Associations Among Hospital Capacity, Utilization, and Mortality of U.S. Medicare Beneficiaries, Controlling for Sociodemographic Factors

Associations Among Hospital Capacity, Utilization, and Mortality of U.S. Medicare Beneficiaries, Controlling for Sociodemographic Factors

Elliott S. Fisher

Objective. To explore whether geographic variations in Medicare hospital utilization rates are due to differences in local hospital capacity, after controlling for socioeconomic status and disease burden, and to determine whether greater hospital capacity is associated with lower Medicare mortality rates.

Data Sources/Study Setting. The study population: a 20 percent sample of 1989 Medicare enrollees. Measures of resources were based on a national small area analysis of 313 Hospital Referral Regions (HRR). Demographic and socioeconomic data were obtained from the 1990 U.S. Census. Measures of local disease burden were developed using Medicare claims files.

Study Design. The study was a cross-sectional analysis of the relationship between per capita measures of hospital resources in each region and hospital utilization and mortality rates among Medicare enrollees. Regression techniques were used to control for differences in sociodemographic characteristics and disease burden across areas.

Principal Findings. The per capita supply of hospital beds varied by more than twofold across U.S. regions. Residents of areas with more beds were up to 30 percent more likely to be hospitalized, controlling for ecologic measures of socioeconomic characteristics and disease burden. A greater proportion of the population was hospitalized at least once during the year in areas with more beds; death was also more likely to take place in an inpatient setting. All effects were consistent across racial and income groups. Residence in areas with greater levels of hospital resources was not associated with a decreased risk of death.

Data Collection/Extraction Methods. Data on the study population were obtained from Medicare enrollment (Denominator File) and hospital claims files (MedPAR) and U.S. Census files.

Conclusions. Residence in areas of greater hospital capacity is associated with substantially increased use of the hospital, even after controlling for socioeconomic characteristics and illness burden. This increased use provides no detectable mortality benefit.

Key Words. Hospital bed supply, hospitalization, mortality, Medicare For the past 50 years, researchers have documented marked geographic variations in the amount of hospital care received by Americans (Chassin, Brook, and Park 1986; Connell, Day, and LoGerfo 1981; Krakauer et al. 1996; Lembcke 1952; Lewis 1969; Wennberg and Gittelsohn 1973). And the costs associated with greater use of the hospital are substantial: per capita spending on acute hospital care in 1993 was only $914 in Seattle, Washington and $851 in New Haven, Connecticut, but was $1,475 in Chicago, Illinois and $1,202 in Boston, Massachusetts (Wennberg 1996).

The underlying cause of these variations, however, has been the subject of debate. Some researchers have emphasized the role of local hospital capacity (Wennberg, Freeman, and Culp 1987; Wennberg and Gittelsohn 1973), while others have stressed the importance of differences in socioeconomic and health status (Billings, Zeitel, Lukomnik, et al. 1993; Hofer, Wolfe, Tedeschi, et al. 1998); a few have argued that local bed supply has no influence on utilization (Bindman, Grumbach, Osmond, et al. 1995). The importance of the debate lies in the policy implications of the alternative hypotheses. If differences in utilization are due to different levels of need, then the variations observed in hospital utilization are not a cause for concern. However, if hospital capacity and utilization relative to underlying need do vary across communities, then the possibility arises to achieve substantial cost savings without harming health.

In this study, we address two questions: first, whether the differences in utilization experienced by Medicare populations living in areas with a greater per capita bed supply are due to differences in population characteristics (such as income or disease burden); and second, whether, after controlling for these population characteristics, those who live in areas of greater bed supply have a detectable benefit in terms of lower mortality rates.

METHODS

Study Population

The study population was a 20 percent sample of Medicare enrollees age 65 and over who were eligible for Medicare Part A (hospital insurance) during calendar year 1989 (mortality analyses) or 1990 (utilization analyses) identified from the Medicare enrollment (“Denominator”) file (Mitchell, Bubolz, Paul, et al. 1994). In this age group, Medicare provides coverage to over 95 percent of the U.S. population (Fisher, Baron, Malenka, et al. 1990). We excluded members of risk-bearing health maintenance organizations (4.0 percent of eligible enrollees) because their discharge records were incomplete, and those residing in zip codes for which no census data were available (0.5 percent of eligible enrollees). The final study population was 5.53 million (representing 27.6 million Medicare enrollees).

MEASURES

Dependent Variables. Discharges were ascertained using Medicare Part A hospital discharge files (“MedPAR” file) (Mitchell, Bubolz, Paul, et al. 1994). We used the enrollment file to ascertain all deaths during 1989 for the study population.

Population Characteristics. Individual level data were available for age, race, and gender from the Medicare Denominator File. Because Medicare’s files include only four racial designations (white–87.5 percent, black–7.4 percent, other–2.0 percent, and unknown–3.l percent), we collapsed these categories to black and non-black (which includes unknown). Data from the 1990 U.S. Census, measured at the level of the postal zip code, were used to estimate individuals’ living arrangements (rural, suburban, urban), education, income, marital, and employment status (for the entire resident population); median household income and poverty status (age 65 and over); and three measures of disability (also for the population over age 65): those with limitations in self-care needs, limitations in mobility, and work disability. We also used the hospital discharge file to control for differences across Hospital Service Areas (HSAs) in illness rates. We selected specific events for which hospitalization is a proxy for the incidence of disease and not a reflection of greater hospital capacity (Fisher et al. 1994; Wennberg, Freeman, and Culp 1987; Wennberg, McPherson, and Caper 1984): hospitalizations for hip fracture, cancer of the colon or lung treated surgically, acute myocardial infarction, and stroke. The age-, sex-, and race-adjusted HSA level incidence rates for these conditions were used as an additional measure of disease burden.

Definition of Hospital Service Areas and Hospital Referral Regions. These were defined as previously described (Wennberg 1996; Wennberg and Gittelsohn 1973). Briefly, 1989 hospital discharges for Medicare enrollees 65 and older were used to assign each zip code to the hospital of predominant use, with the additional requirement of geographic contiguity. In metropolitan areas, hospitals with overlapping service areas were grouped into single HSAs. These individual HSAs were then aggregated into 313 Hospital Referral Regions (HRRs) based on travel patterns for cardiothoracic and neurologic surgery.

Measures of Hospital Resources (Table 1). Our primary measure of the amount of hospital resources available to the Medicare population was the in-area bed supply, a simple tally of the number of staffed hospital beds in hospitals within the HRR divided by the resident population in 1990. This is the most appropriate measure for analyses that examine the associations between bed availability and utilization. We also employed this measure for our analyses of mortality. The per capita supply of in-area beds, however, has serious limitations as a measure of resource inputs for the Medicare population, failing to account for variations in occupancy rates, border crossing, and the share of local hospital resources received by the population under age 65. Any absence of an association between greater inputs and improved mortality could simply reflect imprecision in this measure. We therefore repeated the mortality analyses using two additional measures: (1) the age-sex-race-adjusted inpatient day rate in the HRR, a direct measure of the hospital resources provided to the Medicare population in the inpatient setting; and (2) the inpatient day rate in the six months prior to death among those who died (calculated using deaths that occurred in 1994, to avoid including any deaths among the current study population). The patient day rate will be confounded by local illness rates (and will be interpretable as a measure of resource inputs only if adequate adjustment across areas in Medicare beneficiaries’ risk of death is achieved). The inpatient day rate in the last six months of life, however, should be unconfounded by area level illness rates, as the population in whom the patient day rate is measured will be similarly ill (in terms of their risk of death) across geographic areas.

Physicians. We also controlled for the local per capita supply of clinically active physicians in each HSA based on data obtained from the American Medical Association and the American Osteopathic Association for 1993.

Regression Analyses

Unit of Analysis. For the regression models examining mortality, the unit of analysis was the individual Medicare enrollee. For the models examining discharges, the unit of analysis was the five-year age-sex-race and zip code stratum, since we did not distinguish among multiple hospitalizations to the same person. Individual level data were available for age, race, gender, and mortality. Other covariates were measured at the zip code, HSA, or HRR level.

Regression Models. To estimate the relationship between in-area bed supply and hospital discharge rates (e.g., Figure 1), we used Poisson regression (McCullagh and Nelder 1983), weighting by the number of subjects in each age-sex-race and zip code stratum, controlling for all demographic, socioeconomic, and health status variables described earlier. To estimate the relationship between the three measures of hospital resources and Medicare mortality, we used logistic regression (Hosmer and Lemeshow 1989), controlling for the same demographic, socioeconomic, and health status variables. For computational efficiency, we used strata based on five-year age groups, race, sex, and zip code, weighting each stratum by its number of Medicare enrollees. The dependent variable was the stratum-specific 1989 mortality rate. The models are available from the authors. Point estimates and confidence intervals for the odds ratios and rate ratios were obtained by transforming the corresponding regression parameters in the usua l fashion. We incorporated variance over-dispersion in the estimates of all standard errors to account for any clustering of deaths or hospitalizations within strata, as well as for multiple hospitalizations to the same patient (Donner and Donald 1987).

We conducted separate regressions for blacks and non-blacks and for two income strata among non-blacks. The models for blacks excluded HSAs with fewer than 500 black Medicare enrollees. Low-income groups were defined as those living in zip codes where white median family income was less than $14,000, approximately 23 percent of the non-black Medicare population. We also repeated the regressions in strata based on the size of the HSA, in HSAs with teaching hospitals, and in HSAs with Medicare HMO penetration of 3 percent or less. Similar results were obtained. All tests were performed at the 5 percent level of significance and were two-sided.

RESULTS

Association Between Hospital Bed Supply and Hospital Utilization

Residence in areas with greater per capita numbers of hospital beds is associated with higher hospitalization rates, predominantly for medical (nonsurgical) causes of admission (Figure 1). Medicare enrollees living in areas with more beds are more likely to be admitted to the hospital at least once during the year. Among those who die, death is more likely to occur in the hospital in areas with a greater supply of beds. Although differences in utilization are apparent across income and racial groups, the association between increased capacity and utilization is found in each income and race group.

Associations Between Hospital Resources and Mortality

The associations between in-area hospital beds and mortality were weak and mostly non-significant at conventional levels of statistical significance (Table 2). In no subgroup of the population was a significant mortality benefit from increased local bed availability detected. We repeated these analyses using the other measures of hospital resource inputs and again found no benefit from increased inputs overall (Table 3) or in any age, race, or income subgroup of the population (data not shown). Regardless of the dependent variable used in the analysis, no mortality benefit from increased hospital bed availability or propensity to use the hospital was found.

DISCUSSION

Our analyses of Medicare enrollees’ hospital utilization indicate that, although sociodemographic characteristics influence utilization, differences in sociodemographic characteristics do not explain the large differences in utilization seen across U.S. regions. These findings are consistent with prior research (Connell, Day, and LoGerfo 1981; Fisher et al. 1994; Wennberg, Freeman, and Culp 1987; Wennberg et al. 1989; Wennberg and Gittelsohn 1973). The influence of capacity on utilization operates across the spectrum of illness, influencing the proportion of the population hospitalized and the intensity of terminal care. These findings all suggest that the higher rates of hospitalization in areas of greater per capita bed supply are largely “discretionary” in the sense that similar patients living in low-bedded areas are more often treated outside the hospital.

We recognize that our cross-sectional analyses of the association between increased capacity and mortality represent only a preliminary effort to examine the health outcomes associated with treating these discretionary cases in the inpatient setting as opposed to the outpatient setting. For Medicare enrollees in general and for the subgroups examined, the association between increased hospital capacity and mortality was either flat or positive. In no subgroup were greater inputs of hospital resources associated with lower mortality.

Our inability to detect a beneficial effect of the increased utilization induced by greater hospital capacity could be due to residual confounding. We tried to assess this possibility in several ways. First, we controlled for differences in underlying socioeconomic characteristics and disease burden. Although we used ecologic measures of education, income, poverty status, and disability, such ecologic measures have been shown to provide valid estimates of individual attributes in studies based on census-tract level data (Geronimus, Bound, and Neidert 1996; Krieger 1992) and reasonable proxies in studies based on zip code level data (Gornick, Eggers, Reilly, et al. 1996; Hofer, Wolfe, Tedeschi, et al. 1998). Second, the model that estimated the marginal effects of just those hospital inputs associated with a greater intensity of end-of-life care showed no beneficial effects. As discussed in Table 1, it is possible that the differences in the estimates of the odds ratios between Models 1 and 3 (in Table 3) rep resent the effect of residual confounding in Model 2. At the same time, it is also possible that the difference between Models 2 and 1 reflects the effect of measurement error, as in-area beds represents a poor measure of the marginal inputs of hospital care to the Medicare population. Third, our models gave consistent results across strata based on population characteristics (age, race, and income) and in strata selected on the basis of health system attributes (models restricted to HSAs with teaching hospitals or to HRRs with a Medicare HMO penetration of 3 percent or less).

Other limitations further underscore the need for caution in interpreting the results of our mortality analyses. The only outcome examined was mortality; it is possible that increased hospital resources are associated with improved functional status or quality of life. Our study also says nothing about the relationship between investments in non-hospital resources and outcomes. It would also be inappropriate to generalize these findings to other populations, such as Medicare beneficiaries enrolled in risk-bearing managed care organizations or to those under age 65. The uninsured may be particularly vulnerable to policies that seek to reduce local acute care capacity, (Bindman, Keane, and Lurie 1990) given their dependence on hospital uncompensated care.

Nevertheless, our findings underscore the need to shift the focus of future research from the causes to the consequences of variations in hospital utilization: what are the benefits (or harms) of the increased utilization caused by greater capacity? Several recent studies suggest that harm is possible (Brennan, Leape, Laird, et al. 1991; Guadagnoli, Hauptman, Ayanian, et al. 1995; Weinberger, Oddone, and Henderson 1996). Could the positive association between increased hospital capacity and mortality that appears in several of our models represent a causal relationship? Given the limitations of our data set, our analyses can do no more than raise the possibility. Further work will be required to learn how and when capacity-driven utilization leads to benefit or harm.

Address correspondence to Elliott S. Fisher, M.D., M.P.H., Professor of Medicine and Community and Family Medicine, Center for the Evaluative Clinical Sciences, Dartmouth Medical School, 7251 Strasenburgh Hall, Hanover NH 03755. Dr. Fisher is also Co-Director of the VA Outcomes Group, Dept. of Medicine, VA Medical Center, White River Junction VT. John E. Wennberg, M.D., M.P.H. is the Peggy Y. Thompson Professor for Evaluative Clinical Sciences and Director, Center for the Evaluative Clinical Sciences, Dartmouth Medical School. Therese A. Stukel, Ph.D. is Professor of Community and Family Medicine (Biostatistics), Center for the Evaluative Clinical Sciences, Dartmouth Medical School and Professor of Mathematics, Dartmouth College. Jonathan S. Skinner, Ph.D. is Senior Research Associate, Center for the Evaluative Clinical Sciences, Dartmouth Medical School and the John French Professor and Vice-Chair of Economics at Dartmouth College. Sandra M. Sharp, S.M. is a Research Associate, Center for the Evaluative Clin ical Sciences, Dartmouth Medical School. Jean L. Freeman, Ph.D. is Associate Professor, Department of Geriatrics, The University of Texas Medical Branch, Galveston. Alan M. Gittelsohn, Ph.D. is Visiting Professor of Community and Family Medicine, Center for the Evaluative Clinical Sciences, Dartmouth Medical School. This article, submitted to Health Services Research on June 23, 1998, was revised and accepted for publication on March 5, 1999.

This work was originally supported by a grant from the Robert Wood Johnson Foundation.

REFERENCES

Billings, J., L Zeitel, J. Lukomnik, T. S. Carey, A. E. Blank, and L. Newman. 1993. “Impact of Socioeconomic Status on Hospital Use in New York City.” Health Affairs 12 (1): 162-73.

Bindman, A. B., K. Grumbach, D. Osmond, M. Komaromy, K. Vranizan, N. Lurie, J. Billings, and A. Stewart. 1995. “Preventable Hospitalizations and Access to Health Care.” Journal of the American Medical Association 274 (4): 305-11.

Bindman, A. B., D. Keane, and N. Lurie. 1990. “A Public Hospital Closes: Impact on Patients’ Access to Care and Health Status [see comments].” Journal of the American Medical Association 264 (22): 2899-904.

Brennan, T. A., L. L. Leape, N. M. Laird, L. Hebert, A. R. Localio, A. G. Lawthers, J. P. Newhouse, P. C. Weiler, and H. H. Hiatt. 1991. “Incidence of Adverse Events and Negligence in Hospitalized Patients: Results of the Harvard Medical Practice Study I.” The New England Journal of Medicine 324 (6): 370-76.

Chassin, M. R., R. H. Brook, and R. E. Park. 1986. “Variations in the Use of Medical and Surgical Services by the Medicare Population.” The New England Journal of Medicine 314 (5): 285-90.

Connell, F. A., R. W. Day, and J. P. LoGerfo. 1981. “Hospitalization of Medicaid Children: Analysis of Small Area Variations in Admission Rates.” American Journal of Public Health 71(6): 606-13.

Donner, A., and A. Donald. 1987. “Analysis of Data Arising from a Stratified Design with the Cluster as Unit of Randomization.” Statistics in Medicine 6 (1): 43-52.

Fisher, E.S.,J. A. Baron, D.J. Malenka, J. Barrett, and T. A. Bubolz. 1990. “Overcoming Potential Pitfalls in the Use of Medicare Data for Epidemiologic Research.” American Journal of Public Health 80 (12): 1487-90.

Fisher, E. S., J. E. Wennberg, T. A. Stukel, and S. M. Sharp. 1994. “Hospital Readmission Rates for Cohorts of Medicare Beneficiaries in Boston and New Haven.” The New England Journal of Medicine 331 (15): 989-95.

Geronimus, A.T.,J. Bound, and L.J. Neidert. 1996. “On the Validity of Using Census Geocode Characteristics to Proxy Individual Socioeconomic Characteristics.” Journal of the American Statistical Association 91(434): 529-37.

Gornick, M. E., P. W. Eggers, T. W. Reilly, R. M. Mentnech, L. K. Fitterman, L. E. Kucken, and B. C. Vladeck. 1996. “Effects of Race and Income on Mortality and Use of Services Among Medicare Beneficiaries [see comments].” The New England Journal of Medicine 335 (11): 791-99.

Guadagnoli, E., P. J. Hauptman, J. Z. Ayanian, C. L. Pashos, B. J. McNeil, and P. D. Cleary. 1995. “Variation in the Use of Cardiac Procedures After Acute Myocardial Infarction.” The New England Journal of Medicine 333 (9): 573-78.

Hofer, T. P., R. A. Wolfe, P.J. Tedeschi, L. F. McMahon, and J. P.. Griffith. 1998. “Use of Community Versus Individual Socioeconomic Data in Predicting Variation in Hospital Use.” Health Services Research 33 (2, Part I): 243–59.

Hosmer, D. W, and S. Lemeshow. 1989. Applied Logistic Regression. New York: Wiley and Sons.

Krakauer, H., I. Jacoby, M. Millman, and J. E. Lukomnik. 1996. “Physician Impact on Hospital Admission and on Mortality Rates in the Medicare Population.” Health Services Research 31(2): 191–211.

Krieger, N. 1992. “Overcoming the Absence of Socioeconomic Data in Medical Records: Validation and Application of a Census-based Methodology.” American Journal of Public Health 82 (5): 703–10.

Lembcke, P. A. 1952. “Measuring the Quality of Medical Care Through Vital Statistics Based on Hospital Service Areas: I. Comparative Study of Appendectomy Rates.” American Journal of Public Health 42 (March): 376–86.

Lewis, C. E. 1969. “Variations in the Incidence of Surgery.” The New England Journal of Medicine 281 (16): 880–84.

McCullagh, P., and J. Nelder. 1983. Generalized Linear Models. New York: Chapman and Hall.

Mitchell, J. B., T. Bubolz, J. E. Paul, C. L. Pashos, J. J. Escarce, L. H. Muhlbaier, J. M. Wiesman, W. W. Young, R. S. Epstein, and J. C. Javitt. 1994. “Using Medicare Claims for Outcomes Research.” Medical Care 32 (7, Supplement): JS38–51.

Weinberger, M., E. Z. Oddone, and W. G. Henderson. 1996. “Does Increased Access to Primary Care Reduce Hospital Readmissions?” The New England Journal of Medicine 334 (22): 1441–47.

Wennberg, J. 1996. The Dartmouth Atlas of Health Care. Hanover, NH: American Hospital Press, Inc.

Wennberg, J. E., J. L. Freeman, and W. J. Culp. 1987. “Are Hospital Services Rationed in New Haven or Over-utilized in Boston?” Lancet 1 (8543): 1185–89.

Wennberg, J. E., J. L. Freeman, P.. M. Shelton, and T. A. Bubolz. 1989. “Hospital Use and Mortality Among Medicare Beneficiaries in Boston and New Haven.” The New England Journal of Medicine 321 (17): 1168–73.

Wennberg, J. E., and A. M. Gittelsohn. 1973. “Small Area Variations in Health Care Delivery.” Science 182 (117): 1102–108.

Wennberg, J. E., K. McPherson, and P. Caper. 1984. “Will Payment Based on Diagnosis-Related Groups Control Hospital Costs?” The New England Journal of Medicine 311 (5): 295–300.

Measures of Hospital Resources Used

in the Analyses

Measure (Units) How Defined

In-area beds The number of staffed acute care

(beds per 1000) hospital beds within the region,

divided by the resident general

population.

Mean: 3.6

Range: 1.1-6.6

Inpatient day rate The number of inpatient days per

(days per capita) year experienced by the Medicare

population divided by the resident

Medicare population (age, sex,

race adjusted).

Mean: 2.8

Range: 1.4-4.6

Inpatient day rate The number of inpatient days

in last six months experienced by Medicare enrollees

of life in their last six months of life

(days per capita) (age, sex, race adjusted).

Mean: 11.2

Range: 4.4-22.9

Measure (Units) Comment

In-area beds Measures local availability of beds.

(beds per 1000) Poor measure of hospital resources

for the Medicare population

because of variations in occupancy

rates, border crossing, and the

share of hospital care received by

the non-Medicare population.

Inpatient day rate Most direct measure of the level of

(days per capita) hospital resources provided to the

Medicare population. Differences

in use across regions, however,

will be highly confounded by local

illness rates.

Inpatient day rate Measures the propensity to use

in last six months the hospital independent of local

of life illness rates, because the persons

(days per capita) used to calculate the measure are

likely to be similarly ill across

areas: all are destined to die in

the next six months.

Associations Between the In-Area Bed Supply and the Odds

of Death in 1989 for Medicare Enrollees by Age, Race, and Income

Odds of Death Associated

with Increase of One Bed

per 1000 Population

Mortality Rate for Medicare Enrollees

(Per 1000) [*] odds ratio [+]

All Medicare Enrollees 50.8 1.003

Stratified by Age

Age 65-74 27.1 1.007

Age 75-84 61.2 1.007

Age 85 and over 148.0 0.994

Stratified by Race

Black [++] 56.7 1.009

Non-Black 50.3 1.002

Stratified by Income [ss]

Low income 52.0 0.987

Middle and high income 50.5 1.008

95% CI

All Medicare Enrollees (0.997, 1.009)

Stratified by Age

Age 65-74 (0.996, 1.018)

Age 75-84 (0.997, 1.017)

Age 85 and over (0.982, 1.006)

Stratified by Race

Black [++] (0.987, 1.031)

Non-Black (0.996, 1.009)

Stratified by Income [ss]

Low income (0.974, 1.001)

Middle and high income (1.000, 1.015)

(*.)Crude mortality rate (per 1000) for all Medicare enrollees over age 65 and not enrolled in risk-based HMO.

(+.)Odds ratio for linear trend derived from logistic regression model, controlling for age, race, and sex (individual level variables); income, education, disability, urban/rural location (zip code level variables); illness rates (HSA level variables); and region of the country. The odds ratio provides an estimate of the average increase in the odds of death for each additional bed per 1000 across U.S. HRRs.

(++.)Model for blacks excluded HSAs with fewer than 500 blacks (age 65 years and over).

(ss.)Mortality rate and models are for non-blacks only.

Associations Between Different Measures of

Hospital Resources Across Strata of U.S.

Distribution and Medicare Mortality Rate

odds ratio [*] 95% CI

Model 1

In-area hospital beds (Model 1)

As continuous variable 1.003 (0.997, 1.009)

Across strata

Under 2.5 1.000

2.5 to 3.5 1.022 (1.004, 1.040)

3.5 to 4.5 1.035 (1.015, 1.056)

4.5 and over 1.030 (1.009, 1.052)

Model 2

Inpatient days per capita

As continuous variable 1.076 (1.064, 1.089)

Across strata

Lowest quartile 1.000 —

Second quartile 1.030 (1.011, 1.049)

Third quartile 1.083 (1.061, 1.106)

Highest quartile 1.114 (1.084, 1.145)

Model 3

Inpatient days in six months prior to death

As continuous variable 1.007 (1.004, 1.010)

Across strata

Lowest quartile 1.000 —

Second quartile 1.040 (1.024, 1.057)

Third quartile 1.079 (1.062, 1.097)

Highest quartile 1.071 (1.049, 1.093)

(*.)Odds ratio derived from logistic regression model, controlling for age, race, and sex (individual level variables); income, education, disability, urban/rural location (zip code level variables); illness rates (HSA level variables); and region of the country. Odds ratio provides estimate of the average increase in odds of death across HRRs at the specified stratum of the U.S. distribution.

COPYRIGHT 2000 American College of Healthcare Executives

COPYRIGHT 2000 Gale Group