare HMOs * Year Proportion of African American discharges * Year Proportion of Hispanic discharges * Year Proportion of other race discharges * Year Average income (in $1,000s) * Year Hospital inpatient capacity per 1,000 population * Year Average distance from patient to hospital ZIP codes * Year Population density * Year Primary care physician capacity per 1000 population * Year Proportion of PCSA that is rural * Year Proportion of elderly below the FPL in the PCSA * Year Intercept R2
ˇ 0.037 0.035 0.011 0.005 0.001 0.146 0.158 0.059 0.075 0.115 0.012 9.19E-07 8.28E-08 0.000029 2.17E-07 0.007 0.011 0.047 0.044 0.024 0.003 0.005 0.007 0.004 0.041 0.011 0.044 0.034 0.063 0.044 3.60E-07 5.84E-10 0.00002 2.58E-07 0.002 0.013 0.115 0.347 0.3676
p Value 0.000 0.000 0.000 0.285 0.804 0.000 0.001 0.110 0.000 0.000 0.475 0.000 0.910 0.612 0.089 0.314 0.015 0.235 0.299 0.000 0.748 0.236 0.196 0.574 0.355 0.868 0.239 0.028 0.001 0.029 0.040 0.999 0.787 0.109 0.809 0.027 0.017 0.000
Note. Standard errors of the regression estimates are heteroscedasticity consistent. ACSC D ambulatory care sensitive conditions; PCSA D primary care service area; NY D New York; CA D California; AZ D Arizona; NJ D New Jersey; MA D Massachusetts; MD D Maryland; HMO D health maintenance organization; FPL D federal poverty level.
183
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
184
J. BASU ET AL.
played important role in ACSC hospitalizations over time. The proportion of elderly under poverty exerted a stronger influence over time, being positively associated with ACSC hospitalization rates both years. Likewise, higher average area income levels played a significant role in pulling ACSC rates down and this effect became stronger over time. Among other findings, the rural urban continuum (measured by population density and proportions in rural area) changed the signs of coefficients between 1995 and 2005 to have caused declines in ACSC admission rates. The 1995 and 2005 results were further analyzed to examine the factors with significant changes in relative contributions over time (Table 3). The chow test was conducted on 1995 to 2005 pooled data to assess whether there were structural changes in the two elderly regression models over time (Greene, 1993). The chow test result (F D 5:73, prob < 0.001) indicated that, overall, the effects of the different parameters, measured by regression coefficients, changed over time. To assess the changes in relative contributions of individual factors, we used pooled regression models to examine the interaction coefficients of the independent variables with time. The partial results from this model are presented in Table 3. The results of this analysis in Table 3 highlight several interesting findings. The “other races” showed significantly higher association. The marginal contribution of rural population on ACSC admissions declined significantly over time, as was the contribution of the PCP density. On the other hand, population density became a stronger predictor of ACSC rates. None of the other sociodemographic or policy-driven factors was found to be significantly more or less predictive of ACSC admission rates over time. Somewhat different findings were obtained by using a state fixed effects model (Table 4) that shows the multivariate results for pooled 1995 and 2005 data with state fixed effects. In this model, we use state dummies, year and state interactions, and interactions of each explanatory variable with year. The model differs from the one presented before (Table 3) in terms of better controlling for unobserved differences across states. We present the full model with interaction coefficients to isolate the changes in marginal contributions of the explanatory variables over time (please note that separate estimates for 1995 and 2005 can be obtained from this table). The interaction coefficients indicate that several variables had statistically significant changes in marginal contributions over time to the ACSC rates, namely, race, income, and rurality. African Americans and Hispanics had smaller marginal contributions to ACSC rate over time, whereas the contributions of other races to ACSC rate increased significantly. In contrast to previous models (Tables 2 and 3), the negative association of income with ACSC rate was found significantly weaker over time, whereas stronger association of ACSC rate with poverty rate was observed. The marginal contributions of rurality declined significantly over time.
DISCUSSION Using small area analysis, the study identified several predictors from multivariate models for individual years which are consistent with previous research and as hypothesized. For example, the study showed that ACSC admission rates were inversely related to the availability of local PCPs in both years. This is consistent with previous research using larger area level units (HSAs or counties) that reported negative associations between the two variables (Parchman & Culler, 1994), particularly for adults (Friedman & Basu, 2001), although some studies also found weak to no association between primary care supply and ACSC admissions (Grumbach et al., 1997). We also found that managed care was associated with a decline in ACSC admissions for the elderly in both years, which was also consistent with previous research (Basu & Mobley, 2007). We found minorities, aged elderly, and poverty to be associated with higher ACSC rates, all of which are consistent with previous research findings related to socioeconomic status and ACSC
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
185
hospitalizations (e.g., Begley, Slater, Engel, & Reynolds, 1994; Billings et al., 1993; Blustein et al., 1998). ACSC admission rates for the elderly declined significantly as a whole in the six states. By conducting two cross-sectional studies as well as combined analyses for 2 years, we were able to assess how the effects of the explanatory variables on ACSC hospitalization rates changed over time. These findings, along with data presented in Table 1, would suggest some potential forces behind the observed reduction in ACSC admissions over time. Our descriptive statistics and multivariate results indicate that the period 1995 to 2005 was characterized by several changes in the socioeconomic environment. Our findings suggest that, although the percentage of hospitalizations increased among the minority elderly population between 1995 and 2005, this did not appear to have significantly driven the ACSC hospitalization rates upward. This is because the association between minorities and ACSC admission rates did not commensurately increase. The association between the proportions of Hispanics and African Americans with ACSC rate actually dropped between 1995 and 2005. Although the proportions of Hispanics among hospitalized elderly population increased from 4.5% to 6.8%, the noticeably smaller effect of Hispanics may indicate improved access to primary care among this subgroup over time. Although “other races” showed significantly higher marginal contributions over time, the relative size of this minority group did not increase to have a sizable impact on ACSC rates. Overall, our findings indicate that the racial ethnic minorities commonly associated with higher ACSC rates did not play a significant role in driving ACSC rates up in our study states. On the other hand, older age groups had a stronger influence on increasing ACSC rates over time. As the younger elderly group (65–74) became smaller over time, and their marginal contributions declined, those in the age 85 and older group as well as their contribution to ACSC rates were likely to become stronger. These older age groups are also those likely to have higher rates of preventable hospitalizations. The two socioeconomic factors, the poverty rate among the elderly, and average per capita household income in PCSAs, influenced the changes in ACSC rates in different ways. The average area income levels increased over time, although it was found to be a weaker predictor of ACSC rates in 2005 once unobserved variances across the states was controlled (Table 4). On the other hand, poverty levels became a stronger predictor of ACSC rates; though poverty rates in general declined. Because there was a decline over time in the proportions of elderly under the FPL, the growth of ACSC admissions attributable to poverty was somewhat moderated. The geographic location was found to be an important factor explaining reductions in ACSC rates. The proportion of population in rural areas significantly declined, as was the marginal contribution of rural population on ACSC admissions, indicating better geographic access over time. Because rurality was associated with higher ACSC rates, the declining size and influence of rurality jointly could have led to an overall decline in ACSC rates. The two factors, PCP capacity and Medicare managed care penetration rates, were of primary interest to us because they expanded in size and were likely to be related to lower ACSC hospitalization. The mean levels of both increased in size substantially between 1995 and 2005, whereas the relative contributions did not improve. Although experiencing a major spurt of growth, the role of increased penetration of managed care in the Medicare market was not clear. Although associated with reductions in ACSC rates in both years, its relative marginal contribution fell somewhat over time. A likely reason could be the growth in PFFS plans after 2003, with less emphasis than HMO models on care coordination and prevention, which might have made these plans less willing and efficient in providing primary care. On the other hand, the Medicare HMO plans in 1995 were probably more organized and effective in delivering primary care. The decline in the coefficients in primary care density between 1995 and 2005 could be explained by increased supply and distribution of primary care physicians over the last decade and a potential decline in additional benefits from further increase. Another factor could be the increasing role
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
186
J. BASU ET AL.
of nonphysician personnel, international medical graduates, as well as community health centers in primary care delivery, which was not captured in the primary care variable we constructed. Several limitations of the study should be acknowledged. The six states examined are not fully representative of the entire U.S. population and market conditions. They were chosen based on data availability appropriate for multivariate analysis. Another limitation could be that the study examines only one outcome indicator for the quality of primary care. Although our quality indicator is quite important regarding total social cost of care and a signal of effective management of chronic illness, other measures could be examined. This is an important area of ongoing research. In terms of methods, data limitations did not allow us to adjust well for severity of elderly population morbidity in the comparative analysis for 1995 to 2005, which may cause omitted-variables bias in the impact of managed care penetration, to the extent that managed care plans adversely or beneficially select patients by illness severity. However, we used other area variables as proxies for severity, such as poverty level, age composition, racial composition, average income, as well as average distance from patient’s ZIP-code to hospital ZIP-code (Basu & Friedman, 2007).
CONCLUSION The study uses small area analysis to examine the predictors of ACSC hospitalization rates and their potential roles over 1995 to 2005. The period 1995 to 2005 is particularly significant in terms of changes in elderly hospitalization patterns as well as in socioeconomic environments. The study indicates that improvements in socioeconomic conditions and geographic access may have helped improve the quality of primary care received by the elderly over the last decade (and reduced ACSC hospitalization rates). This change was observed among the minorities, particularly among Hispanics and African Americans. Several policy-driven factors, for example, Medicare managed care and PCP density, were not found significantly more effective over time.
NOTES 1. In 2003, a pilot project was implemented which offered Preferred Provider Organizations (PPOs) with drug coverage to the elderly in 21 states, see Pope et al. (2005). 2. There are some gaps in the ZIP code tabulation area (ZCTA) coverage in very rural areas where no ZIP codes existed in 1990 or 2000, and there are no accompanying census data for these areas. These gaps correspond to the addresses for rural patients with no ZIP code (or giving a PO box that has no latitude or longitude). We did not encounter any problems, because all patients kept in the data set had residential ZIP codes. 3. The census data is nested by ZCTA inside the PCSAs, but the county data may span several PCSAs and cross their borders.
REFERENCES Ansari, Z., Laditka, J. N., & Laditka, S. B. (2006). Access to health care and hospitalization for ambulatory care sensitive conditions. Medical Care Research and Review, 63(6), 719–741. Begley, C. E., Slater, C. H., Engel, M. J., & Reynolds, T. F. (1994). Avoidable hospitalizations and socio-economic status in Galveston County, Texas. Journal of Community Health, 19(5), 377–387. Basu, J., & Friedman, B. (2001). Preventable illness and out-of-area travel of children in New York counties. Health Economics, 10, 67–68. Basu, J., Friedman, B., & Burstin, H. (2002). Primary care, HMO enrollment, and hospitalization for ambulatory care sensitive conditions. Medical Care, 40(12), 1260–1269.
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
AMBULATORY CARE SENSITIVE HOSPITALIZATIONS
187
Basu, J., & Mobley, L. (2007). Do HMOs reduce preventable hospitalizations for Medicare beneficiaries? Medical Care Research and Review, 64(5), 544–567. Basu, J., & Friedman, B. (2007). A reexamination of distance as a proxy for severity of illness and implications for differences in utilization by race/ethnicity. Health Economics, 16, 687–701. Billings, J., Zeitel, L., Lukomnik, J., Carey, T. S., Blank, A. E., & Newman, L. (1993). Impact of socioeconomic status on hospital use in New York City. Health Affairs, 12, 162–173. Blustein, J., Hanson, K., & Shea, S. (1998). Preventable hospitalizations and socioeconomic status. Health Affairs, 17(2), 177–189. Chandra, A., & Skinner, J. (2003). Geography and racial health disparities (NBER Working Paper No. W9513). Cambridge, MA: National Bureau of Economic Research. Retrieved from http://www.uber.org/papers/w9513 Coughlin, S., Thompson, T., Seeff, L., Richards, T., & Stallings, F. (2002). Breast, cervical, and colorectal carcinoma screening in a demographically defined region of the southern U.S. Cancer, 95, 2211–2222. Folland, S. A., Goodman, A. C., & Stano, M. (1997). The economics of health care (2nd ed.). Upper Saddle River, NJ: Prentice Hall. Friedman, B., & Basu, J. (2001). Health insurance, primary care, and preventable hospitalization of children in a large state. American Journal of Managed Care, 7(5), 473–481. Goodman, D. C., Fisher, E. S., Gittelsohn, A., Chang, C. H., & Fleming, C. (1994). Why are children hospitalized? The role of non-clinical factors in pediatric hospitalizations. Pediatrics, 93(6, Pt 1), 896–902. Goodman, D. C., Mick, S. S., Bott, D., Stukel, T., Chang, C., Marth, N., : : : Carretta, H. J. (2003). Primary care service areas: A new tool for the evaluation of primary care services. Health Services Research, 38(1), 287–310. Green, W. H. (1993). Econometric analysis (2nd ed.). New York: MacMillan. Grumbach, K., Vranizan, K., & Bindman, A. B. (1997). Physician supply and access to care in urban communities. Health Affairs, 16(1), 71–85. Homer, C. J., Szilagyi, P., Rodewald, L., Bloom, S. R., Greenspan, P., Yazdgerdi, S., : : : Perrin, J. M. (1996). Does quality of care affect rates of hospitalization for childhood asthma? Pediatrics, 98(1), 18–23. Kozak, L., Hall, M. J., & Owings, M. F. (2001). Trends in avoidable hospitalizations, 1980–1998. Health Affairs, 20(2), 225–232. Krakauer, H., Jacoby, I., Millman, M., & Lukomnik, J. E. (1996). Physician impact on hospital admission and on mortality rates in the Medicare population. Health Services Research, 31(2), 191–210. Makuc, D., Hugland, B., Ingram, D. D., Kleinman, J. C., & Feldman, J. J. (1991). Health service areas for the United States. Vital and Health Statistics, 2, Nov.(112), 1–102. McCall, N., Brody, E., Mobley, L., & Subramanian, S. (2004, April). Investigation of increasing rates of hospitalization for ambulatory care sensitive conditions (avoidable hospitalizations) among Medicare beneficiaries. Final report. Retrieved from http://www.cms.hhs.gov/Reports/Downloads/McCall_2004_3.pdf Meade, M. S., Florin, J. W., & Gesler, W. M. (1988). Medical geography. New York, NY: Guilford Press. Millman, M. (Ed.). (1993). Access to health care in America. Washington, DC: National Academy Press. Mobley, L., Root, E., Anselin, L., Gracia, N., & Koschinsky, J. (2006, May 15). Spatial analysis of elderly access to primary care services. International Journal of Health Geographics, 5(1), 19. Retrieved from http://www.ij-healthgeographics. com/content/5/1/19 Mobley, L. R., Kuo, T., & Andrews, L. S. (2008). How sensitive are multilevel regression findings to defined area of context? A case study of mammography use in California. Medical Care Research and Review, 65(3), 315–337. Parchman, M. L., & Culler, S. (1994). Primary care physicians and avoidable hospitalization. Journal of Family Practice, 39(2), 123–128. Pope, G., Olmsted, E. A., Kautter, J., Mobley, L. R., & Greenwald, L. M. (2005, April). The Medicare preferred provider organization demonstration: Plan offerings and enrollment. Retrieved from http://www.cms.hhs.gov/HealthCareFinancing Review/downloads/06Springpg95.pdf Ricketts, T. C., Savitz, L. A., Gesler, W. M., & Osborne, D. N. (1994). Geographic methods for health services research. Lanham, MD: University Press of America. Riley, G. F., Potosky, A. L., Lubitz, J. D., & Brown, M. L. (1994). Stage of cancer at diagnosis for Medicare HMO and fee-for-service enrollees. American Journal Public Health, 84(10), 1598–1604. Rizzo, J. (2005). Are HMOs bad for health maintenance? Health Economics, 14(11), 1117–1131. Schreiber, S., & Zielinsky, T. (1997). The meaning of ambulatory care sensitive admissions: Urban and rural perspectives. Journal of Rural Health, 13(4), 276–284. Virnig, B., Scholle, S., Chou, A., & Shih, S. (2007). Efforts to reduce racial disparities in Medicare managed care must consider the disproportionate effects of geography. American Journal of Managed Care, 13, 51–56. Weissman, J. S., Gatsonis, C., & Epstein, A. M. (1992). Rates of avoidable hospitalization by insurance status in Massachusetts and Maryland. Journal of the American Medical Association, 268(17), 2388–2394.
188
J. BASU ET AL.
APPENDIX Table A1. Independent Variables and Data Sources Variables
Downloaded by [The University of Manchester Library] at 03:54 12 October 2014
Proportion of elderly age 65 to 74
Proportion of elderly age 75 to 84
Proportion of elderly age 85 and older Proportion of elderly: Medicare FFS plans
Proportion of elderly: Medicare-managed care plans
Proportion of White discharges
Proportion of African American discharges Proportion of Hispanic discharges
Proportion of other race discharges
Average income Hospital inpatient capacity per 1,000 population Average distance from patient to hospital ZIP codes Population density PCP capacity per 1000 population
Proportion of elderly below the federal poverty level in the PCSA Proportion of rural
Description Number of elderly residents age 65–74 years divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents age 75–84 years divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents age 85 and older divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents with Medicare FFS insurance divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of elderly residents with Medicare-managed care insurance divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of White elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of African American elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of Hispanics elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Number of other races elderly residents divided by all elderly residents from the PCSA discharged anywhere in the state for any condition Average household income of PCSA Total inpatient days in hospitals in the PCSA per 1,000 population of the PCSA Average distance in miles between patient’s ZIP code and hospital ZIP code in each PCSA Thousands of total population per square mile land areaa Family practitioners, internists, and specialists expressed as counts per 1,000 U.S. populationa;b Proportion of elderly below federal poverty levela Proportion of population living in rural census tractsa
Source HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID
HCUP-SID American Hospital Association HCUP-SID U.S. Census, HRSA AMA/AOA Masterfiles, ARF and HRSA U.S. Census, HRSA US Census, HRSA
HCUP-SID D Healthcare Cost and Utilization Project; PCSA D primary care service area; FFS D Fee for Service; HRSA D Health Resources and Services Administration; AMA/AOA D American Medical Association/American Osteopathic Association; PCP D primary care physician; ARF D Area Resource Files. a For HRSA data and more explanations of the variables, see http://datawarehouse.hrsa.gov/. b 1995 data are from Area Resource Files (ARF), county level, interpolated to represent PCSA.
