|
|
| |||||||||||||||||||||||||||||||
Background Geographic variations in the use of hospital services are associated with differences in the availability of hospital beds. There continues to be uncertainty about the extent to which unmeasured case-mix differences explain these findings. Previous research showed that the number of occupied beds per capita in Boston was substantially higher than the number of occupied beds per capita in New Haven, Connecticut, and that overall rates of hospital utilization were higher for Boston residents than for New Haven residents.
Methods We used Medicare claims data to study cohorts of Medicare beneficiaries 65 years of age or older and residing in Boston or New Haven who were initially hospitalized for one of five indications (acute myocardial infarction, stroke, gastrointestinal bleeding, hip fracture, or potentially curative surgery for breast, colon, or lung cancer). Residents of Boston or New Haven who were discharged between October 1, 1987, and September 30, 1989, were enrolled in the cohort corresponding to the earliest such admission and followed for up to 35 months.
Results The relative rate of readmission in Boston as compared with New Haven was 1.64 (95 percent confidence interval, 1.53 to 1.76) for all cohorts combined, with a similarly elevated rate for each of the five clinical cohorts and each age, sex, and race subgroup examined. Hospital-specific readmission rates varied substantially among the hospitals in Boston and were higher than those in New Haven. No relation was found between mortality (during the first 30 days after discharge or over the entire study period) and either community or hospital-specific readmission rates.
Conclusions Regardless of the initial cause of admission, Medicare beneficiaries who were initially hospitalized in Boston had consistently higher rates of readmission than did Medicare beneficiaries hospitalized in New Haven. Differences in the severity of illness are unlikely to explain these findings. One possible explanation is a threshold effect of hospital-bed availability on decisions to admit patients.
Prior research comparing hospital services in Boston and in New Haven, Connecticut, has highlighted the issues involved. The two communities are similar demographically, and teaching hospitals account for more than 85 percent of the admissions in both cities1. Excluding beds occupied by nonresidents, people in the Boston hospital-service area occupied 4.5 beds per 1000 population, whereas people in the New Haven hospital-service area occupied 2.9 beds per 1000 population1. Overall mortality rates in the two communities were nearly identical,3 as were the rates of admission for specific conditions for which physicians agree on the need for hospitalization (e.g., myocardial infarction),3 which suggested that the rates of underlying illness were similar. The overall rates of admission in Boston were nearly 50 percent higher, and the average length of stay in Boston was about 7 percent longer1. There continues to be uncertainty about the extent to which unmeasured case-mix differences explain geographic variations in hospitalization rates.
In the United States, patients with certain conditions, such as hip fractures, myocardial infarction, stroke, and gastrointestinal bleeding, are almost always hospitalized. The rates of admission for such conditions, which are determined principally by the illness rates themselves, differ by less than 10 percent between Boston and New Haven1. We therefore studied cohorts of Medicare beneficiaries in the two cities, comprising patients initially hospitalized for one of five such clear-cut indications. The cohorts were selected to minimize base-line differences between the two areas or among the hospitals. We followed patients in these cohorts for up to 35 months after their initial admission in order to examine patterns of hospital readmission and survival.
Methods
Sources of Data
We used routine hospital discharge data (Medicare Part A) and the corresponding enrollment data (Denominator File) maintained by the Health Care Financing Administration4. The discharge files contain a record for each hospitalization of a Medicare beneficiary, regardless of where it occurred, including abstracted clinical and administrative data. The Denominator Files contain data on all people in the Medicare program, including dates of birth and death, sex, race, and residential ZIP Code. Unique identifiers allow these two data sources to be linked. Data on the numbers of hospital beds and the rates of occupancy were obtained from the American Hospital Association5.
Study Population
The people studied were Medicare beneficiaries 65 years of age or older who resided in the Boston or New Haven hospital-service area and who were discharged between October 1, 1987, and September 30, 1989, after hospitalization for acute myocardial infarction, stroke, gastrointestinal bleeding, hip fracture, or potentially curative surgery for colon, lung, or breast cancer. The Boston hospital-service area, which includes Boston, Chelsea, Revere, Brookline, and a portion of Newton, Massachusetts, had a population of approximately 706,000 in 1990. The New Haven hospital-service area, which includes New Haven, West Haven, East Haven, North Haven, Orange, Woodbridge, Bethany, Hamden, North Branford, Branford, Guilford, and Madison, Connecticut, had a population of approximately 382,000 in 1990. People over the age of 99 years or of unknown race were excluded. Cohorts were assembled according to the cause of the admission. The diagnosis-related groups (DRGs) and International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes used to define the clinical cohorts are listed in the Appendix. Patients with more than one eligible admission were assigned to the cohort corresponding to their first discharge. (For example, a patient initially hospitalized for the treatment of cancer and subsequently hospitalized because of a myocardial infarction was assigned to the cancer cohort.) Patients were followed from the date of the index admission until the date of their death, departure from the area, enrollment in a health maintenance organization, or disenrollment from Medicare Part A or until the end of the study period (August 31, 1990), whichever occurred first.
Statistical Analyses
The analyses of readmission according to place of residence entailed three steps. First, we calculated unadjusted readmission rates for each six-month interval after the index admission, according to the place of residence (Boston or New Haven), for each of the five clinical cohorts and for all the cohorts combined (Figure 1). For the combined cohorts readmission rates were also calculated according to age (either 65 to 74 years of age or 75 or over), sex, and race (white or nonwhite), according to the place of residence (Figure 2). For each six-month interval, the number of cohort members surviving to the midpoint of the interval was the denominator for the readmission rate; the numerator was the number of admissions during the interval.
|
|
Third, we combined all the cohorts and used the same analytic approach to determine cause-specific readmission rates in Boston and New Haven, according to the major diagnostic category of the DRG to which each readmission was assigned, excluding diagnostic categories with fewer than 50 readmissions (Figure 3).
|
We used logistic regression (for 30-day mortality) and the Cox proportional-hazards model10 to calculate the relative survival of each cohort, according to the index hospital and controlling for age, sex, race, clinical cohort, and coexisting conditions, as compared with the survival of the Yale-New Haven Hospital cohort. We also used the Cox model to examine the association between the readmission rate and survival for up to three years, by including in the model the relative rate of readmission (Table 1) for each hospital's cohort.
|
Results
Cohort Enrollment and Follow-up
Between October 1, 1987, and September 30, 1989, a total of 5969 Medicare beneficiaries in Boston and 2696 Medicare beneficiaries in New Haven were hospitalized for one of the five conditions specified in the study and were eligible for enrollment. Before the end of the study period (August 31, 1990), 2355 members of the Boston cohort (39.5 percent) died, as did 1036 members of the New Haven cohort (38.4 percent). Mortality rates were similar in Boston and New Haven for each six-month interval after the index admission. For an additional 122 Bostonians (2.0 percent) and 21 New Haven residents (0.8 percent), follow-up was terminated before the end of the study, primarily because of departure from the area.
Readmissions According to Place of Residence
Regardless of the initial diagnosis, readmission rates were significantly higher in Boston than in New Haven (Figure 1). For all cohorts combined, the readmission rate in Boston was 1.64 times higher (95 percent confidence interval, 1.53 to 1.76). For each clinical cohort, the readmission rates in Boston were higher than those in New Haven in almost every six-month interval after the index admission. Readmission rates in Boston were also higher than those in New Haven for all age, sex, and race subgroups (Figure 2) and for most causes of readmission (Figure 3), whether or not the cause of readmission was related to the cause of the index admission. For example, separate analyses (data not reported) showed that in the cohort of patients in Boston who were initially hospitalized because of myocardial infarction, the relative risk of readmission because of cardiovascular disease was 1.74 (95 percent confidence interval, 1.43 to 2.11); for all other medical conditions, the relative risk of readmission was 1.90 (95 percent confidence interval, 1.51 to 2.39).
Readmissions According to Index Hospital
We ranked the major teaching hospitals in each area that had at least 150 index admissions during the study, according to the adjusted relative rate of readmission among their patient cohorts (Table 1). Yale-New Haven Hospital had the lowest rate of readmission. Massachusetts General Hospital had the lowest readmission rate among the Boston hospitals (relative risk, 1.50; 95 percent confidence interval, 1.30 to 1.73). Three hospitals in Boston (Carney Hospital, St. Elizabeth's Hospital, and Boston University Medical Center) had readmission rates that were almost twice as high as those of Yale-New Haven Hospital and the Hospital of St. Raphael, also in New Haven.
The proportion of readmissions to the index hospital ranged from a low of 62 percent at Boston City Hospital to a high of 90 percent at the Hospital of St. Raphael in New Haven. Although the number of hospital beds per capita is smaller in New Haven than in Boston, occupancy rates at the New Haven and Boston hospitals were similar during the study period. There was no correlation between the relative readmission rate and the occupancy rate at each hospital (r = -0.34, P = 0.31 by Pearson's correlation coefficient). The higher readmission rates at the Boston hospitals were not associated with shorter lengths of stay. The mean length of stay for the index admissions in Boston was 14.8 days, as compared with 13.5 days in New Haven (P = 0.001 by Student's t-test), and for the readmissions, the mean length of stay was 11.3 days in Boston and 11.2 days in New Haven (P = 0.71 by Student's t-test).
Readmission and Mortality
Higher rates of readmission were not associated with either lower or higher mortality rates. Mortality rates within 30 days after the index admission did not differ significantly between the two cities (relative risk in Boston as compared with New Haven, 0.90; 95 percent confidence interval, 0.79 to 1.04). The overall mortality for the entire period of the study was essentially the same (relative risk in Boston as compared with New Haven, 0.98; 95 percent confidence interval, 0.91 to 1.05). There was no relation between hospital-specific 30-day mortality rates and hospital readmission rates (r = -0.06, P = 0.86). Higher hospital-specific readmission rates were not associated with improved overall survival (data not shown).
Discussion
We found that Medicare beneficiaries who were hospitalized in Boston for myocardial infarction, stroke, hip fracture, gastrointestinal bleeding, or potentially curative surgery for cancer had a 64 percent higher average readmission rate over the subsequent three years than Medicare beneficiaries hospitalized in New Haven for these conditions. The relative risk of hospitalization was higher in each subgroup of the Boston cohort, whether defined on the basis of patients' demographic characteristics (age, sex, or race) or clinical characteristics at the time of admission. When readmission rates were examined at the hospital level, substantial variations were found between Boston and New Haven and among the Boston hospitals.
We believe that differences in the severity of illness among the patient populations in the two cities are unlikely to explain these findings. If they were due to differences in the severity of illness, all the following would have to be true. First, differences in the severity of illness in these cohorts would have to be unrelated to 30-day mortality rates, an outcome that has been used to develop severity-of-illness models12,13. Second, for each clinical condition, Boston residents would have to be sicker than their counterparts in New Haven, and the magnitude of this difference would have to be consistent (e.g., myocardial infarctions would have to be 1.6 times more severe and cancer 1.6 times more advanced among Boston residents). Finally, the most severely ill patients in Boston would have to be systematically underrepresented at some of the city's most prominent hospitals. This seems unlikely.
The differences we observed are not explained by the substitution of nursing home care for inpatient care. According to the 1990 U.S. Census, 7.5 percent of Boston residents 65 years of age or older are in nursing homes, as compared with 5.6 percent of New Haven residents 65 or older. Outpatient care might also be substituted for inpatient care. In a previous study of variations in physician services in the United States in 1989, we found no evidence of the substitution of outpatient for inpatient services across geographic regions14. Moreover, Medicare beneficiaries in Boston had higher per capita expenditures on outpatient physician services ($697) in 1989 than did beneficiaries in New Haven ($647)14. Our findings cannot be explained by differences in the use of Veterans Affairs (VA) hospitals. Using Veterans Affairs data for 1989, we found that 1.8 percent of the Boston cohort was admitted to such hospitals, as compared with 1.6 percent of the New Haven cohort. Finally, the increased rate of readmission has been consistent over time. An analysis of comparable data from 1984 through 19876 showed that the relative readmission rate in Boston as compared with New Haven was 1.48 (95 percent confidence interval, 1.36 to 1.61), and the correlation between hospital-specific readmission rates in the earlier period and those presented here is 0.90 (P<0.001).
We have previously shown an association between the availability of hospital beds and overall hospital discharge rates1,3,6. The patterns of readmission among subgroups of Medicare beneficiaries and over time suggest a possible explanation: a threshold effect of hospital-bed availability on clinical decision making, in which available hospital resources and clinical judgments combine to determine per capita hospitalization rates.
For most groups of people, the availability of hospital resources is determined by public and private investment decisions, competition, and patterns of regionalization that have been developed over many years with little knowledge of the relative size or health status of the population served (especially at the level of the individual hospital). In contrast, health maintenance organizations that own their hospitals appear to limit the number of beds available to approximately 2 per 1000 enrollees15.
Clinical judgments are then made in the context of whatever hospital resources are available. For some conditions, such as hip fracture or acute myocardial infarction, the incidence of illness essentially determines the need for hospital care. Hospitalization rates for these conditions differ according to the diagnosis (there are more myocardial infarctions than hip fractures) but vary little according to the geographic area16 and are nearly identical in Boston and New Haven3. For most conditions, however, there are no categorical rules that determine the need for admission. The diagnosis clearly influences physicians' decisions: in each city, readmission rates differed according to the diagnosis. But the availability of hospital beds also appears to exert an important influence on readmission. The relative rate of readmission in Boston as compared with New Haven was about 60 percent higher in each of the demographic and clinical subgroups examined (Figure 1 and Figure 2). The consistency of the increased relative rates of readmission in Boston as compared with New Haven suggests that the greater availability of hospital beds increases the probability of admission as a consequence of any outpatient or emergency room visit.
The threshold effect does not appear to work as a simple sorting mechanism on the basis of the relative severity of illness. If it did, then the excess admissions in Boston would be accounted for by patients who were less severely ill. Data from our earlier research, however, show that the process is more complex. We found that, as compared with New Haven residents, a larger overall proportion of Boston residents were admitted to the hospital, they were admitted more frequently, and overall case-fatality rates were lower3. On average, therefore, hospital inpatients in Boston were less severely ill than their counterparts in New Haven. At the same time, however, 40 percent of all deaths among Medicare beneficiaries in Boston occur in the hospital, as compared with 31 percent among Medicare beneficiaries in New Haven3. This suggests that physicians in Boston, as compared with those in New Haven, admit more patients who are less severely ill as well as more patients who are less severely ill as well as more patients close to death.
Our data do not support the hypothesis that poor decision making on the part of physicians explains the higher hospital admission rates in Boston. A threshold effect offers what we believe is a simpler and more plausible explanation: physicians in both cities make difficult judgments in caring for sick people, with different levels of available per capita hospital resources.
In our study, a survival benefit for Medicare beneficiaries in Boston was not associated with the higher hospitalization rate for the city as a whole or for any individual hospital. Although hospitalization may offer other benefits besides prolonging life, it also has well-recognized complications,17 particularly in the elderly18. We acknowledge, however, that a study of two cities, involving a relatively small number of hospitals, can neither prove the existence of a threshold effect of hospital-bed availability nor evaluate the association between the rate of hospitalization for acute care and mortality.
Supported by grants from the Agency for Health Care Policy and Research (R18-HS05745) and the Health of the Public Program of the Pew Charitable Trusts and the Rockefeller Foundation.
We are indebted to Robert Glynn, Grace Lu-Yao, and Alan Gittelsohn for their assistance with the statistical analyses, to Chiang-hua Chang for assistance with the analyses, and to Kristy Patterson and Holly DeBlois for assistance in the preparation of the manuscript.
Source Information
From the Veterans Affairs Medical Center, White River Junction, Vt. (E.S.F.), and the Department of Medicine (E.S.F.) and Center for the Evaluative Clinical Sciences (E.S.F., J.E.W., T.A.S., S.M.S.), Dartmouth Medical School, Hanover, N.H.
Address reprint requests to Dr. Fisher at the Dartmouth-Hitchcock Medical Center, HB 7250, 319 Strasenburgh Hall, Hanover, NH 03755-3862.
References
The definition of the five cohorts was based on either DRGs or specific combinations of ICD-9-CM diagnosis (DX) and procedure (SX) codes. The cohorts of patients with acute myocardial infarction, stroke, and gastrointestinal bleeding included all patients with hospital records containing the following DRGs: acute myocardial infarction, DRG 121, 122, or 123; stroke, DRG 14; and gastrointestinal bleeding, DRG 174 or 175. The only exclusions were patients with acute myocardial infarction who had been discharged after a hospital stay of less than five days (n = 166). The cohort of patients with hip fractures included all patients with discharge records containing one of the following ICD-9-CM DX codes: 820.0 to 820.03, 820.09, 820.1 to 820.12, 820.19, 820.2 to 820.22, 820.3 to 820.32, 820.8, or 820.9. The cohort of patients with cancer admitted for surgery included all patients with discharge records containing the following combinations of ICD-9-CM SX and DX codes: for colon cancer, SX 45.7 to 45.8 or 48.4 to 48.6 and DX 153 or 154; for lung cancer, SX 32.29 to 32.59 and DX 162; and for breast cancer, SX 85.21 to 85.23, 85.31 to 85.36, or 85.41 to 85.48 and DX 174.
| |||||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2008 Massachusetts Medical Society. All rights reserved. |
Previous
